Article Text
PDF
Abstract
Objectives Investigate trends in continuity of care with a general practitioner (GP) before and during the COVID-19 pandemic. Identify whether continuity of care is associated with consultation mode, controlling for other patient and practice characteristics.
Design Retrospective cross-sectional and longitudinal observational studies.
Setting Primary care records from 389 general practices participating in Clinical Practice Research Datalink Aurum in England.
Participants In the descriptive analysis, 100 000+ patients were included each month between April 2018 and April 2021. Modelling of the association between continuity of care and consultation mode focused on 153 475 and 125 298 patients in index months of February 2020 (before the pandemic) and February 2021 (during the pandemic) respectively, and 76 281 patients in both index months.
Primary and secondary outcomes measures The primary outcome measure was the Usual Provider of Care index. Secondary outcomes included the Bice-Boxerman index and count of consultations with the most frequently seen GP.
Results Continuity of care was gradually declining before the pandemic but stabilised during it. There were consistent demographic, socioeconomic and regional differences in continuity of care. An average of 23% of consultations were delivered remotely in the year to February 2020 compared with 76% in February 2021. We found little evidence consultation mode was associated with continuity at the patient level, controlling for a range of covariates. In contrast, patient characteristics and practice-level supply and demand were associated with continuity.
Conclusions We set out to examine the association of consultation mode with continuity of care but found that GP supply and patient demand were much more important. To improve continuity for patients, primary care capacity needs to increase. This requires sufficient, long-term investment in clinicians, staff, facilities and digital infrastructure. General practice also needs to transform ways of working to ensure continuity for those that need it, even in a capacity-constrained environment.
- Retrospective Studies
- PUBLIC HEALTH
- Primary Care
- Quality in health care
- STATISTICS & RESEARCH METHODS
- Patient-Centered Care
Data availability statement
Data may be obtained from a third party and are not publicly available. The data analysed in this study are available for approved research studies from Clinical Practice Research Datalink (CPRD).
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
- Retrospective Studies
- PUBLIC HEALTH
- Primary Care
- Quality in health care
- STATISTICS & RESEARCH METHODS
- Patient-Centered Care
STRENGTHS AND LIMITATIONS OF THIS STUDY
We used a large-scale routinely collected patient-level data set from a nationally representative pool of English general practices.
We developed a conceptual model identifying a variety of potential patient and practice-level factors likely to influence continuity, but unobserved confounding may still affect our results.
We estimated the association between relational continuity of care and patient and practice-level factors before and during the COVID-19 pandemic, comparing findings from three different measures of continuity and two different modelling approaches.
We focused on relational continuity of care with a general practitioner only for patients with at least three consultations in a year and could not account for changes in management or informational continuity.
Data quality may be an issue, as consultation mode is not consistently recorded in the data and our method for deriving consultation mode has not yet been externally validated.
Introduction
Continuity of care is a fundamental characteristic of general practice; associated with a range of better health and care outcomes.1–5 These include adherence to treatment; uptake of preventative care; decreased emergency department use, hospitalisation and mortality; and patient satisfaction.6–13 For some patients, continuity may be particularly beneficial. Those with long-term health conditions,13–15 mental health conditions,16 polypharmacy,17 palliative care,3 vulnerable patients,18 and older and younger age groups19 20 are all more likely to value and benefit from continuity.
Continuity of care in general practice is complex and multifaceted but can be thought of as the extent to which a person experiences coordinated clinical care throughout their journey through health services. This can include relational continuity—an ongoing relationship with a clinical team, or member of a clinical team; informational continuity—the sharing of patient records and information within a clinical team to aid healthcare provision; and management continuity—coordination of care across discrete episodes.21 In this study, we focus on relational continuity. Relational continuity can be further broken down into longitudinal continuity—relating to seeing the same healthcare professional over time; and interpersonal continuity—relating to building an ongoing therapeutic relationship between a patient and healthcare professional. Patients who see the same healthcare professional consistently have higher satisfaction with their care and feel that the healthcare professional has a better understanding of their needs.12 Nevertheless, studies suggest there has been a gradual decline in relational continuity in primary care in England in recent years.22–24 Policy-makers have responded by recommending practices prioritise continuity25 26 and many practices have designed for continuity as part of their care navigation model, further enhancing their processes through the use of digital tools.26 27
Before the COVID-19 pandemic, policies were being put in place to increase the use of digital tools for care navigation and triage, building capacity and expanding supporting services such as community pharmacies. These included the 2019 National Health Service (NHS) Long Term Plan which committed to patients having the right to access their practice digitally by 2023/202428; and the 2019/2020 five-year contract framework for general practitioners (GPs), which required practices to offer web-based consultation systems by 2020/2021 and introduced the Additional Roles Reimbursement Scheme to significantly expand the primary care workforce by 2023/2024.29 At the start of the pandemic, NHS England advised all general practices to move to a total triage model where every patient contacting the practice is triaged before an appointment and, if clinically appropriate, to use remote consultations (by telephone, video or online message) to protect patients and staff from the risk of exposure to COVID-19.30 31 This led to a dramatic increase in the use of remote consultations.
If implemented well, a good care navigation and triage model, alongside expanded supporting services, can be used to help manage high demand in primary care and improve continuity.27 Understanding patient needs and preferences at the point of contact provides more opportunity for flexible delivery of care, including the use of remote consultations. Many patients appreciate being able to remotely consult with their GP, saving time and travel costs, or difficulties taking time away from work or caring responsibilities.32 However, there is little evidence for the impact these new general practice access models have on continuity of care.33
In this study, we describe trends in relational continuity of care both before and during the pandemic. We then exploit the exogenous shock afforded by the dramatic increase in remote consultations during the pandemic to investigate the association between consultation mode (remote or face to face) and relational continuity in English general practice.
Aims and objectives
The study aimed to examine trends in continuity of care before and during the COVID-19 pandemic before robustly estimating the association between consultation mode and relational continuity of care experienced by patients in English general practice. To control for confounding, a conceptual model was developed to identify factors which may influence continuity, at the patient, practice and regional level.
We assessed the sensitivity of the results to outcome measure specification, confounders and model assumptions/specifications by using three different measures of relational continuity of care, staged inclusion of confounders, and both cross-sectional and longitudinal modelling approaches.
Cross-sectional, random-intercept models were used to identify whether there were substantive differences in the between-patient association of consultation mode and continuity, both before and during the pandemic. Longitudinal, fixed effects models were used to identify the within-patient association of changes in consultation mode with continuity.
The results provide evidence as to whether the large increase in remote consultation use during the pandemic led to changes in relational continuity of care for NHS patients in England. The results are discussed with reference to the literature on remote and digital-first care. Potential implications for primary care provision and future research are outlined.
For the sake of brevity, we use the terms ‘continuity’ and ‘relational continuity of care’ interchangeably throughout.
Methods
Data sources
We conducted a retrospective, observational study spanning April 2018 to April 2021 using data from the Clinical Practice Research Datalink (CPRD) Aurum data resource.34 35 CPRD Aurum is a managed data resource based on routinely collected data from GP practices in England and Northern Ireland which use the EMIS Web software system.36 EMIS Web is an electronic health record management system used for everything from booking consultations to recording clinical findings, prescribing and referrals. In June 2021, there were over 1300 practices regularly sending data to CPRD, with over 13 million patients registered in these practices. Patients in participating practices are broadly representative of registered patients at practices in England. Data were pseudonymised but linked at the patient level to data on deprivation from the Office for National Statistics (ONS) and Hospital Episode Statistics (HES, used to supplement the ethnicity coding available in CPRD Aurum).37
Eligibility criteria
CPRD practices in England with a last data collection of March 2021 were eligible for inclusion. A small set of practices (29) were purposively selected due to inclusion in a separate study being undertaken by the Health Foundation.38 An additional 371 practices were randomly sampled (without replacement) to make a total of 400 practices. Practices were sampled from a distinct list of practice IDs using the ‘sample’ function in R.39 Due to data quality concerns, 11 practices were subsequently discarded, resulting in a final total of 389 practices (online supplemental figure 1). Eligible patients were originally those who had an acceptable data quality according to CPRD; were registered at one of the 400 originally sampled practices; had at least 25 months of continuous registration from 1 March 2016 onwards; were recorded as male or female gender (other groups risked identification due to very low counts); and were eligible for HES/ONS linkage. Patients were sampled at this point (without replacement) from a distinct list of patient IDs using the ‘sample’ function in R.39 Further eligibility requirements were then imposed: patients needed their ethnicity to be recorded in CPRD or linked HES records; have a valid measure of home locality deprivation recorded; and be registered at a practice with no data quality concerns (the retained 389 practices). For inclusion in any month, the patient needed to be at least 2 years old by the month-end; continuously registered at the practice for the 2 years prior to the month-end; and have 3 or more consultations in the 12 months prior to the month-end. For the descriptive analysis, samples varied by month but were all above 100 000. For the cross-sectional analysis, which focused on the index months of February 2020 (before the pandemic) and February 2021 (during the pandemic), 153 475 and 125 298 patients were eligible, respectively. For the longitudinal modelling, which included patients eligible in both of these index months, 76 281 patients were included (online supplemental figure 2). More detail on eligibility and sample selection is available in online supplemental material.
Supplemental material
Coding consultations
CPRD Aurum contains data on a wide range of practice staff activities. To identify consultations and their mode, we extended the method used by Watt et al.40 Only records of consultations carried out by GPs were included. Modes of consultation included face to face in the practice or at home, or remotely by telephone, online message or video. Where the mode was not specified, we assumed the consultation was delivered face to face, as deviations from this default were specifically coded as such prior to the pandemic.
To reduce the possibility of duplication, consultations with the same GP were limited to one each day, with face-to-face consultations taking precedence where multiple modes were recorded for same-day consultations. It is important to note that work is ongoing by CPRD to categorise consultations in EMIS Web. For example, only the most commonly occurring text values are categorised in the consultation ‘source’ field. Approximately 5.5% of records were ‘awaiting coding’ in both the consultation ‘source’ and ‘medical code’ field.35 These records were ignored.
Continuity of care outcome measures
To aid comparability with previous research,41–43 we analysed two of the most commonly used measures of relational continuity of care: the Usual Provider of Care Index (UPCI)44 and the Bice-Boxerman Index (BBI).45 The UPCI is the percentage of consultations with the most frequently seen GP and ranges from 0 to 1, exclusive of the zero limit. The BBI is a measure of dispersion of consultations between GPs and ranges from 0 to 1, inclusive. We also analysed the count of consultations with the most frequently seen GP without transforming it to an index. This provided a useful comparison to the UPCI, by allowing direct analysis of the count data on which the UPCI is based, without modelling a derived percentage.
All outcomes were calculated using 12 months of consultation data prior to the end of each month. Patients needed to have at least three consultations to be included in a month. This criterion was enforced to provide a minimum number of values (3) the measures could take. More detail on the outcome measure specifications is available in supplemental material.
Remote Consultation Index
Our main predictor of interest was the proportion of consultations with a GP that were conducted remotely, which we call the Remote Consultation Index (RCI). The RCI was calculated using 12 months of consultation data prior to the end of each month.
Patient-level covariates
We incorporated a range of patient-level covariates. These included gender (male or female); age in years; ethnicity, coded into five categories46; the home locality Index of Multiple Deprivation (IMD) decile47; the total consultations experienced each month; the number of referrals to specialist care in the year prior to each month; and the number of long-term conditions the patient had in the two years prior to each month. The number of long-term conditions was calculated based on the logic of the Cambridge Multimorbidity Score48 applied to a set of Quality Outcomes Framework conditions.49 Continuous/interval covariates were mean-centred and scaled to an SD of 1, to aid interpretation of model coefficients.
Practice-level covariates
We also incorporated a variety of practice-level covariates. These included the Strategic Health Authority region50 and whether the practice area was rural or urban51 and several derived variables, which were calculated from the complete CPRD Aurum data for the eligible practices and aimed to measure GP supply and patient demand at the practice level. These included: the number of unique GPs that had given consultations during the period (‘number of GPs’); the number of unique patients that had received consultations during the period (‘number of patients’); the mean number of days GPs provided consultations (ie, mean days per GP) during the period (‘GP days’); the average number of consultations per patient during the period (‘patient consultations’) and the average number of patient consultations per GP day during the period (‘consultations per GP day’). We calculated these derived variables using all data in the CPRD Aurum dataset for the selected practices (ie, not using a sample of patients). Continuous/interval covariates were mean-centred and scaled to an SD of 1, to aid interpretation of model coefficients.
Analytical approach
A conceptual model52 was developed to understand the relationship between consultation mode and continuity of care and identify potential confounding variables and effect modifiers (online supplemental figure 3). Descriptive analyses were used to characterise the samples used in the analyses, conduct preliminary investigations of the RCI and outcome measures, and visualise changes in these measures over time. More detail on the conceptual model and descriptive analyses is available in supplemental material.
Cross-sectional regression models were estimated to identify the between-patient association of RCI with the outcomes both before and during the pandemic in selected index months of February 2020 and February 2021. Longitudinal regression models were estimated to identify the within-patient association and test for time-specific confounding. In the cross-sectional models, a range of control variables based on the conceptual model were included in stages to identify and reduce confounding. In the longitudinal models, four model specifications were used to test for the effect of changes in consultation mode, period effects and their interaction, while controlling for patient-level fixed effects.
For the UPCI and BBI, we used logistic regression. For the count of consultations with the most frequently seen GP, we used Poisson regression with total consultations as an offset/exposure term and a zero-truncated distribution (because the most frequently seen GP must be seen at least once). We also tested for overdispersion using a negative binomial model.53 For the UPCI and BBI, resulting coefficients were presented as odds ratios (OR); for the count of consultations, incidence rate ratios (IRR) were presented.
The cross-sectional regression models included a practice-level random-intercept to account for unmeasured practice-level heterogeneity.54 The longitudinal models included only time-varying covariates and used both patient-level fixed effects55 to account for time-invariant, unmeasured patient-level heterogeneity, and (in all but the first specification) a period fixed effect56 to also account for time-specific confounding. Model fit was compared using the Akaike information criterion.57 Diagnostic plots and variance inflation factors (VIFs)58 were used to identify any issues with the fit of the preferred models. All analyses were undertaken using R V.4.0.359 and RStudio V.1.4.1103.60 More detail on the analytical approach is available in supplemental material.
Information governance
All data were stored and analysed in the Health Foundation Secure Data Environment (SDE). The SDE is an ISO27001-compliant,61 non-networked data environment, restricted to authorised individuals. The evaluation team had no access to identifiable patient data. All outputs transferred out of the SDE were cleared in accordance with statistical disclosure control procedures.62
Patient and public involvement
Patients and/or the public were not involved in the design, conduct, reporting, or dissemination plans of this research.
Results
In every month between April 2018 and April 2021, at least 100 000 patients were included in the descriptive analyses of trends. For the cross-sectional modelling, 153 475 patients from 389 practices were included in the February 2020 index month (before the pandemic), and 125 298 patients from 388 practices were included in the February 2021 index month (during the pandemic). The longitudinal sample included 76 281 patients from 388 practices for whom the required measures could be calculated at both February 2020 and February 2021 index months.
Descriptive analysis
Before the pandemic, continuity of care (as measured by the mean UPCI) had dropped by approximately 2 percentage points between April 2018 and February 2020 (figure 1A). During the pandemic, continuity of care remained stable increasing slightly from 0.51 (SD=0.22) in February 2020 to 0.52 (SD=0.22) in February 2021 (online supplemental table 1). There were demographic, socioeconomic and regional differences in continuity of care (figure 1B–F and online supplemental figure 4). These differences did not vary notably over the study period except in relation to age where differences between age groups narrowed considerably (figure 1B) during the pandemic. Similar trends were seen for continuity of care as measured by BBI (online supplemental table 1 and online supplemental figures 5-11). See supplemental material for a full discussion of trends.
Time series of mean usual provider of care index (%) by month and selected patient characteristics: (A) all patients; (B) by age group; (C) by gender; (D) by Index of Multiple Deprivation (IMD) quintile (where 1=least deprived and 5=most deprived); (E) by number of long-term conditions (LTCs); (F) by number of consultations. Shaded areas indicate 95% confidence intervals. QOF, Quality Outcomes Framework.
The average proportion of consultations delivered remotely was steadily increasing prior to the pandemic (online supplemental figure 12) but then increased dramatically from less than a quarter of all consultations (0.23, SD=0.24) just before the pandemic to more than three quarters (0.76, SD=0.23) during the pandemic (online supplemental table 1). There were also demographic, socioeconomic and regional differences in the average proportion of consultations delivered remotely. For example, the use of remote consultations was greater for patients who had more consultations in total (online supplemental figures 13–18). During the pandemic, these differences narrowed dramatically for all patient groups.
Patients consulting during the pandemic (index month February 2021) tended to have poorer health on average than patients consulting before the pandemic (index month February 2020). On average, they had more previous referrals (0.60 mean previous referrals in February 2021 compared with 0.52 in February 2020), greater multimorbidity (1.04 vs 0.97 conditions) and more consultations in total (6.63 vs 6.32, online supplemental table 1). There were no notable sociodemographic or regional differences in the types of patients consulting before and during the pandemic (online supplemental table 2).
There were also differences at the practice level between these two periods (online supplemental table 3 and online supplemental table 4). Fewer patients consulted their GP during the pandemic (5792 vs 6838), but they consulted slightly more often on average (3.95 vs 3.90). There were also fewer consultations per GP day during the pandemic (16.70 vs 19.70). In terms of supply, fewer GPs were working during the pandemic (25.25 vs 27.10), but they were working more days on average (64.47 vs 59.64).
Figure 2 shows histograms of the RCI, UPCI and BBI, overlaid at February 2020 and February 2021 index months. The difference in the distribution of the RCI between the two periods is clearly apparent. In contrast, there was little change between the two periods in the distributions of the UPCI and BBI.
Histograms of the remote consultation, usual provider of care and Bice-Boxerman indices by index month.
Scatter plots of the UPCI and BBI by RCI for each of the index months show many mass points where large numbers of patients had the same values (online supplemental figure 19). This was due to most of the included patients having between 3 and 5 consultations in total per year, from which the measures were calculated. Ordinary least squares lines fitted to the plotted data were nearly horizontal, indicating little in the way of a bivariate association with the RCI. Before the pandemic (index month February 2020), there was a slight negative association, with higher RCI values associated with slightly lower continuity. Generalised additive model fitted lines showed a greater deviation from the horizontal at high RCI values before the pandemic and low values during the pandemic (index month February 2021); however, these regions of the plots contained far lower numbers of patients.
Cross-sectional modelling
Figure 3 shows estimates and 95% confidence interals (CI) for the best fitting models. Before the pandemic, there was no significant between-patient association of the RCI with either the UPCI (OR 95% CI 0.98 to 1.01) or the BBI (OR 95% CI 0.98 to 1.02). During the pandemic, there was a small negative association with both the UPCI (OR 95% CI 0.95 to 0.99) and BBI (OR 95% CI 0.94 to 0.98). For the count model (UPC consultations), a very small negative association was detected in both years (February 2020: IRR 95% CI 0.97 to 0.99; February 2021: IRR 95% CI 0.98 to 0.99).
Point estimates and 95% confidence intervals of exponentiated model coefficients for the preferred cross-sectional models by outcome measure and index month. Odds ratios are presented for the UPC index and Bice-Boxerman index. Incidence rate ratios are presented for the UPC consultations. The reference category for ethnicity is white. GPs, general practitioners; IMD, Index of Multiple Deprivation; UPC, Usual Provider of Care. Note: 'UPC consultations' is the count of consultations with the most frequently seen GP, modelled using a Poisson model.
At the patient level, age, gender, ethnicity and multimorbidity were significantly associated with continuity of care, with similar trends in both February 2020 and February 2021. Older patients generally experienced higher levels of continuity, whereas female or black patients and those with more comorbidities experienced slightly lower levels of continuity.
However, continuity was most strongly associated with practice-level characteristics. In terms of demand, there was a small negative association of mean patient consultations with continuity for the UPCI and BBI, and the number of patients receiving consultations at the practice was also negatively associated with continuity. However, a positive interaction between GP and patient numbers suggested that this effect could be ameliorated somewhat by higher supply of GPs. The mean number of days GPs worked and consultations per GP day were positively associated with continuity.
Diagnostic plots highlighted some issues with the Pearson residuals for both the UPCI and BBI at the patient level, reflecting the mass points in these measures (online supplemental figure 19). In contrast, the residuals for the count models were as expected and closely followed the theoretical distribution. VIFs did not suggest any problematic collinearity in any of the models estimated.
Longitudinal modelling
Figure 4 shows the main estimates for the four within-patient, fixed effects models. In the first model specification, which did not incorporate the period effect, there was a very small, positive association of RCI with each of the outcomes (OR 1.016, 1.026 and IRR 1.005, for the UPCI, BBI and count model, respectively). However, larger effects were estimated when the RCI was substituted with a dummy indicator for the pandemic period in the second model (OR 1.075, 1.098 and IRR 1.032). In the third model, both fixed effects were included. The association with the dummy indicator remained (OR 1.076, 1.087 and IRR 1.038), whereas the effect for RCI was no longer significant. An interaction between the RCI and dummy indicator was tested in the fourth model, but was not significant, and so the third model was preferred. These results provide evidence that any within-patient change in continuity was due to period effects and not a direct result of changes in the use of remote consultations.
Point estimates and 95% confidence intervals of the main exponentiated model coefficients for the fixed effects models, by outcome measure. Odds ratios are presented for the UPC index and Bice-Boxerman index. Incidence rate ratios are presented for the UPC consultations. UPC, Usual Provider of Care. Note: 'UPC consultations' is the count of consultations with the most frequently seen GP, modelled using a Poisson model.
Figure 5 shows all estimates for the preferred model specification for each outcome. There were many similarities in the covariate estimates for each of the outcomes. Total consultations, multimorbidity, the number of patients, and mean patient consultations at the practice were negatively associated with continuity. However, an interaction suggested that the negative association between the number of patients and continuity was ameliorated as the number of practice GPs increased. The mean number of consultations per GP day was positively associated with continuity. Diagnostic plots and VIFs did not identify any problems relating to model fit or collinearity. More detailed results are available in supplemental material.
Point estimates and 95% confidence intervals of the exponentiated model coefficients for the preferred fixed effects models, by outcome measure. Odds ratios are presented for the UPC index and Bice-Boxerman index. Incidence rate ratios are presented for the UPC consultations. GP, general practitioner; UPC, Usual Provider of Care. Note: 'UPC consultations' is the count of consultations with the most frequently seen GP, modelled using a Poisson model.
Discussion
Statement of findings
We present the first comprehensive analysis of trends in continuity of care in general practice, from before to during the pandemic. We also robustly identify the association between continuity of care and consultation mode, using the pandemic as an exogenous shock. Like others, we found that the use of remote consultations in English general practice dramatically increased during the pandemic.31 Relational continuity of care declined steadily between 2012 and 2017.22 We found that this trend continued until the start of the pandemic. However, during the pandemic, we found that practices were able to maintain continuity of care and actively prioritise it for those groups who likely needed it most, despite the challenges at that time.
We also observed other differences in general practice activity during the pandemic compared with before. Fewer GPs were working more days but delivering fewer consultations each day to fewer patients. Also, patients receiving care in February 2021 had poorer health on average than patients receiving care in February 2020. As well as demand being lower during the pandemic (as patients avoided or postponed seeking care during lockdown) these results are also consistent with reports suggesting that organisational changes during the pandemic, such as stricter triage, greater use of remote consultations, and the expansion of support services, freed up GPs' time, allowing them to maintain a focus and spend more time on older patients and those identified as high risk.31 63
We found little evidence that consultation mode had a meaningful impact on continuity. Between-patient regression models identified small, negative associations of remote consultation with continuity during the pandemic. Longitudinal, within-patient models suggested that differences in continuity experienced by patients over the 2 years was related to the effect of the pandemic rather than directly due to the change in the use of remote consultations. Given the dramatic increase in remote consultations during the pandemic, these results are consistent with other reports suggesting that practices can find ways to organise the balance of needs for remote and face to face consultation needs in ways that preserve continuity.64
We also explored the impact of other patient-level and practice-level factors on continuity. Patient age, gender, deprivation, ethnicity and comorbidity all impacted continuity of care. Our results suggest that when interpreting differences in continuity of care between practices or over time, care should be taken to account for any underlying differences in patient populations which may vary between practices or periods.
Importantly, we found that measures of GP supply and demand had the strongest influences on continuity. These results suggest that if practices do not have sufficient GPs consistently available with the capacity to provide consultations when their patients request them, it is less likely that a usual GP will be available. Thus, a plausible explanation for the slight increase in continuity seen during the pandemic is that the reduction in demand for consultations during this period permitted greater freedom for patients to be aligned to their usual GPs. More work is required to look at other factors which are not measured here but which might also have contributed to improved continuity, including enhanced navigation and triage and the expansion of support services and multidisciplinary team working.
Strengths and weaknesses
We used a large-scale, routinely collected, nationally representative patient-level data set. We used a novel method to derive consultation mode, which combines information from the time of booking with observations made during the consultation, to improve the accuracy of our results. We developed and populated a conceptual model that identified a variety of potential influences on continuity. We compared the association between continuity of care and a variety of practice-level and patient-level factors, both before and during the COVID-19 pandemic, using three different measures of relational continuity of care and two different modelling approaches. All approaches yielded comparable results, suggesting that our findings are robust.
We were only able to focus on relational continuity of care with a GP and could not account for changes in management or informational continuity. Data quality may affect our results to some extent, as consultation mode is not consistently recorded in the data, and our method for deriving consultation mode has not been externally validated. However, the increase that we observed in the proportion of remote consultations during the pandemic is consistent with other sources, giving us confidence in our methods. Although we included a wide range of factors in our models, there may still be unobserved confounding relating to variables that we could not measure (eg, differences in practice operating models). Our model specifications may be overly simple, relying predominantly on main effects; the conceptual model contained indirect paths and particular concepts that are inherently complex. However, we decided to err on the side of interpretability, mostly avoiding higher order model terms and interactions.
Comparison with previous studies
The specific impact of remote consulting approaches on continuity has rarely been addressed.33 This is the first study to explicitly examine the association between consultation mode and relational continuity in English general practice. Previous studies looking at the impact of remote consulting have mainly focused on outcomes affecting primary care access, including patient and staff experience64–67 and workload,31 68 69 as well as health inequalities64 69 70 and quality of care in relation to clinical risks,71 doctor–patient communication64 and personalised care.72
Implications
Mode of consultation did not have a significant impact on continuity levels. However, consultation mode is just one element of a practice operating model. Future work should examine which features of a practice operating model can improve continuity despite a capacity constrained environment. This particularly applies to practices operating models with advanced care navigation and triage processes; specifically, those that enhance collecting information at the point of contact which can be used to determine continuity needs, and enable practices to direct the patient to the right person.
Issues of demand and capacity were the most important drivers of continuity of care. If policy-makers want to improve continuity for patients, primary care capacity needs to increase. This requires sufficient, long-term investment in clinicians, staff, facilities and digital infrastructure. Alongside this, as demand continues to rise, general practice also needs to transform ways of working to ensure continuity for those that need it.
Future research
The parameterisation of practice-level covariates could be developed further to better understand how supply and demand can influence continuity. We would recommend future research on continuity make more use of count models, rather than solely modelling indices like the UPCI and BBI.
More generally, the adoption of modern general practice models with an explicit focus on enhancing continuity is an area that is under-researched. Quasi-experiments could compare practices implementing new general practice access and care navigation models that are continuity-focused with practices that are not explicitly aiming to manage and enhance continuity. By doing so, it may be possible to identify whether substantive changes in continuity are possible without increasing capacity in general practice.
Data availability statement
Data may be obtained from a third party and are not publicly available. The data analysed in this study are available for approved research studies from Clinical Practice Research Datalink (CPRD).
Ethics statements
Patient consent for publication
Ethics approval
Research ethics:The study protocol73 was approved through the CPRD’s Research Data Governance Process (study number 21_000505). Formal ethical approval was not sought, as the study was based on retrospective analysis of existing pseudonymised administrative data for the purpose of identifying the population-level association between remote consultation and continuity.
Acknowledgments
The authors would like to thank Rebecca Fisher for feedback on the coding of CPRD Aurum consultation data and Stefano Conti for review of our statistical analysis plan. We also thank Therese Lloyd, Kaat de Corte, Jake Beech from the Health Foundation and Minal Bakahi and Jean Ledger from NHS England, for their helpful and insightful comments.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
Contributors GC designed the original study protocol. EC created the initial sample from CPRD Aurum. EV assisted in developing the consultations coding method. JH prepared the practice-level variables. CF undertook initial, descriptive, time-series analyses. WP undertook all analysis presented in this article and prepared the manuscript. GC was study guarantor and confirms that this paper is an accurate representation of the study findings. The authors were solely responsible for the decision to submit this manuscript for publication.
Funding The study was funded by The Health Foundation. Registered charity number 286967. Grant number N/A.
Competing interests None declared.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.