五运六气预测模型与现代流行病学模型:谁更准?

Written by

in

Language

Evidence level: B

五运六气的价值不在于取代现代预测模型,而在于它提供了一种”气候先兆 → 滞后发病”的框架,与现代 DLNM 等模型的底层逻辑不谋而合。

正文

把五运六气当”天气预报 + 疾病预报”来看,是现代人最容易产生的误解。它的本质是《黄帝内经》建立的一套”干支纪年 → 气候节律 → 人群易感”的周期推演系统:把一年分为六气(每气约两个月),把六十年分为五运轮转,主张异常气候出现后大约三年可能酿成疫病——这就是”三年化疫”说。问题来了:这套古代模型和今天靠大数据训练出来的流行病学预测模型,到底哪个更准?答案比想象中更有意思:两者不是一个赛道,但底层逻辑惊人地相似。

先看现代模型怎么预测。今天的传染病预测主流是分布式滞后非线性模型(DLNM),它承认两件事:气候对疾病的影响是非线性的(不是越冷越危险,而是存在最适区间),并且是滞后的(今天的天气要隔几天甚至几周才反映到发病曲线上)。一篇对巴巴多斯登革热的研究用 DLNM 建模,把干旱指数和最低气温纳入暴露-滞后-反应函数,对暴发的正确识别率达到 86%,明显高于传统监测的 64%(PMID 30016319)。另一篇对中国重庆流感的研究同样发现,多种气象因子(温度、湿度、气压、风速)与流感活动存在强关联,且存在不同滞后天数(PMID 33534840)。国内团队还发展出概率型流感早期预警模型,直接拟合流感样病例阳性率来做前瞻预警(PMID 40769217)。这套体系的共同点是:用几十年的大样本数据,让模型自己”学习”气候-疾病的关系形状。

再看五运六气模型做了什么。中医界其实早就做过定量尝试。一篇对北京地区 35 年流感发病率的回顾性研究,把月发病数据与同期及前 1-4 年的气象数据放进反向传播神经网络,建立了六个”流感-气象”预测模型,其中第五号模型(纳入前三年气候数据加当年数据)预测准确率最高,达到 49%,并且发现流感高峰出现在第六气(12-1 月的干冷时段)——这恰好对应《内经》”疫病发生在异常气候三年后”的论断(该研究发表于《Journal of Traditional Chinese Medical Sciences》2018,DOI 10.1016/j.jtcms.2018.06.003)。类似的还有北京结核病 15 年研究(PMID 26525546)和过敏性鼻炎-气象非线性回归研究(PMID 27329149),结论一致:当年及前几年的气象因子对疾病发病有统计显著的预测力。此外,2024 年一篇泛癌研究用五运六气定义的先天体质对肿瘤做分子分型,发现不同运气体质在无进展生存期、免疫细胞丰度、药物敏感性上均有显著差异(PMID 39041169),提示运气框架甚至能延伸到个体化医学。

把这些放在一起看,会得出三个关键结论:

第一,五运六气的”滞后效应”被现代数据证实。 DLNM 的滞后思想(暴露后数天到数月的延迟发病)与”三年化疫”是同一逻辑的不同时间尺度。现代模型在周尺度上验证了滞后,而北京流感研究在年级尺度上部分验证了”三年化疫”——两者互补而非矛盾。

第二,预测精度差距真实存在。 五运六气模型的最高准确率(49%)远低于现代模型的 86%,原因很直白:古代模型只有干支、五行、六气这几个离散变量,而现代模型可以吸纳温度、湿度、气压、风速、降水、紫外线、人群流动等几十个连续变量。这不是五运六气”错”,而是它的输入信息量被锁死在古代天文历法里。

第三,真正的价值在”先兆”而非”精度”。 五运六气的不可替代性在于:它能在疫情发生前,仅凭干支推演给出”哪一类气候异常会在未来催生哪一类疫病”的定性预警,为资源调配争取时间窗口。2020 年新冠期间,就有学者指出 2017 丁酉年”柔干失刚”对应”三年化疫”的伏燥先兆。这种”提前数年”的尺度,是现代模型(通常只做 2-8 周的短期预报)覆盖不到的。理想的架构不是二选一,而是干支粗筛 + 大数据精算:用运气框架做长期定性预警,用 DLNM 做短期定量预报。

证据与局限

已验证: 五运六气框架下的流感/结核病预测模型确实有效(北京 35 年流感研究最高 49% 准确率,PMID 26525546 结核病、27329149 鼻炎均达统计显著);气候-疾病关联存在非线性与滞后性已被 DLNM 研究广泛证实(PMID 30016319、33534840);运气体质在肿瘤分子分型上存在组学差异(PMID 39041169)。

部分验证: “三年化疫”的三年滞后尺度仅有北京流感研究的间接支持,尚无多中心、跨疾病的系统验证;五运六气模型的预测能力主要停留在回顾性拟合,前瞻性验证很少。

待验证: 五运六气 + DLNM 混合模型能否在相同数据集上显著优于纯现代模型;干支变量能否为现代模型提供增量预测价值;”三年化疫”的机制(气候-免疫-病原协同)在分子层面尚未阐明。

相关阅读

证据等级: B
参考文献: PMID 39041169, PMID 26525546, PMID 27329149, PMID 42104306, PMID 30016319, PMID 33534840


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。

Five Elements and Six Qi Prediction Model vs Modern Epidemiological Models: Who Is More Accurate?

Evidence level: B

The value of Five Phases and Six Seasons (Wu Yun Liu Qi) does not lie in replacing modern predictive models, but in providing a framework of “climatic precursors → delayed onset,” which aligns fundamentally with the underlying logic of modern Distributed Lag Non-linear Models (DLNM).

Main Text

Viewing Five Phases and Six Seasons as a “weather forecast + disease forecast” is the most common misconception among modern readers. Its essence is a cyclical deduction system established in the Huangdi Neijing linking “Stem-Branch years → climatic rhythms → population susceptibility”: it divides the year into six Qi (each approximately two months), cycles the Five Phases over sixty years, and posits that abnormal climates may precipitate epidemics approximately three years later—the so-called “Three Years into Epidemic” theory. The question arises: compared to today’s epidemiological prediction models trained by big data, which is more accurate? The answer is more interesting than imagined: they are not on the same track, yet their underlying logic is strikingly similar.

Let us first examine how modern models predict. The mainstream approach for infectious disease prediction today is the Distributed Lag Non-linear Model (DLNM). It acknowledges two facts: the impact of climate on disease is non-linear (not simply “colder is worse,” but rather exists within an optimal range) and is lagged (today’s weather reflects on the incidence curve only after several days or weeks). A study on dengue fever in Barbados using DLNM modeling incorporated drought indices and minimum temperatures into an exposure-lag-response function, achieving a correct identification rate of 86% for outbreaks, significantly higher than the traditional monitoring rate of 64% (PMID 30016319). Another study on influenza in Chongqing, China, similarly found strong associations between multiple meteorological factors (temperature, humidity, atmospheric pressure, wind speed) and influenza activity, with varying lag days (PMID 33534840). Domestic teams have also developed probabilistic early warning models for influenza that directly fit positive rates of influenza-like cases for prospective warnings (PMID 40769217). The commonality of this system is that it uses decades of large-sample data to allow the model to “learn” the shape of the climate-disease relationship.

Now let us examine what the Five Phases and Six Seasons model accomplishes. The TCM community has long attempted quantification. A retrospective study reviewing 35 years of influenza incidence rates in Beijing incorporated monthly incidence data along with meteorological data from the same year and the preceding 1–4 years into a backpropagation neural network. It established six “influenza-meteorology” prediction models; Model No. 5 (incorporating climate data from the previous three years plus the current year) achieved the highest prediction accuracy at 49%. Furthermore, it found that influenza peaks occurred during the Sixth Qi (the dry-cold period of December–January), which precisely corresponds to the Neijing assertion that “epidemics occur three years after abnormal climates” (published in Journal of Traditional Chinese Medical Sciences, 2018, DOI 10.1016/j.jtcms.2018.06.003). Similar findings include a 15-year study on tuberculosis in Beijing (PMID 26525546) and a nonlinear regression study on allergic rhinitis and meteorology (PMID 27329149), both concluding that meteorological factors from the current year and previous years possess statistically significant predictive power for disease incidence. Additionally, a 2024 pan-cancer study defined individual constitutions based on Five Phases and Six Seasons to perform molecular subtyping for tumors, finding significant differences in progression-free survival, immune cell abundance, and drug sensitivity across different constitution types (PMID 39041169), suggesting the framework can even extend to personalized medicine.

Viewing these together yields three key conclusions:

First, the “lag effect” of Five Phases and Six Seasons is confirmed by modern data. The lag concept in DLNM (delayed onset from days to months after exposure) and the “Three Years into Epidemic” theory represent the same logic at different time scales. Modern models validate lag at a weekly scale, while the Beijing influenza study partially validates “Three Years into Epidemic” at a yearly scale—complementary rather than contradictory.

Second, a genuine gap in prediction accuracy exists. The highest accuracy of the Five Phases and Six Seasons model (49%) is far lower than that of modern models (86%). The reason is straightforward: the ancient model relies only on discrete variables such as Stem-Branch, Five Phases, and Six Qi, whereas modern models can incorporate dozens of continuous variables including temperature, humidity, atmospheric pressure, wind speed, precipitation, UV radiation, and human mobility. This is not an error in Five Phases and Six Seasons; rather, its input information volume is locked within ancient astronomical calendars.

Third, true value lies in “precursors” rather than “accuracy.” The irreplaceability of Five Phases and Six Seasons lies in its ability to provide qualitative warnings based on Stem-Branch deduction before an epidemic occurs, identifying “which type of climatic abnormality will foster which type of epidemic in the future,” thereby争取ing time windows for resource allocation. During the 2020 COVID-19 pandemic, scholars pointed out that the “soft stem losing rigidity” in the 2017 Ding You year corresponded to the latent dryness precursor of “Three Years into Epidemic.” This scale of “several years in advance” exceeds the coverage of modern models (which typically provide short-term forecasts of 2–8 weeks). The ideal architecture is not a binary choice but Stem-Branch coarse screening + Big Data precise calculation: using the luck framework for long-term qualitative warnings and DLNM for short-term quantitative forecasting.

Evidence and Limitations

Verified: Predictive models for influenza and tuberculosis within the Five Phases and Six Seasons framework are indeed effective (highest 49% accuracy in the Beijing 35-year influenza study; statistical significance reached for tuberculosis PMID 26525546 and rhinitis PMID 27329149); the non-linear and lagged nature of climate-disease associations has been widely confirmed by DLNM research (PMID 30016319, 33534840); constitution types based on Five Phases show genomic differences in tumor molecular subtyping (PMID 39041169).

Partially Verified: The three-year lag scale of “Three Years into Epidemic” has only indirect support from the Beijing influenza study; there is no multi-center, cross-disease systematic verification yet; the predictive capability of the Five Phases and Six Seasons model remains largely at retrospective fitting, with few prospective validations.

To Be Verified: Whether a mixed model combining Five Phases and Six Seasons with DLNM significantly outperforms pure modern models on the same dataset; whether Stem-Branch variables can provide incremental predictive value for modern models; the mechanism (climate-immune-pathogen synergy) of “Three Years into Epidemic” remains unelucidated at the molecular level.

Related Reading

Evidence Level: B
References: PMID 39041169, PMID 26525546, PMID 27329149, PMID 42104306, PMID 30016319, PMID 33534840


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 *