关于「实践与预注册检验」,《井穴揿针与卒中后认知》给出的定量结论是什么?

结论:第八章给出四步核查表与一项预注册三臂试验:唯一主要比较为真揿针对假揿针(两组都每天按揉三次),唯一主要结局为第 4 周 MoCA 的协方差分析,最小关注效应 1.5 分,每组 72 例可分析、85 例随机,共 255 例,前额 α 只作预先指定的中介分析,判定规则写成代码。书中以标注框区分已发表的证据、可运行的 Python 计算(16 段程序)、原典与作者的模型和假说;除注明的文献数据外,书中数据都是合成的。

数据口径:卒中后在双手十二井穴贴上长 0.9 毫米的揿针并每天按揉,能否借指端密集的感觉传入改善认知?本书是教学与研究用的数学模型与方法学讨论,不是医疗建议。 书中以标注框区分已发表的证据、可运行的 Python 计算(16 段程序)、原典与作者的模型和假说;除注明的文献数据外,书中数据都是合成的。

English version

对照数据表

下表数字由本章 Python 程序原样打印,可重新运行复现。 程序只用 Python 标准库与结论吻合的打印值: 1.5, 255, 72, 85

normal theory: n64.005 -> 65 per arm (residual SD 2.60)
simulated power, n65/arm: 0.894 (s.e. 0.007); treated arm at 30: 14%
simulated power, n68/arm: 0.898 (s.e. 0.007); treated arm at 30: 14%
simulated power, n72/arm: 0.919 (s.e. 0.006); treated arm at 30: 14%
simulated power, n76/arm: 0.930 (s.e. 0.006); treated arm at 30: 14%
smallest n with power - 2 s.e. >0.90: 72; type I error there 0.057
planned72 analysable per arm; 15 % loss -> 85 randomised per arm,
255 in three arms (11:1)
sensitivity, analysable per arm (normal theory):
SESOI \ rho0.3 0.5 0.7
1.0 points174 143 98
1.5 points78 65 44
2.0 points44 37 26
查看完整验算程序
# Chapter 8. Sample size, genuine vs sham press needles (both pressed 3 times a day).
# Primary outcome: MoCA at week 4 by ANCOVA (baseline, centre). Planning assumptions:
# SD 3.0, baseline-outcome correlation 0.5, SESOI 1.5 points, two-sided 0.05, power 90 %.
# n per arm = 2 (z_0.975 + z_0.90)^2 s^2 (1 - r^2) / d^2 + z_0.975^2 / 4, then checked by
# simulation with integer scores, the 18-25 inclusion window and the ceiling at 30.
import math, random, statistics

Z = statistics.NormalDist().inv_cdf
SD, RHO, D = 3.0, 0.5, 1.5

def n_formula(sd=SD, rho=RHO, d=D, power=0.90):
    za, zb = Z(0.975), Z(power)
    return 2 * (za + zb) ** 2 * sd ** 2 * (1 - rho ** 2) / d ** 2 + za ** 2 / 4

def one_trial(rng, n, delta, mean_post=25.0):
    rows = []
    for arm in (1, 0):
        for _ in range(n):
            while True:
                b = rng.gauss(21.5, 3.0); bo = round(b)
                if 18 <= bo <= 25:
                    break
            e = rng.gauss(0, SD * math.sqrt(1 - RHO ** 2))
            y = mean_post + delta * arm + RHO * (b - 21.5) + e
            rows.append((arm, bo, min(30, round(y))))
    n2 = len(rows); mx = math.fsum(r[1] for r in rows) / n2
    # ANCOVA via residualising on baseline within the pooled sample
    X = [(1.0, r[0], r[1] - mx) for r in rows]; Y = [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)
    z, df = Z(0.975), n2 - 3
    tcrit = z + (z ** 3 + z) / (4 * df)                 # t quantile, Cornish-Fisher
    return abs(beta[1] / math.sqrt(s2 * inv[1][1])) > tcrit, \
        sum(1 for r in rows if r[0] == 1 and r[2] == 30) / n

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]]

if __name__ == "__main__":
    n0 = n_formula()
    print(f"normal theory: n = {n0:.3f} -> {math.ceil(n0)} per arm"
          f" (residual SD {SD * math.sqrt(1 - RHO ** 2):.2f})")
    rng = random.Random(2026); chosen = None
    for n in (math.ceil(n0), 68, 72, 76):
        res = [one_trial(rng, n, D) for _ in range(2000)]
        pw = sum(r[0] for r in res) / len(res); se = math.sqrt(pw * (1 - pw) / 2000)
        top = math.fsum(r[1] for r in res) / len(res)
        print(f"   simulated power, n = {n}/arm: {pw:.3f} (s.e. {se:.3f});"
              f" treated arm at 30: {top:.0%}")
        if chosen is None and pw - 2 * se >= 0.90:
            chosen = n
    t1 = sum(one_trial(rng, chosen, 0.0)[0] for _ in range(2000)) / 2000
    print(f"   smallest n with power - 2 s.e. >= 0.90: {chosen}; type I error there"
          f" {t1:.3f}")
    plan = math.ceil(chosen / 0.85)
    print(f"planned: {chosen} analysable per arm; 15 % loss -> {plan} randomised"
          f" per arm,\n   {3 * plan} in three arms (1:1:1)")
    print("sensitivity, analysable per arm (normal theory):")
    print("   SESOI \\ rho        0.3    0.5    0.7")
    for d in (1.0, 1.5, 2.0):
        print(f"   {d:4.1f} points    " + "".join(
            f"{math.ceil(n_formula(rho=r, d=d)):7d}" for r in (0.3, 0.5, 0.7)))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《指—脑皮层放大、远端微针传入与卒中后认知网络重建的计算神经学》第 8 章「实践与预注册检验」(版本 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.23839915

DOI: 10.5281/zenodo.23839915 · 阅读本章所在专著 · 5DL-B-052 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026052,
  author    = {Fan, Yafu},
  title     = {Finger–Brain Cortical Magnification, Distal Microneedle Afference and Computational Neurology of Post-Stroke Cognitive Network Reconstruction},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  doi       = {10.5281/zenodo.23839915},
  url       = {https://doi.org/10.5281/zenodo.23839915}
}

本书其他章节结论