The study analysed a convenience sample of observational, routinely collected data. Between July 2018 and April 2020, the Australian Orthopaedic Association National Joint Replacement Registry (AOANJRR) conducted a PROMs pilot study, collecting pre- and 6 months post-operative PROMs data from patients undergoing hip, knee or shoulder arthroplasty. Patients from 43 hospitals across Australia were included. Different hospital types (high and low volume, metropolitan and regional, private and public) and geographical regions across all six Australian states and one territory (ACT) were represented. The study was nested within the AOANJRR, a national registry that validates more than 97.8% of all arthroplasty procedures performed in Australia [7]. The Australian Orthopaedic Association National Joint Replacement Registry (AOANJRR) was established in 1999 and achieved complete national coverage of all hospitals in 2003. No funding was received for this study.
PROMs data included joint-specific (Oxford Knee Score and Oxford Hip Score) pain and function scores [8], EQ5D-5L quality of life survey [9], joint pain, pre-operative expectations and post-operative satisfaction and perceived change [10]. PROMs data were entered directly using a purpose-built, web-based platform, either by the patient (in clinic or by following links provided by smartphone message or email) or by staff who contacted patients by telephone. Telephone follow-up was only used for patients who did not directly enter data electronically into the web-based platform in clinic or following electronic reminders, or when no means for electronic follow up was recorded. All electronic reminders and telephone calls and all post-operative follow up were administered centrally, within the registry, not at individual hospitals. Individual PROMs data were matched to routine registry data pertaining to the relevant arthroplasty procedure (procedure date, type of procedure, procedure side, age, body mass index [BMI] and American Society of Anaesthesiologists [ASA] physical status classification [11]). Patient characteristics were restricted to the variables available to the registry.
Patient inclusion in the PROMs program involved a two-stage process. The first step was registration in the web-based platform, which could be performed by staff or the patient and consisted of minimal data entry (patient name, date of birth, contact details [email and/or phone numbers], joint, side, surgeon, and hospital) and provision of electronic consent. This was followed by electronic PROMs data entry, which occurred at the time of registration or later, using electronic reminders (email or text message) and telephone follow-up for non-responders. A responder was defined as a patient who answered at least one question.
Due to the two-stage process, sample incompleteness was derived from two sources: patients who underwent arthroplasty but were not registered in the PROMs platform, and patients who underwent surgery and were registered, but did not enter PROMs data. Therefore, sample completeness for each hospital was defined in two different ways:
-
1.
‘Registration’ completeness (the number of patients registered divided by the total number of procedures performed). This provided a measure of the potential completeness of the program at each hospital site and overall.
-
2.
‘Response’ completeness (the number of responders divided by the total number of procedures performed). This included loss of data due to a lack of patient registration (registration completeness, above) and failure to respond to invitations triggered by the registration process. This provided a measure of the actual completeness of the program at each site and overall. Response completeness can be measured for any data collection event (e.g., pre-operative and 6 months post-operative)
This analysis is restricted to elective (non-fracture) primary arthroplasty procedures. The total number of procedures (the denominator) was derived from routine registry data. Representativeness for each hospital was measured with respect to the following patient characteristics: age, gender, BMI and ASA score for each joint (hip, knee and shoulder). ASA score was dichotomised into grades 1–2 and grades 3–5, due to low numbers available for individual analysis for ASA grades 1, 4 and 5.
The completeness rates for each hospital were regressed on summary patient characteristics (mean or percentage) for each hospital. Linear regression was performed for all hospitals, with separate models for each measure of completeness (‘registration’ completeness and ‘response’ completeness), and separate models for each patient characteristic. The significance level was set at 0.05. Given there were two time points for data collection, response completeness was analysed separately for response rates pre-operatively and 6 months post-operatively. These analyses were repeated in multiple linear regression in separate models using all patient factors for each outcome. For each model, the model assumptions were checked by examining standard diagnostic plots. Missing data were not imputed as missingness in the outcome (i.e., completeness) was the dependent variable and the population was restricted to those providing (non-missing) baseline data.