This is a population-based cohort study using routinely collected Hospital Episode Statistics data in England from 1 January 1999 to 31 December 2020.

Records for all patients undergoing elective shoulder replacement surgery in England were available from the Hospital Episode Statistics (HES) Admitted Patient Care (APC) database managed by NHS Digital. The HES APC database provides universal coverage of all inpatient and day case activity carried out by NHS hospitals and NHS-funded care in England and contains demographic data, medical diagnoses, and procedural and administrative information. Data submission from hospital providers is mandatory to ensure accurate reimbursement for all activity performed. Data were linked to the Civil Registration Mortality database. Population estimates by age, sex, and year within each Government Office Region (GOR) were obtained from the Office for National Statistics (ONS) and linked to the HES data for analysis [10]. National population projections per 5-year age groups and sex were obtained from the ONS for the years 2021 through 2050 [11].

The study dataset consisted of all episodes for included patients, linkable by a valid pseudonymised patient identifier. The index operative episode was identified as the first episode containing a procedure for a shoulder replacement per side. Subsequent shoulder replacement procedures on the same side were identified as repeat (revision) surgery. Revisions included in this study were restricted to those linked to elective primary procedures that met the eligibility criteria. The three types of shoulder replacement procedures (humeral hemiarthroplasty [HA], conventional total shoulder replacement [TSR], and reverse total shoulder replacement [RTSR]), as well as revisions, were identified from combinations of primary/revision and anatomy OPCS-4 codes (see Additional file 1). The GOR of residence for each shoulder replacement procedure was identified from the patient’s outward code (first part of the postcode). While GORs closed in 2011, this regional geography is maintained for statistical purposes and is referred to as ‘regions’. Patient socioeconomic status was assigned using the Index of Multiple Deprivation (IMD). This is a combined measure of deprivation capturing income, employment, education, health, crime, barriers to housing and services, and living environment domains [12]. IMD overall rankings were used to categorise patients into five IMD groups from the most deprived 20% to the least deprived 20%. Population data stratified by IMD fifths were only available from 2001 onwards (IMD areas were created in line with the 2001 Census).

Serious adverse events (SAEs) were defined as medical complications severe enough to require admission to hospital including pulmonary embolism, myocardial infarction, lower respiratory tract infection, acute kidney injury, urinary tract infection, cerebrovascular events, and all-cause death [9]. SAEs were identified using ICD-10 codes and categorised into those occurring within 30 or 90 days from the index procedure.

The NHS HRG4 + 2022/23 national costs grouper was used to generate Healthcare Resource Group (HRG) codes for each index operative spell [13]. Each operative spell may consist of one or more episodes, including inpatient activity before or after the operative episode, enabling the capture of all inpatient activity related to that index procedure. HRGs were valued using the 2020–2021 NHS Reference Costs to generate the reimbursement value of each procedure to the hospital provider based on the National Reimbursement System [14].

All patients aged 18 years and older who had an OPCS-4 code for a primary shoulder replacement were eligible for inclusion in the study. Patients were excluded if the main indication for surgery was acute trauma or malignancy, based on ICD-10 diagnostic codes. Patients were excluded if their surgical history was inconsistent (i.e. their date of revision or death predated their primary surgery) or contained duplicates.

Several of the top ten research uncertainties from the 2015 James Lind Alliance Priority Setting Partnership on shoulder surgery related to shoulder replacements [15]. A Patient Advisory Panel for this study highlighted the importance of equitable access to shoulder replacement services across the country to reduce travel for elective surgery. We therefore also planned to analyse the availability of surgical units providing shoulder replacements in each region.

Descriptive statistics were used to summarise patient demographics. Population data from the ONS were used to calculate standardised incidence rates by year of treatment, stratified by region, IMD fifth, age band, and sex, following the methodology of the Association of Public Health Observatories, using direct age and sex standardisation [16]. Age- and sex-standardised risks were calculated for SAEs within 30 and 90 days of surgery. Risks for each type of SAE were also analysed separately. For revision surgery, we were interested in the net failure of the implant, and so the Kaplan–Meier estimator was used to estimate the risk of revision at 1, 3, 5, and 10 years following primary shoulder replacement. Flexible parametric survival models were used to estimate the age- and sex-adjusted risk of revision at each follow-up period as the proportional hazards assumption for these variables did not hold (precluding analysis using a simpler Cox model) [17].

Service provision for each region was evaluated by calculating the number of surgical units providing shoulder replacement surgery per 100,000 population for each region of treatment (surgical unit density) per year and comparing this to the regional incidence of elective primary shoulder replacement procedures. The rate of travel for treatment was calculated by comparing the region of patient residence to that of treatment (the hospital provider’s region).

Two different scenarios were considered to calculate projections for shoulder replacement surgery demand. Scenario 1 used an age- and sex-standardised incidence rate that was held constant at the 2019 levels (preceding the COVID-19 pandemic) while scenario 2 used a linear extrapolation of the age- and sex-standardised incidence rate for the study period up to 2019 [18]. For scenario 2, separate linear regression models were fit to historical data for each 5-year age band and sex cohort, using year of surgery as a covariate, and predictions derived for future years. Data from 2020 were not used for forecasting due to the marked effect of the COVID-19 pandemic on surgical volume. The corresponding incidence rates for each scenario were applied to national population projection data from the ONS to calculate the expected standardised incidence and hence absolute volume predictions for 2021 through 2050. The forecast cost was calculated by applying the 5-year age band and sex-stratified mean costs for 2019 to the surgical volume projections. All historic and forecast costs are presented in 2021 GBP, as all admissions were valued according to the latest available NHS Reference Costs (2020–2021) at the time of conducting the study.

Data for either region or IMD were missing for a total of 1029 patients (1.3% of the study dataset). No data was missing for any other variables included in this study. These records were excluded, and a complete case analysis was undertaken (see Additional file 1 for data flowchart and for baseline characteristics and outcomes of observations with missing data) [19, 20]. A total of 6% of procedures did not generate a valid HRG code, so historical costs only reflect 94% of shoulder replacement surgery undertaken. The geographic information system, QGIS V.3.82, was used to graphically summarise standardised incidence rates for each region in England, per year [21]. Study findings are reported in accordance with the REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) recommendations (see Additional file 2) [22]. Stata V.16.1 (StataCorp) was used to perform all statistical analyses [23].