Independent variables
Selection of potential independent variables was driven by two conceptual frameworks: the SEARCH framework, in which population factors are important determinants of health outcomes,12 and a framework describing the mechanisms of primary care that influence population mortality, developed using evidence from low-, middle-, and high-income countries.13 A landmark review in 2005 described six mechanisms of primary care accounting for its beneficial effects on population health.10 The authors of the current study’s new framework categorises 23 mechanisms into five groups:
Comprehensiveness, the lifelong care of all individuals, is dependent on health system policies and was excluded as the study concerns a single health system. Potential variables with correlations of ≥0.4 with other variables were excluded (Supplementary Box S3 and Supplementary Tables S1–S3).
The measure of deprivation used in the current study was the 2019 version of the Index of Multiple Deprivation (IMD),14 which used 39 indicators from seven domains (income, employment, health, education, housing, crime, and environment). Practice IMD values were obtained from the National General Practice Profiles system;9 these were estimated by taking a weighted average of the IMD scores for each lower-level super output area in which a given practice had registered patients. Practices’ NHS commissioning region15 was included as there could be regional variations of health services’ provision and life expectancy.
Data from the Quality and Outcomes Framework (QOF),16 collected from general practice records, provided practice list sizes, disease prevalence, and clinical performance variables (Supplementary Box S4). Disease prevalences were used to represent population morbidity. Mean prevalence was calculated from the annual rates for each year 2015 to 2019, with conditions excluded if their definitions changed during this period. In the current study, the authors initially selected conditions that could plausibly affect whole-population life expectancy. For example, chronic obstructive pulmonary disease (COPD) was included rather than asthma because there are many more deaths from COPD than from asthma.17 The selected conditions were coronary heart disease (CHD), stroke, diabetes, hypertension, COPD, and cancer. The prevalences of all these conditions were combined to create another potential measure of morbidity and the authors also considered the percentage of patients reporting having a long-standing condition, derived from the General Practice Patient Survey (GPPS) (https://www.gp-patient.co.uk). However, only diabetes prevalence was retained, as all the other morbidity variables were highly correlated with each other or with ethnicity variables (Supplementary Table S2).
Variables relating to clinical care performance that could plausibly affect life expectancy were derived from QOF indicators that were defined consistently throughout 2015 to 2019. After eliminating highly correlated variables, four were retained — the percentages of patients:
aged ≥45 years who had blood pressure recorded in the preceding 5 years;
with COPD who were vaccinated against flu;
with diabetes whose last International Federation of Clinical Chemistry HbA1c was ≤59 mmol/mol; and
with CHD on antiplatelet or anticoagulant therapy.
Data from the GPPS between 2015 and 2019 provided measures of ethnicity and patient reports of access and continuity. GPPS questionnaires are sent annually to samples of patients in every practice nationwide (Supplementary Box S5). Samples are weighted to resemble population characteristics within each practice, accounting for factors that include age, gender, ethnicity, and marital status. The survey was substantially revised in 2018, with changes to some variable definitions and to the sample surveyed. Data, therefore, from the first 3 years only (2015–2017) were used in the current study. The ethnicity categories used were the five bandings nationally recommended (White, Asian, Black, mixed, and other).18 The GPPS did not report the exact figure if the percentage of people in a particular practice belonging to an ethnic group was <0.5%. These values were coded as 0%. As ethnicity categories were strongly correlated with each other, only White ethnicity was retained.
After checking for correlation between variables, for the access variable in the current study the percentage of patients booking an appointment who were seen on the same day was selected (Supplementary Table S1). For the continuity variable, the product of the percentage of people who had a doctor they preferred to see and the combined percentage of those reporting being able to see that doctor always, almost always, or a lot of the time was used.19 Only 2015 data were used for continuity because of the cumulative problem of missing data for this variable over several years. GPPS-reported rates of smoking were excluded in the current study as these were correlated with IMD 2019.
Funding was expressed as NHS payments per patient (Supplementary Box S6).20 General practice workforce data are published using a standardised reporting system.21 Information is provided on four staff groups: GPs, nurses, administrative staff, and direct patient care staff (such as dispensers and assistants), published as both head counts and full-time equivalents (FTEs); full-time work was defined as 37.5 h per week.21 FTEs were used in the current study. Direct patient care staff were excluded from the study as these data were incomplete. From within the other groups, fully qualified GPs, GP registrars, advanced nurse practitioners (ANPs), other practice nurses (excluding ANPs), and receptionists were selected. GP registrars and ANPs consult with patients and may influence life expectancy. Receptionists were included as fewer could affect access to clinical staff.
