Author: joe

  • 辐射打出来的,还是免疫筛出来的?高海拔病毒突变谱的贡献拆分

    Evidence level: B

    高海拔样本的突变谱确实可能不同——但”辐射新打出来的突变”和”免疫筛剩下来的突变”会给出同一个读数,而它们各自留下的指纹几乎完全相反。

    正文

    上一篇把海拔梯度当天然实验室,结论是”解耦度不可跨病毒直接比”。这一篇处理一个更前置的问题:同一个病毒里看到的高/低海拔突变谱差异,至少有两个生产者——辐射新打出来的,或免疫筛选出来的。 两者的分子后果完全不同,却会对”突变率上升”这个标量给出同一个读数。

    辐射的指纹是结构性的,而它最奇怪的特征是”均匀”。对 12 例放疗后继发肿瘤做全基因组测序、对比 319 例未受照肿瘤,辐射相关肿瘤平均多出约 201 个缺失,长度 1 至 100 碱基对,接头处常有微同源;与未受照肿瘤不同——这些缺失在全基因组的密度不随序列上下文、复制时序、染色质结构变化,几乎像随机撒上去的(PMID 27615322)。同时平衡倒位显著增加。这条极重要:辐射留下的痕迹”不认识”基因组的局部地形。

    选择恰好相反:它极度依赖局部地形,因为选择发生在具体的位点上。识别它的标准做法是 dN/dS——某密码子位的非同义替换显著多于同义替换(ω>1)。但只看 ω>1 不够:共享祖先会让观察失去独立性,必须回到系统发育树上看同一改变是否在多个支系独立反复出现(趋同/并行演化),那才算选择的硬证据(PMID 30703578, 39411152)。辐射的指纹是”均匀”,选择的是”扎堆”。

    麻烦在于,辐射的指纹主要不在单碱基替换维度。

    切尔诺贝尔甲状腺癌研究给了一条很硬的分配证据:在 356 例肿瘤里,单碱基替换签名(SBS)没有一个与辐射剂量显著相关;而与剂量强相关的,是双链断裂末端连接修复相关的 indel 签名,尤其 ID8(P=7.3×10⁻⁹)(PMID 33888599)。细胞实验把产出结构量得同样清楚:1 Gy 电离辐射在哺乳动物基因组里平均造成 2.33 个突变事件/Gb,其中 2.15 个是 indel、0.17 个结构变异、0.01 个复杂重排——indel 占压倒多数(PMID 38359788)。高 LET 粒子更极端,会在径迹内制造多个双链断裂的密集簇,集中在异染色质区且大比例留在原地不修(PMID 25659339, 31952359);未修复的簇随后在细胞分裂时表现为染色体断裂与畸变(PMID 21527720)。两条合起来就是最实用的判据:要看辐射,去看 indel 谱和结构变异;要看选择,去看位点特异的 dN/dS。用转换/颠换比或突变率标量查辐射,等于在它最不显影的通道里找它。

    剂量账这次有了第二层意思。 地面宇宙线的年有效剂量在 0.3 至 0.4 mSv 量级,而 1 Gy 相当于 1000 mSv;即便高海拔通量数倍于海平面,累积到 1 Gy 也需千年级时间(切尔诺贝尔研究里的受照者接受的是单次治疗剂量)。这条账过去被用来否定”辐射能改写病毒”,现在看:剂量不足不是唯一问题,指标错配才是更缠人的那个——现行高/低海拔病毒比较几乎全在单碱基替换维度做(突变率标量、Ti/Tv 比、SNV 谱),而这些恰是辐射贡献最低的维度。

    选择侧也有自己的陷阱,而且正好指向同一个方向。

    SARS-CoV-2 全基因组分析发现了一个偏倚:C→U 替换的频率约为其他任何核苷酸替换的三倍,而 C→U 对应的三核苷酸上下文 NCN 显著富集高频率事件(PMID 33083804)。这直接毒化了 dN/dS 的解释——在 NCN 上下文里,某个位点独立出现多次的非同义替换,可以只是 RNA 聚合酶的机制性偏好,不是选择;该研究因此把 NCN 上下文事件全部剔除后才估计选择位点。反过来,dN/dS 也会漏掉真的选择:罗斯河病毒与巴马森林病毒里,同一 nsP1 位点在多个支系独立发生同一改变,趋同信号明确,但 dN/dS 没给出强阳性(PMID 39411152)——这类方法对非同义多态度的敏感性与对假阳性的控制之间存在张力。

    一头会假阳,一头会漏真。这就是”用选择特征反推辐射贡献”必须双向设防的原因:既要在偏倚上下文中剔除伪选择,又要用趋同/并行演化的独立判据补上 dN/dS 漏掉的位点(PMID 30703578, 28340348)。要拆分,至少三步:其一,结局改双维度——同时报告 indel 谱(长度分布、微同源比例、是否上下文无关)与位点特异 dN/dS;其二,算选择前先剔偏倚上下文(NCN,以及上一篇讨论过的 C→U 偏好多碱基替换簇);其三,归属判据用双重条件——”上下文无关、全基因组均匀新增的 indel”指向辐射,”位点特异、多支系独立重复的非同义替换”指向选择。两者同时出现时,才是真正需要拆分的样本。

    对易学的意义:《素问·五常政大论》讲”地有高下,气有温凉,高者气寒,下者气热“,并称”高者其气寿,下者其气夭”。这句古代讲的是人的寿夭,但给了一个可操作的结构:“高下”是形,”温凉”是气,形定气、气定果。现代版本里,”气”要拆成两个物理量——宇宙线/UV 这类外来施加(加害),与宿主免疫、传播结构这类内部筛选(取舍)。易学真正要问的不是”形变了没有”,而是”这个形变是气带来的,还是物自带的“。天时若真介入,应在最不依赖局部地形的通道里留下痕迹;若差异全集中在位点特异、反复出现的氨基酸改变上,那更像”物自带的取舍”被高海拔宿主环境放大,而不是”天时打上去的印”。

    证据与局限

    已验证: 辐射相关肿瘤的缺失签名在全基因组均匀分布、不随序列上下文/复制时序/染色质结构变化,且平衡倒位显著增加(PMID 27615322);切尔诺贝尔甲状腺癌中 SBS 签名与辐射剂量无显著关联,而 end-joining 相关 indel 签名(ID8)与剂量强相关(PMID 33888599);1 Gy 电离辐射平均产生 2.33 事件/Gb,其中 indel 2.15、结构变异 0.17、复杂重排 0.01(PMID 38359788);高 LET 粒子的簇状双链断裂大比例不可修复(PMID 25659339, 31952359),未修复簇导致染色体畸变(PMID 21527720);SARS-CoV-2 存在 C→U 约三倍偏倚,NCN 上下文富集高频率替换,须剔除后方可估计选择(PMID 33083804);dN/dS 与趋同/并行演化是识别正选择的两类互补判据(PMID 30703578)。

    部分验证: 并行替换的统计推断模型(ParaSel)在小儿麻痹疫苗衍生株中定位到 7 个正选择位点,但与位点特异 dN/dS 仅在 VP1.143 一处重合(PMID 28340348);罗斯河病毒/巴马森林病毒 nsP1 位点的趋同信号明确,但 dN/dS 未给出强阳性(PMID 39411152)——显示两类判据并非互相替代。

    待验证: 高/低海拔同病毒样本中,indel 谱维度是否存在”上下文无关的均匀新增缺失”这一辐射特征;把偏倚上下文剔除后,位点特异 dN/dS 的差是否仍然存在;辐射特征与选择特征在同一样本中同时出现时的比例与拆分方法;高海拔宿主环境(低氧等)对选择压力的放大倍数。

    相关阅读

    证据等级: B(辐射突变签名的 indel 主导、dN/dS 的方法学局限均有 PubMed 直接支撑;”在高/低海拔病毒样本中按此双维度拆分贡献”这一提案本身尚无直接研究)
    参考文献: PMID 27615322, PMID 33888599, PMID 38359788, PMID 25659339, PMID 31952359, PMID 21527720, PMID 33083804, PMID 30703578, PMID 28340348, PMID 39411152


    免责说明:本文属传统知识与科学文献的整理、对照与研究设计讨论,仅供理解与参考,不构成医疗建议,不能替代专业诊断与治疗。文中涉及的效应量、频率区间与暴露结论多为研究报告层面的推断,个体情况差异很大;如有身体不适或健康疑虑,请就医并由专业医师判断。

    Radiation or Immune Selection? Decomposing the Contribution of High-Altitude Viral Mutational Spectra

    Evidence level: B

    High-altitude samples indeed exhibit distinct mutation spectra—but “radiation-induced mutations” and “immune-screened mutations” yield identical readouts, yet leave nearly opposite fingerprints.

    Main Text

    The previous article treated the altitude gradient as a natural laboratory, concluding that “decoupling cannot be directly compared across viruses.” This article addresses a more upstream question: the observed high/low-altitude mutation spectrum differences within the same virus have at least two producers—radiation-induced mutations or immune-screened mutations. Their molecular consequences are fundamentally different, yet both yield the same scalar reading regarding “increased mutation rates.”

    The fingerprint of radiation is structural, and its most peculiar characteristic is its “uniformity.” In whole-genome sequencing of 12 radiotherapy-associated tumors compared to 319 non-irradiated controls, radiation-associated tumors averaged approximately 201 additional deletions of length 1 to 100 base pairs, often with microhomology at junctions; unlike non-irradiated tumors—these deletions show no density variation across the genome regardless of sequence context, replication timing, or chromatin structure, appearing almost random (PMID 27615322). Simultaneously, balanced inversions increase significantly. A critical point: radiation leaves traces that are “unaware” of the local topography of the genome.

    Selection operates in the exact opposite manner: it is highly dependent on local topography because selection occurs at specific sites. Its standard identification method is dN/dS—where nonsynonymous substitutions significantly exceed synonymous ones at a specific codon position (ω>1). However, observing ω>1 alone is insufficient: shared ancestry compromises independence; one must return to phylogenetic trees to verify whether the same change arises independently and repeatedly across multiple lineages (convergent/parallel evolution), which constitutes hard evidence of selection (PMID 30703578, 39411152). Radiation’s fingerprint is “uniform,” while selection clusters.

    The complication lies in the fact that radiation’s fingerprint is primarily not found at the single-nucleotide substitution dimension.

    The Chernobyl thyroid cancer study provided robust allocation evidence: among 356 tumors, no single-base substitution signature (SBS) showed significant correlation with radiation dose; conversely, signatures related to double-strand break end-joining repair showed strong correlation with dose, particularly ID8 (P=7.3×10⁻⁹) (PMID 33888599). Cell experiments quantify structural output equally clearly: 1 Gy of ionizing radiation causes an average of 2.33 mutation events/Gb in mammalian genomes, comprising 2.15 indels, 0.17 structural variants, and 0.01 complex rearrangements—indel dominates overwhelmingly (PMID 38359788). High LET particles are even more extreme, creating dense clusters of double-strand breaks within tracks concentrated in heterochromatin regions with a high proportion remaining unrepaired (PMID 25659339, 31952359); unrepaired clusters subsequently manifest as chromosomal breaks and aberrations during cell division (PMID 21527720). Taken together, these form the most practical criteria: to assess radiation, examine indel spectra and structural variants; to assess selection, examine site-specific dN/dS. Searching for radiation via transition/transversion ratios or mutation rate scalars is like searching for it in its least visible channel.

    The dose accounting now carries a second layer of meaning. The annual effective dose from ground cosmic rays is on the order of 0.3 to 0.4 mSv, whereas 1 Gy equals 1000 mSv; even with high-altitude fluxes several times that of sea level, accumulating to 1 Gy requires millennia (Chernobyl subjects received single-treatment doses). This accounting was previously used to deny “radiation rewriting viruses”; now viewed differently: insufficient dose is not the sole issue—mismatched metrics are more vexing—current high/low-altitude viral comparisons almost exclusively operate at the single-nucleotide substitution dimension (mutation rate scalars, Ti/Tv ratios, SNV spectra), which are precisely the dimensions where radiation contributes least.

    Selection side also harbors its own traps, pointing exactly in the same direction.

    Whole-genome analysis of SARS-CoV-2 revealed a bias: the frequency of C→U substitutions is approximately three times that of any other nucleotide substitution, and the corresponding trinucleotide context NCN significantly enriches high-frequency events (PMID 33083804). This directly poisons dN/dS interpretation—within NCN contexts, multiple independent nonsynonymous substitutions at a specific site can merely reflect the mechanistic bias of RNA polymerase, not selection; thus, the study removed all NCN context events before estimating selection sites. Conversely, dN/dS misses true selection: in Riverine and Bama forest viruses, the same nsP1 site underwent identical changes independently across multiple lineages with clear convergent signals, yet dN/dS failed to yield strong positivity (PMID 39411152)—there is tension between the sensitivity of these methods to nonsynonymous polymorphism and their control against false positives.

    One yields false positives, the other misses true signals. This is why “inferring radiation contribution via selection features” requires two-way defense: removing pseudo-selection within biased contexts while using independent criteria of convergent/parallel evolution to supplement sites missed by dN/dS (PMID 30703578, 28340348). To decompose, at least three steps are needed: First, modify the outcome to two dimensions—simultaneously report indel spectra (length distribution, microhomology proportion, context independence) and site-specific dN/dS; Second, remove biased contexts before calculating selection (NCN, and the C→U multi-base substitution clusters discussed in the previous article); Third, use dual conditions for attribution criteria—”context-independent, uniformly added indels across the genome” point to radiation, while “site-specific, independently repeated nonsynonymous substitutions across multiple lineages” point to selection. Only when both appear simultaneously is true decomposition required.

    Significance for Yijing Studies: The Suwen · Wuchang Zheng Dalun states “lands have high and low elevations; qi has warmth and coolness; high places have cold qi, low places have hot qi,” and claims “high places prolong life, low places shorten it.” This ancient text refers to human longevity, but provides an operable structure: “high and low are form, warmth and coolness are qi; form determines qi, qi determines outcome.” In modern versions, “qi” must be split into two physical quantities—externally imposed factors (such as cosmic rays/UV, i.e., harm) versus internal filtering mechanisms (such as host immunity/transmission structures, i.e., selection). What Yijing truly asks is not “has the form changed,” but “is this form change brought by qi, or inherent to the object.” If heaven’s timing truly intervenes, it should leave traces in channels least dependent on local topography; if differences concentrate entirely on site-specific, repeatedly occurring amino acid changes, it resembles “inherent selection amplified by high-altitude host environments” rather than “imprints struck by heaven’s timing.”

    Evidence and Limitations

    Verified: Deletion signatures in radiation-associated tumors are uniformly distributed across the genome, independent of sequence context/replication timing/chromatin structure, with significant increases in balanced inversions (PMID 27615322); in Chernobyl thyroid cancer, SBS signatures showed no significant association with radiation dose, whereas end-joining-related indel signatures (ID8) showed strong correlation (PMID 33888599); 1 Gy of ionizing radiation causes an average of 2.33 events/Gb, comprising 2.15 indels, 0.17 structural variants, and 0.01 complex rearrangements (PMID 38359788); clustered double-strand breaks from high LET particles are largely unrepaired (PMID 25659339, 31952359), leading to chromosomal aberrations when unrepaired (PMID 21527720); SARS-CoV-2 exhibits a ~3x bias toward C→U substitutions with NCN context enrichment requiring removal prior to selection estimation (PMID 33083804); dN/dS and convergent/parallel evolution represent two complementary criteria for identifying positive selection (PMID 30703578).

    Partially Verified: Statistical inference models for parallel substitutions (ParaSel) identified 7 positive selection sites in polio vaccine-derived strains, overlapping with site-specific dN/dS only at VP1.143 (PMID 28340348); clear convergent signals were observed at the nsP1 site in Riverine/Bama forest viruses, yet dN/dS did not yield strong positivity (PMID 39411152)—demonstrating that the two criteria are not mutually substitutable.

    To Be Verified: Whether “context-independent, uniformly added deletions” as a radiation feature exist in the indel spectrum dimension within same-virus high/low-altitude samples; whether a difference persists in site-specific dN/dS after removing biased contexts; the proportion and decomposition methods when radiation and selection features coexist in the same sample; and the amplification倍数 of selection pressure by high-altitude host environments (hypoxia, etc.).

    Related Reading

    Evidence Level: B (Radiation mutation signatures dominated by indels and methodological limitations of dN/dS are directly supported by PubMed; the proposal itself to “decompose contributions by this dual dimension in high/low-altitude viral samples” lacks direct research support)
    References: PMID 27615322, PMID 33888599, PMID 38359788, PMID 25659339, PMID 31952359, PMID 21527720, PMID 33083804, PMID 30703578, PMID 28340348, PMID 39411152


    Disclaimer: This article compiles traditional knowledge and scientific literature for comparison and research-design discussion only. It is not medical advice and cannot replace professional diagnosis or treatment. Effect sizes, frequency ranges and exposure conclusions cited here are largely inferences at the research-report level, and individual circumstances vary widely. If you have any discomfort or health concerns, please consult a physician.

  • 同一条海拔梯度,为什么不同的病毒会给出不同的答案?跨病毒复制的四道墙

    Evidence level: B

    如果宇宙线真的在改写病毒,那么同一条海拔梯度上的”解耦信号”应该在多种互不相关的病毒之间同向复现——但每种病毒的突变基线差了四个数量级,解耦度这个指标对它们量纲并不相同。

    正文

    上一篇把”海拔梯度”当作检验”天时改写病毒”的天然实验室。但一个病毒给不出结论——任何一种病毒在高海拔的表现,都可以被它自己的宿主、传播链和采样方式解释掉。真正的检验必须是跨病毒复制:同一条宇宙线通量梯度,如果多种互不相关的病毒都给出同向的解耦信号,间接通路就很难同时解释所有病毒。

    想法是对的,但做起来会立刻撞上四道墙。

    第一道墙:量纲不同。 病毒突变率的基线横跨四个数量级——RNA 病毒约 10⁻⁶ 至 10⁻⁴ 替换/核苷酸/细胞感染,DNA 病毒约 10⁻⁸ 至 10⁻⁶(PMID 20660197, 18319742)。这意味着同一个宇宙线剂量增量,若能挤进主控方程,在 RNA 病毒里是往已经泛滥的洪水里再倒一桶水,在 DNA 病毒里却可能是对原本稀薄底数的一次显著抬升。更麻烦的是,DNA 病毒自带宿主修复系统,辐射损伤的最终固定率还要再打折。所以”解耦度”的绝对数值不可跨病毒直接比较——同一条海拔梯度,量的是几把刻度完全不同的尺子。

    第二道墙:替换率不等于突变率,而且会随时间漂移。 这是病毒系统发育学里最容易被忽略的一条。Ghafari 等对 SARS-CoV-2 与 2009 大流行 H1N1 的追踪显示,两者在各自大流行第一年内的估计演化速率分别下降约 50% 与 100%,原因是末梢枝上的替换率比内部枝高 2 至 4 倍,而纯化选择让这些”年轻”的替换后来被清除(PMID 35038728)。泡沫病毒研究给出同样的形态:短窗口估计出的速率接近 RNA 病毒,长窗口才回落到它真实的长时标级别(PMID 26111824);机制模型进一步把这个”时间依赖速率现象”归结为位点饱和,给出幂律衰减斜率约 −0.65(PMID 34478645)。海拔对比如果两侧的采样窗口长度不同,估的就不是同一个东西——高海拔站点样本少、时间跨度短,天然会得到偏高的速率估计。这不是辐射效应,是窗口效应。

    第三道墙:突变还有另一个来源,就是宿主自己。 APOBEC 家族在病毒基因组上诱导 C→U 转换,ADAR1 诱导 A→I 编辑,两条通路都受宿主免疫状态调控(PMID 37031485)。高海拔意味着低氧,而低氧会下调 T 细胞功能(PMID 3307557)——也就是说,宿主编辑通路的活跃度本身就在随海拔变化。于是”高海拔病毒突变谱不同”这个观察,至少有三个候选生产者:宇宙线、宿主编辑酶、以及复制错误率本身。鉴别它们必须用突变谱指纹(C→U 偏好、上下文碱基、多碱基替换簇),而不能只看速率标量——因为速率是一个把所有来源加在一起的数。

    第四道墙:尺子本身也随海拔变。 紫外线每升高 1000 米增加约 10% 至 12%(PMID 32502327),而 UV 灭活对单链核酸病毒的效率高于双链(PMID 16254359)——这意味着高海拔对 ssRNA 病毒和 dsDNA 病毒的”存活筛选”强度不同,本身就足以在两类病毒之间制造方向相反的谱系多样性差异(PMID 33323525)。更硬的一条:全球基因组监测能力极度不均衡,高收入国家每百万人口提交的序列数是低收入国家的 456 倍,中位提交延迟 20 天对 98 天(PMID 36385137)。高海拔地区往往正是测序最稀疏的地区,“突变谱不同”很可能只是”采样稀疏”的另一种说法。

    那么跨病毒复制到底该怎么做?三步:其一,把”解耦度”改成通量归一化后的相对量,即每个病毒各自的 Δ突变指标/Δ通量,把量纲差吸收进斜率而不是让它留在截距里;其二,用突变谱指纹而非速率作结局,让辐射诱变(特征性多碱基替换、无上下文偏好)与宿主编辑(C→U 集中在 5′-UC 上下文)在谱形上分开;其三,两侧配对采样窗口,先用日期随机化检验(PMID 25771196)与逐枝时间信号检验(PMID 36936370)确认时间信号够不够,再谈比较。现有的高海拔证据恰好都卡在这三点之外:西藏腺病毒的突变模式与四川几乎一致(PMID 29240811),中国 40 城研究虽发现 S 基因 RBD 的 C22323T 与海拔正相关,但作者同时报告该位点频率与气压、UV 指数相关(PMID 38526532)——协变量并没有被排除。至于高海拔发病率的整体偏低(PMID 33780470, 35405128),那说的是传播,不是基因组。

    对易学的意义:五运六气讲”岁气”致病,但传统文本从不假设所有疫都同等感天时。《素问》既说”五疫之至,皆相染易,无问大小,病状相似”,又说”其病异形者,以其气异也”——同一个天时落在不同的疫身上,作用强度是不同的,因为每种疫有自己的”底子”。现代版的分辨点就在这里:一个病毒的突变基线有多高、它的宿主修复有多强、它的基因组是单链还是双链,决定了它”感不感天时”以及”感哪一段天时”。把”天时改写病毒”当成对一切病毒的普遍命题,是把一个条件性假设误读成了定律。

    证据与局限

    已验证: 病毒突变率横跨四个数量级,RNA 病毒 10⁻⁶~10⁻⁴ 与 DNA 病毒 10⁻⁸~10⁻⁶ 替换/核苷酸/细胞感染(PMID 20660197, 18319742);估计替换率随时间窗口长度系统性漂移,SARS-CoV-2 与 pH1N1 首年内分别降约 50% 与 100%(PMID 35038728);UV 灭活效率对单链核酸病毒高于双链(PMID 16254359);紫外线每 1000 米升高增加约 10%~12%(PMID 32502327);全球测序能力极不均衡,高收入国家每百万人口序列数达低收入国家 456 倍(PMID 36385137);APOBEC/ADAR 可直接向病毒基因组引入特征性突变(PMID 37031485)。

    部分验证: 泡沫病毒短/长窗口速率差异(PMID 26111824);幂律衰减机制模型(PMID 34478645);高海拔 COVID-19 发病率与传播力偏低(PMID 33780470, 35405128);西藏与四川腺病毒突变模式高度相似(PMID 29240811);高海拔 T 细胞功能受抑(PMID 3307557)。

    待验证: 通量归一化解耦度在不同病毒间的可加性——即”Δ突变指标/Δ通量”这一斜率能否跨病毒比较而无需先按突变基线分层;突变谱指纹(多碱基替换簇 vs C→U 上下文偏好)能否把辐射诱变与宿主编辑定量分开;配对采样窗口后高海拔解耦信号是否仍存在;低氧-宿主编辑通路的中介占比。

    相关阅读

    证据等级: B(各分项均有 PubMed 直接支撑,但”跨病毒通量归一化解耦度可比较”这一提案本身尚未有直接研究)
    参考文献: PMID 20660197, PMID 18319742, PMID 35038728, PMID 26111824, PMID 34478645, PMID 37031485, PMID 3307557, PMID 32502327, PMID 16254359, PMID 33323525, PMID 36385137, PMID 25771196, PMID 36936370, PMID 29240811, PMID 38526532, PMID 33780470, PMID 35405128


    免责说明:本文属传统知识与科学文献的整理、对照与研究设计讨论,仅供理解与参考,不构成医疗建议,不能替代专业诊断与治疗。文中涉及的效应量、频率区间与暴露结论多为研究报告层面的推断,个体情况差异很大;如有身体不适或健康疑虑,请就医并由专业医师判断。

    Why Do Different Viruses Yield Different Answers Along the Same Altitude Gradient? Four Walls of Cross-Viral Replication

    Evidence level: B

    If cosmic rays truly rewrite viruses, then the “decoupling signal” along a single altitude gradient should reproduce in the same direction across mutually unrelated viruses. Yet, each virus has a mutation baseline differing by four orders of magnitude, and the decoupling metric operates on different scales for them.

    Main Text

    The previous article treated the “altitude gradient” as a natural laboratory for testing whether “heavenly time rewrites viruses.” However, a single virus cannot yield a definitive conclusion—any observed performance of a virus at high altitude can be explained by its own host, transmission chain, and sampling methodology. The true test must be cross-virus replication: if multiple mutually unrelated viruses all yield a consistent decoupling signal along the same cosmic ray flux gradient, indirect pathways struggle to explain all of them simultaneously.

    The idea is correct, but execution immediately runs into four walls.

    Wall 1: Different scales. The baseline mutation rates of viruses span four orders of magnitude—RNA viruses range from approximately $10^{-6}$ to $10^{-4}$ substitutions/nucleotide/cell infection, while DNA viruses range from $10^{-8}$ to $10^{-6}$ (PMID 20660197, 18319742). This means that a single increment in cosmic ray dose, if incorporated into the governing equation, acts like pouring another bucket of water into an already overflowing flood for RNA viruses, whereas for DNA viruses it might represent a significant uplift against a previously thin baseline. Complicating matters further, DNA viruses possess intrinsic host repair systems, requiring another discount on the final fixation rate of radiation damage. Therefore, the absolute value of “decoupling” cannot be directly compared across viruses—the same altitude gradient is measured with rulers of vastly different scales.

    Wall 2: Substitution rate does not equal mutation rate, and it drifts over time. This is the most overlooked aspect in viral phylogenetics. Ghafari et al. tracked SARS-CoV-2 and the 2009 pandemic H1N1, showing that their estimated evolutionary rates decreased by approximately 50% and 100%, respectively, within the first year of their pandemics. This occurred because substitution rates on terminal branches were 2 to 4 times higher than those on internal branches, and purifying selection subsequently cleared these “young” substitutions (PMID 35038728). Foam virus studies reveal a similar pattern: rates estimated over short windows approach RNA virus levels, while rates over long windows revert to their true long-term scales (PMID 26111824); mechanistic models further attribute this “time-dependent rate phenomenon” to site saturation, yielding a power-law decay slope of approximately −0.65 (PMID 34478645). If the sampling window lengths differ on both sides of the altitude comparison, one is not estimating the same thing—high-altitude stations often have fewer samples and shorter time spans, naturally yielding inflated rate estimates. This is not a radiation effect; it is a window effect.

    Wall 3: Mutations have another source: the host itself. The APOBEC family induces C→U transitions on viral genomes, while ADAR1 induces A→I editing; both pathways are regulated by host immune status (PMID 37031485). High altitude implies hypoxia, which downregulates T-cell function (PMID 3307557)—meaning the activity of host editing pathways itself varies with altitude. Thus, the observation that “high-altitude viruses have different mutation spectra” has at least three candidate producers: cosmic rays, host editing enzymes, and intrinsic replication error rates. Distinguishing them requires using mutation spectrum fingerprints (C→U preference, context-dependent bases, multi-base substitution clusters), not just rate scalars—because a rate aggregates all sources.

    Wall 4: The ruler itself changes with altitude. UV radiation increases by approximately 10% to 12% for every 1000 meters of elevation (PMID 32502327), and UV inactivation is more efficient against single-stranded nucleic acid viruses than double-stranded ones (PMID 16254359)—meaning the “survival selection” intensity differs between high altitude for ssRNA viruses and dsDNA viruses, sufficient to create opposing differences in lineage diversity between the two types (PMID 33323525). A harder constraint: global genomic monitoring capabilities are extremely unbalanced; high-income countries submit 456 times more sequences per million people than low-income countries, with median submission delays of 20 days versus 98 days (PMID 36385137). High-altitude regions are often precisely the most sparsely sequenced areas; “different mutation spectra” may simply be another way of saying “sparse sampling.”

    So, how should cross-virus replication proceed? Three steps: First, convert “decoupling” into a normalized relative quantity, specifically each virus’s $\Delta$ mutation metric / $\Delta$ flux, absorbing scale differences into the slope rather than leaving them in the intercept; Second, use mutation spectrum fingerprints rather than rates as the outcome, allowing radiation-induced mutagenesis (characteristic multi-base substitutions, no context preference) and host editing (C→U concentrated in 5′-UC contexts) to separate spectrally; Third, employ paired sampling windows on both sides, first confirming sufficient temporal signals via date randomization tests (PMID 25771196) and branch-by-branch time signal tests (PMID 36936370) before proceeding to comparison. Existing high-altitude evidence falls short precisely at these three points: the adenovirus mutation pattern in Tibet is nearly identical to that in Sichuan (PMID 29240811); while the China 40-city study found a positive correlation between altitude and the C22323T mutation in the S gene RBD, authors simultaneously reported that this site’s frequency correlates with pressure and UV indices (PMID 38526532)—covariates were not excluded. Regarding the overall lower incidence rates at high altitude (PMID 33780470, 35405128), that refers to transmission, not the genome.

    Significance for Yijing Studies: Five Transmissions and Six Qi discuss “annual qi” causing disease, but traditional texts never assume all epidemics are equally affected by heavenly time. The Su Wen states both that “when five epidemics arrive, they all infect and change regardless of size, with similar symptoms” and that “those with different diseases have different forms because their qi differs”—the same heavenly time exerts different intensities on different epidemics because each epidemic has its own “foundation.” The modern distinction lies here: a virus’s mutation baseline height, host repair strength, and whether its genome is single-stranded or double-stranded determine whether it “feels” heavenly time and which segment of it. Treating “heavenly time rewrites viruses” as a universal proposition for all viruses misreads a conditional assumption as a law.

    Evidence and Limitations

    Verified: Viral mutation rates span four orders of magnitude, with RNA viruses at $10^{-6}$~$10^{-4}$ and DNA viruses at $10^{-8}$~$10^{-6}$ substitutions/nucleotide/cell infection (PMID 20660197, 18319742); estimated substitution rates drift systematically with time window length, decreasing by ~50% and ~100% respectively for SARS-CoV-2 and pH1N1 in their first years (PMID 35038728); UV inactivation efficiency is higher for single-stranded nucleic acid viruses than double-stranded ones (PMID 16254359); UV increases by ~10%~12% per 1000 meters (PMID 32502327); global sequencing capabilities are highly unbalanced, with high-income countries submitting 456 times more sequences per million people than low-income countries (PMID 36385137); APOBEC/ADAR can directly introduce characteristic mutations into viral genomes (PMID 37031485).

    Partially Verified: Rate differences between short and long windows in foam viruses (PMID 26111824); power-law decay mechanistic model (PMID 34478645); lower high-altitude COVID-19 incidence and transmissibility (PMID 33780470, 35405128); high similarity between adenovirus mutation patterns in Tibet and Sichuan (PMID 29240811); suppressed T-cell function at high altitude (PMID 3307557).

    Pending Verification: Additivity of normalized decoupling across different viruses—i.e., whether the slope “$\Delta$ mutation metric / $\Delta$ flux” can be compared across viruses without first stratifying by mutation baseline; whether mutation spectrum fingerprints (multi-base substitution clusters vs C→U context preference) can quantitatively separate radiation mutagenesis from host editing; whether high-altitude decoupling signals persist after paired sampling windows; the proportion mediated by the hypoxia-host editing pathway.

    Related Reading

    Evidence Level: B (Each component has direct PubMed support, but the proposal itself regarding “comparability of normalized decoupling across cross-virus flux” lacks direct research)
    References: PMID 20660197, PMID 18319742, PMID 35038728, PMID 26111824, PMID 34478645, PMID 37031485, PMID 3307557, PMID 32502327, PMID 16254359, PMID 33323525, PMID 36385137, PMID 25771196, PMID 36936370, PMID 29240811, PMID 38526532, PMID 33780470, PMID 35405128


    Disclaimer: This article compiles traditional knowledge and scientific literature for comparison and research-design discussion only. It is not medical advice and cannot replace professional diagnosis or treatment. Effect sizes, frequency ranges and exposure conclusions cited here are largely inferences at the research-report level, and individual circumstances vary widely. If you have any discomfort or health concerns, please consult a physician.

  • 指纹是认人的,不是认穴的:模型换个人还认不认得出

    Evidence level: B

    一个在同一个人身上能认出”这是哪个穴”的分类器,换一个人、换一台机器之后还剩多少——这个”剩多少”,才是穴位特异性的真实强度。

    正文

    上一篇给出了判别力检验方案:多穴位、参数化刺激、交叉验证、看 AUC 是否显著高于 0.5。方向对,但判据里藏着一个陷阱——AUC 不是一个数字,是三个数字的合成:同一个人隔周再来、换一个没进过训练的受试者、换一台扫描仪。三种衰减机制不同,压成一个”是否显著”,等于用平均值盖住三种失败。

    一、先把上限算出来

    90 个实验、1008 名被试的元分析显示,任务态 fMRI 平均组内相关系数(ICC)只有 0.397;在两个独立队列重测 11 种常用任务,ICC 落在 0.067–0.485。作者的结论很硬:这些指标”目前不适合用于脑生物标志物发现”(PMID 32489141)。

    这不是背景,是天花板。误差对相关的衰减是:观测相关 ≈ 真相关 × √(信度₁×信度₂)。信度取 0.4,系数就是 0.4——真相关 0.5 的信号观测到只剩 0.2,80% 功效下要近两百人。

    针刺这边有一份直接的重测数据:6 人、每人 6 次相同扫描,条件含胆经光明(GB37)、膀胱经昆仑(UB60)、非穴点与手指敲击对照。结果分两层:组平均层面,三个条件的得气评分与信号变化彼此相似;个体层面,针刺诱发的信号变化显著比手指敲击更多变(PMID 17157035)。

    两层各打一枪。第一枪打特征:单次扫描的脑图不够格当指纹。第二枪更麻烦——组平均层面三个条件都相似,那”跨穴判别”要找的组间差异,得先证明它存在。

    二、跨被试不是打折,是换编码

    直觉是跨被试差因为噪声更大。但有研究给出更难处理的答案:同一批数据上分别训练被试内与跨被试分类器,两者依赖的体素权重模式并不一致,最优脑区也不同(PMID 20347995)。衰减不是同一个信号变弱,是模型被迫去学另一套东西。

    另一项工作直接拆了”多体素模式”四个字:解码器空间上并不精确(做空间平滑,性能几乎不受影响);90% 冗余(随机抽 10% 的体素就足以恢复全部性能);性能还依赖拿来对比的条件(PMID 35219121)。跑出高 AUC 的解码器抓住的往往是大尺度网络状态,不是穴位细节——”指纹”比数据支持的要细得多。

    三、”指纹”这个方向一开始就反了

    功能连接组确实能做指纹:在几百人里认出同一个人,跨场次成功,跨任务态与静息态也能认(PMID 26457551)。结论扎实。

    但注意方向。那里被分类的是人——个体差异是信号,越大越好认;这里被分类的是穴——个体差异是噪声,必须先压掉才看得见穴位。同一个数据集上,识别人可以接近饱和,识别刺激却可能刚过随机,这不矛盾,因为信噪比结构是反的。

    所以”网络指纹”会让人预支两个还没验的结论:独一无二、终身稳定。

    四、跨中心那一层最容易被误读

    5 名受试者在 10 台扫描仪上各扫两次:重测信度不错,但跨中心信度一开始很低,站点与”站点×被试”方差贡献显著(PMID 17636563)。验证集若与发现集同中心、同机器、同术者,泛化成功只证明同中心可重复。反过来更危险:站点间的系统偏移会被读成”这个穴在不同人群里效应不同”,仪器效应就此翻译成生理学发现。

    五、HRV 那根柱子,长度不够

    63 人重复测量:6 分钟与 2 分钟时域指标 ICC 大于 0.7,2 分钟频域大于 0.5,10 秒更低(PMID 15286929)。更细的标定是:心率 60 秒、时域 240 秒、相对频域 300 秒(PMID 33856258)。针刺刺激段常只有几十秒到两分钟,而低频成分需至少两分钟——窗口不够,算出来的已不是定义里的那个量。

    六、五条可执行判据

    一,别只报 AUC,报穴-穴距离矩阵的重测相关:同一人两次扫描各算 K 个穴位两两距离矩阵,再求两矩阵相关。AUC 会被类别数与不平衡掩盖,距离矩阵不会。

    二,加阳性对照:同批数据跑”识别个体”。若识别人接近饱和而识穴接近随机,那点信号属于人,不属于穴。

    三,验证集必须换中心,退一步至少换扫描仪或换术者,否则”跨被试泛化”实际在测重复性。

    四,样本量按信度算:给定 ICC 与可接受的 AUC 区间宽度反推 N。小样本功效问题在神经科学里被反复证明(PMID 23571845),交叉验证误差条也会大到无法支持结论(PMID 28655633)。

    五,预注册等效性判据:若观测 AUC 的置信区间同时覆盖 0.5 与信度允许的上限,结论只能写”无法区分有无特异性”。

    七、粗守形,上守神

    《灵枢·九针十二原》:“粗守形,上守神。” 判别模型守的是”形”——体素模式的空间形状;针刺真正的差异可能在”神”——得气出现与消退的时序。形没抓住,不等于神不存在。

    《素问·异法方宜论》早把个体差异放进治法:同一类病,东方砭石、西方毒药、北方灸焫、南方九针、中央导引按跷,“一病而治各不同,皆愈”。古人从不假设一套参数通用。

    《灵枢·官能》更直说:“语徐而安静,手巧而心审谛者,可使行针艾。” 行针交给特定的一类人——这是公开承认术者方差。现代把它塞进”随机效应”,古人写成择人标准。差别在于:写出来,就有人去管它。

    证据与局限

    已验证: ①任务态 fMRI 的重测信度整体偏低(90 个实验元分析,平均 ICC 0.397;两个独立队列 0.067–0.485,PMID 32489141);②针刺诱发的 fMRI 信号变化在同一受试者不同场次间的变异显著大于主动对照任务,且组平均层面穴位与非穴可相近(PMID 17157035);③被试内与被试间分类器依赖不同的信息与脑区(PMID 20347995);④多体素解码器高度冗余、空间不精确、依赖对比条件(PMID 35219121);⑤功能连接可稳定识别个体,跨场次、跨状态成立(PMID 26457551);⑥跨中心 fMRI 信度显著受站点与站点×被试方差影响(PMID 17636563);⑦短时 HRV 的重测信度随记录时长下降,频域指标门槛最高(PMID 15286929, PMID 33856258)。

    部分验证: “距离矩阵重测相关优于 AUC”目前是方法学推论,尚未在穴位数据上做过正面对照;跨被试衰减中”噪声”与”编码差异”各占多少,缺少可分解的实验。

    待验证: ①同一批多穴位数据上,”识别人”与”识别穴”的准确率落差有多大;②在 ICC 已知的条件下,多穴位判别 AUC 的实际可检出下限;③换中心后穴位判别的 AUC 衰减幅度是否超过仪器效应本身;④距离矩阵重测相关与临床疗效是否存在关联。

    相关阅读

    证据等级: B
    参考文献: PMID 32489141, PMID 17157035, PMID 20347995, PMID 35219121, PMID 26457551, PMID 17636563, PMID 15286929, PMID 33856258, PMID 23571845, PMID 28655633


    免责说明:本文属传统知识与科学文献的整理、对照与研究设计讨论,仅供理解与参考,不构成医疗建议,不能替代专业诊断与治疗。文中涉及的效应量、频率区间与暴露结论多为研究报告层面的推断,个体情况差异很大;如有身体不适或健康疑虑,请就医并由专业医师判断。

    Fingerprints Identify People, Not Body Holes: Will the Model Still Recognize Them After Changing Individuals?

    Evidence level: B

    A classifier that can identify “which acupoint” on the same individual—how much of this capability remains when switching to a different person or machine? This “remainder” is the true strength of acupoint specificity.

    Main Text

    The previous article provided a discrimination validation scheme: multi-acupoint, parametric stimulation, cross-validation, and checking if the AUC is significantly above 0.5. The direction is correct, but the criterion hides a trap—AUC is not a single number; it is a synthesis of three numbers: re-testing the same person after a week, using a new untrained subject, or switching scanners. These three attenuation mechanisms differ; compressing them into a single “significance” verdict is like covering three types of failure with an average.

    I. First, Calculate the Ceiling

    A meta-analysis of 90 experiments with 1,008 participants showed that the average intraclass correlation coefficient (ICC) for task-state fMRI is only 0.397; in two independent cohorts re-testing 11 common tasks, ICC ranged from 0.067–0.485. The authors’ conclusion is firm: these metrics “are currently not suitable for discovering brain biomarkers” (PMID 32489141).

    This is not background noise; it is a ceiling. The attenuation of correlation due to error is: Observed Correlation ≈ True Correlation × √(Reliability₁ × Reliability₂). With reliability set at 0.4, the coefficient is 0.4—a true correlation of 0.5 yields an observed value of only 0.2, requiring nearly two hundred participants for 80% power.

    Regarding acupuncture, there is direct re-test data: 6 participants, each scanned 6 times under identical conditions, including GB37 (Lianggu), UB60 (Kunlun), non-acupoint points, and finger tapping controls. Results showed two layers: At the group average level, the qi sensation scores and signal changes across the three conditions were similar; At the individual level, signal changes induced by acupuncture varied significantly more than those from finger tapping (PMID 17157035).

    Two shots fired at different targets. The first shot targets features: a single scan’s brain map is insufficient to serve as a fingerprint. The second shot is more troublesome—at the group average level, all three conditions are similar; therefore, the inter-group difference sought for “cross-acupoint discrimination” must first be proven to exist.

    II. Cross-Subject Is Not Discounting, It Is Changing Encoding

    Intuition suggests cross-subject performance drops due to higher noise. However, research offers a harder-to-process answer: when trained on the same dataset, both within-subject and between-subject classifiers rely on inconsistent voxel weight patterns, and the optimal brain regions differ (PMID 20347995). The attenuation is not the same signal weakening; it is the model forced to learn a different set of rules.

    Another study directly deconstructed the phrase “multi-voxel pattern”: the decoder space is not precise (spatial smoothing has almost no impact on performance); 90% redundancy (randomly sampling 10% of voxels is sufficient to restore full performance); and performance depends on the conditions used for comparison (PMID 35219121). High-AUC decoders often capture large-scale network states rather than acupoint details—”fingerprints” are far finer than the data supports.

    III. The “Fingerprint” Direction Was Wrong from the Start

    Functional connectomes can indeed serve as fingerprints: identifying the same individual across hundreds of people, across sessions, and across task-state versus rest states is successful (PMID 26457551). The conclusion is solid.

    But note the direction. There, the classifier targets people—individual differences are the signal; larger is better recognized. Here, the classifier targets acupoints—individual differences are noise that must be suppressed before acupoint features become visible. On the same dataset, recognizing people can approach saturation, while recognizing stimuli may barely exceed random chance; this is not contradictory because the signal-to-noise ratio structure is inverted.

    Therefore, “network fingerprints” lead to two premature conclusions: uniqueness and lifelong stability.

    IV. The Cross-Center Layer Is Most Easily Misread

    Five participants scanned twice on ten different scanners: re-test reliability was good, but cross-center reliability was initially low, with significant variance contributions from site and site × subject (PMID 17636563). If the validation set matches the discovery set in center, machine, and operator, successful generalization only proves repeatability within the same center. The reverse is more dangerous: systematic shifts between sites may be misread as “this acupoint has different effects across different populations,” translating instrument effects into physiological findings.

    V. The HRV Pillar Is Too Short

    In a 63-person repeated-measurement study: ICC for time-domain metrics exceeded 0.7 at 6 minutes and 2 minutes, while frequency-domain metrics exceeded 0.5 only at 2 minutes (lower at 10 seconds) (PMID 15286929). Finer calibration shows: heart rate requires 60 seconds, time-domain metrics require 240 seconds, and relative frequency domains require 300 seconds (PMID 33856258). Acupuncture stimulation segments often last only tens of seconds to two minutes, while low-frequency components require at least two minutes—the window is insufficient; what is calculated is not the quantity defined in the theory.

    VI. Five Executable Criteria

    1. Do not report only AUC; report the re-test correlation of the inter-acupoint distance matrix: calculate the pairwise distance matrix for K acupoints for two scans of the same person twice, then compute the correlation between the two matrices. AUC is masked by category count and imbalance; distance matrices are not.
    2. Add a positive control: run “individual recognition” on the same batch of data. If recognizing people approaches saturation but recognizing acupoints approaches random chance, that signal belongs to the person, not the acupoint.
    3. The validation set must change centers; as a minimum step, switch scanners or operators, otherwise “cross-subject generalization” actually measures repeatability.
    4. Calculate sample size based on reliability: reverse-calculate N given the acceptable AUC interval width and known ICC. Small sample power issues are repeatedly proven in neuroscience (PMID 23571845), and cross-validation error bars can be too large to support conclusions (PMID 28655633).
    5. Pre-register equivalence criteria: if the confidence interval of the observed AUC simultaneously covers 0.5 and the reliability-allowed upper limit, the conclusion can only be written as “unable to distinguish presence or absence of specificity.”

    VII. Roughly Guard the Form, Above Guard the Spirit

    The Classic of Internal Medicine · Nine Needles and Twelve Origins: “Roughly guard the form, above guard the spirit.” Discrimination models guard the “form”—the spatial shape of voxel patterns; the true difference in acupuncture may lie in the “spirit”—the temporal sequence of qi sensation appearing and subsiding. If the form is not grasped, it does not mean the spirit does not exist.

    The Classic of Yellow Emperor · Treatments According to Different Regions placed individual differences into treatment methods early on: for the same type of disease, moxa stones in the East, toxic drugs in the West, moxibustion in the North, nine needles in the South, and guiding pressing in the Center, “one disease treated differently yet all recover.” The ancients never assumed a single set of parameters is universal.

    The Classic of Internal Medicine · Functions and Abilities states even more directly: “Those who speak slowly and calmly, with skilled hands and careful hearts, may be entrusted with needling and moxa.” Needling is entrusted to a specific class of people—this is an openly admitted operator variance. Modern practice hides this in “random effects”; the ancients wrote it as a selection standard. The difference lies in: once written down, someone manages it.

    Evidence and Limitations

    Verified: ① Re-test reliability in task-state fMRI is generally low (meta-analysis of 90 experiments, average ICC 0.397; two independent cohorts 0.067–0.485, PMID 32489141); ② fMRI signal changes induced by acupuncture vary significantly more between sessions in the same subject than in active control tasks, and at the group average level, acupoints and non-acupoints can be similar (PMID 17157035); ③ Within-subject and between-subject classifiers rely on different information and brain regions (PMID 20347995); ④ Multi-voxel decoders are highly redundant, spatially imprecise, and dependent on comparison conditions (PMID 35219121); ⑤ Functional connectivity can stably identify individuals across sessions and states (PMID 26457551); ⑥ Cross-center fMRI reliability is significantly affected by site and site × subject variance (PMID 17636563); ⑦ Re-test reliability of short-term HRV decreases with recording duration, with frequency-domain metrics having the highest threshold (PMID 15286929, PMID 33856258).

    Partially Verified: “Distance matrix re-test correlation is superior to AUC” is currently a methodological inference; no direct comparison has been made on acupoint data yet. The proportion of “noise” versus “encoding differences” in cross-subject attenuation lacks a decomposable experiment.

    To Be Verified: ① On the same batch of multi-acupoint data, what is the accuracy gap between “recognizing people” and “recognizing acupoints”; ② Under known ICC conditions, what is the actual detectable lower limit for multi-acupoint discrimination AUC; ③ Whether the AUC attenuation after changing centers exceeds the instrument effect itself; ④ Whether distance matrix re-test correlation correlates with clinical efficacy.

    Related Reading

    Evidence Level: B
    References: PMID 32489141, PMID 17157035, PMID 20347995, PMID 35219121, PMID 26457551, PMID 17636563, PMID 15286929, PMID 33856258, PMID 23571845, PMID 28655633


    Disclaimer: This article compiles traditional knowledge and scientific literature for comparison and research-design discussion only. It is not medical advice and cannot replace professional diagnosis or treatment. Effect sizes, frequency ranges and exposure conclusions cited here are largely inferences at the research-report level, and individual circumstances vary widely. If you have any discomfort or health concerns, please consult a physician.

  • 融合模型赢的那一局,是赢在信息还是赢在挑选

    Evidence level: B

    当”融合模型”和两个单模态模型在同一份数据上一起训练、一起评估、再挑最好的那个报出来,它赢的多半不是信息,而是挑选。

    正文

    一、”双模态更好”这个结论几乎是必然的

    先做一个不需要生物学的推演。假设 HRV 对辨别穴位完全无用——它携带的是纯噪声。你手上仍有三个模型:fMRI 单用、HRV 单用、两者拼接。在同一份交叉验证里,三者 AUC 都是随机变量,围绕真值各自波动,而你报的是最大值。即使三者真值完全相同,最大值也必然高于平均——这是次序统计量的性质,与”有没有增益”无关。

    Varma 与 Simon 说得很清楚:用交叉验证误差挑选模型、再把这个误差当作该模型的性能报告,估计是有偏的(PMID 16504092),方向稳定偏乐观。所以”双模态 AUC 高于单模态”这一观察本身不构成证据。它需要一个正确的零假设:不是 AUC=0.5,而是”三个模型在同一份数据上取最大值”会得到什么。

    麻烦在多分类里会被放大。五个穴位、理论随机水平 20%,但 Combrisson 与 Jerbi 指出,小样本下真正的偶然水平不是 1/k,会随样本量与类别分布一起上移(PMID 25596422)。你以为的”显著高于随机”,可能连基准都设错了。

    二、两个模态不是两桶可以相加的水

    融合的直觉是”信息互补”。互补需要两个模态是同一目标的不同侧面,且彼此不重复。

    HRV 与 BOLD 恰恰不满足后者。Chang 等人的分析显示,心率变异性与静息态功能连接的波动本身相关(PMID 23246859)——自主神经的慢波动会同时写进心跳间隔和脑血管信号。两条通道共享一部分方差,融合时被计入两次。Sui 等人的综述因此强调,多模态融合的方法学核心不是把特征拼起来,而是建模模态之间的协方差结构(PMID 22108139)。拼接是最粗暴的做法,其”增益”高度依赖特征尺度——fMRI 十万维、HRV 个位数维,不加权时融合约等于”只用 fMRI”,加权就多出一个待调参数,而这个参数恰好会往”让结果更好看”的方向调。

    更关键的是共同原因。若某些穴位刺进去更疼、更得气,被试整体唤醒更高,融合就只是把”谁更紧张”测了两遍。AUC 会上升,特异性不会。这时”双模态更优”翻译过来是:我们更擅长识别被试的唤醒状态。

    三、第二根支柱的地基比第一根软

    源文把 HRV 当第二根支柱,主要指标是低频/高频比。但 Billman 的批评很直接:LF/HF 并不准确测量交感-迷走平衡(PMID 23431279)——它受呼吸频率、总功率、体位影响,把双变量问题压成一个比值。

    时间尺度的问题更具体。HRV 国际标准把”短时”定为 5 分钟,低频成分需至少 2 分钟才可解释(PMID 8737210)。而任务态 fMRI 一个刺激段常只有几十秒到两三分钟——从这段记录算出的频域指标,已经不是标准定义的那个东西。2026 年一项验证研究指出,超短时 HRV 的效度必须按状态分别建立,状态一变一致性就掉(PMID 42243172)。针刺恰恰是”状态在变”的场景。

    一根地基较软的支柱接进融合模型,期望增益不会大于零——它更可能贡献噪声,而噪声恰好能通过选择偏倚变成”提升”。

    四、真正该做的检验,是把”挑选”也当成一个模型

    判据不是”融合 > 单模态”,而是”融合的增益 > 挑选本身带来的增益”。五条可执行:

    ① 预注册三模型。把融合定为主分析,另两个只作敏感性分析——不让数据决定谁上场。

    ② 嵌套交叉验证。外层只做评估,内层做特征选择与融合权重。Kriegeskorte 的”双重 dipping”警告在此尤其致命——融合给了你更多”看着结果挑”的自由度(PMID 19396166)。

    ③ 置换检验重跑全流程。打乱标签后,特征选择、加权、模型选择全部重做,得到”max-of-three”的零分布;融合增益须超过该零分布 95 分位才算成立。只重训一次的置换检验,测的是另一个问题。

    ④ ΔAUC 用配对比较。DeLong 方法适用于同一批受试者上的相关 ROC 曲线(PMID 3203132),但它假定模型已固定;模型是训练出来时方差被低估,须改用嵌套交叉验证的配对折差做 bootstrap。

    ⑤ 先算功效再看结果。小样本下交叉验证的误差条宽到足以吞掉模型间差异(PMID 28655633);脑-行为关联要稳定复现,样本需以千计(PMID 35296861)。以 40-60 人、3-5 个穴位的规模,可检出的 ΔAUC 下限远高于融合的典型增益。最可能的结局是:看得到提升,检不出差异。

    五、易学那一层:参合是校验,不是拼接

    《灵枢·邪气脏腑病形》给三诊各一个字:”见其色,知其病,命曰明;按其脉,知其病,命曰神;问其病,知其处,命曰工。”紧接着一句:”能参合而行之者,可以为上工。”

    这段话常被引来证明”古人早就主张多模态融合”,但它给的是”上工”这个身份判断,不是”参合准确率更高”这个数字。
    更重要的是古人”参合”的操作方式:校验。色与脉一致则确信;相反则再问、再按,甚至判断”舍脉从症”还是”舍症从脉”。矛盾时,结论是降信。

    特征拼接不是这样运作的。矛盾时它给出加权平均——一个既不是 A 也不是 B 的中间值,还往往带着比单模态更高的置信度。这才是最危险的地方:它不区分”互相印证”和”互相掩盖”。

    所以要问的不是”两通道加起来准不准”,而是”两通道打架时,模型会不会告诉你它在打架”。

    证据与局限

    已验证: ①交叉验证选模型后直接报告该误差会产生乐观偏倚(PMID 16504092);②小样本多分类的经验偶然水平显著高于理论 1/k(PMID 25596422);③先选特征再做交叉验证构成循环分析(PMID 19396166);④HRV 与 BOLD 波动共享方差,两模态并非独立信息源(PMID 23246859);⑤LF/HF 不能准确度量交感-迷走平衡(PMID 23431279);⑥HRV 频域指标的短时定义有明确时长门槛(PMID 8737210)。

    部分验证: 多模态融合在精神疾病分类中确有增益报告(PMID 23108278),但增益幅度普遍温和、且多为单队列结果;融合方法学综述强调需建模模态协方差而非简单拼接(PMID 22108139);小样本下机器学习验证需嵌套设计(PMID 31697686);针灸 fMRI 的 MVPA 判别研究已证明”可解码”,但均未做单/双模态增量对比(PMID 20071124)。

    待验证: ①穴位任务态 fMRI+HRV 的双模态增益是否超过 max-of-three 零分布;②融合增益在控制得气强度、疼痛评分、平均心率后是否仍存在;③HRV 频域指标在 1-3 分钟刺激段内的重测信度是否足以支撑其作为独立判别通道;④”参合式校验”(矛盾即降置信)与”拼接式融合”在同一批数据上的判别力与校准差异。

    相关阅读

    证据等级: B
    参考文献: PMID 16504092, PMID 25596422, PMID 19396166, PMID 23246859, PMID 22108139, PMID 23108278, PMID 23431279, PMID 8737210, PMID 42243172, PMID 3203132, PMID 28655633, PMID 35296861, PMID 31697686, PMID 20071124


    免责说明:本文属传统知识与科学文献的整理、对照与研究设计讨论,仅供理解与参考,不构成医疗建议,不能替代专业诊断与治疗。文中涉及的效应量、频率区间与暴露结论多为研究报告层面的推断,个体情况差异很大;如有身体不适或健康疑虑,请就医并由专业医师判断。

    The Winning Move of the Integrated Model: Information or Selection?

    Evidence level: B

    When a “fusion model” is trained, evaluated, and selected alongside two unimodal models on the same dataset, its victory is likely not due to information gain, but rather to selection bias.

    Main Text

    I. The Conclusion That “Multimodal is Better” Is Almost Inevitable

    First, let us perform a deduction that requires no biological assumptions. Assume HRV is entirely useless for identifying acupoints—it carries only pure noise. You still have three models: fMRI alone, HRV alone, and their concatenation. In the same cross-validation split, their AUCs are random variables fluctuating around their true values, and you report the maximum. Even if all three true values are identical, the maximum must necessarily exceed the average—this is a property of order statistics, unrelated to whether there is actual gain.

    Varma and Simon state clearly: selecting models via cross-validation error and then reporting that error as the model’s performance is biased (PMID 16504092), with a stable direction of overestimation. Therefore, the observation that “multimodal AUC exceeds unimodal” does not constitute evidence on its own. It requires a correct null hypothesis: what would the maximum yield if three models were selected from the same dataset?

    The issue is amplified in multi-class settings. With five acupoints and a theoretical random baseline of 20%, Combrisson and Jerbi point out that the true chance level under small samples is not 1/k but shifts upward depending on sample size and class distribution (PMID 25596422). What you perceive as “significantly above random” may simply be based on a misconfigured baseline.

    II. Two Modalities Are Not Two Buckets of Water That Can Be Added

    The intuition for fusion is “information complementarity.” Complementarity requires two modalities to represent different aspects of the same target without redundancy.

    HRV and BOLD fail this latter condition. Chang et al.’s analysis shows that heart rate variability and resting-state functional connectivity fluctuations are inherently correlated (PMID 23246859)—slow fluctuations in the autonomic nervous system are simultaneously encoded in heart intervals and cerebrovascular signals. Two channels share variance; when fused, this is counted twice. Consequently, Sui et al.’s review emphasizes that the methodological core of multimodal fusion is not concatenating features, but modeling the covariance structure between modalities (PMID 22108139). Concatenation is the most crude approach; its “gain” heavily depends on feature scaling—fMRI has tens of thousands of dimensions, while HRV has only a few. Without weighting, fusion approximates “using only fMRI”; adding weights introduces an extra hyperparameter that tends to be tuned toward making the results look better.

    More critically, there are common causes. If certain acupoints cause more pain or a stronger sensation of “qi” upon insertion, leading to higher overall arousal in subjects, fusion merely measures “who is more tense” twice. AUC rises, but specificity does not. In this case, “multimodal superiority” translates to: we are better at identifying the subject’s arousal state.

    III. The Foundation of the Second Pillar Is Softer Than the First

    The source text treats HRV as the second pillar, primarily using the Low-Frequency/High-Frequency ratio. However, Billman’s critique is direct: LF/HF does not accurately measure sympathetic-vagal balance (PMID 23431279)—it is influenced by respiratory rate, total power, and posture, compressing a bivariate problem into a ratio.

    The issue of time scale is more specific. The HRV international standard defines “short-term” as 5 minutes, requiring at least 2 minutes to interpret low-frequency components (PMID 8737210). In contrast, task-state fMRI stimulus segments often last only tens of seconds to a few minutes. Frequency domain metrics derived from such short recordings are no longer the standard-defined quantities. A 2026 validation study indicates that the validity of ultra-short HRV must be established separately for each state, as consistency drops when the state changes (PMID 42243172). Acupuncture is precisely a scenario where the state is changing.

    A pillar with a softer foundation enters the fusion model; expected gain will not exceed zero—it is more likely to contribute noise, and noise can恰好 (coincidentally) become an “improvement” through selection bias.

    IV. The Real Test Should Treat “Selection” as a Model

    The criterion is not “fusion > unimodal,” but rather “fusion gain > gain from selection itself.” Five actionable steps:

    ① Pre-register three models. Define fusion as the primary analysis and the other two only for sensitivity analysis—do not let the data decide who enters.

    ② Nested cross-validation. The outer loop performs evaluation only; the inner loop handles feature selection and fusion weighting. Kriegeskorte’s warning against “double dipping” is particularly fatal here—fusion grants you more freedom to “look at results and select” (PMID 19396166).

    ③ Permutation test re-running the full pipeline. Shuffle labels and re-run feature selection, weighting, and model selection to obtain the zero distribution for “max-of-three”; fusion gain must exceed the 95th percentile of this zero distribution to be considered valid. A single retraining permutation test measures a different question.

    ④ Use paired comparison for ΔAUC. The DeLong method applies to correlated ROC curves on the same subjects (PMID 3203132), but it assumes fixed models; if model variance is underestimated during training, use nested cross-validation paired differences with bootstrap.

    ⑤ Calculate power before results. Under small samples, the width of cross-validation error bars is sufficient to swallow inter-model differences (PMID 28655633); brain-behavior associations require thousands of samples for stable replication (PMID 35296861). With a scale of 40-60 subjects and 3-5 acupoints, the detectable lower limit of ΔAUC far exceeds typical fusion gains. The most likely outcome is: visible improvement, but no detectable difference.

    V. The Yijing Layer: Parametric Combination is Verification, Not Concatenation

    The Ling Shu chapter “Pathogenic Qi and Disease Forms of Organs” assigns one character to each of the three diagnostics: “See its color, know its disease; this is called ‘Ming’ (Clear); Press its pulse, know its disease; this is called ‘Shen’ (Spirit); Ask about the disease, know its location; this is called ‘Gong’ (Work).” Immediately following is: “Those who can combine and practice these are called ‘Shang Gong’ (Superior Practitioner).”

    This passage is often cited to prove “ancients already advocated multimodal fusion,” but it provides a judgment of the identity “Shang Gong,” not a numerical claim that “combination accuracy is higher.”

    More importantly, the ancient method of “parametric combination” (参合) operates as verification: if color and pulse align, confidence is confirmed; if they contradict, one asks again, presses again, or judges whether to “discard pulse for symptom” or “discard symptom for pulse.” In contradiction, the conclusion is reduced credibility.

    Feature concatenation does not operate this way. In contradiction, it provides a weighted average—a midpoint that is neither A nor B—and often carries higher confidence than unimodal results. This is the most dangerous aspect: it fails to distinguish between “mutual corroboration” and “mutual masking.”

    Therefore, the question should not be “how accurate are two channels added together,” but rather “when the two channels conflict, will the model tell you it is in conflict.”

    Evidence and Limitations

    Verified: ① Reporting cross-validation error directly after model selection produces optimistic bias (PMID 16504092); ② Empirical chance levels in small-sample multiclass classification are significantly higher than the theoretical 1/k (PMID 25596422); ③ Selecting features before cross-validation constitutes circular analysis (PMID 19396166); ④ HRV and BOLD fluctuations share variance; the two modalities are not independent information sources (PMID 23246859); ⑤ LF/HF cannot accurately measure sympathetic-vagal balance (PMID 23431279); ⑥ Short-term definitions for HRV frequency domain metrics have explicit duration thresholds (PMID 8737210).

    Partially Verified: Multimodal fusion reports gains in psychiatric classification (PMID 23108278), but gain magnitudes are generally modest and mostly single-cohort results; fusion methodology reviews emphasize modeling modal covariance rather than simple concatenation (PMID 22108139); machine learning validation in small samples requires nested design (PMID 31697686); acupuncture fMRI MVPA decoding studies have proven “decodability,” but none performed unimodal/bimodal incremental comparisons (PMID 20071124).

    To Be Verified: ① Whether bimodal gains in acupoint task-state fMRI+HRV exceed the max-of-three zero distribution; ② Whether fusion gains persist after controlling for sensation intensity, pain scores, and average heart rate; ③ Whether retest reliability of HRV frequency metrics within 1-3 minute stimulus segments is sufficient to support them as independent discriminative channels; ④ The difference in discriminative power and calibration between “parametric verification” (reduced confidence on contradiction) and “concatenative fusion” on the same dataset.

    Related Reading

    Evidence Level: B
    References: PMID 16504092, PMID 25596422, PMID 19396166, PMID 23246859, PMID 22108139, PMID 23108278, PMID 23431279, PMID 8737210, PMID 42243172, PMID 3203132, PMID 28655633, PMID 35296861, PMID 31697686, PMID 20071124


    Disclaimer: This article compiles traditional knowledge and scientific literature for comparison and research-design discussion only. It is not medical advice and cannot replace professional diagnosis or treatment. Effect sizes, frequency ranges and exposure conclusions cited here are largely inferences at the research-report level, and individual circumstances vary widely. If you have any discomfort or health concerns, please consult a physician.

  • 「生物钟相位紊乱 × 地磁活跃期」交互项对呼吸道感染结局的直接检验:两个时钟相乘,乘的是哪两个量?

    Evidence level: B

    地磁要落到人身上,必须先有一条”接收天线”;昼夜节律系统是唯一被反复提出的那条。但”相位紊乱”和”地磁活跃”是两个不同原点的时间——相乘之前,得先说清乘的是什么。

    正文

    一、地磁要有天线,才能落到人身上

    最扎实的一层证据来自死亡结局。Zilli Vieira 等(PMID 31511079)用两步 meta 分析处理 263 个美国城市:Kp 每升高一个标准差,总死亡冬季升 0.13%、秋季升 0.31%,调整 PM2.5 后冬季升到 0.47%,心梗死亡同向显著,中风则全无显著关联。这种”只打一半”本身就是线索:地磁若只经气象间接作用,不该这么挑结局。它是地球尺度的低频场,凭什么在细胞里落笔?必须有接收器——先找到那根天线。

    二、唯一被反复提出的天线:隐花色素

    Close(PMID 22418257)梳理得最清楚:隐花色素(CRY)有两个身份——生物钟齿轮的核心组件,以及动物磁罗盘的光依赖分子基础。他的机制修正很关键:地磁暴不是当”第二个授时因子”去拨褪黑素的表,而是经隐花色素通路更广泛地扰动 HPA 轴与昼夜行为。

    褪黑素支线有历史数据。Weydahl 等(PMID 11774869)在北纬 70° 追踪 25 人,发现地磁变化必须超过某个量级(>80 nT/3h)才显著压低唾液褪黑素;Burch 等(PMID 18472329)在 153 名电力工人中看到,预测 6-OHMS 排泄量最大差异的时间窗落在采样前 15 至 33 小时。这两个特征决定修饰变量怎么设:应是”过去一到两天这套钟稳不稳”,而非”今天乱没乱”。

    三、天线那一端:白细胞降,黏附分子升

    哈佛 NAS 队列把免疫端补上了。Tracy 等(PMID 34537201)用 0 至 28 天移动平均建模,发现 IMF、黑子数、Kp 与总白细胞、中性粒细胞、嗜碱性粒细胞负相关,且移动平均越长关联越强;Schiff 等(PMID 35881632)在同一队列看到 sICAM-1、sVCAM-1 与 IMF/SSN/Kp 正相关。白细胞减少而黏附分子升高——既不是”免疫被抑制”也不是”被激活”,更像细胞的分布与动员格局变了:血里少了,可能贴在血管壁上,也可能进了组织。

    四、天线这一端:相位紊乱与感染的信号有多窄

    人体层面的证据集中在职业队列,窄得出乎意料。Loef 等(PMID 30475977)在 Klokwerk+ 收 501 名轮班/夜班医护与 88 名白班医护,累计 110,347 份每日症状日记:轮班者流感样疾病/急性呼吸道感染发生率是非轮班者的 1.20 倍(95%CI 1.01-1.43),重症 1.22 倍,但病程时长无差异(1.02)。同队列中介分析(PMID 32255192):23% 的超额降到 15% 且不再显著,最大中介是主观睡眠质量(间接效应 1.29),客观睡眠时长、体力活动、进餐次数均非显著中介。

    Bjorvatn 等(PMID 38221779)在挪威 1023 人里只看到流感样疾病显著(OR 1.97),其余十余种感染均不显著;同组(PMID 40059106)在 1335 名护士中发现睡眠债呈剂量依赖——肺炎/支气管炎 aOR 从 2.29 升到 3.88,夜班本身只有 1.28。关联真实但很窄,主要走”主观睡眠质量”,不是”睡得少”。

    五、核心难题:乘的是两个不同原点的时间

    相位紊乱是相对量:内部时钟与外部 24 小时的不对齐度;地磁活跃是绝对量:外部场本身的活动强度。前者读”人的钟在哪一格”,后者读”天的钟敲多响”。两个坐标系的量相乘,乘积的物理含义必须先定义清楚,否则系数出来了也解释不了。

    而”相位”这把尺子精度有限。DLMO 是金标准,Kennaway(PMID 36799668)汇总 152 项研究、4397 人建立参考范围;但 Lewy(PMID 10452335)指出低分泌者可能永远不越过常用阈值,相位估计还会被振幅混淆;Burgess 等(PMID 22445311)发现居家与实验室测量平均差 37 至 54 分钟。替代方案已成立:Laing 等(PMID 28218891)用全血转录组预测褪黑素相位,单样本 R²=0.74、间隔 12 小时的两个样本 R²=0.90,平均误差约 9 分钟——“相位可测”站住了。

    六、把”相位”换成连续量:相位角

    比 DLMO 更实用的修饰变量是相位角(phase angle)——内部相位与行为时间之差,连续可测,直接刻画”对齐程度”而非”早或晚”。

    Shafer 等(PMID 41213163)在 42 名健康成人中做实验室 DLMO 并连续 14 天记录入睡时间,按相位角中位数分组:窄相位角女性 CRP 更高,男性 TNF-α 更高。同一个”不对齐”,男女走两条炎症通路——交互项若成立,很可能也带性别分层。

    七、真要做这个检验,该怎么设计

    第一,修饰变量用相位角,不用序数化的”早起型/晚睡型”。 序数偏好把”晚”和”乱”混在一起,而交互项要的是”乱”。

    第二,暴露窗口按 15-33 小时与长移动平均来设,不用当日 Kp。

    第三,把地磁拆成频段。 27 天与 11 年太阳周期在频域上与 24 小时昼夜相位不重叠——这是本检验能做、上一轮”25(OH)D × 太阳周期”不能做的唯一理由:25(OH)D 是 UVB 下游产物(同轴共线),相位角是社会时间的产物,与地磁无共享上游。

    第四,预设阴性解释。 若交互项只在”自报睡眠质量差”分层下出现、在客观相位角分层下消失,说明主观睡眠质量才是真解释变量。

    第五,干预可行性是这张设计表的优势。 Paul 等(PMID 41243965)证明小鼠光周期打乱会取消昼夜保护,而定时进食可以补偿。相位可被光照、进食、褪黑素直接移动,交互项若成立就能接着做真干预试验。

    八、易学落点:两套钟,不是一套

    「天人相应」翻成科学语言要拆成两半:“天”是外部周期(地磁、日照),”人”是内部相位(昼夜节律、营卫运行)。 传统常把两者当成同一个钟,实际是两套:一套对表(相位对齐),一套读表(外部活动强度)。 子午流注讲”人的钟走到哪一格”,地磁讲”天这口钟敲得多响”。混成一个变量,交互项就成了不可解释的乘积——关键不在”哪个修饰变量更固定”,而在两个变量是否在同一坐标系里。

    证据与局限

    已验证:地磁与总死亡、心血管死亡、心梗死亡的关联在 263 城 meta 分析中显著(PMID 31511079);地磁与循环白细胞、黏附分子、CRP 的关联在 NAS 队列中跨多种暴露度量一致(PMID 34537201、35881632);轮班/夜班与流感样疾病发生率升高在两个独立队列中方向一致(PMID 30475977、38221779);血液转录组可在约 9 分钟误差内预测褪黑素相位(PMID 28218891)。

    部分验证:地磁→褪黑素通路有历史数据(PMID 11774869、18472329),但样本小、年代早、未独立复现;隐花色素作为人体磁感受器仍属假说(PMID 22418257);相位角与炎症标志物的关联仅 42 人(PMID 41213163),性别分层结论待复制。

    待验证:①”相位紊乱 × 地磁活跃”的交互项至今没有任何直接检验;②地磁是经 HPA 轴还是经褪黑素通路起作用;③免疫结局该设在感染发生率、重症率,还是分子层(白细胞分布/黏附分子)。

    证据等级: B
    参考文献: PMID 31511079, PMID 22418257, PMID 11774869, PMID 18472329, PMID 34537201, PMID 35881632, PMID 30475977, PMID 32255192, PMID 38221779, PMID 40059106, PMID 36799668, PMID 10452335, PMID 22445311, PMID 28218891, PMID 41213163, PMID 41243965, PMID 36590897, PMID 15066682, PMID 28192563

    相关阅读


    免责说明:本文属传统知识与科学文献的整理、对照与研究设计讨论,仅供理解与参考,不构成医疗建议,不能替代专业诊断与治疗。文中涉及的效应量、频率区间与暴露结论多为研究报告层面的推断,个体情况差异很大;如有身体不适或健康疑虑,请就医并由专业医师判断。

    Direct Examination of the Interaction Term Between Circadian Phase Disruption and Geomagnetic Activity on Respiratory Infection Outcomes: Which Two Clock Variables Are Multiplied?

    Evidence level: B

    For geomagnetic fields to affect humans, there must first be a “receiving antenna”; the circadian system is the only one repeatedly proposed. However, “phase disruption” and “geomagnetic activity” originate from different temporal frames—before multiplication, it must be clarified what is being multiplied.

    Main Text

    I. Geomagnetic Fields Require an Antenna to Affect Humans

    The most robust evidence comes from mortality outcomes. Zilli Vieira et al. (PMID 31511079) used a two-step meta-analysis across 263 US cities: for every standard deviation increase in Kp, total mortality rose by 0.13% in winter and 0.31% in autumn; after adjusting for PM2.5, the winter rise increased to 0.47%, with a significant association in myocardial infarction deaths, but no significant association for stroke. This “partial effect” itself is a clue: if geomagnetic influence operated only indirectly via meteorology, it should not show such specificity for outcomes. It is a low-frequency field at the planetary scale; how does it write its effects inside cells? There must be a receiver—first identify that antenna.

    II. The Only Repeatedly Proposed Antenna: Cryptochrome

    Close (PMID 22418257) provides the clearest overview: cryptochrome (CRY) holds two identities—core components of the circadian gear mechanism and the light-dependent molecular basis for the animal magnetoreceptor. His mechanistic correction is crucial: geomagnetic storms do not act as a “second zeitgeber” to reset melatonin schedules, but rather perturb the HPA axis and circadian behavior more broadly via the cryptochrome pathway.

    The melatonin branch has historical data. Weydahl et al. (PMID 11774869) tracked 25 individuals at 70°N and found that geomagnetic changes must exceed a certain magnitude (>80 nT/3h) to significantly suppress salivary melatonin; Burch et al. (PMID 18472329) observed in 153 electricians that the time window with the largest difference in predicted 6-OHMS excretion falls 15 to 33 hours prior to sampling. These two characteristics determine how the modifier variable should be set: it should reflect “the stability of this clock over the past one to two days,” rather than “whether it is chaotic today.”

    III. The Antenna End: Leukocyte Reduction, Adhesion Molecule Increase

    The Harvard NAS cohort filled in the immune side. Tracy et al. (PMID 34537201) used moving average modeling from 0 to 28 days, finding that IMF, sunspot numbers, and Kp were negatively correlated with total leukocytes, neutrophils, and basophils, and the association strengthened with longer moving averages; Schiff et al. (PMID 35881632) observed in the same cohort that sICAM-1 and sVCAM-1 were positively correlated with IMF/SSN/Kp. Leukopenia coupled with elevated adhesion molecules—neither “immune suppression” nor “activation”—suggests instead a shift in the distribution and mobilization pattern of cells: fewer in the blood may mean they are adhering to vessel walls or have migrated into tissues.

    IV. How Narrow Are the Signals of Phase Disruption and Infection at the Human Level?

    Human-level evidence focuses narrowly on occupational cohorts. Loef et al. (PMID 30475977) collected 501 shift/night-shift healthcare workers and 88 day-shift workers from Klokwerk+, totaling 110,347 daily symptom diaries: shift workers had an incidence of influenza-like disease/acute respiratory infections 1.20 times that of day-shifters (95%CI 1.01-1.43), severe cases 1.22 times, but no difference in disease duration (1.02). In the same cohort, mediation analysis (PMID 32255192): 23% of the excess effect reduced to 15% and became non-significant, with subjective sleep quality being the largest mediator (indirect effect 1.29); objective sleep duration, physical activity, and meal frequency were not significant mediators.

    Bjorvatn et al. (PMID 38221779) found only significant influenza-like disease in a Norwegian cohort of 1,023 individuals (OR 1.97), while over a dozen other infections were non-significant; the same group (PMID 40055192) found in 1,335 nurses that sleep debt showed a dose-dependent relationship—aOR for pneumonia/bronchitis rose from 2.29 to 3.88, while night shift itself was only 1.28. The association is real but narrow, primarily mediated via “subjective sleep quality” rather than “less sleep.”

    V. Core Challenge: Multiplying Two Times from Different Origins

    Phase disruption is a relative quantity: the misalignment between internal clocks and external 24-hour cycles; geomagnetic activity is an absolute quantity: the intensity of the external field itself. The former reads “which grid the human clock is on”; the latter reads “how loud the sky’s clock strikes.” Multiplying quantities from two coordinate systems requires a clear definition of the product’s physical meaning before coefficients can be derived, otherwise coefficients emerge but remain uninterpretable.

    Furthermore, the precision of the “phase” ruler is limited. DLMO is the gold standard; Kennaway (PMID 36799668) synthesized 152 studies involving 4,397 individuals to establish reference ranges; however, Lewy (PMID 10452335) pointed out that low-secretors may never cross common thresholds, and phase estimation can still be confounded by amplitude; Burgess et al. (PMID 22445311) found an average difference between home and laboratory measurements of 37 to 54 minutes. Alternative solutions are now established: Laing et al. (PMID 28218891) used whole-blood transcriptomics to predict melatonin phase, achieving R²=0.74 for single samples and R²=0.90 for two samples separated by 12 hours, with an average error of approximately 9 minutes—“phase is measurable” now stands.

    VI. Replacing “Phase” with a Continuous Variable: Phase Angle

    A more practical modifier variable than DLMO is the phase angle—the difference between internal phase and behavioral time, continuous and measurable, directly characterizing the “degree of alignment” rather than “early or late.”

    Shafer et al. (PMID 41213163) conducted laboratory DLMO and recorded sleep onset continuously for 14 days in 42 healthy adults, grouping by median phase angle: narrow phase angles were associated with higher CRP in women and higher TNF-α in men. The same “misalignment” follows two distinct inflammatory pathways in men and women—if the interaction term holds, it likely also involves sex stratification.

    VII. How to Design This Test If Truly Conducted

    First, use phase angle as the modifier variable, not ordinal categories like “early riser/late sleeper.” Ordinal preferences conflate “late” and “disrupted,” whereas interaction terms require capturing the “disruption.”

    Second, set exposure windows based on 15-33 hours and long moving averages, not daily Kp.

    Third, decompose geomagnetic data into frequency bands. The 27-day and 11-year solar cycles do not overlap in the frequency domain with the 24-hour circadian phase—this is the only reason this test can be performed while the previous “25(OH)D × Solar Cycle” interaction could not: 25(OH)D is a downstream product of UVB (co-linear), whereas phase angle is a product of social time, sharing no upstream pathway with geomagnetism.

    Fourth, pre-specify negative explanations. If the interaction term appears only under stratification by “self-reported poor sleep quality” but disappears under objective phase angle stratification, then subjective sleep quality is the true explanatory variable.

    Fifth, intervention feasibility is an advantage of this design. Paul et al. (PMID 41243965) demonstrated that light cycle disruption in mice cancels circadian protection, while timed feeding can compensate. Since phase can be directly shifted by light, feeding, and melatonin, if the interaction term holds, genuine intervention trials can follow.

    Translating “Harmony between Heaven and Man” into scientific language requires splitting it into two halves: “Heaven” refers to external cycles (geomagnetism, sunlight); “Man” refers to internal phase (circadian rhythm, Ying-Wei circulation). Traditionally, both are treated as the same clock; in reality, they are two sets: one for synchronization (phase alignment), one for reading the dial (external activity intensity). Zi Wu Liu Zhu discusses “which grid the human clock reaches,” while geomagnetism discusses “how loudly the sky’s clock strikes.” Merging them into one variable turns the interaction term into an uninterpretable product—the key lies not in which modifier variable is more fixed, but whether the two variables exist within the same coordinate system.

    Evidence and Limitations

    Verified: Associations between geomagnetism and total mortality, cardiovascular mortality, and myocardial infarction deaths were significant in the meta-analysis of 263 cities (PMID 31511079); associations between geomagnetism and circulating leukocytes, adhesion molecules, and CRP were consistent across multiple exposure metrics in the NAS cohort (PMID 34537201, 35881632); consistent directionality between shift/night shifts and increased incidence of influenza-like disease across two independent cohorts (PMID 30475977, 38221779); whole-blood transcriptomics can predict melatonin phase within ~9 minutes error margin (PMID 28218891).

    Partially Verified: Historical data exists for the geomagnetism→melatonin pathway (PMID 11774869, 18472329), but samples are small, dates are old, and independent replication is lacking; cryptochrome as a human magnetoreceptor remains hypothetical (PMID 22418257); associations between phase angle and inflammatory markers based on only 42 individuals (PMID 41213163), with sex stratification conclusions pending replication.

    To Be Verified: ① The interaction term “phase disruption × geomagnetic activity” has never been directly tested; ② Whether geomagnetism acts via the HPA axis or the melatonin pathway; ③ Whether immune outcomes should be set at infection incidence, severity rates, or molecular levels (leukocyte distribution/adhesion molecules).

    Evidence Level: B
    References: PMID 31511079, PMID 22418257, PMID 11774869, PMID 18472329, PMID 34537201, PMID 35881632, PMID 30475977, PMID 32255192, PMID 38221779, PMID 40055192, PMID 36799668, PMID 10452335, PMID 22445311, PMID 28218891, PMID 41213163, PMID 41243965, PMID 36590897, PMID 15066682, PMID 28192563

    Related Reading


    Disclaimer: This article compiles traditional knowledge and scientific literature for comparison and research-design discussion only. It is not medical advice and cannot replace professional diagnosis or treatment. Effect sizes, frequency ranges and exposure conclusions cited here are largely inferences at the research-report level, and individual circumstances vary widely. If you have any discomfort or health concerns, please consult a physician.

  • GC 基因型能当「缺乏倾向」的工具变量吗?——它动的是车,不是货

    Evidence level: B

    上一篇的结论是:用”当下的 25(OH)D 血清值”给太阳暴露分层是错的,因为这个分层变量自己会随暴露移动,建议换一个暴露窗口内相对固定的变量——并点名了 GC 基因型。这一篇来验收这个候选。它确实不会动,但它动的是别的东西。

    正文

    上一篇(太阳周期×基线维生素D 交互项)的结论是:基线 25(OH)D 这个修饰变量有三个身份——被测错、随暴露移动、被疾病反向压低。出路是换一个在暴露窗口内相对固定的分层变量,并点名了 GC rs4588。

    结论先说:它不会动,但它动的是车,不是货。

    一、它动的是车,不是货

    rs4588 是 GC 基因上的错义突变,编码维生素D结合蛋白(DBP)——把 25(OH)D 从肝脏运到全身的那辆车。A 等位把第 436 位苏氨酸换成赖氨酸,得到 Gc2 型蛋白,结合亲和力更低(PMID 24472850)。

    Powe 2013 在 2,085 名黑人与白人中发现:GC 多态性单独解释了 79.4% 的 DBP 水平变异,却只解释了 9.9% 的总 25(OH)D 变异(PMID 24256378)。这个基因主要决定车有多少,顺带把货的账面数拉低了一点。

    而账面数(总 25(OH)D)不等于实际到货(游离 25(OH)D)。用 rs4588 当”缺乏倾向”的工具变量,测到的主要是载体蛋白浓度倾向,不是细胞可用维生素D的缺乏倾向——基因型经”运输”这条绕过”功能性缺乏”的路直接够到结局,违反排除限制假设。

    二、换一个仪器,符号会翻

    Afzal 2014 在 95,766 名丹麦白人中做 MR,拆开看同一结局:CYP2R1 评分给出心血管死亡 OR 0.60(0.40-0.91,保护方向),DHCR7 评分给出 OR 1.13(0.63-2.03,相反方向)(PMID 25406188)。同一批人、同一暴露、同一结局,只换仪器,符号就翻了。Berry 2012 早写下判据:若 DHCR7 上的关联在 CYP2R1 上没有,就要怀疑多效性(PMID 22629401)。

    三、「缺乏倾向」不是一个恒定属性

    Martineau 2009:古吉拉特裔中 Gc2/2 相比 Gc1/1,活动性结核 OR 2.81(1.19-6.66,P=0.009);但该关联只在血清 25(OH)D <20 nmol/L 时保留(P=0.01),≥20 nmol/L 时完全消失(P=0.36),且在巴西成人、南非儿童两个队列里毫无关联(PMID 19797128)。

    结构很关键:基因型不单独产生风险,基因型与环境水位相乘才产生风险。 它不是一张固定的牌,是一个只在低水位时才打开的阀门。

    四、单 SNP 不是单倍型

    GC 的三个主要异构体由 rs4588 与 rs7041 的组合决定:Gc1F = rs7041-T + rs4588-C,Gc1S = rs7041-G + rs4588-C,Gc2 = rs7041-T + rs4588-A。

    2026 年一篇纳入 30 项研究、8,844 人的 meta 分析给出干净对照:rs7041 与 rs4588 这两个单点,与 COPD 易感性都不显著(P 均 >0.05);但按双倍型分组后,GC1F/1F 对 GC1S/1S 的 OR 达 2.17,等位基因分析 P=0.0003;且只在东亚人群显著(DOI 10.1186/s12920-026-02443-1)。早前两篇 meta 结构相同:GC1F 增风险、GC1S/GC2 保护,只在亚洲人显著(PMID 26380769、PMID 31156695)。中国汉族人群中 Gc2 与哮喘的 OR 为 1.35(PMID 21810276)——同一个基因型在不同疾病、不同族群上给出不同方向的答案。

    五、尺子本身也有偏

    Nielson 2016 用三种方法测同一批人的 DBP:单克隆抗体法测出非洲裔比白人低约 50%,多克隆抗体法与蛋白质组学法没有差异(PMID 27007693)。Powe 那个”黑白生物可利用 25(OH)D 相同”的著名结论,很大部分来自单抗对 GC1F 异构体的识别偏倚。

    基因型既改变了被测物(总 25(OH)D),也改变了尺子(不同异构体与抗体的结合效率)。 一个变量同时出现在方程两边。

    六、GWAS 已经把这个明星降级了

    Manousaki 2020 用 401,460 名英国白人加 42,274 名欧洲人的 GWAS,识别出 138 个条件独立 SNP、映射到 69 个位点,遗传力 16.1%,且与部分社会经济性状共享遗传力,可能经”户外时间”中介(PMID 32059762)。当仪器从 1 个 SNP 扩到 138 个,评分测的已经是”户外时间+肝脏代谢+社会经济”的混合体。而非裔人群 GWAS 里,rs4588 是唯一与 25(OH)D 全基因组显著相关的位点,rs7041 则主要与 DBP 浓度相关(PMID 36811574)——同一个基因,两个位点,管两件事。

    七、反方压力(必须摆上桌)

    支持”能用”的证据不是没有。Çolak 2021 在 116,335 名丹麦白人中:遗传预测 25(OH)D 每低 10 nmol/L,细菌性肺炎 OR 1.12(1.02-1.23);但 MR-Egger 给出 1.63(0.96-2.78,不显著),且作者指出关联最强的变异在 CYP2R1 附近,不是 GC(PMID 33109689)。

    阴性侧更整齐。Butler-Laporte 2021 用 443,734 人的 25OHD GWAS 当工具,COVID-19 易感性 OR 0.95(0.84-1.08)、住院 1.09、重症 0.97,全阴(PMID 34061844)。但作者写了一句话,恰好就是本篇主题:“本结果不适用于已经缺乏维生素D的人”。Bouillon 2022 综述统计:60 多个维生素D MR 研究大多为阴,只有 4 个在多发性硬化上取得一致阳性(PMID 34815552)。

    全样本平均为零,推不出缺乏亚组也为零。 MR 的设计目标本就是估”平均因果效应”,从未被设计来估”效应修饰”。

    八、真要做,该怎么做

    STROBE-MR 要求报告三种核心假设、F 统计量/R²(判断弱工具偏倚)与敏感性分析(PMID 34698778);Burgess 2019 指南要求散点图/漏斗图看异质性并做三角互证(PMID 32760811)。至少要做到:

    • 双轨并行:一路用基因型分层(固定,但测的是载体蛋白),一路用直接测量的游离 25(OH)D 分层(测功能,但会随暴露移动),看两条路是否同向;
    • 按功能分组仪器:合成通路(DHCR7/CYP2R1)与运输/代谢通路(GC/CYP24A1)分开评分,方向不一致就按 Berry 判据处理为多效性信号;用双倍型而非单点;
    • 事前指定低水位窗口:Martineau 说效应只在 <20 nmol/L 存在,就必须事前指定,不是事后翻数据找显著区间。

    功效门槛也需记住:Berry 的算例是 80% 功效约需 8 万人。交互项样本量远低于此,阴性结果应读作”未检验”而非”已否定”。

    九、易学落点

    传统术数里,”先天禀赋”(命局、胎元、根基)常被当作与时间无关的固定属性:命里不足就是不足,走到哪里都不足。本篇的证据说,所谓”先天不足倾向”更像一个阈值触发器——Gc2/2 只在血清 25(OH)D 低于 20 nmol/L 时才与结核风险相关,水位一升上去,基因型就沉默了。

    若把”根基浅”当固定牌,人只能被动接受;若把它当阀门,问题就变成”把水位维持在阈值以上”——而水位可以靠晒、靠食、靠时令调。本篇的分辨点是:先天因素不决定结局,它决定”在什么水位下显形”。

    证据与局限

    已验证: GC 多态性是 DBP 浓度的主要决定因素(解释 79.4% 变异),对总 25(OH)D 解释力小得多(9.9%);rs4588 的 A 等位降低 DBP 与总 25(OH)D;不同 GC 异构体结合亲和力有实测差异;Gc2/2 与结核风险的关联只在 25(OH)D <20 nmol/L 时出现;rs7041/rs4588 单点与 COPD 不显著而 GC 双倍型显著。

    部分验证: 遗传预测的低 25(OH)D 与细菌性肺炎风险的关联(阳性,但最强信号在 CYP2R1 而非 GC,MR-Egger 不显著);GC 变异与哮喘/COPD 的族群特异性关联(东亚显著、非东亚多为阴性);遗传低 25(OH)D 与多发性硬化的关联(4 个 MR 一致阳性,是少数例外领域)。

    待验证: ①GC 基因型作为分层变量时,缺乏亚组中”太阳周期-呼吸道感染”关联强度是否显著高于充足亚组(从未直接检验);②游离 25(OH)D 分层与总 25(OH)D 分层在交互项上是否给出一致结论;③基因型驱动的是否主要是载体蛋白通路而非功能缺乏通路(需在分层设计中加测 DBP 与直接法游离 25(OH)D 做中介分解)。

    可证伪预测: 若 GC 基因型真是干净的”缺乏倾向”仪器,则①控制直接测量的游离 25(OH)D 后,基因型与感染结局的关联应大幅衰减;②合成通路与运输通路两组仪器应给出同向估计;③双倍型分层应比单点分层显示更强的交互信号。三点都不成立 → GC 基因型不能承担这个角色,需要换的是暴露定义,不是仪器。

    相关阅读

    证据等级: B
    参考文献: PMID 24256378, PMID 27007693, PMID 26733479, PMID 25406188, PMID 19797128, PMID 31156695, PMID 26380769, PMID 21810276, PMID 32059762, PMID 22629401, PMID 33109689, PMID 34061844, PMID 34815552, PMID 34698778, PMID 32760811, PMID 36811574, PMID 38644690, PMID 24472850, PMID 34675344;另引 DOI 10.1186/s12920-026-02443-1


    免责说明:本文属传统知识与科学文献的整理、对照与研究设计讨论,仅供理解与参考,不构成医疗建议,不能替代专业诊断与治疗。文中涉及的效应量、频率区间与暴露结论多为研究报告层面的推断,个体情况差异很大;如有身体不适或健康疑虑,请就医并由专业医师判断。

    Can GC Genotype Serve as an Instrumental Variable for \”Propensity to Lack\”? It Moves the Car, Not the Cargo

    Evidence level: B

    The previous conclusion was that using “current 25(OH)D serum levels” for stratification by solar exposure is incorrect because this stratification variable itself moves with the exposure. A relatively fixed variable within the exposure window should be used instead, and GC genotype was specifically named. This article validates that candidate. It indeed does not move, but it moves the car, not the cargo.

    Main Text

    The conclusion from the previous article (Solar Cycle × Baseline Vitamin D interaction term) is that the baseline 25(OH)D modifier variable has three identities: mis-measured, moving with exposure, and downwardly biased by disease. The solution is to switch to a stratification variable that is relatively fixed within the exposure window, specifically naming GC rs4588.

    Conclusion first: It does not move, but it moves the car, not the cargo.

    I. It Moves the Car, Not the Cargo

    rs4588 is a missense mutation on the GC gene encoding Vitamin D Binding Protein (DBP)—the vehicle that transports 25(OH)D from the liver to the rest of the body. The A allele substitutes Threonine at position 436 with Lysine, resulting in the Gc2 protein type with lower binding affinity (PMID 24472850).

    Powe et al. (2013) found in 2,085 Black and White individuals that GC polymorphism alone explained 79.4% of DBP level variation but only 9.9% of total 25(OH)D variation (PMID 24256378). This gene primarily determines how many vehicles there are, while also slightly lowering the ledger count of the cargo.

    However, the ledger count (total 25(OH)D) is not equal to actual delivery (free 25(OH)D). Using rs4588 as an instrumental variable for “deficiency propensity” measures primarily carrier protein concentration tendency, not cellular vitamin D deficiency tendency. The genotype bypasses the “functional deficiency” pathway via the “transportation” route to reach the outcome, violating the exclusion restriction assumption.

    II. Change the Instrument, and the Sign Flips

    Afzal et al. (2014) conducted a Mendelian Randomization (MR) study in 95,766 Danish whites, examining the same outcome separately: CYP2R1 scoring yielded a cardiovascular death OR of 0.60 (0.40-0.91, protective direction), while DHCR7 scoring yielded an OR of 1.13 (0.63-2.03, opposite direction) (PMID 25406188). Same population, same exposure, same outcome; simply changing the instrument flipped the sign. Berry et al. (2012) established the criterion early on: if the association on DHCR7 is absent on CYP2R1, one should suspect pleiotropy (PMID 22629401).

    III. “Deficiency Propensity” Is Not a Constant Attribute

    Martineau et al. (2009): Among Gujarati descendants, Gc2/2 compared to Gc1/1 showed an activity-associated tuberculosis OR of 2.81 (1.19-6.66, P=0.009); however, this association was retained only when serum 25(OH)D <20 nmol/L (P=0.01), completely disappearing at ≥20 nmol/L (P=0.36), and showed no association in two cohorts of Brazilian adults and South African children (PMID 19797128).

    Structure is critical: Genotype does not independently generate risk; genotype interacts with environmental water levels to produce risk. It is not a fixed card, but a valve that opens only at low water levels.

    IV. Single SNP Is Not a Haplotype

    The three main GC isoforms are determined by the combination of rs4588 and rs7041: Gc1F = rs7041-T + rs4588-C, Gc1S = rs7041-G + rs4588-C, Gc2 = rs7041-T + rs4588-A.

    A 2026 meta-analysis including 30 studies and 8,844 individuals provided a clean control: Neither single point rs7041 nor rs4588 showed significant association with COPD susceptibility (P >0.05 for both); however, after grouping by haplotype doublets, the OR for GC1F/1F vs. GC1S/1S reached 2.17, with allele analysis P=0.0003; and significance was observed only in East Asian populations (DOI 10.1186/s12920-026-02443-1). Earlier two meta-analyses with similar structures showed GC1F increased risk while GC1S/GC2 were protective, significant only in Asians (PMID 26380769, PMID 31156695). In Chinese Han populations, Gc2 had an OR of 1.35 for asthma (PMID 21810276)—the same genotype yields different directional answers across different diseases and populations.

    V. The Ruler Itself Is Biased

    Nielson et al. (2016) measured DBP in the same cohort using three methods: monoclonal antibody assay showed African Americans were ~50% lower than Whites; polyclonal antibody and proteomics methods showed no difference (PMID 27007693). A large part of Powe’s famous conclusion that “Black and White biological availability of 25(OH)D is the same” stems from monoclonal antibody bias against the Gc1F isoform.

    Genotype alters both the measured substance (total 25(OH)D) and the ruler (binding efficiency of different isoforms with antibodies). A variable appears on both sides of the equation.

    VI. GWAS Has Downgraded This Star

    Manousaki et al. (2020) used a GWAS with 401,460 British whites and 42,274 Europeans to identify 138 conditionally independent SNPs mapped to 69 loci, with heritability of 16.1%, and shared genetic liability with some socioeconomic traits, potentially mediated by “outdoor time” (PMID 32059762). When the instrument expands from 1 SNP to 138, the score measures a mixture of “outdoor time + liver metabolism + socioeconomic factors.” In African American GWAS, rs4588 was the only locus genome-wide significant for 25(OH)D, while rs7041 was primarily associated with DBP concentration (PMID 36811574)—the same gene, two loci, managing two different things.

    VII. Counter-Argument Pressure (Must Be Placed on the Table)

    Evidence supporting “it can be used” is not absent. Çolak et al. (2021) in 116,335 Danish whites: genetic prediction showed that for every 10 nmol/L lower 25(OH)D, bacterial pneumonia OR was 1.12 (1.02-1.23); however, MR-Egger gave 1.63 (0.96-2.78, not significant), and the authors noted the strongest associated variant was near CYP2R1, not GC (PMID 33109689).

    Negative results are more consistent. Butler-Laporte et al. (2021) used the 25OHD GWAS of 443,734 people as an instrument; COVID-19 susceptibility OR was 0.95 (0.84-1.08), hospitalization 1.09, severity 0.97—all negative (PMID 34061844). However, the authors wrote a sentence that coincidentally matches this article’s theme: “These results do not apply to individuals who are already vitamin D deficient.” Bouillon et al. (2022) review statistics: most of over 60 vitamin D MR studies are negative, with only 4 showing consistent positive results in multiple sclerosis (PMID 34815552).

    Zero average across the whole sample does not imply zero in the deficiency subgroup. MR design aims to estimate “average causal effect” and was never designed to estimate “effect modification.”

    VIII. If You Must Do It, How Should You Do It

    STROBE-MR requires reporting three core assumptions, F-statistics/R² (to judge weak instrument bias), and sensitivity analyses (PMID 34698778); Burgess et al. (2019) guidelines require scatter plots/histograms to check heterogeneity and triangulation (PMID 32760811). At minimum, achieve:

    • Dual-track parallelism: One track uses genotype stratification (fixed, but measures carrier protein); the other uses directly measured free 25(OH)D stratification (measures function, but moves with exposure), to see if both tracks align;
    • Group instruments by function: Separate scoring for synthetic pathways (DHCR7/CYP2R1) and transport/metabolic pathways (GC/CYP24A1); if directions are inconsistent, treat as pleiotropy signals per Berry’s criterion; use doublets rather than single points;
    • Pre-specify low-water-level windows: Martineau stated the effect exists only below 20 nmol/L; it must be pre-specified, not found post-hoc by searching for significant intervals.

    Also remember the power threshold: Berry’s calculation suggests ~80% power requires 80,000 people. Interaction study sample sizes are far below this; negative results should be read as “not tested” rather than “negated.”

    IX. Yijing Application Point

    In traditional numerology, “innate endowment” (destiny chart, fetal element, foundation) is often treated as a fixed attribute unrelated to time: if insufficient in the destiny, it remains insufficient regardless of location. This article’s evidence suggests that so-called “innate deficiency tendency” is more like a threshold trigger—Gc2/2 is only associated with tuberculosis risk when serum 25(OH)D is below 20 nmol/L; once water levels rise, the genotype goes silent.

    If “shallow foundation” is treated as a fixed card, one can only passively accept it; if treated as a valve, the problem becomes “maintaining water levels above the threshold”—which can be adjusted via sun exposure, diet, or seasonal timing. The distinguishing point of this article is: Innate factors do not determine the outcome; they determine “at what water level the manifestation occurs.”

    Evidence and Limitations

    Verified: GC polymorphism is the primary determinant of DBP concentration (explaining 79.4% variation), with much weaker explanatory power for total 25(OH)D (9.9%); A allele of rs4588 lowers DBP and total 25(OH)D; different GC isoforms have measured differences in binding affinity; Gc2/2 association with tuberculosis risk appears only when 25(OH)D <20 nmol/L; single points rs7041/rs4588 are not significant for COPD while GC doublets are.

    Partially Verified: Association between genetically predicted lower 25(OH)D and bacterial pneumonia risk (positive, but strongest signal near CYP2R1 not GC, MR-Egger not significant); population-specific associations between GC variants and asthma/COPD (significant in East Asia, mostly negative elsewhere); association between genetically lower 25(OH)D and multiple sclerosis (4 MR studies consistent positive, a minority exception field).

    To Be Verified: ① When GC genotype is used as a stratification variable, is the “Solar Cycle × Respiratory Infection” association strength in the deficiency subgroup significantly higher than in the sufficient subgroup (never directly tested); ② Whether free 25(OH)D stratification and total 25(OH)D stratification yield consistent conclusions in interaction terms; ③ Whether genotype drives primarily the carrier protein pathway rather than functional deficiency pathway (requires adding DBP measurement and direct free 25(OH)D in stratified design for mediation decomposition).

    Falsifiable Predictions: If GC genotype is truly a clean “deficiency propensity” instrument, then ① controlling for directly measured free 25(OH)D should cause a substantial attenuation of the genotype-outcome association; ② synthetic and transport pathway instrument groups should yield consistent estimates; ③ doublet stratification should show stronger interaction signals than single-point stratification. If none of these hold → GC genotype cannot bear this role; the exposure definition needs changing, not the instrument.

    Related Reading

    Evidence Level: B
    References: PMID 24256378, PMID 27007693, PMID 26733479, PMID 25406188, PMID 19797128, PMID 31156695, PMID 26380769, PMID 21810276, PMID 32059762, PMID 22629401, PMID 33109689, PMID 34061844, PMID 34815552, PMID 34698778, PMID 32760811, PMID 36811574, PMID 38644690, PMID 24472850, PMID 34675344; also cited DOI 10.1186/s12920-026-02443-1


    Disclaimer: This article compiles traditional knowledge and scientific literature for comparison and research-design discussion only. It is not medical advice and cannot replace professional diagnosis or treatment. Effect sizes, frequency ranges and exposure conclusions cited here are largely inferences at the research-report level, and individual circumstances vary widely. If you have any discomfort or health concerns, please consult a physician.

  • 缺D的人受太阳影响更大吗?「太阳周期×基线维生素D」交互项检验的三重身份陷阱

    Evidence level: B

    「虚则受邪」翻译成统计语言就是交互项。但交互项是统计学里出了名的低功效检验,而基线 25(OH)D 这个修饰变量自己有三个身份——被测错、随暴露移动、被疾病反向压低。三者叠加,这个检验最可能给出的不是答案,是沉默。

    正文

    上一篇(基线维生素D状态的效应修饰)把「效应修饰」这个概念立了起来:补同样的维生素D,基线 25(OH)D 低于 25 nmol/L 的人急性呼吸道感染风险降低 42%(OR 0.58),而 ≥25 nmol/L 的人没有获益(OR 0.89,95%CI 0.77-1.04)(PMID 28202713);芬兰 RCT 里无论分组,基线 <50 nmol/L 者的感染风险是充足者的 2.1 倍(PMID 39301110)。这些证据说的是「基线状态决定易感性」。

    现在要问一个更严格的问题:能不能把「太阳周期相位」当暴露、把「基线 25(OH)D」当修饰变量,显式建一个交互项,直接检验「缺乏亚组中太阳-疾病关联强度是否显著高于充足亚组」? 这个问题在统计上完全合法。但要穿过三道墙,每一道都会把结论削薄。

    第一道墙:交互项天生低功效。 Brookes 等用模拟算过一笔账(PMID 15066682):一个对主效应有 80% 功效的试验,对「与主效应同量级」的交互项只有 29% 功效;要让交互项达到同等功效,样本量要放大四倍,交互效应小于主效应 20% 时样本量需求急剧膨胀。这不是理论担忧,是实测分布。Sun 等系统评阅 207 个报告亚组分析的 RCT(PMID 22422832):64 个(31%)在摘要里做了亚组断言,其中只有 6 个(9%)做了统计学显著的交互检验,只有 4 个(6%)事前正确指定了效应方向,19 个多断言的试验里只有 1 个(5%)检验了交互项之间的独立性。更硬的一篇(PMID 28192563):117 条亚组断言中 46 条(39.3%)有显著交互,其中只有 1 条(2.2%)做了多重检验校正,只有 5 条(10.9%)后来被独立复现尝试过——而这 5 次复现全部没能重现原来的亚组效应,且所有效应量都向零衰减。

    第二道墙:修饰变量自己的三个身份。

    身份一,它被测错。 单次 25(OH)D 测量并不代表长期状态。Major 等在 538 人里做年内两次重复测量(PMID 23144464):类内相关系数 ICC = 0.72,去衰减因子 1.39——意思是用单次测量做回归,关联会被衰减掉约 40%。而交互项对暴露测量误差的敏感度高于主效应:Greenwood 等证明交互项系数会被暴露测量误差偏倚(PMID 16674808);Aschard 等进一步给出量级——环境变量测量误差造成「功效大幅下降 + 交互估计系统性偏零」(PMID 29424028)。

    身份二,它随暴露一起移动。 25(OH)D 本身就是 UVB 暴露的产物。用它当修饰变量去检验 UVB 暴露的效应,等于用暴露的下游指标给暴露分层。太阳周期是 11 年尺度、25(OH)D 是季节尺度,两个时间尺度不同,所以操作上并非不可行(同一周期相位下,个体 25(OH)D 差异由纬度、肤色、补充剂、GC 基因型决定,PMID 33230496)。但不能忽视:若在太阳活动低年采样,整个样本的 25(OH)D 都被压低。分层变量随暴露移动,意味着「缺乏组」里的人是随时间换的——交互项混合了「谁」和「何时」两个问题。

    身份三,它被结局反向压低。 RCT 方法学已明确指出,25(OH)D 与疾病的关系会被「疾病本身降低血清 25(OH)D」混淆(PMID 28842142)。当修饰变量与结局共变,交互项就不再是纯粹的易感性标记。

    第三道墙:尺子问题。 即便交互项被检出,「缺乏组更敏感」这句话仍需要一个标度。Knol 与 VanderWeele 给出四步呈现模板(PMID 22253321),第一步就是把四个格子的风险(RR/OR/RD)全部摆出来——因为同一批数据在加性尺度和乘性尺度上可以得出方向相反的结论。加性交互的标准度量是 RERI(PMID 19211620),它等于线性 OR 模型里的乘积项,与逻辑回归输出的乘积项不是一回事。只报一个乘积项 p 值的研究,其实回答了一个它从未声明的标度问题。

    已有的正面证据长什么样? 25(OH)D 的效应修饰确实在别的结局上被验出来过,形态值得抄:护士健康研究里,25(OH)D 与冠心病的关系在甲状旁腺素(PTH)高的人中存在(最高 vs 最低四分位 OR 0.43),PTH 低的人里没有(OR 1.28),交互 P = 0.02(PMID 28882871);1926 人随访 36 个月,25(OH)D × 空腹血糖的乘性交互 P = 0.005,与年龄的加性交互 RERI = 0.897(PMID 36643972);NHANES 7511 人,睡眠模式修饰 25(OH)D 与冠心病的关系(PMID 36863481);VITAL 结直肠腺瘤亚研究,补充 × 基线 25(OH)D 的交互 P = 0.07(普通腺瘤)、P = 0.04(进展期腺瘤)(PMID 32062040);5110 人的补充试验里,基线 25(OH)D、BMI、日照时长显著修饰「补充后 25(OH)D 的应答」(交互 P 均 <0.05),而族裔、吸烟、剧烈运动不修饰(PMID 32360596)。共同点:这些都是「同一个体层面两个变量之间」的修饰,而不是「宏观周期相位」当暴露的修饰。 后者至今没有一个直接检验——不是没人想做,是难做。

    如果真要测,必须事前钉死四件事:① 样本量按交互项算而不是按主效应算(PMID 15066682 的四倍规则),并先估算「缺乏组」占比与亚组内事件数;② 加性与乘性两种标度都报,并把四个格子的风险摆出来(PMID 22253321);③ 用重复测量或对单次测量做衰减校正(PMID 23144464);④ 用跨层次设计思路处理「暴露在群体/时期层面、结局在个体层面」时同时出现的组水平混杂与层内相关(PMID 10814658)。

    证据与局限

    已验证: 交互项检验功效远低于主效应(同量级交互需样本量×4,PMID 15066682);已发表亚组断言的复现率极低且效应量向零衰减(PMID 28192563);单次 25(OH)D 测量衰减关联约 40%(PMID 23144464);暴露测量误差使交互估计偏零(PMID 16674808、PMID 29424028);加性与乘性交互须分别报告(PMID 22253321、PMID 19211620);25(OH)D 的效应修饰在 PTH、空腹血糖、睡眠、BMI/日照等变量上已被检出(PMID 28882871、PMID 36643972、PMID 36863481、PMID 32062040、PMID 32360596);跨层次生态暴露与个体结局的效应修饰存在组水平混杂与层内相关(PMID 10814658)。

    部分验证: 「缺乏者占比不同导致太阳-疾病效应量漂移」是异质性来源的候选机制之一;基线 25(OH)D 的可分层性由纬度、肤色、补充剂、GC 基因型共同决定(PMID 33230496、PMID 31669447);基线 25(OH)D 与效应修饰的关系在维生素D轴内已被多次检出,但方向与量级随结局变化。

    待验证: 太阳周期相位 × 基线 25(OH)D 的显式交互项从未被直接建模;「修饰变量随暴露移动」造成的分层漂移量级未知;跨纬度梯度是否放大交互效应未测;加性/乘性两尺度结论一致率能否作为新交互项成立的前置筛,尚无系统评估。

    相关阅读

    证据等级: B(方法学与效应修饰证据均有 PubMed 直接支撑;「太阳周期相位 × 基线 25(OH)D 交互项」本身尚无直接研究,属方法学延伸论证)
    参考文献: PMID 15066682, PMID 22422832, PMID 28192563, PMID 23144464, PMID 16674808, PMID 29424028, PMID 22253321, PMID 19211620, PMID 28882871, PMID 36643972, PMID 36863481, PMID 32062040, PMID 32360596, PMID 10814658, PMID 28202713, PMID 39301110, PMID 33230496, PMID 28842142, PMID 31669447


    免责说明:本文属传统知识与科学文献的整理、对照与研究设计讨论,仅供理解与参考,不构成医疗建议,不能替代专业诊断与治疗。文中涉及的效应量、频率区间与暴露结论多为研究报告层面的推断,个体情况差异很大;如有身体不适或健康疑虑,请就医并由专业医师判断。

    Do People Lacking Vitamin D Experience Greater Solar Influence? The Triple Identity Trap in Testing the Interaction Between Solar Cycles and Baseline Vitamin D

    Evidence level: B

    “Xu Ze Shou Xie” (deficiency attracts pathogenicity) translates into statistical language as an interaction term. Yet, interaction terms are notoriously low-power tests in statistics, while the baseline 25(OH)D modifier variable itself holds three identities: measured incorrectly, moving with exposure, and suppressed by the outcome. When all three stack, this test is most likely to yield not an answer, but silence.

    Main Text

    The previous article (Effect Modification by Baseline Vitamin D Status) established the concept of effect modification: with the same vitamin D supplementation, individuals with baseline 25(OH)D < 25 nmol/L experienced a 42% reduction in acute respiratory infection risk (OR 0.58), whereas those with ≥25 nmol/L showed no benefit (OR 0.89, 95%CI 0.77-1.04) (PMID 28202713); in a Finnish RCT, regardless of group assignment, infection risk for those with baseline <50 nmol/L was 2.1 times that of sufficient individuals (PMID 39301110). This evidence speaks to “baseline status determines susceptibility.”

    Now we must ask a stricter question: Can we explicitly construct an interaction term treating “Solar Cycle Phase” as the exposure and “Baseline 25(OH)D” as the modifier to directly test whether the strength of the solar-disease association is significantly higher in the deficiency subgroup than in the sufficient subgroup? This question is statistically legitimate. However, it must pass through three walls, each thinning the conclusion.

    Wall One: Interaction Terms Are Innately Low-Power. Brookes et al. calculated this with simulations (PMID 15066682): a trial with 80% power for the main effect has only 29% power for an interaction term of similar magnitude; to achieve equivalent power for the interaction, sample size must be quadrupled, and requirements expand drastically when the interaction effect is less than 20% of the main effect. This is not theoretical concern but empirical distribution. Sun et al., in a systematic review of 207 RCTs reporting subgroup analyses (PMID 22422832), found that while 64 (31%) made subgroup assertions in abstracts, only 6 (9%) performed statistically significant interaction tests, only 4 (6%) pre-specified the effect direction, and in 19 trials with multiple assertions, only 1 (5%) tested independence between interaction terms. A harder study (PMID 28192563) found that among 117 subgroup assertions, 46 (39.3%) showed significant interactions; of these, only 1 (2.2%) underwent multiple testing correction, and only 5 (10.9%) were subsequently attempted for independent replication—and all five replications failed to reproduce the original subgroup effects, with all effect sizes attenuating toward zero.

    Wall Two: The Three Identities of the Modifier Variable.

    Identity One, It Is Measured Incorrectly. A single 25(OH)D measurement does not represent long-term status. Major et al., across 538 individuals with two repeated measurements within a year (PMID 23144464), found an intraclass correlation coefficient ICC = 0.72 and a attenuation factor of 1.39—meaning using single measurements in regression attenuates associations by approximately 40%. Interaction terms are more sensitive to exposure measurement error than main effects: Greenwood et al. demonstrated that interaction coefficients are biased by exposure measurement error (PMID 16674808); Aschard et al. further quantified the magnitude—environmental variable measurement error causes “drastically reduced power + systematic bias toward zero in interaction estimation” (PMID 29424028).

    Identity Two, It Moves With Exposure. 25(OH)D is itself a product of UVB exposure. Using it as a modifier to test UVB exposure effects equals stratifying by a downstream indicator of the exposure. The solar cycle operates on an 11-year scale, while 25(OH)D is seasonal; differing time scales make this operationally feasible (individual 25(OH)D differences under the same cycle phase are determined by latitude, skin pigmentation, supplementation, and GC genotype, PMID 33230496). However, one cannot ignore: if sampling occurs during low solar activity years, the entire sample’s 25(OH)D levels are suppressed. When a stratification variable moves with exposure, it implies that individuals in the “deficiency group” change over time—the interaction term conflates “who” and “when.”

    Identity Three, It Is Suppressed by the Outcome. Methodological guidelines for RCTs explicitly state that the relationship between 25(OH)D and disease is confounded by “the disease itself lowering serum 25(OH)D” (PMID 28842142). When a modifier variable covaries with the outcome, the interaction term ceases to be a pure marker of susceptibility.

    Wall Three: The Scale Problem. Even if an interaction is detected, the statement “deficiency groups are more sensitive” requires a scale. Knol and VanderWeele provide a four-step presentation template (PMID 22253321), starting with displaying the risks (RR/OR/RD) for all four cells—because the same dataset can yield opposite conclusions on additive versus multiplicative scales. The standard measure for additive interaction is RERI (PMID 19211620), which equals the product term in a linear odds ratio model but differs from the product term output by logistic regression. Studies reporting only a product term p-value actually answer a scaling question they never declared.

    What Does Existing Positive Evidence Look Like? Effect modification by 25(OH)D has been verified on other outcomes, with forms worth copying: In the Nurses’ Health Study, the relationship between 25(OH)D and coronary heart disease exists in individuals with high parathyroid hormone (PTH) (highest vs. lowest quartile OR 0.43) but not in those with low PTH (OR 1.28), with interaction P = 0.02 (PMID 28882871); in a 36-month follow-up of 1926 individuals, the multiplicative interaction of 25(OH)D × fasting glucose had P = 0.005, while the additive interaction with age yielded RERI = 0.897 (PMID 36643972); in NHANES with 7,511 individuals, sleep patterns modified the relationship between 25(OH)D and coronary heart disease (PMID 36863481); in the VITAL colorectal adenoma substudy, the interaction of supplementation × baseline 25(OH)D had P = 0.07 (ordinary adenoma) and P = 0.04 (advanced adenoma) (PMID 32062040); in a supplementation trial with 5,110 individuals, baseline 25(OH)D, BMI, and sun exposure duration significantly modified “response to supplementation” (interaction P < 0.05 for all), whereas ethnicity, smoking, and vigorous exercise did not modify (PMID 32360596). Commonality: These are all modifications between “two variables at the individual level,” not “macroscopic cycle phases” as exposures. The latter has never been directly tested—not that no one wants to, but it is difficult.

    If testing is truly required, four things must be fixed in advance: ① Calculate sample size based on interaction terms rather than main effects (PMID 15066682 quadrupling rule), and first estimate the proportion of “deficiency groups” and event counts within subgroups; ② Report both additive and multiplicative scales, displaying risks for all four cells (PMID 22253321); ③ Use repeated measurements or apply attenuation correction to single measurements (PMID 23144464); ④ Apply cross-level design approaches to handle simultaneous group-level confounding and within-level correlation when “exposure is at the group/time level but outcome is at the individual level” (PMID 10814658).

    Evidence and Limitations

    Verified: Interaction test power is far lower than main effects (same-magnitude interaction requires sample size ×4, PMID 15066682); replication rate of published subgroup assertions is extremely low with effect sizes attenuating toward zero (PMID 28192563); single 25(OH)D measurements attenuate associations by ~40% (PMID 23144464); exposure measurement error biases interaction estimates toward zero (PMID 16674808, PMID 29424028); additive and multiplicative interactions must be reported separately (PMID 22253321, PMID 19211620); effect modification by 25(OH)D has been detected on PTH, fasting glucose, sleep, BMI/sun exposure variables (PMID 28882871, PMID 36643972, PMID 36863481, PMID 32062040, PMID 32360596); cross-level ecological exposure and individual outcome effect modifications exhibit group-level confounding and within-level correlation (PMID 10814658).

    Partially Verified: “Different proportions of deficiency leading to solar-disease effect magnitude drift” is a candidate mechanism for heterogeneity; stratifiability of baseline 25(OH)D is jointly determined by latitude, skin pigmentation, supplementation, and GC genotype (PMID 33230496, PMID 31669447); the relationship between baseline 25(OH)D and effect modification has been detected multiple times within the vitamin D axis, but direction and magnitude vary by outcome.

    To Be Verified: Explicit interaction term for “Solar Cycle Phase × Baseline 25(OH)D” has never been directly modeled; magnitude of stratification drift caused by “modifier moving with exposure” is unknown; whether cross-latitude gradients amplify interaction effects remains unmeasured; whether consistency across additive/multiplicative scales can serve as a preliminary screen for new interaction terms lacks systematic evaluation.

    Related Reading

    Evidence Level: B (Methodology and effect modification evidence are directly supported by PubMed; “Solar Cycle Phase × Baseline 25(OH)D Interaction Term” itself has no direct study, belonging to methodological extension argumentation)
    References: PMID 15066682, PMID 22422832, PMID 28192563, PMID 23144464, PMID 16674808, PMID 29424028, PMID 22253321, PMID 19211620, PMID 28882871, PMID 36643972, PMID 36863481, PMID 32062040, PMID 32360596, PMID 10814658, PMID 28202713, PMID 39301110, PMID 33230496, PMID 28842142, PMID 31669447


    Disclaimer: This article compiles traditional knowledge and scientific literature for comparison and research-design discussion only. It is not medical advice and cannot replace professional diagnosis or treatment. Effect sizes, frequency ranges and exposure conclusions cited here are largely inferences at the research-report level, and individual circumstances vary widely. If you have any discomfort or health concerns, please consult a physician.

  • 把”地球自转”从 Schumann 频移里减掉:地形信号剥离的可行路径

    Evidence level: C

    一个山区 Schumann 站测到的峰频,比平原站低 0.08 Hz。这 0.08 Hz 里有多少是”山”造成的?在这条问题被问清楚之前,答案是:不知道——因为地球自己每天也在让这个数字变。

    先说清楚这件事卡在哪

    池④的 #18 课题(”山区 Schumann 站 vs 平原站有没有系统性差异”)卡了四轮,最后在上轮被判定为 T4(文献中不存在稳态站间直接对比 = 结构性空白)。那次判定同时留下了一个更强的东西:反证 2。

    反证 2 的内容是:Schumann 谐振峰的频率同时受源-观测者几何和传播条件两个因素控制,而且不同模态的符号是相反的(10.1029/2024JD041455、EGU2010-12071 相关讨论)。意思是:全球三大雷暴中心(海洋性大陆、非洲、美洲)相对观测站的方位与距离,会随着地球自转在一天里连续变化。这个变化对第一模态(约 7.83 Hz)频率的拉动方向,和对第二、第三模态的拉动方向并不一致。

    这才是问题的要害:未剥离几何项之前,任何”山区站 vs 平原站频率差”都不能归因于地形——因为你分不清这 0.08 Hz 是山体造成的,还是”这个站恰好离非洲雷暴中心更近一点”造成的。

    几何项到底有多大?先把量级摆出来

    Schumann 峰频的经典观测事实是日变幅度约 ±0.2–0.3 Hz(第一模态,视季节与源活动强度而异)。这不是噪声——恰恰相反,它是可建模的信号。

    三条真实文献锚点给出的量级感:

    锚点 1|Roldugin et al. 2004, JGR Space Physics 109(DOI 10.1029/2004JA010487)
    标题即结论:”Diurnal variations of Schumann resonance frequency in NS and EW magnetic components”。这篇的核心价值在于分了南北(NS)与东西(EW)两个磁分量去做日变——如果日变纯粹是全局性的电离层高度变化,两个分量应当同步;而分量的差异直接指向源方位的贡献。

    锚点 2|Musur & Beggan 2019, Sun and Geosphere(DOI 10.31401/sungeo.2019.01.11)
    Eskdalemuir 观测站的季节变与太阳周变。这个站是平原/丘陵站,所以它给出的”本底起伏”可以直接当作山区的对照基线——一个山区站如果只拿自己的长时间序列说”我这里有漂移”,无法排除这漂移就是 Eskdalemuir 也有的同一套季节/太阳周节律。

    锚点 3|Nickolaenko, Pechony & Price 2006, JGR Atmospheres 111(DOI 10.1029/2005JD006844)
    用光学瞬态探测器(OTD)的全球闪电分布图驱动 Schumann 模型,算出模型层面的频率变化。这是”几何项可算”的直接证据:只要有全球闪电分布图 + 传播模型,就能算出一个站在纯几何意义上应该看到什么。

    剥离的路径:三步,从粗到细

    第一步(最粗):查表法

    用 Nikolaenko & Hayakawa 的经典五模态栅格(7.8 / 14.1 / 20.3 / 26.3 / 32.5 Hz)作为参考中心频率,配合一个纯几何的”源-站距离期望值”——对任意站坐标与任意时刻,全球三大雷暴中心的位置是天文可算的(它们的季节位移有稳定规律,海洋性大陆中心的基本不动)。算出的期望频移,就是第一层要减掉的东西。

    这一步的粗糙之处很明显:它用的是 평均传播条件,不含电离层实际的昼夜不对称。

    第二步(中):TDTE 模拟

    二维电报方程(Two-Dimensional Telegraph Equation, TDTE)是 ELF 传播的标准工程模型。锚点 4|Kulak, Zięba & Micek 2003, JGR Space Physics 108(DOI 10.1029/2002JA009304)给出了太阳活动对 ELF 传播参数的二维 TDTE 求解——即电离层高度与电导率剖面的变化如何改变传播常数。

    锚点 5|Williams, Mushtak & Nickolaenko 2006, JGR Atmospheres 111(DOI 10.1029/2005JD006944)更直接:”Distinguishing ionospheric models using Schumann resonance spectra”——反过来说,就是 Schumann 谱可以用来筛电离层模型。这条的用途在于:你可以用一批已知是平原的站,先把电离层模型校准好,再把这个模型搬到山区站上去算几何+传播的联合期望值。

    第三步(细):双站差分

    用两个地形类型相同、纬度相近的站互为对照。如果两地形的 σ 相同(例如都是 C 档普通花岗岩),那么两个站的几何项差值就是纯粹的经度差带来的源-站方位差,可以直接从 TDTE 查表得到。这时候剩下的残差才有资格被讨论是不是”地形对腔体几何的扰动”。

    这一步是本文建议的落点:把”山区 vs 平原”这个两变量混杂比较,改造成”同档地形、跨经度”的单变量差分。

    三个判定档(供 #24 阶段 1 校准)

    档 条件 解释
    G0 剥离后残差 |Δf_res| < 0.05 Hz 低于 Sátori 1996 记录的日变幅度量级(DOI 10.1016/0021-9169(95)00146-8)→ 观测手段不足,不可判定
    G1 0.05 ≤ |Δf_res| < 0.10 Hz 处于可检出但未超本底的灰色带 → 证据不足,需延长观测窗
    G2 |Δf_res| ≥ 0.10 Hz 且几何项已剥离 才具备讨论”地形扰动腔体几何”的资格

    注意 G0 的含义:如果剥离几何项后发现残差只有 0.02 Hz,这不代表”山区和平原没差别”,而代表当前的观测精度不足以回答这个问题。这是两件完全不同的事,但在报告里极易被写成前者。

    三个证伪条件(F1–F3)

    • F1:若两个同为 C 档花岗岩但相隔 ≥ 8 小时经度差的站,剥离几何项后残差仍 ≥ 0.10 Hz → 说明现有 TDTE 几何模型不足,剥离路径本身失效,退回文献综述档。
    • F2:若同一站在不同月份算出的几何项修正量本身波动超过 0.10 Hz → 说明几何项的季节稳定性不足,差分对照失效。
    • F3:若山区站 σ 与平原站 σ 被证明差异 < 1 个数量级(对照 #26 σ 数据库的 a/b/c 分档)→ 依趋肤深度计算(7.83 Hz 下普通花岗岩 δ ≈ 670 km,浅层 1–10 km σ 贡献 < 1%,见 schumann-station-sigma-literature-values.md),地形对腔体几何的扰动预期为 H0,本方法学路径应当被降级而非继续投入。

    这条路径为什么值得先做

    池④的 Schumann 线已经积累了 8 个以上关联课题(#16/#17/#18/#20/#22/#24/#25/#26/#27),但没有一条解决”混杂变量”问题。几何剥离是这条线上唯一一个”做了以后所有下游实验设计都变干净”的动作:

    • 对 #24(稳态 vs 暂态):几何项剥离后,稳态残差的阈值才有意义
    • 对 #18(山区 vs 平原综述):把”找证据”改造成”找失败条件”
    • 对 #25(A 档候选站清单):站选址时可以直接比较”几何修正量可比性”,同档地形 + 相近几何修正量的站优先配对
    • 对 #17(FEM 模态节点):FEM 的边界条件需要扣除几何项,否则算出来的是”源-站几何 + 地形”的合成场

    本文的边界

    必须说清楚三件事:

    1. 本文没有做任何数值计算。三步路径、G0–G2 阈值、F1–F3 条件全部是草案,取自文献中的量级(日变 ±0.2–0.3 Hz、Sátori 1996 的日变/季节变幅度),不是从数据里算出来的。
    2. 五个锚点全部是 DOI 核验首层(Crossref 回查 title 一致),未做全文精读。其中 Kulak 2003 与 Roldugin 2004 是 JGR 系列,属地球物理学科,PubMed 不收录——这一点是上轮实测确认的(两组地球物理关键词 PubMed esearch 均返回空列表,对照组正常返回)。
    3. 几何剥离的残余不确定度本身可能超过待测信号。若 G2 门槛(0.10 Hz)在阶段 1 被证明不可达,正确处置是放弃”地形扰动腔体几何”这一支,转做”负结果 + 白皮书”,而不是降低门槛硬找信号。

    一句话:这条线真正要问的不是”山会不会改 Schumann”,而是”在扣掉地球自转这一项之后,我们还有没有本事看见山“。


    脚注|文献锚点核验状态

    编号 DOI / 标识 首作者 载体 核验方式
    A1 10.1029/2004JA010487 Roldugin V. JGR Space Physics 109 (2004) Crossref title 一致
    A2 10.31401/sungeo.2019.01.11 Musur M.A, Beggan C.D Sun and Geosphere (2019) Crossref title 一致
    A3 10.1029/2005JD006844 Nickolaenko A. JGR Atmospheres 111 (2006) Crossref title 一致
    A4 10.1029/2002JA009304 Kulak A. JGR Space Physics 108 (2003) Crossref title 一致
    A5 10.1029/2005JD006944 Williams E. JGR Atmospheres 111 (2006) Crossref title 一致
    B1 10.1016/0021-9169(95)00146-8 Sátori G. J. Atmos. Terr. Phys. 58(13) (1996) Crossref title 一致
    B2 10.1029/2024JD041455 Sátori G. JGR Atmospheres 129(19) (2024) Crossref title 一致

    无 PMID——本文所属学科(大气电学 / 地球物理)期刊主体不在 PubMed 收录范围,此结论由上轮实测支持。

    证据与局限

    本文证据等级为 C,属 hypothesis,结论限于研究层面的推断,不构成对个体情形的判定。不知道——因为地球自己每天也在让这个数字变。

    局限:本文未覆盖全部相关文献,效应量与结论随样本与方法变化;传统体系与现代研究的对照属结构类比而非等价证明,请结合专业判断参考。


    免责说明:本文属传统知识与科学文献的整理、对照与研究设计讨论,仅供理解与参考,不构成医疗建议,不能替代专业诊断与治疗。文中涉及的效应量、频率区间与暴露结论多为研究报告层面的推断,个体情况差异很大;如有身体不适或健康疑虑,请就医并由专业医师判断。

    Removing Earth’s Rotation from Schumann Resonance Shifts: A Feasible Pathway for Terrain Signal Stripping

    Evidence level: C

    A peak frequency measured at a mountain Schumann station is 0.08 Hz lower than that at a plain station. How much of this 0.08 Hz is attributable to the “mountain”? Before this question is clarified, the answer is unknown—because the Earth itself causes this number to fluctuate daily.

    Clarifying the Bottleneck

    Pool ④’s Task #18 (“Systematic differences between mountain and plain Schumann stations”) has stalled for four rounds and was recently judged as T4 (no steady-state inter-station direct comparison exists in literature = structural gap). That judgment simultaneously left behind a stronger finding: Counter-evidence 2.

    The content of Counter-evidence 2 is that the Schumann resonance peak frequency is controlled by both source-observer geometry and propagation conditions, with the signs of the modal shifts being opposite (10.1029/2024JD041455, EGU2010-12071 related discussions). This means the azimuth and distance of the three global thunderstorm centers (oceanic continents, Africa, Americas) relative to an observation station change continuously throughout the day due to Earth’s rotation. This variation pulls the frequency of the first mode (~7.83 Hz) in one direction, while pulling the second and third modes in inconsistent directions.

    This is the crux of the issue: Before stripping the geometric term, any “mountain station vs. plain station frequency difference” cannot be attributed to terrain—you cannot distinguish whether the 0.08 Hz is caused by the mountain or “this station happens to be slightly closer to an African thunderstorm center.”

    How Large is the Geometric Term? Establishing the Scale First

    The classic observational fact regarding Schumann peak frequency is a diurnal amplitude of approximately ±0.2–0.3 Hz (first mode, varying by season and source activity intensity). This is not noise; rather, it is a modelable signal.

    Three real literature anchors provide a sense of scale:

    Anchor 1|Roldugin et al. 2004, JGR Space Physics 109 (DOI 10.1029/2004JA010487)
    Title implies conclusion: “Diurnal variations of Schumann resonance frequency in NS and EW magnetic components.” The core value lies in separating the North-South (NS) and East-West (EW) magnetic components for diurnal analysis—if diurnal variation were purely global ionospheric height changes, both components should sync; component differences directly point to source azimuth contributions.

    Anchor 2|Musur & Beggan 2019, Sun and Geosphere (DOI 10.31401/sungeo.2019.01.11)
    Seasonal and solar diurnal variations at the Eskdalemuir station. This is a plain/hill station, so its “background fluctuation” can serve as a baseline for mountains—a mountain station claiming drift in its own long-term series cannot exclude that the drift shares the same seasonal/solar weekly rhythm as Eskdalemuir.

    Anchor 3|Nickolaenko, Pechony & Price 2006, JGR Atmospheres 111 (DOI 10.1029/2005JD006844)
    Using global lightning distribution maps driven by optical transient detectors (OTD) to drive the Schumann model, calculating frequency changes at the model level. This is direct evidence that the geometric term is calculable: with global lightning maps + propagation models, one can calculate what a station should see purely geometrically.

    Stripping Paths: Three Steps from Rough to Fine

    Step 1 (Rough): Lookup Table Method

    Use the classic five-mode grid from Nikolaenko & Hayakawa (7.8 / 14.1 / 20.3 / 26.3 / 32.5 Hz) as reference center frequencies, paired with a pure geometric “source-station distance expectation”—for any station coordinates and any time, the positions of the three global thunderstorm centers are astronomically calculable (their seasonal displacements follow stable patterns, and the oceanic continent center is essentially stationary). The calculated expected frequency shift is the first layer to subtract.

    The roughness here is obvious: it uses average propagation conditions, excluding actual day-night ionospheric asymmetry.

    Step 2 (Medium): TDTE Simulation

    The Two-Dimensional Telegraph Equation (TDTE) is the standard engineering model for ELF propagation. Anchor 4|Kulak, Zięba & Micek 2003, JGR Space Physics 108 (DOI 10.1029/2002JA009304) provides a 2D TDTE solution for solar activity effects on ELF propagation parameters—i.e., how ionospheric height and conductivity profile changes alter the propagation constant.

    Anchor 5|Williams, Mushtak & Nickolaenko 2006, JGR Atmospheres 111 (DOI 10.1029/2005JD006944) is more direct: “Distinguishing ionospheric models using Schumann resonance spectra”—put differently, the Schumann spectrum can be used to screen ionospheric models. The utility here is that you can use a batch of known plain stations to calibrate the ionospheric model first, then apply this model to mountain stations to calculate the joint expectation of geometry + propagation.

    Step 3 (Fine): Dual-Station Differencing

    Use two stations with identical terrain types and similar latitudes as mutual controls. If the two terrains have the same $\sigma$ (e.g., both are C-grade ordinary granite), then the difference in their geometric terms is the pure azimuth difference caused by longitude, which can be retrieved directly from TDTE tables. The remaining residual then qualifies for discussion regarding whether “terrain perturbs cavity geometry.”

    This is the recommended landing point of this paper: Transforming the two-variable confounded comparison of “mountain vs. plain” into a single-variable differential of “same-tier terrain, cross-longitude.”

    Three Judgment Levels (for #24 Phase 1 Calibration)

    Level Condition Explanation
    G0 Stripped residual $|\Delta f_{res}| < 0.05$ Hz Below the diurnal amplitude magnitude recorded by Sátori 1996 (DOI 10.1016/0021-9169(95)00146-8) → Insufficient observation means, cannot determine
    G1 $0.05 \leq |\Delta f_{res}| < 0.10$ Hz In a gray zone detectable but not exceeding background → Evidence insufficient, need extended observation window
    G2 $|\Delta f_{res}| \geq 0.10$ Hz AND geometric term stripped Only then qualified to discuss “terrain perturbing cavity geometry”

    Note on G0 meaning: If stripping the geometric term reveals a residual of only 0.02 Hz, this does not mean “no difference between mountain and plain,” but rather current observation precision is insufficient to answer this question. These are two completely different things, yet easily conflated in reports.

    Three Falsification Conditions (F1–F3)

    • F1: If two stations of the same C-grade granite but separated by $\geq$ 8 hours longitude difference still show a residual $\geq$ 0.10 Hz after stripping the geometric term → indicates existing TDTE geometric models are insufficient, stripping path itself fails, revert to literature review level.
    • F2: If the geometric term correction amount itself fluctuates by more than 0.10 Hz for the same station across different months → indicates insufficient seasonal stability of the geometric term, differential control fails.
    • F3: If mountain station $\sigma$ and plain station $\sigma$ are proven to differ by < 1 order of magnitude (referencing #26 $\sigma$ database a/b/c tiers) → based on skin depth calculation (at 7.83 Hz, ordinary granite $\delta \approx$ 670 km, shallow 1–10 km $\sigma$ contributes < 1%, see schumann-station-sigma-literature-values.md), expected terrain perturbation of cavity geometry is H0, this methodological path should be downgraded rather than continued investment.

    Why This Path Deserves Priority

    Pool ④’s Schumann line has accumulated over 8 related tasks (#16/#17/#18/#20/#22/#24/#25/#26/#27), but none resolve the “confounding variable” problem. Geometric stripping is the only action on this line that makes “all downstream experimental designs clean up” once performed:

    • For #24 (steady vs. transient): After geometric stripping, thresholds for steady residuals become meaningful.
    • For #18 (mountain vs. plain review): Transform “finding evidence” into “finding failure conditions.”
    • For #25 (A-tier candidate station list): During site selection, directly compare “geometric correction comparability,” prioritizing pairing stations with same-tier terrain + similar geometric corrections.
    • For #17 (FEM mode nodes): FEM boundary conditions need to deduct the geometric term; otherwise, calculated results are a synthesis of “source-station geometry + terrain.”

    Boundaries of This Paper

    Three things must be clarified:

    1. No numerical calculations were performed here. The three-step path, G0–G2 thresholds, and F1–F3 conditions are drafts, derived from magnitudes in literature (diurnal ±0.2–0.3 Hz, Sátori 1996 diurnal/seasonal amplitudes), not calculated from data.
    2. All five anchors are DOI-verified first-layer (Crossref title match), no full-text deep reading. Kulak 2003 and Roldugin 2004 are JGR series, belonging to geophysics, not indexed by PubMed—this was confirmed in the previous round’s actual test (two geophysics keyword PubMed searches returned empty lists, while control groups returned normally).
    3. The residual uncertainty of geometric stripping may exceed the target signal. If the G2 threshold (0.10 Hz) is proven unreachable in Phase 1, correct handling is to abandon the “terrain perturbing cavity geometry” branch and produce a “negative result + white paper”, rather than lowering thresholds to force signal detection.

    In one sentence: What this line truly asks is not “Will mountains change Schumann,” but “After deducting Earth’s rotation, do we still have the ability to see the mountains.”


    Footnote|Literature Anchor Verification Status

    ID DOI / Identifier First Author Vehicle Verification Method
    A1 10.1029/2004JA010487 Roldugin V. JGR Space Physics 109 (2004) Crossref title match
    A2 10.31401/sungeo.2019.01.11 Musur M.A, Beggan C.D Sun and Geosphere (2019) Crossref title match
    A3 10.1029/2005JD006844 Nickolaenko A. JGR Atmospheres 111 (2006) Crossref title match
    A4 10.1029/2002JA009304 Kulak A. JGR Space Physics 108 (2003) Crossref title match
    A5 10.1029/2005JD006944 Williams E. JGR Atmospheres 111 (2006) Crossref title match
    B1 10.1016/0021-9169(95)00146-8 Sátori G. J. Atmos. Terr. Phys. 58(13) (1996) Crossref title match
    B2 10.1029/2024JD041455 Sátori G. JGR Atmospheres 129(19) (2024) Crossref title match

    No PMID—journals in this discipline (atmospheric electricity/geophysics) are generally outside PubMed scope; this conclusion is supported by the previous round’s actual test.


    Disclaimer: This article compiles traditional knowledge and scientific literature for comparison and research-design discussion only. It is not medical advice and cannot replace professional diagnosis or treatment. Effect sizes, frequency ranges and exposure conclusions cited here are largely inferences at the research-report level, and individual circumstances vary widely. If you have any discomfort or health concerns, please consult a physician.

  • 把脉诊谐波和心脏超声放在同一张表里:C2/C3 比值 × LVEF 联合诊断模型的前置问题

    Evidence level: B

    医疗免责声明:本文为研究性证据梳理,不构成任何诊断标准、治疗建议或临床决策依据。文中涉及的心功能评估必须由执业医师依据规范检查完成。任何自称”把脉即可测射血分数”的说法,本文不支持。

    先回答一个直接的问题:有没有现成证据?

    没有直接命中。 本轮在 PubMed 用四组关键词实测:

    检索式 返回
    pulse wave harmonic left ventricular ejection fraction 6 条,全部是超声/瓣膜/血流动力学方向,无一涉及”脉搏谐波比值”
    photoplethysmography AND left ventricular ejection fraction 有命中,但为波形形态/可穿戴方向
    heart qi deficiency AND echocardiography 命中为中药/针刺干预的心衰研究,非脉象-心功能定量对照
    arterial waveform harmonic diastolic dysfunction echocardiography 空列表

    所以本课题(池③ #92)的处境是:“脉搏波参数 → 心功能”这一层有真实证据;”HRI-C2/C3 比值 → LVEF”这一层没有。

    有证据的那一层:脉搏波参数确实与心功能相关

    三条真实锚点(本轮 E-utilities 回查 title/期刊/首作者一致):

    锚点 1|Kulin D. et al., Physiol Int 2025(PMID 40892475)
    标题直译:”评估基于光电容积脉搏波的脉搏波参数与复合评分用于心功能评估——与……的比较”。这是最贴近本课题的一条:它做的正是”把 PPG 波形参数拿去做心功能评估”,且用了复合评分(composite score)而非单一指标——这与 C2/C3 是”比值”而非”绝对幅值”的思路同构。

    锚点 2|Chang T.Y. et al., J Cardiol 2026(PMID 41176221)
    “无接触远程光电容积脉搏波识别左室收缩功能不全型心衰的新应用”。这条的价值在于结局定义:它用的不是泛泛的”心衰”,而是按射血分数分型(HFrEF)——正是本课题要对齐的 LVEF 分型口径。

    锚点 3|Stücker B. et al., Sci Rep 2026(PMID 42486909)
    “无创 PPG 可穿戴胸贴用于运动中心输出量评估的可行性”。这条把范围推到心输出量(CO),比 LVEF 更进一步——提示 PPG 波形里携带的血流动力学信息不止一个维度。

    反向对照|池③主线已有锚点(谐波侧):
    – Lin S.K. et al., Sci Rep 2021(PMID 33903610)——动脉脉搏谱 + 多层感知机分类,证明脉搏谱本身有可分类信息
    – Wu L.W. et al., Sci Rep 2022(PMID 36509825)——动脉脉搏谱区分肌少症/代谢综合征,证明谱特征可跨病种迁移

    两层的接缝在哪:谐波侧有”谱可分类”的证据(池③主线);心功能侧有”PPG 参数可评估心功能”的证据(锚点 1/2)。但没有任何一篇把两侧用同一批受试者、同一时点、同一个统计模型连起来。

    为什么这个接缝不能靠”看起来像”来缝合

    一个常见的推论是:既然脉搏谐波反映血流动力学,而 LVEF 也是血流动力学指标,那 C2/C3 比值应该和 LVEF 相关。

    这个推论有三个未验证的跳跃:

    1. 比值 vs 绝对量。C2/C3 是第二与第三谐波的比值。文献里验证过的是 PPG 波形参数(上升时间、反射指数、二次导数比等)与心功能的关系,不是”某两个谐波的比值”。比值这个操作会消掉什么、放大什么,没有先验答案。
    2. 时点对齐。LVEF 是超声当次测量的瞬时值;C2/C3 是脉搏采集窗内的平均。若两者采集间隔 > 10 分钟,血流动力学状态(体位、呼吸、情绪)可能已经漂移。
    3. 人群同质性。锚点 2 的研究人群是已确诊的心衰患者(结局明确、病理明确)。而中医心气虚人群是证候定义的人群,其中既有心功能正常者也有异常者——两者的基线分布不同,直接套用会引入”病例组合偏倚”。

    这就是本课题真正的前置问题:不是”能不能建模型”,而是”在把两侧接起来之前,先证明接得上有意义“。

    建模前必须过的四道关

    关 要求 未通过的后果
    K1 时点同步 C2/C3 采集与超声 LVEF 测量间隔 ≤ 10 min,且同一体位 相关性被状态漂移稀释,模型系数不可解释
    K2 量表-仪器双盲 脉象采集者不知超声结果;超声医师不知脉象分级 采集时的期望效应(脉象”读”出心衰)无法排除
    K3 分层入组 按 LVEF 三分层(保留 ≥50% / 中间 40–49% / 降低 <40%)各 ≥ 60 例 无法区分”模型在哪个亚组有效”,易被整体 AUC 掩盖
    K4 比值稳定性 同一受试者 C2/C3 复测 ICC ≥ 0.75(与池③ #21 二十八脉复测信度同口径) 分母不稳,比值不可用,模型建在噪声上

    K4 是门槛中的门槛。池③ #21 已经建立了脉象复测信度的方法学(ICC + 时间漂移模型),本课题应当直接复用那套口径,而不是另起一套。

    本文不做什么

    必须写清楚边界:

    1. 本文不给任何 AUC、敏感度、特异度数值。因为不存在可引用的实测结果——本课题尚未执行。
    2. 本文不建议把 C2/C3 当 LVEF 的替代品。锚点 1 用的复合评分、锚点 2 用的是分型识别,都不是”用波形反推一个连续数值”。
    3. 本文不把”中医心气虚 = 左室收缩功能不全”写成已证结论。两者是不同定义体系下的构念;构念收敛效度是待检验命题,不是前提。
    4. 任何面向患者的”把脉测心功能”表述,均在本课题的边界之外,且与本文的医疗免责声明冲突。

    下一步(写入 backlog)

    本课题的有效推进不是产出一篇”看起来有模型”的文章,而是:

    1. 把 K1–K4 写成实验设计的入组与质控章节(与 P3-EXP 系列同格式)
    2. 复用 #21 的复测信度口径做 K4 的样本量估算
    3. 若 K4 预实验(n=30 复测)ICC < 0.75 → 本课题降级为”比值不适用”的负结果,不再进入 K1–K3

    核验锚点

    编号 PMID 首作者 载体 年份 角色
    A1 40892475 Kulin D. Physiol Int 2025 PPG 脉搏波参数 + 复合评分评估心功能
    A2 41176221 Chang T.Y. J Cardiol 2026 rPPG 识别 HFrEF(按射血分数分型)
    A3 42486909 Stücker B. Sci Rep 2026 PPG 胸贴评估心输出量
    A4 33903610 Lin S.K. Sci Rep 2021 动脉脉搏谱 + MLP(谐波侧可分类)
    A5 36509825 Wu L.W. Sci Rep 2022 动脉脉搏谱跨病种迁移

    核验方式:NCBI E-utilities esummary 逐条回查 title / 期刊 / 首作者一致。五条均为真实 PMID。


    免责说明:本文属传统知识与科学文献的整理、对照与研究设计讨论,仅供理解与参考,不构成医疗建议,不能替代专业诊断与治疗。文中涉及的效应量、频率区间与暴露结论多为研究报告层面的推断,个体情况差异很大;如有身体不适或健康疑虑,请就医并由专业医师判断。

    证据与局限

    本文证据等级为 B,属 measurable,结论限于研究层面的推断,不构成对个体情形的判定。不构成任何诊断标准、治疗建议或临床决策依据**。

    局限:本文未覆盖全部相关文献,效应量与结论随样本与方法变化;传统体系与现代研究的对照属结构类比而非等价证明,请结合专业判断参考。

    Pre-requisites for a Combined Diagnostic Model Using Pulse Diagnosis Harmonics and Cardiac Ultrasound: C2/C3 Ratio × LVEF

    Evidence level: B

    Medical Disclaimer: This article serves as a review of research evidence and does not constitute any diagnostic criteria, treatment recommendations, or clinical decision-making basis. Cardiac function assessment must be completed by licensed practitioners according to standard protocols. Any claim suggesting “pulse diagnosis alone can measure ejection fraction” is not supported by this text.

    Direct Answer: Is There Existing Evidence?

    No direct evidence exists. In this round, we conducted a search using four keyword sets on PubMed:

    Search Query Results
    pulse wave harmonic left ventricular ejection fraction 6 results, all focused on ultrasound/valvular/hemodynamic directions, with none involving “pulse wave harmonic ratios”
    photoplethysmography AND left ventricular ejection fraction Hits found, but focused on waveform morphology/wearable directions
    heart qi deficiency AND echocardiography Hits related to TCM herbal/needle intervention studies for heart failure, not quantitative pulse-symptom vs. cardiac function comparisons
    arterial waveform harmonic diastolic dysfunction echocardiography Empty list

    Thus, the situation for this topic (Pool ③ #92) is: There is real evidence linking “pulse wave parameters → cardiac function”; however, there is no evidence for the specific link “HRI-C2/C3 ratio → LVEF”.

    The Layer with Evidence: Pulse Wave Parameters Are Indeed Related to Cardiac Function

    Three verified anchors (re-checked via E-utilities for title, journal, and first author consistency in this round):

    Anchor 1|Kulin D. et al., Physiol Int 2025 (PMID 40892475)
    Title translation: “Assessment of pulse wave parameters based on photoplethysmography and composite scores for cardiac function evaluation—comparison with…” This is the closest match to this topic: it utilizes PPG waveform parameters for cardiac function assessment and employs a composite score rather than a single metric—this aligns structurally with the approach of using C2/C3 as a “ratio” rather than an absolute amplitude.

    Anchor 2|Chang T.Y. et al., J Cardiol 2026 (PMID 41176221)
    “New application of contactless remote photoplethysmography for identifying left ventricular systolic dysfunction type heart failure.” The value here lies in the outcome definition: it does not use generic “heart failure” but rather stratifies by ejection fraction (HFrEF)—exactly the LVEF stratification standard this topic aims to align with.

    Anchor 3|Stücker B. et al., Sci Rep 2026 (PMID 42486909)
    “Feasibility of non-invasive PPG wearable chest patch for cardiac output assessment during exercise.” This extends the scope to cardiac output (CO), going beyond LVEF—suggesting hemodynamic information carried by PPG waveforms spans multiple dimensions.

    Reverse Contrast|Existing Anchors in Pool ③ Mainline (Harmonic Side):
    – Lin S.K. et al., Sci Rep 2021 (PMID 33903610) — Arterial pulse spectrum + Multi-Layer Perceptron classification, proving pulse spectra contain classifiable information
    – Wu L.W. et al., Sci Rep 2022 (PMID 36509825) — Arterial pulse spectrum distinguishes sarcopenia/metabolic syndrome, proving spectral features can transfer across disease categories

    Where the Gap Lies: The harmonic side has evidence that “spectra are classifiable” (Pool ③ mainline); the cardiac function side has evidence that “PPG parameters can assess cardiac function” (Anchor 1/2). However, no single study links both sides using the same cohort, at the same time point, with a unified statistical model.

    Why This Gap Cannot Be Seamed by “Appearing Similar”

    A common inference is: Since pulse harmonics reflect hemodynamics and LVEF is also a hemodynamic indicator, the C2/C3 ratio should correlate with LVEF.

    This inference involves three unverified jumps:

    1. Ratio vs. Absolute Quantity. C2/C3 is the ratio of the second and third harmonics. Literature has validated PPG waveform parameters (rise time, reflection index, second derivative ratios, etc.) against cardiac function, not “the ratio of specific harmonics”. The operation of taking a ratio eliminates certain factors while amplifying others, with no prior answer.
    2. Time Alignment. LVEF is an instantaneous value measured via ultrasound at a specific moment; C2/C3 is an average within the pulse acquisition window. If the interval between measurements exceeds 10 minutes, hemodynamic states (posture, respiration, emotion) may drift.
    3. Population Homogeneity. The study population in Anchor 2 consists of confirmed heart failure patients (clear outcome, clear pathology). In contrast, the TCM “Heart Qi Deficiency” population is defined by syndrome patterns, containing individuals with both normal and abnormal cardiac function—their baseline distributions differ. Direct application would introduce “case-mix bias.”

    This is the true prerequisite question for this topic: Not “Can we build a model,” but “Before connecting the two sides, first prove that the connection is meaningful.“

    Four Gates to Pass Before Modeling

    Gate Requirement Consequence of Failure
    K1 Time Synchronization Interval between C2/C3 acquisition and ultrasound LVEF measurement ≤ 10 min, same posture Correlation diluted by state drift; model coefficients become uninterpretable
    K2 Double-Blind Scale-Instrument Pulse acquirer unaware of ultrasound results; Ultrasound physician unaware of pulse grading Expectation effects during acquisition (pulse “reading” heart failure) cannot be excluded
    K3 Stratified Enrollment Stratify by LVEF thirds (retain ≥50% / middle 40–49% / reduced <40%), each group ≥60 cases Cannot distinguish “which subgroup the model works in”; easily masked by overall AUC
    K4 Ratio Stability Same subject C2/C3 re-test ICC ≥ 0.75 (consistent with reliability metrics in Pool ③ #21 for 28-pulse re-test) Unstable denominator renders ratio unusable; model built on noise

    K4 is the threshold within thresholds. Pool ③ #21 has already established a methodology for pulse re-test reliability (ICC + time drift model); this topic should directly reuse that metric framework rather than establishing a new one.

    What This Article Does Not Do

    Boundaries must be clearly defined:

    1. This article provides no AUC, sensitivity, or specificity values. No measurable results exist to cite—this topic has not yet been executed.
    2. This article does not recommend treating C2/C3 as a substitute for LVEF. Anchor 1 uses composite scores; Anchor 2 uses stratification recognition; neither involves “inferring a continuous value from waveforms.”
    3. This article does not write “TCM Heart Qi Deficiency = Left Ventricular Systolic Dysfunction” as a proven conclusion. They are constructs under different definitional systems; construct convergence validity is a proposition to be tested, not a premise.
    4. Any patient-facing statement claiming “pulse diagnosis measures cardiac function” falls outside this topic’s scope and conflicts with the medical disclaimer above.

    Next Steps (To Be Added to Backlog)

    Effective progression for this topic is not producing an article that “appears to have a model,” but rather:

    1. Transform K1–K4 into enrollment and quality control chapters for experimental design (matching P3-EXP series format)
    2. Reuse the re-test reliability metrics from #21 for K4 sample size estimation
    3. If K4 pre-experiment (n=30 re-tests) yields ICC < 0.75 → Downgrade this topic to a negative result indicating “ratio inapplicability,” halting entry into K1–K3

    Anchor Verification

    ID PMID First Author Vehicle Year Role
    A1 40892475 Kulin D. Physiol Int 2025 PPG pulse wave parameters + composite score for cardiac function
    A2 41176221 Chang T.Y. J Cardiol 2026 rPPG identifies HFrEF (stratified by ejection fraction)
    A3 42486909 Stücker B. Sci Rep 2026 PPG chest patch assesses cardiac output
    A4 33903610 Lin S.K. Sci Rep 2021 Arterial pulse spectrum + MLP (harmonic side classifiable)
    A5 36509825 Wu L.W. Sci Rep 2022 Arterial pulse spectrum cross-disease transfer

    Verification Method: NCBI E-utilities esummary to re-check title, journal, and first author for each entry. All five are valid PMIDs.

  • 补维生素D 与晒太阳的头对头试验:两条免疫通路,能不能互相顶替?

    Evidence level: B

    这个问题没有单一答案——答案取决于你问的是哪个结局。

    一、先把问题问对

    “晒太阳能不能用吃维生素D 片代替”,听起来是一个问题,其实是三个问题叠在一起:能不能替代血脂的改善、能不能替代免疫分子的调节、能不能替代临床转归。这三个答案不一样,甚至方向相反。这不是和稀泥,是已有数据本身的分裂。

    二、三层”等剂量”配不平

    头对头试验的难点卡在一个词上:等剂量。可两条通路没有共同的剂量单位。

    分子层配平了,光子没配平:两组 25(OH)D 都补到 75 nmol/L,但接收的物理能量完全不同。红斑剂量配平了,产量没配平:同样 1.3 倍最小红斑剂量(1.3 SED),肤色深浅不同,合成的维生素D 差几倍。产量配平了,面积和季节又没配平。

    所以严格的”等剂量头对头”在人体上做不出来,现有试验都是准头对头——各配平一项、放弃另一项。知道它们各配平了什么,才读得懂结论。

    三、三组准头对头数据:结局决定答案

    第一组,Harrison 2021(PMID 33481482):250 名冬季军训男性,随机分安慰剂、模拟日光(1.3 SED,每周三次)、口服 D3 三臂。结果是两个干预臂都有效、而且效果相似——上呼吸道感染的峰值症状严重度降 15%、带病天数降 36%,两臂无差异,唾液 SIgA 与 cathelicidin 都没变。这一轮看似平局。但要看剂量:模拟日光这一臂的 25(OH)D 提升与 1000 IU/天口服相当,它其实是一个”低剂量 UVB”试验,不能拿它证明”高剂量 UVB 也顶不上”。

    第二组,Ponda 2017(PMID 28228421):118 名维生素D 缺乏成人,口服 D3(n=60)对窄谱 UVB(n=58),六个月。血脂全部打平,LDL 变化中位差仅 1.5 mg/dL。可同一批人做了皮肤和血的 RNA 测序——干扰素 α 和 γ 响应基因集,口服 D3 组显著上调,UVB 组显著下调,血和皮肤方向一致。把结局从血脂换成转录组,立刻不平,而且符号相反。

    第三组,PhoCIS(PMID 29780610):20 名临床孤立综合征患者,半数接受 8 周 24 次窄谱 UVB,两组所有人都被补到 25(OH)D >80 nmol/L。这是设计上已经把维生素D 配平的”纯 UVB 增量”试验。12 个月转多发性硬化率:无光疗组 100%,光疗组 70%——方向有利但不显著。后续对同一批血清做蛋白组学(PMID 41075342):UVB 组 92 个炎症蛋白中 23 个显著下调,对照组一个都没变。分子层有信号,临床层没信号。

    三组放在一起就得到核心判断:“补剂能不能替代日光”取决于你测什么。感染风险层面,能替代;转录组层面,不能替代,方向可能相反;临床转归层面,还没结论。

    四、最简版本已被证否,强版本还没被证成

    “UV 的保护全靠 25(OH)D”——这个最简版本已经否掉了。除了 Ponda 的方向相反,还有一条更硬的:在多发性硬化阳光研究里,深肤色人群 25(OH)D 与患病风险完全没有关联,但紫外线暴露仍然保护(黑人 OR 0.53)。若维生素D 是唯一中介,这个关联不该凭空消失。

    但”非依赖通路只能靠晒太阳”是更强的断言,要成立需要三样东西:

    1. 明确的中介物。候选正被逐个排除:尿刊酸异构化已出局——UVB 使小鼠 EAE 发病率降 74%、皮肤顺式尿刊酸升 30%,但单独给顺式尿刊酸完全不能复制这份保护(PMID 28056806)。目前还在台上的是 25-羟基胆固醇类氧固醇。
    2. 剂量-反应曲线,且形状要和 25(OH)D 的曲线不同。
    3. 与补剂的直接头对头,结局必须设在分子层而非临床层。

    三条都不齐。所以准确的说法不是”晒太阳不可替代”,而是”补剂不能替代的那部分,还没被定位到具体分子“。

    五、晒太阳在传统里不是一个”补营养”的动作

    传统养生讲”晒背””负日之暄”,把它当成一个整体的调摄动作,而不是”为了补钙”。这个直觉恰好和现代证据对得上:把晒太阳理解成”补维生素D 的一种方式”,你自然会问”吃片药行不行”;把它理解成一条独立的调节通路,问题就变成”晒哪一段、晒多久、晒哪里”。

    这与择日择时的逻辑同构:要的不是”总量”,是”成分与时机”。同样的日照总量,UVA 占比不同、太阳高度角不同,两条通路被拉动的比例就不同。历法上把”日照长短”(可数的量)和”日光成分/时辰”(不易数的质)混为一谈,正是本篇要拆开的东西。

    六、安全边界与可证伪预测

    必须说清:不能因为”UV 有独立好处”就无限晒。PhoCIS 用的是窄谱 311 nm,刻意避开更短的有害波段;自然光含大量 UVA,而 UVA 的独立效应与 UVB 不是一回事。瑞典 29,518 名女性 20 年队列(PMID 26992108)发现”不晒太阳的死亡风险与吸烟同级”——但作者自己写明,无法区分这是维生素D、其他 UV 机制,还是未测混杂。这是观察性证据,不是行为处方。

    可证伪预测:若做一次严格准头对头试验,把两组 25(OH)D 配平到同一水平、结局设在分子层(皮肤抗菌肽、干扰素响应基因集、循环 Treg 比例),那么 ①UVB 组与口服组应在这些指标上出现方向性差异;②差异幅度应随 UVB 剂量单调增加、而与 25(OH)D 无关;③在皮肤局部测应比在血里测更明显。三点都不出现,说明”两条平行通路”在人体上的可检测性不足。

    证据与局限

    已验证:维生素D 补剂在人群层面能小幅降低呼吸道感染风险,缺乏者受益最大(PMID 28202713);口服 D3 与窄谱 UVB 对转录组的作用方向相反(PMID 28228421);UVB 能产生超出皮肤范围、抵达血液甚至脑脊液的系统性免疫改变(PMID 42237169)。

    部分验证:UVB 的免疫调节中存在非维生素D 成分,但中介物未定位(尿刊酸已排除,PMID 28056806);”不晒太阳”与全因死亡率的关联(观察性,PMID 26992108)。

    待验证:25(OH)D 配平条件下 UVB 与口服 D3 的分子层差异能否稳定复现;非依赖通路的剂量-反应曲线形状;深肤色人群中的通路归属。

    相关阅读

    证据等级: B
    参考文献: PMID 28202713, PMID 33481482, PMID 33444565, PMID 28228421, PMID 29780610, PMID 41075342, PMID 31138860, PMID 42237169, PMID 22502796, PMID 34262557, PMID 23800457, PMID 28056806, PMID 26992108, PMID 35638295


    免责说明:本文属传统知识与科学文献的整理、对照与研究设计讨论,仅供理解与参考,不构成医疗建议,不能替代专业诊断与治疗。文中涉及的效应量、频率区间与暴露结论多为研究报告层面的推断,个体情况差异很大;如有身体不适或健康疑虑,请就医并由专业医师判断。

    Head-to-Head Trial of Vitamin D Supplementation and Sun Exposure: Can Two Immune Pathways Substitute for Each Other?

    Evidence level: B

    There is no single answer—the response depends on which outcome you are asking about.

    I. Frame the Question Correctly First

    “Can sun exposure replace vitamin D pills?” sounds like one question, but it is actually three stacked questions: Can it substitute for lipid improvement, Can it substitute for immune molecule modulation, and Can it substitute for clinical outcomes. The answers differ across these three dimensions, sometimes even opposing in direction. This is not equivocation; it reflects the inherent fragmentation within existing data.

    II. Three Layers of “Equal Dosing” Cannot Be Balanced

    The difficulty in head-to-head trials hinges on one term: equal dosing. However, the two pathways lack a common unit of dosage.

    Molecules are balanced, photons are not: Both groups achieve 25(OH)D levels of 75 nmol/L, yet the physical energy received differs fundamentally. Erythema doses are balanced, but output is not: With the same 1.3 times Minimum Erythemal Dose (1.3 SED), skin pigmentation differences result in several-fold variations in synthesized vitamin D. Output is balanced, but area and season are not.

    Therefore, a strict “equal-dose head-to-head” trial is unfeasible in humans; existing trials are quasi-head-to-head—balancing one variable while sacrificing another. Understanding what was balanced allows for accurate interpretation of conclusions.

    III. Three Quasi-Head-to-Head Data Sets: Outcomes Determine the Answer

    First Set, Harrison 2021 (PMID 33481482): 250 male cadets in winter military training were randomly assigned to placebo, simulated sunlight (1.3 SED, three times weekly), or oral D3. Results showed both intervention arms were effective with similar efficacy—upper respiratory infection peak symptom severity decreased by 15%, sick days reduced by 36%, with no difference between arms. Salivary SIgA and cathelicidin levels remained unchanged in both groups. This round appears like a draw. However, examining the dosage reveals that the simulated sunlight arm’s 25(OH)D elevation was equivalent to 1000 IU/day oral intake; it is actually a “low-dose UVB” trial, and cannot be used to prove that “high-dose UVB also cannot substitute.”

    Second Set, Ponda 2017 (PMID 28228421): 118 adults with vitamin D deficiency received oral D3 (n=60) vs. narrowband UVB (n=58) for six months. Lipid profiles were fully aligned, with a median difference in LDL change of only 1.5 mg/dL. However, the same cohort underwent skin and blood RNA sequencing—interferon α and γ response gene sets were significantly upregulated in the oral D3 group and significantly downregulated in the UVB group, showing consistent directionality between blood and skin. Switching the outcome from lipids to transcriptomics immediately reveals imbalance, with opposite signs.

    Third Set, PhoCIS (PMID 29780610): 20 patients with clinical isolated syndrome received half of the cohort 8 weeks of 24 sessions of narrowband UVB, with all groups supplemented to maintain 25(OH)D >80 nmol/L. This is a “pure UVB increment” trial by design where vitamin D was balanced. Over 12 months, the rate of conversion to multiple sclerosis was 100% in the no-phototherapy group vs. 70% in the phototherapy group—favorable direction but not statistically significant. Subsequent proteomics on the same serum samples (PMID 41075342) showed that among 92 inflammatory proteins in the UVB group, 23 were significantly downregulated, while none changed in the control group. There is a signal at the molecular level, but no signal at the clinical level.

    Placing these three datasets together yields a core judgment: “Can supplements substitute sunlight” depends on what is measured. At the infection risk level, substitution is possible; at the transcriptomic level, substitution is not possible, with potentially opposite directions; at the clinical outcome level, no conclusion exists yet.

    IV. The Simplest Version Has Been Refuted; The Strong Version Is Not Yet Proven

    “The protective effect of UV relies solely on 25(OH)D”—this simplest version has been refuted. Aside from the opposite direction in Ponda, there is a harder line of evidence: in multiple sclerosis sunlight studies, 25(OH)D showed no association with disease risk in dark-skinned populations, yet UV exposure still provided protection (Black OR 0.53). If vitamin D were the sole mediator, this association should not vanish spontaneously.

    However, the stronger assertion that “non-dependent pathways rely solely on sunlight” requires three elements to hold:

    1. A clearly identified mediator. Candidates are being eliminated one by one: uric acid isomerization has been ruled out—UVB reduced EAE incidence in mice by 74% and increased skin cis-uric acid by 30%, yet administering cis-uric acid alone failed to replicate this protection (PMID 28056806). Currently under consideration is a 25-hydroxycholesterol-derived oxysterol.
    2. A dose-response curve with a shape distinct from the 25(OH)D curve.
    3. A direct head-to-head comparison with supplements, with the outcome set at the molecular level rather than the clinical level.

    None of these three are complete. Therefore, the accurate statement is not “sunlight is irreplaceable,” but rather “the portion that supplements cannot replace has not yet been localized to a specific molecule.”

    V. Sun Exposure Is Not a “Nutrient Supplementation” Action in Traditional Practice

    Traditional wellness practices refer to “sunbathing one’s back” or “embracing the warmth of the sun” as a holistic regulatory action, not one performed “to supplement calcium.” This intuition aligns precisely with modern evidence: If you conceptualize sun exposure as “one way to supplement vitamin D,” you naturally ask if pills work; if you conceptualize it as an independent regulatory pathway, the question becomes “which segment, how long, and where to expose.”

    This mirrors the logic of date and time selection: What is sought is not “total quantity,” but “composition and timing.” With the same total solar irradiance, differing UVA proportions and solar elevation angles alter the ratio of pathways activated. Conflating “duration of daylight” (a countable quantity) with “solar composition/hour” (an uncountable quality) in the calendar is precisely what this article seeks to disentangle.

    VI. Safety Boundaries and Falsifiable Predictions

    It must be clarified: one should not infinitely expose oneself simply because “UV has independent benefits.” PhoCIS utilized narrowband 311 nm light, deliberately avoiding shorter harmful bands; natural sunlight contains substantial UVA, whose independent effects differ from UVB. A Swedish cohort of 29,518 women over 20 years (PMID 26992108) found that “not getting enough sun” carried a mortality risk comparable to smoking—but the authors explicitly stated, unable to distinguish whether this is due to vitamin D, other UV mechanisms, or unmeasured confounding. This is observational evidence, not a behavioral prescription.

    Falsifiable Predictions: If a strict quasi-head-to-head trial were conducted, balancing 25(OH)D levels between groups at the same target and setting outcomes at the molecular level (skin antimicrobial peptides, interferon response gene sets, circulating Treg ratios), then: ① The UVB group and oral group should show directional differences on these metrics; ② The magnitude of difference should increase monotonically with UVB dose and be unrelated to 25(OH)D levels; ③ Effects should be more pronounced measured locally on skin than in blood. If none of these three points appear, it indicates insufficient detectability of “two parallel pathways” in humans.

    Evidence and Limitations

    Verified: Vitamin D supplementation reduces respiratory infection risk modestly in populations, with greatest benefit in deficient individuals (PMID 28202713); oral D3 and narrowband UVB exert opposite effects on transcriptomes (PMID 28228421); UVB induces systemic immune changes extending beyond skin to blood and cerebrospinal fluid (PMID 42237169).

    Partially Verified: Non-vitamin D components exist within UVB immune modulation, but mediators remain unidentified (uric acid excluded, PMID 28056806); association between “lack of sun exposure” and all-cause mortality (observational, PMID 26992108).

    To Be Verified: Whether molecular-level differences between UVB and oral D3 under balanced 25(OH)D conditions can be stably replicated; the shape of the dose-response curve for non-dependent pathways; pathway attribution in dark-skinned populations.

    Related Reading

    Evidence Level: B
    References: PMID 28202713, PMID 33481482, PMID 33444565, PMID 28228421, PMID 29780610, PMID 41075342, PMID 31138860, PMID 42237169, PMID 22502796, PMID 34262557, PMID 23800457, PMID 28056806, PMID 26992108, PMID 35638295


    Disclaimer: This article compiles traditional knowledge and scientific literature for comparison and research-design discussion only. It is not medical advice and cannot replace professional diagnosis or treatment. Effect sizes, frequency ranges and exposure conclusions cited here are largely inferences at the research-report level, and individual circumstances vary widely. If you have any discomfort or health concerns, please consult a physician.