Tag: 因果推断

  • 体质作暴露:孟德尔随机化如何把”体质→疾病”从相关变成因果

    Evidence level: A

    体质研究最大的软肋是”只能证明相关、不能证明因果”——而孟德尔随机化给出了一条出路:把体质当成”暴露”,用遗传位点作工具变量,借”基因彩票”的自然随机分配来推断因果。这条路,韩国已经走通了第一步。

    正文

    中医体质学说有个反复被质疑的点:痰湿体质和代谢病、阳虚体质和抑郁症之间”有关联”,但这个关联到底是”体质导致疾病”,还是”疾病让人显得像某种体质”?过去所有的横断面研究都答不了这个问题——因为观察性研究永远无法排除反向因果和混杂因素。体质研究真正缺的不是数据,而是一把能切进因果关系的刀。

    这把刀就是孟德尔随机化(MR)。 它的原理听起来像魔术,其实是个天然实验:人的基因变异在受精那一刻随机分配,不受后天生活习惯、社会经济地位、甚至疾病本身的影响。如果你能用”决定体质的遗传位点”当工具变量(IV),那么”基因→体质”这一段是确定无疑的,于是顺着”基因→体质→疾病”这条链推断因果,就绕开了所有后天混杂。这就是为什么 #194 的多中心 GWAS 那么关键——体质必须有可重复的遗传位点,MR 才有工具可用;而 #195 的 PRS 证明了”先天禀赋”可以量化,MR 则是把这种量化推进到因果层面的下一步。

    一个合格的体质工具变量必须同时满足三个条件:相关性(位点确实关联体质)、独立性(位点不受”体质→疾病”通路之外的混杂影响)、排他性(位点只能通过体质影响疾病,不能直接作用于结局)。第三条最难,因为基因常有”一因多效”——所以现代 MR 的标配是一整套敏感性分析:MR-Egger 检验水平多效性、加权中位数对异常 SNP 稳健、留一法确认结果不是被单个位点带偏。换句话说,体质 MR 不是”跑一个回归就完事”,而是一场围绕因果推断的证伪演练,每一个假设都要有对应的检验来保护。

    韩国已经做出了直接示范。 2024 年,韩国韩医学研究院(KIOM)发表了一项真正意义上的”体质作暴露”研究(PMID 38927518):他们用四象体质(太阴/少阴/少阳/太阳)相关的遗传位点作工具变量,对卒中的四个亚型做孟德尔随机化。结果很有冲击力——太阴型(TE,代谢病风险最高的类型)的遗传倾向显著增加心源性栓塞性卒中(CES)风险(β=4.190,p=0.035);而少阴型(SE)的遗传倾向则显著降低 CES 风险(β=−9.336,p=3.87×10⁻⁵)。两个体质类型对同一个疾病结局给出了方向相反的因果效应,而且效应方向与四象体质已知的代谢特征完全吻合。这已经不是”体质和病有关系”,而是”你是什么体质,遗传上就在多大程度上推向某种病”——因果推断意义上的结论。

    中介链证据也在同步补上。 台湾生物库 11,030 人的研究(PMID 39257026)发现,抑郁症的多基因风险对抑郁发病的效应,有 10.21% 经由阳虚体质传导、8.41% 经由血瘀体质传导——”基因→体质→疾病”的中介路径首次在中医体质上被量化。而两步 MR 中介分析的技术模板也已经成熟:2024 年两项研究分别证明 LDL-C 介导 GLP1 受体激动剂对心衰的保护效应 4.23%(PMID 39198854)、炎症因子 LIFR 介导脂质-心衰通路 23.5%–25.2%(PMID 39145416)。把这些拼起来,九种体质做”体质→中间生物标志物→疾病”两步 MR 的全部工具都已就位。

    剩下的是三道硬门槛。 第一,工具变量强度:体质 GWAS 目前样本量小(韩国 1222 人,PMID 22394158),能通过 F>10 检验的稳健 IV 数量不足,这是最现实的瓶颈;第二,水平多效性:一个 SNP 可能同时影响多个表型,必须用 MR-Egger、加权中位数、MR-PRESSO 做敏感性分析排除;第三,体质定义的一致性:韩国四象与中医九种体质并不等同,跨体系复现是必答题。这三道门槛,恰好就是”体质作暴露”从韩国示范走向中医九种体质的施工图。

    值得注意的还有一条正在打开的新路:2023 年《中医杂志》的系统评述(DOI 10.13288/j.11-2166/r.2023.05.002)明确提出,MR 在中医领域最大的应用空间不是验证某个方药,而是”简化证型、理清不同病证间的关系,实现异病同治”——也就是把体质/证候本身当作可因果推断的暴露变量来研究。这说明”体质作暴露”不是一个孤立的脑洞,而是整个中医现代化因果研究的方向共识。对九种体质而言,最现实的第一步,是回到 #194 提到的多中心 GWAS:先把痰湿、阳虚、阴虚这三个证据最厚的体质做到万人级样本,拿到一批稳健位点,然后跑第一轮”体质→代谢指标→疾病”的两步 MR。韩国用四象体质证明了这条路走得通,中医九种体质能不能复现,就是下一步最值得盯的问题。

    证据与局限

    已验证: 以体质相关 SNP 作工具变量的 MR 设计已被真实执行(四象体质→卒中亚型,PMID 38927518);体质作为”基因→疾病”中介节点的因果证据存在(阳虚 10.21%、血瘀 8.41%,PMID 39257026);两步 MR 中介方法学成熟(PMID 39198854, 39145416);体质遗传位点已有多项 GWAS 基础(PMID 19954339, 22394158, 22889377)。

    部分验证: 中医九种体质(区别于韩国四象体质)尚无直接以体质为暴露的双样本 MR 研究;现有体质 GWAS 样本量小(千人级),IV 强度不足是普遍问题。

    待验证: ①九种体质 GWAS 位点能否构建通过 F>10 与多效性检验的稳健 IV 集;②”体质→代谢标志物→疾病”两步 MR 的中介比例与稳定性;③四象体质 MR 结果能否在中医九种体质上跨体系复现。

    相关阅读

    证据等级: A
    参考文献: PMID 38927518, PMID 39257026, PMID 39198854, PMID 39145416, PMID 22394158, PMID 19954339, PMID 22889377


    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。

    Phenotype as Exposure: How Mendelian Randomization Transforms the Association from \”Constitution to Disease\” into Causality

    Evidence level: A

    The greatest vulnerability in constitution research is the inability to prove causality despite demonstrating association—yet Mendelian Randomization (MR) offers a pathway: treating constitution as “exposure” and using genetic loci as instrumental variables (IVs) to infer causality through the natural random allocation of a “genetic lottery.” South Korea has already navigated this path successfully in its initial phase.

    Main Text

    A recurring point of scrutiny regarding Traditional Chinese Medicine (TCM) constitution theory is the observed “association” between phlegm-dampness constitution and metabolic disorders, or yang-deficiency constitution and depression. However, does the constitution cause the disease, or does the disease manifest as if the individual possesses a specific constitution? Cross-sectional studies have historically failed to answer this question because observational designs cannot rule out reverse causality or confounding factors. The true deficit in constitution research is not data, but a scalpel capable of cutting into causal inference.

    This scalpel is Mendelian Randomization (MR). Its mechanism may appear symbolic, but it functions as a natural experiment: genetic variants are randomly allocated at fertilization, unaffected by lifestyle habits, socioeconomic status, or the disease itself. If one uses “genetic loci determining constitution” as an IV, the link from “gene → constitution” is deterministic; thus, tracing the chain “gene → constitution → disease” infers causality while bypassing all postnatal confounders. This explains why the multi-center GWAS in #194 is critical—constitution must possess reproducible genetic loci for MR to have an instrument available; meanwhile, #195 demonstrated that “innate endowment” can be quantified via Polygenic Risk Scores (PRS), and MR represents the next step advancing this quantification to the causal level.

    A valid constitution IV must satisfy three conditions: relevance (the locus genuinely associates with constitution), independence (the locus is unaffected by confounders outside the “constitution → disease” pathway), and exclusion restriction (the locus affects the outcome only through constitution, not directly). The third condition is most difficult due to “pleiotropy” (one gene affecting multiple traits); therefore, modern MR standards include a suite of sensitivity analyses: MR-Egger tests for horizontal pleiotropy, weighted median robustness against outlier SNPs, and leave-one-out methods to confirm results are not driven by a single locus. In short, constitution MR is not merely “running one regression,” but a falsification exercise centered on causal inference where every hypothesis requires corresponding tests for protection.

    South Korea has provided a direct demonstration. In 2024, the Korean Institute of Oriental Medicine (KIOM) published a genuine “constitution as exposure” study (PMID 38927518): they used genetic loci associated with the Four Temperaments (Taiyin/Shaoyin/Shaoyang/Taiyang) as IVs to perform MR on four subtypes of stroke. The results were impactful—the genetic propensity for the Taiyin type (associated with the highest metabolic disease risk) significantly increased the risk of cardioembolic stroke (CES) ($\beta=4.190$, $p=0.035$); conversely, the genetic propensity for the Shaoyin type (SE) significantly decreased CES risk ($\beta=-9.336$, $p=3.87\times10^{-5}$). Two constitution types yielded causally opposite effects on the same disease outcome, with effect directions fully consistent with known metabolic characteristics of the Four Temperaments. This transcends mere “association between constitution and disease”; it establishes that “what constitution you are genetically predisposes you to a specific disease”—a conclusion in the sense of causal inference.

    Mediating chain evidence is also being filled simultaneously. A study of 11,030 individuals from the Taiwan BioBank (PMID 39257026) found that 10.21% of the effect of depression’s polygenic risk on disease onset is mediated via yang-deficiency constitution, and 8.41% via blood stasis constitution—quantifying the “gene → constitution → disease” mediation pathway for the first time in TCM constitution. Furthermore, technical templates for two-step MR mediation analysis are now mature: two studies in 2024 proved that LDL-C mediates 4.23% of the protective effect of GLP1 receptor agonists on heart failure (PMID 39198854) and that the inflammatory factor LIFR mediates 23.5%–25.2% of the lipid-heart failure pathway (PMID 39145416). Combining these, all tools for conducting “constitution → intermediate biomarker → disease” two-step MR across nine constitutions are now in place.

    Remaining are three hard thresholds. First, IV strength: current constitution GWAS sample sizes are small (Korea: 1,222 individuals; PMID 22394158), resulting in insufficient robust IVs passing the $F>10$ test, representing the most realistic bottleneck. Second, horizontal pleiotropy: a single SNP may influence multiple phenotypes, requiring sensitivity analyses via MR-Egger, weighted median, and MR-PRESSO to exclude such effects. Third, consistency in constitution definition: Korea’s Four Temperaments are not equivalent to the nine TCM constitutions; cross-system replication is a mandatory question. These three thresholds precisely map the blueprint for moving from Korea’s demonstration to the nine TCM constitutions as “exposure.”

    Notably, a new path is opening: a 2023 systematic review in the Journal of Traditional Chinese Medicine (DOI 10.13288/j.11-2166/r.2023.05.002) explicitly states that MR’s largest application space in TCM is not validating specific formulas, but “simplifying syndromic patterns, clarifying relationships between different disease-syndromes, and achieving ‘same treatment for different diseases’”—treating constitution/syndrome itself as a causally inferable exposure variable. This indicates that “constitution as exposure” is not an isolated idea but a consensus direction for modernizing TCM causal research. For the nine constitutions, the most practical first step is returning to the multi-center GWAS mentioned in #194: first, build a population-level sample (ten thousand+) for the three constitutions with the strongest evidence (phlegm-dampness, yang-deficiency, yin-deficiency) to obtain a set of robust loci, then run the first round of two-step MR for “constitution → metabolic marker → disease.” Korea has proven this path is viable; whether the nine TCM constitutions can replicate these results is the most critical question to watch next.

    Evidence and Limitations

    Verified: MR designs using constitution-associated SNPs as IVs have been executed in reality (Four Temperaments → Stroke subtypes, PMID 38927518); causal evidence exists for constitution as a “gene → disease” mediator node (yang-deficiency 10.21%, blood stasis 8.41%, PMID 39257026); two-step MR mediation methodology is mature (PMID 39198854, 39145416); multiple GWAS foundations exist for constitution genetic loci (PMID 19954339, 22394158, 22889377).

    Partially Verified: There are no direct two-sample MR studies using the nine TCM constitutions (distinct from Korea’s Four Temperaments) as exposure; existing constitution GWAS sample sizes are small (thousands), with insufficient IV strength being a common issue.

    To Be Verified: ① Can GWAS loci for the nine constitutions construct a robust IV set passing $F>10$ and pleiotropy tests? ② What are the mediation proportions and stability of “constitution → metabolic marker → disease” two-step MR? ③ Can MR results from the Four Temperaments be replicated across the nine TCM constitutions in a cross-system manner?

    Related Reading

    Evidence Level: A
    References: PMID 38927518, PMID 39257026, PMID 39198854, PMID 39145416, PMID 22394158, PMID 19954339, PMID 22889377


    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.

  • 体质多基因风险评分:从先天禀赋到可量化的因果链

    Evidence level: A

    中医所说的”先天禀赋”,在现代遗传流行病学中可以被多基因风险评分(PRS)精确量化——而且从基因→体质→疾病的因果链,正被纵向队列与孟德尔随机化方法逐步打通。

    正文

    中医体质学说有个核心主张:人的体质差异一部分”禀受于先天”,出生时就已经写进身体里,决定你容易得什么病、对什么环境敏感。过去这句话只能停留在经验层面——”先天禀赋”看不见摸不着,更像玄学。但过去十年遗传流行病学的发展,恰恰给这个古老概念配上了量化的尺子:多基因风险评分(PRS)。

    PRS 的思路很直白:复杂疾病不是单个基因决定的,而是成百上千个微小效应的基因变异累加的结果。把一个人携带的所有风险等位基因按效应大小加权求和,得到一个分数,就是这个人的”基因彩票”数字。芬兰 FinnGen 队列 13.5 万人的研究给出了这个数字的威力:与平均 PRS 相比,高 PRS 人群的终身疾病风险高出 21%–38%,冠心病、糖尿病、房颤等五种常见病的发病年龄平均提前 4–9 年(PMID 32273609)。换句话说,”先天禀赋”在多大程度上影响你什么时候得病,现在可以直接算出数字。

    这个逻辑放到中医体质上,会产生一个更深的发现。台湾生物库 11,030 人的一项研究,把抑郁症的多基因风险评分(MDD PRS)与中医体质分类(BCQ 量表)放在同一个模型里检验(PMID 39257026)。结果有三个层次:第一,MDD PRS 每升高一个标准差,阳虚体质(OR 1.07)、阴虚体质(OR 1.07)、血瘀体质(OR 1.06)的几率都显著上升——基因风险高的个体,确实更倾向于呈现”偏颇体质”;第二,阳虚体质(OR 2.07)和血瘀体质(OR 1.65)本身显著增加抑郁症风险;第三,也是最关键的一步——中介分析显示,MDD 多基因风险对抑郁症发病的效应,有 10.21% 经由阳虚体质传导、8.41% 经由血瘀体质传导。这是”基因→体质→疾病”因果链的直接统计证据:体质不只是疾病的相关标记,它可能是基因风险向疾病转化路径上的一个真实中间节点。

    那么”体质是不是病因”这个问题,还能再进一步吗?这里需要引入孟德尔随机化(MR)。MR 利用一个天然实验:基因变异在受精时随机分配,不受后天环境、生活习惯、社会经济地位等混杂因素影响。把基因变异当作”工具变量”,就可以在观察数据里模拟随机对照试验,推断暴露与结局的因果关系,而不只是相关关系。中医体质能否作为 MR 中的暴露或中介变量,取决于一个前提——体质必须能被可靠地、量化地测量。2026 年一项 1,472 人的研究展示了这条路的可行性:研究者用验证性因子分析和项目反应理论把中医”证候负荷”构造成连续潜变量,再套用 target trial emulation(目标试验模拟)+ 双重机器学习框架做因果推断,发现证候负荷每升高一个标准差,甘油三酯 z 分数上升 0.18、LDL-C 上升 0.11、C反应蛋白上升 0.16,负对照结果为零,E值 1.54–1.92 表明结果对未测混杂具有中等稳健性(PMID 42519806)。这是中医证候量化第一次完整走进现代因果推断的方法学框架。

    把三层证据拼起来,一条清晰的链路浮现:体质是可遗传的中间表型(GWAS 已定位相关位点)→ PRS 可以把”先天禀赋”变成可计算的连续分数 → 纵向队列和中介分析显示体质确实是基因风险传导的通路 → 孟德尔随机化与目标试验模拟提供了把”体质→疾病”从相关升级为因果的严格工具。对”先天禀赋”这个说法而言,这意味着一场身份转变:它从一句只能信的经验判断,变成一个可以检验、可以分层、甚至可以干预的临床工具。无锡队列 5,024 人 14 年随访给出了干预的可行性证据:高 PRS 人群如果把生活方式从糟糕(0–2 项健康因素)改进到理想(4–6 项),2 型糖尿病 10 年发病率从 6.77% 降到 3.28%(PMID 37432247)——基因风险高不是判决,而是提醒。

    证据与局限

    已验证: 体质(阳虚/阴虚/血瘀)与 MDD 多基因风险显著相关,且 MDD PRS→MDD 的效应部分由阳虚(10.21%)、血瘀(8.41%)体质中介(PMID 39257026);高 PRS 使常见病发病提前 4–9 年(PMID 32273609);中医证候负荷可量化并进入 target trial emulation 因果框架,与血脂/CRP 显著关联(PMID 42519806)。

    部分验证: “体质→疾病”的孟德尔随机化研究仍稀缺——多数 MR 用于中医药疗效与肠道菌群,直接以体质为暴露的 MR 设计尚待开展;台湾研究为横断面中介,因果时序需纵向数据确认。

    待验证: ①体质量表能否像 PRS 一样构建”体质多基因评分”并在独立队列复现;②以体质为暴露的孟德尔随机化设计(用 GWAS 定位的体质相关位点作工具变量);③体质 PRS×生活方式干预的随机对照试验,检验”先天禀赋可被后天修正”的幅度边界。

    相关阅读

    证据等级: A
    参考文献: PMID 39257026, PMID 37432247, PMID 32273609, PMID 42519806


    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。

    Polygenic Risk Scores: From Innate Endowment to Quantifiable Causal Chains

    Evidence level: A

    In Traditional Chinese Medicine (TCM), “inherent endowment” refers to the constitution determined at birth. In modern genetic epidemiology, this can be precisely quantified via Polygenic Risk Scores (PRS)—and longitudinal cohorts combined with Mendelian Randomization are progressively establishing the causal chain from genes → constitution → disease.

    Main Text

    A core tenet of TCM constitution theory is that individual differences in constitution are partly “inherently endowed,” encoded at birth and determining susceptibility to specific diseases and sensitivity to environmental factors. Historically, this concept remained empirical—”inherent endowment” was intangible and often perceived as akin to metaphysics. However, the development of genetic epidemiology over the past decade has provided a quantitative metric for this ancient concept: the Polygenic Risk Score (PRS).

    The logic of PRS is straightforward: complex diseases are not determined by a single gene but result from the cumulative effect of hundreds or thousands of minor genetic variants. By summing weighted risk alleles carried by an individual based on their effect sizes, one obtains a score representing that person’s “genetic lottery.” A study of 135,000 individuals in the Finnish FinnGen cohort demonstrated the power of this metric: compared to average PRS, high-PRS populations exhibited a 21%–38% higher lifetime disease risk and an average onset age advancement of 4–9 years for five common conditions, including coronary heart disease, diabetes, and atrial fibrillation (PMID 32273609). In other words, the extent to which “inherent endowment” influences disease onset timing can now be calculated directly.

    Applying this logic to TCM constitution yields a deeper insight. A study involving 11,030 individuals from the Taiwan BioBank integrated MDD PRS with TCM constitution classification (BCQ scale) within a unified model (PMID 39257026). The results revealed three levels of evidence: First, for every standard deviation increase in MDD PRS, the likelihood of Yang Deficiency constitution (OR 1.07), Yin Deficiency constitution (OR 1.07), and Blood Stasis constitution (OR 1.06) increased significantly—individuals with higher genetic risk indeed tended to manifest “imbalanced constitutions.” Second, Yang Deficiency (OR 2.07) and Blood Stasis (OR 1.65) constitutions themselves significantly increased the risk of depression. Third, and most critically, mediation analysis indicated that 10.21% of the effect of MDD PRS on depression onset was mediated through Yang Deficiency constitution and 8.41% through Blood Stasis constitution. This provides direct statistical evidence for the “genes → constitution → disease” causal chain: constitution is not merely a correlate of disease but may represent a genuine intermediate node in the pathway from genetic risk to clinical manifestation.

    Can the question “Is constitution a cause?” be further explored? Here, Mendelian Randomization (MR) must be introduced. MR utilizes a natural experiment: genetic variants are randomly allocated at fertilization, independent of postnatal environmental factors, lifestyle habits, or socioeconomic status. By treating genetic variants as “instrumental variables,” one can simulate randomized controlled trials within observational data to infer causal relationships between exposure and outcome rather than mere correlation. Whether TCM constitution can serve as an exposure or mediator variable in MR depends on a prerequisite: constitution must be reliably and quantitatively measured. A 2026 study involving 1,472 individuals demonstrated the feasibility of this approach: researchers constructed a continuous latent variable for TCM “syndrome burden” using confirmatory factor analysis and item response theory, then applied a target trial emulation plus double machine learning framework for causal inference. They found that for every standard deviation increase in syndrome burden, z-scores for triglycerides rose by 0.18, LDL-C by 0.11, and C-reactive protein by 0.16; null results for negative controls indicated robustness against unmeasured confounding (E-value 1.54–1.92) (PMID 42519806). This marks the first time TCM syndrome quantification has fully entered modern causal inference methodological frameworks.

    Synthesizing these three layers of evidence reveals a clear pathway: constitution is a heritable intermediate phenotype (GWAS has identified relevant loci) → PRS transforms “inherent endowment” into a calculable continuous score → longitudinal cohorts and mediation analyses confirm constitution acts as a conduit for genetic risk → Mendelian Randomization and target trial emulation provide rigorous tools to upgrade the “constitution → disease” relationship from correlation to causation. For the concept of “inherent endowment,” this signifies a shift in identity: it transitions from an empirical judgment based on faith into a testable, stratifiable, and potentially intervention-driven clinical tool. Evidence for intervention feasibility comes from the Wuxi cohort of 5,024 individuals followed over 14 years: high-PRS populations improved their lifestyle from poor (0–2 healthy factors) to ideal (4–6 healthy factors), reducing the 10-year incidence of type 2 diabetes from 6.77% to 3.28% (PMID 37432247)—high genetic risk is not a verdict, but a warning.

    Evidence and Limitations

    Verified: Significant association between constitution (Yang/Yin/Blood Stasis) and MDD PRS, with partial mediation of the MDD PRS→MDD effect via Yang Deficiency (10.21%) and Blood Stasis (8.41%) constitutions (PMID 39257026); high PRS associated with earlier onset of common diseases by 4–9 years (PMID 32273609); TCM syndrome burden quantifiable and integrated into target trial emulation causal frameworks with significant associations to lipid profiles and CRP (PMID 42519806).

    Partially Verified: Mendelian Randomization studies specifically examining “constitution → disease” remain scarce—most MR studies focus on TCM efficacy and gut microbiota; direct MR designs using constitution as exposure are still pending; the Taiwan study was cross-sectional regarding mediation, requiring longitudinal data to confirm causal temporality.

    To Be Verified: ① Can constitution scales be used to construct a “constitution PRS” replicable in independent cohorts? ② MR designs using constitution as exposure (utilizing GWAS-identified loci as instrumental variables)? ③ Randomized controlled trials examining the interaction of constitution PRS × lifestyle interventions to determine the magnitude boundary of whether “inherent endowment” can be modified postnatally.

    Related Reading

    Evidence Level: A
    References: PMID 39257026, PMID 37432247, PMID 32273609, PMID 42519806


    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.