Biostatistics and Epidemiology: Diagnostic Test Performance and Study Design
1. Sensitivity and Specificity
Sensitivity measures the proportion of truly diseased individuals correctly identified by the test, while specificity measures the proportion of truly non-diseased individuals correctly identified as negative 1, 2.
From a 2×2 table:
- Sensitivity = TP/(TP + FN) - the ability to detect disease when present
- Specificity = TN/(TN + FP) - the ability to rule out disease when absent
These are intrinsic test characteristics that remain stable regardless of disease prevalence in the population being tested 1.
2. Positive Predictive Value (PPV)
PPV represents the proportion of individuals with a positive test who actually have the disease 1, 2.
PPV = TP/(TP + FP)
Critical caveat: PPV is heavily dependent on disease prevalence. As demonstrated in 1, with 80% sensitivity and 99% specificity:
- At 1% prevalence: PPV = 45% (55% false positives)
- At 10% prevalence: PPV = 90% (10% false positives)
This means the same test performs dramatically differently based on the population tested.
3. Negative Predictive Value (NPV)
NPV represents the proportion of individuals with a negative test who truly do not have the disease 1, 2.
NPV = TN/(TN + FN)
Like PPV, NPV varies with prevalence 1:
- At 1% prevalence: NPV >99% (0.2% false negatives)
- At 10% prevalence: NPV = 98% (2% false negatives)
Important distinction: While sensitivity and specificity are classification values that remain constant, predictive values change with disease prevalence and must be stratified by exposure groups even when misclassification is non-differential 3.
4. The 2×2 Diagram
Disease Present Disease Absent
Test Positive TP (a) FP (b) | Total Positive = a+b
Test Negative FN (c) TN (d) | Total Negative = c+d
___________________________________________
Total a+c b+d | Total = a+b+c+dMathematical formulas:
Sensitivity = a/(a+c)
Specificity = d/(b+d)
PPV = a/(a+b)
NPV = d/(c+d)
Where:
- a = True Positives (TP)
- b = False Positives (FP)
- c = False Negatives (FN)
- d = True Negatives (TN)
The denominators differ critically: sensitivity/specificity use disease status (columns), while PPV/NPV use test results (rows) 2.
5. Sampling Technique and Disadvantages
For prevalence calculations and diagnostic test validation, stratified sampling with oversampling of low-prevalence conditions is typically used 4.
Key disadvantages:
Bias in PPV estimation: Failure to account for varying prevalence of diagnostic codes within groups leads to underestimation of PPV, as low-prevalence codes are more likely false positives 4
Overestimation of sensitivity: Oversampling claims within low-prevalence diagnostic groups relative to those outside leads to inflated sensitivity and underestimated specificity 4
Non-representative samples: If the validation sample is not representative of the target population in terms of disease prevalence, absolute test values, or demographic characteristics, predictive values cannot be accurately transported to clinical practice 5
Requires weighted adjustment: PPV must be calculated as a weighted average across diagnostic codes when stratified sampling is used 4
Critical pitfall: Applying a single, unstratified predictive value across exposure strata generates biased estimates even when misclassification is non-differential 3.
6. Type of Data: Age
Age is continuous quantitative data (ratio scale).
Age can be analyzed as:
- Continuous variable: Maintains full information, allows assessment of non-linear associations 5
- Categorical variable: When grouped (e.g., age ranges), loses precision but may be clinically meaningful
The continuous nature should be preserved when possible, with allowance for non-linear effects in predictive models 5.