Design matters in patient-level prediction: evaluation of a cohort vs. case-control design when developing predictive models in observational healthcare datasets

Journal of Big Data

Table 2 Different types of prediction problems and examples of how they fit the cohort design

Prediction type	Target population	Outcome	Example target cohort inclusion criteria and index
Prediction type	Target population	Outcome	Target cohort criteria	Index
Disease onset	General population	Disease (e.g., depression)	A visit (outpatient or inpatient) during 2010, > 365 days observation in database, age ≥ 8, no prior illness	First valid visit in 2010
Disease progression	Early-stage disease patients	Advanced stage disease	Diagnosed with disease, > 365 days observation in database	Initial disease record date
Treatment choice	Patients dispensed treatment 1 or 2	Treatment 1	Dispensed treatment 1 or 2, > 365 days observation in database	First recorded date of treatment 1 or 2
Treatment response	Patients dispended a treatment	Desired effect (e.g., disease cured)	Dispensed treatment at adequate therapeutic level, > 365 days observation in database	First recorded date of treatment
Treatment safety	Patients dispended a treatment	An adverse event	Dispensed treatment, > 365 days observation in database	First recorded date of treatment
Treatment adherence	Patients dispended a treatment	> X% days covered during some follow-up	Dispensed treatment, > 365 days observation in database	First recorded date of treatment