II. Definitions
- Factors affecting Positive Predictive Value
- Prevalence of disease
- Accuracy of test
III. Types: Bias
IV. Criteria: Ideal Screening Test
- Disease Features
- Disease significantly impacts public health
- Intermediate probability of disease
- Detection occurs before a critical point
- Critical point occurs before clinical diagnosis
- Screened patient is still asymptomatic
- Diagnosis would not otherwise occur this early
- Endometrial Cancer's critical point is too late
- Critical point occurs in time to affect outcome
- Disease must be detected early enough for cure
- Lung Cancer's critical point occurs too early
- Critical point occurs before clinical diagnosis
- Test Features
- Screening Test tolerated by patients
- High Test Sensitivity to detect asymptomatic disease
- Best quality for Screening Test
- Test criterion or threshold is set low to minimize False Negatives (at the expense of increased False Positives)
- High Test Specificity
- Best quality for confirmatory test
- Test criterion or threshold is set high to minimize False Positives (at the expense of increased False Negatives)
- Screened Population Features
- Disease has high enough Prevalence to allow screening
- Medical care available if Screening Test positive
- Patient willing to undergo further evaluation
V. Evaluation
- Assume you know the disease state
- Test Sensitivity (Test Recall)
- Given the patient has the disease,
- What is the probability of a true positive test
- Test Specificity
- Given the patient does not have the disease,
- What is the probability of a true negative test
- Probability of test given disease
- P(test | disease) = Test Sensitivity / Test Specificity
- Where P (A | B) = Probability of A given B
- Test Sensitivity (Test Recall)
- Assume you know the test result
- Positive Predictive Value (PPV, Test Precision)
- Given a positive test,
- What is the probability that the patient has the disease
- Negative Predictive Value (NPV)
- Given a negative test
- What is the probability that the patient does not have the disease
- Probability of disease given test
- P(disease | positive test) = PPV / NPV
- P(disease | positive test) via Bayes Theorem = P(positive test | disease) * P(disease) / P(positive test)
- where P(positive test | disease) = Test Sensitivity
- where P(disease) = Prevalence of disease in the tested cohort
- where P(positive test) = Probability of positive test in the tested cohort
- Positive Predictive Value (PPV, Test Precision)
VI. References
- Desai (2014) AMIA Board Review, Clinical Decision Making
- Gates (2001) Am Fam Physician 63(3):513-22 [PubMed]
- MacLean (1996) Med Clin North Am 80(1):1-14 [PubMed]
- Nielsen (1999) Med Clin North Am 83(6):1323-37 [PubMed]