Relationship between general practice capitation funding and the quality of primary care in England: a cross-sectional, 3-year study

Objective To explore the relationship between general practice capitation funding and quality ratings based on general practice inspections. Design Cross-sectional study pooling 3 years of primary care administrative data. Setting UK primary care. Participants 7310 practices (95% of all practices) in England which underwent Care Quality Commission (CQC) inspections between November 2014 and December 2017. Main outcome measures CQC ratings. Ordered logistic regression methods were used to predict the relationship between practice capitation funding and CQC ratings in each of five domains of quality: caring, effective, responsive, safe and well led, together with an overall practice rating. Results Higher capitation funding per patient was significantly associated with higher CQC ratings across all five quality domains: caring (OR 1.14, 95% CI 1.04 to 1.23), effective (OR 1.08, 95% CI 1.00 to 1.16), responsive (OR 1.09, 95% CI 1.02 to 1.17), safe (OR 1.11, 95% CI 1.05 to 1.18), well led (OR 1.13, 95% CI 1.06 to 1.20) and overall rating (OR 1.13, 95% CI 1.06 to 1.19). Conclusion Higher capitation funding was consistently associated with higher ratings across all CQC domains and in the overall practice rating. This study suggests that measured dimensions of the quality of care are related to the underlying capitation funding allocated to each general practice, implying that additional capitation funding may be associated with higher levels of primary care quality.

Conclusion: Higher capitation funding per patient was consistently associated with higher ratings across all CQC domains and in the overall practice rating. This study suggests that measured dimensions of the quality of care are related to the underlying funding allocated to each general practice, implying that additional funding may be associated with higher levels of primary care quality.

Strengths and limitations
 A cross-sectional study covering three years of primary care data  The definition of primary care quality used in this study was multidimensional, based on inspection findings and covering patient safety, patient experience, clinical effectiveness.
 The association between the achievement of quality ratings and practice funding was explored, adjusted for known confounders  Although based on a near complete sample of general practices in England, bias may have been introduced by data coding and recording errors  Longer term and prospective studies are required to strengthen causal inferences Introduction Improving the quality of care is a major focus of UK government health policy (1). High quality health care has three main components: clinical achievement, patient experience and patient safety(2). There is wide variation between general practices in the achievement of clinical care quality indicators and patient reported satisfaction (3).
It is important to understand whether variations in the quality of care provided across practices are related to variations in their funding. Healthcare quality regulation of healthcare in England is currently undertaken by the Care Quality Commission, focuses on outcomes for patients and has a wide range of enforcement powers, including closure and deregistration of services, if essential standards are not met. (4) Studies of the relationship between quality and funding in English general practices have largely focussed on the Quality and Outcomes Framework (QOF) which rewards practices for higher quality care, as defined by the achievement of clinical process and outcome targets.
The QOF has had limited impact on reducing secondary care costs (5) or improving primary care performance (6,7). In terms of financial incentivisation, the QOF accounted for approximately 7.8% of funding received by general practices in England in 2016 (8). In contrast, capitation payments represent the largest proportion of funding to general practice (54% in 2016) and are related to the number of registered patients in each practice (8), adjusted for factors thought to increase the demand on primary care services (9). Other Greater capitation spending on general practices has been found to be associated with reductions in secondary care usage and costs, and increased patient satisfaction (11). Studies have also shown that leadership within the practice organisation plays a key role in the delivery of high quality care (12). Until recently, nationally derived metrics of measures of inspection-based primary care quality were unavailable. Since October 2014 all general practices have been subjected to inspections by the Care Quality Commission (CQC) (13), (4).
The CQC reports on the extent to which practices are caring, effective, responsive to the needs of patients, safe, and well-led (4,14) and also combines these five domains to produce an overall practice rating. In this study, we assess the relationship of practice capitation funding with overall CQC ratings and with the individual CQC domains.

Data sources
We linked practice-level data on NHS payments to general practice (15) to CQC inspection ratings (14), NHS administrative datasets, General and Personal Medical Services Statistics (16), and small area Census and socio-economic data from Neighbourhood Statistics(17).

Care Quality Commission Ratings
CQC ratings are based on publicly available data (such as QOF and General Practice Patient Survey(18)), practice inspections and interviews with patients and staff (14). We used CQC ratings for practices with completed CQC reports first inspected between November 2014 and  Table 1; each is rated on a 4-point scale.

Practice data
Data for all general practices in England were obtained from the General and Personal Medical Services database, for 2014/15, 2015/16 and 2016/17 financial years (16). Our use of practice based demographic data followed a previously used methodology (19).
We defined practice funding as the capitation payment per patient for each financial year.
Capitation payments are weighted to reflect factors affecting GP workload (age, gender, patients in nursing and residential homes, small area measures of morbidity), rurality and an index of local staff costs which affect the cost of providing services (9).

Sample
We linked inspected practices (n=7310) with funding data for their year of inspection. We excluded atypical practices with ≤750 registered patients (n=10) or ≤500 patients per FTE GP (n= 8) following a previously used method(22). Practices with recorded negative (n= 2) or zero funding (n = 52) were excluded. The final analysis sample consisted of 7238 practices.

Data Analysis
Analysis was at GP practice level. Since the CQC rating outcomes are ordered categories we used ordered logistic regression to model the relationship between funding and the practice CQC ratings (23). Separate models were estimated for each domain.
The key explanatory variable was funding per 100 patients. In addition to patient and practice characteristic covariates, the regression models included year effects to allow for inspection year and annual general practice funding uplifts. We accounted for local area effects by adjusting for clustering at Clinical Commissioning Group (CCG) level. Multicollinearity was tested for by calculating the Variance Inflation Factor (VIF) and variables with a value for VIF>5 were excluded. The proportional odds assumption of the ordered logit model was also tested (24). We report the odds ratio from the estimated models. We also calculated the average marginal effects of funding on the predicted probabilities of achieving overall ratings of "outstanding" and "inadequate" for all practices. We also compare the predicted probabilities of an "outstanding" overall rating for training versus non-training practices, single-handed versus multi-handed practices, and rural versus urban practices.
STATA 14 (StataCorp, College Station, TX) was used for all statistical analyses.

Patient involvement
Funding for this study included funding of a dedicated patient involvement group. Patients were involved in developing plans for the study design, approving the outcome measures and commenting on the potential impact of outcomes. A lay summary was also provided.

Results
Summary statistics for the main characteristics of the general practices are provided in Table   2. Mean funding per registered patient increased from £77.49 in 2014/5 to £83.17 in 2016/7 ( Table 3).
The distribution of practice ratings across each quality domain is shown in Figure 1. A total of 79% (n = 5774) of practices achieved an overall rating of 'Good', while only 4% (294) achieved an overall rating of 'Outstanding'. 'Inadequate' ratings varied across the domains, from 1% (caring domain) to 6% (safety domain), with an 'Inadequate' rating of 4% for overall performance.  and for the overall practice rating.  Table 5 reports odds ratios for all the explanatory variables in the overall practice quality rating model with all the additional explanatory variables. In addition to higher practice funding, rural practice and training practice status were significantly associated with higher overall practice ratings. Conversely, lower CQC overall practice rating were associated with higher levels of deprivation, the proportions of patients from black and Asian ethnicities, and singlehanded practice status.
The odds ratio on capitation funding per 100 patients from the full models for each CQC domain are shown in Table 6. Higher funding was significantly associated with higher CQC ratings across all five quality domains. We used the results from the ordered logistic regression models with the full set of explanatory variables to calculate the probability of achieving an overall practice rating of 'Outstanding' or 'Inadequate'. Figure 3 shows the average predicted probability of achieving the practice rating for each funding level calculated using actual values of practice non-funding characteristics (year effects, patient characteristics & practice characteristics). Higher funding was associated with reduced probability of achieving an 'Inadequate' ratings and increased probability of an 'Outstanding' quality rating.
We also compared the probability of achieving an 'Outstanding' rating at different levels of GP funding for training versus non-training practices (Figure 4), for single-handed versus group practices ( Figure 5), and for rural versus urban practices ( Figure 6). At all levels of funding training practices have higher probabilities than non-training, group practices have higher probabilities than single handed practices, and rural practices have higher probabilities than urban practices. In all cases higher capitation funding is associated with higher probabilities of an Outstanding rating.

Sensitivity Analyses
The Brant test (24)

Discussion
This study has demonstrated that higher capitation funding is associated with significantly higher overall practice quality ratings and ratings across all individual domains.
Practice characteristics such as training practice and group practice status were also associated with higher quality ratings, representing primary care structures which support higher quality of care. However, some factors related to the registered practice population, such as urban location, social deprivation and larger proportions of ethnic minority patients were negatively associated with the practice quality of care rating. Many of these factors are already known to be negatively associated with reported patient satisfaction (25) and QOF achievement (26). Including them in the model led to a stronger association of practice funding with practice quality rating. The likely reason for this is that practice capitation funding is positively correlated with patient characteristics which have negative effects on the quality rating. Thus including these patient characteristics in the model removes a source of bias from omitted variables which would otherwise tend to underestimate the positive association of funding with the quality rating. F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   11 This is the first study to explore the relationship between practice-level capitation funding and CQC ratings. We have been able to estimate the probability of achieving higher CQC ratings with additional funding. A variety of sensitivity analyses included in the methodology have confirmed the robustness of the ordered logistic regression modelling. These findings are based on a near complete sample of general practices across England. Using data linkages from a wide range of sources, and multilevel statistical models, this study has been able to demonstrate the independent effect of practice funding and practice characteristics on quality ratings, which might otherwise be confounded in single-level analyses.

Strengths and weaknesses of the study
However, there are several limitations. Routinely collected data are subject to coding and recording errors. As with all observational studies, significant associations, even if large, may not be causal. Although a wide range of potential confounders were included in the models, residual confounding cannot be excluded.

Comparison with existing literature
These findings add to those of a previous study which found that increased general practice funding was associated with reduced emergency hospital admissions and A&E attendances (11). Further, it has demonstrated that the current capitation funding formula may contribute to the persistence of the inverse care law with deprived areas experiencing lower quality of care, as defined by inspection ratings (27). Consistent with our study, others have found that GP practice funding correlate negatively with healthcare need predictors such as deprivation and non-white ethnicity (28). Previous studies have also demonstrated that greater GP workload may be associated with higher levels of social deprivation and with a higher proportion of Asian patients (29). Similarly, practices with a greater proportion of ethnically diverse patients reported worse patient experience (30). Though rural practice were more likely to achieve an 'Outstanding' rating, consultation rates and rurality do not appear to be associated (29).

Implications for policy and practice
This work provides further evidence of the association between general practice funding and the quality of primary care. A causal association is plausible and supports the argument that increased quality and safety of patient care may be achieved through additional investment.
The recently published NHS Long Term Plan(31) outlines proposals to offer substantial increases to capitation payment together with an emphasis on inter-practice cooperation through the formation of primary care networks. Both factors are likely to further influence the relationship between funding and the quality of primary care and will require further study. The NHS Long Term Plan also emphasises the importance of collaborations between the CQC and practices working in local areas implying collective responsibility for improving the quality of care in localities. Our findings suggest that revisions to the primary care capitation formula are necessary to ensure that more funding is provided in areas of high deprivation and ethnic minority populations reduce inequalities in the quality of care.

Unanswered questions and future research
Future research should extend these findings to subsequent 3-year cycles of quality  Tables   Table 1: The five key domains for CQC Inspections   Domain  Description  Safe Patients are protected from abuse and avoidable harm

Abstract:
Objective To explore the relationship between general practice capitation funding and quality ratings based on general practice inspections.
Design Cross-sectional study pooling three years of primary care administrative data.
Setting UK primary care. Introduction Improving the quality of care is a major focus of UK government health policy(1). High quality health care has three main components: clinical achievement, patient experience and patient safety(2). There is wide variation between general practices in the achievement of clinical care quality indicators and patient reported satisfaction (3,4). The QOF has had limited impact on reducing secondary care costs (6) or improving primary care performance (7,8). In terms of financial incentivisation, the QOF accounted for approximately 7.8% of funding received by general practices in England in 2016 (9). In contrast, capitation payments represent the largest proportion of funding to general practice (54% in 2016) and are related to the number of registered patients in each practice (9), adjusted for factors thought to increase the demand on primary care services (10). Other components of general practice funding include additional payments for postgraduate training, the provision of additional clinical services ('enhanced services') and various Greater capitation spending on general practices has been found to be associated with reductions in secondary care usage and costs, and increased patient satisfaction (12). Studies have also shown that leadership within the practice organisation plays a key role in the delivery of high quality care (13). Until recently, nationally derived metrics of inspection-based primary care quality were unavailable. Since October 2014 all general practices have been subjected to inspections by the Care Quality Commission (CQC) (14), (5). The CQC reports on the extent to which practices are caring, effective, responsive to the needs of patients, safe, and well-led (5,15) and also combines these five domains to produce an overall practice rating. These five domains incorporate components of clinical achievement, patient experience and patient safety(2). In this study, we assess the relationship of practice capitation funding with overall CQC ratings and with the individual CQC domains. We aimed to examine the relationship between practice funding and the quality of care as determined by inspection-based quality assessment. Analysis of total practice funding would have introduced confounding through inclusion of quality-related payments. We therefore used capitation funding as our measure of practice funding since this financial indicator is independent of financial rewards associated with quality achievement such as the QOF and other national and local incentive schemes.

Data sources
We linked practice-level data on NHS payments to general practice identifiers(16), CQC inspection ratings (15), NHS administrative datasets, General and Personal Medical Services

Care Quality Commission Ratings
CQC ratings are based on publicly available data (such as QOF and General Practice Patient Survey (19)), practice inspections, interviews with patients and staff, complaints, clinical record reviews, reviews of practice documents and policies (15). We used CQC ratings for practices with completed CQC reports first inspected between November 2014 and December 2017 (n = 7310, 95% of all practices). Practice ratings were obtained from the CQC; these data are publicly available on request. For practices which required reinspection only the first inspection score was included in the analysis. The five domains of quality described by CQC inspections are summarised in Table 1; each is rated on a 4-point scale.

Practice data
Data for all general practices in England were obtained from the General and Personal  inspection and the year in which the practice was inspected.

Sample
We linked inspected practices (n=7310) with funding data for their year of inspection. We excluded atypical practices with ≤750 registered patients (n=10) or ≤500 patients per FTE GP (n= 8) following a previously used method (23). Practices with recorded negative (n= 2) or zero funding (n = 52) were excluded. The final analysis sample consisted of 7238 practices.

Data Analysis
Analysis was conducted at GP practice level. Since the CQC rating outcomes are ordered categories we used ordered logistic regression to model the relationship between funding and the practice CQC ratings (24). Separate models were estimated for each domain.
The key explanatory variable was capitation funding per patient. Capitation funding per patient is reported in standard deviation units. In addition to patient and practice characteristic covariates, the regression models included year effects to allow for inspection year and annual general practice funding uplifts. We accounted for local area effects by adjusting for clustering at Clinical Commissioning Group (CCG) level. Multicollinearity was tested for by calculating the Variance Inflation Factor (VIF) and variables with a value for VIF>5 were excluded. The proportional odds assumption of the ordered logit model was also tested (25). We report the odds ratio from the estimated models.
We calculated the average marginal effects of funding on the predicted probabilities of achieving overall ratings of "outstanding" and "inadequate" for all practices. We also compared the predicted probabilities of an "outstanding" overall rating at different practice capitation funding levels for training versus non-training practices, single-handed versus multi-handed practices, and rural versus urban practices. STATA 14 (StataCorp, College Station, TX) was used for all statistical analyses. commenting on the potential impact of outcomes. A lay summary was also provided.

Results
Summary statistics for the main characteristics of the general practices are provided in Table   2. Mean practice capitation funding per registered patient increased from £77.49 in 2014/5 to £83.17 in 2016/7 ( Table 3). The mean capitation funding per patient across the CQC inspection period was £79.48. The standard deviation of the mean capitation funding per patient was £22.00.
The distribution of practice ratings across each quality domain is shown in Figure 1. A total of 79% (n = 5774) of practices achieved an overall rating of 'Good', while only 4% (294) achieved an overall rating of 'Outstanding'. 'Inadequate' ratings varied across the domains, from 1% (caring domain) to 6% (safety domain) and 4% (overall). Figure 2 shows the difference in capitation funding for practices with the lowest quality rating compared to those with the highest quality rating. In each domain, 'Inadequate' practices received less capitation funding. Using an independent group t-test this difference was found to be significant for three domains (caring, safe, well-led) and for the overall practice rating.   Table 4). In addition to higher practice capitation funding, rural practice and training practice status were significantly associated with higher overall practice ratings. For example, the adjusted odds ratio of a training practice achieving an 'outstanding' The odds ratios for capitation funding per patient from the full models for each CQC domain are shown in Table 6. Higher capitation funding was significantly associated with higher CQC ratings across all five quality domains.
We used the results from the ordered logistic regression models with the full set of explanatory variables to calculate the probability of achieving an overall practice rating of 'Outstanding' or 'Inadequate' at different levels of capitation funding. Figure 3 shows the average predicted probability of achieving an 'Outstanding' rating for a range of per capita funding levels. The probabilities are the average of the estimated probabilities for each practice calculated at each funding level using actual values of the practice non-funding characteristics (year effects, patient characteristics and practice characteristics). Figure 4 shows the average predicted probability of achieving an 'Inadequate' practice rating. Higher capitation funding was associated with reduced probability of achieving an 'Inadequate' rating and increased probability of an 'Outstanding' quality rating. At capitation payments above £100 per patient, practices have a greater probability of being rated as "Outstanding" rather than "Inadequate".
We also compared the probability of achieving an 'Outstanding' rating at different levels of practice capitation funding for training versus non-training practices ( Figure 5), for singlehanded versus group practices (Figure 6), and for rural versus urban practices (Figure 7). At all levels of funding, the probability of achieving an 'Outstanding' rating is higher for training practices than non-training practices, for group practices than single handed practices, and rural practices than urban practices. In all cases higher capitation funding is associated with higher probabilities of an Outstanding rating.

Sensitivity Analyses
The Brant test (25)

Discussion
This study has demonstrated that higher capitation funding is associated with significantly higher overall practice quality ratings and ratings across all individual domains.
Practice characteristics such as post-graduate training practice and group practice status were also associated with higher quality ratings, representing primary care structures which support higher quality of care. However, some factors related to the registered practice population, such as urban location, social deprivation and larger proportions of ethnic minority patients were negatively associated with the practice quality of care rating. Many of these factors are already known to be negatively associated with reported patient satisfaction (26) and QOF achievement (27). Including them in the model led to a stronger association of practice capitation funding with practice quality rating. The likely reason for this is that

Strengths and weaknesses of the study
This is the first study to explore the relationship between practice-level capitation funding and practice quality as measured by CQC ratings. The findings are based on a near complete sample of general practices across England. Using data linkages from a wide range of sources, and multilevel statistical models, this study has been able to demonstrate the independent effects of practice funding and practice characteristics on quality ratings, which might otherwise be confounded in single-level analyses. A variety of sensitivity analyses have confirmed the robustness of the ordered logistic regression modelling.
However, there are several limitations. Routinely collected data are subject to coding and recording errors. As with all observational studies, significant associations, even if large, may not be causal. Although a wide range of potential confounders were included in the models, confounding by omitted variables cannot be excluded.

Comparison with existing literature
These findings complement those of a previous study which found that increased general practice capitation funding was associated with reduced emergency hospital admissions and Accident and Emergency attendances (12). In a country level European analysis it was found that systems relying on capitation funding were more responsive than those based on fee for service or mixed payment (28). However, analysis of Scottish general practices suggests that capitation funding may contribute to the persistence of the inverse care law with deprived areas experiencing lower quality of care, as defined by inspection ratings (29). Consistent with our study, others have found that GP practice funding is negatively correlated with healthcare need predictors such as deprivation and non-white ethnicity (30). Previous studies have also demonstrated that greater GP workload may be associated with higher levels of social deprivation and with a higher proportion of Asian patients (31). Similarly, practices with a greater proportion of ethnically diverse patients reported worse patient experience (32). Our work is also consistent with a recent study which demonstrated that GPs co-located with other GPs and professionals had improved outcomes compared with single-handed GP practices such as broader provision of technical procedures, wider coordination with secondary care and increased collaboration among different providers (33).
Our study was based on funding data for general practices but was unable to study the relationship between quality ratings and individual GP income. However, values for overall 'profit' per practice are expected to become available in due course. Other studies have confirmed that incentives based on personal income may influence both quality achievement and productivity (34).

Implications for policy and practice
This work provides further evidence of the association between general practice capitation funding and the quality of primary care. A causal association is plausible and supports the influence the relationship between funding and the quality of primary care and will require further study. Our findings suggest that revisions to the primary care capitation formula are necessary to ensure that additional funding is provided in urban areas of high deprivation and ethnic minority populations in order to address quality of care inequalities.

Unanswered questions and future research
Future research could extend similar analyses to subsequent 3-year cycles of quality inspection. A longitudinal approach, relating changes in funding to changes in outcomes, is likely to provide more accurate estimates of the effect of funding. Complementary qualitative analysis is likely to provide insight into mechanisms underlying the link between better funded practices and higher quality rating achievement.

Conclusion
Higher capitation funding was consistently associated with higher overall and domain quality ratings yielded by CQC inspections. This study suggests that measured and inspected dimensions of the quality of care are related to the underlying funding allocated to each general practice, implying that additional funding may be associated with higher levels of primary care quality.

What is already known on this topic
 Few studies on the relationship between capitation funding and quality Patients are protected from abuse and avoidable harm

Effective
Care, treatment and support achieves good outcomes, helps patients to maintain quality of life and is based on the best available evidence Caring Staff involve and treat patients with compassion, kindness, dignity and respect Responsive Services are organised so that they meet patients' needs Well-led The leadership, management and governance of the organisation make sure it's providing high-quality care that's based around the individual needs, that it encourages learning and innovation, and that it promotes an open and fair culture Adapted from: CQC. The five key questions we ask (36) 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59          Conclusion: Higher capitation funding was consistently associated with higher ratings across all CQC domains and in the overall practice rating. This study suggests that measured dimensions of the quality of care are related to the underlying capitation funding allocated to each general practice, implying that additional capitation funding may be associated with higher levels of primary care quality.

Strengths and limitations
 A cross-sectional study covering three years of primary care data  The definition of primary care quality used in this study was multidimensional, based on inspection findings and covering patient safety, patient experience, clinical effectiveness.
Greater capitation spending on general practices has been found to be associated with reductions in secondary care usage and costs, and increased patient satisfaction (12). Studies have also shown that leadership within the practice organisation plays a key role in the delivery of high quality care (13). Until recently, nationally derived metrics of inspection-based primary care quality were unavailable. Since October 2014 all general practices have been subjected to inspections by the Care Quality Commission (CQC) (14), (5). The CQC reports on the extent to which practices are caring, effective, responsive to the needs of patients, safe, and well-led (5,15) and also combines these five domains to produce an overall practice rating. These five domains incorporate components of clinical achievement, patient experience and patient safety(2). In this study, we assess the relationship of practice capitation funding with overall CQC ratings and with the individual CQC domains. We aimed to examine the relationship between practice funding and the quality of care as determined by inspection-based quality assessment. Analysis of total practice funding would have introduced confounding through inclusion of quality-related payments. We therefore used capitation funding as our measure of practice funding since this financial indicator is independent of financial rewards associated with quality achievement such as the QOF and other national and local incentive schemes.

Care Quality Commission Ratings
CQC ratings are based on publicly available data (such as QOF and General Practice Patient Survey (19)), practice inspections, interviews with patients and staff, complaints, clinical record reviews, reviews of practice documents and policies (15). We used CQC ratings for practices with completed CQC reports first inspected between November 2014 and December 2017 (n = 7310, 95% of all practices). Practice ratings were obtained from the CQC; these data are publicly available on request. For practices which required reinspection only the first inspection score was included in the analysis. The five domains of quality described by CQC inspections are summarised in Table 1; each is rated on a 4-point scale.

Data Analysis
Analysis was conducted at GP practice level. Since the CQC rating outcomes are ordered categories we used ordered logistic regression to model the relationship between funding and the practice CQC ratings (24). Separate models were estimated for each domain.
The key explanatory variable was capitation funding per patient (measured in standard deviation units). We also include patient and practice characteristic covariates, thereby reducing the risk of bias from the omission of variables which might affect the CQC rating and are correlated with practice capitation funding. The regression models included year effects to allow for inspection year and annual general practice funding uplifts. We accounted for local area effects by adjusting for clustering at Clinical Commissioning Group (CCG) level.
Multicollinearity was tested for by calculating the Variance Inflation Factor (VIF) and variables with a value for VIF>5 were excluded. The proportional odds assumption of the ordered logit model was also tested (25). We report the odds ratio from the estimated models.
We calculated the average marginal effects of funding on the predicted probabilities of achieving overall ratings of "outstanding" and "inadequate" for all practices. We also compared the predicted probabilities of an "outstanding" overall rating at different practice capitation funding levels for training versus non-training practices, single-handed versus multi-handed practices, and rural versus urban practices. STATA 14 (StataCorp, College Station, TX) was used for all statistical analyses.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   8 Funding for this study included funding of a dedicated patient involvement group. Patients were involved in developing plans for the study design, approving the outcome measures and commenting on the potential impact of outcomes. A lay summary was also provided.

Results
Summary statistics for the main characteristics of the general practices are provided in Table   2. Mean practice capitation funding per registered patient increased from £77.49 in 2014/5 to £83.17 in 2016/7 ( Table 3). The mean capitation funding per patient across the CQC inspection period was £79.48. The standard deviation of the mean capitation funding per patient was £22.00.
The distribution of practice ratings across each quality domain is shown in Figure 1. A total of 79% (n = 5774) of practices achieved an overall rating of 'Good', while only 4% (294) achieved an overall rating of 'Outstanding'. 'Inadequate' ratings varied across the domains, from 1% (caring domain) to 6% (safety domain) and 4% (overall). Figure 2 shows the difference in capitation funding for practices with the lowest quality rating compared to those with the highest quality rating. In each domain, 'Inadequate' practices received less capitation funding. Using an independent group t-test this difference was found to be significant for three domains (caring, safe, well-led) and for the overall practice rating.   Table 4). In addition to higher practice capitation funding, rural practice and training practice status were significantly associated with higher overall practice ratings. For example, the adjusted odds ratio of a training practice achieving an 'outstanding' The odds ratios for capitation funding per patient from the full models for each CQC domain are shown in Table 6. Higher capitation funding was significantly associated with higher CQC ratings across all five quality domains.
We used the results from the ordered logistic regression models with the full set of explanatory variables to calculate the probability of achieving an overall practice rating of 'Outstanding' or 'Inadequate' at different levels of capitation funding. Figure 3 shows the average predicted probability of achieving an 'Outstanding' rating for a range of per capita funding levels. The probabilities are the average of the estimated probabilities for each practice calculated at each funding level using actual values of the practice non-funding characteristics (year effects, patient characteristics and practice characteristics). Figure 4 shows the average predicted probability of achieving an 'Inadequate' practice rating. Higher capitation funding was associated with reduced probability of achieving an 'Inadequate' rating and increased probability of an 'Outstanding' quality rating. At capitation payments above £100 per patient, practices have a greater probability of being rated as "Outstanding" rather than "Inadequate".
We also compared the probability of achieving an 'Outstanding' rating at different levels of practice capitation funding for training versus non-training practices ( Figure 5), for singlehanded versus group practices (Figure 6), and for rural versus urban practices (Figure 7). At all levels of funding, the probability of achieving an 'Outstanding' rating is higher for training practices than non-training practices, for group practices than single handed practices, and rural practices than urban practices. In all cases higher capitation funding is associated with higher probabilities of an Outstanding rating.

Sensitivity Analyses
The Brant test(25) assesses the proportional odds assumption that the distance between each category is equivalent. Four of the variables included in our model (region, proportion of patients aged 0-4 years, contract type and single-hander status) did not meet the assumption of proportionality of the odds ratios. However, our variable of interest, capitation funding per patient, did not violate the proportional odds assumption. A partial proportional odds model excluding these four variables, estimated by generalised ordered logistic regression, yielded similar results to our main model: higher capitation funding was significantly associated with increase probability of achieving an 'Outstanding' rating (OR 1.14, 95% CI: 1.04 to 1.25).

Discussion
This study has demonstrated that higher capitation funding is associated with significantly higher overall practice quality ratings and ratings across all individual domains.
Practice characteristics such as post-graduate training practice and group practice status were also associated with higher quality ratings, representing primary care structures which support higher quality of care. However, some factors related to the registered practice population, such as urban location, social deprivation and larger proportions of ethnic minority patients were negatively associated with the practice quality of care rating. Many of these factors are already known to be negatively associated with reported patient satisfaction (26) and QOF achievement (27). Including them in the model led to a stronger association of practice capitation funding with practice quality rating. The likely reason for this is that

Strengths and weaknesses of the study
This is the first study to explore the relationship between practice-level capitation funding and practice quality as measured by CQC ratings. The findings are based on a near complete sample of general practices across England. Using data linkages from a wide range of sources, and multilevel statistical models, this study has been able to demonstrate the independent effects of practice funding and practice characteristics on quality ratings, which might otherwise be confounded in single-level analyses. A variety of sensitivity analyses have confirmed the robustness of the ordered logistic regression modelling.
However, there are several limitations. Routinely collected data are subject to coding and recording errors. As with all observational studies, significant associations, even if large, may not be causal. Although a wide range of potential confounders were included in the models, confounding by omitted variables cannot be excluded.

Comparison with existing literature
These findings complement those of a previous study which found that increased general practice capitation funding was associated with reduced emergency hospital admissions and Accident and Emergency attendances (12). In a country level European analysis it was found that systems relying on capitation funding were more responsive than those based on fee for service or mixed payment (28). However, analysis of Scottish general practices suggests that capitation funding may contribute to the persistence of the inverse care law with deprived areas experiencing lower quality of care, as defined by inspection ratings (29). Consistent with our study, others have found that GP practice funding is negatively correlated with healthcare need predictors such as deprivation and non-white ethnicity (30). Previous studies have also demonstrated that greater GP workload may be associated with higher levels of social deprivation and with a higher proportion of Asian patients (31). Similarly, practices with a greater proportion of ethnically diverse patients reported worse patient experience (32). Our work is also consistent with a recent study which demonstrated that GPs co-located with other GPs and professionals had improved outcomes compared with single-handed GP practices such as broader provision of technical procedures, wider coordination with secondary care and increased collaboration among different providers (33).
Our study was based on funding data for general practices but was unable to study the relationship between quality ratings and individual GP income. However, values for overall 'profit' per practice are expected to become available in due course. Other studies have confirmed that incentives based on personal income may influence both quality achievement and productivity (34).

Implications for policy and practice
This work provides further evidence of the association between general practice capitation funding and the quality of primary care. A causal association is plausible and supports the influence the relationship between funding and the quality of primary care and will require further study. Our findings suggest that revisions to the primary care capitation formula are necessary to ensure that additional funding is provided in urban areas of high deprivation and ethnic minority populations in order to address quality of care inequalities.

Unanswered questions and future research
Future research could extend similar analyses to subsequent 3-year cycles of quality inspection. A longitudinal approach, relating changes in funding to changes in outcomes, is likely to provide more accurate estimates of the effect of funding. Complementary qualitative analysis is likely to provide insight into mechanisms underlying the link between better funded practices and higher quality rating achievement.

Conclusion
Higher capitation funding was consistently associated with higher overall and domain quality ratings yielded by CQC inspections. This study suggests that measured and inspected dimensions of the quality of care are related to the underlying funding allocated to each general practice, implying that additional funding may be associated with higher levels of primary care quality.