Risk-Adjusted Cohort Study — Observational Epidemiology with Confounding Control

A standard cohort study compares crude outcome rates between exposed and unexposed groups without formally accounting for confounders. A risk-adjusted cohort study adds a statistical adjustment step — regression modelling, propensity scoring, or weighting — to control for measured differences between groups at baseline. Risk adjustment is now considered standard practice in observational epidemiology whenever groups differ on important covariates.

Neither is universally superior. Multivariable regression is efficient when the outcome is relatively common and the number of confounders is modest relative to sample size (a rough rule: at least 10 outcome events per covariate). Propensity score methods are preferred when there are many confounders relative to outcomes (common with rare events), when you want to balance covariates transparently, or when you need to estimate average treatment effects in specific target populations. Both require the same fundamental assumption: no unmeasured confounding.

No. Risk adjustment controls only for measured confounders. Randomization balances both measured and unmeasured variables. When unmeasured confounders are likely — for instance, when indication for treatment is driven by clinician judgement or patient preference — even the most careful adjustment leaves residual bias. Sensitivity analyses (e.g., E-values) can quantify how large this bias would need to be to change conclusions, but they cannot rule it out.

Sample size depends on outcome frequency, the number of covariates to be adjusted, the expected effect size, and the adjustment method. For Cox regression a commonly cited rule of thumb is at least 10–20 outcome events per covariate included in the model. Propensity score methods shift the focus to achieving balance and may tolerate more covariates, but very small exposed or unexposed groups still limit precision. A formal a priori power calculation is recommended.

The E-value, introduced by VanderWeele and Ding (2017), quantifies the minimum strength of association that an unmeasured confounder would need to have with both the exposure and the outcome — on the risk ratio scale — to fully explain away an observed association. It provides readers with a transparent benchmark for judging robustness: a large E-value means the result is harder to dismiss as residual confounding. Reporting it has become increasingly expected in high-quality observational epidemiology.