The methods and results are described as recommended in the Strengthening of Reporting in Observational studies in Epidemiology (STROBE) statement .
Setting, participants and data sources
For this study, we used the Mayo Total Joint Registry that prospectively collects data on all joint replacements performed at the Mayo Clinic, Rochester, MN, USA. Data in the registry include patient demographics, operative diagnosis, surgery and implant details, dates of evaluation, reoperations and complications, current radiographs and pain and functional assessments [13, 14]. All patients who undergo THA at the Mayo Clinic are requested to complete a validated Mayo Hip questionnaire preoperatively and at two- and five-year follow-ups. The Mayo hip questionnaire has face, content and construct validity and test-retest reliability [15–17]. The pain and function questions are similar to those in the validated Harris Hip Score , the most widely used questionnaire in THA patients. The Mayo Hip questionnaires are mailed to the patients, administered during the clinic visit or by telephone by experienced, dedicated joint registry staff. Questionnaire data have been captured electronically starting in 1993. Patients were included in this study if they had undergone a primary THA from 1993 to 2005 and had responded to the Mayo hip survey at the two-year or five-year follow-up. Patients of all ages were included. The Institutional Review Board (IRB) at the Mayo Clinic, Rochester, MN, USA approved the study. Since it was a database study, the IRB waived the need for an informed consent.
Predictors of interest and covariates
The operative diagnosis was the main variable of interest, categorized as osteoarthritis (OA), rheumatoid or inflammatory arthritis (RA), avascular necrosis of bone (AVN) and other. This was based on preoperative diagnosis (based on history, clinical examination, medications and the results of radiographic and other studies) as well as operative findings. We adjusted for covariates that included known and suspected correlates of pain and function after THA. Covariates were demographics (age, gender), body mass index (BMI), American Society of Anesthesiologist (ASA) class, distance from the medical center and implant fixation (uncemented, hybrid/cemented) obtained from the Total Joint Registry and linked databases. Anxiety, depression and medical comorbidity using the validated Deyo-Charlson index  were based on the presence of International Classification of Diseases-ninth revision, common modification (ICD-9-CM) codes in the Mayo Clinic electronic databases, derived from administrative and clinical records. Distance from the medical center was included, since Mayo Clinic provides THA to local residents as well as referred patients traveling from far, who may have different disease severity and expectations, and both can impact pain and function outcomes [20–22]. In addition, we adjusted all functional outcome models for preoperative functional limitation assessed by preoperative limitation in seven activities of daily living (ADLs) and the pain models for preoperative pain severity, respectively.
Outcomes of interest
Study outcomes were PROs of moderate-severe ADL limitation and moderate-severe pain obtained from self-reported validated Mayo hip questionnaires at two years or five years (reference, no/mild categories) after THA. We defined these composite outcomes a priori as undesirable outcomes of THA as in previous studies [23, 24], since THA is done primarily to relieve pain and improve function. Similar pain and ADL questions were also administered preoperatively; details are in Additional file 1: Table S1. For patients with multiple procedures, the latest observation for the index hip arthroplasty that was available prior to an additional procedure (and qualified for the two- or the five-year time point) was used as a conservative approach.
Patients self reported limitations in seven key ADLs that specifically assessed index hip function, including walking, climbing stairs, putting on shoes/socks, picking up objects from the floor, sitting in a chair, getting in/out of the car and rising from a chair to a standing position. For four ADLs (walking, climbing stairs, sitting and rising from a chair), limitations were categorized into ‘no’, ‘mild’, ‘moderate’ or ‘severe’ limitation. The remaining three ADLs (putting on shoes/socks, picking up objects from the floor and getting in/out of the car), that did not have a response corresponding to the ‘mild’ category, were categorized into ‘no’, ‘moderate’ or ‘severe’ limitation, as previously . The presence of three or more ADLs with moderate or severe limitation was classified as overall moderate to severe ADL limitation (reference, all other categories), as previously described .
Postoperative index hip pain was assessed with a single question on the hip questionnaire, namely ‘Do you have pain in the hip in which the joint was replaced? no pain, slight, moderate, severe.’ This validated question [15–17] is similar to the pain question in the Harris Hip Score, a commonly used THA outcome instrument that is valid, reliable and sensitive to change [25–27].
Bias and sample size
We tried to minimize confounding bias by including several covariates previously known or suspected to be associated with pain and ADL limitation after THA including the preoperative status, but recognize that residual confounding is a limitation of cohort study design. We accounted for correlation of observations (due to bilateral THA in patients, simultaneously or sequentially) using appropriate statistical methods. We anticipated non-response to be higher at five than at two years, and acknowledge this as a study limitation limiting the generalizability of results. We included a large enough sample to study pain and ADL limitation without having too long of a study period (to avoid significant secular trends in implants and procedure) and, therefore, chose all eligible patients from 1993 to 2005. No formal sample size calculations were done.
We used univariate and multivariable-adjusted logistic regression models to assess the association between the operative diagnosis and moderate-severe ADL limitation and moderate-severe pain two and five years after primary THA. The multivariable models included age, gender, BMI, ASA class, distance from the medical center, implant fixation, Deyo-Charlson index, anxiety and depression as well as respective preoperative variable/s – preoperative pain for pain outcome and preoperative limitation in seven ADLs for ADL limitation outcome. Individual ADL limitations were only examined as exploratory analyses, to avoid multiple comparisons and results are presented in Additional file 1. Odds ratios (OR), 95% confidence intervals (CI), and P-values are reported. Subgroup analyses were done for patients younger and older than 65 years to assess the association of diagnosis with pain and ADL outcomes, since patients with AVN or RA are expected to be younger than those with OA. We performed sensitivity analyses by restricting the study sample to surgery from 1998 to 2005, to examine whether change in RA management in the recent years impacted the noted associations in the main analyses.
All regression analyses used a generalized estimating equations (GEE) approach to adjust the standard errors for the correlation between observations on the same subject due to both hips having been replaced. Responder and non-responder characteristics were compared using logistic regression analyses, which were pre-specified to include demographics, comorbidity, implant-related factors and underlying diagnosis. We decided a priori not to impute any missing data and to treat them as missing. A P-value <0.05 was considered significant. Analyses were done using SPSS, version 21 (Chicago, IL, USA).
The Mayo Clinic Institutional Review Board approved this study and all investigations were conducted in conformity with ethical principles of research.