Understanding bias in EMS STEMI data: a national service-evaluation study of deprivation, behavioural pathways and inequality metrics in Wales

Background Composite deprivation indices are widely used in health services research but may introduce endogeneity when health-related indicators are included in the deprivation measure itself. This study examined whether socioeconomic gradients in emergency cardiac events reflect true disease burden or are distorted by index composition and behavioural pathway selection. Methods We conducted a national evaluation using linked electronic patient care records from the Welsh Ambulance Services NHS Trust (2021–2025). Deprivation gradients in ST-elevation myocardial infarction (STEMI) and out-of-hospital cardiac arrest (OOHCA) were analysed across 999 and 111 pathways using three WIMD variants: full, Health-domain–excluded (“Minus Health”), and Income-only. Inequality was quantified using Slope and Relative Indices of Inequality (SII, RII), with quadratic terms testing for non-linearity. Results Among 12,241 EMS-attended cases (2,515 STEMI; 9,726 OOHCA), STEMI showed an inverse-U deprivation gradient that peaked in mid-quintiles, particularly for 111 users, while OOHCA displayed a strong, linear increase with deprivation. Removing the Health domain altered STEMI gradients but had minimal impact on OOHCA, suggesting measurement bias in the former. The 999 STEMI pathway showed a more consistent, monotonic gradient aligned with biological risk, whereas the 111 pathway was less stable and more affected by behavioural factors such as help-seeking and symptom appraisal. Conclusions Inequalities in EMS STEMI data reflect both true deprivation-linked disease and behavioural capture effects shaped by index design. Behaviour-independent comparators like OOHCA and the use of Health-excluded indices offer more valid assessments of inequality. Non-linear modelling is essential, as standard linear metrics (SII, RII) may obscure complex relationships. Crucially, these findings suggest that no dataset is behaviourally neutral: even within a single EMS system, pathway-specific behaviours influence visibility and gradient shape. Accurate interpretation of service-based health data must therefore account for both measurement structure and access dynamics to avoid misrepresenting need and misallocating resources.