关于「两句问话:把红脸史当作一个分类器」,《红脸基因与宴请血醇曲线》给出的定量结论是什么?
结论:第六章把红脸问卷当作分类器,阳性预测值 0.8084。
数据口径:本书用纯标准库的程序把每一步写成可以核对的算术:参数凡未注明出处者均为示意,所有数字由书中程序生成。 本书不是医疗建议,不提供任何形式的安全饮用量;书中的血醇曲线用于安排用车与住宿,绝不可用于判断能否驾驶——饮酒之后不驾车。
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: 0.8084
flush questions against genotyping, prevalence 0.36
panel sens spec PPV NPV LR+
men, flush survey 0.90 0.88 0.8084 0.9399 7.50
women, flush survey 0.88 0.92 0.8609 0.9316 11.00
ethanol patch test 0.90 0.90 0.8351 0.9412 9.00
the same questions in populations with fewer carriers
prevalence PPV NPV wrong labels per 100 asked
0.05 0.2830 0.9941 11.90
0.10 0.4545 0.9875 11.80
0.20 0.6522 0.9724 11.60
0.36 0.8084 0.9399 11.28
0.50 0.8824 0.8980 11.00
a host screening a table of ten guests, prevalence 0.36
outcome expected guests
carrier, answers yes 3.2400
carrier, answers no 0.3600
non-carrier, answers yes 0.7680
non-carrier, answers no 5.6320
guests the questions would mislabel: 1.1280
the asymmetry that matters at a banquet
a yes costs the host one poured cup; a no that is wrong
costs a carrier the plateau of 24.00 uM for the evening
so the question is asked of everyone and the answer is
never required to be a yes查看完整验算程序
# Chapter 6. p61: The two-question flush survey as a classifier.
# Sensitivity 0.90 and specificity 0.88 for Japanese men, and 0.88
# and 0.92 for Japanese women, are the questionnaire figures quoted
# by Brooks and colleagues (2009, PLoS Med 6(3):e1000050) against
# genotyping. The prevalence scan covers East Asian populations; the
# 0.36 figure is the share of East Asians who flush, from the same
# paper. Everything below is Bayes' rule, so the arithmetic can be
# checked by hand.
import math
PREV = 0.36
PANELS = [("men, flush survey", 0.90, 0.88),
("women, flush survey", 0.88, 0.92),
("ethanol patch test", 0.90, 0.90)]
def predictive(prev, sens, spec):
tp = prev * sens
fp = (1.0 - prev) * (1.0 - spec)
fn = prev * (1.0 - sens)
tn = (1.0 - prev) * spec
return tp / (tp + fp), tn / (tn + fn)
print("flush questions against genotyping, prevalence {:.2f}"
.format(PREV))
print("panel sens spec PPV NPV LR+")
for name, se, sp in PANELS:
ppv, npv = predictive(PREV, se, sp)
print("{:21s} {:5.2f} {:5.2f} {:7.4f} {:7.4f} {:6.2f}"
.format(name, se, sp, ppv, npv, se / (1.0 - sp)))
print("")
print("the same questions in populations with fewer carriers")
print("prevalence PPV NPV wrong labels per 100 asked")
for pv in (0.05, 0.10, 0.20, 0.36, 0.50):
ppv, npv = predictive(pv, 0.90, 0.88)
wrong = 100.0 * (pv * 0.10 + (1.0 - pv) * 0.12)
print("{:10.2f} {:8.4f} {:8.4f} {:20.2f}"
.format(pv, ppv, npv, wrong))
print("")
print("a host screening a table of ten guests, prevalence 0.36")
print("outcome expected guests")
tp = 10.0 * PREV * 0.90
fn = 10.0 * PREV * 0.10
fp = 10.0 * (1.0 - PREV) * 0.12
tn = 10.0 * (1.0 - PREV) * 0.88
for label, v in (("carrier, answers yes", tp),
("carrier, answers no", fn),
("non-carrier, answers yes", fp),
("non-carrier, answers no", tn)):
print("{:30s} {:15.4f}".format(label, v))
print("guests the questions would mislabel: {:.4f}"
.format(fn + fp))
print("")
print("the asymmetry that matters at a banquet")
print("a yes costs the host one poured cup; a no that is wrong")
print("costs a carrier the plateau of {:.2f} uM for the evening"
.format(24.0))
print("so the question is asked of everyone and the answer is")
print("never required to be a yes")唯一数据源与引用
范雅福(Fan, Y.). (2026). 《乙醛脱氢酶二型红脸基因、酒量耐受致癌悖论与商务宴请血醇曲线母典》第 6 章「两句问话:把红脸史当作一个分类器」(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-191 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026191,
author = {Fan, Yafu},
title = {ALDH2 Flushing Genetics, the Tolerance-Carcinogenesis Paradox and Blood-Alcohol Curves at Business Banquets},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}