Clinical epidemiology and measures of effect — branching viva

Opening (candidate)

I would approach these two papers by asking which measure of effect each design allows, and then checking whether the headline figure is the right one for that design. For the case-control study I would build the 2x2 table and compute the odds ratio myself, then ask whether the outcome is common enough to make reading it as a relative risk misleading. For the trial I would compute the absolute risk reduction and the number needed to treat from the event rates, read the hazard ratio alongside the Kaplan-Meier curve, and check whether the confidence interval crosses the null before trusting any of the figures. [7] [9]

Branch A — The odds ratio and the rare-outcome assumption

Examiner: The case-control study reports an odds ratio of 4 for an outcome that affects 30 percent of the population. Can you read that as a four-fold increase in risk? [4]

Candidate: No. The odds ratio approximates the relative risk only when the outcome is rare, under about 10 percent. At 30 percent prevalence the outcome is common, so the odds ratio is always further from 1 than the relative risk, and reading the odds ratio of 4 as a relative risk of 4 overstates the association. The true relative risk is closer to 1, and I would report it that way, or convert the odds ratio using the baseline risk where the method allows. [4] [5]

Branch B — Absolute benefit and the number needed to treat

Examiner: The trial reports a 50 percent relative risk reduction, with the outcome in 200 of 1,000 controls and 100 of 1,000 treated. Walk me through the absolute benefit. [2]

Candidate: The risk in controls is 20 percent and the risk in the treated group is 10 percent, so the absolute risk reduction is 10 percent, or 0.10. The number needed to treat is the reciprocal of the absolute risk reduction, 1 divided by 0.10, which is 10. So for every ten children treated, one event is prevented. The 50 percent relative reduction sounds dramatic, but the honest number for a family is ten. [2] [3]

Branch C — Baseline risk and tailoring the number needed to treat

Examiner: Your patient's baseline risk is 2 percent, not 20 percent. Does the number needed to treat change? [2]

Candidate: Yes, and substantially. The same 50 percent relative reduction applied to a 2 percent baseline gives an absolute risk reduction of 1 percent and a number needed to treat of 100, ten times larger than in the trial. The number needed to treat is the reciprocal of the absolute risk reduction, so it inherits the baseline risk, and I would recompute it from the child's own baseline risk rather than borrow the trial average. [2] [3]

Branch D — The hazard ratio and the survival curve

Examiner: The survival analysis reports a hazard ratio of 0.7. What does that mean, and what would you check? [6]

Candidate: A hazard ratio of 0.7 means the treated group experienced the outcome at 70 percent of the rate of the control group across the follow-up period. The hazard ratio compares the instantaneous event rates of the two survival curves and respects censoring, but it is an average over time, so it can hide curves that separate late or cross. Before acting on it I would inspect the Kaplan-Meier plot to confirm the benefit is spread evenly and that the proportional-hazards assumption holds, and I would read the confidence interval against the null value of 1. [6] [7]

Branch E — The confidence interval crossing the null

Examiner: In one subgroup the confidence interval's lower bound crosses unity. A colleague calls it a positive result. Your response? [8]

Candidate: I would correct the claim. For ratio measures the null value is 1, so a confidence interval that crosses unity is compatible with values both above and below the null, meaning the result is indeterminate rather than positive, however large the point estimate. I would state plainly that the data do not exclude harm or no effect in that subgroup, and I would not let momentum carry a decision the data do not support. [8] [7]

Close

Confirm understanding with teach-back, leave a written summary of the computed measures with their confidence intervals, name the next contact, and document the design, the matched measure, the point estimate, the interval, the absolute effect and number needed to treat, the certainty rating, and the shared decision. [9] [2]