5D Library › 复合补剂相互作用、时间疗法给药窗口与代谢酶竞争 · English

第八章 依从性,以及那道闸

没被遵守的排程什么也不交付

前几章的每一个数字都假定排程被遵守了。Claxton、Cramer 与 Pierce 关于给药方案与实测依从性的系统综述,发现了每个开过门诊的人都会预料的事:依从性随每日服用次数的增加而下降,而下降最陡的是一天之中那些中段的次数,因为它们没有一顿饭或一次就寝可以依附 (Claxton et al. 2001)。一个忽略这件事的最优化,优化的是错的那个量。

证据

已发表。 Claxton、Cramer 与 Pierce 综述了给药方案与实测依从性之间的关系,发现依从性随每日服用次数增加而下降 (Claxton et al. 2001)。Ito 等人给出那个相互作用方程,其上界支配着哪些同服药物要紧 (Ito et al. 1998)。

推导

# Chapter 8. p15: A schedule that is not kept delivers nothing. The
# three-slot optimum of p11 scores 0.8605 when every slot is taken, and
# the single breakfast pile scores 0.5394. But a slot in the middle of a
# working day is the one people miss, and a slot that is missed delivers
# nothing at all for the items in it. The per-slot sums below are read
# off p11's solution. Thirty days are simulated for four thousand
# people, each slot taken or missed independently on each day, seed 96.
import random

SEED, PEOPLE, DAYS, N_ITEMS = 96, 4000, 30, 11
SLOT_SUM = (5.6499, 1.0000, 2.8160)     # from p11's best schedule
SLOT_NAME = ("breakfast", "afternoon", "evening")
ADHERE = (0.92, 0.62, 0.80)
ONE_SLOT_SUM = 0.5394 * N_ITEMS         # p11's all-at-breakfast score
ONE_SLOT_ADHERE = 0.95

def q(v, a):
    s = sorted(v)
    return s[int(a * (len(s) - 1))]

def run(rng, sums, adhere):
    got = 0.0
    for _ in range(DAYS):
        for s, p in zip(sums, adhere):
            if rng.random() < p:
                got += s
    return got / (DAYS * N_ITEMS)

if __name__ == "__main__":
    rng = random.Random(SEED)
    three = [run(rng, SLOT_SUM, ADHERE) for _ in range(PEOPLE)]
    one = [run(rng, (ONE_SLOT_SUM,), (ONE_SLOT_ADHERE,))
           for _ in range(PEOPLE)]
    print("   three slots, adherence {}: perfect score {:.4f}"
          .format(ADHERE, sum(SLOT_SUM) / N_ITEMS))
    print("   realised: median {:.4f}, 10% to 90% {:.4f} to {:.4f}"
          .format(q(three, 0.5), q(three, 0.1), q(three, 0.9)))
    print("   one slot, adherence {:.2f}: perfect score {:.4f}"
          .format(ONE_SLOT_ADHERE, ONE_SLOT_SUM / N_ITEMS))
    print("   realised: median {:.4f}, 10% to 90% {:.4f} to {:.4f}"
          .format(q(one, 0.5), q(one, 0.1), q(one, 0.9)))
    print("   the three-slot schedule still wins by {:+.4f}"
          .format(q(three, 0.5) - q(one, 0.5)))
    wins = sum(1 for a, b in zip(three, one) if b > a)
    print("   people for whom the one-slot pile did better: {} of {}"
          .format(wins, PEOPLE))
    never = (SLOT_SUM[0] + SLOT_SUM[2]) / N_ITEMS
    print("   three slots with the afternoon never taken at all:"
          " {:.4f}".format(never))
    print("   which is still above the one-slot perfect score of {:.4f},"
          .format(ONE_SLOT_SUM / N_ITEMS))
    print("   so no level of afternoon adherence makes the single pile")
    print("   the better choice, and the question is only whether the")
    print("   middle slot is worth keeping at all")
    print("   loss from adherence alone, three slots: {:.4f}"
          .format(1.0 - q(three, 0.5) / (sum(SLOT_SUM) / N_ITEMS)))
    print("   loss from adherence alone, one slot: {:.4f}"
          .format(1.0 - q(one, 0.5) / (ONE_SLOT_SUM / N_ITEMS)))
    print("   the middle slot carries only {:.4f} of the day's value"
          .format(SLOT_SUM[1] / sum(SLOT_SUM)))
    print("   so moving its single item into the evening would cost")
    print("   little and remove the slot people miss, which is the")
    print("   practical recommendation of this chapter")
#    three slots, adherence (0.92, 0.62, 0.8): perfect score 0.8605
#    realised: median 0.7362, 10% to 90% 0.6924 to 0.7735
#    one slot, adherence 0.95: perfect score 0.5394
#    realised: median 0.5214, 10% to 90% 0.4855 to 0.5394
#    the three-slot schedule still wins by +0.2148
#    people for whom the one-slot pile did better: 0 of 4000
#    three slots with the afternoon never taken at all: 0.7696
#    which is still above the one-slot perfect score of 0.5394,
#    so no level of afternoon adherence makes the single pile
#    the better choice, and the question is only whether the
#    middle slot is worth keeping at all
#    loss from adherence alone, three slots: 0.1445
#    loss from adherence alone, one slot: 0.0333
#    the middle slot carries only 0.1056 of the day's value
#    so moving its single item into the evening would cost
#    little and remove the slot people miss, which is the
#    practical recommendation of this chapter

在清晨依从 0.92、午后 0.62、傍晚 0.80 的情况下,纸面上得 0.8605 的三槽排程实现中位 0.7362,十分位到九十分位为 0.6924 到 0.7735。仅依从性一项造成的损失是 0.1445,而依从 0.95 的单一堆是 0.0333。依从性的代价大于本书中任何单一的相互作用。

然而它并不颠倒排序。即便午后那一槽从来没被服过,三槽排程仍交付 0.7696,高于一份完美的单一早餐堆所能交付的 0.5394,而四千个模拟的人里没有一个在单一堆上表现更好。结论不是排程没用,而是中间那一槽没用。它承载当日价值的 0.1056,接近它唯一那一项(锌)所贡献的量,而它正是人们会漏掉的那一槽。把这一项挪进傍晚,放弃的很少,却移除了那个失效点,这正是第三章已经以另一种理由提示过的。

假说 · 尚未验证

作者的诠释(尚未验证): 对补剂排程而言,对几乎所有人来说每日最优的时段数是两个,因为第三个时段的边际交付小于它所花掉的依从性;因此已发表的一日三次排程,相对它们本想改进的两槽版本,实际交付更少。什么会证明它错: 在一项以药片计数与状态标志物为终点、比较两槽与三槽排程的试验里,如果三槽那一组尽管计数依从性更低却交付了更多可测营养素,那么这里的边际计算就是错的。

谁不可以服用这个柜子

最后一项计算,是任何以本书为基础的产品都必须放在最前面的那一项。这些项目里有几项在常用药物存在时并不中性,而其中两对可能造成严重伤害。

推导

# Chapter 8. p16: Who must not take this cupboard. Several items in the
# stack are not neutral in the presence of common medicines, and two of
# the interactions can kill. The prevalences below are rounded from
# published population figures for adults in a developed country and
# are applied independently, which overstates how many people carry
# exactly one flag. For each flag the item it conflicts with is named,
# because the useful output is not a number but a list of pairs a
# reader can check against their own medicine cabinet. Twenty thousand
# synthetic adults, seed 96.
import random

SEED, N = 96, 20000
STOP = (
    ("warfarin",                 0.0130, "vitamin K2"),
    ("a direct oral anticoagulant", 0.0220, "high-dose fish oil"),
    ("kidney function below 30",  0.0060, "magnesium and potassium"),
    ("digoxin",                   0.0040, "calcium"),
    ("lithium",                   0.0030, "caffeine"),
    ("a thiazide diuretic",       0.0900, "calcium and vitamin D3"),
    ("haemochromatosis",          0.0030, "iron"),
)
CARE = (
    ("levothyroxine",             0.0700, "calcium, iron, magnesium"),
    ("a proton pump inhibitor",   0.0900, "iron and vitamin B12"),
    ("a tetracycline or quinolone", 0.0250, "calcium, zinc, iron, magnesium"),
    ("metformin",                 0.0700, "vitamin B12"),
    ("an SSRI",                   0.1100, "curcumin and high-dose fish oil"),
    ("a statin",                  0.2000, "curcumin through the same enzyme"),
    ("pregnant or breastfeeding", 0.0300, "vitamin A, iodine, caffeine"),
)

if __name__ == "__main__":
    rng = random.Random(SEED)
    stop_n = care_n = clear_n = multi = 0
    sc = [0] * len(STOP)
    cc = [0] * len(CARE)
    for _ in range(N):
        s = [rng.random() < p for _, p, _ in STOP]
        c = [rng.random() < p for _, p, _ in CARE]
        for i, v in enumerate(s):
            sc[i] += v
        for i, v in enumerate(c):
            cc[i] += v
        if any(s):
            stop_n += 1
            if sum(s) > 1:
                multi += 1
        elif any(c):
            care_n += 1
        else:
            clear_n += 1
    print("   {} synthetic adults".format(N))
    print("   absolute conflicts: what the person takes, and with what")
    for (name, p, item), k in zip(STOP, sc):
        print("   {:<28}{:<9.4f}{}".format(name, k / N, item))
    print("   conflicts of timing or dose, needing a doctor's view:")
    for (name, p, item), k in zip(CARE, cc):
        print("   {:<28}{:<9.4f}{}".format(name, k / N, item))
    print("   carrying at least one absolute conflict: {:.4f}"
          .format(stop_n / N))
    print("   of those, with more than one: {:.4f}".format(multi / stop_n))
    print("   needing a doctor's view but not an outright stop: {:.4f}"
          .format(care_n / N))
    print("   free to take the cupboard as written: {:.4f}"
          .format(clear_n / N))
    print("   so the gate stops or refers {:.4f} of adults"
          .format((stop_n + care_n) / N))
    worst = max(zip(STOP, sc), key=lambda z: z[1])
    print("   the commonest absolute conflict is {} at {:.4f}"
          .format(worst[0][0], worst[1] / N))
    print("   and it conflicts with {}".format(worst[0][2]))
    print("   the two that can kill are warfarin against vitamin K2,")
    print("   carried by {:.4f}, and an anticoagulant against a high"
          .format(sc[0] / N))
    print("   dose of fish oil, carried by {:.4f}".format(sc[1] / N))
    print("   together {:.4f} before allowing for people who have both"
          .format((sc[0] + sc[1]) / N))
#    20000 synthetic adults
#    absolute conflicts: what the person takes, and with what
#    warfarin                    0.0123   vitamin K2
#    a direct oral anticoagulant 0.0221   high-dose fish oil
#    kidney function below 30    0.0057   magnesium and potassium
#    digoxin                     0.0042   calcium
#    lithium                     0.0032   caffeine
#    a thiazide diuretic         0.0896   calcium and vitamin D3
#    haemochromatosis            0.0026   iron
#    conflicts of timing or dose, needing a doctor's view:
#    levothyroxine               0.0675   calcium, iron, magnesium
#    a proton pump inhibitor     0.0909   iron and vitamin B12
#    a tetracycline or quinolone 0.0246   calcium, zinc, iron, magnesium
#    metformin                   0.0694   vitamin B12
#    an SSRI                     0.1113   curcumin and high-dose fish oil
#    a statin                    0.1948   curcumin through the same enzyme
#    pregnant or breastfeeding   0.0274   vitamin A, iodine, caffeine
#    carrying at least one absolute conflict: 0.1346
#    of those, with more than one: 0.0368
#    needing a doctor's view but not an outright stop: 0.4017
#    free to take the cupboard as written: 0.4637
#    so the gate stops or refers 0.5363 of adults
#    the commonest absolute conflict is a thiazide diuretic at 0.0896
#    and it conflicts with calcium and vitamin D3
#    the two that can kill are warfarin against vitamin K2,
#    carried by 0.0123, and an anticoagulant against a high
#    dose of fish oil, carried by 0.0221
#    together 0.0344 before allowing for people who have both

在 20000 个合成成年人中,0.1346 至少带有一项绝对冲突,而在这些人里又有 0.0368 带有一项以上。另有 0.4017 带有需要医生判断而非禁止的情况,于是剩下 0.4637 可以照原样服用这个柜子。这道闸拦下或转诊了 0.5363 的成年人。

注

可能致命的那两对,以及最常见的那一对。 维生素 K2 直接对抗华法林,一个在服华法林的人如果不告诉任何人就加上它,可能失去抗凝;本样本中有 0.0123 在服华法林。大剂量鱼油与直接口服抗凝药并用会叠加出血风险;有 0.0221 在服后者。两者合计 0.0344,而这还没扣除同时具备两者的人。最常见的绝对冲突并不是这两者中的任一个,而是噻嗪类利尿剂,占 0.0896,它与钙及维生素 D3 并用可能升高血钙。最常见的时机冲突是左甲状腺素,它在肠道里与钙、铁、镁结合,必须与它们相隔数小时;它被列在转诊名单而不是停止名单里,只因为分开确实解决了它。

独立性假设高估了恰好带一项标志的人数,低估了带若干项的人数,因为这些情况会聚集。这并不让这道闸变小。真正要紧的那个数字,也就是被直接拦下的 0.1346,数的是带有任何一项标志的人,聚集只把它降低一点点,却让它真正拦下的那些人病得更重。

小结

依从性把最优三槽排程从 0.8605 拉到中位 0.7362,损失 0.1445,大于本书中算出的任何相互作用;但排程仍然胜出,因为即便午后那一槽从未被服过,它也交付 0.7696,对单一堆的 0.5394。中间那一槽只承载当日价值的 0.1056,应当被取消。安全闸直接拦下 0.1346 的成年人,另外转诊 0.4017,剩下 0.4637。所以本书的算术是对一个机制的描述与对其大小的一个界,而 p16 里那张冲突清单才是读者真正应当据以行动的部分,行动的方式是把它带给药师。

参考文献

Claxton, Ami J., Joyce Cramer, and Charles Pierce. 2001. “A Systematic Review of the Associations between Dose Regimens and Medication Compliance.” Clinical Therapeutics 23 (8): 1296 to 1310.
Ito, Kiyomi, Tomoyuki Iwatsubo, Satoshi Kanamitsu, Kunihiro Ueda, Hiroshi Suzuki, and Yuichi Sugiyama. 1998. “Prediction of Pharmacokinetic Alterations Caused by Drug-Drug Interactions: Metabolic Interaction in the Liver.” Pharmacological Reviews 50 (3): 387 to 412.

Fan, Y. (2026). Polypharmacy Supplement Interactions, Chronotherapy Dosing Windows and Cytochrome Enzyme Competition: Michaelis-Menten Inhibition Matrices, Chelation Absorption Losses and Circadian Receptor Sensitivity [复合补剂相互作用、时间疗法给药窗口与代谢酶竞争] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090250 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).