关于「测量链:组织间液滞后、MARD 与传感器偏差」,《连续血糖曲线与最小模型》给出的定量结论是什么?

结论:第六章把平均绝对相对差换算成误差带:相对标准差 0.1429,在 150 mg/dL 处单读数误差带为正负 42.0。

数据口径:本书不是医疗建议,不宣称任何功效,不提供任何饮食、禁食或用药安排;任何自我试验须先咨询执业医师。 书中参数为示意;全部计算由书中程序生成。

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: 0.1429, 150, 42.0

(1) relative sd for MARD 0.1140.1429
95% band of one reading at 150 mg/dL: +/- 42.0 mg/dL
(2) mixed error: relative sd 0.1327 + additive sd 6 mg/dL
range (mg/dL) share of readings MARD
below 700.061 0.133
70-1800.816 0.114
above 1800.122 0.108
overall0.114
(3) sensor bias sd 0.10: ratio of two meals
same sensor: bias cancels; two sensors95% range 0.758 to 1.319
查看完整验算程序
# Chapter 6. p62: what a mean absolute relative difference (MARD) means.
# MARD = mean of |sensor - reference| / reference. 0.114 is the overall
# MARD reported by Bailey et al. 2015 for one factory-calibrated sensor
# against capillary reference (72 participants; 86.7 % of results in
# zone A of the consensus error grid); everything else is illustrative.
# (1) purely relative normal error: MARD = s * sqrt(2 / pi)
# (2) relative + additive error (6 mg/dL), tuned to the same overall
#     MARD on an illustrative day of reference values (log-normal,
#     median 120 mg/dL, sd of log 0.35): MARD by glucose range
# (3) a bias shared by all readings of one sensor cancels in a
#     within-sensor ratio of two meals but not across two sensors
import math
import random

TARGET = 0.114
s = TARGET / math.sqrt(2.0 / math.pi)
print("(1) relative sd for MARD %.3f: %.4f" % (TARGET, s))
print("    95%% band of one reading at 150 mg/dL: +/- %.1f mg/dL"
      % (1.96 * s * 150.0))
rng = random.Random(20261118)
N = 40000
REF = [120.0 * math.exp(rng.gauss(0.0, 0.35)) for _ in range(N)]
Z1 = [rng.gauss(0.0, 1.0) for _ in range(N)]
Z2 = [rng.gauss(0.0, 1.0) for _ in range(N)]


def mard(r, a, lo=0.0, hi=1e9):
    v = [abs(x * r * z1 + a * z2) / x for x, z1, z2 in zip(REF, Z1, Z2)
         if lo <= x < hi]
    return math.fsum(v) / len(v), len(v)


lo_r, hi_r = 0.0, 0.3
for _ in range(40):
    mid = (lo_r + hi_r) / 2.0
    if mard(mid, 6.0)[0] < TARGET:
        lo_r = mid
    else:
        hi_r = mid
r = (lo_r + hi_r) / 2.0
print("(2) mixed error: relative sd %.4f + additive sd 6 mg/dL" % r)
print("    range (mg/dL)   share of readings   MARD")
for lo, hi, name in ((0, 70, "below 70"), (70, 180, "70-180"),
                     (180, 1e9, "above 180")):
    m, n = mard(r, 6.0, lo, hi)
    print("    %-13s       %.3f            %.3f" % (name, n / N, m))
print("    overall                             %.3f" % mard(r, 6.0)[0])
sb = 0.10
sd_lr = math.sqrt(2.0) * sb
print("(3) sensor bias sd %.2f: ratio of two meals" % sb)
print("    same sensor: bias cancels; two sensors: 95%% range"
      " %.3f to %.3f" % (math.exp(-1.96 * sd_lr), math.exp(1.96 * sd_lr)))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《连续血糖曲线下面积、双室胰岛素最小模型与生酮代谢切换动力学母典》第 6 章「测量链:组织间液滞后、MARD 与传感器偏差」(版本 1.0.0). 5D Library. DOI 待发布

DOI 待发布 · 专著页正式发布后上线(本站暂无专著页) · 5DL-B-286 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026286,
  author    = {Fan, Yafu},
  title     = {Areas under Continuous Glucose Curves, the Two-Compartment Minimal Model of Insulin Action and the Dynamics of the Switch to Ketone Metabolism},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论