Presentation (CTI)

OBJECTIVES: Heterogeneity in patient characteristics is frequently cited as justification for individual patient simulation (IPS) models. This conflates heterogeneity with stochastic simulation and obscures situations in which cohort-based models, applied to individuals or subgroups, are preferable. We examine when heterogeneity necessitates IPS, when cohort models remain sufficient, and how heterogeneous cohorts should be sampled and results presented to support transparent decision-making.
METHODS: We distinguish three modelling dimensions: (i) representation of heterogeneity (observed covariates and their joint distribution), (ii) simulation architecture (cohort vs individual), and (iii) stochasticity. We compare cohort models applied at the individual level, stratified cohort models, and IPS models against policy-relevant criteria: preservation of correlated covariates; computational burden; and interpretability of results. We examine approaches to sampling heterogeneous cohorts to maintain correlation between risk factors. Findings are illustrated using a recently presented Scottish Cardiovascular Policy Model.
RESULTS: Heterogeneity alone does not require IPS. When outcomes are deterministic conditional on baseline covariates and transitions depend only on current state and time, cohort models evaluated across individuals or finely defined strata recover identical expected values with greater transparency and lower computational cost. IPS becomes necessary when path dependence, risk factor-based treatment choice, or endogenous events introduce within-individual stochasticity that cannot be represented analytically. Naïve independent sampling of covariates materially distorts risk distributions and outcomes; preserving correlation is essential regardless of simulation type. Presenting results solely as population means conceals distributional consequences of heterogeneity; reporting of outcome distributions and subgroup effects materially improves interpretability and policy relevance.
CONCLUSIONS: The choice between cohort and individual simulation models should be driven by structural requirements, not the mere presence of heterogeneity. Clear separation of heterogeneity, correlation, and stochastic simulation can prevent unnecessary model complexity while improving transparency. Policy models should justify simulation architecture explicitly, preserve correlated risk structures, and report distributional outcomes alongside means.