5D Library › 内感受解离、突显网络切换与内脏自主神经修复 · English

第八章 实践与预注册检验

注

本章的核查表与试验方案用于研究与阅读文献,不是医疗建议。 核查表不能告诉你是否应当做某种练习,只能帮助你在读到一项说法时把问题问清楚。试验方案是供独立研究机构参考的设计草案,必须经伦理委员会审查、在临床试验注册平台登记,并由有资质的团队实施。任何盆腔、泌尿、消化或生殖系统的症状,以及胸闷、心悸、头晕、呼吸困难,都应先由医生评估。

第一部分 四步核查表

前七章可以浓缩为一份四步核查表,用于阅读任何「某种注意练习改善某种生理状态」的说法。

  1. 链条。 这个说法有几个环节?每一环是证据、模型,还是类比?有没有哪一环在方向上与它所依据的数学相反(第一、二章)?模型中的「证明」是否被当成了经验结论(第六章)?
  2. 解剖。 被说成原因的那条通路,解剖上到得了被说成结果的那个器官吗?所用的测量与所主张的结论是同一个系统吗(第四章)?
  3. 可识别性。 所用的测量在原则上能不能确定所主张的量?有没有一组不同的参数能给出同样的测量而蕴含不同的结论(第六章)?
  4. 证据形式。 是个案还是对照试验?若是个案,进入练习的时点是怎样决定的,序列自相关处理了吗(第六章)?若是试验,对照臂在时长、姿势、注意力与接触上匹配了吗,还是等待名单(第三章、本章第二部分)?

把核查表用于本书开头的那条链条,结果如下表。这不是对练习的总体评价,而是说明在这四个方面,那条链条能支持到哪里。

步骤 那条链条的情况 能支持的结论
链条 六环;第一环方向相反(警觉先验下后验上升),第三环不增加信息(d′d' 与增益无关),第二环的代价(熵率下降、首达时间变长)未提 作为机制陈述不成立;作为「某类练习可能有用」的提示成立,但理由与原说法给的不同
解剖 迷走神经不支配盆腔器官;盆腔出路是骶部的,且「骶部副交感」本身在争论中 第四环为事实错误;关于盆腔的结论不能由心率变异度支持
可识别性 四组参数重现同一条心脏曲线,蕴含的平均内脏传导度 0.506 到 0.884 「心率变异度上升故内脏灌注恢复」是假设,不是推论;心脏通道只能识别反应的时点
证据形式 以个案为主 无效应时朴素检验在 28.5% 的个案中宣称获益;似然比约 2.2,弱证据

第二部分 预注册的两臂随机对照试验

问题与设计选择

在身体警觉偏高的成年人中,一项为期 6 周、以「注意身体内部而不解释它」为内容的练习,与在时长、姿势、呼吸引导、指导者接触与材料篇幅上结构匹配的外部注意对照练习相比,第 6 周的身体警觉分数是否更低,且差别至少达到最小关注效应?

六点设计选择,每一点都来自前面某一章的结果。

第一,只设一个主要比较:练习对结构匹配的主动对照。不与等待名单比较。理由是 Goyal 等人的系统综述显示,正念类项目相对主动对照的效应是小到中等,而相对非主动对照要大得多 (Goyal et al. 2014);Van Dam 等人系统地讨论了该领域的夸大问题 (Van Dam et al. 2018)。与等待名单比较测的是「做点什么比什么都不做好」,这不是本书要问的问题。

第二,只设一个主要结局:第 6 周的身体警觉分数,以基线分数与中心为协变量作协方差分析 (Vickers and Altman 2001; Senn 2006)。不以任何自主神经指标为主要结局,有两个独立的理由:第六章程序 p09 的结构不可识别性,以及第四章程序 p05 的呼吸混淆与本章程序 p14 的信度上界。不以心跳计数任务为任何结局,理由见第一章 (Zamariola et al. 2018)。

第三,结局在另一天的独立评估中测量,取两次评估的平均。理由是第三章程序 p11:课末测量到的主要是一个到第二天就消失的快过程,而混淆的方向取决于练习间隔;程序 p14 给出取两次的理由。

第四,预先指定一个阴性对照结局:一个本框架预测不应改变的量(例如与内感受无关的一般性生活满意度单题,或一项与练习内容无关的认知任务成绩)。它不计入把握度计算,它的作用是识别泛化的预期效应:如果它的变化达到主要效应的一半,那么「什么都动了」是更简单的解释。

第五,预期与可信度必须测量并匹配:用可信度/预期问卷 (Devilly and Borkovec 2000) 在第 1 周测量两臂,并在判定规则中作为限定条件。结构匹配的对照不等于预期匹配的对照,而这件事只能测量,不能假设。

第六,由独立机构实施:方案、分析代码与判定规则在第一例入组前公开登记,作者不参与招募、干预、评估与分析。

方案

要素 规定
实施与登记 由与作者无关的研究团队实施;伦理委员会批准;在公开的临床试验注册平台登记;按 SPIRIT/CONSORT 撰写与报告 (Chan et al. 2013; Schulz, Altman, and Moher 2010);不从作者的社群、学员或客户中招募
参与者 25–60 岁;身体警觉量表筛查分数达到预先设定的切点;能独立完成每日练习记录;排除当前的惊恐障碍、创伤后应激障碍、精神病性或解离症状、癫痫、重度心血管或呼吸系统疾病、孕妇、以及正在接受以内感受为内容的心理治疗者;所有筛查与排除由中心的临床人员判断。任何躯体症状者先转诊评估,不纳入
分层 按中心、基线身体警觉(两层)与表象生动度(两层)分层;表象生动度分层的理由见第三章:无自发视觉表象者对以表象为核心的指导语不适用 (Zeman, Dewar, and Della Sala 2015)
随机化 中央网络随机系统,1:1,区组大小可变且不公开;分析时调整分层因素
干预臂 每日一次、固定时长的内感受注意练习,指导语为注意身体内部的感觉与自发出现的表象而不作解释、不作命名;每日练习次数与间隔固定并记录(理由见第三章程序 p11:间隔是干预的参数)
对照臂 结构完全匹配的外部注意练习:相同时长、相同姿势、相同呼吸引导、相同的指导者接触次数与时长、相同篇幅的书面材料,内容为注意外部的声音或视觉对象。姿势与肌肉张力的匹配是必需的,理由见第五章末节的力学替代解释
设盲 参与者不能对练习内容设盲,但对研究假设设盲:两臂的材料都不说明预期方向;结局评估者与统计分析人员设盲;可信度/预期在第 1 周测量 (Devilly and Borkovec 2000)
主要结局 第 6 周身体警觉分数(第 6 与第 7 周两次独立评估的平均);协方差分析,协变量为基线分数与中心;主要比较:干预臂对对照臂
阴性对照结局 预先指定的一个与内感受无关的量;报告其估计与区间,不计入把握度
次要结局 多维内感受觉察量表各分量表 (Mehling et al. 2012);第 2 周与第 6 周的表象流转移熵(Miller—Madow 估计、个人均值、取样间隔固定,见第三章程序 p13);任务切换代价(见第二章假说);自然呼吸窗口内的 RMSSD 与高频功率,与同一窗口的呼吸频率并列报告(第四章);不良事件,特别是症状加重与解离体验。均为描述性或探索性分析
敏感性分析 按方案集分析;随机缺失假设下的多重插补与偏离该假设的情景分析;以基线身体警觉为效应修饰的预先指定检验(检验第六章的边界条件:常驻警觉先验高时练习是否反而无益)
等效性判读 若主要效应落在 ±SESOI 之内,按等效性检验的框架报告为「没有值得据此改变做法的差别」,而不是「无效应」(Lakens, Scheel, and Isager 2018)

最小关注效应与样本量

最小关注效应定为 0.35 个标准差。理由是:半个标准差是常被引用的「最小重要差别」的经验界值 (Norman, Sloan, and Wyrwich 2003),而 0.35 低于它,是一个保守(更难达到)的选择;小于这一差别,不值得为每天的练习时间改变做法。基线与结局的相关取 0.60,单次评估的信度取 0.75。这些都是需要在预试验中核对的规划假设。

先看信度的后果,因为它决定了整个试验的规模。

推导

# Chapter 8. Reliability sets a ceiling on any effect a trial can see. If the
# primary outcome has reliability rho_xx, an observed standardised effect is
# attenuated:
#   d_observed = d_true * sqrt(rho_xx)
# and the sample size needed goes up as 1 / rho_xx. For an outcome averaged
# over k occasions, the Spearman-Brown formula gives the reliability of the
# average, so the program answers a design question: is it cheaper to add
# participants or to add measurement occasions per participant? Exact
# formulas, no simulation.
import math
from statistics import NormalDist

Z = NormalDist().inv_cdf

def n_per_arm(d, rho_xx=1.0, power=0.90, alpha=0.05, rho_base=0.0):
    """ANCOVA with a baseline covariate correlated rho_base with the
    outcome."""

    d_obs = d * math.sqrt(rho_xx)
    za, zb = Z(1 - alpha / 2), Z(power)
    return 2 * (za + zb) ** 2 * (1 - rho_base ** 2) / d_obs ** 2 + za ** 2 / 4

def spearman_brown(rho1, k):
    return k * rho1 / (1 + (k - 1) * rho1)

if __name__ == "__main__":
    # smallest effect of interest, SD units
    D = 0.35
    print(
        "Smallest effect of interest: {:.2f} SD. ANCOVA, two-sided 0.05, 90 %"
        " power,".format(D)
    )
    print("baseline-outcome correlation 0.60.\n")
    print("   reliability   attenuated d   n per arm   n for two arms")
    for r in (1.00, 0.90, 0.80, 0.70, 0.60, 0.50):
        n = n_per_arm(D, rho_xx=r, rho_base=0.60)
        print("   {:11.2f} {:14.3f} {:11d} {:16d}"
              .format(r, D * math.sqrt(r), math.ceil(n), 2 * math.ceil(n)))

    print("\nHalving reliability from 0.90 to 0.50 nearly doubles the trial")
    print("({} -> {} per arm)."
          .format(math.ceil(n_per_arm(D, 0.90, rho_base=0.60)),
                  math.ceil(n_per_arm(D, 0.50, rho_base=0.60))))

    print(
        "\nAdd occasions or add people? Single-occasion reliability 0.60; the"
    )
    print("outcome is the mean of k occasions, each costing one visit.")
    print(
        "\n   k   reliability of the mean   n per arm   total visits per arm"
    )
    best = None
    for k in (1, 2, 3, 4, 6, 8, 12):
        r = spearman_brown(0.60, k)
        n = math.ceil(n_per_arm(D, rho_xx=r, rho_base=0.60))
        visits = n * k
        if best is None or visits < best[1]:
            best = (k, visits)
        print("   {:2d} {:24.3f} {:11d} {:22d}".format(k, r, n, visits))
    print(
        "\n   Total visits are minimised at k = {} ({} visits per arm): extra"
        .format(best[0], best[1])
    )
    print(
        "   occasions buy reliability along a flattening Spearman-Brown curve"
        " while"
    )
    print(
        "   costing visits linearly. But if the scarce resource is PEOPLE"
        " rather"
    )
    print(
        "   than visits - the usual case when recruitment is the bottleneck"
        " - the"
    )
    print(
        "   ranking reverses: k = 2 cuts the sample from {} to {} per arm"
        " for one".format(
            math.ceil(n_per_arm(D, spearman_brown(0.60, 1), rho_base=0.60)),
            math.ceil(n_per_arm(D, spearman_brown(0.60, 2), rho_base=0.60)),
        )
    )
    print(
        "   extra visit each. The Chapter 8 protocol takes two occasions for"
        " that"
    )
    print(
        "   reason, and states the assumption rather than leaving it"
        " implicit."
    )

    print(
        "\nThe same arithmetic bounds the autonomic secondary outcomes."
        " Programme"
    )
    print(
        "p05 showed that RMSSD moves with breathing pattern; if a measure's"
    )
    print(
        "session-to-session reliability is 0.4 in free breathing, then even a"
        " true"
    )
    print(
        "0.35 SD effect appears as {:.2f} SD and needs {} per arm - more"
        " than the".format(
            D * math.sqrt(0.4), math.ceil(n_per_arm(D, 0.40, rho_base=0.60))
        )
    )
    print(
        "primary outcome. That is a second reason, besides the"
        " identifiability"
    )
    print(
        "argument of p09, for not making an autonomic index the primary"
        " outcome."
    )

    print("\nWhat reliability is needed for a given trial size? Inverting:")
    print(
        "   n per arm available   minimum reliability for 90 % power at"
        " 0.35 SD"
    )
    for n in (60, 80, 100, 140, 200):
        za, zb = Z(0.975), Z(0.90)
        need = 2 * (za + zb) ** 2 * (1 - 0.60**2) / ((n - za**2 / 4) * D**2)
        print("   {:19d}   {:.3f}{}".format(n, need,
              "  (not attainable: exceeds 1)" if need > 1 else ""))
# Smallest effect of interest: 0.35 SD. ANCOVA, two-sided 0.05, 90 % power,
# baseline-outcome correlation 0.60.
#
#    reliability   attenuated d   n per arm   n for two arms
#           1.00          0.350         111              222
#           0.90          0.332         123              246
#           0.80          0.313         139              278
#           0.70          0.293         158              316
#           0.60          0.271         184              368
#           0.50          0.247         221              442
#
# Halving reliability from 0.90 to 0.50 nearly doubles the trial
# (123 -> 221 per arm).
#
# Add occasions or add people? Single-occasion reliability 0.60; the
# outcome is the mean of k occasions, each costing one visit.
#
#    k   reliability of the mean   n per arm   total visits per arm
#     1                    0.600         184                    184
#     2                    0.750         148                    296
#     3                    0.818         136                    408
#     4                    0.857         130                    520
#     6                    0.900         123                    738
#     8                    0.923         120                    960
#    12                    0.947         117                   1404
#
#    Total visits are minimised at k = 1 (184 visits per arm): extra
#    occasions buy reliability along a flattening Spearman-Brown curve while
#    costing visits linearly. But if the scarce resource is PEOPLE rather
#    than visits - the usual case when recruitment is the bottleneck - the
#    ranking reverses: k = 2 cuts the sample from 184 to 148 per arm for one
#    extra visit each. The Chapter 8 protocol takes two occasions for that
#    reason, and states the assumption rather than leaving it implicit.
#
# The same arithmetic bounds the autonomic secondary outcomes. Programme
# p05 showed that RMSSD moves with breathing pattern; if a measure's
# session-to-session reliability is 0.4 in free breathing, then even a true
# 0.35 SD effect appears as 0.22 SD and needs 276 per arm - more than the
# primary outcome. That is a second reason, besides the identifiability
# argument of p09, for not making an autonomic index the primary outcome.
#
# What reliability is needed for a given trial size? Inverting:
#    n per arm available   minimum reliability for 90 % power at 0.35 SD
#                     60   1.860  (not attainable: exceeds 1)
#                     80   1.389  (not attainable: exceeds 1)
#                    100   1.109  (not attainable: exceeds 1)
#                    140   0.790
#                    200   0.552

信度直接设定了上界,因为观测到的标准化效应被衰减为 dobs=dρxxd_{\text{obs}} = d\sqrt{\rho_{xx}},这是经典测验理论中的衰减公式 (Nunnally and Bernstein 1994):在 0.35 个标准差的最小关注效应下,信度 1.00 需要每臂 111 例,信度 0.50 需要 221 例。关于「加人还是加测量次数」,答案取决于哪种资源稀缺:按总访视数计,k=1k = 1 最省;按人数计,k=2k = 2 把每臂从 184 降到 148,只多一次访视。本方案取两次,并把这个假设写明。最后一段给出一个有用的反向计算:每臂只有 100 例时,需要的信度超过 1,也就是无论测量多好都不可能达到 90% 把握度——这类试验在规模不足时不应开展。

推导

# Chapter 8. Sample size for the pre-registered two-arm trial. Single primary
# comparison: non-judgmental interoceptive attention against a structurally
# matched external-focus control. Single primary outcome: body-vigilance score
# at week 6 by ANCOVA (baseline score, site). Planning assumptions: SESOI 0.35
# SD, baseline-outcome correlation 0.60, single-occasion reliability 0.75,
# outcome = mean of 2 occasions, two-sided 0.05, 90 % power. Normal theory
# first, then a simulation with an integer-scored scale, a floor, and the
# screening cut-off that selects participants into the trial.
import math
import random
from statistics import NormalDist

Z = NormalDist().inv_cdf
D, RHO_BASE, REL1, K_OCC = 0.35, 0.60, 0.75, 2
ITEMS, ITEM_MAX = 12, 6                  # 12 items scored 0-6: range 0-72
# screening cut-off for elevated vigilance
SCREEN_MIN = 28

def reliability(k):
    return k * REL1 / (1 + (k - 1) * REL1)

def n_formula(d=D, rho_xx=1.0, rho_base=RHO_BASE, power=0.90, alpha=0.05):
    d_obs = d * math.sqrt(rho_xx)
    za, zb = Z(1 - alpha / 2), Z(power)
    return 2 * (za + zb) ** 2 * (1 - rho_base ** 2) / d_obs ** 2 + za ** 2 / 4

def invert3(S):
    a, b, c = S[0]
    d, e, f = S[1]
    g, h, i = S[2]
    det = a * (e * i - f * h) - b * (d * i - f * g) + c * (d * h - e * g)
    return [
        [(e * i - f * h) / det, (c * h - b * i) / det, (b * f - c * e) / det],
        [(f * g - d * i) / det, (a * i - c * g) / det, (c * d - a * f) / det],
        [(d * h - e * g) / det, (b * g - a * h) / det, (a * e - b * d) / det],
    ]

# population before screening, scale points
TRUE_MEAN, TRUE_SD = 26.0, 9.0

def one_trial(rng, n, delta_sd):
    """delta_sd is the true effect in SD units, applied to the treated arm."""
    rows = []
    delta = delta_sd * TRUE_SD
    err_sd = TRUE_SD * math.sqrt(1 - RHO_BASE ** 2)
    meas_sd = TRUE_SD * math.sqrt((1 - REL1) / REL1 / K_OCC)
    for arm in (1, 0):
        got = 0
        while got < n:
            lat_b = rng.gauss(TRUE_MEAN, TRUE_SD)
            obs_b = lat_b + rng.gauss(0, meas_sd)
            b = min(ITEMS * ITEM_MAX, max(0, round(obs_b)))
            if b < SCREEN_MIN:
                continue                 # screened out, as in the real trial
            lat_y = (TRUE_MEAN + RHO_BASE * (lat_b - TRUE_MEAN)
                     - delta * arm + rng.gauss(0, err_sd))
            y = min(
                ITEMS * ITEM_MAX, max(0, round(lat_y + rng.gauss(0, meas_sd)))
            )
            rows.append((arm, b, y))
            got += 1
    n2 = len(rows)
    mb = math.fsum(r[1] for r in rows) / n2
    X = [(1.0, float(r[0]), r[1] - mb) for r in rows]
    Y = [float(r[2]) for r in rows]
    S = [
        [math.fsum(x[i] * x[j] for x in X) for j in range(3)]
        for i in range(3)
    ]
    v = [math.fsum(x[i] * y for x, y in zip(X, Y)) for i in range(3)]
    inv = invert3(S)
    beta = [math.fsum(inv[i][j] * v[j] for j in range(3)) for i in range(3)]
    res = [
        y - math.fsum(bb * xx for bb, xx in zip(beta, x))
        for x, y in zip(X, Y)
    ]
    s2 = math.fsum(r * r for r in res) / (n2 - 3)
    df = n2 - 3
    z = Z(0.975)
    # Cornish-Fisher t quantile
    tcrit = z + (z ** 3 + z) / (4 * df)
    # sign: lower score is better
    t = -beta[1] / math.sqrt(s2 * inv[1][1])
    at_floor = sum(1 for r in rows if r[0] == 1 and r[2] == 0) / n
    return t > tcrit, at_floor

if __name__ == "__main__":
    rel = reliability(K_OCC)
    print(
        "outcome = mean of {} occasions; single-occasion reliability {:.2f}"
        " -> {:.3f}".format(K_OCC, REL1, rel)
    )
    n0 = n_formula(rho_xx=rel)
    print("normal theory: n = {:.3f} -> {} per arm (attenuated d = {:.3f})"
          .format(n0, math.ceil(n0), D * math.sqrt(rel)))

    rng = random.Random(2026)
    chosen = None
    print("\nsimulation with integer scores, a floor at 0 and the screening")
    print("cut-off at {}:".format(SCREEN_MIN))
    for n in (math.ceil(n0), 150, 165, 180):
        res = [one_trial(rng, n, D) for _ in range(1500)]
        pw = sum(r[0] for r in res) / len(res)
        se = math.sqrt(pw * (1 - pw) / len(res))
        fl = math.fsum(r[1] for r in res) / len(res)
        print(
            "   n = {:3d}/arm: power {:.3f} (s.e. {:.3f}); treated arm at"
            " floor {:.1%}".format(n, pw, se, fl)
        )
        if chosen is None and pw - 2 * se >= 0.90:
            chosen = n
    if chosen is None:
        chosen = 180
    t1 = sum(one_trial(rng, chosen, 0.0)[0] for _ in range(1500)) / 1500
    print("   smallest n with power - 2 s.e. >= 0.90: {}".format(chosen))
    print(
        "   error rate there with no effect: {:.3f} against a nominal 0.025,"
        " since".format(t1)
    )
    print(
        "   the simulated rejection counts only the favourable direction"
        " against a"
    )
    print(
        "   two-sided critical value; the reported trial test is two-sided at"
        " 0.05."
    )

    plan = math.ceil(chosen / 0.85)
    print(
        "\nplanned: {} analysable per arm; 15 % loss -> {} randomised"
        "\nper arm, {} in all".format(chosen, plan, 2 * plan)
    )

    print(
        "\nsensitivity, analysable per arm (normal theory, reliability"
        " {:.2f}):".format(rel)
    )
    print("   SESOI \\ rho_base      0.4     0.6     0.8")
    for d in (0.25, 0.35, 0.50):
        print("   {:4.2f} SD          ".format(d) + "".join(
            "{:8d}".format(math.ceil(n_formula(d=d, rho_xx=rel, rho_base=r)))
            for r in (0.4, 0.6, 0.8)))

    print(
        "\nThe trial is powered for the PRIMARY comparison only. The"
        " pre-specified"
    )
    print(
        "negative-control outcome (a measure the framework predicts"
        " should NOT"
    )
    print(
        "change) is reported with its interval and is not powered; its role"
        " is to"
    )
    print(
        "show whether a generic expectancy effect moved everything, and a"
        " change"
    )
    print(
        "there of similar size to the primary effect voids the primary claim."
    )
# outcome = mean of 2 occasions; single-occasion reliability 0.75 -> 0.857
# normal theory: n = 129.051 -> 130 per arm (attenuated d = 0.324)
#
# simulation with integer scores, a floor at 0 and the screening
# cut-off at 28:
#    n = 130/arm: power 0.847 (s.e. 0.009); treated arm at floor 0.1%
#    n = 150/arm: power 0.893 (s.e. 0.008); treated arm at floor 0.1%
#    n = 165/arm: power 0.915 (s.e. 0.007); treated arm at floor 0.1%
#    n = 180/arm: power 0.946 (s.e. 0.006); treated arm at floor 0.1%
#    smallest n with power - 2 s.e. >= 0.90: 165
#    error rate there with no effect: 0.021 against a nominal 0.025, since
#    the simulated rejection counts only the favourable direction against a
#    two-sided critical value; the reported trial test is two-sided at 0.05.
#
# planned: 165 analysable per arm; 15 % loss -> 195 randomised
# per arm, 390 in all
#
# sensitivity, analysable per arm (normal theory, reliability 0.86):
#    SESOI \ rho_base      0.4     0.6     0.8
#    0.25 SD               331     253     143
#    0.35 SD               170     130      74
#    0.50 SD                84      64      37
#
# The trial is powered for the PRIMARY comparison only. The pre-specified
# negative-control outcome (a measure the framework predicts should NOT
# change) is reported with its interval and is not powered; its role is to
# show whether a generic expectancy effect moved everything, and a change
# there of similar size to the primary effect voids the primary claim.

正态理论给出每臂 129.051,取整 130。模拟随后加入整数计分、0 的下限与筛查切点:每臂 130 例时把握度 0.847、150 例 0.893、165 例 0.915、180 例 0.946;按「把握度减两个标准误不低于 0.90」的规则取每臂 165 例可分析。按 15% 的失访与无法评估率,每臂随机 195 例,两臂共 390 例。敏感性表显示:最小关注效应取 0.25 个标准差时每臂需 253 例;若基线与结局的相关只有 0.40,0.35 个标准差的效应需 170 例。

按第 33 本第六章的结果,「90% 把握度」指的是在真实效应恰好等于 0.35 个标准差时,得出「练习较好」(置信区间排除零)这类结论的概率约为 90%;在这种情况下,最强的结论「至少高出 0.35 个标准差」只有约 2.5% 的机会出现。

判定规则

推导

# Chapter 8. Decision rule, fixed before unblinding. Primary comparison:
# non-judgmental interoceptive attention versus the structurally matched
# external-focus control. D = adjusted difference in the week-6 body-vigilance
# score (ANCOVA: baseline, site), expressed in SD units, with its 95 %
# confidence interval. NEGATIVE D favours the practice (a lower score is
# better). SESOI 0.35 SD. Gates, checked before anything else: more than 20 %
# of randomised participants without a primary outcome; allocation revealed to
# outcome assessors before database lock. Pre-specified qualifiers: the
# credibility/expectancy check (if the two arms differ by more than 1 point on
# a 9-point credibility scale, expectancy is not matched) and the
# NEGATIVE-CONTROL outcome (if it moved by at least half the primary effect, a
# generic expectancy effect is the simpler explanation). An adverse-event
# excess in either arm is reported first, whatever the efficacy result.
SESOI = 0.35

def verdict(d, lo, hi, missing=0.08, cred_gap=0.3, neg_ctrl=0.05,
            revealed=False, ae_excess=False):
    pre = "ADVERSE-EVENT SIGNAL REPORTED FIRST;\n" if ae_excess else ""
    for bad, why in (
        (
            missing > 0.20,
            "{:.0%} without primary outcome (limit 20 %)".format(missing),
        ),
        (revealed, "allocation revealed to assessors before lock"),
    ):
        if bad:
            return "{}D = {:+.2f}: NOT INTERPRETABLE - {}".format(pre, d, why)
    if hi <= -SESOI:
        v = "benefit of at least the SESOI"
    elif hi < 0:
        v = "benefit, but not shown to reach the SESOI"
    elif lo > 0:
        v = "the control arm did better"
    elif lo >= -SESOI and hi <= SESOI:
        v = "no difference worth acting on"
    else:
        v = "unclear: the interval is too wide to decide"
    notes = []
    if cred_gap > 1.0:
        notes.append(
            "credibility differed by {:.1f} points: expectancy is not"
            "\n      matched, so part of D may be expectancy".format(cred_gap)
        )
    # the negative control only qualifies a claimed benefit; with no benefit
    # to explain away there is nothing for it to undercut
    if hi < 0 and abs(d) > 1e-9 and abs(neg_ctrl) >= 0.5 * abs(d):
        notes.append(
            "negative control moved {:+.2f} SD, at least half of D:"
            "\n      a generic expectancy effect is the simpler explanation"
            .format(neg_ctrl)
        )
    out = "{}D = {:+.2f} ({:+.2f} to {:+.2f}): {}".format(pre, d, lo, hi, v)
    for n in notes:
        out += "\n    qualifier: " + n
    return out

CASES = [
    # clear benefit past SESOI
    (-0.52, -0.80, -0.38),
    # benefit, below SESOI
    (-0.22, -0.41, -0.03),
    # nothing worth acting on
    (-0.05, -0.24, 0.14),
    # control did better
    (0.30, 0.09, 0.51),
    (-0.20, -0.70, 0.30),                                    # too wide
    # expectancy not matched
    (-0.48, -0.74, -0.22, 0.08, 2.4),
    # negative control moved
    (-0.44, -0.70, -0.18, 0.08, 0.3, -0.26),
    # too much missing data
    (-0.50, -0.78, -0.22, 0.27),
    # adverse-event signal
    (-0.10, -0.29, 0.09, 0.08, 0.3, 0.02, False, True),
]

if __name__ == "__main__":
    for c in CASES:
        print(verdict(*c))
    print(
        "\nSecondary outcomes (the autonomic indices, the imagery-stream"
        " entropy of"
    )
    print(
        "p13, the questionnaire subscales) are reported with intervals and"
        " never"
    )
    print(
        "change the verdict. Per programme p09, no autonomic index is read as"
    )
    print(
        "evidence about visceral perfusion; per p05, every autonomic index is"
    )
    print(
        "reported alongside the respiration rate recorded in the same window."
    )
    print(
        "\nThe example numbers above are invented, to show what each branch"
        " prints."
    )
# D = -0.52 (-0.80 to -0.38): benefit of at least the SESOI
# D = -0.22 (-0.41 to -0.03): benefit, but not shown to reach the SESOI
# D = -0.05 (-0.24 to +0.14): no difference worth acting on
# D = +0.30 (+0.09 to +0.51): the control arm did better
# D = -0.20 (-0.70 to +0.30): unclear: the interval is too wide to decide
# D = -0.48 (-0.74 to -0.22): benefit, but not shown to reach the SESOI
#     qualifier: credibility differed by 2.4 points: expectancy is not
#       matched, so part of D may be expectancy
# D = -0.44 (-0.70 to -0.18): benefit, but not shown to reach the SESOI
#     qualifier: negative control moved -0.26 SD, at least half of D:
#       a generic expectancy effect is the simpler explanation
# D = -0.50: NOT INTERPRETABLE - 27% without primary outcome (limit 20 %)
# ADVERSE-EVENT SIGNAL REPORTED FIRST;
# D = -0.10 (-0.29 to +0.09): no difference worth acting on
#
# Secondary outcomes (the autonomic indices, the imagery-stream entropy of
# p13, the questionnaire subscales) are reported with intervals and never
# change the verdict. Per programme p09, no autonomic index is read as
# evidence about visceral perfusion; per p05, every autonomic index is
# reported alongside the respiration rate recorded in the same window.
#
# The example numbers above are invented, to show what each branch prints.

前置条件中任何一项不满足(超过 20% 的随机参与者没有主要结局、结果锁定前分组向评估者泄露),无论主要结局如何,报告都写「不可解释」。任一臂严重不良事件较多时,安全信号写在最前面。只有当置信区间上限不高于 −0.35-0.35 个标准差时,报告才写「获益至少达到最小关注效应」;上限低于 0 但高于 −0.35-0.35 时写「有获益,但未显示达到最小关注效应」;下限高于 0 时写「对照臂较好」;区间落在 ±0.35\pm 0.35 之内时写「没有值得据此改变做法的差别」;其余情形写「无法判断」。两条限定语:可信度相差超过 1 分时,预期未匹配;阴性对照结局的变化达到主要效应的一半时,泛化的预期效应是更简单的解释。

这项试验不能说明什么

第一,它只检验一种指导语、一种时长、一种为期 6 周的安排,不能推广到别的练习。第二,它完全不检验第四、五、六章所讨论的生理链条:它没有测量内脏或盆腔的血流,因此不能支持也不能否证关于内脏修复的任何说法。想检验那一部分,需要与心脏观测同时进行的内脏床区直接测量(第六章),而本方案不包含这种测量。第三,参与者不能对练习内容设盲,所以预期的贡献只能通过阴性对照与可信度测量部分地识别,不能排除。第四,结论不能推及有躯体疾病、惊恐障碍或创伤后应激障碍的人群,这些人被排除在外。第五,6 周结局不回答长期效果。第六,如果结果是「没有值得据此改变做法的差别」,那就是本试验的结论,不能用次要结局、亚组或自主神经指标替代。

伦理与利益冲突

把注意力集中于身体内部对一部分人会加重症状 (Paulus and Stein 2010; Khalsa et al. 2018),所以不良事件的采集必须主动进行,特别包括症状加重、惊恐发作与解离体验;设立独立的数据与安全监查委员会;参与者可随时退出;任何在试验期间出现躯体症状者转诊临床评估并退出干预。无论结果正负,方案、统计分析计划、分析代码与汇总结果都将公开,去标识化的个体数据按数据共享声明提供;作者不参与招募、干预、评估与分析。

利益披露

作者以 5D Library 的著作(包括本书)作为其社群课程、升维牌卡(5D Ascension Cards,Dimensional Shift Card Decks)与 Yafu AI 知识库的方法论基础,因此与本书主题有商业利益关系。作者并在社群课程中讲授呼吸、聆听与注意力练习类内容,本书讨论的正是这一类练习的机制与检验方法,利益关系直接。本书不推销任何课程、牌卡或服务,不给出任何练习处方、时长或疗程建议;第八章的试验设计为由独立于作者企业的研究人员实施和分析,不从作者的社群、学员或客户中招募;书中所有代码与研究方案均公开,供任何人检验。作者不提供医疗服务,不销售任何器械、补充剂或检测。

本章小结

四步核查表从链条、解剖、可识别性与证据形式四个方面阅读「注意练习改善生理状态」的说法;用于本书开头的链条时,它显示第一环方向相反、第三环不增加信息、第四环是事实错误、而关键参数不可由所用测量识别,个案证据的似然比约 2.2。预注册的两臂试验以练习对结构匹配的主动对照为唯一主要比较,以第 6 周身体警觉分数的协方差分析为唯一主要结局,最小关注效应 0.35 个标准差,每臂 165 例可分析、195 例随机,共 390 例;结局在另一天的两次独立评估中测量,并预先指定一个阴性对照结局与可信度/预期的测量,判定规则写成代码。自主神经指标与表象流熵只作次要结局,并按前面各章的限定报告。该试验不检验内脏血流的那一段链条,书中写明这一点。

参考文献

Chan, An-Wen, Jennifer M. Tetzlaff, Douglas G. Altman, Andreas Laupacis, Peter C. Gøtzsche, Karmela Krleža-Jerić, Asbjørn Hróbjartsson, et al. 2013. “SPIRIT 2013 Statement: Defining Standard Protocol Items for Clinical Trials.” Annals of Internal Medicine 158 (3): 200–207. https://doi.org/10.7326/0003-4819-158-3-201302050-00583.
Devilly, Grant J., and Thomas D. Borkovec. 2000. “Psychometric Properties of the Credibility/Expectancy Questionnaire.” Journal of Behavior Therapy and Experimental Psychiatry 31 (2): 73–86. https://doi.org/10.1016/S0005-7916(00)00012-4.
Goyal, Madhav, Sonal Singh, Erica M. S. Sibinga, Neda F. Gould, Anastasia Rowland-Seymour, Ritu Sharma, Zackary Berger, et al. 2014. “Meditation Programs for Psychological Stress and Well-Being: A Systematic Review and Meta-Analysis.” JAMA Internal Medicine 174 (3): 357–68. https://doi.org/10.1001/jamainternmed.2013.13018.
Khalsa, Sahib S., Ralph Adolphs, Oliver G. Cameron, Hugo D. Critchley, Paul W. Davenport, Justin S. Feinstein, Jamie D. Feusner, et al. 2018. “Interoception and Mental Health: A Roadmap.” Biological Psychiatry: Cognitive Neuroscience and Neuroimaging 3 (6): 501–13. https://doi.org/10.1016/j.bpsc.2017.12.004.
Lakens, Daniël, Anne M. Scheel, and Peder M. Isager. 2018. “Equivalence Testing for Psychological Research: A Tutorial.” Advances in Methods and Practices in Psychological Science 1 (2): 259–69. https://doi.org/10.1177/2515245918770963.
Mehling, Wolf E., Cynthia Price, Jennifer J. Daubenmier, Mike Acree, Elizabeth Bartmess, and Anita Stewart. 2012. “The Multidimensional Assessment of Interoceptive Awareness (MAIA).” PLOS ONE 7 (11): e48230. https://doi.org/10.1371/journal.pone.0048230.
Norman, Geoffrey R., Jeff A. Sloan, and Kathleen W. Wyrwich. 2003. “Interpretation of Changes in Health-Related Quality of Life: The Remarkable Universality of Half a Standard Deviation.” Medical Care 41 (5): 582–92. https://doi.org/10.1097/01.MLR.0000062554.74615.4C.
Nunnally, Jum C., and Ira H. Bernstein. 1994. Psychometric Theory. 3rd ed. New York: McGraw-Hill.
Paulus, Martin P., and Murray B. Stein. 2010. “Interoception in Anxiety and Depression.” Brain Structure and Function 214: 451–63. https://doi.org/10.1007/s00429-010-0258-9.
Schulz, Kenneth F., Douglas G. Altman, and David Moher. 2010. “CONSORT 2010 Statement: Updated Guidelines for Reporting Parallel Group Randomised Trials.” BMJ 340: c332. https://doi.org/10.1136/bmj.c332.
Senn, Stephen. 2006. “Change from Baseline and Analysis of Covariance Revisited.” Statistics in Medicine 25 (24): 4334–44. https://doi.org/10.1002/sim.2682.
Van Dam, Nicholas T., Marieke K. van Vugt, David R. Vago, Laura Schmalzl, Clifford D. Saron, Andrew Olendzki, Ted Meissner, et al. 2018. “Mind the Hype: A Critical Evaluation and Prescriptive Agenda for Research on Mindfulness and Meditation.” Perspectives on Psychological Science 13 (1): 36–61. https://doi.org/10.1177/1745691617709589.
Vickers, Andrew J., and Douglas G. Altman. 2001. “Analysing Controlled Trials with Baseline and Follow up Measurements.” BMJ 323 (7321): 1123–24. https://doi.org/10.1136/bmj.323.7321.1123.
Zamariola, Giorgia, Pierre Maurage, Olivier Luminet, and Olivier Corneille. 2018. “Interoceptive Accuracy Scores from the Heartbeat Counting Task Are Problematic: Evidence from Simple Bivariate Correlations.” Biological Psychology 137: 12–17. https://doi.org/10.1016/j.biopsycho.2018.06.006.
Zeman, Adam, Michaela Dewar, and Sergio Della Sala. 2015. “Lives Without Imagery — Congenital Aphantasia.” Cortex 73: 378–80. https://doi.org/10.1016/j.cortex.2015.05.019.

Fan, Y. (2026). Interoceptive Decoupling, Salience-Network Switching and Autonomic Visceral Restoration: From Precision-Weighting in the Default Mode Network and the Redistribution of Vagal Tone to the Neurophenomenology of a Non-Judgmental Imagery Stream, with a Textual Examination of neiguan in the Huangdi Neijing [内感受解离、突显网络切换与内脏自主神经修复] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24020418 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).