Prognostic factors associated with incident primary hip and knee arthroplasty: findings from Record-Wide Association Study (ReWAS) within the Clinical Practice Research Datalink (CPRD)

Problem

There are currently no established models to predict patients’ risk of progressing to joint replacement. A systematic literature review identified 42 prognostic factors but most are not feasible for use in routine practice. As part of a project aiming to derive and validate a prediction model based on routine records in general practice, we undertook a novel hypothesis-free ‘Record-Wide Association Study (ReWAS)’.

Approach

96,450 and 165,413 incident primary hip and knee arthroplasty cases between 1992 - 2013 were identified from the Clinical Practice Research Datalink using 3-year run-in period. An age-, gender- and practice-matched control with ≥1 consultations in the past 3 years was selected for each case using risk-set sampling. For ReWAS, firstly, we identified all third-level Read codes and third-level sections within the British National Formulary which had been recorded in ≥1% of cases in the 3 years prior to date of arthroplasty as potential prognostic factors (n=6,109 and 325 respectively); secondly, we used conditional logistic regression to estimate their association with outcome, with ‘hits’ defined as those factors with a population attributable risk (PAR) ≥ 1% or ≤ -1% and with significant Bonferroni-corrected P-values; thirdly, we asked a clinical review panel (6 GPs + 1 lay member) to interpret the ‘hits’ based on clinical relevance / plausibility. Agreement was defined as ≥3 panel-members reaching the same decision. Finally, we repeated the ReWAS method in a case-control study with 874 incident knee/hip arthroplasty cases and 4,370 matched controls, derived from a regional dataset - Consultations in Primary Care Archive (CiPCA).

Findings

We identified 36 ‘hits’, i.e. potential prognostic factors recorded in ≥ 1% of cases, and having PAR ≥ 1% /≤ -1% and P-value < 1.539×10-4 (32 for primary hip arthroplasty; 33 for primary knee arthroplasty (29 associated with both outcomes)). The clinical review panel agreed on the clinical relevance/plausibility of 27 of these: (A) General: ethnicity, family history of (osteo)arthritis; (B) Lifestyle: body mass index, diet consultation, smoking status, alcohol intake, physical activity; (C) Comorbidity: cardiovascular diseases, treated hypertension, asthma, mental disorders, diabetic foot; (D) Musculoskeletal disorders: rheumatoid arthritis, previous fracture/ injury, osteoarthritis and allied disorders, knee effusion, low back pain; (E) Pharmacological/nonpharmacological treatments for OA/pain: corticosteroids, glucocorticoids, antidepressants, opioid analgesia, oral NSAID/Cox-2, topical NSAID, knee/hip intra-articular injection, physiotherapy referral; (F) Other prescribed medicines: sex hormones, prostaglandins & oxytocics (as gastroprotection). 14 of these 27 factors were successfully replicated in the regional dataset.

Consequences

A ReWAS approach is feasible for identifying potential prognostic factors for primary hip/knee arthroplasty in primary care electronic health records. Further research will investigate whether the ‘hits’ from such an approach, when combined in multivariable prediction model, can help GPs discriminate patients at high-risk of future joint replacement.