两套脉位映射判别AUC矩阵的疾病叠加混杂控制协议:多病共存协变量建模与纯组子集验证的对比

Written by

in

Language

Evidence level: C

脉位映射判别矩阵的对角线如果”整片发亮”,要先问一句:是脉位真映出来的,还是多病共存画上去的。

正文

寸关尺分部候脏与藏医六点候脏共享同一个隐含前提:一个脉位点主要携带一个脏腑的信息。上一轮我们给这个前提定了检验标准——对角线强度指数,即”用某点信号判某脏异常”的 AUC 应显著高于非对角线。但这个检验有一个尚未处理的混杂源:多病共存。

真实门诊人群里,同时患两三种慢病的人占很大比例。一位同时有高血压、糖尿病、肾病的受试者,其心、脾、肾对应的脉位点都可能同时出现异常读数——他的脉不是”单点发亮”,而是”整片亮”。这时无论模型判哪个脏,阳性都可能命中,对角线与非对角线的差距被疾病叠加稀释,甚至被反向伪造:非对角线的 AUC 也虚高。这正是诊断方法学里经典的谱偏倚问题——Ransohoff 与 Feinstein 在 1978 年就系统指出,诊断试验的准确率随受试者谱系变化,病例谱越”纯”准确率越好看,而这不是方法变好了,是样本变简单了(PMID 692598)。用”多病共存的人”测脉位映射,等于在污染谱系上做判别力评估。

控制这个混杂有两条可对比的路线。

路线A:协变量建模。 保留全队列(含多病共存者)训练判别模型,把逐脏腑的共病状态作为协变量显式纳入模型,或在估计阶段做协变量校正——Janes 与 Pepe 给出了协变量校正 ROC 曲线与 AUC 的规范估计和比较方法,校正后可直接报告”混杂校正前后的对角线强度指数变化”(PMID 20161343)。

路线B:纯组子集验证。 只保留”仅单系统异常”的受试者子集,重算同一套 AUC 矩阵。纯组的对角线强度就是这台映射在无叠加条件下能达到的天花板。全队列 AUC 与纯组 AUC 之差(判别力衰减量)即疾病叠加混杂的实测规模。作为参照,已有单病种脉波判别研究——高血压判别 AUC 约 0.83 至 0.86(PMID 30534557)、多囊卵巢综合征的桡动脉脉波判别(PMID 37957660)——采用的多是相对”纯”的病例对照样本,它们本身就近似纯组端的上界。

两条路线互补:路线A回答”混杂能否被统计移除”,路线B回答”移除混杂人群后天花板在哪”。若校正后估计与纯组估计一致,说明协变量模型抓住了叠加结构;若仍不一致,则提示共病不是简单加法——一个系统的病可能改变另一个系统在脉位上的表达方式。这恰好对应易学”五行传变”的可操作版本:一脏病传多脏,在数据上就是共病结构对脉位读数的交互效应,可用交互项建模检验。而多病共存本身的组合结构近年已有数据驱动的系统刻画(疾病进展图谱,PMID 38992158;多病共存的复合诊断指数,PMID 40874812),说明逐脏腑协变量的结构设计有据可依。

实施与验收口径。 主终点定为混杂校正前后对角线强度指数的变化方向与幅度,并预注册假设:多病共存比例每升高一成,未校正对角线优势出现可测衰减。次终点有三:判别力衰减量与共病负担的剂量-反应曲线、协变量校正估计与纯组估计的一致性区间、脏腑间交互项的显著性。样本量须反向推算——纯组(仅单系统异常)受试者是稀缺资源,若其自然占比低,全队列招募量会随所需纯组规模成倍放大,这也是这类设计最常见的成本瓶颈。登记多病状态时不建议只记”有无慢病”这种粗颗粒:逐脏腑的异常标记才能对应”哪几个点位被同时点亮”,粗颗粒协变量会把结构性信息平均掉。

证据与局限

已验证:谱偏倚是诊断试验准确率波动的已确认来源(PMID 692598);协变量校正 AUC 有规范的估计与比较方法(PMID 20161343);单病种脉波判别存在可复现的性能基准(PMID 30534557、PMID 37957660)。

部分验证:多病共存的组合结构可被数据驱动刻画(PMID 38992158、PMID 40874812),为协变量设计提供参照;但共病状态如何改变脉象信号本身,尚缺系统实证。

待验证:协变量校正与纯组子集两条路线的 AUC 一致性;判别力衰减量与共病负担的剂量-反应关系;脏腑间传变交互项是否显著。纯组子集是稀缺样本,须先实测多病共存率再反推总招募量。

相关阅读

证据等级: C
参考文献: PMID 692598, PMID 20161343, PMID 30534557, PMID 37957660, PMID 38992158, PMID 40874812

证据等级:A=有PubMed直接支撑,B=科学类比/间接证据,C=传统文化共识/初步探索

L1 网络状态:该主题尚未进入 L1 对应网络——CORRESPONDENCE_NETWORK.md 与 EDGE_TABLE_science-yixue.md 中均无「脉位映射判别AUC矩阵 × 疾病叠加混杂控制」对应边(NETWORK 中”判别”相关条目均为 E-061 呼吸节律/压力反射主题,EDGE_TABLE 对脉位判别/混杂/多病关键词零命中)。本文按 C 级证据撰写,待研究内核补边后重新定级。


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。

Two-Pulse Position Mapping Discrimination AUC Matrix Disease Superposition Confounding Control Protocol: Comparison of Multidisease Covariate Modeling and Pure Subset Validation

Evidence level: C

If the diagonal of the pulse position mapping discrimination matrix appears “entirely illuminated,” one must first ask: is this reflection generated by true pulse position mapping, or is it an artifact of superimposed diseases?

Body

The寸关尺分部候脏 (a method mapping specific pulse positions to individual organs) and the Tibetan medicine six-point organ assessment share an implicit premise: a single pulse position primarily carries information about one organ. In the previous round, we established validation criteria for this premise—the diagonal strength index, defined as the AUC for “using a point’s signal to judge an organ’s abnormality” should be significantly higher than the non-diagonal. However, this validation has an unaddressed confounding source: coexisting diseases.

In real outpatient populations, individuals with two or three chronic conditions represent a large proportion. A subject simultaneously suffering from hypertension, diabetes, and kidney disease may exhibit abnormal readings at the pulse positions corresponding to the heart, spleen, and kidney—all at once. Their pulse is not “one point shining,” but rather “the whole area lit up.” Consequently, regardless of which organ the model identifies, a positive result may occur by chance; the gap between diagonal and non-diagonal AUCs is diluted by disease superposition, or even falsely inflated in reverse: the non-diagonal AUC also appears artificially high. This is precisely the classic spectrum bias problem in diagnostic methodology—Ransohoff and Feinstein systematically identified this in 1978, noting that diagnostic test accuracy varies with the subject spectrum; the “purer” the case spectrum, the better the accuracy appears, not because the method improved, but because the sample became simpler (PMID 692598). Using “people with coexisting diseases” to test pulse-position mapping is equivalent to assessing discriminative power on a contaminated spectrum.

Controlling this confounding offers two comparable routes.

Route A: Covariate Modeling. Retain the full cohort (including those with coexisting diseases) to train the discriminative model, explicitly incorporating the co-morbidity status for each organ as covariates, or performing covariate adjustment during the estimation phase. Janes and Pepe provided standardized methods for ROC curves and AUC estimation with covariate adjustment, allowing direct reporting of “changes in the diagonal strength index before and after confounding correction” (PMID 20161343).

Route B: Pure Subset Validation. Retain only the subset of subjects with “abnormalities in a single system” and recalculate the same AUC matrix. The diagonal strength of this pure group represents the ceiling achievable by this mapping under non-superimposed conditions. The difference between the full-cohort AUC and the pure-group AUC (the decrement in discriminative power) quantifies the actual scale of disease superposition confounding. As a reference, existing single-disease pulse-wave discriminative studies—hypertension discrimination AUC approximately 0.83 to 0.86 (PMID 30534557), polycystic ovary syndrome radial artery pulse discrimination (PMID 37957660)—mostly employ relatively “pure” case-control samples, which themselves approximate the upper bound of the pure group.

The two routes are complementary: Route A answers “can statistical methods remove confounding,” while Route B answers “where is the ceiling after removing the confounded population.” If the post-adjustment estimate aligns with the pure-group estimate, it indicates the covariate model captured the superimposed structure; if they remain inconsistent, it suggests co-morbidity is not merely additive—a disease in one system may alter how another system expresses itself at the pulse position. This precisely corresponds to the operational version of Yijing Studies’ “Five Phases transmission” (traditional consensus: the dynamic interaction of Five Tones representing organ systems); in data terms, this is the interaction effect of co-morbidity structure on pulse-position readings, which can be tested via interaction terms. Furthermore, the compositional structure of coexisting diseases itself has recently been systematically characterized using data-driven approaches (disease progression maps, PMID 38992158; composite diagnostic indices for co-morbidity, PMID 40874812), providing empirical basis for designing covariates by organ.

Implementation and Acceptance Criteria. The primary endpoint is defined as the direction and magnitude of change in the diagonal strength index before versus after confounding correction, with a pre-registered hypothesis: for every 10% increase in the proportion of coexisting diseases, a measurable attenuation occurs in the uncorrected diagonal advantage. Secondary endpoints include three metrics: (1) the dose-response curve of discriminative power decrement and co-morbidity burden; (2) the consistency interval between covariate-adjusted estimates and pure-group estimates; and (3) the significance of interaction terms between organs. Sample size must be calculated inversely—the pure group (subjects with single-system abnormalities) represents a scarce resource; if their natural proportion is low, the required full-cohort recruitment size will multiply accordingly, representing the most common cost bottleneck in such designs. When registering co-morbidity status, it is not recommended to record only “presence/absence of chronic disease” at a coarse granularity; instead, organ-specific abnormality markers must be recorded to correspond with “which specific points are simultaneously lit up,” as coarse-grained covariates would average out structural information.

Evidence and Limitations

Verified: Spectrum bias is a confirmed source of fluctuation in diagnostic test accuracy (PMID 692598); covariate-adjusted AUC possesses established estimation and comparison methodologies (PMID 20161343); pulse wave discrimination for single diseases exhibits reproducible performance benchmarks (PMID 30534557, PMID 37957660).

Partially Verified: The combination structure of coexisting conditions can be characterized by data-driven approaches (PMID 38992158, PMID 40874812), providing reference for covariate design; however, systematic empirical evidence remains lacking regarding how co-morbidity states alter the pulse signal itself.

Pending Verification: Consistency between AUC values derived from covariate adjustment versus pure subgroup subsets; the dose-response relationship between discrimination decay and disease burden; significance of organ interaction terms. Pure subgroup subsets represent scarce samples, necessitating preliminary measurement of co-morbidity rates before reverse-calculation of total recruitment volume.

Related Reading

Evidence level: C
References: PMID 692598, PMID 20161343, PMID 30534557, PMID 37997660, PMID 38992158, PMID 40874812

Evidence level: A = directly supported by PubMed, B = scientific analogy / indirect evidence, C = traditional consensus / preliminary exploration.

L1 network status: This topic has not yet entered the L1 corresponding network—neither CORRESPONDENCE_NETWORK.md nor EDGE_TABLE_science-yixue.md contains an edge for “pulse-position mapping discriminative AUC matrix × disease superposition confounding control” (the “discrimination” entries in the NETWORK are limited to E-061 respiratory rhythm / pressure-reflex themes, and the EDGE_TABLE shows zero hits for pulse-position discrimination/confounding/multi-disease keywords). This article is written at evidence level C; it will be reclassified once the research core adds the corresponding edges.


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.

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *