python機器學習生物信息學系列課（博主錄制）：http://dwz.date/b9vw
項目聯系：QQ :231469242
https://wenku.baidu.com/view/8d506a9cda38376baf1fae84.html
sensitivity 敏感性：患病且檢測陽性概率
specificity特異性：健康且檢測陰性概率
ppv（positive predict value）陽性預測值:指機器檢查后，患病概率多高，這和疾病流行程度有關
npv（negative predict value）陰性預測值:指機器檢查后，非患病概率多高，這和疾病流行程度有關
但現實中機器會把很多正常人也檢查出陽性，所以沒病的人也可能檢查出陽性
所以陽性預測值和陰性預測值可以很好預測你檢查結果后，患病概率
陽性還和疾病流行程度有關，疾病是罕見病，即使機器檢測陽性，概率也不會很高
陽性預測率：當你被檢查出陽性時，你想知道你真的患病概率。
陽性預測率=患病且檢查出陽性數量/（患病且被診斷陽性數量+不患病且被診斷陽性數量）
陰性預測率：檢查陰性時，健康概率
陰性預測率=不患病且檢查出陰性數量/（患病且被診斷陰性數量+不患病且被診斷陰性數量）
案例中，總人口1000人（A+B+C+D），90%患病率，機器敏感度低50%，但陽性預測值高90%
如果疾病流行度很低，10%感染率，雖然敏感度50%，但陽性預測率只有10%
陰性時，用於排除疾病
用於前期篩選工具
敏感度和疾病流行程度無關
For example, pregnancy tests have a high sensitivity: when a woman is pregnant,
the probability that the test is positive is very high.
In contrast, an indicator for an attack with atomic weapons on the White House
should have a very high specificity: if there is no attack, the probability that the
indicator is positive should be very, very small.
While sensitivity and specificity characterize a test and are independent of
prevalence, they do not indicate what portion of patients with abnormal test results
are truly abnormal. This information is provided by the positive/negative predictive
value (PPV/NPV). These are the values relevant for a doctor diagnosing a patient:
when a patient has a positive test result, how likely is it that the patient is in fact sick?
Unfortunately, as Fig. 7.8 indicates, these values are affected by the prevalence of the
disease. The prevalence of a disease indicates how many out of 100,000 people are
affected by it; in contrast, the incidence gives the number of newly diagnosed cases
per 100,000 people. In summary, we need to know the prevalence of the disease as
well as the PPV/NPV of a test to provide a sensible interpretation of medical test
results.
Take for example a test where a positive test results implies a 50% chance
of having a certain medical condition. If half the population has this condition,
a positive test result tells the doctor nothing. But if the condition is very rare, a
positive test result indicates that the patient has a fifty–fifty chance of having this
rare condition—a piece of information that is very valuable.
Figure 7.8 shows how the prevalence of a disease affects the interpretation
of diagnostic results, with a test with a given specificity and sensitivity: a high
prevalence of the disease increases the PPV of the test, but decreases the NPV;
and a low prevalence does exactly the opposite. Figure 7.9 gives a worked example.
疾病流行率與陽性預測值關系密切
即使敏感度很高，但ppv也可能低，因為機器預測陽性錯誤的數量太多。所以敏感度也不是唯一判斷金指標
python金融風控評分卡模型和數據分析微專業課（博主親自錄制視頻）：http://dwz.date/b9vv
