Evidence level: C
前一篇把「练与不练一个样」的闸门挂在「Δφ 的 MDC95 与等效界谁大」上。可 MDC 是算术公式,φ 是角度。脚踩在 0° 两侧时公式先坏,坏的方向恰是把同步最好的人判成不可测。
正文
第一层,前提不成立。 MDC95 的标准写法是 1.96×√2×SEM,SEM 由测量值标准差与重测信度推出(PMID 18569858 示范:髋外展肌力 ICC 0.96、SEM 2 牛顿、MDC95 折成体重 5.4%)。公式有个不言自明的前提:被测量能在数轴上直接加减。胸腹相位角 φ 不满足。φ 是罗盘上的一根指针,0° 两边不是「正」与「负」,是同一方向的左右肩。
第二层,跨零让变异凭空变大。 一个人胸腹几乎同步,φ 在 +2° 与 −2° 之间抖。生理上这是最稳的状态,算术上差值算成 4°,按角度只有 2°。抖动没变,标准差翻倍,MDC95 跟着虚高。结果是同步越好的人越容易被算成测不出来。这不是随机噪声,是有方向的选择性偏倚:它淘汰的不是脏数据,恰恰是最干净的那批。
第三层,还有个硬天花板。 Lissajous 图法算出的相位角天生关在 ±90° 里。一项对 2290 个呼吸周期的分析给出实测范围:相位角 −86.90° 到 +88.4°,而同一数据集改用「全局相位延迟」可到 −179.75° 至 +178.54°,且相位角与环面积都无法正确分类严重偏离与矛盾运动的呼吸(PMID 38829281)。超过 90° 那一段,也就是反相与矛盾运动、临床上最要紧的那一段,被折回去了。在截断的尺子上谈最小可检测变化,数学上不成立。
第四层,圆形口径自己也不干净。 换 1−R、圆方差、von Mises κ 不自动解决问题。平均合成向量长度(即相位锁定值、Kuramoto 序参量)的期望值随样本量变化:只有样本趋于无穷才趋近 0,于是不同样本量之间比较会引入偏差,须重归一化才可比(PMID 37756609)。落到实操就一句:同一个人的可检测性,取决于你数了 20 个呼吸还是 200 个。
第五层,稳定性指标有前提,前提常不成立。 用相对相位 SD 当稳定性,需两段振荡锁定在某值、且围着固定点做单峰波动。骨盆-胸廓研究用圆形统计专门检验,发现步态里两条都不满足,作者因此否定标准相对相位指标,改用连续傅里叶相位差(PMID 12297252)。胸腹 φ 是同类量,新手本就不锁定在固定相位,前提只会更松。
第六层,口径不同判定就不同,是实测过的。 同一批 COPD 病人、同一份数据,用相位角判出 15 例胸腹不同步,用相位偏移只判出 7 例;运动时 14 对 8(PMID 28183817)。不是差几个百分点,是差一倍。
第七层,反方写在明处。 有研究直接比较呼吸感应体积描记与光电体积描记测的相位角,两法数值非常接近,静息与不同运动强度均无统计学差异,Bland-Altman 落在一致性界限内(PMID 28767680)。这条很关键:仪器不是主要问题,问题在统计口径。所以处理方式是重算,不是重做实验。
第七层半,加样本量救不了。 这个偏差不是噪声,是系统性的:n 从 90 加到 900,跨零的折叠效应一分不减,因为每个人都在 0° 附近被折。唯一能缓解的是换个度量方式。同一个量在个体层与组层算出来的可检测水平本就差一个数量级(SDCind 10.10 对 SDCgroup 0.86,PMID 28758322),口径错误加上层级别混用,会把这两个数字搅成一团。
第八层,怎么测。 纯模拟:Monte Carlo 合成已知 κ 的 φ 序列,均值 μ 从 0° 扫到 ±60°,分含跨零与不含跨零,每条件 20/50/200 次呼吸,按两态相关 r 分 0.5/0.7/0.9/0.97 四档生成配对样本;分别用算术差值 SD 与圆形离散度推 MDC95,比两者对真实 κ 的恢复偏差。真实数据:≥90 人静息 RIP 记录,同体位同带位,双口径重算,统计同一批人里被判「差值落在等效界内」的比例差多少。
第九层,四格判据。 ① μ→0° 且 r≥0.9 的格子里算术 MDC95 高估超 30% 并随 μ 减小单调放大 ⇒ 算术判定在同步最好的人群中系统性失效,凡未报 φ 均值的 MDC 结论不可用于两态判读。② 两口径对同一批人的判定不一致率 >20%(哪怕只集中 μ<15° 子组)⇒ 闸门改双口径并列,加「不一致即判不可检验」。③ 偏差 <10% 且一致率 >90% ⇒ 算术 MDC 可沿用,本文担忧撤回。
证据与局限
已验证: 相位角的 Lissajous 算法存在 ±90° 硬性取值范围,实测 2290 个周期跨度 −86.90° 至 +88.4°,同数据集全局相位延迟可覆盖 ±179° 区间(PMID 38829281);相位角与相位偏移对同一批数据的分类可差一倍(15 对 7、14 对 8,PMID 28183817);平均合成向量长度存在样本量依赖的系统偏差(PMID 37756609);相对相位 SD 的锁定与单峰前提在真实数据中可被证伪(PMID 12297252)。
部分验证: 两种体积描记技术的相位角一致性良好(PMID 28767680),提示仪器非误差主因;但该结论仅覆盖静息与运动、样本 27 人,不能外推到长时间静息或极低幅度胸腹运动。个体级与组级可检测变化可差一个数量级(SDCind 10.10 对 SDCgroup 0.86,PMID 28758322),这一量级差在角度量上是否同样成立尚无实测。
待验证: 算术 MDC95 在 φ 跨零时的高估幅度(30% 还是 300%)无任何实测;圆形口径的 MDC 是否真对 κ 恢复更好,只有理论推导与模拟空间,缺人体数据对照;「最少记录多少个呼吸才能让两口径收敛」未标定。相位估计精度本身还受采样设计影响(非等间隔采样显著改变相位参数方差,PMID 40952326),而现有胸腹研究几乎不报采样间隔与呼吸次数,这层混杂尚未拆开。
相关阅读
证据等级: C
参考文献: PMID 38829281, PMID 28183817, PMID 37756609, PMID 12297252, PMID 28767680, PMID 28758322, PMID 18569858, PMID 40952326, PMID 29991414
L1 网络状态:
CORRESPONDENCE_NETWORK.md节点表仅在物理侧登记「相位 / 节律」一项(备注原文「周期性耦合的关键」),生理侧八项(HRV、皮温、呼吸率、EMG、EEG、胃肠蠕动、气感报告、体质分型)全为横断面观测项,三域均无「圆形离散度 / von Mises κ / 最小可检测变化 / 跨零 / 测量误差带」这类测量学节点;EDGE_TABLE_science-yixue.md为概念反抽统计(文章 774 篇 / 唯一 PMID 3542),grep「相位 / MDC / 圆形 / 测量误差 / 信度 / 等效 / κ」仅命中节点表同名行,无 L1/L2/L3 层级与 active/open/conflict 状态判定;边表止于 E-050(层级 L1、状态 open,现代证据栏原文「尚无任何可测来源」)。该主题尚未进入 L1 对应网络,故本文证据等级封顶为 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。
Difference MDC Cannot Directly Judge Circularity: At the Moment Phi Crosses Zero, Arithmetic Formulas First Misclassify the Best Performer as \”Undetectable\
Evidence level: C
The previous article hung the gate on whether “practice yields no different from non-practice” based on whether the MDC95 of Δφ is larger than the equivalence bound. However, MDC is an arithmetic formula, while φ is an angle. When stepping on either side of 0°, the formula breaks first; the direction of failure precisely misclassifies those with the best synchronization as unmeasurable.
Main Text
First Layer, the premise is invalid. The standard notation for MDC95 is $1.96 \times \sqrt{2} \times \text{SEM}$, where SEM is derived from the standard deviation of measurements and test-retest reliability (demonstrated in PMID 18569858: hip abduction muscle strength ICC 0.96, SEM 2 Newtons, MDC95 converted to 5.4% of body weight). The formula has an obvious, unstated premise: the measured variable can be directly added or subtracted on a number line. The thoraco-abdominal phase angle φ does not satisfy this. φ is a pointer on a compass; the two sides of 0° are not “positive” and “negative,” but rather left and right shoulders of the same direction.
Second Layer, crossing zero inflates variability. When an individual’s thoraco-abdominal motion is nearly synchronous, φ fluctuates between +2° and −2°. Physiologically, this is the most stable state; arithmetically, the difference calculates to 4°, whereas on an angular scale it is only 2°. The fluctuation does not change, but the standard deviation doubles, causing MDC95 to become artificially inflated. The result is that those who are more synchronized are more likely to be calculated as unmeasurable. This is not random noise, but a directional selective bias: it eliminates not dirty data, but precisely the cleanest batch.
Third Layer, there is also a hard ceiling. The Lissajous method inherently confines phase angles within ±90°. An analysis of 2,290 breathing cycles yielded a measured range: −86.90° to +88.4°, whereas using “global phase delay” on the same dataset extended the range to −179.75° to +178.54°, and phase angles failed to correctly classify severe deviations and contradictory breathing motions compared to ring area (PMID 38829281). The segment beyond 90°, representing anti-phase and contradictory motion—the most clinically critical segment—is folded back. Discussing the minimum detectable change on a truncated ruler is mathematically invalid.
Fourth Layer, the circular metric itself is not clean. Switching to $1-R$, circular variance, or von Mises κ does not automatically resolve the issue. The expected value of the mean resultant vector length (i.e., phase locking value, Kuramoto order parameter) varies with sample size: it only approaches 0 as the sample size tends to infinity. Consequently, comparing across different sample sizes introduces bias requiring re-normalization for comparability (PMID 37756609). In practice, the sentence is simple: an individual’s detectability depends on whether you counted 20 or 200 breaths.
Fifth Layer, stability indicators have premises that often fail. Using relative phase SD as a stability metric requires two oscillations locked at specific values and exhibiting unimodal fluctuations around fixed points. A pelvic-thoracic study using circular statistics specifically for verification found that neither condition is met in gait; thus, authors rejected standard relative phase indicators in favor of continuous Fourier phase difference (PMID 12297252). Thoraco-abdominal φ is a similar quantity; novices are not locked at fixed phases to begin with, making the premise even looser.
Sixth Layer, different metrics yield different judgments, verified empirically. With the same cohort of COPD patients and the same dataset, phase angles identified 15 cases of thoraco-abdominal asynchrony, while phase offset identified only 7; during exercise, the counts were 14 versus 8 (PMID 28183817). It is not a difference of a few percentage points, but a factor of two.
Seventh Layer, the counter-argument is explicit. Some studies directly compared phase angles measured by respiratory inductive plethysmography versus photoplethysmography; the values were very close, with no statistical differences at rest or across different exercise intensities, and Bland-Altman plots fell within agreement limits (PMID 28767680). This is critical: the instrument is not the primary issue; the problem lies in the statistical metric. Therefore, the solution is recalculation, not re-experimentation.
Seventh Layer and a Half, increasing sample size does not save it. This bias is not noise but systematic: increasing n from 90 to 900 does not reduce the folding effect at zero crossing because every individual is folded near 0°. The only mitigation is changing the metric. The detectability level calculated at the individual versus group level for the same quantity differs by an order of magnitude ($\text{SDC}{\text{ind}}$ 10.10 vs $\text{SDC}$ 0.86, PMID 28758322); combining metric errors with level mixing muddles these two numbers together.}
Eighth Layer, how to measure. Pure simulation: Monte Carlo synthesis of φ sequences with known κ, sweeping mean μ from 0° to ±60°, divided into conditions with and without crossing zero, generating paired samples at 20/50/200 breaths each with correlation coefficients r at four levels (0.5/0.7/0.9/0.97); compare MDC95 derived from arithmetic difference SD versus circular dispersion against recovery bias for true κ. Real data: ≥90 subjects at rest with RIP recordings, same posture and banding position, recalculate with both metrics, statistically compare the proportion judged as “difference falls within equivalence bounds” across the same batch of people.
Ninth Layer, four-grid criteria. ① In the grid where $\mu \to 0^\circ$ and $r \geq 0.9$, arithmetic MDC95 overestimates by >30% and monotonically increases as μ decreases ⇒ arithmetic judgment systematically fails in populations with best synchronization; MDC conclusions not reporting φ mean are invalid for two-state interpretation. ② Inconsistency rate between two metrics for the same batch of people >20% (even if focusing only on $\mu < 15^\circ$ subgroup) ⇒ switch to dual metrics in parallel, add “inconsistency implies untestable.” ③ Bias <10% and consistency rate >90% ⇒ arithmetic MDC can be retained; concerns in this article warrant retraction.
Evidence and Limitations
Verified: The Lissajous algorithm for phase angles has a hard ±90° range; measured data spanned −86.90° to +88.4° across 2,290 cycles, while global phase delay on the same dataset covered ±179° (PMID 38829281); phase angles and phase offsets can differ by a factor of two in classifying the same dataset (15 vs 7, 14 vs 8, PMID 28183817); mean resultant vector length exhibits systematic bias dependent on sample size (PMID 37756609); the locking and unimodal premises of relative phase SD can be falsified in real data (PMID 12297252).
Partially Verified: Phase angle consistency between two volume plethysmography techniques is good (PMID 28767680), suggesting instruments are not the primary error source; however, this conclusion covers only rest and exercise with 27 subjects and cannot be extrapolated to long-duration rest or extremely low-amplitude thoraco-abdominal motion. Individual-level versus group-level detectable changes can differ by an order of magnitude ($\text{SDC}{\text{ind}}$ 10.10 vs $\text{SDC}$ 0.86, PMID 28758322); whether this order-of-magnitude difference holds for angular metrics lacks empirical verification.}
To Be Verified: The magnitude of overestimation by arithmetic MDC95 when φ crosses zero (30% or 300%) has no empirical measurement; whether circular metrics truly better recover κ is only available through theoretical derivation and simulation, lacking human data comparison; “minimum number of breaths required for convergence between two metrics” is not calibrated. Phase estimation precision itself is affected by sampling design (non-uniform sampling significantly alters phase parameter variance, PMID 40952326), and existing thoraco-abdominal studies rarely report sampling intervals or breath counts; this confounding factor remains unseparated.
Related Reading
- Is “Two-State Difference” Accurate: That Gate Must First Pass This Check
- Breathing Stability Cannot Be Judged Solely by Amplitude
- “Practice Yields No Different from Non-Practice”: Requires an Equivalence Bound First
- Resonance Consciousness Field: Research Pool One
Evidence Level: C
References: PMID 38829281, PMID 28183817, PMID 37756609, PMID 12297252, PMID 28767680, PMID 28758322, PMID 18569858, PMID 40952326, PMID 29991414
L1 Network Status: The
CORRESPONDENCE_NETWORK.mdnode table registers only “Phase/Rhythm” on the physical side (note: original text “key to periodic coupling”), while all eight items on the physiological side (HRV, skin temperature, breathing rate, EMG, EEG, gastrointestinal motility, qi sensation reports, constitution typing) are cross-sectional observations; there are no measurement nodes for “circular dispersion/von Mises κ/minimum detectable change/cross-zero/measurement error band” across all three domains. TheEDGE_TABLE_science-yixue.mdis a conceptual reverse statistical extraction (774 articles / unique PMID 3542); grepping “phase/MDC/circular/measurement error/reliability/equivalence/κ” only hits identical rows in the node table without L1/L2/L3 hierarchy or active/open/conflict status determination; edge tables stop at E-050 (Level L1, status open, modern evidence column originally “No measurable sources yet“). This topic has not entered the L1 corresponding network, thus this article’s evidence level is capped at 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