Evidence level: B
骨关节炎说”没关”,痛经说”有关”。这不是两个病给出了两个答案,而是两个研究根本不是同一个问题。
正文
得气强度→脑网络调制→临床疗效,这条三段链是针刺机制研究里最漂亮的一个假设:刺激产生感觉,感觉改变网络,网络改变疼痛。它已被源文立为待验证的完整中介模型。现在有人拿两个病种来对照:骨关节炎(OA)里得气与镇痛无关(p=0.49),痛经里得气与镇痛相关。于是很容易得出一个结论——中介强度有疾病特异性,痛经这条链通,OA 不通。
这个结论下得太快。两篇放在一起先要问”可比吗”,答案几乎是否定的,理由四层,一层比一层基本。
第一层:阳性/阴性是两套对照强度的产物。 2018 年 Vickers 的个体患者数据元分析(29 试验、20 827 人)给出过一句很硬的话:试验之间效应量的差异,主要由对照组接受的治疗决定,而不是由针刺本身的特征决定(PMID 29198932)。这直接决定了 OA 那个”阴性”的分量。Hinman 2014 的 JAMA 试验里,毫针相对假针 12 周时疼痛只差 −0.4 单位(95%CI −1.2~0.4),但相对不治疗对照差 −1.1(−1.8~−0.4)(PMID 25268438)——针不是没用,是它赢不了那个假针。而痛经那篇的两组是”深刺粗针加手法”对”浅刺细针不手法”(PMID 31223331),后者是天然偏弱的对照,组间差必然拉大。一个是针打不过强假针,一个是拿自己设计的弱对照比。这两个标签比的不是中介强弱,是假针对照做得像不像。
第二层:两个”疗效”不在同一个时间窗。 OA 试验的结局是四周疗程结束时的疼痛变化;痛经那篇测的是起针当时、起针后 10、20、30 分钟(PMID 31223331)。前者是累积疗程效应,后者是即时镇痛。即时镇痛和疗程效应背后大概率不是同一条通路——前者更依赖内源性镇痛与注意调制,后者还要经过组织修复、习惯化与期望重塑。用一个即刻的量去当四周结局的中介,或者反过来,都不是”中介强弱不同”,是链条长度不同。
第三层:得气不是一个标量,符号甚至会在同一研究内部翻转。 痛经那篇最要命的不是”相关”,而是相关的方向在两组里相反:得气组里得气强度与镇痛负相关,非得气组里正相关(PMID 31223331)。同一份数据、同一个量表、同一个穴位,方向翻了。原因写得也明白——得气含两簇方向相反的东西:酸痛、胀满随强度上升而疗效下降,电窜、放散则相反。文章最终结论是”中度得气比强得气更能延长镇痛“,也就是一条倒 U 曲线。而 OA 那篇用的是 Park 问卷总分加”有没有得气”的二分类(PMID 20964256),用一维总分去平均一个多维、且方向互抵的构念,无论真实关系是正是负,都会被压成”p=0.49″。拿一个倒 U 曲线的左段和一个被压平的总分去比”中介强度”,比的其实是两把不同的尺子。
第四层:中介效应量本身没有跨研究的共同单位。 自然间接效应的点识别依赖 cross-world independence,这个假设不可检验(PMID 33512846);不做该假设时只能拿到部分识别区间(PMID 38962169)。加上效应量的非可压缩性——在 OR 尺度上调整协变量引起的效应变化并非中介(PMID 33314251)——”中介比例 30%”这类数字一旦换了结局量表就不可移植。更实际的问题是测量误差:单次得气评分是瞬时量,其重测信度若像多数症状量表那样在 0.5-0.7,间接效应会按信度被系统性稀释(回归稀释,PMID 37336529),而两个病种用的量表不同,稀释幅度就不同——连”谁的中介更强”的排序都不可靠,更不用说比较其数值。
所以本篇的判决是:“跨病种中介强度比较”这个设计,在现有研究之间做不成,只在一个统一样本的试验内部做得成。 可行的做法是把疾病从”两个研究”改成”一个研究里的一个调节变量”:一份方案、一个得气量表(且必须拆成不适簇/放散簇/非痛感簇分别进模型,不合并总分)、一个脑网络指标、一个固定时间窗,按病种分层随机,检验的是”病种×得气”的交互项——而不是拿两篇论文的中介比例相减。中介报告必须同时给出部分识别区间而非点估计,并声明结局尺度。
易学落点
《灵枢·终始》讲针刺的次序:”一刺则阳邪出,再刺则阴邪出,三刺则谷气至,谷气至而止。”最后四个字值得停一下——“至而止”。古人没有说”谷气越多越好”,说的是到了就停。痛经那篇的倒 U 曲线,恰好在数据层面重复了这四个字:强得气并不优于中度得气。
更要紧的是,古典文本里的”得气”从来不是一把连续刻度尺。它至少是三种不同性质的变量:《灵枢·九针十二原》”刺之要,气至而有效”——是有/无的二值事件;《灵枢·终始》”谷气至而止”——是何时该停的时点判断;《灵枢·行针》分四类人”或神动而气先针行,或气与针相逢,或针已出气独行,或数刺乃知”——是时序类型的分类变量(该篇已论)。这三种里,一种都不是”强度 0-10 分”。把得气压缩成一个可以进回归的连续量,是现代研究方法的需求,不是古典概念的结构。 于是当数据给出矛盾结果时,我们其实是在追问一个被自己改造过的变量。
证据与局限
已验证:Vickers 2018 的对照驱动结论、Hinman 2014 的针不优于假针、White 2010 的得气与 OA 镇痛无关、Hu 2019 的组内符号翻转与倒 U 结论,均有原文支撑。
部分验证:得气的多维成分结构在痛经中已被观察到,但在其他病种是否同构未知;测量误差对间接效应的稀释幅度尚无针刺领域的实测值。
待验证:跨病种中介的调节效应(”病种×得气”交互)从未在同一方案内被检验;得气量表各成分簇的重测信度与最小可检出差值均未标定。
相关阅读
证据等级: B
参考文献: PMID 29198932, PMID 25268438, PMID 20964256, PMID 31223331, PMID 33512846, PMID 38962169, PMID 33314251, PMID 37336529
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。
One Chain, Two Diseases: The Mediating Role of Qi Reception, Network, and Efficacy—Can We Compare Across Disease Categories?
Evidence level: B
Osteoarthritis (OA) says “no link”; dysmenorrhea says “yes link.” These are not two diseases giving two answers; the two studies address fundamentally different questions.
Main Text
The three-stage chain—Qi reception intensity → brain network modulation → clinical efficacy—is arguably the most elegant hypothesis in acupuncture mechanism research: stimulation generates sensation, sensation alters networks, and networks alter pain. It has been established in the original literature as a complete mediator model awaiting verification. Now, two disease categories are used for comparison: Qi reception shows no association with analgesia in OA ($p=0.49$), but a significant association in dysmenorrhea. Consequently, it is easy to conclude that mediator strength exhibits disease specificity—the chain is open in dysmenorrhea but closed in OA.
This conclusion is drawn too hastily. Before juxtaposing the two studies, one must ask “Are they comparable?” The answer is almost certainly no, for four reasons, each more fundamental than the last.
First Layer: Positive/Negative findings are products of two different control intensities. Vickers’ 2018 individual patient data meta-analysis (29 trials, 20,827 participants) made a definitive statement: between-trial effect size differences are primarily determined by the control treatment received, not by characteristics of acupuncture itself (PMID 29198932). This directly determines the “negative” component in the OA study. In Hinman’s 2014 JAMA trial, at 12 weeks, pain scores differed by only −0.4 units between fine needles and sham needles (95%CI −1.2~0.4), but by −1.1 units compared to no-treatment controls (−1.8~−0.4) (PMID 25268438)—the needle is not useless; it simply loses against the sham. Conversely, in the dysmenorrhea study, the comparison was “deep insertion with coarse needles and manipulation” versus “shallow insertion with fine needles without manipulation” (PMID 31223331), where the latter represents a naturally weaker control, inevitably widening the group difference. One study compares a needle against a strong sham; the other compares against a weak, self-designed control. These labels do not compare mediator strength; they compare the quality of the control group.
Second Layer: The two “efficacy” outcomes operate outside different time windows. The OA trial measured pain change at the end of a four-week course; the dysmenorrhea study measured effects immediately upon needle removal, and at 10, 20, and 30 minutes post-removal (PMID 31223331). The former reflects cumulative course effects, while the latter reflects immediate analgesia. Immediate analgesia and course effects likely rely on different pathways—the former depends more on endogenous analgesia and attentional modulation, while the latter involves tissue repair, habituation, and expectation reshaping. Using an immediate measure as a mediator for a four-week outcome, or vice versa, does not indicate “different mediator strengths”; it indicates different chain lengths.
Third Layer: Qi reception is not a scalar; even the sign can flip within the same study. The most critical issue in the dysmenorrhea study is not “association,” but the direction of association reverses between groups: in the Qi-reception group, intensity correlates negatively with analgesia, while in the non-Qi-reception group, it correlates positively (PMID 31223331). The same data, same scale, same acupoint—yet the direction flips. The authors explain this clearly: Qi reception contains two clusters of opposing effects; efficacy decreases with rising intensity for soreness/distension, but increases for electric dispersal/spreading. The study concludes that “moderate Qi reception extends analgesia better than strong Qi reception,” implying an inverted U-curve. In contrast, the OA study used the total score of the Park questionnaire plus a binary classification of “presence/absence of Qi” (PMID 20964256). Using a one-dimensional total score to average a multi-dimensional construct with opposing directional components compresses both positive and negative relationships into “p=0.49.” Comparing the left segment of an inverted U-curve against a flattened total score is not comparing “mediator strength”; it is comparing two different rulers.
Fourth Layer: Mediator effect sizes lack a common unit across studies. Natural indirect effects rely on cross-world independence, an assumption that is untestable (PMID 33512846); without it, only partial identification intervals can be obtained (PMID 38962169). Furthermore, effect sizes are non-compressible: changes in effects caused by covariate adjustment on an odds ratio scale are not mediated effects (PMID 33314251). Consequently, figures like “mediator proportion of 30%” are not portable when the outcome scale changes. A more practical issue is measurement error: single-timepoint Qi scoring is an instantaneous measure. If its test-retest reliability is around 0.5–0.7, similar to many symptom scales, the indirect effect will be systematically diluted by regression dilution (PMID 37336529). Since different scales are used in the two disease categories, the dilution magnitude differs—even the ranking of “whose mediator is stronger” is unreliable, let alone comparing their numerical values.
Therefore, the verdict for this paper is: Comparing mediator strength across disease categories cannot be performed between existing studies; it can only be done within a single study’s trial. A feasible approach is to reframe the disease from “two studies” to “a moderator variable within one study”: a single protocol, a single Qi scale (which must be disaggregated into discomfort clusters/dispersal clusters/non-painful sensation clusters into separate models, not merged into a total score), a single brain network metric, and a fixed time window, with randomization stratified by disease category. This tests the “disease × Qi” interaction term—rather than subtracting mediator proportions from two papers. Mediator reports must provide partial identification intervals rather than point estimates and explicitly declare the outcome scale.
Yijing Application Points
The Ling Shu · Zhong Shi discusses the sequence of acupuncture: “One insertion drives out Yang evil; two insertions drive out Yin evil; three insertions bring the valley Qi. Stop when valley Qi arrives.” The final four characters, “arrive and stop,” deserve pause. Ancient texts did not say “more valley Qi is better”; they stated to stop upon arrival. The inverted U-curve in the dysmenorrhea study replicates these four characters at the data level: strong Qi reception is not superior to moderate Qi reception.
More importantly, “Qi reception” in classical texts is never a continuous ruler. It comprises at least three distinct variable types: Ling Shu · Jiu Zhen Shi Er Yuan‘s “The key to needling is when Qi arrives, efficacy follows”—a presence/absence binary event; Ling Shu · Zhong Shi‘s “Stop when valley Qi arrives”—a timing judgment of when to cease; and Ling Shu · Xing Zhen‘s classification of four types of people (“or spirit moves before needle travel, or Qi meets needle, or needle exits but Qi travels alone, or multiple insertions reveal it”)—a temporal type categorical variable (discussed previously). None of these are “intensity scores from 0 to 10.” Compressing Qi reception into a continuous variable suitable for regression is a demand of modern research methodology, not the structural nature of the classical concept. Thus, when data yields contradictory results, we are actually interrogating a variable that has been transformed by ourselves.
Evidence and Limitations
Verified: Vickers’ 2018 conclusion driven by control design, Hinman’s 2014 finding that needles are not superior to sham needles, White’s 2010 finding of no association between Qi and OA analgesia, and Hu’s 2019 finding of intra-group sign reversal and inverted U-conclusion are all supported by original texts.
Partially Verified: The multi-dimensional component structure of Qi reception has been observed in dysmenorrhea but remains unknown whether it is isomorphic in other conditions; no empirical values exist in the acupuncture field regarding the dilution magnitude of measurement error on indirect effects.
To Be Verified: The moderating effect of cross-disease mediation (“disease × Qi” interaction) has never been tested within a single protocol; test-retest reliability and minimum detectable differences for each component cluster of the Qi scale remain uncalibrated.
Related Reading
- Qi Intensity → Brain Network → Efficacy: Three Pieces of Evidence for a Mediator Chain
- Two Scales on One Needle: Convergence of Subjective Qi and Objective Manipulation
- Research Pool: Biological Frequency Code
Evidence Level: B
References: PMID 29198932, PMID 25268438, PMID 20964256, PMID 31223331, PMID 33512846, PMID 38962169, PMID 33314251, PMID 37336529
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