Population-representative inference for primary and secondary outcomes in extended case-cohort designs

Abstract Background The case-cohort design is useful for obtaining population-representative inference with smaller samples, but the performance when dealing with secondary outcomes and extended follow-up remains unclear. We compare the case-cohort design with the full-cohort design across various outcomes, including additional case-groups and extended follow-up, illustrated using the iPSYCH study. Methods From Danish nationwide registers, we identified the full-population cohort of all individuals born 1981–2008 (n = 1,657,449) and their clinical diagnoses until 2021. The iPSYCH case-cohort sample (n = 141,265) includes specific case-groups with specific psychiatric disorders diagnosed 1994–2015 (n = 93,608) and a random-population subcohort (n = 50,615). We applied inverse probability weights to estimate person-years at risk, age-specific incidence rates, absolute risks, and incidence rate ratios in various scenarios for primary outcomes (affective disorder, bipolar disorder, schizophrenia, autism, and attention-deficit hyperactivity disorder) and secondary outcomes (epilepsy, anxiety, migraine, asthma, diabetes, injury, traumatic brain injury, substance use disorder, and death). Results Weighted estimates based on the iPSYCH sample aligned with those in the full cohort for both primary and secondary outcomes. For example, weighted absolute risks by age 40 were analogous to full-cohort estimates for both affective disorder (7.9% [7.5–8.3] versus 8.0% [7.9–8.1]) and epilepsy (2.3% [2.1–2.5] versus 2.3% [2.2–2.3]). Similarly, weighted incidence rates and incidence rate ratios mimicked full-cohort estimates. Conclusion The extended case-cohort design yields valid estimates of age-specific incidence rates, absolute risks, person-years at risk, and incidence rate ratios for primary and secondary outcomes, even with multiple case-groups and extended follow-up.