三因辨证选声 vs 泛用声疗的效应量,能在多中心重复吗?

Written by

in

Language

Evidence level: B

单中心阳性不是疗法有效的证明——声音治疗文献里的效应量从 0.36 到 1.54 满天飞、异质性 I² 冲到 97%,”辨证选声比一张方子治百病强”这个增量,必须拿到多中心协议下去检验会不会被洗掉。

正文

先给结论:“辨证组 vs 泛用组”的头对头试验设计已经成立(见《三因辨证选声 vs 泛用声疗的头对头 RCT》),但那个设计只回答了”有没有差异”;科学上更硬的问题是”这个差异能不能重复”。 单中心样本量小、实施者是同一拨人、人群是同质的,阳性结果天然膨胀。要把藏医”辨证选声”从”一个中心有效”推向”这个疗法真的有效”,必须做多中心重复验证——而这恰恰是声音治疗领域最薄弱的环节。

看现代证据先看反面教材。2015 年一项九中心、120 名痴呆患者的随机对照试验:在九个意大利机构同时开展,比较主动音乐治疗、个体化听音乐、标准照护三组,结果行为与情绪结局三组无显著差异,只有组内随时间改善(PMID 26289682)。这不是音乐没用的孤例——它说明单中心的乐观发现在多中心、多实施者的真实条件下,会被稀释到统计上站不住。声音治疗 meta 分析的异质性数据更直白:2024 年乳腺癌音乐干预 meta 中,焦虑、抑郁、疼痛结局的异质性分别高达 I²=93.6%、87.3%、97.1%(PMID 38321719)——接近完全不可比。这种量级的异质性意味着”效应量”几乎没有意义,除非我们把它拆开:差异来自处方、人群、还是实施?

再给一个整体视角。2025 年音乐治疗焦虑多水平 meta 分析纳入 51 项试验、93 个效应量:总体中等效应 g=0.357,但生理终点只有 g=0.153 且不显著(PMID 40547443)。也就是说”声音缓解焦虑”在问卷上成立、在生理指标上存疑——这提醒藏医试验的双终点设计里,HRV 频域这类生理终点可能天然比症状量表更难出阳性,多中心后尤其如此。2020 年抑郁音乐 meta(55 项 RCT)进一步暴露”处方类型是最大调节变量”:音乐疗法 SMD=-0.66,音乐医学 SMD=-1.33,再创造性疗法 SMD=-1.41,差距近一倍(PMID 33206656)。同样是”听音乐”,换一种开法,效应量差一倍——这正是”辨证选声”要证明的增量,也是它最容易被多中心稀释的东西。

但个性化/对证的音频并非都活不下来。2016 年一项盲法随机对照试验,50 名耳鸣患者随机接受”按个体耳鸣频谱定制修改的音乐”或”未修改的古典音乐”,随访 12 个月,定制组在耳鸣困扰、严重度、功能损害上显著优于对照组,且效应持续(PMID 27838685)。这说明只要处方依据足够硬(个体频谱特征),效应可以跨时程存活——多中心要做的不是否认增量,而是检验”三因辨证”是否也是这么硬的处方依据。

那么,藏医版多中心重复该怎么做?核心是统一协议 + 中心作随机效应 + 报告预测区间:①处方库标准化——把三因声处方做成”频谱指纹”级别的定义(隆型低缓稳定音、赤巴型清凉下行音、培根型温暖振奋音,各配固定音阶、频段、时长、响度曲线),由中央生成统一音频文件分发到各中心,杜绝”每个中心自己放音乐”的实施漂移;②中心选择——至少 4 个中心,跨藏区、内地、海外藏医机构,天然覆盖三因人群分布差异;③三因量表跨中心校准——统一量表与双盲辨证流程,先做评分者间一致性(ICC)再放行;④统计分析——中心作为随机效应,报告辨证组 vs 泛用组的 Δ效应量、I²、95% 预测区间,预测区间穿过 0 就说明增量不稳定。三种可判读结局:增量稳定为正=三因辨证携带真实特异信息;增量消失但两组都优于假声=声音有用、辨证无增量,藏医处方特异性需重审;全组无效=回到声频参数映射层查设计。

证据与局限

已验证: 声音治疗作为泛类有中等效应但异质性极高(焦虑 g=0.357、生理终点不显著);处方类型改变效应量近一倍;个性化频谱音乐可在盲法下持续 12 个月有效;多中心可稀释单中心阳性(九中心痴呆试验三组无差异)。

部分验证: “辨证/个体化音频优于统一音频”在中医五行音乐、耳鸣定制音乐中有阳性头对头证据,但均非藏医三因框架、且未做跨中心效应量重复。

待验证: 三因辨证组 vs 泛用组的 Δ效应量在跨中心协议下是否稳定(预测区间是否穿过 0);异质性中中心、人群三因分布、实施各占多少;生理终点(HRV 频域)在多中心下是否仍可测出增量。

相关阅读

证据等级: B(”辨证音频优于泛用音频”有头对头 RCT 与 meta 支撑,”藏医三因辨证增量在多中心稳定”尚无直接实证,属科学类比)
参考文献: PMID 26289682, PMID 38321719, PMID 40547443, PMID 33206656, PMID 27838685


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。

Comparing the Effect Sizes of Three-Factor Differentiation-Based Sound Selection Versus Generalized Sound Therapy: Can It Be Replicated in a Multi-Center Study?

Evidence level: B

A positive result from a single center does not prove therapeutic efficacy. Effect sizes in sound therapy literature range wildly from 0.36 to 1.54, with heterogeneity ($I^2$) reaching 97%. The claim that “syndrome-differentiated selection is superior to a one-size-fits-all prescription” must be subjected to multi-center protocol validation to determine if this incremental benefit is diluted.

Main Text

Conclusion: A head-to-head trial design comparing “syndrome-differentiated groups” vs. “generic groups” is established (see Head-to-Head RCT: Tibetan Syndrome-Differentiated vs. Generic Sound Therapy), but this design only answers “is there a difference?” The scientifically more rigorous question is “can this difference be replicated?” Small single-center samples, identical implementers, and homogeneous populations naturally inflate positive results. To move Tibetan medicine’s “syndrome-differentiated sound selection” from “effective at one center” to “truly effective,” multi-center replication is mandatory—a weakness currently prevalent in the sound therapy field.

Lessons from Modern Evidence: The Counter-Example

Looking at modern evidence, first examine the negative example. A 2015 nine-center, 120-subject randomized controlled trial in dementia patients conducted simultaneously across nine Italian institutions compared active music therapy, individualized listening, and standard care. Results showed no significant differences between the three groups for behavioral and emotional outcomes, with improvement only observed within groups over time (PMID 26289682). This is not an isolated case of music failing; it demonstrates that optimistic findings from single centers are diluted under real-world multi-center, multi-implementer conditions until they become statistically unsupported. Sound therapy meta-analyses provide even more direct evidence of heterogeneity: in a 2024 meta-analysis of music interventions for breast cancer, heterogeneity for anxiety, depression, and pain outcomes reached $I^2=93.6\%$, $87.3\%$, and $97.1\%$ respectively (PMID 38321719)—approaching complete non-comparability. At this magnitude of heterogeneity, “effect sizes” hold little meaning unless decomposed: does the variation stem from prescriptions, populations, or implementation?

A Holistic Perspective

A 2025 multi-level meta-analysis of music therapy for anxiety included 51 trials and 93 effect sizes. The overall moderate effect was $g=0.357$, but physiological endpoints showed only $g=0.153$ and were not significant (PMID 40547443). Thus, “sound alleviating anxiety” holds in questionnaire data but is questionable regarding physiological metrics. This suggests that in Tibetan medicine’s dual-endpoint design, physiological endpoints like HRV frequency domains may naturally yield fewer positive results than symptom scales, especially post-multi-center validation. A 2020 depression music meta-analysis (55 RCTs) further revealed “prescription type” as the largest moderator: Music Therapy $SMD=-0.66$, Music Medicine $SMD=-1.33$, and Re-creative Therapy $SMD=-1.41$, with a gap nearly double (PMID 33206656). The same “listening to music,” yet with different prescriptions, yields effect sizes differing by nearly a factor of two—precisely the incremental value “syndrome-differentiated sound selection” must prove, yet also the most vulnerable to dilution in multi-center settings.

Survival of Personalized/Audiotherapy Based on Diagnosis

However, personalized/diagnostic audio does not always fail. A 2016 blinded randomized controlled trial assigned 50 tinnitus patients to either “music modified according to individual tinnitus spectral characteristics” or “unmodified classical music,” followed up for 12 months. The customized group showed significantly better outcomes in tinnitus distress, severity, and functional impairment compared to the control group, with effects persisting (PMID 27838685). This indicates that provided the prescription basis is sufficiently robust (individual spectral features), effects can survive across long durations—multi-center trials should not deny the incremental value but rather verify whether “Three Factors Differentiation” constitutes an equally robust prescription basis.

Designing Multi-Center Replication for Tibetan Medicine

How should multi-center replication be structured for the Tibetan version? The core lies in standardizing protocols, treating centers as random effects, and reporting prediction intervals:
1. Standardize Prescription Libraries: Define Three Factors sound prescriptions at the level of “spectral fingerprints” (e.g., Rong type: low-frequency stable tones; Chitta type: cooling descending tones; Beken type: warming uplifting tones), each paired with fixed scales, frequency bands, durations, and loudness curves. Generate unified audio files centrally and distribute them to all centers to prevent implementation drift where “each center plays its own music.”
2. Center Selection: Include at least 4 centers across Tibetan regions, mainland China, and overseas Tibetan medical institutions to naturally cover differences in Three Factors population distributions.
3. Cross-Center Calibration of Three Factors Scales: Standardize scales and double-blind differentiation protocols. Conduct inter-rater reliability checks (ICC) before approval.
4. Statistical Analysis: Treat centers as random effects. Report the $\Delta$ effect size between the syndrome-differentiated group and the generic group, along with $I^2$ and 95% prediction intervals. If the prediction interval crosses zero, the incremental benefit is unstable.

Three interpretable outcomes emerge:
* Stable positive increment = The syndrome-differentiated approach carries genuine specific information.
* Increment disappears but both groups outperform sham sound = Sound therapy is useful, but syndrome differentiation adds no incremental value; Tibetan prescription specificity requires re-examination.
* Entire group ineffective = Return to the sound frequency parameter mapping layer to check design flaws.

Evidence and Limitations

Verified: Sound therapy as a broad category shows moderate effects but extreme heterogeneity (anxiety $g=0.357$, physiological endpoints non-significant); prescription types alter effect sizes by nearly a factor of two; personalized spectral music remains effective for 12 months under blinding; multi-center trials can dilute single-center positives (nine-center dementia trial showed no difference across three groups).

Partially Verified: “Syndrome-differentiated/individualized audio superiority over unified audio” has positive head-to-head evidence in Traditional Chinese Medicine Five Elements music and tinnitus customized music, but neither uses the Tibetan Three Factors framework nor replicates effect sizes across centers.

To Be Verified: Whether the $\Delta$ effect size between syndrome-differentiated and generic groups remains stable under cross-center protocols (does the prediction interval cross zero); the proportion of heterogeneity attributable to centers, Three Factors population distributions, and implementation; whether physiological endpoints (HRV frequency domains) can still detect incremental benefits in multi-center settings.

Related Reading

Evidence Level: B (“Syndrome-differentiated audio superiority over generic audio” is supported by head-to-head RCTs and meta-analyses; “stability of Tibetan Three Factors differentiation increment in multi-center settings” lacks direct empirical evidence and relies on scientific analogy).
References: PMID 26289682, PMID 38321719, PMID 40547443, PMID 33206656, PMID 27838685


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 *