Policies and strategies to retain and support the return of experienced GPs in direct patient care: the ReGROUP mixed-methods study

Aim 1

To develop a conceptual framework and comprehensive assessment of factors associated with GPs’ decisions to (1) quit direct patient care, (2) take career breaks from general practice and (3) return to general practice after a career break. There are two objectives:

1. to conduct a systematic review of existing literature to describe factors affecting these decisions in the UK and other high-income countries

2. to conduct a census survey of GPs in south-west England to provide a sampling frame to provide qualitative evidence from GPs intending to quit direct patient care and those who are currently taking or who are considering taking a career break with a view to identifying factors affecting quitting intentions, and to identify potentially modifiable factors relevant to these groups of GPs.

Aim 2

To identify the potential content and assess the evidence supporting key potential components of policies and strategies aimed at retaining experienced GPs and/or supporting the return of GPs to direct patient care following a career break. There are two objectives:

1. to outline the content of policies and strategies that will support the retention of these groups of GPs in direct patient care

2. to prioritise, using an expert panel and validated methodology [RAND/UCLA Appropriateness Method (RAM)], the proposed policies and strategies in respect of their feasibility and effectiveness.

Aim 3

To identify practices that may face supply–demand workforce imbalances at the macro (regional) and micro (general practice/GP) level within the next 5 years, with a view to strategically targeting relevant policies and strategies. There are two objectives:

1. drawing on a range of data, including the previously mentioned survey, to specify, develop and test the approach necessary to identify supply–demand imbalance at the level of individual practices

2. to use the approach developed in aim 3, objective 1, to identify general practices in south-west England (an area with broad representation of practice settings) at risk of workforce shortages owing to early retirement from direct patient care among experienced GPs and owing to GPs planning, or currently taking, a career break.

Aim 4

To assess the acceptability and feasibility of implementing the policies and strategies. The objective is:

1. to gather feedback from key stakeholders on the acceptability and likelihood of implementing the policies and strategies at a local level.

Current problems in the general practitioner workforce

There is said to be a ‘crisis’ in general practice in the UK on account of GPs leaving direct patient care or reducing their hours, and many others intending to do so. 40 A survey of UK GPs by the Commonwealth Fund in 2015 found that, at that time, nearly 30% of GPs planned to leave general practice within 5 years. 41 As well as those planning to retire, this included a substantial minority intending to switch medical specialty and others aiming to completely change career. Table 1 summarises the proportion of UK GPs who were aiming to leave general practice or direct patient care within the next 5 years from various studies.

In the three earlier surveys, conducted in England and Scotland between 1998 and 2001, between 14% and 22% of GPs said that they were likely to leave direct patient care within 5 years. By 2014 and 2015, the proportion of GPs in the UK saying that they would leave general practice in the next 5 years varied from 29% to 42% in different regions of England, with proportions of 29% and 35.3% in two surveys that randomly sampled from all UK GPs.

General practitioners appear to be more stressed and more dissatisfied than ever before, and more so than GPs and primary care physicians (PCPs) in all other countries surveyed. 41,43 This is happening at a time of increasing demand for primary care services, owing to demographic changes, such as the ageing population, and to more health care shifting away from hospital settings and disease specialists. There is, therefore, an urgent need to understand what factors are driving GPs to leave patient care, and which GPs may be targeted by interventions that aim to improve the retention of GPs in patient care.

Justification and review question

Although there have been recent systematic reviews of the effectiveness of strategies to retain primary care doctors in the workforce, or of strategies to reduce specific determinants of quitting (such as ‘burnout’), this is the first comprehensive systematic review to identify and describe the factors that underlie GPs’ decisions and intentions about quitting, taking career breaks or reducing their work hours. Our review question was ‘what are the factors which affect GPs’ decisions to quit direct patient care (including reducing their time commitment to it), take career breaks from general practice or return to general practice after a career break?’.

The review protocol was registered on the PROSPERO prospective register as PROSPERO CRD42016033876.

Searches

We sought published articles and ‘grey’ literature published in English from 1990 onwards. The first search identified published, unpublished and grey literature studies and was run in a variety of relevant databases. The second search drew on supplementary search methods (e.g. forward and backward citation chasing, web searches of relevant organisations) to locate unpublished studies and grey literature. The grey literature searches are fully described in Appendix 3.

The following databases were searched in January 2016 with update searches in April 2016: MEDLINE, MEDLINE In-Process & Other Non-Indexed Citations, PsycINFO, Healthcare Management Information Consortium (HMIC), The Cochrane Library, Applied Social Sciences Index of Abstracts (ASSIA) and Web of Science.

Inclusion criteria and processes

Process of identification and selection of studies

The titles and abstracts of search results were screened against these criteria (an initial sample was independently screened by two reviewers to establish consistency in exclusion decisions). Titles and abstracts that could not be excluded were sought as full-text articles, and the inclusion criteria were applied to these. The process and resulting search hits are shown in Figure 2.

FIGURE 2.

Preferred Reporting Items for Systematic Reviews and Meta-Analyses flow diagram showing the process of study selection. a, See Appendix 5 for a list of full-text exclusions with reasons.

Data extraction and quality assessment

Strategy for data synthesis

Overview

Table 2 shows the key characteristics of the 22 survey studies of UK GPs published since 1990. Most of the studies were cross-sectional surveys and surveyed the actual quitting or quitting intentions of GPs in a particular year or month. The earliest survey gathered data from UK GPs in 1991–4,64 whereas five of the most recent surveys were conducted in 2014–15. 17,41,43,44,55 The sample sizes of the surveys of UK GPs ranged from 40 to 4421.

The bottom third of Table 2 shows the key characteristics of the 12 survey studies of non-UK GPs, all but two of which report surveys conducted since 2006. The results and synthesis of these 12 non-UK studies are not reported further in this chapter (see Report Supplementary Material 1).

Geographical coverage of the surveys

Of the 22 surveys of UK GPs, seven surveyed GPs from the whole of the UK,41,43,53,56,64–66 two surveyed GPs from England and Scotland combined,45,63 two surveyed GPs from the whole of England,59,62 five surveyed GPs in a particular region or city of England17,44,52,61,76 and five surveyed GPs in Scotland. 42,46,54,57,58

Characteristics of general practitioners surveyed

Most of the surveys of UK GPs were of all practising GPs in the country or a region, regardless of age, year of qualification or contract type/mode of employment (Table 2). However, some surveyed only GP principals (i.e. practice partners)54,58,62 or recruited only GP principals and non-principals (not locums). 42,46,63 Only four studies surveyed GPs who had already quit patient care. 52,59,64,66

Most UK studies reported a mean age (or median age band) of between 40 and 55 years. However, one study had much younger respondents, because it set out to survey recently qualified GPs. 64 Two surveys specifically targeted older GPs: Chambers et al. 54 surveyed only unrestricted principals aged > 55 years (mean age not reported), whereas Luce et al. 61 surveyed GPs aged ≥ 45 years (mean age 52 years). Across the UK studies, female GPs accounted for between 29% and 75% of the GPs surveyed and tended to account for a higher proportion of GPs in the later surveys.

Overview of the UK questionnaire surveys

Overview of the qualitative research studies and data

Five papers based on qualitative semistructured interviews with practising or retired British GPs were retrieved; all studies had been conducted in England. A further qualitative semistructured interview study, with practising GPs who were working part-time in clinical practice in Australia, was found but is not reported in this chapter (see www.journalslibrary.nihr.ac.uk/programmes/hsdr/1419602/#/). The main characteristics of these studies, and the GPs interviewed within them, are shown in Table 4.

Two of the papers reporting studies from England were based on almost the same set of interviews. 17,18 The 2016 study by Doran et al. 51 focused on why GPs had left medical practice, whereas the other four studies were wholly or dominantly with practising GPs. All of the semistructured interviews were quality appraised using an adapted Wallace checklist and found to be of reasonable76,80 to good17,18,51,79 quality.

The synthesis is presented as a series of linked themes, each of which belongs to one of five categories. The five main categories of explanatory theme were undoable/unmanageable (including workload and related pressures), morale, impact of organisational changes, projected future and multiple options and strategies. This analytical framework is summarised in Table 5.

What then follows is a heavily abridged version of the qualitative synthesis; some subthemes are mentioned very briefly and there are no illustrative quotations. The full write-up of the synthesis of qualitative studies, with direct GP quotations, is available in our separate full report of the systematic review (see Report Supplementary Material 1).

Explanatory model and narrative

The themes and detailed factors emerging from the qualitative synthesis of semistructured interview studies conducted in the UK were also used to construct a pictorial explanatory model (Figure 5). It provides an overview of the key contexts and factors that were found to relate to GPs quitting or intending to quit patient care. The applicability of the explanatory model was confirmed following feedback on the model during meetings with PPI representatives and co-investigators.

FIGURE 5.

Flow diagram of factors contributing to GPs leaving general practice. Admin, administrative; IT, information technology.

There are three main ‘domains’ in the flow diagram: (1) factors associated with low levels of job satisfaction, (2) factors associated with high levels of job satisfaction and (3) factors associated with linked to the doctor–patient relationship. In addition, the overarching historical context to these factors is that the career path, pressures and expectations of GPs in the UK have changed considerably since the 1990s. Today’s GP is expected to be a member of a wider MDT commissioned to deliver national standards of care and has a role barely recognisable to the one many older GPs remember (e.g. GP partners typically staying in one practice for most of their career, with less regulation and an expectation of autonomy).

Strengths of the review methods

This systematic review was conducted by an experienced and collaborative review team, addressing clear review questions and using a prespecified and published systematic review protocol (CRD42016033876) and search strategy. 47 We worked closely with experienced information specialists to design the most effective possible searches for obtaining bibliographic and web sources. We found substantially more includable studies than a recent literature review that had a similar review question. 17

In terms of involving relevant stakeholders in the review, several GPs were involved in the development of the review protocol, and two GPs were closely involved in the conduct of the review (a GP trainee, and an experienced GP and Professor of Primary Care). Patients were also involved, both as co-investigators in the wider mixed-methods project and through contributing to a PPI workshop, which presented and discussed emerging findings from the qualitative and quantitative evidence syntheses. Reassuringly, the PPI workshop both endorsed and expanded our understanding of our emerging interpretations (see Appendix 10 for a fuller description of that discussion).

In terms of quality assurance, we used either two reviewers making independent inclusion/exclusion decisions or checking by a second reviewer with an independent assessment of a sample of included studies (for data extraction and quality assessment) and used established study quality assessment tools.

Limitations of the review methods

Although this review captured information about specific retention strategies, where they provided the background to a particular survey, or if the strategy was suggested on the basis of an included survey’s findings, it should be noted that identifying and evaluating GP retention strategies was not the primary aim of this systematic review and, therefore, we did not explicitly search for studies that either evaluated or described different approaches to GP retention. Although we used the most appropriate and established quality assessment tools available, the tool we adapted and used for the quantitative survey studies had a few limitations. First, there was no separate score/assessment of study design and reporting quality. Second, our judgements about the applicability of findings to GPs in the UK NHS in 2016 were based on subjectively weighing up information such as the age of the study data, the geographical scope of the study, whether or not it was limited to certain types of GP (e.g. older/younger, practice principals/non-principals) and whether or not there were differences between respondent characteristics and target GP population characteristics.

The study pragmatically focused on either survey-based (mainly quantitative) studies or qualitative interview studies. However, some of the survey studies also collected and reported some qualitative data, in the form of free-text answers to open questions (one survey study was purely a collation of such data56). We did not incorporate or separately content analyse these disparate data.

In December 2016, another qualitative research study that should have been included – unfortunately missed by our grey literature searches – came to our attention. 81 The detailed findings and four groups of ‘deeper frustrations’ identified in this study largely corroborated the insights from our synthesis of qualitative studies, including the finding that workload was the overarching factor causing GPs to leave.

Limitations of the current evidence base

Most studies examined only the association of a single quitting construct (e.g. intention to reduce work hours) with a few variables, one variable at a time. There were only four comprehensive survey studies that used multivariable analyses. The most commonly used quitting construct/question was GP intention to retire/quit direct patient care within 5 years, which was used in eight studies. But even this apparently standard question was asked in a number of slightly different ways, and questions relating to actual or preferred part-time working varied even more. Only three UK studies and no non-UK studies had any focus on why GPs take career breaks, and no studies explicitly examined the benefits and determinants of flexible working arrangements. Few studies explicitly assessed the role of GP health or ageing, or potential mediating factors, such as ‘commitment’ or ‘emotional exhaustion’ (a key component of instruments for assessing ‘burnout’), which may attenuate or accelerate the effects of job satisfaction on intentions to quit.

Most questionnaire surveys of GPs provide a snapshot of factors in a particular year and region. They therefore capture the absolute levels of perceived factors, but do not capture prior levels or prior expectations. What is clear from both the quantitative and qualitative studies is that many of the causes of quitting or deciding to work part-time have a temporal element (i.e. they relate to widening gaps between initial expectations of being a GP and current reality, or the cumulative effect of recurrent organisational changes over many years).

There were only five qualitative studies of GPs (four based on interviews with UK GPs). 17,18,51,76,77 Two of them, by Hutchins76 and Newton,80 pre-date a number of substantial changes in the organisation of general practice and the remuneration of GPs and general practices, which may limit their generalisability to UK general practice in 2016.

The ‘Removals Log’

The list of GPs provided by the National Performers List was also supplemented by a small additional list maintained by NHS England local area teams, containing the names of all GPs who are due to be removed from the List, through notification of their intention to resign, retire, relinquish their licence to practise or cease to practise in their current region (the ‘Removals Log’).

NHS England permitted the research team access to this list of 152 GPs who had been removed (or were shortly to be removed) from the National Performers List between April 2015 and March 2016. Of these 152 GPs, 77 were still on the List (i.e. they had not yet been removed and we already had their contact details), six were not to be contacted [one was deceased, four were mandatory removals from the List, one had been suspended by the General Medical Council (GMC)] and two had moved onto the Welsh Performers List. We requested postal and/or e-mail addresses from NHS England for the remaining 97 GPs, eight of which were not supplied (one was a mandatory removal from the List, two had no contact details available, four had retired owing to ill health and one had retired owing to personal circumstances).

We therefore received additional contact details for a total of 89 of the 152 Removals Log GPs. Sixty-five of the 89 had a postal address, all of which were home addresses, and 51 of these 65 also had an e-mail address. The remaining 24 had an e-mail address only.

Questionnaire

The questionnaire was based on that used within earlier work, modified to increase the number of questions from 11 to 24, including rewording of four questions to reflect alignment of wording with other questionnaires of broadly similar intent, and by providing clear definitions for key concepts in the questionnaire, including a career break, taking steps towards changing work–life balance and defining a clinical session. 17 The questionnaire (see Appendix 11) comprised items that asked GPs about their career intentions, reporting on the likelihood that they would permanently leave direct patient care within the next 2 years or within the next 5 years. 82 GPs were also asked to report the likelihood that they would take a career break within the next 5 years, or that they would reduce their weekly average hours spent in direct patient care during this time period. GPs rated the likelihood of these events from ‘very likely’ to ‘very unlikely’ using a four-point scale. The questionnaire also included a question about current level of morale and captured general demographic data: gender, age, ethnicity, region and year of graduation, current GP employment status (e.g. partner, salaried), number and pattern of sessions worked in a typical week and involvement in delivering out-of-hours care.

Data collection

General practitioners were sent study materials through the post to either their practice or their home address, and also by e-mail, if available. 82 The questionnaire was available for completion by post or online. The survey was supported by a comprehensive strategy of publicising the research through routine newsletters and circulars of relevant organisations and networks, including Local Medical Committees (LMCs), Clinical Research Networks (CRNs), HEE South West, the RCGP, UEMS and the South-West Academic Health Sciences Network (AHSN).

If a GP returned multiple online or postal surveys, only the first response received by the research team was analysed. Postal response data were double entered and discrepancy checking was undertaken. Response data were stored securely and without participant names or addresses.

Patient involvement

Although the study participants were GPs rather than patients, patient representatives contributed to the design of the survey. 82 The planned work was presented to the wider project’s PPI group, by way of sharing the process and to check the integrity of the work, and the group provided supportive feedback. 83 The survey results were presented at a project management group meeting, which included PPI representatives who directly contributed to interpreting and contextualising the results.

Statistical analysis

Differential response rates between different groups of GPs would potentially introduce bias into crude survey findings. To counter this, we employed non-response weights. 82 Inverse probability weights were calculated based on three factors: age (< 40, 40–49, 50–54, 55–59 and ≥ 60 years), gender (male and female) and role (partner, salaried and locum/other). These factors were used as they were the only ones consistently recorded for both responders and non-responders. By employing these weights, it was estimated what responses would have been received with a 100% response rate under the assumption that non-responders would have responded similarly to GPs of the same age, gender and role. Logistic regression was used to investigate the association between responses to questions regarding future career intentions (permanently leaving direct patient care within the next 2 and 5 years, taking a career break within the next 5 years and reducing average hours spent in direct patient care within the next 5 years). Each of the four sets of responses was dichotomised into ‘very likely’ and ‘likely’ versus other responses. Initially, unadjusted associations were examined for effects attributable to the explanatory factors of gender, age, country of qualification, ethnicity, role/position and rating of current morale. Subsequently, regression models adjusting simultaneously for all explanatory factors were used to examine adjusted associations. Similar models were used with reported morale as the outcome (but not including morale as an explanatory factor). Regression analyses were restricted to those respondents with complete data on gender, age, country of qualification, ethnicity, role/position and rating of current morale.

Supplementary analysis

Interactions were explored between various factors in the models. 82 Although some of these were found to be statistically significant, the magnitude of the interaction terms was generally small and did not alter the interpretation of the data, with one exception commented on in the results. For this reason, the more complex interaction models have not been reported here.

In addition, the possibility was considered that some groups of GPs, for example female GPs, may not report the intention to reduce hours spent in direct patient care because they already work fewer hours, on average, than other groups. Therefore, two supplementary analyses were conducted, including either a binary variable indicating part-time working (defined as working fewer than eight sessions per week) or a continuous variable detailing the reported number of sessions worked per week.

All analyses were conducted in Stata^® version 14.2 (StataCorp LP, College Station, TX, USA).

Career intentions

Of the 2248 responding GPs, 55 had already permanently left direct patient care (Table 7); 16 had selected ‘none of the above’ to reflect their working status in direct patient care (i.e. they were not currently working in direct patient care, nor were they on a career break, but they had not permanently quit) and were removed from further analysis. A further 58 GPs reported being currently on a career break. Differential response by age and role meant that GP groups reporting an intention to quit direct patient care were somewhat over-represented, introducing a small bias into the crude results, which is accounted for in the weighted percentages.

Of the 2179 GP participants included in the analysis,82 473 (weighted percentage 20.3%, 95% CI 18.7% to 22.0%) reported a high likelihood of quitting direct patient care (‘likely’ or ‘very likely’) within the next 2 years, and 863 (weighted percentage 36.7%, 95% CI 34.7% to 38.8%) within the next 5 years (see Table 7). There were 1252 participants (weighted percentage 56.7%, 95% CI 54.5% to 58.8%) who reported being likely or very likely to reduce their working hours, and 770 (weighted percentage 36.3%, 95% CI 34.3% to 38.4%) who reported the intention to take a career break within the next 5 years. Considered together, 1535 participants (weighted percentage 69.9%, 95% CI 67.9% to 71.8%) reported that they were likely/very likely to pursue a career intention (one or more of the four presented) that would potentially adversely affect the workforce available in south-west England within the next 5 years. The majority of participants also had low levels of morale, with a substantially greater proportion of participants (1195; weighted percentage 54.4%, 95% CI 52.2% to 56.5%) reporting ‘low’ or ‘very low’ levels of morale than of those reporting ‘high’ or ‘very high’ levels of morale (305 participants; weighted percentage 14.2%, 95% CI 12.7% to 15.7%).

Associations between general practitioner characteristics and career intentions/morale

Figure 6 illustrates the reported intention to quit direct patient care of responding GPs broken down by gender and each year of GP age, aggregating likely and very likely responses together. Reported quit intentions are strongly related to age, remaining low (< 20%) among younger GPs (≤ 45 years) for quitting direct patient care in both of the next 2 and 5 years. Both of these outcomes show a sharp rise from the age of 52 years, with the proportion stating they were likely to quit in 5 years rising to almost 90% by the age of 56 years. The proportion anticipating taking a career break as ‘likely’ was highest for the youngest GPs, especially among younger women; nearly 9 out of 10 female GPs aged 30 years reported the intention to take a career break, presumably on account of anticipated maternity leave.

FIGURE 6.

Career intentions to quit direct patient care by gender and age. (a) Male; and (b) female. Reproduced with permission from Fletcher et al. 82 Quitting patient care and career break intentions among general practitioners in south-west England: findings of a census survey of general practitioners. BMJ Open 2017;7:e015853. https://doi.org/10.1136/bmjopen-2017-015853. This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/.

Reproduced with permission from Fletcher et al. 82 Quitting patient care and career break intentions among general practitioners in south-west England: findings of a census survey of general practitioners. BMJ Open 2017;7:e015853. doi: https://doi.org/10.1136/bmjopen-2017-015853. This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/.

Of the 2179 respondents reporting that their current role involved direct patient care, 2119 (97%) provided complete information on gender, age, region of qualification, ethnicity, role/position and rating of current morale and were included in the regression analyses. Tables 8 and 9 show the results of the unadjusted and adjusted logistic regression analyses, respectively, for career intentions. In unadjusted analyses, there is strong evidence (p < 0.001 for all) that gender and role are associated with being likely to report the intention to leave patient care within both 2 and 5 years, with female GPs being less likely to quit patient care than male GPs [odds ratio (OR) 0.55, 95% CI 0.46 to 0.66, for leaving patient care in 5 years], and with locum GPs being the most likely to quit patient care and salaried GPs least likely. However, these associations are likely to be confounded, for example by the fact that younger doctors were more likely than older doctors to be female. Once adjustment is made for all other factors, the effects of these characteristics on intentions to quit are no longer significant. However, after adjustment for other variables, women were still substantially less likely to report intending to reduce hours spent in direct patient care (OR 0.58, 95% CI 0.47 to 0.71), and locum GPs remained more likely to report intending to take a career break than GPs in other roles.

TABLE 8 - Unadjusted associations between career intentions and GP attributes

ORs of > 1 indicate that GPs in this category are more likely to leave direct patient care/reduce hours/take a career break than those in the reference category.

This group has been excluded from the analysis owing to perfect prediction (i.e. all > 70-year-olds reported being likely to leave direct patient in 5 years and no mixed-ethnicity GPs reported being likely to leave direct patient care in 2 years).

Reproduced with permission from Fletcher et al. 82 Quitting patient care and career break intentions among general practitioners in south-west England: findings of a census survey of general practitioners. BMJ Open 2017;7:e015853. https://doi.org/10.1136/bmjopen-2017-015853. This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/.

Characteristic	How likely is it that you will permanently leave direct patient care within the next 2 years?		How likely is it that you will permanently leave direct patient care within the next 5 years?		How likely is it that you will reduce your weekly average hours spent in direct patient care within the next 5 years?		How likely is it that you will take a career break (or another career break) within the next 5 years?
	ORa (95% CI)	p-value	ORa (95% CI)	p-value	ORa (95% CI)	p-value	ORa (95% CI)	p-value
Gender
Male	Reference	< 0.001	Reference	< 0.001	Reference	< 0.001	Reference	0.046
Female	0.51 (0.41 to 0.63)		0.55 (0.46 to 0.66)		0.48 (0.41 to 0.58)		1.20 (1.00 to 1.44)
Age (years)
< 40	0.75 (0.48 to 1.16)	< 0.001	0.49 (0.34 to 0.70)	< 0.001	1.24 (0.98 to 1.56)	< 0.001	2.34 (1.83 to 2.98)	< 0.001
40–49	Reference		Reference		Reference		Reference
50–54	1.65 (1.12 to 2.43)		4.00 (3.02 to 5.28)		2.04 (1.58 to 2.63)		1.24 (0.95 to 1.63)
55–59	9.32 (6.72 to 12.92)		30.04 (21.04 to 42.90)		5.00 (3.74 to 6.69)		1.59 (1.22 to 2.08)
60–69	22.22 (14.43 to 34.22)		91.60 (41.88 to 200.36)		7.38 (4.58 to 11.88)		1.61 (1.11 to 2.35)
≥ 70	45.86 (12.84 to 163.84)		b		3.86 (1.23 to 12.10)		1.55 (0.56 to 4.32)
Country of qualification
UK/Ireland	Reference	0.450	Reference	0.132	Reference	0.921	Reference	0.470
Europe	0.56 (0.27 to 1.14)		0.60 (0.35 to 1.02)		0.85 (0.52 to 1.39)		0.65 (0.37 to 1.12)
South Asia	0.88 (0.29 to 2.66)		0.43 (0.14 to 1.32)		1.10 (0.45 to 2.70)		0.95 (0.38 to 2.39)
Other	1.05 (0.47 to 2.32)		0.98 (0.51 to 1.90)		0.95 (0.50 to 1.80)		0.88 (0.45 to 1.73)
Ethnic group
White	Reference	0.300	Reference	0.032	Reference	0.581	Reference	0.610
Mixed	b		0.60 (0.26 to 1.36)		0.74 (0.35 to 1.57)		1.57 (0.74 to 3.31)
Asian	0.87 (0.48 to 1.57)		0.43 (0.25 to 0.76)		1.27 (0.78 to 2.05)		1.00 (0.61 to 1.63)
Black	3.59 (0.89 to 14.41)		1.49 (0.37 to 5.99)		2.23 (0.45 to 11.07)		0.60 (0.12 to 2.99)
Other	0.77 (0.22 to 2.69)		0.68 (0.24 to 1.96)		1.36 (0.50 to 3.70)		1.61 (0.62 to 4.18)
Role/position
GP partner	Reference	< 0.001	Reference	< 0.001	Reference	< 0.001	Reference	< 0.001
Salaried GP	0.61 (0.46 to 0.83)		0.43 (0.34 to 0.55)		0.63 (0.51 to 0.78)		1.25 (1.00 to 1.56)
Locum GP	1.58 (1.18 to 2.11)		1.08 (0.82 to 1.41)		1.34 (1.02 to 1.76)		1.72 (1.32 to 2.24)
Other	1.12 (0.61 to 2.08)		0.72 (0.40 to 1.26)		0.79 (0.47 to 1.34)		0.85 (0.48 to 1.52)
Reported morale
Very low	0.99 (0.45 to 2.17)	< 0.001	1.85 (0.81 to 4.19)	< 0.001	2.58 (1.19 to 5.59)	< 0.001	2.24 (1.01 to 4.95)	< 0.001
Low	0.42 (0.19 to 0.91)		0.78 (0.35 to 1.75)		1.31 (0.62 to 2.77)		1.14 (0.52 to 2.49)
Neither low or high	0.30 (0.14 to 0.65)		0.62 (0.28 to 1.40)		0.82 (0.39 to 1.73)		0.81 (0.37 to 1.78)
High	0.40 (0.18 to 0.90)		0.62 (0.27 to 1.42)		0.54 (0.25 to 1.16)		0.55 (0.24 to 1.24)
Very high	Reference		Reference		Reference		Reference

TABLE 9 - Adjusted associations between career intentions and GP attributes. Adjustment was made for all factors shown in this table

ORs of > 1 indicate that GPs in this category are more likely to leave direct patient care/reduce hours/take a career break than those in the reference category.

This group has been excluded from the analysis owing to perfect prediction (i.e. all > 70-year-olds reported being likely to leave direct patient in 5 years and no mixed-ethnicity GPs reported being likely to leave direct patient care in 2 years).

Reproduced with permission from Fletcher et al. 82 Quitting patient care and career break intentions among general practitioners in south-west England: findings of a census survey of general practitioners. BMJ Open 2017;7:e015853. https://doi.org/10.1136/bmjopen-2017-015853. This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/.

Characteristic	How likely is it that you will permanently leave direct patient care within the next 2 years?		How likely is it that you will permanently leave direct patient care within the next 5 years?		How likely is it that you will reduce your weekly average hours spent in direct patient care within the next 5 years?		How likely is it that you will take a career break (or another career break) within the next 5 years?
	ORa (95% CI)	p-value	ORa (95% CI)	p-value	ORa (95% CI)	p-value	ORa (95% CI)	p-value
Gender
Male	Reference	0.181	Reference	0.810	Reference	< 0.001	Reference	0.019
Female	0.83 (0.64 to 1.09)		1.03 (0.80 to 1.33)		0.58 (0.47 to 0.71)		1.27 (1.04 to 1.55)
Age (years)
< 40	0.89 (0.56 to 1.41)	< 0.001	0.58 (0.40 to 0.85)	< 0.001	1.47 (1.14 to 1.90)	< 0.001	2.63 (2.02 to 3.42)	< 0.001
40–49	Reference		Reference		Reference		Reference
50–54	1.70 (1.14 to 2.55)		4.73 (3.49 to 6.41)		2.11 (1.61 to 2.77)		1.31 (0.98 to 1.73)
55–59	11.69 (8.18 to 16.72)		44.15 (29.77 to 65.48)		5.62 (4.12 to 7.66)		1.79 (1.35 to 2.38)
60–69	36.87 (22.53 to 60.34)		194.43 (84.32 to 448.34)		9.58 (5.76 to 15.95)		2.03 (1.35 to 3.04)
≥ 70	110.61 (25.85 to 473.26)		b		4.93 (1.47 to 16.55)		1.83 (0.61 to 5.48)
Country of qualification
UK/Ireland	Reference	0.193	Reference	0.048	Reference	0.829	Reference	0.738
Europe	0.60 (0.27 to 1.37)		0.48 (0.23 to 0.99)		0.98 (0.57 to 1.70)		0.73 (0.41 to 1.29)
South Asia	0.21 (0.04 to 1.15)		0.13 (0.02 to 0.97)		0.61 (0.19 to 1.96)		0.98 (0.31 to 3.05)
Other	0.83 (0.30 to 2.31)		0.94 (0.39 to 2.25)		0.83 (0.40 to 1.71)		1.08 (0.53 to 2.22)
Ethnic group
White	Reference	0.109	Reference	0.142	Reference	0.209	Reference	0.483
Mixed	b		2.84 (1.15 to 7.04)		1.32 (0.60 to 2.92)		1.74 (0.79 to 3.81)
Asian	1.58 (0.70 to 3.54)		0.84 (0.38 to 1.86)		1.89 (1.04 to 3.46)		0.87 (0.48 to 1.57)
Black	6.07 (1.00 to 36.99)		2.26 (0.29 to 17.71)		2.89 (0.48 to 17.28)		0.54 (0.10 to 2.87)
Other	0.45 (0.10 to 2.16)		0.45 (0.08 to 2.43)		1.17 (0.39 to 3.45)		1.57 (0.56 to 4.41)
Role/position
GP partner	Reference	0.177	Reference	0.153	Reference	0.005	Reference	< 0.001
Salaried GP	1.10 (0.76 to 1.59)		0.81 (0.58 to 1.14)		0.97 (0.75 to 1.24)		1.15 (0.89 to 1.47)
Locum GP	1.55 (1.04 to 2.30)		1.37 (0.91 to 2.07)		1.74 (1.26 to 2.39)		2.09 (1.55 to 2.81)
Other	0.91 (0.39 to 2.15)		0.73 (0.29 to 1.86)		1.19 (0.64 to 2.20)		1.15 (0.62 to 2.13)
Reported morale
Very low	5.77 (1.96 to 17.00)	< 0.001	13.07 (3.72 to 45.88)	< 0.001	6.74 (2.76 to 16.47)	< 0.001	3.49 (1.49 to 8.14)	< 0.001
Low	1.87 (0.65 to 5.38)		3.89 (1.13 to 13.34)		3.27 (1.38 to 7.74)		1.50 (0.66 to 3.45)
Neither low nor high	0.86 (0.30 to 2.47)		2.15 (0.63 to 7.39)		1.75 (0.74 to 4.14)		0.93 (0.40 to 2.12)
High	0.80 (0.27 to 2.39)		1.29 (0.36 to 4.60)		0.86 (0.36 to 2.08)		0.59 (0.25 to 1.40)
Very high	Reference		Reference		Reference		Reference

Age was a very strong predictor of reported intention to leave direct patient care, a finding that persisted after adjustment (e.g. 60- to 69-year-olds vs. 40- to 49-year-olds: OR 194.4, 95% CI 84.3 to 448.3 for leaving patient care in 5 years). Age was also a strong predictor of the intention to reduce hours in the adjusted model, with older GPs more likely to report this intention than younger GPs. Female GPs, GPs aged < 40 years and locum GPs were most likely to report the intention to take a career break after adjustment for other factors. A model including an interaction between age and gender showed that the effect of GPs aged < 40 years, being more likely to report intentions to take a career break was strongest in female GPs, consistent with the pattern shown in Figure 6 (results not shown).

Self-reported morale was a strong predictor of all four outcomes shown in Tables 8 and 9. In the unadjusted analysis, a ‘U-shaped’ relationship is seen for all four quitting outcomes, with those with ‘very high’ levels of morale being more likely to report intentions to leave direct patient care/reduce hours/take a career break than those with ‘high’ or ‘neither low nor high’ levels of morale (and in the case of leaving patient care within 2 years, more likely than any other morale category), but less likely to intend to leave/reduce hours/take a career break than those with ‘very low’ levels of morale. This U-shaped relationship largely disappears after adjustment, however, and it is those GPs with ‘very low’ levels of morale who are most likely to report intending to leave direct patient care, reduce hours in patient care or to take a career break. Moreover, the change in odds between ‘low’ and ‘very low’ levels of morale is particularly strong. Table 10 shows the results of the logistic regression analysis modelling of factors associated with low or very low levels of morale. Only age and practice role show any evidence of being associated with reported morale (p < 0.001 for both). Those aged 50–54 years are most likely to report low levels of morale. In respect of role, GP partners are the most likely to report low levels of morale, followed by salaried GPs, and locums and other GPs are the least likely groups to report low levels of morale.

TABLE 10 - Unadjusted and adjusted associations between self-reported low or very low levels of morale and GP attributes

ORs of > 1 indicate that GPs in this category are more likely to report low or very low levels of morale than those in the reference category. Adjustment was made for all factors shown in this table.

Reproduced with permission from Fletcher et al. 82 Quitting patient care and career break intentions among general practitioners in south-west England: findings of a census survey of general practitioners. BMJ Open 2017;7:e015853. https://doi.org/10.1136/bmjopen-2017-015853. This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/.

Characteristic	Association
	Unadjusted		Adjusted
	ORa (95% CI)	p-value	ORa (95% CI)	p-value
Gender
Male	Reference	0.134	Reference	0.231
Female	0.87 (0.72 to 0.72)		0.90 (0.76 to 0.76)
Age (years)
< 40	0.73 (0.57 to 0.57)	< 0.001	0.69 (0.55 to 0.55)	< 0.001
40–49	Reference		Reference
50–54	1.06 (0.81 to 0.81)		1.10 (0.85 to 0.85)
55–59	0.85 (0.66 to 0.66)		0.92 (0.71 to 0.71)
60–69	0.38 (0.26 to 0.26)		0.33 (0.23 to 0.23)
≥ 70	0.07 (0.01 to 0.01)		0.04 (0.01 to 0.01)
Country of qualification
UK/Ireland	Reference	0.157	Reference	0.194
Europe	0.74 (0.45 to 0.45)		0.73 (0.45 to 0.45)
South Asia	0.97 (0.32 to 0.32)		0.97 (0.40 to 0.40)
Other	0.50 (0.25 to 0.25)		0.55 (0.29 to 0.29)
Ethnic group
White	Reference	0.534	Reference	0.721
Mixed	0.82 (0.38 to 0.38)		0.82 (0.39 to 0.39)
Asian	1.59 (0.89 to 0.89)		1.31 (0.81 to 0.81)
Black	1.64 (0.37 to 0.37)		0.82 (0.20 to 0.20)
Other	1.01 (0.36 to 0.36)		0.73 (0.28 to 0.28)
Role/position
GP partner	Reference	< 0.001	Reference	< 0.001
Salaried GP	0.75 (0.59 to 0.59)		0.67 (0.54 to 0.54)
Locum GP	0.40 (0.30 to 0.30)		0.33 (0.25 to 0.25)
Other	0.22 (0.12 to 0.12)		0.20 (0.11 to 0.11)

Supplementary analyses exploring the contribution of current working patterns to any association with intentions to reduce hours spent in direct patient care did explain some of the difference between genders. However, even after adjustment for current working patterns, female GPs were still substantially less likely to report intentions to reduce hours spent in direct patient care than their male counterparts (results not reported).

Main findings

A high proportion of GPs – around two in every five – currently working within direct patient care in south-west England reported an intention to permanently quit direct patient care within the next 5 years, this being one in five within the next 2 years. Further depletion of the GP workforce in this region of England through the reduction of weekly average hours or through taking a career break within the next 5 years also emerged as an impending risk. Overall, 7 out of every 10 GPs in this region reported a career intention that, if implemented, would adversely affect the GP workforce capacity in south-west England through GPs leaving direct patient care, reducing hours spent in direct patient care or by taking a career break within the next 5 years.

Older age is highly predictive of a GP’s intentions to permanently quit direct patient care and to reduce hours. The intention to quit was independent of the GP’s role (partner, salaried or locum) and gender. The gender differences observed among younger GPs in respect of intended career breaks is a particular issue because, currently, 69% of the GP trainee workforce is female,84 because a substantial proportion of younger women returning to clinical care do so on the basis of reduced hours,1 and because doctors who wish to return to clinical practice may face significant bureaucratic obstacles in doing so. 85 The decision to reduce hours or to take a career break varied with employment status, with locum GPs being most likely to report intentions to reduce hours or to take a career break. In a situation in which GPs who are not partners in a practice do in fact leave direct patient care, the system of care appears vulnerable, with added strain likely to be placed on GPs who are partners in their practice.

A key issue is whether or not the intention to leave is translated into reality, and few studies have explored this important question. However, Hann et al. 59 explored this question with nearly 1200 UK GPs who were followed up over 5 years and identified a strong relationship between a stated intention to leave patient care and actually doing so, reflected in a 4.5-fold difference in odds of leaving between those with no stated intention to leave patient care and those reporting a high intention to do so.

The majority of responding GPs in our study reported low levels of morale, with < 15% reporting high levels of morale. GP morale was identified as an important predictor of future career intentions, particularly when morale was very low.

Strengths and limitations

Ensuring the alignment of wording with other surveys of similar intent allows the opportunity to compare and contrast results. Recent high-quality surveys support our estimate of the proportion of GPs intending to leave general practice within the next 5 years. 17,41,43,44 Unlike most other similar surveys, which tend to survey samples of GPs,19,86 we undertook a census survey of all GPs currently eligible to provide patient care across a large area, providing a cross-sectional overview of the quit intentions of all GPs in south-west England. The response rate was high (67%), reflecting both rigorous planning and implementation of the survey and interest among GPs in the subject matter of workforce challenges. Furthermore, we have reduced the potential for non-response bias by weighting our estimates to account for differential response rates by age, gender and practice role. The overall programme of work, of which this survey is one element, includes the planned development of a predictive risk modelling tool to enumerate risk of future workforce undersupply at local levels, thus offering the potential of providing data that will be of value to health-care planners and policy-makers. In addition, our study relates to a key current area of national interest and concern.

The study has a number of limitations. The survey was conducted among GPs in a single region of England, which may not be entirely typical of England as a whole. For example, a higher proportion of GPs in the region are trained in the UK and a lower proportion are partners (see Appendix 13, Table 37). Furthermore, differences may exist in respect of population mix and practice setting. 87 The region is often informally regarded as having desirable living and working conditions. 88 It seems likely that other regions of England, with more inner-city areas with higher levels of deprivation, are likely to face even greater challenges in respect of the GP workforce than in the findings we have presented here. A further limitation is that the research was undertaken within a rapidly changing GP workforce policy environment, for example the implementation in 2015–16 of a 10-point plan agreed between authoritative UK health, governmental and professional bodies concerned about GP workforce issues. 89 Finally, this survey was cross-sectional, rather than longitudinal in design, and so we are unable to report on the actual translation of GPs’ reported career intentions into actual career decisions, although previous work has demonstrated intention to quit to be a strong predictor of actually leaving. 59

Implications for research

Our survey is part of a multimethod suite of work commissioned by the NHS, seeking to inform evidence-based policies and strategies in a key area of national concern, which will involve stakeholder consultations with policy-makers who have oversight of GP workforce planning. 83 Future research should include an assessment of the predictive validity of GPs’ reported career intentions by longitudinal follow-up of GPs, an investigation of how reducing hours or better flexible working arrangements may influence GP retention, and research exploring the potential for extended clinical roles among important allied health-care professionals in the clinical workforce. Such groups might include nurses, physiotherapists and pharmacists, and developing such extended roles for these professionals within the primary care team may offer a rapid and potentially effective contribution to the alleviation of current GP workforce pressures. Research into the determinants and impact of working reduced hours should also encompass other aspects of flexible working arrangements in general practice, and how such arrangements may improve job satisfaction and work–life balance and thereby potentially increase GP retention in the longer term. This work also provides insight into those groups of GPs that might best be targeted with interventions aiming to improve workforce retention.

Implications for policy and practice

Concerns regarding the GP workforce are now recognised widely by authoritative bodies in the UK, leading to recent policy statements and initiatives89 and to the establishment of a Primary Care Workforce Commission by HEE8. Our research can inform this policy agenda by providing an estimate of the likely proportion of south-west GPs intending to leave direct patient care within the next 5 years through permanently quitting patient care or by taking a career break. Our findings are likely to be of relevance in other areas of England that may face even greater challenges in respect of the GP workforce. Retaining the direct patient care capacity of established GPs represents an immediate challenge for health-care planners. Recognising and addressing issues of low GP morale is an area of importance, and addressing the issue may not only improve how GPs feel about their work but also potentially be associated with improvements in patients’ experience of the care received. 90 In the light of problems of recruiting to general practice91 and the wider workforce problems, the UK government recently announced an increase in the number of training places for GPs89 and an increase in the number of medical training places in the UK. 92 However, these medium- to long-term solutions do not address the immediate workforce crisis. In contrast, policy initiatives to support the retention of the existing GP workforce, the rapid development of and support for new models of care, and the implementation of policies aimed at alleviating workload pressures for GPs are potential short- to medium-term solutions that need to be urgently considered. Developing and implementing relevant policies and strategies to achieve those ambitions will be important for the alleviation of the immediate pressures faced by the GP workforce and the populations of patients they serve.

Conclusions

Health care in England faces imminent challenges in respect of GP workforce capacity. This survey identifies the magnitude of the problem in south-west England and highlights the important role of GP morale as one important factor contributing to that challenge. Acknowledgement of the magnitude of the problems is urgently required, along with implementation and monitoring of relevant policy and strategy. Failure to do so will risk serious adverse effects on the capacity and ability of the NHS to provide effective primary care to the UK population.

Aim

The aim of the qualitative workstream was to identify the potential content of policies and strategies aimed at retaining experienced GPs and/or supporting the return of GPs following a career break. To do this, an exploration of the lived experience of GPs was required, including the investigation of remaining/leaving decision-making processes among GPs.

Approach

We incorporated a thematic analysis of CQC practice report data to explore examples of potentially good and poor practice in south-west England in respect of the quality of care delivered, followed by semistructured interviews with GPs and key stakeholders. The CQC strand aimed to inform the content of interview questions and to provide contextual information to inform the RAM expert panel workstream. Interviews were undertaken to explore the lived experiences and perspectives of GPs and the impact of GP quitting decisions on practice management and organisation, and to further inform the content of policies and strategies.

Care Quality Commission inspection reports

The aim was to use CQC report data to produce a briefing document that could:

• inform the content of interviews with GPs

• provide contextual information and case examples to inform workstream 4.

Full details of the methodology are provided in Appendix 14. There were five key stages to working with the CQC report data:

1. identification of general practices in south-west England that had been rated by the CQC, and their overall rating category

2. familiarisation with the approach, content and report layout

3. use of an iterative approach to determine the best method for extracting information and examples from the reports

4. identification of a maximum variation sample consisting of CQC reports of general practices in south-west England

5. extraction and summarising of report data.

Identifying the number of general practices in south-west England that had been reported on by the Care Quality Commission

Care Quality Commission report data are publicly accessible (see www.cqc.org.uk/content/how-get-and-re-use-cqc-information-and-data#directory; accessed 30 November 2017).

Sampling took place in January 2016 using the CQC data set that had been completed up to 1 December 2015. There were 442 active location practices in the ReGROUP catchment area, of which 227 (51.4%) had been inspected, rated and had their inspection results made publicly available (Table 11).

Sampling Care Quality Commission reports for analysis

A maximum variation sample was taken based on practice list size and urban or non-urban locality. The sample included 8 out of 16 (50%) ‘outstanding’ reports and 6 out of 8 (75%) ‘requires improvement’ and ‘inadequate’ reports (see Appendix 14, Tables 38 and 39, for full details). Practices rated ‘good’ were not included as it was agreed by the workstream 3 team that a sufficient number and range of examples could be drawn from those reports at the extremes of good and poor practice. Pragmatic considerations guided the number of sampled reports (Table 12).

Interviews

Key findings from Care Quality Commission inspection reports

See Appendix 14 for detailed findings.

Recruitment and interviews

Key findings from interviews

To contextualise the findings, interview questions included an exploration and reiteration of the problems faced by GPs. It was clear that these were not ‘standalone’ issues but complex interactions of GP experiences. The problems/issues were divided into five main topics:

1. workload

2. GP health and well-being

3. support and relationships

4. finances, investment and fiscal reward

5. change, uncertainty and the future.

Three underlying themes were identified concerning the GPs’ experiences and perceptions of their working lives and environment:

1. identity and value

2. fear and risk

3. choice and volition.

General practitioners described a range of ‘push’ and ‘pull’ factors and how these influenced their own staying/quitting intentions and behaviours. Comments were also provided by stakeholders regarding GP experiences.

The following sections provide further details of the topics and themes related to the issues or problems identified by the participants and quitting decisions. These are followed by participants’ suggested ‘solutions’ to these issues, and the inherent tensions and contradictions that they identified within these.

Underlying themes that affect decisions to remain in or leave direct patient care

The analysis also identified three underlying themes that gave more in-depth understanding of the lived experiences of the GPs and helped to contextualise decision-making about remaining in or leaving direct patient care. The underlying themes were:

1. identity and value

2. fear and risk

3. choice and volition.

Figure 7 illustrates the multidirectional relationships between the problems and solutions, the underlying themes, and also the influence of tensions and contradictions (see Tensions and contradictions). The underlying themes are presented in detail with illustrative quotations (with supplementary quotations in Appendix 21).

FIGURE 7.

Illustrative model of the complexity and multilayered relationships between problems, solutions, GP perspectives and experiences.

Strengths and limitations

Patient and public involvement group comments

A meeting with five members of the PPI group was held (May 2017) to discuss the findings of the qualitative workstream and to gain their perspectives on the underlying themes.

The group expressed general sympathy towards, and understanding of, the pressures that GPs can experience, and noted the potential negative impact on patients of GPs being under pressure. The PPI members noted that there is an opportunity, and desire, for patients to be more involved in supporting GPs and the organisation and delivery of primary care. There was agreement that more involvement and inclusion of patient participation groups (PPGs) could benefit GPs: positive interactions with patient representatives could help to reduce GP anxiety (e.g. about complaints). However, it was also noted that, for PPGs to be of value to GPs, there was a need for practice staff and patient representatives to be perceived as ‘all being on the same side’, and for GPs and non-clinical staff to trust patients as part of the practice team. The PPI members felt that there was a role for PPGs to be supportive of GPs, namely helping GPs to feel more valued. They identified a ‘positive feedback loop of fear’: when doctors change their working practices, patients change their habits in response, but without clear communication this leads to anxiety, fear and irritation in both groups. The PPI members noted that more involvement and inclusion of patients, along with good communication, could aid the relationship between patients and practice staff and, thus, help to address patient demands and expectations. The PPI members suggested that involving PPGs as part of the practice team could help the identification of models and examples of good practice that could then be shared by PPGs with other practices.

The discussion also highlighted some of the limitations or disadvantages of PPGs: they can be used as platforms for disgruntled patients, they may not be very representative of the local population and they can feel unwanted by the practice staff (clinical and/or non-clinical). Thus, the PPI group identified an opportunity for proactive recruitment of PPG members and promotion of a culture of mutual support and trust.

How workstream 3 adds to ReGROUP

The findings from the qualitative workstream informed the qualitative modelling and verification of findings from the evidence synthesis of qualitative studies, the background contextual information and the development of statements and subgroups used for the RAM workstream. They also helped to inform the content and direction of the stakeholder consultations.

Recommendations directly arising from workstream 3 are provided in Appendix 25.

Conclusion

The breadth of issues influencing GPs’ decisions to leave direct patient care has been extensively reported. This current research found that the same issues are still identified by GPs, at a time of rapid introductions of policy and strategy aimed at ameliorating the workforce crisis. This study indicated that, in order to address the issues, each issue should not be taken in isolation. Rather, the findings from this study showed that (1) factors and issues need to be addressed collectively, (2) there are inherent tensions and contradictions within potential solutions that need to be considered and (3) there is also a need to address GPs’ lived experiences of their work and role in the current health-care climate. The lived experience was illustrated through three underlying themes that emerged from the data: (1) identity and value, (2) fear and risk and (3) choice and volition.

The application of theories and models may help to further an understanding of the implications of GPs’ feelings and experiences (the underlying themes) on GP retention. Use of evidence from occupational and workplace literature adds weight to the argument that solutions should not be ‘sticking plasters’ but rather implementable policies and strategies that will help to (1) increase the perceived value and clarify the identity and future of general practice, (2) reduce the levels of fear and risk that GPs are experiencing (to acceptable and manageable levels) and (3) provide GPs with feasible, acceptable and sustainable routes to remaining in direct patient care.

The findings from this workstream add to the ReGROUP study, providing further insight into ways to help retain the experienced GP workforce and giving a ‘voice’ to GPs at this pivotal time of change and initiative.

Study design

A RAM panel was used to assess the appropriateness and feasibility of candidate policies and strategies and involved two rounds of data collection via an online survey. RAM panels typically consist of between 7 and 15 members, selected on the basis of their expertise in the topic under consideration. 111

We aimed to recruit 15 panel members to allow for attrition and to secure responses from at least seven respondents completing both rounds of data collection.

Eligibility criteria

The composition of our panel was limited to GP partners and GPs working in national stakeholder organisations with a role in workforce planning, as they represent an influential group in primary care who are directly responsible for managing GP recruitment/retention. These participants are also the intended recipients of the policy and strategies, and are employers of other GPs and primary care teams with responsibility for direct patient care within their general practices. By maintaining homogeneity in the composition of panel membership, there is a higher likelihood that the panel would reach consensus and the findings would be of wider applicability to GPs with managerial and operational responsibilities in primary care.

As challenges to GP recruitment and retention might plausibly vary by geographical setting,117 we planned to recruit GP participants from general practices located in south-west England (urban and rural areas) and London (high population density urban areas), as well as GPs working in relevant national stakeholder organisations including the RCGP and the BMA.

Sampling procedures

Potential participants were identified through different mechanisms for the south-west of England, for London and for the national stakeholder organisations.

In the south-west of England, the National Performers List of all GPs registered to practise in south-west England was used (see Chapter 3). Participants in the qualitative research were excluded to avoid individuals being invited to give substantial time in the same research project. We then purposively sampled 34 GP partners from the south-west, ensuring that an equal number of GPs were identified from urban and rural settings.

As the National Performers List was not available for London, we constructed a sampling frame from publicly available information. The process included collating information on all CCGs in London and then restricting the sample to 12 CCGs with the highest population densities. A list of all practices, including the number of GP partners at each practice, was then collated using information published by 12 CCGs. The sample of 16 practices on the list was then selected at random and the names of individual GPs were extracted from the websites for the selected general practices. In each practice, one of the partners was picked by the researcher to be contacted.

The south-west of England and London samples were then combined into one list (50 GPs) and 25 individuals were selected at random (16 from the south-west and 9 from London) and invited to participate.

Representatives from national stakeholder organisations were identified using expert knowledge and snowballing techniques. We identified key organisations and individuals involved in recruitment and retention issues for GPs such as the Deaneries, BMA, RCGP and the Nuffield Institute. A total of seven individuals were identified and approached to take part, with individuals asked to nominate another person from their organisation to take part if they were unable to do so themselves.

Participant recruitment

Potential participants were sent a recruitment pack consisting of a covering letter from the research team and a participant information sheet. Invitations were sent electronically when possible and by post for all other potential participants. A reminder invitation pack was sent within 2 weeks. Interested participants were provided with a link to an online form and asked to complete it to indicate their willingness to participate. Following agreement to participate, members were provided with further information via e-mail and informed of the dates of the two rounds of data collection (see Appendix 26).

Data collection

Data analysis

In line with the RAM approach, a descriptive analysis of data emanating from round 1 for all panel members was undertaken. 111 For each statement, the panel median score was calculated and categorised into a median band classification with the following classifications applied: 1–3 (inappropriate or unfeasible), 4–6 (uncertain) and 7–9 (appropriate or feasible). As an even number of panel members could yield panel median ratings in decimal places, this was taken into account in the analysis by adjusting the median band classifications. As our statements were about informing policy directions as opposed to clinical decisions, the ‘relaxed’ and simpler approach was used (see Fitch et al. 111). We applied classification boundaries that would make a statement less likely to be included in the uncertain/unequivocal classification (appropriate as 1.0–3.5; inappropriate as 6.5–9.0).

For interpretation, the final outcomes were then determined based on the level of agreement between panel members. Here, consensus was defined as no more than two panel members rating the statement outside the classification in which the group median score was located. Thus, to be considered ‘inappropriate’/’unfeasible’ or ‘appropriate’/’feasible’ the group median score was required to fall within the defined range. Irrespective of the group median score, if the panel did not reach consensus then the statement was deemed to be of uncertain or equivocal value.

Although an interim analysis of appropriateness was undertaken after the first round, this was used only to identify a subset of statements rated (with consensus) as ‘appropriate’ in order to inform the feasibility ratings in round 2. The primary analysis is based on the analysis of round 2 panel ratings. As the statements were repeated for different subgroups, in some cases there were different ratings for a given statement based on the subgroup (e.g. a rating of appropriateness or feasibility when applied to GPs not nearing retirement age and uncertain when applied to GPs nearing retirement age). In these cases, the policy and strategy relating to the statement was included in the category for appropriateness (biased towards showing that consensus was reached), and the differences in ratings for the subgroups were presented in the findings. This resulted in policies and strategies being included in the results as having reached consensus even if this did not take place for all the subgroups included in the statements. These are referred to in Results as differential outcomes.

Analysis of policies and strategies for appropriateness and feasibility

Statement ratings in rounds 1 and 2

In round 1, 30 statements were rated as appropriate, three as inappropriate and 67 as uncertain (Figure 8). Included within the ‘uncertain’ statements were 25 for which the median panel score was within the ‘potentially appropriate’ band and eight in the ‘potentially inappropriate’ band, but with lack of consensus; therefore, they remained as ‘uncertain’ in the round 2 results. As panel members were requested to rate feasibility for the statements that have been rated as appropriate (with agreement) in round 1, 30 statements were identified for feasibility rating in the second round.

FIGURE 8.

Summary of rounds 1 and 2 ratings for appropriateness and feasibility of the 100 statements. a, The 30 statements rated as ‘appropriate’ and with consensus were the same 30 statements put forward for feasibility rating in round 2. Light blue shading indicates the numbers of statements for which the panel reached consensus (whether ‘appropriate’/‘inappropriate’ or ‘feasible’/‘not feasible’). Light green shading indicates the numbers of statements for which consensus was not reached regarding appropriateness/feasibility, or for which consensus was reached that statements were ‘uncertain’ regarding appropriateness/feasibility. Dark blue shading indicates the numbers of statements for which the panel reached consensus that a statement was either ‘inappropriate’ or ‘not feasible’.

From rounds 1 to 2 (when panel members were invited to review and change their previous ratings if they wished), consensus was reached for an additional 11 (out of 67) statements that had previously been classified as ‘uncertain’ in the first round. All these statements had a median panel score within the ‘potentially appropriate’ median score band in round 1 and reached consensus in round 2, resulting in all of these statements being judged as ‘appropriate’. The average change in the median appropriateness score was small (increase of 0.13 points, minimum of –1 point, maximum of 1.5 points). The classifications changed for only 3 out of 100 statements, with two statements reclassified from ‘uncertain’ to ‘potentially appropriate’ and one statement from ‘potentially appropriate’ to ‘uncertain’ in round 2. Owing to a lack of consensus for all three statements in both rounds, the ratings for these remained ‘uncertain’ in the final analysis.

Each policy or strategy may have multiple ratings if statements were generated for different subgroups (e.g. GP ages in relation to retirement, or optional vs. compulsory implementation). The results in the next section are presented at the level of policies and strategies, rather than individual statements.

Main findings

This study rated appropriateness and feasibility for a set of potential policies and strategies that might be expected to facilitate the retention of GPs in providing direct patient care through prevention of quitting or through substantially reducing their commitment. The final set of these potential policies and strategies was selected following review of the research literature and current and planned government policies, and consensus from the ratings of panel members. Overall, most of the policies and strategies rated as ‘appropriate’ were also rated as ‘feasible’. Many of these policies and strategies related to the provision of personal or professional support, such as ‘health and well-being programmes’ to support GPs who were returning to work, which aim to manage their re-entry into the workforce; flexible working; and targeting career stages in the first 5 years or when nearing retirement. When considered at a national level, there was recognition that early self-reporting of practice at-risk status in respect of retention issues was appropriate and feasible. Such interventions may offer potential for enhanced systematic identification of vulnerable practices, drawing on information from routinely available data sets and local intelligence and may facilitate recognition of practice progression to ‘vulnerable’ or ‘at-risk’ status.

Taking into consideration the possibility that policies and strategies might be implemented on a ‘compulsory’ or ‘optional’ basis, panellists were more likely to rate those involving optional implementation as appropriate. This was particularly noted for the compulsory introduction of portfolio working as part of the GP career pathway: implementation was determined to be both appropriate and feasible when introduced as an optional part of the career pathway (as is the case in the current system) but inappropriate when presented as a ‘compulsory’ measure.

Some of the policies and strategies considered to be both appropriate and feasible and already presented above relate to HR management, which is currently managed by each general practice or by a federation. However, when the panellists were presented with policies and strategies relating to the introduction of external HR management to support the implementation of areas such as flexible working and professional development (policies and strategies 42 to 46 in Appendix 27), the panel median scores were in the ‘inappropriate’ band but without consensus, and therefore, following our methodology, these were rated as ‘uncertain’. This may be indicative of a reluctance for panellists (many of whom were recruited in their capacity as GP partners) to support a loss of direct control over the implementation of policies and strategies or management of staff.

When the policies and strategies related to operational functions and day-to-day management of the general practice, they could be considered as protective measures to reduce work-related stressors. These include incentivising and supporting the set-up and operationalisation of new approaches in under-doctored areas, managing high or intense workload by limiting the number of consultations that GPs carry out in a given day or in increasing consultation lengths (all identified as appropriate). Others related to ways in which contracts may be used, for example in terms of reducing the financial risk of general practice ownership, establishing contracts based on programmed activities across several general practices to increase flexibility or in using a national campaign to manage patients’ expectations of general practice. However, although panellists deemed these policies and strategies to be appropriate, there was uncertainty for some of these in terms of feasibility. This is not a surprising finding given that previous research and policy literature have highlighted these issues as affecting patient experience123 and there is still little in terms of tangible progress in addressing these concerns. Nevertheless, the importance of these protective measures to retain experienced GPs in direct patient care has been highlighted throughout this research study and the issue of feasibility requires further exploration.

Research in context

In considering relevant research, it is important to consider some of the limited viewpoints relating to health-care workforce recruitment and retention literature to date. For example, Barriball et al. 121 reviewed the primary research and policy practices for all health-care workforce recruitment and retention in Europe, focusing on educational interventions, financial incentives, professional and personal support, and regulation interventions, as would be expected using the usual recruitment and retention categorisations. Their research and policy is aligned with the international guidelines for improving access to health workers in remote and rural areas,124 often used in health workforce research125,126 and also influential in the UK at a national level. 127 However, this study addresses the recognised gap between consideration and incorporation of these wider contexts in the day-to-day running of general practices and workloads in informing the development of policies and strategies targeting retention among the experienced GP workforce in UK primary care. 120,122 These contexts are particularly important in primary care where GPs in their role as GP partners are responsible for managing workforce issues, as well as for having an interest in and at least shared responsibility for the contractual implications associated with service delivery as part of owning general practices.

There are indications in the General Practice Workforce Action Plan (focusing on England) that a system-wide approach is required. 99 However, the nature and magnitude of any interdependencies of the key areas of workload, practice infrastructure and care (re)design to GP workforce and retention are not yet defined or quantified. At the policy level, the focus is on levers such as skill mix and technology to reduce the workload pressures on GPs. 128 These could result in GPs spending more time on direct patient care and potentially focusing their care on a subset of patients with complex conditions. Implementing these operational-level changes without protective measures on managing the cognitive workload associated with patient consultations may result in increased stressors129,130 and may contribute to ‘push’ factors, with a resultant negative effect on retention. The polices and strategies related to operational functions and the day-to-day management of general practices rated as appropriate (but not necessarily feasible) in this study take into account some of these protective measures and can be used to widen the scope of the current policies when considering the reduction of work pressures on GPs to support retention in direct patient care.

Strengths and limitations

This study used the RAM, widely adopted as a formal group consensus method,131 to undertake a systematic consensus exercise as part of health workforce policy development. Applied for the first time in health workforce policy research, the RAM has been shown to represent a viable approach within this context. The opportunities afforded by this methodology include the ability to present a large number of statements with multiple layers (in this case subgroups) to establish a more refined set of policies and strategies for consideration at national and local levels. There are nuances with the use of this methodology, for example in terms of the tight criteria for reaching consensus, which leads to classifications of uncertainty when a different RAM panel or a larger group may tip the balance into a consensus. In addition, the inability to tease out why some policies and strategies were appropriate or feasible over and above others with subtle differences creates challenges for interpreting the findings. It has already been argued in earlier workstreams that retaining GPs in direct patient care requires a multilayered and multifactored approach. This study builds on these findings and provides a set of policies and strategies with consensus that they are appropriate (and mostly feasible) for implementation as part of interventions to retain GPs in direct patient care.

The main limitations include the selection of just one professional group for the panel, the paucity of evidence regarding the effectiveness of policies and strategies aimed at retaining GPs within the workforce, the exclusion of skill mix/multidisciplinary working policies and strategies in the statements, and the use of appropriateness and feasibility as the rating scales. Each of these is discussed in turn.

It might be argued that, because the panellists were GP partners, the ratings could be biased given their dual role as both beneficiaries of the policies and strategies and as employers who may have to manage the potential burden of implementation as employers. A panel specifically chosen because of their in-depth knowledge about GPs who have left the profession or have transferred from or avoided GP partnership to provide care as locums, salaried doctors or who are just working in the out-of-hours teams may have yielded different results. By targeting GP partners, we had an opportunity to understand and meet the needs of a group of health-care professionals who are influential in facilitating or undermining the implementation of any such policies or strategies. Future studies might consider using panels of individuals from a wider variety of backgrounds working in primary care, including commissioners/workforce planners or other GPs (such as locums or associates, or GPs employed under ‘retainer’ arrangements). To balance the size of the panel (recommended as between 7 and 15 panel members for RAM111), multiple panels may be required to accommodate heterogeneity without distilling the representation from any given group to an individual member.

Similarly, panels drawn from different geographical areas and/or GPs drawn from different types of practice may also be beneficial in relation to the transferability of our findings to more varied general practice configurations.

This panel method required GPs to rate the appropriateness of policies based on the best available evidence combined with their professional judgement. As this study developed, it became apparent that there was a paucity of evidence across a number of policy areas regarding the effectiveness of strategies aimed at maximising GP retention, particularly for those targeting management and HR functions. GPs were provided with the best available evidence, but it is acknowledged that, in some areas, the panellists may have had to rely mostly on their professional judgement. Although less than ideal, as it is assumed that panellists will weight their judgements based on the strength of evidence available, we believe this approach to be defensible. Indeed, it could be argued that many of the emergent policies and strategies currently being introduced in the UK to aid GP retention are based on limited evidence.

Another limitation related to the decision taken by the research team to exclude policies and strategies regarding multidisciplinary working and role substitution and skill mix from the statements presented to the panel. This decision was partly because of the ongoing implementation of new roles (e.g. introduction of clinical pharmacists), but was also influenced by a desire to reduce participant burden by constraining the number of statements presented. Even though there is recognition that skill mix is likely to be of importance in addressing recruitment and retention issues from the perspective of increasing supply to meet demand,128 detailed research that takes into account the impact of such policies and strategies on job satisfaction and GP retention is required. Thus, the detail was judged to be beyond the scope of this immediate piece of work.

Finally, we acknowledge that the use of ‘appropriateness’ and ‘feasibility’ as the rating scales could be called into question compared with ratings on ‘acceptability’. In developing this workstream, we recognised that multiple alternatives may be used. To reduce the potential variation in the interpretation of the rating scales, we provided panellists with clear definitions on how to apply the ratings of appropriateness and feasibility. Such ratings might also be limited by the complex context in which interventions are likely to be implemented; in Chapter 2, it was shown that GP decision-making to stay in direct patient care is influenced by a range of competing ‘push’ and ‘pull’ factors. Similarly, feasibility may be viewed differently based on the panellist’s experience of working in or managing particular types of practices, although this should be ameliorated by the requirement for panels to reach consensus for statements to be deemed appropriate or feasible.

Conclusion

Understanding the views of the GP partners and involving them as those who work on implementation in their role as employers is rarely acknowledged in UK policy research and development. In the absence of a wealth of contextually relevant effectiveness research to directly inform policy decisions on retaining GPs in direct patient care in the UK, the systematic and panel approach provides important focus on 16 potentially appropriate and feasible policies and strategies for implementation at the national/regional level, general practice level and GP level. These 16 areas of policy and strategy formed the basis of the material that was presented for consideration by key stakeholder organisations within the stakeholder consultations workstream.

Conceptual model of supply–demand imbalance

To identify practices at risk of supply–demand imbalance, we first sought to define the concept (Figure 9). The model incorporates two domains and provides a potential framework for evaluating risk of imbalance. The first domain is the ratio of workload (demand) to GP workforce availability (supply); this ratio is shown on the x-axis of Figure 9. Neither workload nor workforce should be considered in isolation, as a large practice may have a large workload but may also a have large workforce. For this reason, we considered the ratio of the two quantities, which can be considered as standardised across different practice sizes. It was also judged that the consideration of this ratio in isolation might produce erroneous inferences. For example, practices with a high workload may be meeting patient demand through innovative and efficient service delivery systems. This discussion was reflected through the incorporation of a marker of a practice’s ability to meet patient demand (y-axis). It was considered that the ability of a practice to meet patient demand under conditions of high workload would not necessarily reflect supply–demand imbalance. On the contrary, such a scenario may be judged by some to imply a desirable efficiency in service delivery (although this may not necessarily reflect desirable working conditions from the GP’s perspective).

FIGURE 9.

Conceptual model of supply–demand imbalance. The gradient in colour indicates efficiency in service delivery.

Considering GP workload-to-workforce ratio and the ability to meet patient demand at the practice level, Figure 9 shows the potential to identify several ‘types’ of practice reflected in the nine internal cells. The following descriptions should be seen only within the context of this model and we recognise that other factors exist that cover the wider concept of service delivery and quality.

Practices with a high workload-to-workforce ratio and good ability to meet patient demand (see Figure 9, bottom right) might be considered as delivering the most efficient practice.

When the ability to meet patient demand is good but the workload-to-workforce ratio is low (see Figure 9, bottom left), this may be considered to reflect a situation whereby supply and demand is imbalanced but not on account of a shortage of workforce. Although identification of these practice types may highlight practices that may benefit from operations improvement focused on increasing efficiency, these do not represent a direct detriment to patient care and are therefore considered to be outside the scope of this work.

When workload-to-workforce ratio is moderate and the ability to meet patient demand is identified as ‘medium’ (see Figure 9, centre cell), we considered this to reflect a ‘stable’ or ‘sustainable’ practice. We note, however, that this classification assumes that the overall situation is itself sustainable, which may not be the case in times of national GP workforce difficulties.

When the ability to meet patient demand is ‘poor’ and the workload-to-workforce ratio is ‘low’ (see Figure 9, top left), this may be considered to reflect less efficient practice rather than to reflect an imbalance between supply and demand.

Within this framework, it is those practices that have a poor ability to meet patient demand in association with a high workload-to-workforce ratio (see Figure 9, top right) that may be considered to represent an adverse imbalance of supply and demand (‘undersupply’). Within our model there are two routes for a practice to move away from the ‘undersupply’ category: either by increasing the workforce while maintaining efficiency (moving to the centre cell) or by increasing the efficiency within the existing workforce (moving towards the bottom right). We assumed that continuing to evidence a poor ability to meet patient demand was judged not to be a viable option.

We focused only on the GP workforce and neither nurses nor other allied health-care professionals were included in our consideration. This was for two main reasons. First, on a conceptual level, the primary focus of the ReGROUP study was the GP workforce. Second, a practical reason relating to operationalising the model meant that some estimate of nurse and GP equivalence (i.e. each nurse is ‘worth’ X% of a GP) would be necessary. Such a metric would be difficult to justify on the basis of published literature, and for this reason we did not attempt to do so. Rather, our approach was based on the idea that the use of nurses and other health-care professionals may be considered a way to achieve practice-level efficiencies and permit effective use of the GP workforce consistent with service delivery system design. 132 Practice-level efficiencies are likely to have an impact on the ability to meet patient demand even in the presence of a high workload-to-workforce ratio.

Operationalisation of the conceptual model

To operationalise the conceptual model summarised in Figure 9, we needed to identify measurable quantities relating to the two domains (when the workload-to-workforce ratio domain comprises two measurable quantities). Estimates of workload were based on the practice registered population on the basis that this represents a proxy for demand placed on the practices. To account for the fact that the needs of patients vary by age, with older patients and children having the highest expressed need, we incorporated weights based on the age and gender composition of the practice population. 2,133 A further adjustment was incorporated to account for additional needs associated with deprived populations. 134,135 Estimates of workforce were based on the FTE GP workforce of the practice. Finally, the ability to meet patient demand was quantified using a measure of patient access, based on the widely adopted national GP Patient Survey (GPPS) in the belief that access is an important measure, reflecting the ease with which patients might engage with the primary health-care system. 136

Overview

The investigation of supply–demand imbalance at a general practice level requires the synthesis of data from a number of sources, and the explicit consideration of a number of factors (Figure 10). A detailed description of how we operationalised these factors is provided below. A predictive risk model (to predict the risk of a practice being in a state of ‘undersupply’ within 5 years) was developed by assessing the associations between current (2016) ‘undersupply’ status and data (when available) from 2012 on the factors included in the model. The rationale for this approach was to obtain factor weightings informed by evidence developed on past data. This model was then used to identify practices and areas in south-west England that are likely to experience a supply–demand imbalance (‘undersupply’) in the future.

FIGURE 10.

Factors used in the construction of the risk prediction model. ONS, Office for National Statistics.

Data sources and preparation

Except where specified, national data for England were obtained and processed. For a summary of data sources and data flow used in the modelling process, see Figure 14.

Mapping the current situation

Figure 11 summarises the distribution of practices in England and in south-west England across the nine categories defined within our conceptual model on the basis of their 2016 data. Practices with access scores in the top third nationally were over-represented in south-west England, with 57% of practices in this region falling in that category. There was also an under-representation of south-west England among practices with the highest workload (only 22% of practices in the region were classified as in the top third of practices nationally). As a result, the percentage of practices defined as being in undersupply was considerably lower in south-west England (5.1%) than in England as a whole (13.5%). We also note that the percentage of practices with good patient access and a low workload per GP was much higher in south-west England (24.5%) than in England overall (13.2%).

FIGURE 11.

Distribution of practices in England and in south-west England across categories defined by our conceptual model of supply–demand imbalance. SW, south-west.

Table 20 provides a description of practices in south-west England, comparing those defined as currently in undersupply and those not currently in undersupply in accordance with our conceptual model. There was no evidence that either list size or list size adjusted for deprivation and weighted for population varied between the two groups. However, there was evidence that practices in undersupply had fewer FTE GPs, indicating that the differences in workload are driven more by the supply of GP workforce than the demand of the registered patient population. Practices in undersupply also had lower patient experience scores, not only for access (which forms part of our definition of undersupply) but also for continuity of care and overall satisfaction. Practices currently in undersupply also tended to have a higher ratio of nurse FTE to GP FTE, had fewer patients over the age of 65 years, served more deprived populations and were more likely to be in urban areas. There was no evidence (p = 0.34) that the distribution of practices currently in undersupply varied by CCG.

Predictive risk model

The predictive risk model was developed using 2012 data from 6398 practices across England (where possible) to predict current (2016) undersupply status. The regression coefficients for the logistic model used are shown in Table 21. Predictive risk model coefficients were estimated using 2012 data when possible to estimate the independent association with 2016 undersupply status. A negative coefficient in the model implies a reduced risk of future undersupply as the value of the variable increases. Because our model was designed to be predictive rather than explanatory, it is difficult to directly interpret coefficients as they may well share predictive power with other variables with which they have a high degree of collinearity. In particular, we note that continuity of care (as measured by GPPS) and rurality of practices were not statistically significant predictors of future undersupply status, although we did retain these variables in the model. This is not to say that they are not themselves associated with undersupply status, but that other variables in the model already account for their predictive contribution. Furthermore, we note that the interaction between the expected proportion of GP FTEs still working in patient care and the ratio of nurse FTE to GP FTE had a relatively large p-value (0.177). In initial modelling (before excluding practices on the basis of data quality; see Practice registered population), this variable had a smaller p-value (0.06), indicating that there was some evidence that it was worth including. When exclusions were applied, the coefficient did not change meaningfully. This fact, combined with the a priori expectation that the effect of expected future GP workforce would be dependent on recruitment, provided sufficient support to maintain the interaction term.

We applied the predictive risk model to the 2012 data on which it was developed to estimate historical predictive risk scores for each practice in England with available data. A comparison of these predictive risk scores to the 2016 supply–demand categorisation was used to assess the predictive value of the model. Figure 12 shows the ROC curve from this comparison. The area under the curve is 0.759. The ROC curve from a model including only the defining factors (GPPS access scores and adjusted weighted list size per FTE) was 0.718, suggesting that the full model provided a modest, but meaningful, improvement in predictive value.

FIGURE 12.

Receiver operating characteristic curve for the predictive risk model based on the national data used to build the model. Area under ROC curve = 0.7586.

Future risk predictions

Applying the risk prediction model to data from 2016, seeking to predict the risk of future supply–demand imbalance for individual practices in south-west England, we obtained risk scores for 368 practices with available data remaining after applying exclusions. This risk prediction was based on the projected future workforce from responses to the ReGROUP survey. The median probability of future supply–demand imbalance across practices was 5.4% (IQR 2.8–10.0%). In total, 40 (10.9%) practices had a risk of > 20% and 12 (3.3%) had a risk of > 50%. In the absence of any clear separation in the distribution, we defined the highest 25% of practices as being at ‘high risk’ (> 10% risk) and the lowest 25% of practices as being at ‘low risk’ (< 2.8% risk). Using these cut-off points we rated the practices in south-west England. Table 22 shows the characteristics of those practices in south-west England classified as being at high risk of future supply–demand imbalance compared with other practices. In contrast to the current situation shown in Table 21, there was no evidence (p = 0.445) that the total GP FTE varies between high/other risk classification. There was evidence, however, that all other descriptive factors varied between those practices classified as being at ‘high risk’ when compared with other practices. The practices at ‘high risk’ of supply–demand imbalance over the next 5 years tend, currently, to have larger list sizes, to have a higher nurse workforce (relative to GP workforce), to serve more deprived and younger populations, to be in urban areas and to have considerably worse GPPS scores. There was also variation between CCGs, with over half of practices in North Somerset being classified as ‘high risk’ compared with only 15% in Northern, Eastern and Western Devon.

Stress-testing scenarios

Further risk predictions were calculated using the expected future workforce based on GP age and gender alone and through exploring a number of hypothetical scenarios building on the baseline scenario described immediately above. For each of these scenarios, we classified each practice in terms of relative risk and absolute risk. Relative risk classifications are defined in accordance with the position of each practice in the distribution of risk scores for all practices for a given scenario. Relative risk classifications are used to illustrate how the ordering of practices, in terms of risk, changes under different scenarios. We classify the 25% of practices with the highest risk scores (under a given scenario) as ‘high relative risk’ and the 25% of lowest practices with the lowest risk scores as ‘low relative risk’. Absolute risk classifications are used to illustrate how the actual risk of supply–demand imbalance increases with increased stresses on the system. We classify practices in terms of absolute risk using the cut-off points used in the baseline predictions (i.e. a risk of being in a state of supply–demand imbalance in 5 years’ time of 10% and 2.8%). Figure 13 illustrates the changes to relative risk (see Figure 13a) and absolute risk (see Figure 13b) using these classifications. In this figure, each practice is represented by a horizontal bar. The ordering of each practice (vertical position) is the same for each scenario based on the rank ordering of each practice in accordance with the baseline risk prediction. For each scenario, the colouring of each practice illustrates the relative or absolute risk classification (see Figure 13a and Figure 13b, respectively) such that changes in colour indicate changes in risk classification.

FIGURE 13.

Rating of practices in south-west England from different risk prediction scenarios a–d using cut-off points defined by (a) the quartiles of each prediction (relative risk); and (b) the quartiles of the baseline prediction (absolute risk). a, Risk prediction as for baseline, but using ages and genders of GPs alone rather than including responses to ReGROUP survey. In each case the practices are ordered by the baseline scenario.

Comparing the baseline prediction in which responses to the ReGROUP survey were used to predict the future GP workforce remaining in patient care with a prediction in which GP age and gender alone was used, very little difference was observed in practices categorised as being either at ‘high relative risk’ or ‘high absolute risk’ of undersupply (seen in Figure 13 as limited reclassification of practices, correlation of ranks = 0.999).

‘Scenario a’ was intended to illustrate the effect of a modest increase in the difficulty in recruitment of GPs to replace those leaving (where the model coefficient for the expected proportion of GP FTEs remaining in patient care is increased by 2). For the most part, those practices classified as being at ‘high relative risk’ in the baseline scenario remained so under this scenario, as can be seen by the limited changes in Figure 13a (correlation in ranks between scenario a and baseline = 0.97). However, there was a dramatic increase in the number of practices with a predicted absolute risk of future undersupply of > 10%, illustrated in Figure 13b (scenario a), in which there is an increase in the number of practices coloured dark green. Increasing the model coefficient for the expected proportion of GP FTEs remaining in patient care further (by four from baseline) illustrates the potential effect if the recruitment of GPs was to become much harder (scenario b). Although a greater disturbance in the classification of practices was observed, the reclassification in terms of relative risk was still relatively modest (see Figure 13a, scenario b, correlation in ranks between scenario b and baseline = 0.90). Conversely, the reclassification in terms of absolute risk (see Figure 13b, scenario b) was significantly greater; the majority of practices had a predicted risk above 10% (see Figure 13a, scenario b). Practices at increased risk of undersupply from difficulty in recruiting GPs were likely to be those where current doctors are older. However, little difference was seen in the type of practice affected under these scenarios (see Appendix 35).

As with scenarios a and b, increasing the projected practice population resulted in only modest changes in respect of which practices are classified as being at ‘high relative risk’. Only a small relative increase was seen when comparing scenarios c and d (see Figure 13a, correlation in ranks between scenario c and baseline = 0.99 and scenario d and baseline = 0.98). However, substantial changes were seen in the number of practices with an absolute risk of undersupply of > 10% (see Figure 13b, scenarios c and d). Appendix 35 shows the descriptive features of practices classified as ‘high relative risk’ under these two scenarios of increased practice populations. In general, the type of practice classified as being at ‘high relative risk’ was similar to that seen under other scenarios (baseline, a and b), but under the scenarios of increased projected populations, those with more GP FTEs were also more likely to be at high risk. Combining the effect of scenarios a and c resulted in relative risk classifications closer to the baseline predictions than that for scenario a alone. However, in terms of absolute risk, more practices had a risk of > 10% (see Figure 13b, scenarios a and c). When scenarios b and d were combined (illustrating a situation in which it was much harder to recruit GPs combined with an increased practice population of 40%), it was evident that nearly all practices (88%) exceeded a 10% absolute risk of supply–demand imbalance within 5 years, with only nine practices (2.4%) classified as being at ‘low absolute risk’ using the cut-off points derived from the baseline predictions. Descriptive statistics for those practices at ‘high relative risk’ under these scenarios are shown in Appendix 35. It should be noted that we would not expect these substantial increases in practice populations to apply to all practices, but this illustrates those that would be at high risk if such an increase did occur.

Summary of main findings

The aim of this research was to identify NHS general practices that may face an undersupply in GP workforce within the next 5 years. To illustrate this, we developed and evaluated a conceptual model of workforce supply–demand imbalance and, based on this, produced a risk prediction model informed by historical data that could be used to predict an individual practice’s risk of being in a state of undersupply in 5 years’ time. The risk prediction model produced a range of risk scores across south-west England and might be considered to have a fair to good discriminatory ability in this context (based on the ROC curve analysis). Applying this to current data suggests that the practices at highest risk of an undersupply of GP workforce in the future have, on average, larger patient lists, employ more nurses relative to GPs, serve more deprived and younger populations and have considerably worse patient experience ratings. Furthermore, we found that there was geographical variation in those practices being classified as at ‘high risk’, with the majority of practices in North Somerset being classified as being at ‘high risk’.

We then modelled scenarios in which the recruitment of GPs becomes harder and/or practice populations increase dramatically beyond what would be expected from historical trends (such as might be expected with a new housing development). These scenarios do increase absolute risk dramatically, but by and large it is the same practices in all scenarios that are at highest risk of future undersupply of GP workforce. This almost certainly reflects the fact that those practices most likely to have problems in the future are those currently experiencing difficulties. This can be seen in the relatively good predictions from a simple model including only defining factors based on routinely available data (i.e. workload-to-workforce ratio and access scores), which had an area under the ROC curve that was not dramatically less than in the full model. In particular, we found that the inclusion of findings from our survey of GPs’ career intentions had very little impact on the predictions, compared with using expected future workforce projections based only on data regarding GPs’ genders and ages.

During the course of our research, a serious situation emerged in respect of the GP workforce in Plymouth,146 a city within the geographical area of our research interest. An opportunistic secondary analysis of our data (Box 1), identified findings that, although preliminary, tended to support the validity of the model we described.

Implications

We have demonstrated that it is possible to make reasonable predictions of an individual general practice’s future risk of undersupply of GP workforce with respect to its patient population. With ongoing GP workforce issues in south-west England and nationally, many local models are being developed to identify potentially ‘at-risk’ practices. However, unlike the model we present here, it is not clear to what extent these models are evidence based or to what extent their limitations are recognised by the users of the models. The predictions are inherently limited by the data that are available. Improvements in data going forwards will help the situation, particularly if data on practice recruitment are released. However, it will be some time before robust historical data exist that can be used for the model development process outlined here. If models such as the one outlined here are to be produced and used, it is important that high-quality data continue to be collected. The predictions produced by our model may allow targeting of interventions to retain and attract GPs either in specific practices or in specific regions currently at high risk of problems driven by workforce supply. Although the model we present here provides fair discrimination, much could be achieved by focusing efforts on those practices currently experiencing difficulties.

Although a policy of targeted interventions may have a place, most practices are likely to be at a high risk of workforce undersupply when faced with a substantial increase in demand from an increased patient population combined with major difficulties in recruiting GPs. For this reason, local knowledge of drivers of increased practice populations, such as housing developments, will be key to being able to suitably apply targeted interventions. Even in south-west England where workload and ability to meet patient demand are better than in England overall, most practices are vulnerable to recruitment challenges. Given this, overarching policies and strategies may be more effective than targeted ones, especially if areas of population growth are not known.

Recruitment

Target stakeholder groups included representatives of CCGs, LMCs, NHS England Local Area Teams, AHSNs and PPI groups, as well as representatives from national medical and government organisations including the Department of Health and Social Care (DHSC), the BMA, the RCGP, HEE, NHS Employers and the GMC. Members of our project advisory board provided key stakeholders from within HEE, NHS England, the BMA and the RCGP, and these were supplemented by extensive web searches for names and contact details of CCG/LMC chairpersons and others in workforce planning roles within national organisations or who had published on the subject of GP workforce in recent years.

The project management group reviewed the list of potential participants to ensure that no key groups or individuals had been omitted. All participants were e-mailed a formal invitation containing web-page links to register attendance at a meeting in June 2017 in either London or Leeds. Non-respondents were followed up by e-mail at regular intervals, and a new invitation was sent to any new individuals suggested by those unable to attend. Of 121 individuals approached, 48 registered to attend one of the meetings, of whom 44 (36% of those approached) attended on the day (22 at each meeting) (see Appendix 36).

Formulating the discussion topics

Both meetings involved discussion of the same topics to facilitate comparisons and identify differences between regions on emergent ideas. We identified 11 policy/strategy topics for discussion.

The discussion topics were derived mainly from the key policies and strategies prioritised by the RAM workstream (see Chapter 5) where the RAM panel had reached overall agreement that these policies and strategies were ‘appropriate’ and some where there was also consensus on them being ‘feasible’.

Topics 1–10 encompass the 16 policies and strategies that had been rated by the RAM panel (with consensus) as ‘appropriate’ and ‘feasible’ (Table 23).

A further six topics were then added for consideration within the stakeholder consultations. Topics 11–13 had been rated (in the form of statements) by the RAM panel as being ‘uncertain’ or with no consensus regarding appropriateness (‘external HR interventions and monitoring/support’) or as being ‘appropriate’ but ‘uncertain’ regarding feasibility (‘maximum number of consultations for GPs’ and ‘making consultations longer’). Topics 14 and 15 had been suggested as areas for potential inclusion in the RAM during the initial statement development work, but had not been taken forward for rating by the RAM panel. Topic 16 was added as an area of methodology for supporting implementation of policies and strategies, or for future research.

From the 16 topics in Table 24, we formed three broad categories (Box 2). Eleven discussion topics were formed from the 16 draft topic titles by combining the two topics concerning identifying the ‘at-risk’ (of workforce undersupply) status of general practices and a new focus on under-doctored areas (1 and 2); the two topics concerning portfolio working (7 and 8); the two topics concerning incentives and support packages for GPs (4 and 5); and also the three topics covering support and planned exits for GPs nearing retirement and the implications of losing previous pension incentives (9, 15 and 16).

In addition, graphic facilitation was seen as an innovative and potentially useful means of capturing some of the process and key discussion points from the stakeholder events and so was undertaken at each meeting with the aim of capturing the key themes generated by the group discussions.

Data collection and analysis

We adopted a consistent structure for the two meetings (see Appendix 37). Following presentation of our research findings, stakeholders were asked to work in their allocated round-table groups, seated at tables of seven or eight people. During round-table discussions, each group covered between one and three policy and strategy areas in each 45-minute discussion session.

We anticipated that stakeholders would take a broad view of the feasibility of policy and strategy implementation, noting barriers and facilitators that might impede or enhance the utility of our research to the front-line NHS. Discussions were facilitated by two project team members at each of the table groups and were structured as follows:

• What factors could have an impact – positively or negatively – on the implementation of the emergent recommendation?

• Please comment on feasibility and acceptability of the emergent recommendation.

• What are the structures or groups that would need to be involved in introducing change to this area?

• What are the three key next steps to move this agenda forward?

Facilitators recorded stakeholders’ comments on sticky notes (and/or stakeholders did this themselves) and added them to flip charts during the discussion, which were referred back to during the closing summary session. In addition, stakeholders were provided with a poster summarising the findings of the systematic review, and facilitated whole-group discussion throughout the event and at the end of the day’s proceedings.

Post-event processing of collected information

Following discussion among the research team, Emily Fletcher summarised the key messages that had been collected on the sticky notes on flip charts by examining those comments that fell within the following broad categories: (1) positive and negative factors for implementation; (2) comments regarding feasibility and acceptability; (3) region-specific (i.e. ‘north’ or ‘south’) issues raised; and (4) key actions required, including those organisations that needed to be involved to make changes to the policy area. This summary was initially produced by Emily Fletcher and was cross-checked with members of the project team who had acted as table group facilitators (JLC, RA, CS, SD, SR, RC and AA).

Strengths and limitations

A strength of this work was the successful recruitment and involvement of representatives from key stakeholder organisations, including some in very senior and influential roles within medicine and in health more widely, and within major professional, government, policy and related organisations. Importantly, we incorporated the participation of PPI representatives at both meetings. A core member of the project’s PPI group acted as a facilitator for the round-table discussions in London, and two members of the public in Leeds representing a local practice PPG and the University of Manchester PPI network actively participated as stakeholders. This directly supported the view of the project’s PPI group that patients and members of the public should take part in the conversations and could bring their experience as patients or from other fields of work or industry.

The meetings followed an identical structure and used focused discussions on answering specific questions, and yielded clear statements on both the perceived feasibility of the strategies and potential specific actions to be taken by particular organisations or groups.

Although the RAM workstream excluded the consideration of policies and strategies regarding MDT working, role substitution and skill mix, and also the implications of recent changes in pension policies, this stakeholder consultation workstream did seek views on these areas. These were felt to be important areas for consideration by stakeholder participants, owing to the ongoing implementation of new roles and the likely impact of changes to pension arrangements on the number of older GPs considering early retirement.

Limitations include the fact that our findings comprise a summary list of the views captured across both meetings; the finer detail of individual comments are not reported. Nor have we systematically attempted to draw out the potential complementarity or inevitable tensions between some of the proposed retention strategies (e.g. between having a daily maximum limit on the number of consultations and increasing the typical length of GP consultations). In addition, in order to make good use of stakeholders’ time at the meetings and to present sufficient background material on the findings of the whole of the ReGROUP project, we needed to restrict the number of discussion topics that were possible to cover in detail. Furthermore, each of the 11 potential strategies was discussed in detail by only one table group of stakeholders at each meeting. Nevertheless, an attempt to implement change in 11 policy and strategy areas at the same time would cause disruption; further consideration of prioritising these areas to avoid creating additional pressure on the existing GP workforce is needed.

Conclusions

Following identification of policies and strategies that are likely to support the retention of GPs in direct patient care through the work described in the preceding chapters, we organised two successful stakeholder engagement events, which proved dynamic, supporting wide-ranging but focused discussion. The consultation has identified potentially important areas for policy and strategy development and has crystallised which workforce partner groups and stakeholder organisations might be best placed to lead on innovation and change. However, most of these areas will require some form of evaluation and so this work is a first step towards testing potentially important areas and gaining preliminary evidence on feasibility and acceptability.

Patient and public involvement group recruitment and composition

The project PPI group was recruited from local PPI networks. People who had attended the initial workshop were invited to continue participating. The final group consisted of two men and five women, representing individuals with experience of a range of long-term physical and mental health conditions and experience of regular contact with GP/primary care services. Some of the group also had experience as carers for elderly relatives or for children with life-altering health conditions. One member had extensive experience as a lay representative within CCGs, and with the Quality Outcome Framework as a lay assessor of general practices.

Initial patient and public involvement meeting

At the outset of the project (January 2016), the PPI group was invited to a half-day workshop facilitated by the PPI lead (JW), the project manager (EF), a qualitative researcher (AS) and a systematic review researcher.

Workstream 1: systematic review

A meeting was held between the PPI lead (JW), the systematic review lead (RA) and one of the systematic review researchers to consider how best to involve the PPI group in the review. Given the time constraints of the project and the need to produce findings promptly, the review process was under way before the PPI group was in place. It was decided that the PPI group could have the most impact by being involved in sharing and discussing emerging findings from the review with the systematic review team.

A half-day workshop was held in June 2016, attended by Rob Anderson, the project information specialist, one of the systematic review researchers and the PPI group.

Workstream 2: census survey of south-west general practitioners

The survey instrument used in this project was based on that used in preliminary work, the design of which had been informed by that study’s PPI group. 17 The current PPI group were, however, sent the survey via e-mail for additional comments.

Workstream 3: qualitative research

An initial planning meeting was held between the PPI lead (JW) and the two qualitative researchers (AS and RT). As a result of this, the qualitative researchers decided to produce a briefing document for the PPI group with key areas on which they wished to receive PPI feedback.

The PPI group were offered the opportunity to attend a half-day qualitative research training course: ‘An Introduction to Qualitative Research’ delivered by the PenCLARHC (http://clahrc-peninsula.nihr.ac.uk/event/making-sense-of-evidence-an-introduction-to-qualitative-research1; accessed 4 December 2018). Several members attended. Others had either previously attended this course or contributed to PPI in qualitative research in other studies.

Workstream 4: RAND/UCLA Appropriateness Method

In the early stages of the RAM workstream, the RAM researcher (RC) circulated explanatory documents to the PPI group. These included a plain English ‘Introduction to the RAM Process’ and a completed example from a previous research study.

Two planning meetings were held before the PPI workshop, attended by the RAM researcher (RC) and PPI lead (JW), and with a member of the PPI group (AA). Given the novelty and complexity of the process, it proved useful for Alex Aylward (with background experience of working as a lay representative with CCGs and general practices) to inform the development of the PPI process.

In October 2016, a half-day workshop was led by Rupatharshini Chilvers and Jo Welsman and attended by members of the PPI group. At this stage, the RAM categories and policy statements were in development, and thus there was scope to incorporate PPI into both the contextual information around the statements and the statements themselves that would be provided to RAM panel members. PPI was considered less relevant in the formal RAM panel process and analysis owing to its prescriptive and quantitative nature.

Workstream 5: workforce predictive risk modelling

Given the nature of the work involved in this workstream, no formal input from the whole PPI group was sought. However, Alex Aylward was involved in project management meetings and participated in discussions regarding results and interpretation, and commented on drafts of the write-up.

Workstream 6: stakeholder consultations

Alex Aylward received facilitator training and facilitated round-table discussions at the London stakeholder consultation meeting. In addition, two experienced PPI representatives from Leeds were recruited to attend the Leeds stakeholder consultation meeting as stakeholders in their own right to participate in the round-table discussions of emergent policies and strategies for retention, as described in Chapter 7. Both PPI representatives provided positive feedback via e-mail on the success of the meeting and on their experience of being involved in the conversations.

Results meeting

In preparation for the stakeholder consultation meetings, the research team rehearsed presentations of the project’s results at a meeting attended by two PPI members, who fed back to help finalise presentation content.

Dissemination

Three PPI members attended an informal workshop to review their participation and input throughout the project, and also to hear the whole project’s conclusions. They were asked for their ideas regarding dissemination to non-academic/clinical groups. Following discussion, it was suggested that a short film or infographic could be produced, highlighting the issues of GP retention and the ways in which the public might support their GPs, based on the main findings of the project. In terms of routes for dissemination, PPI members suggested that these may include practice waiting room screens/websites, circulation to PPGs, CCGs and various patient care forums and condition-specific charities.

Report writing

Alex Aylward co-authored this PPI report. The report was circulated to the wider PPI group for comments and amendments.

In addition, the Plain English summary for this final HSDR report was circulated to all members of the PPI group for comments and amendments. The following key points were changed/incorporated/recommended:

• ‘quit’ was replaced by ‘leave’ in order to avoid negative connotations associated with GPs leaving the profession

• ‘stakeholder’ was replaced by ‘interested parties’ and ‘people with a specific interest in the subject’

• ‘workstreams’ was replaced by ‘activities’.

Objective

This systematic review aimed to answer the following question: what are the factors in the UK and other high-income countries that affect GPs’ decisions to (1) quit direct patient care (including reducing their time commitment to it), (2) take career breaks from general practice and (3) return to general practice after a career break?

General practitioner is used here to denote any medically qualified clinical professional (doctor) whose main clinical role is community based, outside hospital, and who is a primary care specialist rather than a clinical specialist focusing on a disease or disease process.

Methods

Searches identified both published articles and ‘grey’ literature in English and from 1990 onwards, using both conventional bibliographic searches of relevant databases (MEDLINE, MEDLINE in Process, PsycInfo, HMIC, Cochrane, ASSIA, Web of Science) and complementary methods (web searching, forwards and backwards citation chasing). Searches were conducted in January 2016 and updated in April 2016 and yielded 5227 records after deduplication. The review used prespecified inclusion and exclusion criteria and was registered on PROSPERO.

Included studies

We included 34 survey-based (mostly quantitative; 22 studies from the UK) and five qualitative interview-based (four from the UK) studies.

Results

General practitioners in the UK leave general practice for a very wide range of factors: both negative job-related ‘push’ factors and positive, leisure-, retirement- and home-life related ‘pull’ factors. Although some factors clearly operate at an individual, personal level (such as the financial ability to retire, health, family and marital circumstances, or good/poor relationships with practice partners), other factors operate at the level of the general practice, local area, the whole profession or the national health system (e.g. media portrayal of GPs, service reform and performance targets, CQC inspections and professional revalidation).

Conclusions

Four closely related, job-related factors play a major part in decision-making about both early retirement and part-time working: workload, job (dis)satisfaction, work-related stress and work–life balance. However, many other detailed factors either underlie these higher-level factors or are more important for a significant minority of GPs. These factors could provide a focus for developing a variety of GP retention initiatives.

Objectives

Given recent concerns regarding general practice workforce capacity, we aimed to describe GPs’ career intentions, especially those that might have an impact on GP workforce availability over the next 5 years.

Design

Census survey, conducted between April and May 2016 using postal and online responses, of all GPs on the NHS National Performers List and eligible to practise in primary care. Two reminders were used as necessary.

Setting

South-west England (population 3.5 million), a region with low overall socioeconomic deprivation.

Participants

A total of 2248 out of 3370 eligible GPs (67% response rate).

Outcomes

Reported likelihood of permanently leaving or reducing hours spent in direct patient care or of taking a career break within the next 5 years, and present morale weighted for non-response.

Results

Responders included 2179 GPs engaged in patient care. Of these, 863 (37% weighted, 95% CI 35% to 39%) reported a high likelihood of quitting direct patient care within the next 5 years. Overall, 1535 (70% weighted, 95% CI 68% to 72%) respondents reported a career intention that would negatively affect GP workforce capacity over the next 5 years, through permanently leaving or reducing hours spent in direct patient care, or through taking a career break. GP age was an important predictor of career intentions; sharp increases in the proportion of GPs intending to quit patient care were evident from the age of 52 years. Only 305 (14% weighted, 95% CI 13% to 16%) GPs reported high levels of morale, whereas 54% (weighted, 95% CI 52% to 56%) reported low levels of morale. Low levels of morale was particularly common among GP partners. Current morale strongly predicted GPs’ career intentions; those with very low levels of morale were particularly likely to report intentions to quit patient care or to take a career break.

Conclusions

A substantial majority of GPs in south-west England report low levels of morale. Many are considering career intentions that, if implemented, would adversely affect GP workforce capacity within a short time period.

Objective

To help identify the potential content of policies and strategies aimed at retaining experienced GPs and/or supporting the return of GPs following a career break, through an exploration of the lived experience of GPs, including their remaining/leaving decision-making.

Design and setting

Thematic analysis of CQC practice report data to explore examples of good and poor practice in south-west England, followed by semistructured interviews with GPs and other primary care stakeholders in south-west England. Transcribed interviews were analysed thematically.

Participants

Forty-one GPs identified from GP workforce census survey returns: seven retired GPs; eight GPs aged 50–59 years who were intending to retire; 11 GPs on a career break or intending to take one; nine GPs aged 35–49 years who had quit or were intending to quit direct patient care; and six GPs who did not intend to quit and had good morale. In addition, 19 stakeholders were opportunistically sampled from primary-care-related settings (local and regional) in the south-west.

Results

Examples of good practice were extracted from the CQC report data and these were used to develop interview prompts and questions, and to inform workstream 4. The interview findings showed that (1) factors and issues need to be addressed collectively, (2) there are inherent tensions and contradictions within potential solutions that need to be considered and (3) there is also a need to address GPs’ lived experiences of their work and role in the current health-care climate. The lived experience was illustrated through three underlying themes that emerged from the data: identity and value, fear and risk, and choice and volition.

Conclusions

Use of evidence from occupational and workplace literature adds weight to the argument that solutions should not be ‘sticking plasters’ but rather implementable policies and strategies that will help to (1) increase the perceived value of GPs and clarify the identity and future of general practice, (2) reduce the levels of fear and risk that GPs are experiencing (to acceptable and manageable levels) and (3) provide GPs with feasible, acceptable and sustainable routes to remaining in direct patient care.

Background

The development of policies and strategies to retain experienced GPs in direct patient care in the UK must acknowledge the current policy and practice contexts within which GPs are working. Using this contextual information and research evidence, we generated a set of potential policies and strategies aimed at maximising GP retention and tested their appropriateness and feasibility for implementation.

Methods

Using the RAM, a panel of GP partners and GPs working in national stakeholder organisations assessed the appropriateness of potential policies and strategies in two rounds of an online survey. The policies and strategies were those that could plausibly maximise GP retention within a 5-year window. Supported by an evidence briefing paper, panellists evaluated 54 potential policies and strategies that could be targeted at either national/regional, general practice or individual GP levels. Ratings were analysed for consensus and categorised based on appropriateness (appropriate, uncertain, inappropriate). A subset of statements rated as ‘appropriate’ after the first round of the online survey was also assessed for feasibility (feasible, uncertain, unfeasible) in the second round.

Results

The panel deemed 24 out of 54 policies and strategies as appropriate and 16 out of 19 as feasible. Policies and strategies were endorsed from a number of areas and considered suitable for implementation at national/regional, general practice or GP levels.

Conclusion

This systematic approach identified 24 potential policies and strategies deemed to be appropriate for improving retention of GPs in the workforce, the majority of which were deemed feasible. Of these, six were for implementation at national/regional level, four at general practice level and 14 at GP level. Future research evaluating the appropriateness and feasibility of other policies and strategies falling outside the scope of our research (e.g. primary care skills mix) is warranted.

Objectives

Through other workstreams, the wider ReGROUP project has identified and proposed some potential policies and strategies targeting the retention of the experienced GP workforce. To most effectively apply these it would be beneficial to identify those practices at highest risk of facing a supply–demand imbalance. We therefore aimed to develop a modelling framework to identify such practices.

Design

A hybrid modelling approach was used to predict imbalance based on practice factors including current workload, current GPPS scores, nurse workforce, projected populations and the projected fraction of existing GP workforce remaining in direct patient care. A predictive model was developed using historical data (2012–16), and current data (2016), which were used to predict future risk. The utility of surveying GP quitting intentions was explored.

Setting/participants

Predictive model development used data for all general practices in England, whereas predictions of future supply–demand risk status were restricted to south-west England.

Results

Based on historical data, the predictive model had fair to good discriminatory ability to predict which practices faced supply–demand imbalance (area under the ROC curve of 0.759). Predictions using current data suggested that practices at highest risk of future supply–demand imbalance on average have larger patient lists, employ more nurses, serve more deprived and younger populations, and have considerably worse patient experience ratings. Findings from a survey of GPs’ career intentions had very little impact on the predictions of future supply–demand risk status, when compared with using expected future workforce projections based only on routinely available data on GPs’ genders and ages.

Conclusions

It is possible to make reasonable predictions of an individual general practice’s future risk of undersupply of GP workforce with respect to its patient population. However, the predictions are inherently limited by the data available. Although targeted interventions may have a place, blanket policies may be more effective, given that most practices in south-west England are likely to be ‘at risk’ if faced with substantial increases in demand combined with major difficulties in recruiting GPs.

Objectives

Other ReGROUP project workstreams have identified potential policies and strategies to support the retention of GPs. From the stakeholder consultations we aimed to gather preliminary evidence from workforce partners and key stakeholder organisations on the feasibility and acceptability of implementing the policy changes and initiatives proposed.

Design

Two consultation meetings were organised at which stakeholders were invited to listen to the research evidence produced by the ReGROUP project. Discussions on 11 policy and strategy areas took place in small round-table groups, facilitated by project team members, and comments were captured using sticky notes and flip charts.

Setting

London and Leeds, UK.

Participants

Stakeholders comprised regional representatives of CCGs, LMCs, NHS England Local Area Teams, AHSNs and PPI groups, and also included representatives from national bodies such as the BMA, RCGP, HEE, NHS Employers and GMC.

Outcomes

Stakeholders explored the practicalities of implementing change to the proposed areas of policy and strategy to support retention, specifically focusing on barriers to and facilitators of change, feasibility and acceptability, and key actions that could be undertaken by particular organisations to introduce change.

Results

Stakeholder feedback was obtained for three broad areas of policy and strategy: (1) ‘protecting’ GPs and managing expectations of patients, (2) incentives and support mechanisms for GPs and (3) portfolio and wider working arrangements. Tangible actions were identified for national organisations including NHS England, the RCGP, HEE and CEPNs, as well as for CCGs and individual practices, or federations/groups of practices.

Conclusions

A broad range of stakeholders involved in workforce planning had the opportunity to contribute to discussions on emergent policies and strategies to support retention of GPs from our research. Although this work has successfully highlighted specific actions that could be taken by workforce partner groups and stakeholder organisations, it is a first step towards formal evaluation of the feasibility and acceptability of the potential policies and strategies to support retention.

MEDLINE (via OvidSP)

Date range searched: 1946 to January week 3 2016.

Date searched: 29 January 2016.

Searcher: SR.

Hits: 3655.

MEDLINE(R) In-Process & Other Non-Indexed Citations (via OvidSP)

Date range searched: 28 January 2016.

Date searched: 28 January 2016.

Searcher: SR.

Hits: 87.

PsycINFO (via OvidSP)

Date range searched: 1806 to January week 4 2016.

Date searched: 29 January 2016.

Searcher: SR.

Hits: 511.

Health Management Information Consortium (via OvidSP)

Date range searched: 1979 to November 2015.

Date searched: 29 January 2016.

Searcher: SR.

Hits: 417.

Applied Social Sciences Index of Abstracts (via ProQuest)

Data parameters: not applicable.

Date searched: 29 January 2016.

Searcher: SR.

Hits: 214.

Cochrane (Cochrane Collaboration)

Data parameters: CENTRAL – Issue 12 of 12, December 2015; CDSR – Issue 1 of 12, January 2016.

Date searched: 29 January 2016.

Searcher: SR.

Hits: 75.

Web of Science (via Thomson Reuters)

Data parameters: SCI-EXPANDED and SSCI.

Date searched: 29 January 2016.

Searcher: SR.

Hits: 1702.

Number of hits