Drivers of primary care appointment volumes before and after the COVID-19 pandemic: a longitudinal study

Abstract

Background

The COVID-19 pandemic led to significant reconfiguration of primary care systems internationally. For example, a large proportion of face-to-face appointments have been replaced by telephone or online consultations. As a result, the relationship between primary care appointment provision and some of its key determinants, such as workforce and demographic characteristics, are likely to have changed. Conclusions from previous studies which used only pre-pandemic data may no longer be applicable under these new configurations.

Methods

This study aims to investigate whether the relationships between primary care appointment rates and their determinants, including workforce composition and the age structure of registered patients, have changed after the COVID-19 pandemic. We conducted longitudinal analysis on the 106 Clinical Commissioning Groups (CCGs) in England over two periods of 24 months (March 2018 to February 2020 and March 2021 to February 2023). We used fixed-effects regression models to relate monthly general practice appointment rates per registered patient to workforce size and composition and population age structure, and compared the results between the two periods.

Results

In the pre-pandemic period, changes in full time equivalent (FTE) numbers of GP trainees and nurses were the only time-varying variables associated with appointment rates. In the post-lockdown period, the age profile of registered patients became a key determinant of appointment rates. A 1% increase in the proportion of registered patients over 80 years was associated with a 0.165 (38.7%) increase in appointments per patient per month. Changes in FTE numbers of qualified GPs and direct patient care staff were not found to influence the appointment rate in either period.

Conclusion

The relationships between primary care appointment rates and the workforce composition and population age structure have changed following the COVID-19 pandemic. The proportion of registered patients over the age of 80 years is now the most significant time-varying driver of appointment rates. General practices serving older patients may face much higher demand and have a bigger challenge providing sufficient appointments in the future compared to before the pandemic.

Introduction

The COVID-19 pandemic caused severe interruption to the provision of primary care services worldwide, and has led to major changes and reconfiguration of primary care systems [1]. One major change has been the increased use of telephone consultations [2], accompanied by skill-mix expansions within the workforce. Recent data from England shows that the volume of telephone consultations in February 2023 had increased by 105% compared to February 2020, and appointments with a non-GP practice staff surpassed those with general practitioners (GPs) for the first time in late 2022 [3]. Although the volume of appointments provided in England has recovered to its pre-pandemic level as of March 2021, many of the changes such as the sudden increase in the use of telemedicine [4], improved patient access to electronic health records [5], and shifted perceptions of primary care staffs’ professional roles [6], are likely to represent permanent shifts [7].

The volume of appointments delivered is determined by both supply and demand factors, with the gap between the supply of, and demand for, primary care services in England widening in recent years [8, 9]. Previous research has studied the workforce determinants of primary care activity and a range of outcomes, finding that skill-mix changes can increase capacity but have negative effects on patient satisfaction [10, 11], and higher levels of nurses and other direct patient care practitioners (excluding GPs and nurses) to be associated with worse patient reported access to primary care [12]. The lack of access to primary care has been further worsened by the pandemic, which had long-term detrimental impacts on the mental and physical health of patients, increasing demand-side pressures even further [13–15]. However, given the effect of the pandemic on consultation styles, long-term care provision, and healthcare staff [1], results obtained using pre-pandemic data may no longer be generalizable to post-lockdown service provision.

Although the number of appointments was severely affected in the short-term by lockdowns and other restrictions, technological and organizational advancements driven by the pandemic may have enabled an expansion of the supply of primary care services in the long run. On the other hand, the need and demand for care have also increased as a result of the pandemic and the aging population. The overall effects on access to services will depend upon which of these effects is larger. While practice level appointment rates also depend on a wide range of population and organizational characteristics, including level of deprivation and practice rurality, many of these drivers are constant or only change very slowly over time [16]. We therefore focus on the time-varying drivers of appointment volumes, as these are the characteristics which will drive changes in primary care workload and access in the short-term. This paper examines the pre- and post-lockdown effects of the two key time-varying determinants of appointment rates, i.e., workforce and age structure, and compares how these relationships have changed.

Methods

Data

We obtained data on the number of attended appointments per month in England from the Appointments in General Practice data series published by NHS Digital, from March 2018 to February 2023 [3]. The data series contains daily counts of appointments (excluding COVID vaccinations) collected from GP systems at practice level, which can be broken down by staff type (GP or other practice staff), and four consultation modes (face-to-face, telephone, home visit, and video conference/online). The data covers 90.1% of all practices and patients in England as of March 2018 and the coverage has been consistently over 99% since September 2020 [16]. The data are published at the level of Clinical Commissioning Group (CCGs), which, until July 2022, were clinically led statutory NHS bodies responsible for local health service planning and commissioning. CCGs were based on geographical groups of practices, responsible for populations of approximately a quarter of a million people on average [17, 18].

Although CCGs have since been replaced by integrated care boards (ICBs), appointments data are still published at this level under the new name of “sub-ICB locations”. As data is not available at the level of ICBs before November 2022 nor at the level of general practices before October 2022, we conduct this longitudinal study using CCG-level data. As the commissioning reform took place towards the end of our study period, we continue to use the name CCGs throughout. There were originally over 200 CCGs, which were gradually merged into 106 organisations. To ensure that the units we examine are comparable over time, we aggregated data from the merged organizations to obtain a balanced panel of 106 CCGs, defined in terms of the CCG membership configurations as of June 2022.

Data on the following population and workforce characteristics of each CCG were also obtained from the NHS Digital General Practice Workforce data series: full-time-equivalent (FTE) numbers of practice staff by type: GP (qualified), GP trainees, nurse, and all other direct patient care (DPC) staff, number of registered patients, numbers of patients aged under 5, and number of patients aged over 80 [19]. These age groups represent the youngest and oldest 5% of the general population, respectively. The workforce data was only available quarterly in the pre-pandemic period, and so the same numbers are used for all three months of each quarter. All other variables in the first period and all in the second period are monthly data. Direct patient care staff include a wide range of roles, such as healthcare assistants, dispensers, paramedics, and pharmacists. All data used are publicly available on the NHS Digital website [3, 19, 20].

Statistical analysis

We began by splitting the data series into two periods of two years each. The 24 months before the first UK lockdown in March 2020 is defined as the pre-pandemic period i.e., March 2018 to February 2020. The 24 months from March 2021 to February 2023 is defined as the post-lockdown period, the start of which was marked by the publication of government guidance “Road out of lockdown” [21]. The two periods are chosen to be of equal length and contain the same calendar months in order to keep any influences of seasonality in appointment volumes the same in both periods. Given this constraint, 24 months is the maximum length of data that is available for both periods. March 2020 to February 2021 is not included in either period to avoid the impact of severe lockdown restrictions on primary care provision and use.

At CCG-level, we regress the monthly appointment rate per registered patient on the FTE numbers of four types of practice staff per 10,000 registered patients, and the proportions of registered patients that are under 5 and over 80. The appointment rate and workforce inputs are both measured per capita, though we have re-scaled the workforce measures per 10,000 patients as opposed to per patient in order to present the coefficients on a more meaningful scale for interpretation. A fixed-effects linear panel regression model was used to estimate the marginal effects of the explanatory variables on the appointment rate [22]. The fixed-effects model adjusts for bias caused by all unmeasured factors that vary across CCGs and are stable over time, such as any organizational, cultural, geographical, or socio-economic differences between places. We also included monthly dummy variables to account for the strong seasonality in primary care activities. In a supplementary analysis, we conducted the same analysis for appointments with GPs and appointments with other practice staff separately.

The effect of workforce numbers on appointment rates may not be linearly related to the size of the CCG, because there could be economies of scale from cost-saving benefits of CCG mergers and expansion, or there could be diminishing marginal returns as the organisation gets “too big”. Therefore, we created a constant-elasticity model by log-transforming all of the explanatory variables except the proportions of patients under 5 and of over 80, which allows us to interpret the coefficients as percentage effect of the explanatory variables on the appointment rate [23]. The age proportions were scaled up by a factor of 100 to represent percentage points, and are interpreted as the effect of a one percentage point increase in the proportion of patients in a respective age group on the CCG appointment rate.

Formally, our model is specified as follows:

$${y}_{it}=\varvec{\beta}{\mathbf{Staff}}_{it}+\varvec{\gamma}{\mathbf{Age}}_{it}+\varvec{\tau}{\mathbf{Month}}_{t}+{u}_{i}+{\epsilon}_{it}$$

Where $\:{y}_{it}$ is the monthly appointment rate per registered patient of CCG $\:i$ in month $\:t$; $\:{\mathbf{Staff}}_{it}$ represents the log-transformed FTE numbers of the four categories of clinical staff per 10,000 patients of CCG $\:i$ at time $\:t$; $\:{\mathbf{Age}}_{it}$ includes the proportions of patients in the two high-demand age bands (under 5 and over 80 years); $\:{\mathbf{Month}}_{t}$ are 11 month dummy variables, using March as the baseline; $\:{u}_{i}$ is the time-invariant CCG fixed effect and ${\epsilon}_{it}$ is the error term.

We estimated Driscoll & Kraay standard errors using the ‘xtscc’ command in Stata. We used these for all statistical inferences to ensure they are robust to serial correlation and cross-sectional dependence [24, 25]. We measured the goodness-of-fit of the models by the “within” R squared statistic. We used Stata (version 17.0) to conduct all analyses.

Results

Descriptive statistics

Table 1 presents the descriptive statistics of CCG appointment rates, workforce, and the age structure of registered patients in the first and last months of the pre-pandemic and post-lockdown periods. The appointment rate was slightly lower at the end of each period than at the start, likely due to seasonal patterns, but increased between the pre-pandemic and post-lockdown periods. The overall staffing level has also increased over the five-year period, driven by a shift in skill-mix. There were more GP trainees and other DPC staff by the end of the second period, but fewer GPs and a similar number of nurses. The proportion of registered patients under the age of 5 has been decreasing while the proportion of older patients has been increasing steadily. Meanwhile, there is substantial variation in the age structure of CCG populations. Some CCGs have as high as 28.14% of total patients over the age of 65, while for some CCGs this figure is only 9.27%.

The variations in CCG level workforce by staff type are shown in Fig. 1. There are large variations in the composition of different types of staff among CCGs and number of staff per 10,000 patients, so the variations shown in the boxplots are not merely results of differences in CCG size. Also, given the variations in skill-mix composition, CCGs with high levels of a certain type of staff do not necessarily have a high overall level of staffing.

Table 1 Descriptive statistics of CCG-level appointment rates and explanatory variables in the first and last months of the two periods

Fig. 1

Numbers of four types of practice clinical staff employed in CCGs in March 2018 and March 2023

Figure 2 shows the number of attended appointments in England with either type of staff (GP or Other practice staff) and the mode of consultation (face-to-face vs. telephone). The total number of attended appointments has experienced large fluctuations during the pandemic, dropping 32.2% from 22.7 million in March 2020 to 15.4 million in the following month, recovering to 26.6 million in October 2020, and dropping again to 21.3 million in early 2021, before finally getting back to its pre-pandemic level. There has been a clear shift towards providing more telephone consultations since the pandemic. Although the volume of face-to-face appointments with other practice staff has now recovered to its pre-pandemic level, volumes of GP face-to-face appointments have not. Figure 3 compares the trends in appointment volumes with GPs to those with other practice staff in the pre- and post-lockdown periods. In the pre-pandemic period, the two series show parallel upward trends over time. However, in the post-lockdown period, GP appointments are no longer increasing but instead follow a stable flat trend over time. Conversely, the volume of appointments with other practice staff follows a much steeper upward trendline than before the pandemic. Towards the end of our study period, the volume of appointments with other practice staff surpasses those delivered by GPs for the first time.

Fig. 2

Appointment volume in England by staff type and mode (March 2018-February 2023). Note: practices using Cegedim or Informatica systems are not included, as these two systems do not record this information [26]

Fig. 3

Pre- and post-lockdown appointment volumes by staff type

Regression analysis

The regression results modelling the effect of workforce and population age structure on CCG-level appointment rates are presented in Table 2. The coefficients and the significance levels of several explanatory variables are different in the two periods, suggesting that the factors driving appointment rates have changed over time. Before the COVID-19 pandemic, the only explanatory variables found to drive appointment rates significantly were the numbers of FTE GP trainees and nurses. The estimated coefficient for the number of FTE GP trainees is 0.008, indicating that a 1% increase in the number of FTE GP trainees per 10,000 patients is associated with a 0.008 increase in the number of appointments per capita. The number of FTE nurses also had a significant impact, with an estimated coefficient of 0.047, meaning that a 1% increase in the number of FTE nurses per 10,000 patients corresponds to a 0.047 increase in the appointment rate per capita. The age structure of registered patients was not found to be associated with the CCG appointment rate in the pre-pandemic period, suggesting that workforce numbers were the primary driver of appointment volumes. The model has a within $\:{R}^{2}$ of 0.689.

In the post-lockdown period, the number of FTE nurses was again found to be associated with significantly higher appointment volumes, to a similar mangitude as before the pandemic. GP trainees, however, appear to have become a more important driver of appointment rates after the lockdown period. The proportion of registered patients over the age of 80 now becomes the most significant contributor to appointment rates, both in terms of the magnitude of the effect and statistical significance. A one percentage point increase in the proportion of patients aged 80+ is estimated to increase the appointment rate per patient per month by 0.164 appointments. Given that the average rate in the post-lockdown period was 0.426 appointments per person per month, the marginal effect of one extra percentage point of patients over 80 is a 38.7% increase in the overall appointment rate. The percentage of patients aged under 5 is found to have a significant negative effect on the appointment rate, although the magnitude of this effect is very small. The within $\:{R}^{2}$ of the model is 0.797, higher than in the pre-pandemic model and indicating that variations in appointment rates have become more predictable.

The number of FTE GPs and other direct patient care staff were not found to be significantly associated with the appointment rate in either period.

Table 2 Regressions of appointment rates on workforce numbers and population age structure in the pre-pandemic and post-lockdown periods

In supplementary analysis (Table A1), we examined the determinants of appointments with GPs and appointments with other healthcare professionals separately. These results suggest that numbers of GPs and Direct Care Professionals do not increase overall appointment rates in our main analyses because their activity acts as substitutes with appointments delivered by other staff types. The increased effect of the population aged 80 years and over post-lockdown is apparent in both appointments with GPs and appointments with other healthcare professionals, indicating that an aging population will place significantly increased demands on general practices irrespective of their workforce compositions.

Discussion

Summary

Using longitudinal data, this paper examines whether the effects of changes in workforce size and composition and the age structure of registered patients on general practice appointment rates in England has changed as a result of the COVID-19 pandemic. From March 2018 to February 2023, we observed large increases in the number of GP trainees and other DPC staff, the latter of which is a direct result of the Additional Role Reimbursement Scheme introduced in 2019 to encourage the expansion of primary care workforce [27]. The age structure of registered patients has also shifted, reflecting the low birth rate during the pandemic and the surge in adult patient registrations since the pandemic: total registered patients has increased by almost two million since March 2020 while the number of patients under 5 has decreased [28]. This is likely caused by previously unregistered individuals registering to ensure they could receive COVID-19 vaccination, as months of largest increases in registered patients coincide with vaccination rollout [29, 30].

We find that the key determinants of general practice appointment rates have indeed changed, potentially due to the reconfiguration of primary care systems during the pandemic such as the shift towards telephone consultations and changes in workforce skill-mix. In the pre-pandemic model, the population weighted FTE number of nurses was the main contributor to CCG appointment rates, with the number of FTE GP trainees also associated with higher appointment rates. The marginal effect of GP trainees on appointment rates increased following the lockdown period, indicating that this growing staff group has become an increasingly important driver of appointment provision in recent years. This is likely somewhat explained by the relatively larger increase in trainee GPs in the early periods of our analyses, meaning that they, on average, would have been more experienced during the post-lockdown period we examine. Since August 2021, the time trainees spend in general practice has been increased from 18 months to 24 months, which also makes trainees in the second period more productive on average [31].

These findings also point to a wider change in skill-mix, as the traditional qualified GP and Nurse workforce are unable to meet increasing population demands for appointments. Initiatives such as the Quality and Outcomes Framework (QOF) incentivise the provision of care for patients with chronic diseases, often requiring additional appointments for monitoring. The introduction of the QOF enhanced the role of nurses in the management of long-term conditions, but is felt to have decreased continuity of care [32]. Changes to the design of schemes such as the QOF have a direct impact on the skill-mix of staff required in general practice, and therefore also contribute to the evolving nature of service provision, impacting the drivers of appointment volumes.

The proportions of young children or patients aged over 80 were not found to have significant effect on appointment rates before the pandemic. However, in the post-lockdown period, the proportion of patients aged 80+ was found to be the biggest driver of appointment rates of all variables we examined. A one percentage point increase in patients in this age group was associated with a 0.165 increase in appointments per registered patient per month, equivalent to 38.7% of the average appointment rate in the post-lockdown period. Some of this may be a result of long COVID, as older patients are more likely to experience both mental and physical consequences of COVID [33]. However, the magnitude of the increase suggests that it reflects a more general increase in the workload associated with an aging population over time.

Comparison with existing literature

While there is rich literature on general practice workforce, previous research linking this to outcomes of importance to staff, patients, and system outcomes has only been able to examine self-reported proxy measures of access [10, 34]. We instead examine appointment volumes, and find evidence that older patients’ have become a much stronger driver of appointment rates since the pandemic. The burden of long-term post-COVID-19 syndrome on healthcare systems is becoming a major concern as predicted by some earlier research [15, 35]. Results from our supplementary analysis (Table A1) also confirm that there are substitution effects between different types of clinical staff in appointment provision, as has been found previously [16]. This means that an increase in workforce numbers in one staff group, such as GPs, may not result in an increase in the overall volumes of appointments delivered. It should however be noted that these results are associations, and therefore not necessarily causal effects.

It is well known that increases in supply and/or reductions in access costs lead to increased use of healthcare [36]. The introduction of online booking systems and increased availability of telephone appointments introduced into primary care at pace during the COVID-19 pandemic represent one such reduction in access costs. These may explain some of the increase in appointment rates that we observe during the period we examine. However, patient reported satisfaction with access and experience of making an appointment has declined over the same period, suggesting that these changes to supply have not been sufficient to satisfy demand [37].

Strength and limitations

The study uses new data on appointment volumes, which is collected and updated monthly by NHS Digital. The data series include details on general practice appointments such as staff type and mode and provides very good coverages of patients and practices in England as all releases used in this study cover 90% or higher of all registered patients in the nation [3]. While most studies on primary care access and provision use cross-sectional data, we used a panel time-series dataset, which better captures the effects of time-varying factors as the model is not prone to biases caused by unmeasured time-invariant influences.

However, the data suffers from several issues, partly due to its new experimental nature. First, since there are no national standards for the recording of appointments, the data accuracy is subject to the practice standard. Whilst there may be systematic differences in recording standards between different practices and CCGs, these are unlikely to change significantly over time and would therefore be accounted for by the CCG fixed effects included in our analyses. However, due to known inconsistencies in the definition of appointment modes (i.e., face-to-face, telephone, home visit, and video conference/online) by different GP system providers, a nontrivial proportion of appointments were recorded with “unknown” mode [26]. We were therefore unable to examine whether the identified determinants of overall appointment volumes impact some modes of appointments more than others.

Second, the data only covers “scheduled and planned activity recorded on the GP appointment system”, while the other activity and interactions with patients were not reflected [26]. The level of activity recorded therefore represents the lower bound of true CCG activity levels, as for example, emergency and out-of-hour appointments were not always well captured. Third, whilst data on the number of FTE staff employed are broken down into four categories, the appointment data is only categorized as “appointments with a GP” and “appointments with other practice staff”. This prevents us from analysing the specific effects of each group of staff on their respective appointment rates, with the exception of GPs where this data is available.

All other data quality issues listed by NHS Digital were also evaluated and found to be irrelevant to our analyses, either because they only affect later periods of the data that we do not examine (e.g. addition of Primary Care Network data from June 2023) or because they affect data fields that we do not use (e.g. number of unattended appointments, national category mapping, duration of appointment, etc.) [26].

We use fixed effects regression models to account for influences on appointment rates which are stable over time. This approach has the advantage that it enables us to control for all determinants of appointment volumes which vary across CCGs but are constant over time, including determinants for which there are no data. However, it means that we cannot identify individual contributions of these time-invariant factors (such as rurality and deprivation) because they are all combined in the estimated fixed effects.

Implications for research, policy, and practice

Once a long period of practice-level appointment data becomes available, future research could examine the time-varying determinants of appointment rates at the practice-level to investigate whether the post-lockdown relationships we detect are evident at the level of individual general practices. Alternatively, future analyses could estimate the impact of time-invariant variables, such as a CCG’s size and rurality, using a mixed effects model. This would improve the overall understanding of the determinants of GP activity and output, and the relative workload pressures caused by population factors which do and do not change year on year. Researchers should also explore whether the relationships between the explanatory variables and other outcomes of importance to patients, such as continuity of care, have changed, as observed in this study.

Our results show that higher proportion of older patients is the key time-varying driver of appointment rates in the post-lockdown period. The magnitude of this effect is informative for the reconfiguration of primary care services, projections of practice workload, and practice funding allocation. General practices serving older patients are likely to face much higher demand and have a bigger challenge providing sufficient appointments in the future compared to before the pandemic. While the NHS plan to move towards digital-first primary care should facilitate fast and easy access to healthcare services, it is vital to take into consideration older patients’ preference for face-to-face appointments over remote contacts [38, 39]. Practices serving older populations may need to provide additional support and tailor their reconfiguration process to the group driving the largest increases in demand, to ensure equitable access for patients of all ages and to improve patient satisfaction with non-face-to-face appointments.