Associations of primary care workforce composition with population, professional and system outcomes: retrospective cross-sectional analysis

BACKGROUND
The diversification of types of staff delivering primary care may affect professional, population, and system outcomes.


AIM
To estimate associations between workforce composition and outcomes.


DESIGN AND SETTING
Cross-sectional analysis of 6210 GP practices from a range of geographical settings across England in 2019.


METHOD
A multivariable regression analysis was undertaken, relating numbers of staff in four groups - GPs, nurses, healthcare professionals, and health associate professionals - to patient access and satisfaction, quality of clinical care and prescribing, use of hospital services, GP working conditions (subsample of practices), and costs to the NHS. Data were obtained from the GP Patient Survey 2019, Quality and Outcomes Framework, prescribing data, the Hospital Episode Statistics database, the NHS Payments to General Practice 2019/2020, and the Tenth National GP Worklife Survey 2019.


RESULTS
Having additional GPs was associated with higher levels of satisfaction for the GPs themselves and for patients, whereas additional staff of other types had opposite associations with these outcomes. Having additional nurses and health associate professionals was associated with lower costs per prescription but more prescribing activity than having additional staff from the other two groups. Having more GPs was associated with higher costs per prescription and lower use of narrow-spectrum antibiotics compared with the other staff groups. Except for health associate professionals, greater staff numbers were associated with more hospital activity.


CONCLUSION
Professional, population, and system outcomes showed a variety of associations with primary care workforce composition. Having additional nurses was associated with lower quality in some aspects, and higher costs and activity. The association between additional healthcare professionals or health associate professionals and higher costs was less than that for additional GPs, but was also linked to lower patient and GP satisfaction.


Introduction
The supply of general practitioners is under strain in most countries, with workload and other pressures driving burnout and early exit from the workforce. (1)(2)(3)(4) One proposed solution is to broaden the workforce, introducing new types of practitioners to supplement care provided by medically-qualified practitioners. (5,6) The NHS in England has been particularly affected by these pressures (2)(3)(4)7) and has gone further than most countries in seeking to broaden the primary care workforce. A substantial investment of £746 million has been pledged in 2021/22 to employ a wide range of new types of practitioner, including pharmacists, physician associates, care co-ordinators, physiotherapists and mental health support workers.(8) However, there is limited high-quality evidence to support these developments as much of the existing research has been more narrowly focused on the mix of physicians and nurses,(9) on the performance of particular types of practitioners in their roles, (10)(11)(12) and on a narrow range of outcomes. (13) There has been little focus on the wider impacts of employing a more diverse range of practitioners. We do not know whether such practitioners can substitute for general practitioners (GPs), (14) nor do we know their impact on quality of care and patient satisfaction. Existing evidence does not consider cost-effectiveness or the impact on care provision at practice level. (15) Primary care in England is a good setting in which to explore the impact of skill-mix change for several reasons. Firstly, it is delivered by a large number of small practices which have discretion over their employment, leading to considerable variation in workforce composition. Secondly, more detailed data on workforce composition at practice level have recently become available. (16) Finally, national policy is driving further changes in workforce and their potential impacts need to be understood. We use nationwide data from a large sample of practices to examine whether the inputs of GPs, nurses and other healthcare professionals are associated with a wide range of population, professional and system outcomes.

Data
We obtained data on different types of practitioners employed in general practices across England in September 2019 from NHS Digital. Information on headcounts and full-time equivalents are available for 35 categories of professionals providing direct patient care.
We obtained data on outcomes from the following sources: GP Patient Survey; Quality and Outcomes Framework (QOF); Prescribing Data; Hospital Episode Statistics; NHS Payments to General Practice; and the GP Worklife Survey (see Appendix 1). All sources provide practice-level figures for all practices in England except the GP Worklife Survey, which includes a sample of individual GP respondents. (7) We analysed outcomes experienced by the population, practitioners, and the healthcare system. For the population, we consider patient experience of access using data from the GP Patient Survey, and consider clinical quality using measures of QOF points scored and measures of prescribing quality including ratio of broad to narrow antibiotics prescribed and volume of prescribing. For the healthcare system we use Hospital Episode Statistics to look at use of hospital services, including attendance at Emergency Departments, emergency admissions and hospital outpatient referral, and we consider costs to the system based upon payments to practices using NHS Payments Data and prescribing data and costs associated with hospital activity using average unit costs. We address practitioner outcomes in a subset of practices using data from the GP Worklife Survey. These data sources are described in more detail in the supplementary appendix.
We sourced data on a range of population and practice characteristics that may be associated with workforce composition and outcomes and were included as covariates. These included registered population size and age structure, indicators of population healthcare needs, local area income deprivation and average wages, the practice's dispensing status, contract type, rural location, and NHS region.
There were 6,770 practices open in September 2019 with at least 1,000 registered patients. Practice workforce data is self-reported by practices through the National Workforce Reporting System (NWRS). Where practices have not provided any or only incomplete data, NHS Digital imputes values. We only analysed practices that provided a complete set of workforce information, which in September 2019 were 6,296 (93%) practices. Of these, 6,210 practices had full data availability across all covariates and were included in our analysis. For the analyses of the GP Worklife survey, the sample size varies between 1,191 and 1,270 responding GPs.

Analysis
We measured levels of staff input in full-time equivalents (FTEs). The current pattern of employment of new types of clinical staff in England is complicated, with incentive payments available to employ 12 different types of staff (8), including health coaches, clinical pharmacists, nursing assistants and many others. The roles undertaken by these groups vary significantly. In order to make sense of the potential impact of such a wide range of staff on patient care we followed the World Health Organization classification (17), which distinguishes between staff able to provide therapeutic interventions n their own right and those carrying out delegated tasks and groups staff into four broad categories: GPs; Nurses; Health Professionals (excluding GPs and Nurses); and Health Associate Professionals ( Table 1). The Health Professionals category contains staff that make professional assessments or deliver therapeutic interventions, such as pharmacists or physician associates. The Health Associate Professionals category contains any other staff involved in patient treatment, such as healthcare assistants, dispensers, trainees, and apprentices.
[ Table 1 here] We used linear regression to relate practice outcomes to cubic functions of the levels of each staff type. (18,19) We also included interactions between GPs and each of the other three staff types to test for direct substitution (negative interaction coefficient) or direct complementarity (positive interaction coefficient) between staff groups. We standardised all outcome measures using z-scores to aid comparisons. We weighted all analyses by the denominator of the outcome (size of registered population for most outcomes, see Appendix 3).
We summarised the estimated effects of the staff variables by calculating the changes in each outcome for unit-changes in the FTEs of each staff type, holding all other characteristics at their median values for continuous variables and mean values for discrete variables. These are marginal effects of staff changes for the average practice. We also graphed means for the outputs at key percentiles of staff input levels, conditional on the covariates.
We ran similar analysis for total costs, calculated as the sum of the payments made to the practice by the NHS for providing primary care services, the costs of the prescriptions generated by the practice, and the volumes of four types of hospital activity used by the practice's patients costed using national unit-costs.(20) To account for the skewed distribution of these costs, we estimated a generalised linear model (GLM) using a log link function and a gamma distribution. (21) We estimated this model for total costs and for six components of total costs. To check the stability of the coefficients to the inclusion of covariates, we also estimated the model for total costs adding covariates sequentially.
We used Stata version 15.1. We computed Huber-White robust standard errors to allow for heteroscedasticity. We also checked the sensitivity of the results to allowing the errors to be correlated across different outcomes using seemingly unrelated regression equations (22). We measured goodness-of-fit using R-squared and Root Mean Squared Error statistics and estimated the joint significance of the workforce variables using F-tests.

Results
There is considerable variation in staffing levels across practices ( Table 2). The median practice employs 4.3 full-time equivalent GPs, 1.9 Nurses, zero Health Professionals and 1.0 Health Associate Professional.
[ Table 2 here] Full regression results are shown in Appendix 4. For the majority of outcomes, the staff input variables are statistically significant at p<0.05 and mostly at p<0.01. The interaction terms between numbers of GPs and numbers of other staff are generally not statistically significant. This indicates a lack of substitution or complementarity for most outcomes. As exceptions, we find a small degree of complementarity between GPs and Nurses in the achievement of QOF points and between GPs and Health Professionals in relation to outpatient attendances and time since last Nurse appointment. The only substitution effects that were statistically significant were between GPs and Health Associate Professionals for prescribing of narrow-spectrum antibiotics and for total items prescribed.
The effects for the average practice are summarised in Table 3. The full patterns of the associations between staff levels and the outcomes are shown in the margin plots in Appendix 5. We examined the variance inflation factors (VIFs) and these were all below 10 except for the workforce and age category variables, which would be expected to be correlated.
The estimated effects are statistically significant but quite small in magnitude, with a one FTE difference in staff input associated with less than 0.1 standard deviation difference in outcome. Additional GPs are associated with shorter times since patients had a GP appointment and longer times since patients had a Nurse appointment, lower proportions of work that GPs think can be delegated and higher job satisfaction for GPs. Additional nurses have the opposite effects on practice functioning and environment. Health Professionals and Health Associate Professionals have similar effects to Nurses on practice functioning and environment, except Health Professionals have a much larger effect on perceived opportunities for work delegation and Health Associate Professionals are associated with lower GP job satisfaction.
Additional staff of any type is associated with achievement of higher QOF points. More GPs is associated with higher patient satisfaction and more of the other three staff groups is associated with lower patient satisfaction. Figures 1 and 2 show the marginal effect of additional staff members on patient satisfaction with making an appointment and their overall satisfaction with the practice. Additional Nurses and Health Associate Professionals are associated with more prescribing activity but lower average cost per prescription. Additional GPs increase costs per prescription and lower the use of narrow-spectrum antibiotics, while Health Professionals increase the use of narrow-spectrum antibiotics.
Additional staff in primary care is associated with more hospital activity of all types, with the exception of Health Associate Professionals. More Health Professionals have the largest impact on hospital activity, particularly outpatient attendances and emergency admissions.
The analysis of overall practice costs (Table 4) shows that a one FTE increase in Nurses results in the largest increase in costs (£512,000 per practice per year, 95% CI £446,000 to £579,000), followed by GPs (£362,000, 95% CI £326,000 to £398,000) and Health Associate Professionals (£256,000, 95% CI £195,000 to £318,000). The effect of Health Professionals on costs is the smallest and not statistically different from zero (£182,000, 95% CI -£33,000 to £396,000). The effects of different staff groups are similar across the components of costs, except that Health Professionals do not increase prescribing costs and Health Professionals and Health Associate Professionals do not increase hospital activity as much as GPs and Nurses.
The seemingly unrelated regression models and other sensitivity checks confirm our main results (Appendix 6).

Summary
The composition of the workforce in primary care practices is associated significantly with most indicators of professional, population and system outcomes, but the nature of these associations varies across outcomes. The number of GPs is associated positively with GP job satisfaction and with patient satisfaction. The numbers of Nurses, Health Professionals, and Health Associate Professionals relate to these outcomes in the opposite direction. More Nurses and Health Associate Professionals is associated with more prescribing activity but lower average cost prescribing. More GPs is associated with lower use, and more Health Professionals is associated with higher use, of narrow-spectrum antibiotics. With the exception of Health Associate Professionals, more staff in primary care is associated with higher levels of hospital activity per head of population. More staff increases the costs of the practice to the NHS, with the largest effect being for Nurses and the smallest effect being for Health Professionals.
Although the numbers of staff in each group is associated with outcomes, these effects are largely independent of the numbers of staff in the other groups. There are few instances where there is evidence of significant substitution or complementarity between GPs and the other three staff groups.

Strengths and limitations
We used nationwide data from a large number of primary care practices. Because practices are generally small organisations with substantial discretion over how they organise services, there is substantial variation in workforce composition to analyse. We obtained data from several sources and considered professional and patient satisfaction, as well as indicators of healthcare activity and quality. In addition, we considered how workforce composition was associated with costs. We adjusted for several population and practice characteristics associated with workforce composition and with outcomes.
In analysing the relationship between workforce composition and outcomes, we used a flexible regression model. This cubic function has been shown to be an acceptable alternative to the most flexible translog specification used in economic studies of production and less susceptible to multicollinearity in applied work because it requires fewer interaction terms (18,19). Although the cubic production function, like many flexible functions, results in correlated variables we are interested in their combined effect rather than the coefficients on individual measures which may be estimated imprecisely due to multicollinearity.
We relied on workforce data collected by the NHS. We worked closely with the data provider to identify practices where they had imputed values because of known problems with missing or incomplete data and excluded them from our analysis. This affected 7% of practices.
National data on primary care activity, for example numbers of patient consultations, are not currently available at practice level. Instead, we used indicators from the GP Patient Survey on how many months ago respondents had last seen a GP and last seen a Nurse. These showed the expected associations with numbers of GPs and Nurses but better quality data might have shown stronger relationships.
We did not include all aspects of costs that practices create for the National Health Service. However, all six of the cost components that we included showed similar patterns suggesting our results are not driven by one aspect of cost. We are not able to comment on health and social care expenditure more generally; it may be that increased expenditure in primary care results in lower costs in social care, for example.
Although we included several population and practice characteristics that may otherwise have confounded the relationship between workforce composition and outcomes, this is a cross-sectional study and may be prone to bias from unmeasured confounding.

Comparison with existing literature
Studies have shown that GP practices that are well organised tend to deliver high quality care(23) and tasks that can be guided by clinical protocols can be successfully delegated to nurses (e.g. chronic conditions management) (9,24,25). This is consistent with our finding of a small degree of complementarity between GPs and Nurses in the achievement of QOF points. However, it remains unclear whether the overlap in roles performed by nurses described by Lukewich et al also extends to other workers and whether this enhances care (26). Our finding that workforce composition is associated with mixed effects on the use of hospital services is consistent with research highlighting the complexity of this area (27,28). Our finding of both positive and negative outcomes associated with the Health Professional group is consistent with prior evidence highlighting the challenge of managing inter-professional interactions and negotiating role boundaries and regulations (29). Similarly, our results confirm the positive effects of pharmacists on prescribing practices and patient safety (30) . However, broader integration of pharmacists in practice seems to be still lacking (31,32).

Implications for Research and/or practice
Overall, our analysis highlights that workforce composition matters for professional, population and system outcomes. Because labour is such an important input to the production of healthcare, these findings are not surprising. However, they have rarely been investigated at the organisational level before, despite their obvious relevance for decisionmaking and future policy. Our findings demonstrate that different types of healthcare professionals are not simple substitutes for one another, and the quantity and quality of services generated in primary care will depend on who is employed to work in this setting.
We also show that the employment of different types of healthcare professionals has different effects on GPs' experience of their working lives.
We have shown also that employment of more of each of the different types of staff leads to higher costs to the NHS. This also may not seem surprising, but it should be noted that the costs we consider are not directly affected by staff employment costs. We find that employment of more staff means that a practice can claim more payments from the NHS for the services they provide and generate additional costs through increases in prescribing and increases in their populations' use of hospital services. This is particularly the case for the employment of Nurses but seems to be less the case for Health Professionals.
Finally, we find very little evidence of direct interactions between the effects of different types of workers on outcomes at the organisation level. This may indicate that different staff groups are not co-ordinating their work and are working quite independently. There may be gains to be made from encouraging more effective team-working.
We have only examined the effects of workforce composition on the work experiences of GPs because the available survey data is collected only from this staff group. Future work should consider the effects of workforce composition on the working lives of other primary care practitioners.
We have shown significant associations between cross-sectional variations in workforce composition across primary care practices and a range of outcomes and outcomes. As workforce composition continues to diversify over time, it will be important to examine how this affects outcomes and costs.

Author statement
JG wrote the first draft of the manuscript. SS led the cross-disciplinary collaborative development and completion of the study in which this analysis is embedded. JG and IF undertook the analysis. All authors contributed to the writing of the final manuscript. JG is the guarantor.
The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted. JG affirms that the manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned have been explained.

Competing interests declaration
Competing interests: All authors have completed the ICMJE uniform disclosure form and declare: no support from any organisation for the submitted work; no financial relationships with any organisations that might have an interest in the submitted work in the previous three years; no other relationships or activities that could appear to have influenced the submitted work.

Public involvement
Following early discussion with a patient and public involvement and engagement group, two members of the group maintained engagement with the research team throughout the study.

Funding
This study was funded by NIHR Health Services and Delivery Research Programme Project: 17/08/25 -An investigation of the scale, scope, and impact of skill mix change in primary care. The views expressed are those of the authors and not necessarily those of the NIHR or the Department of Health and Social Care.    Table 3. Associations between practice workforce and population, professional and system outcomes