Evidence level: C
一个参数值不值得当终点,不看它「能不能变」,而看它「变了之后能不能和抖动分开」——而这条线,比大多数人以为的要高。
正文
一、MDD 回答的问题,和它不回答的问题
MDD(最小可检测差异)是从 SEM(测量标准误)推出来的:MDC95 ≈ 1.96 × √2 × SEM ≈ 2.77 × SEM。它只回答一件事——这个数变了,是不是仪器自己在抖。它完全不管这个变化有没有生理意义。
问题在于,MDD 这个数本身的质量没人担保。2025 年一项系统综述纳入 49 项功能测试的信度研究,报告了 ICC、SEM、MDC,但用 COSMIN 清单评估后全部被评为「不合格」(PMID 40895367)。也就是说,文献里那些看起来很精确的 MDD 数字,方法学底子普遍是虚的。
二、CoV 尤其不适合直接套 MDD
CoV = SD / 均值,是个比值。它的抽样分布依赖均值大小:均值越接近零,CoV 越爆炸;同一个绝对变异,在低均值个体身上会被算得更大。这意味着「CoV 的 MDD」不是一个跨人群通用的常数——它随基线水平漂移。
更麻烦的是组间比较。CoV 不能像均值那样直接做 t 检验,必须走对数变换(PMID 10588295)。所以「修炼者 CoV 的 MDD」和「新手 CoV 的 MDD」本身就是两个不同的数,拿同一个 MDD 去卡两组,等于用一把会伸缩的尺子量两样东西。
三、真正成熟的框架是 RCV,不是 MDD
临床检验早有更完整的解法:参考变化值 RCV = √2 · Z · √(CV_A² + CV_I²),把分析变异(CV_A)和个体内生物学变异(CV_I)分开算(PMID 21958344)。关键在于 CV_I 必须单独估,而且它对不同指标差一个数量级。
实测数据很直白:同一批慢性肾病患者、同一时间窗内,肾小球滤过标记物的个体内 CV 只有 5.4%–7.4%,而白蛋白尿的 CV 高达 29.7%–50.6%(PMID 30031564)。同样的人、同样的重复测量设计,不同指标的「最小可辨变化」差十倍。这直接否掉了「给所有参数定一个统一 MDD」的做法——SnR 门槛必须逐参数定。
四、SnR > 2 这个数字,出处是空的
「效应量要大于 MDD 的 2 倍」听起来像行业惯例,但它没有生理学依据。一项针对 380 位 RCT 通讯作者的调查显示:选「要检测多大差值」根本没有指南,中位选择是疼痛控制改善 7.0%、死亡率或复发降低 5.0%,而且随基线风险显著变化(PMID 27473336)。连「要检测多大」都是主观拍定的,「2 倍」自然只是拍定值的拍定值。
更糟的是预期效应量系统性偏高。194 项 RCT 的分析发现:阴性结果组预期效应量比实测高 42%(95%CI 32–51),post hoc 检验力只有 20.2%(PMID 26041717)。也就是说,用预实验效应量去除 MDD 得到的 SnR,分子天生偏大。
五、个体反应异质性会让平均效应量失真
就算 SnR 算出来达标,也要小心它掩盖了什么。对最大摄氧量训练反应的批判性综述指出:个体间反应差异极大,存在明确的「高反应者」与「无反应者」(PMID 28097487)。用一个平均效应量去除以 MDD,得到的 SnR 描述的是一个不存在的「平均人」。
呼吸侧的信度数据也提醒同一件事:可穿戴设备测呼吸率,静息 ICC 0.76–0.84(PMID 30694966),看起来很好;但膈肌功能的多指标评估里,峰值类指标 ICC 0.70–0.94,斜率类指标 ICC ≤ 0.20(PMID 41706258)。同一台设备、同一批人,指标形态就决定了信度。若 CoV 的算法里带拟合斜率,直接报废。
六、技术噪声地板有个参照量级
34 个实验室、19 个国家测同一批标准物质:实验室内 CV ≤ 4.2%,实验室间 CV < 14%(PMID 39362843)。这是「纯技术噪声」的量级参照。另外,MRI 换机后重测信度与样本量估计都必须重算(PMID 32032737)——换中心就要重算 MDD,没有继承这回事。
反转:SnR 卡的不是参数,是设计
如果某参数效应量只有 MDD 的 1.5 倍,正确反应不是「换个参数」,而是增加重复测量次数——SEM 按 √n 下降,MDD 跟着降。把 SnR > 2 写成不带重复次数的固定门槛,等于用淘汰参数来掩盖设计不足。四个格子的判法:全部候选 SnR < 2 ⇒ 这个试验不该做;仅 CoV 达标而 %AB 不达标 ⇒ 疑 CoV 的 MDD 被低估;全部达标但个体异质性大 ⇒ 终点改「响应者比例」;达标但跨中心重算后掉到 2 以下 ⇒ 中心效应吃掉效应,先统一协议。
证据与局限
已验证: RCV 框架的数学形式与 CV_A/CV_I 分离原则(PMID 21958344);个体内 CV 的指标间数量级差异(PMID 30031564);MDC 与 SEM 的换算关系及功能测试 MDC 的广泛报告(PMID 18356292, PMID 40895367);呼吸率与膈肌功能指标的 ICC 分层(PMID 30694966, PMID 41706258);实验室间 CV 基准(PMID 39362843);换设备后信度与样本量需重估(PMID 32032737)。
部分验证: SnR > 2 作为终点入选门槛在方法学上合理但无统一出处(PMID 27473336 显示「要检测多大」本身无指南);预期效应量系统性高估已有明确证据(PMID 26041717),但高估幅度是否适用于练功类干预未知;CoV 的组间比较须对数变换(PMID 10588295)在呼吸参数上未直接验证。
待验证: 丹田呼吸场景下 CoV 的个体内 CV 究竟是多少(无任何直接研究);修炼者与新手 CoV 的 MDD 是否真的不同;「气沉」效应量在不同重复次数下的 SnR 曲线;响应者比例终点是否比均值终点更敏感。
相关阅读
证据等级: C
参考文献: PMID 40895367, PMID 21958344, PMID 30031564, PMID 26041717, PMID 28097487, PMID 10588295, PMID 18356292, PMID 30694966, PMID 41706258, PMID 39362843, PMID 32032737, PMID 27473336L1 网络状态:「该主题尚未进入 L1 对应网络」——CORRESPONDENCE_NETWORK.md 节点表只把「相位 / 节律」列为物理侧节点(备注「周期性耦合的关键」)、把「呼吸率 / 呼吸模式」列为生理侧观测项(备注「与节律耦合最直接」),二者均未与易学侧任何节点连成边;节点表中完全没有「变异系数 / 最小可检测差异 / 信噪比 / 测量误差」这类测量学维度节点,边表 3.1–3.4 只登记五行↔五脏/五色/五音/五方/五季/五志、八卦↔五行、天干地支↔五行,无任何以丹田参数、CoV、MDD 或终点选择为源或目标的边。EDGE_TABLE_science-yixue.md 仅有概念计数(丹田 51 篇 / 气功 48 篇 / 呼吸 189 篇),无层级(L1/L2/L3)与状态(active/open/conflict)判定。故本文封顶 C。
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。
MDD Is Not \”Minimum Useful Difference\” but \”Minimum Non-Noise Difference\”: How to Set the Signal-to-Noise Ratio Threshold for Dantian CoV Effect Sizes
Evidence level: C
A parameter value is not worth serving as an endpoint based on whether it “can change,” but rather on whether it “can be distinguished from noise after changing”—and this threshold is higher than most assume.
Main Text
I. The Question MDD Answers and the Questions It Does Not Answer
MDD (Minimum Detectable Difference) is derived from SEM (Standard Error of Measurement): MDC95 ≈ 1.96 × √2 × SEM ≈ 2.77 × SEM. It addresses only one thing: if this number changes, is the instrument itself fluctuating? It completely ignores whether the change has physiological significance.
The issue lies in the quality of the MDD number itself, which is not may support. A 2025 systematic review including 49 reliability studies from functional tests reported ICC, SEM, and MDC, but after assessment using the COSMIN checklist, all were rated “inadequate” (PMID 40895367). In other words, the seemingly precise MDD numbers in the literature generally rest on a weak methodological foundation.
II. CoV Is Especially Unsuitable for Direct Application of MDD
CoV = SD / Mean is a ratio. Its sampling distribution depends on the mean magnitude: as the mean approaches zero, CoV explodes; the same absolute variation is calculated as larger in individuals with lower baseline means. This means the “CoV MDD” is not a universal constant across populations—it drifts with baseline levels.
Complications arise in between-group comparisons. CoV cannot undergo a t-test like the mean; it requires a logarithmic transformation (PMID 10588295). Therefore, the “MDD for CoV in practitioners” and the “MDD for CoV in novices” are fundamentally different numbers. Applying a single MDD to compare two groups is equivalent to using a stretching ruler to measure two different things.
III. The Mature Framework Is RCV, Not MDD
Clinical diagnostics already have a more complete solution: the Reference Change Value (RCV) = √2 · Z · √(CV_A² + CV_I²), separating analytical variation (CV_A) from individual biological variation (CV_I) (PMID 21958344). The key is that CV_I must be estimated separately, and it differs by an order of magnitude across different indicators.
The empirical data is straightforward: in the same cohort of chronic kidney disease patients within the same time window, the individual CV for glomerular filtration markers was only 5.4%–7.4%, whereas the CV for albuminuria reached 29.7%–50.6% (PMID 30031564). The same people, with the same repeated measurement design, show a “minimum discernible change” differing tenfold across different indicators. This directly invalidates the practice of setting a unified MDD for all parameters—the Signal-to-Noise Ratio (SnR) threshold must be determined per parameter.
IV. The SnR > 2 Figure Has No Source
The statement “effect size must be greater than twice the MDD” sounds like industry convention, but it lacks physiological basis. A survey of 380 RCT correspondence authors found: there is no guideline on “how large a difference to detect,” with the median selection being a 7.0% improvement in pain control or a 5.0% reduction in mortality or recurrence, varying significantly with baseline risk (PMID 27473336). Even “how large to detect” is subjectively set; thus, “twice” is merely the square of a subjective setting.
Worse still, expected effect sizes are systematically overestimated. Analysis of 194 RCTs found that expected effect sizes in negative-result groups were 42% higher than observed (95%CI 32–51), with post hoc power only at 20.2% (PMID 26041717). In other words, the SnR derived by dividing the pre-experimental effect size by MDD has a molecularly inflated numerator.
V. Individual Response Heterogeneity Distorts Average Effect Sizes
Even if SnR calculates as meeting the standard, one must be careful what it masks. A critical review of responses to VO2 max training indicates extreme inter-individual variability, with clear “high responders” and “non-responders” (PMID 28097487). Using an average effect size divided by MDD describes a non-existent “average person.”
Reliability data from respiratory side measurements also warns of this: wearable devices measuring respiratory rate show resting ICC 0.76–0.84 (PMID 30694966), which appears good; however, in multi-indicator assessments of diaphragm function, peak-type indicators show ICC 0.70–0.94, while slope-type indicators show ICC ≤ 0.20 (PMID 41706258). The same device and same cohort yield different reliability based on indicator morphology. If the CoV algorithm includes fitted slopes, it should be discarded immediately.
VI. Technical Noise Floors Have a Reference Magnitude
34 laboratories across 19 countries measured the same batch of standard substances: intra-laboratory CV ≤ 4.2%, inter-laboratory CV < 14% (PMID 39362843). This is the reference magnitude for “pure technical noise.” Additionally, re-test reliability and sample size estimation after MRI scanner changes must be recalculated (PMID 32032737)—changing centers requires recalculating MDD; there is no inheritance.
Reversal: SnR Filters Design, Not Parameters
If a parameter’s effect size is only 1.5 times the MDD, the correct response is not “change the parameter,” but increase the number of repeated measurements—SEM decreases by √n, and MDD follows. Writing SnR > 2 as a fixed threshold without repeated counts equals using the elimination of parameters to mask design deficiencies. The four-quadrant judgment method: all candidate SnR < 2 ⇒ the trial should not proceed; only CoV meets standard while %AB fails ⇒ suspect CoV MDD is underestimated; all meet standards but high individual heterogeneity ⇒ change endpoint to “proportion of responders”; meets standards but drops below 2 after cross-center recalculation ⇒ center effects consume the effect, first unify protocols.
Evidence and Limitations
Verified: Mathematical form of the RCV framework and the separation principle of CV_A/CV_I (PMID 21958344); order-of-magnitude differences in individual CV across indicators (PMID 30031564); conversion relationship between MDC and SEM and widespread reporting of functional test MDC (PMID 18356292, PMID 40895367); stratification of ICC for respiratory rate and diaphragm function indicators (PMID 30694966, PMID 41706258); inter-laboratory CV benchmarks (PMID 39362843); need for re-estimation of reliability and sample size after equipment changes (PMID 32032737).
Partially Verified: SnR > 2 as an endpoint inclusion threshold is methodologically reasonable but lacks a unified source (PMID 27473336 shows no guideline on “how large to detect”); systematic overestimation of expected effect sizes has clear evidence (PMID 26041717), but whether the overestimation magnitude applies to training interventions is unknown; between-group comparison of CoV requires logarithmic transformation (PMID 10588295), not directly verified for respiratory parameters.
To Be Verified: What the individual CV for CoV is specifically in the Dantian breathing scenario (no direct research exists); whether MDD for CoV truly differs between practitioners and novices; the SnR curve of “Qi Sinking” effect size across different repetition counts; whether a proportion-of-responder endpoint is more sensitive than a mean endpoint.
Related Reading
- 「Three Modes All Changing in One Direction」: Actually No One Verified This
- Is There a Limit to Qi Sinking in the Dantian? EIT Ventilation Center of Gravity Dose-Response Curve by Practice Duration
- Device Noise Must Be Suppressed Below Effect Size
- Environment-Life Frequency Interface (P0 Pool)
Evidence Level: C
References: PMID 40895367, PMID 21958344, PMID 30031564, PMID 26041717, PMID 28097487, PMID 10588295, PMID 18356292, PMID 30694966, PMID 41706258, PMID 39362843, PMID 32032737, PMID 27473336L1 Network Status: “This topic has not entered the L1 corresponding network”—the CORRESPONDENCE_NETWORK.md node table lists only “Phase/Rhythm” as a physical-side node (note: “key for periodic coupling”) and “Respiratory Rate/Pattern” as a physiological-side observation item (note: “most directly coupled with rhythm”), but neither connects to any node on the Yijing side; there are no nodes in the node table with measurement dimensions such as “Coefficient of Variation/Minimum Detectable Difference/Signal-to-Noise Ratio/Measurement Error,” and Edge Tables 3.1–3.4 only register Five Phases ↔ Five Organs/Five Colors/Five Tones/Five Directions/Five Seasons/Five Emotions, Eight Trigrams ↔ Five Phases, Heavenly Stems and Earthly Branches ↔ Five Phases, with no edges sourced or targeted by Dantian parameters, CoV, MDD, or endpoint selection. EDGE_TABLE_science-yixue.md only contains concept counts (Dantian 51 articles / Qigong 48 articles / Breathing 189 articles), with no hierarchy (L1/L2/L3) or status (active/open/conflict) determination. Therefore, this article is capped at Level C.
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.
Leave a Reply