Data sources

We used a national database of primary care coded electronic medical records from England; the Clinical Practice Research Datalink (CPRD, Gold and Aurum) (15,16). The CPRD includes routinely collected electronic medical records from consenting primary care practices. The CPRD comprises two databases, Aurum and Gold, each containing data from primary care practices using different patient management systems. The CPRD is a dynamic database with the number of contributing primary care practices varying over time. As of September 2018, CPRD Aurum included data for around 7 million patients, representing around 13% of the population of England. As of July 2013, CPRD Gold included data for 4.4 million patients, representing around 7% of the total UK population (15). Further details of the CPRD are described elsewhere (15,16).

The CPRD was linked to secondary care coded electronic medical records, the Hospital Episode Statistics (HES) database (17), and also the Office for National Statistics (ONS) national mortality database.

Identification of incident hip and knee OA

We included in our analysis incident cases of hip and knee OA occurring between 2 January 1998 and 31 March 2019, based on diagnostic codes for hip and knee OA recorded in the primary care records (Supplementary Table 1). The incident date of hip or knee OA was defined as the first occurrence of a hip or knee OA diagnostic code recorded in the primary care record. To mitigate the potential for incorrectly classifying prevalent OA cases as incident cases, we excluded individuals who had a first OA diagnostic code recorded within 12 months of registration with the primary care practice. Since the prevalence of frailty is relatively low among people aged less than 60 years, we included only people aged 60 years or older at the time of OA diagnosis.

Identification of primary total hip and knee arthroplasty

We identified primary THAs and TKAs occurring between 2 January 1997 and 31 March 2019, based on OPCS codes (18) recorded in the HES data. We excluded people who had had a THA or TKA due to fracture, osteonecrosis, rheumatoid arthritis, and malignant neoplasm of bone. In addition, we excluded cases also where the coded primary indication for THA and TKA was used in <0.05% of cases.

Assessment of frailty

Frailty was assessed using the electronic frailty index (eFI) (19). The eFI comprises 36 age-related deficits identified by coded data in primary care electronic medical records and was developed using a standard procedure described by Rockwood, et al (20). Clinical codes were searched in the primary care electronic medical records to identify candidate deficits for inclusion in the eFI recorded based on the deficits included in the Canadian Study of Health and Aging frailty index (21). Further details of the development and initial validation of the eFI are described elsewhere (19). In order to apply the eFI in practices using the SNOMED coding system, we mapped the original eFI Read code lists to SNOMED codes using mapping tables from the National Health Service Data Migration Programme (22). The eFI is calculated as the total number of deficits present in an individual, divided by 36. The deficits included in the eFI is shown in Supplementary Table 2. The eFI was determined at the date of hip or knee OA diagnosis and, based on previously published thresholds, was categorised as fit (eFI≤0.12), mild frailty (0.12<eFI≤0.24), moderate frailty (0.24<eFI≤0.36), and severe frailty (eFI>0.36) (19). The eFI has been validated in multiple databases and criterion validity has been demonstrated by comparing the eFI to other frailty instruments, including the phenotype model of frailty and the Clinical Frailty Scale (19,23,24).

Assessment of deprivation

Deprivation was assessed using the 2019 English Index of Multiple Deprivation (IMD) (25). IMD is a multidimensional measure of neighbourhood-level deprivation based on an individuals’ postcode and in this analysis IMD was categorised based on quintiles. The IMD is calculated based on seven distinct domains of deprivation: income deprivation, employment deprivation, education, skills, and training deprivation, health deprivation and disability, crime, barriers to housing and services, and living environment deprivation (25). Further details about the construction of the IMD are available elsewhere (26). The 2019 English IMD was categorised into quintiles, based on the national distribution of the IMD.

Assessment of body mass index

Since body mass index (BMI) has been linked with the likelihood of receiving THA and TKA (9), we considered BMI as a covariate in our analyses. Data on body mass index (BMI) was extracted from the primary care record. For each individual, we used the BMI value that was recorded closest to the date of incident OA and within 12 months prior to the date of incident OA. For individuals who did not have BMI recorded in the 12 months prior to the date of incident OA, BMI was set to missing.

Assessment of ethnicity

Since ethnicity has been linked with the likelihood of receiving THA and TKA (27), we considered ethnicity as a covariate in our analyses. Ethnicity was derived from data recorded either in the primary care record, or in the HES data and categorised as White, Asian, Black, Mixed, other. Where there was disagreement in ethnicity category between the primary care records and HES, or where ethnicity was not recorded, ethnicity category was set to missing.

Statistical analysis

Descriptive cohort characteristics were calculated and mean (standard deviation (SD)) values were presented for continuous variables and number (%) were presented for categorical variables.

We determined the association between frailty category at the date of OA diagnosis (in all analyses, ‘fit’ was the reference category) and likelihood of receiving THA or TKA using time-to-event, maximum likelihood, competing-risk regression models (Fine and Grey method) (28), with mortality modelled as a competing event. Individuals contributed person-time to the analysis from the first recorded date of hip or knee OA diagnosis (incident OA) until the date of receiving THA or TKA, date of death, the date the individual’s primary care practice stopped contributing data to the CPRD, or 31 March 2019 (end of study period), whichever came first. We adjusted for year of OA diagnosis, age at OA diagnosis, and sex (based on primary care records). We then adjusted additionally for quintile of IMD in a separate model. In separate analyses, we adjusted additionally for BMI (continuous variable) and ethnicity category among individuals who had these variables recorded.

As a supplementary analysis, we modelled the eFI as a continuous variable when looking at the association between the eFI and likelihood of receiving THA or TKA. To allow for possible nonlinear relationships, we included fractional polynomials for the eFI in a Cox regression model (see Supplementary Text).

We determined the association between quintile of IMD (least deprived was the referent group) and frailty category at the time of hip or knee OA diagnosis using multinomial logistic regression. Quintile of IMD was the exposure variable (least deprived quintile was the referent group) and frailty category was the outcome variable. We adjusted the model for year of OA diagnosis, age at OA diagnosis, and sex. Results were presented as relative risk ratios and 95% CIs.

Finally, we determined the association between quintile of IMD (least deprived was the referent group) and likelihood of receiving THA and TKA using competing risk regression models, with mortality modelled as a competing event. We adjusted for year of OA diagnosis, age at OA diagnosis and sex. We then additionally adjusted for frailty category in a separate model. In separate analyses, we additionally adjusted form BMI (continuous variable) and ethnicity category among individuals who had these variables recorded. The results were expressed as subhazard ratios (SHR) and 95% confidence intervals (CI).

Analyses were carried out using Stata/MP v13.1.

The protocol for this work was approved by the Independent Scientific Advisory Committee for CPRD research (protocol number 20_119). CPRD has ethics approval from the Health Research Authority to support research using anonymised patient data.