Abstract
Background Funding shortfalls persist for practices in the most deprived areas, despite capitation formula adjustments.
Aim To evaluate whether deprivation scores predicted practice payment trends between 2019 and 2024.
Design and setting Multivariable analysis was undertaken of English general practices (2018–2019 to 2023–2024), excluding practices with <750 patients or average payments >£500 per patient per year, using published data.
Method A quadratic mixed-effects model was fitted, using cluster-robust standard errors. The outcome was log-transformed average NHS practice payments per patient (net of deductions/reimbursements). The fixed effects were time (categorical), the Index of Multiple Deprivation (IMD) score (higher score indicates greater deprivation), and seven covariates (geographical, population, or organisational). The random effect was practices’ random intercepts.
Results Among 5726 included practices, median payments increased in nominal terms (8.6%) but decreased in real terms (–12.6% consumer price index [CPI] and –9.0% CPI for health). The IMD–payment trend relationship was curvilinear, peaking at IMD 49.8 (1.4% above mean deprivation, IMD 23.2), declining to 0.6% higher at IMD 70.0. More positive payment trends were associated with non-London regions, rurality, greater long-term conditions (LTCs) prevalence, and higher baseline payments; less positive trends were associated with more patients aged <16 years, larger lists, and personal medical services contracts. In interaction models, rurality increased whereas higher LTCs decreased IMD’s impact.
Conclusion Deprivation had a positive but diminishing association with payment trends as deprivation increased, moderated by geography and morbidity. Payment uplifts must match inflation. Funding formulas must better compensate for deprivation and morbidity, address the attenuated positive effect of deprivation in practices with more patients with LTCs, and minimise geographical inequalities.
How this fits in
Because NHS practice payments have been criticised for ‘pro-rich inequity’, this study evaluated the association between deprivation scores and total payment trends in 2018– 2019 to 2023–2024. Inflation-adjusted total payments decreased while the deprivation–payment trend relationship was curvilinear, flattening as scores increased. These findings suggest persistent relative underfunding in the most deprived areas and geographical inequalities, particularly in urban areas and the London region. The mechanisms of these inequalities need further investigation, but remedies should include payment uplifts matching inflation, better alignment with direct deprivation measures, and weighting extended to other payment streams where appropriate.
Introduction
General practices serving socioeconomically disadvantaged populations need additional resources, but have not always received them when compared with those serving more affluent populations.1–9 Since 2004, capitation payments directed to English general practices, accounting for about half of their NHS income,10 have been adjusted. As part of the NHS GP contract, the global sum allocation formula, also known as the Carr-Hill formula, weights capitation funding based on modelled estimates of practices’ patient workloads. These take into account two groups of factors, ‘drivers of workload’ and ‘unavoidable costs’11,12 (Supplementary Table S1). Carr-Hill originally used consultation length as a proxy for workload, which may be flawed as consultations are not necessarily longer in deprived areas.13 However, these areas are likely to experience higher morbidity that drives up healthcare ‘consumption’,14 and often suffer from uneven workforce distribution.15 Carr-Hill has only included proxy, never direct, measures of deprivation and morbidity,16 two important factors associated with health need. Even after age correction, the validity of Carr-Hill’s modelling for additional needs, using the measure’s standardised limited longstanding illness and the standardised mortality ratio for those aged <65 years, has been questioned.11
Associations have been found between higher total general practice payments per patient and greater life expectancy in England,17 and between higher capitation payments and better quality of primary care, as measured by Care Quality Commission ratings.18 In 2016, NHS England recognised that Carr-Hill had limitations and needed to be revised.19 A 2024 government review of the NHS acknowledged a ‘shortfall in funding for practices serving more deprived populations’.20 At least one integrated care board has developed local general practice funding models to address inequities in Carr-Hill.21 In June 2025, the government pledged in its Fit for the Future: 10 Year Health Plan for England to strengthen care in the community and to ‘shift the pattern of health spending.’
22 Achieving fairer funding will include a review of Carr-Hill; however, no details are currently available.
The context has evolved during and following the COVID-19 pandemic: total healthcare expenditures increased, although their ad hoc nature likely contributed to growing inefficiency,23 and there was a potential adverse impact on prevention and the management of long-term conditions (LTCs).24 Workforce reconfigurations have included substantial increases in non-doctor roles but not GPs,25 alongside increased triage, remote consulting,26 and growth in reported morbidity — the annual GP Patient Survey (GPPS) reported a post-pandemic increase in LTC prevalence of 19.6% (from 52.5% in 2021 to 62.8% in 2025).27 An updated investigation is needed into whether the previous suboptimal alignment of total NHS practice payments to deprivation in practice populations4,5 has improved or worsened since the pandemic. The study’s hypothesis was that practice payment trends continued to show little or no relative improvement in practices serving populations with the greatest deprivation; thus, the study’s research question was: how much did deprivation scores predict variations in the longitudinal trends of total NHS practice payments between 2018–2019 and 2023–2024, after adjusting for geographical, population, and organisational factors?
Method
Overview
This longitudinal, ecological multivariable analysis included practice-level data from published summary statistics (Supplementary Table S2). Organisation data service codes were used to define practices within each dataset before merging these.28 All included variables required consistently defined available data across the study period.
Study population
The study population was English general practices from 2018–2019 to 2023–2024 fulfilling the following criteria: receiving NHS payments, classification as GP practices, and open for ≥1 full year. Practices with <750 patients or receiving very high adjusted average NHS payments (>£500 per registered patient per year) were excluded.
Dependent variable
The outcome in each study year was average annual NHS payments per registered patient, from which deductions for pensions, levies, and prescription charge income, and reimbursements for premises payments and drugs were subtracted.10
Independent variables
Where possible, the independent variables’ data were matched to each financial year. Unless otherwise stated, values were included for all 6 years from the continuous variables.
The central independent variable was the 2019 Index of Multiple Deprivation (IMD) score, an overall relative area-based measure of socioeconomic status that combines indicators from seven domains (income, employment, education, health, crime, barriers to housing and services, and living environment).29 Higher IMD scores indicate greater deprivation. The Office for Heath Improvement and Disparities publishes practices’ IMD scores:30 these were patient-weighted, calculated by NHS Digital by averaging the IMD of lower-layer super output areas according to the proportion of a practice’s patients resident in each. The current study used these scores, which reflect the deprivation profile of the practice’s catchment population, to assess their gradient relationship with payment trends.
Geographical, population, and organisational factors were adjusted for. For inclusion, a covariate needed to have a conceptually plausible relationship to either payments or deprivation, using a published population health research framework (Supplementary Figure S1),31 and to not be strongly correlated with others (coefficient <0.4) (Supplementary Figure S2). The latter was to reduce the risk of multicollinearity in the models. Some potential variables with strong correlations found in previous work using these datasets were eliminated.32 Seven independent variables fulfilled the criteria.
Two geographical categorical variables were included: the NHS commissioning region and rurality. England has seven NHS commissioning regions, responsible for the performance of all NHS organisations within their region (East of England, London, Midlands, North East and Yorkshire, North West, South East, and South West).33 London was the reference. The rural urban classification is an official statistical classification, used to distinguish urban (defined as settlements of ≥10 000 population) and rural (everything else) areas.34 Practices were classified as either rural or urban based on their postcode.10 Urban location was the reference.
Two continuous population variables were included: the percentages of practice patients aged <16 years35 and with LTCs from the GPPS27 to represent morbidity. The percentage aged <16 years35 was the only age band that did not correlate strongly with other variables. For morbidity, the authors of the current study considered using one of five potential Quality and Outcomes Framework disease registers,36 but all these correlated with at least one of the other independent variables, most commonly IMD. Also considered was using a self-reported broad ethnicity band from the GPPS,27 as defined in the England and Wales 2021 census,37 but this was moderately strongly correlated with the LTC variable (coefficient = 0.48). The GPPS data are weighted to adjust for response rates and practice population characteristics.
Three organisational variables were included: baseline payments (2018–2019), the practice list size (in 100s), and the type of NHS primary care contract at the start of the study period.10 Practices can have different contracts with the NHS, either general medical services (GMS), personal medical services (PMS), or alternative provider medical services (APMS). GMS was used as the reference. The first two variables were continuous and the latter categorical.
Analyses
In data cleaning, implausible percentage values, >100% or negative, were replaced by missing values.
Descriptive statistics were used to assess continuous variables’ distributions. Very skewed distributions were log-transformed.
In the multivariable analyses, a linear mixed-effects regression model with practice random intercepts was fitted, including only practices with data in all 6 years. The initial model included time, IMD (linear term), and the seven covariates as fixed effects. Following diagnostic checks that suggested potential non-linearity in the deprivation–payment relationship, a quadratic IMD term was added, after mean-centring IMD, to address multicollinearity. Time was treated as categorical (factor) with six levels corresponding to each year (reference 2019), as model comparisons showed this specification provided a superior fit to continuous time (ΔAkaike information criterion [AIC] = –4961; likelihood-ratio test [LRT] χ²[4] = 234.5, P<0.001; Supplementary Table S3). The random effect was the practices' random intercepts, to account for unobserved baseline differences between them. To account for expected heteroscedasticity and clustering, standard errors clustered by practice with small-sample correction (CR2 type) were used to ensure valid inference. Additionally, whether the quadratic relationship differed across the payment distribution was examined by comparing practices in different payment quantiles.
Model diagnostics (residual plots, Cook’s distance) were assessed and robustness confirmed via sensitivity analyses that excluded influential observations (Cook’s distance ≥4/n).
Four supplementary models were fitted to check robustness and explore effect modification: a) testing interaction between IMD and rurality; b) testing interaction between IMD and LTC percentage; c) added percentage of patients of White ethnicity (with unreported values coded as 0%);27 and d) unadjusted (no subtractions) replaced adjusted payments. All these models treated time as categorical and used cluster-robust standard errors.
All models’ performances were assessed for multicollinearity, homogeneity of variance, and normality of residuals.
Statistical analyses were performed using R (version 4.5.1).
Results
Study population
There were 7279 practices registered for NHS payments in 2018–2019. Of these, 479 were transient, inactive, or providing a restricted range of services to very specific populations (such as walk-in clinics), leaving a potential study population of 6800 practices. A further 27 (0.4%) were excluded with <750 patients or with average payments >£500 per patient per year. Thus, 6773 practices were eligible. The models included 5726 practices (84.2% of those eligible) with data in all years (Supplementary Table S4).
Descriptive statistics
Tables 1 and 2 show the descriptive statistics. The distributions of the payments and list size variables were highly skewed, requiring log-transformation for the model.
Table 1. Descriptive statistics of continuous variables (N = 5726 practices with data) Table 2. Descriptive statistics of documented categorical variables (N = 5726) Between 2018–2019 and 2023–2024 the median of average unadjusted payments per patient increased by 8.6% (£117.00 to £127.00) (Table 3). When practices were ranked by IMD score in deciles, the nominal increases in the medians of total payments were larger in the three least deprived deciles than in the three most deprived deciles, and, between 2021–2022 and 2022–2023, the medians decreased nominally in six of the seven most deprived deciles (Supplementary Figure S3). However, when adjusted for inflation between 2019 and 2024, the median decreased in real terms by –12.6% to £102.30 (using the consumer price index [CPI])38 and by –9.0% to £106.47 (using the CPI for health) (Table 3).39
Table 3. Median total payments per patient 2018–2019 to 2023–2024, nominal and inflation-adjusted (both general and health related) Between 2021 and 2024 the mean LTC prevalence increased by 16.00% (52.14% to 60.48%), having been previously stable (Table 1). The percentage of patients aged <16 years declined slightly from 2019–2024 (mean 17.44%–16.44%, standard deviation 3.76%–3.43%). The distribution was clustered, with very few adult-only (0.14%–0.40%) or paediatric-only (0%–0.01%) practices (Supplementary Table S5).
Multivariable analyses
Table 4 presents the final mixed-effects model of practice payment trends (2018–2019 to 2023–2024). Initial model diagnostics suggested non-linearity; adding a quadratic term for IMD improved model fit (χ²[1] = 8.45, P = 0.004) (Figure 1). Both linear (β = 0.0011, 95% confidence interval [CI] = 0.0007 to 0.0014, P<0.001) and quadratic (β = −0.0000 × 10⁻⁵, 95% CI = −0.0000 × 10⁻⁵ to −0.0000 × 10⁻⁶, P = 0.003) IMD terms were significant, indicating diminishing positive payment trends with higher IMD scores. The relationship peaked at IMD 49.8 (1.4% above mean deprivation, mean IMD 23.2), declining to 0.6% higher at IMD 70.0.
Table 4. Linear mixed-model fit by restricted maximum likelihood of NHS practice payments 2018–2019 to 2023–2024 with IMD centred and quadratic IMD terma
Model-adjusted payment trends were non-linear, decreasing by 0.05% between 2019 and 2020 (β = –0.0005, P = 0.71) but increasing by 7.3% in between 2019 and 2024 (β = 0.0734, P<0.001). All regions had more positive trends than London (β = 0.011–0.036, all P<0.05), as did rural versus urban practices (β = 0.0519, P<0.001). Higher percentage of patients aged <16 years was associated with less positive trends (β = −0.0028, P<0.001); higher percentage LTC was associated with slightly more positive trends (β = 0.0008, P<0.001). Higher baseline payments were associated with more positive trends (β = 0.731, P<0.001). Larger list size was associated with slightly less positive trends (β = −0.0001, P<0.001). Compared with GMS contracts, PMS contracts were associated with a marginally significant less positive trend (β = −0.0061, P = 0.05), whereas the association with APMS did not reach significance (β = −0.0394, P = 0.053) (Table 4).
The model explained 73.0% of variance (conditional R²; intraclass correlation 0.33). Diagnostic checks indicated no concerning multicollinearity (all generalised variance inflation factors <2.1; Supplementary Table S6). Residual plots showed acceptable, although not perfect, homogeneity of variance and normality (Supplementary Figure S4).
Sensitivity analyses excluding influential observations (Cook’s distance ≥4/n and ≥5/n thresholds), as well as excluding one particularly influential practice, produced nearly identical coefficient estimates for the key IMD terms (linear: 0.00098–0.00108; quadratic: consistently −0.000021; Supplementary Table S7). To verify that a minor wording change in the 2024 GPPS LTC question did not affect the findings, a sensitivity analysis was conducted treating the 2024 LTC data as missing and refitted the main model. The coefficients for the deprivation terms (IMD linear and quadratic) were identical to six decimal places, and the model fit (AIC) was unchanged (Supplementary Table S8), confirming the robustness of the primary results.
Supplementary analyses
High-payment practices were disproportionately rural (54% versus 14%) with smaller lists (n = 7540 versus n = 9550) (Supplementary Table S9). Stratified analyses showed different IMD–payment relationships as payment levels changed: slightly more positive trends for low-payment practices (β = 0.00035) but slightly less positive trends for high-payment practices (β = –0.00029), with no significant curvature within subgroups (both quadratic P>0.13). The full sample’s curvilinear pattern therefore appears to reflect compositional differences across the payment distribution (Supplementary Figure S5). Thus, the overall quadratic pattern may have resulted from combining groups with different baseline characteristics, consistent with the rurality–IMD interactions.
Four supplementary models tested robustness and identified modifiers of the deprivation–payment relationship (Supplementary Table S10). The positive IMD effect was stronger in rural than urban practices (quadratic interaction P<0.001) (Supplementary Figure S6). It was weaker in practices with higher LTC prevalence (linear interaction P = 0.004) (Supplementary Figure S7). Adding percentage White ethnicity revealed that practices with higher percentage White ethnicity had slightly more positive payment trends (β = 0.0003, P<0.001), although this did not substantially alter the IMD coefficients (Supplementary Table S10). Using unadjusted (no subtractions) payments per patient instead of adjusted payments per patient substantially weakened the association between IMD and payment trends, with the linear effect becoming non-significant (P = 0.101) and the quadratic effect attenuated (P = 0.024).
Discussion
Summary
In answer to the primary research question, statistically, higher deprivation (IMD) scores had a curvilinear association with payment trends between 2018–2019 and 2023–2024; the positive association lessened as deprivation increased. This finding was robust across sensitivity checks; however, during the study period real-terms payments decreased substantially after inflation adjustment (–12.6% retail prices and –9.0% health inflation). Additionally, the model found:
statistically significant geographical disparities independent of deprivation scores, with less positive trends in urban practices and in the London region;
more positive payment trends in practices with higher baseline payments; and
less positive payment trends in practices with a younger age structure.
Supplementary analyses show that the IMD’s effect on payment trends was stronger in rural practices but weaker as the prevalence of LTCs increased, and less positive payment trends in practices with a greater ethnic minority population.
Strengths and limitations
This study has a number of strengths. The model included 84.2% of English practices. Most characteristics showed small standardised mean differences (SMD <0.2) between included and excluded practices (Supplementary Table S11). Regional distribution differed most (SMD 0.40), with higher South West exclusion. However, similar IMD scores and payment levels suggest minimal bias for the curvilinear IMD–payment trend relationship. A robust process was used for selecting independent variables. Total payments was a broader outcome than specific income streams. Multivariable analysis provided a realistic approach for examining payment complexities. The model accounted for practice-level clustering and explained a substantial proportion of the outcome variance (high conditional R²), with diagnostic checks confirming appropriate fit (normally distributed residuals). The supplementary models helped to better understand the main model. The current study updated the evidence, examining payment trends across the pandemic.
Potential limitations included missing data for some variables (beyond the authors’ control). The direct NHS practice payments data do not include other indirect streams or private income, nor do they take account of community and outreach services. Although IMD is a standard research tool and has been used in funding analysis,40 it is a composite variable that could mask some heterogeneity even in very small areas.41 Its inclusion of a health domain (13.5%) could introduce potential confounding, and it does not capture short-term temporal changes. The GPPS addresses low-response rates (usually <30%) by weighting sampling and results to align with practices’ population characteristics, such as age and ethnicity.27 Cluster-robust methods provided reliable P-values and confidence intervals despite heteroscedasticity (expected in large observational data) and clustering, although they may reduce power. Sensitivity analyses excluding influential observations did not alter coefficient estimates or model fit statistics (Supplementary Table S7). Ecological studies find associations but cannot prove causality, including exclusion of potential reverse causality between deprivation and payments.
Comparison with existing literature
Although providing updated evidence, the current findings aligned with the authors’ previous pre-pandemic analyses, showing that increased deprivation was weakly associated with higher practice payments, both cross-sectionally4 and longitudinally,5 and with other studies showing that, overall, practices serving more deprived populations had received little or insubstantial additional payments.6–8 Another study assessed ‘inequity in payments’ to English primary care providers from 2014–2022, comparing multiple definitions of need, and found ‘pro-rich inequity in total payments for most definitions of need’.9 More deprived primary care networks attained fewer performance-related investment and impact fund points in 2022–2023, hence less additional funding.42 The above evidence suggests current payment systems remain insufficiently responsive to increases in diverse measures of either deprivation or need, a situation unchanged post-pandemic.
Implications for research and practice
The study findings have clear practical significance:
the inflation-adjusted decrease in NHS general practice payments, since before the pandemic, will not help practices trying to manage rising workloads from increased LTC prevalence and from the 10 Year Health Plan for England’s proposal to shift care from hospitals into the community;22
the funding formula seems to inadequately address relative underfunding, particularly for practices serving the most deprived areas, although this study’s observational design precludes any causal inferences;
independent of deprivation, the current analysis indicates possible geographical inequalities, particularly in urban areas and the London region, with further investigation needed to better understand the mechanisms; and
the attenuated positive association between deprivation and payment trends in practices with high LTC prevalence could, if unaddressed, risk worsening inequities and the persistence of the inverse care law.2
Potential remedies could include:
ensuring payment uplifts match inflation;
thoroughly reviewing the Carr-Hill formula to better align with direct deprivation measures and reduce potential overcompensation; and
examining other payment streams for extending weighting and for switching performance targets to more work-sensitive metrics, where appropriate.
These actions may help the 10 Year Health Plan for England to achieve its pledge to close health inequalities.
Notes
Funding
No funding was received for this study.
Ethical approval
This research is a retrospective analysis of NHS payments to English general practices using data published in the public domain by UK Government agencies and the NHS. The data are highly aggregated and present no risk to patient confidentiality. This research does not report the results of experiments on humans and/or the use of human tissues. No ethical approval was considered necessary for this research.
Provenance
Freely submitted; externally peer reviewed.
Data
All data and analytical code for this study are publicly available on Zenodo at: 10.5281/zenodo.19636272. The repository includes the raw dataset and the complete R script used for analysis, preserving all iterative steps and exploratory work.
Acknowledgements
The corresponding author used DeepSeek (version V3) as a tool in the preparation of this article. Specifically, DeepSeek was used to: 1) assist with writing and debugging R code for statistical analyses and generating graphical elements (figures) from analysed data; 2) proofread the manuscript for clarity and errors; and 3) assist with responses to technical queries during peer review. The corresponding author retained full control over all analytical decisions, interpretation of results, and final manuscript content. All co-authors have reviewed and approved the final manuscript and are aware of this AI use disclosure.
Competing interests
The authors have declared no competing interests.
