5D Library › 自主神经交感迷走平衡、六经辨证状态空间与多模态标志物反演 · English

第三章 体温节律漂移与核心至末梢温差

节律峰值要几天才看得准

核心体温有一条约二十四小时的节律,平均值(mesor)约 36.9 ℃,振幅约 0.40 ℃,峰值多在傍晚。余弦拟合(cosinor)用 24 小时周期的余弦与正弦做最小二乘,得到平均值、振幅与峰值时刻 (Cornelissen 2014)。真实记录比白噪声难读:峰值时刻每天自身就会游走,活动、进食与睡眠又带来缓慢而前后相关的扰动。p05 把这两种噪声都放进去。

推导

# Chapter 3. p05: Reading a temperature rhythm (cosinor). Core temperature
# is sampled every 10 minutes: mesor 36.9 deg C, amplitude 0.40 deg C,
# peak (acrophase) at 17:00. Real recordings are messier than white
# noise: the peak wanders from day to day (sd 40 min) and activity, meals
# and sleep add slow, correlated noise (AR(1), lag-1 correlation 0.95,
# sd 0.15 deg C). Least squares on cos and sin of the 24 h cycle gives mesor,
# amplitude and acrophase; their spread over 500 repeats (seed 86) shows
# how many days are needed before a 1-hour drift in the peak can be seen.
import random
from math import atan2, cos, pi, sin, sqrt

def series(rng, days, shift_h=0.0):
    out, e = [], rng.gauss(0, 0.15)
    for d in range(days):
        jit = rng.gauss(0, 40 / 60.0)
        for i in range(144):
            t = d * 24 + i / 6.0
            e = 0.95 * e + sqrt(1 - 0.95 ** 2) * rng.gauss(0, 0.15)
            out.append((t, 36.9 + 0.40 * cos(
                2 * pi * (t - 17 - shift_h - jit) / 24) + e))
    return out

def cosinor(data):
    n = len(data)
    c = [cos(2 * pi * t / 24) for t, _ in data]
    s = [sin(2 * pi * t / 24) for t, _ in data]
    y = [v for _, v in data]
    m = sum(y) / n
    # cos and sin are orthogonal over whole days
    b = 2 * sum(ci * (yi - m) for ci, yi in zip(c, y)) / n
    g = 2 * sum(si * (yi - m) for si, yi in zip(s, y)) / n
    amp = sqrt(b * b + g * g)
    peak = (atan2(g, b) * 24 / (2 * pi)) % 24
    return m, amp, peak

if __name__ == "__main__":
    rng = random.Random(86)
    print("   days  mean peak   sd of peak (min)  amplitude  1 h shift seen")
    for days in (1, 2, 3, 7):
        peaks, amps, seen = [], [], 0
        for _ in range(500):
            _, a, p = cosinor(series(rng, days))
            _, _, q = cosinor(series(rng, days, 1.0))
            peaks.append(p)
            amps.append(a)
            seen += (q - p) > 0.5
        mp = sum(peaks) / 500
        sd = sqrt(sum((p - mp) ** 2 for p in peaks) / 499)
        print("   {:4d}  {:9.2f}   {:16.1f}  {:9.3f}  {:14.3f}".format(
            days, mp, 60 * sd, sum(amps) / 500, seen / 500))
#    days  mean peak   sd of peak (min)  amplitude  1 h shift seen
#       1      16.94               60.7      0.414           0.652
#       2      16.99               41.9      0.401           0.728
#       3      17.03               35.5      0.399           0.710
#       7      16.99               23.5      0.396           0.836

只有一天的数据,峰值时刻的标准差是 60.7 分钟,一小时的峰值漂移只有 0.652 的机会被看出来;连续七天后标准差降到 23.5 分钟,看出的机会升到 0.836。振幅估计在各种天数下都接近 0.40(0.396 至 0.414)。结论很直接:用一两天的体温记录来判断「节律前移或后移」并不可靠,至少要一周。

核心至指尖温差需要连同室温一起量

末梢血管的收缩与舒张主要受交感神经控制,指尖皮肤温度因此可以作为血管运动张力的窗口 (Kräuchi et al. 1999)。一个简单的热平衡把指尖皮肤温度放在室温与核心体温之间,舒张程度 vv 越大,指尖越接近核心温度。

推导

# Chapter 3. p06: The core-to-fingertip temperature gradient as a window
# on vasomotor (sympathetic) tone, and why room temperature must be
# measured with it. A simple heat balance puts fingertip skin between room
# and core: T_skin = T_room + (T_core - T_room) * h, where h = 0.25 + 0.6 v
# rises with the dilation v (0 = clamped shut, 1 = wide open). Inverting
# a measured gradient for v with the room temperature known, or only
# assumed to be 25 deg C when it is really 22 to 28, over 4,000 readings
# with 0.3 deg C sensor noise. Seed 86.
import random
from math import sqrt

def skin(core, room, v):
    return room + (core - room) * (0.25 + 0.6 * v)

def invert(grad, core, room):
    h = 1 - grad / (core - room)
    return min(1.0, max(0.0, (h - 0.25) / 0.6))

if __name__ == "__main__":
    rng = random.Random(86)
    print("   dilation v  gradient at 22 / 25 / 28 deg C room")
    for v in (0.1, 0.5, 0.9):
        g = [37.0 - skin(37.0, r, v) for r in (22.0, 25.0, 28.0)]
        print("   {:10.1f}  {:5.2f} / {:5.2f} / {:5.2f}".format(v, *g))
    err_known = err_assumed = 0.0
    for _ in range(4000):
        v, room = rng.random(), rng.uniform(22.0, 28.0)
        g = 37.0 - skin(37.0, room, v) + rng.gauss(0, 0.3)
        err_known += (invert(g, 37.0, room) - v) ** 2
        err_assumed += (invert(g, 37.0, 25.0) - v) ** 2
    print("   rms error of v: room measured {:.3f}, room assumed {:.3f}"
          .format(sqrt(err_known / 4000), sqrt(err_assumed / 4000)))
#    dilation v  gradient at 22 / 25 / 28 deg C room
#           0.1  10.35 /  8.28 /  6.21
#           0.5   6.75 /  5.40 /  4.05
#           0.9   3.15 /  2.52 /  1.89
#    rms error of v: room measured 0.042, room assumed 0.115

同样的舒张程度 v=0.5v = 0.5,在 22 ℃ 的房间里温差是 6.75 ℃,在 28 ℃ 的房间里只有 4.05 ℃。反推 vv 时,如果室温与温差一起量,均方根误差是 0.042;如果只假定室温是 25 ℃,误差升到 0.115,几乎是三倍。

注

记录方案的最低要求: 体温节律至少连续记录七天;每次读取指尖温度时同时记录室温,并在说明中写明测量部位与适应时间。没有室温的「手脚冰凉」读数不能拿来比较不同的日子或不同的人。

小结

体温节律的峰值与振幅可以用余弦拟合读出,但峰值的日间游走与慢变噪声使一天的数据几乎无法判断一小时的漂移;核心至指尖温差能反映血管运动张力,前提是同时量室温。这两项都会在第四章作为六经状态模型的观测量。

参考文献

Cornelissen, Germaine. 2014. “Cosinor-Based Rhythmometry.” Theoretical Biology and Medical Modelling 11: 16.
Kräuchi, Kurt, Christian Cajochen, Esther Werth, and Anna Wirz-Justice. 1999. “Warm Feet Promote the Rapid Onset of Sleep.” Nature 401: 36 to 37.

Fan, Y. (2026). Autonomic Sympathovagal Balance, Six-Meridian State-Space Models and Multimodal Biomarker Inversion: Heart-Rate Variability Spectral Decomposition, Thermoregulatory Phase Drift and Bayesian Syndrome Classification [自主神经交感迷走平衡、六经辨证状态空间与多模态标志物反演] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090227 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).