Does adjusting the Carr-Hill formula, or total GP funding by deprivation data improve accuracy of predicting clinical need?

Abstract

Introduction: Resourcing General Practice (GP) proportionate to need is paramount to delivering equitable, cost-effective care. Approximately half of GP funding follows a capitation model (the Carr-Hill formula), where patient demographics drive calculations of clinical need. Other funding is obtained through pay-for-service, and pay-for-performance mechanisms. Previous research has called for the addition of deprivation data to the Carr Hill formula. However, there is a lack of evidence that this would better align with clinical need. This study aims to assess whether adjusting either the Carr-Hill formula, or total GP funding by an area's deprivation score using the Index of Multiple Deprivation (IMD), leads to a better prediction of clinical need. Methods: This cross-sectional study utilised data from the 32,844 Lower-Super-Output-Area (LSOA) in England in 2021–2022. Weighted average Carr-Hill Index (CHI) and total GP funding were calculated for each LSOA using the number of patients registered at each GP surgery per LSOA. Five measures of clinical need were calculated for each LSOA: Combined Morbidity Index (CMI), predicted total morbidity (including undiagnosed morbidity), emergency department (ED) presentations, total GP appointments and age and sex standardised mortality rates (SMR). For both CHI and GP funding, three sets of generalised linear models were calculated for each outcome variable: 1. unadjusted; 2. adjusted for age; and 3. adjusted for age and IMD. Adjusted R2 value assessed how accurately each model predicted variations in outcome variables. If R2 values increased after adjusting for IMD and age (model 3) compared to age alone (model 2), this would indicate that adjusting funding for IMD is a more accurate measure of clinical need. Results: In age-adjusted models, CHI most accurately predicted CMI (R2=62%), followed by total morbidity (R2=47%). CHI was a moderate predictor for ED admissions and GP appointments (R2=40%, 29% respectively) and a poor predictor for SMR (R2=9%). The introduction of IMD enhanced accuracy of all models, most notably for mortality (R2=63% CMI, 49% total morbidity, 41% ED admissions, 29% GP appointments, and 23% mortality respectively). Total funding showed less robust predictive power for clinical need measured by CMI (R2=47%), total predicted morbidity (R2=30%), ED admissions (R2=30%), and SMR (R2=9.8%), but stronger predictability for GP appointments (R2=33%). Adjusting for IMD resulted in larger improvements in accuracy of all models (R2=53%, 39%, 32%, 23%, and 35% respectively). Sensitivity analysis using pre-pandemic 2019 data confirmed findings. Discussion: These findings offer the most compelling evidence to date that incorporating IMD within either the Carr Hill formula, or total-funding, would result in more accurate distribution of funds in relation to clinical need. Total funding's lower accuracy to predict clinical need compared to CHI indicates inefficiencies introduced by pay-for-service and pay-for-performance mechanisms, and supports broader use of capitation models. Capitation formulas should better account for variations in mortality.