Scientific Reasoning and Data Interpretation
This section tests whether you can read a study the way a clinician should: identify the design, spot bias, and convert a 2 × 2 table into sensitivity, specificity and predictive values. Numbers are supplied in the stem, so the skill is choosing the right denominator.
Core rules
- Variables: independent (manipulated), dependent (measured), controlled (held constant). Randomisation and blinding reduce selection and observer bias.
- Study designs: RCT for causation, cohort for incidence and relative risk, case-control for rare outcomes (odds ratio), cross-sectional for prevalence.
- Sensitivity = TP/(TP + FN); specificity = TN/(TN + FP).
- PPV = TP/(TP + FP); NPV = TN/(TN + FN). Both depend on prevalence, unlike sensitivity and specificity.
- A p-value is the probability of data this extreme if the null hypothesis is true - not the probability that the null is true.
- Confidence intervals convey precision; a wide interval means a small sample.
Worked example
A test is applied to 1,000 people; 100 have the disease. Sensitivity is 90% and specificity 90%. What is the PPV? TP = 0.90 × 100 = 90; FN = 10. FP = 0.10 × 900 = 90; TN = 810. PPV = TP/(TP + FP) = 90/180 = 50%. Half of the positive results are false positives because the disease is uncommon - the same test in a high-prevalence clinic gives a much higher PPV.
Traps
- Reporting sensitivity as PPV; the denominators differ.
- Assuming a highly sensitive test rules out disease regardless of prevalence (it helps, but the post-test probability still depends on it).
- Reading “no significant difference” as proof of no effect.
- Quoting relative risk reduction without the absolute figure, which is far smaller.