出生时刻暴露能达到 OR≥3 吗:全结局扫描的效应量上限审计

Written by

in

Language

Evidence level: B

上一篇算出命理断言需要 OR 4-5 才够门槛。这一篇换个问法:OR≥3 在人类疾病风险因素里到底算什么档次? 答案是——它属于”单基因罕见变异”那一档,多基因与环境暴露普遍落在 OR 1.0-1.3。出生时刻不是”还没测到 OR 3″,而是它本来就不属于会达到 OR 3 的那一类。

正文

审计的第一件事,是先把”OR 3″这个数字放到正确的坐标系里。

OR≥3 在遗传学里是”中等偏强”档,不是普通档。 ACMG/AMP 序列变异解读指南的原话是:具有孟德尔中等效应的变异,其 OR 一般为 3 或更大;高度外显的变异则有极高 OR(例如 APOE E4/E4 相对 E3/E3 的 OR 约 13)(PMID 25741868)。SCHEMA 联盟对 24,248 例精神分裂症与 97,322 名对照做全外显子荟萃,找到 10 个基因的超罕见编码变异达到 OR 3–50,其中 GRIN2A 达 24.1、GRIA3 达 20.1(PMID 35396579)。也就是说,能站上 OR 3 这条线的,是”单个基因功能被严重破坏”这类暴露——它的机制强度与”你出生在哪个时辰”根本不在一个层级。

那出生时刻暴露实测落在哪一档? 现有全结局证据很清楚。

规模最大的一次是 SeaWAS(Boland 等,2015):纽约 174.9 万人的电子病历、1,688 种疾病,逐月做 logistic 回归 + 卡方检验 + FDR 校正,最终只有 55 种疾病与出生月份显著相关(PMID 26041386)。独立复制研究(Li 等,2016)在另一家医院 116.9 万患者中重跑,9 种心血管类关联里复制了 7 种,确认”心血管风险在冬末春初出生者中达峰”(PMID 27624541)。

关键在于效应量。英国 MS 研究在调整时间与地区混杂后,给出出生季节效应中最高的一个点估计:4 月出生者相对 11 月,OR 1.24(PMID 27366989)。UK Biobank 约 45 万人的系统评估更直白:低出生体重的 2 月 vs 9 月 OR 1.23、青春期提前 9 月 vs 7 月 OR 1.22、身材偏矮 12 月 vs 6 月 OR 1.09(PMID 27123493)。瑞典乳糜泻队列 29,096 例的夏季出生 OR 仅 1.06(PMID 23172784)。美国多中心脑肿瘤病例对照(489 例胶质瘤 + 197 例脑膜瘤)发现出生月份与风险呈 12 个月周期波动,峰在 1-2 月、谷在 7-8 月,但同样只是”周期性形态”而非强效应(PMID 15277620)。2024 年 UK Biobank 的出生季节-影像研究,所有显著关联的标准化效应量在 β 0.013–0.026 之间——作者自己的结论是”效应量之小提示季节性对成人健康作用有限”(PMID 38776272)。

换句话说,出生月份/季节的效应量天花板大约在 OR 1.2–1.3。 这跟 OR 3 之间不是”差一点”,而是差一个档次(约 3–10 倍)。

唯一的”越界”报告恰恰是最好的反例。 中国一项单中心研究(Zhang 等,2019)报告 11 月出生 OR 1.390、12 月 1.358、2 月 1.332,冬季相对春季 1.247(PMID 31122721)——这是文献里少见的、接近 OR 1.4 的数字。但看设计:单中心回顾、4,729 例 CAD vs 7,534 名对照、倾向性评分匹配、未做全结局多重检验校正、无外部复制。而它”发现”的月份(11 月、12 月、2 月)与前述多中心研究给出的峰值方向并不完全一致。这正是全表型扫描最经典的陷阱:在 1,688 种疾病 × 12 个月 = 2 万多个检验格子里,最大的那几个点估计天然是噪声堆出来的极值。SeaWAS 的做法是必须过 FDR 校正(因此只剩 55 个),而不是挑最大的那个报出来。

方法学本身还有一层代价。 把暴露从”季节”细化到”时辰”,名义格子从 4 格涨到 12 格,检验数量随之翻三倍,多重比较校正后的通过门槛更高、可发现的显著数只会更少。同时,PheWAS 类研究对电子病历数据的偏倚高度敏感——不同医院的数据质量、编码习惯、就诊流程都会影响结果,这也是为何独立复制被反复强调(PMID 24270849, 25474351)。

那么”时辰粒度全结局扫描”的预期结果是什么? 坦率说,这个扫描至今没有被做过——现有全表型研究全部停在”月份”粒度,因为绝大多数电子病历只记录出生日期、不记录出生时刻。这是本篇要标注的最重要缺口。但从已知的效应量分布可以做出一个有依据的预测:若对时辰粒度做同样的 FDR 校正扫描,显著命中数会少于月份粒度,且”OR≥3″的命中率预期接近零。 理由不是”技术不够”,而是出生时刻属于低效应量环境暴露这一类别——它不像单基因变异那样直接破坏蛋白功能,只是通过光周期、季节感染暴露等弱通路留下痕迹(PMID 29501066 证明其与精神分裂症 PRS 不相关,是独立的环境效应)。而常见变异效应量分布的研究显示,效应越小、数量越多,绝大多数位点落在”小几个数量级”的区间(PMID 22003128)。

所以对命理体系,这次审计给出的是一个类别判断,而不是一个待补的数字。 “出生时刻暴露达到 OR≥3″之所以不成立,不是因为样本还不够大、方法还不够好,而是因为它与”达到 OR 3″的那些暴露不属于同一类别。把审计线设在这里,恰好能把”命理断言需要 OR 4-5″与”出生时刻能提供的 OR 1.2-1.3″之间的差距说清楚——这是一道由暴露类型决定的结构性鸿沟。

证据与局限

已验证: OR≥3 在遗传学中属”孟德尔中等效应”档,典型代表是单基因罕见变异(PMID 25741868, 35396579);出生月份/季节关联的效应量天花板约 OR 1.2–1.3,多项大规模研究报告在 OR 1.06–1.24 区间(PMID 27123493, 27366989, 23172784, 38776272);SeaWAS 在 174.9 万人/1,688 病中经 FDR 校正后仅 55 种显著,并在独立 116.9 万人队列复制(PMID 26041386, 27624541);PheWAS 对电子病历偏倚敏感、依赖独立复制(PMID 24270849, 25474351)。

部分验证: 中国单中心研究报告出生月份 OR 1.33–1.39(PMID 31122721),为文献中最高一档,但设计为单中心回顾、无全结局多重校正、无外部复制,与多中心研究峰值方向不完全一致;常见变异效应量分布研究提示”效应越小数量越多”的普遍形态(PMID 22003128),可外推但非直接证据。

待验证: ① 时辰粒度(而非月份)的全表型关联扫描从未被做过——现有电子病历普遍只有出生日期,这是本篇标注的最大缺口;② 若补做,其 FDR 校正后显著数与 OR≥3 命中率能否落在本文预测区间,需实测;③ 出生时刻效应在非欧洲、非北美人群中的量级分布是否一致,尚无系统评估。

相关阅读

证据等级: B
参考文献: PMID 25741868, PMID 35396579, PMID 26041386, PMID 27624541, PMID 27123493, PMID 27366989, PMID 23172784, PMID 38776272, PMID 31122721, PMID 24270849, PMID 25474351, PMID 29501066, PMID 22003128, PMID 15277620


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 Birth Time Exposure Reach an OR≥3: Audit of Effect Size Upper Limits in Full-Outcome Scan

Evidence level: B

The previous analysis established that a threshold of OR 4–5 is required for fate assertions. This piece reframes the question: What tier does OR ≥ 3 represent within human disease risk factors? The answer is that it belongs to the category of “single-gene rare variants,” whereas polygenic exposures and environmental factors typically fall within an OR range of 1.0–1.3. Birth time is not merely “not yet measured at OR 3”; rather, it fundamentally does not belong to the class of exposures capable of reaching OR 3.

Main Text

The first step in the audit is to place the figure “OR 3” into the correct coordinate system.

In genetics, OR ≥ 3 represents a “moderate-to-strong” tier, not an average one. The ACMG/AMP Sequence Variant Interpretation Guidelines state that variants with Mendelian moderate effects generally have an OR of 3 or greater; highly penetrant variants exhibit very high ORs (e.g., APOE E4/E4 relative to E3/E3 has an OR of approximately 13) (PMID 25741868). The SCHEMA Consortium conducted a whole-exome meta-analysis across 24,248 cases of schizophrenia and 97,322 controls, identifying 10 genes with ultra-rare coding variants reaching an OR of 3–50, with GRIN2A at 24.1 and GRIA3 at 20.1 (PMID 35396579). Thus, exposures capable of reaching the OR 3 threshold are those involving “severe disruption of single gene function”—a mechanism intensity that operates on a completely different tier than “which hour you were born.”

Where does the empirical evidence for birth time exposure actually fall? Existing full-outcome data is clear.

The largest-scale study was SeaWAS (Boland et al., 2015): analyzing electronic medical records of 17.49 million people in New York across 1,688 diseases using monthly logistic regression, chi-square tests, and FDR correction, only 55 diseases showed significant associations with birth month (PMID 26041386). An independent replication study (Li et al., 2016) re-ran the analysis in a separate hospital cohort of 11.69 million patients; 7 out of 9 cardiovascular associations were replicated, confirming that “cardiovascular risk peaks among those born at the end of winter and beginning of spring” (PMID 27624541).

The critical factor is effect size. After adjusting for temporal and regional confounders, the UK Multiple Sclerosis study provided one of the highest point estimates for birth season effects: OR 1.24 for April-born versus November-born (PMID 27366989). The UK Biobank’s systematic assessment of approximately 450,000 individuals was even more direct: OR 1.23 for low birth weight in February versus September, OR 1.22 for early puberty in September versus July, and OR 1.09 for shorter stature in December versus June (PMID 27123493). The Swedish Celiac Disease cohort of 29,096 individuals showed a summer birth OR of only 1.06 (PMID 23172784). A US multicenter case-control study of brain tumors (489 gliomas + 197 meningiomas) found a 12-month cyclical fluctuation in birth month risk, peaking in January–February and troughing in July–August, yet this remained merely a “cyclical pattern” rather than a strong effect (PMID 15277620). A 2024 UK Biobank birth season-imaging study found standardized effect sizes for all significant associations between β 0.013 and 0.026; the authors concluded that “the effect size hints that seasonality has a limited impact on adult health” (PMID 38776272).

In short, the effect size ceiling for birth month/season is approximately OR 1.2–1.3. The gap between this and OR 3 is not “close enough,” but rather a distinct tier difference (approximately 3–10 times).

The only “outlier” report serves as the best counter-example. A single-center Chinese study (Zhang et al., 2019) reported an OR of 1.390 for November, 1.358 for December, and 1.332 for February, with winter relative to spring at 1.247 (PMID 31122721)—a figure close to OR 1.4, which is rare in the literature. However, examining the design: single-center retrospective study, 4,729 CAD cases vs. 7,534 controls, propensity score matching, no correction for multiple testing in full-outcome analysis, and no external replication. Furthermore, the months identified (“found” November, December, February) do not align perfectly with the peak direction provided by the aforementioned multicenter studies. This illustrates a classic trap in full-phenome scanning: across 1,688 diseases × 12 months = over 20,000 test cells, the largest point estimates are naturally generated by noise extremes. SeaWAS required passing FDR correction (leaving only 55 significant findings) rather than simply reporting the largest single estimate.

Methodology itself incurs another cost. Refining exposure from “season” to “hour” increases nominal cells from 4 to 12, tripling the number of tests. Consequently, the threshold for multiple comparison correction becomes higher, and fewer significant findings will be discovered. Additionally, PheWAS-style studies are highly sensitive to biases in electronic medical record data; data quality, coding habits, and clinical workflows across different hospitals can significantly influence results, necessitating repeated emphasis on independent replication (PMID 24270849, 25474351).

So, what are the expected results of a “full-outcome scan at hour granularity”? Frankly, this scan has not yet been conducted—existing full-phenome studies stop at “month” granularity because most electronic medical records record only birth dates, not birth times. This is the most significant gap highlighted in this piece. However, based on known effect size distributions, we can make an evidence-based prediction: if a similar FDR-corrected scan were performed at hour granularity, the number of significant hits would be lower than month granularity, and the expected hit rate for “OR ≥ 3” would approach zero. The reason is not “insufficient technology,” but rather that birth time belongs to the category of low-effect-size environmental exposures—unlike single-gene variants that directly disrupt protein function, it leaves traces only through weak pathways such as photoperiodism or seasonal infection exposure (PMID 29501066 demonstrates its lack of association with schizophrenia PRS, identifying it as an independent environmental effect). Research on common variant effect size distributions shows that smaller effects occur more frequently, with the vast majority of loci falling into intervals several orders of magnitude lower (PMID 22003128).

Therefore, for the Yijing system, this audit provides a categorical judgment rather than a number to be filled. The claim that “birth time exposure reaches OR ≥ 3” is invalid not because the sample size is insufficient or methods are inadequate, but because it does not belong to the same category as exposures reaching OR 3. Setting the audit line here clearly delineates the structural gap between “OR 4–5 required for fate assertions” and “OR 1.2–1.3 provided by birth time”—a gap determined by exposure type.

Evidence and Limitations

Verified: OR ≥ 3 in genetics belongs to the “Mendelian moderate effect” tier, typified by single-gene rare variants (PMID 25741868, 35396579); effect size ceilings for birth month/season associations are approximately OR 1.2–1.3, supported by multiple large-scale studies within the OR 1.06–1.24 range (PMID 27123493, 27366989, 23172784, 38776272); SeaWAS identified only 55 significant associations in 17.49 million people across 1,688 diseases after FDR correction and replicated them in an independent cohort of 11.69 million people (PMID 26041386, 27624541); PheWAS studies are sensitive to biases in electronic medical records and rely on independent replication (PMID 24270849, 25474351).

Partially Verified: A Chinese single-center study reported birth month ORs of 1.33–1.39 (PMID 31122721), representing the highest tier in literature, but was designed as a single-center retrospective study without full-outcome multiple testing correction and without external replication, showing inconsistent peak directions compared to multicenter studies; common variant effect size distribution research suggests a general pattern of “smaller effects occur more frequently” (PMID 22003128), which is extrapolatable but not direct evidence.

To Be Verified: ① Full-phenome association scans at hour granularity (rather than month) have never been conducted—existing electronic medical records typically only contain birth dates, representing the largest gap highlighted here; ② If performed, whether the number of significant findings after FDR correction and the hit rate for OR ≥ 3 fall within the predicted intervals in this article requires empirical testing; ③ Whether the magnitude distribution of birth time effects is consistent across non-European and non-North American populations remains unassessed systematically.

Related Reading

Evidence Level: B
References: PMID 25741868, PMID 35396579, PMID 26041386, PMID 27624541, PMID 27123493, PMID 27366989, PMID 23172784, PMID 38776272, PMID 31122721, PMID 24270849, PMID 25474351, PMID 29501066, PMID 22003128, PMID 15277620


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 *