Study Design and Data Sources
We conducted a population-based cohort study utilizing administrative data from Ontario, Canada. All methods were carried out in accordance with relevant guidelines and regulations. Patients in Ontario are insured under a single-payer system, the Ontario Health Insurance Plan (OHIP) that covers all hip, knee and shoulder total joint replacements.12 All inpatient hospital stays and same day procedures are reported in the Discharge Abstract Database (inpatient procedures) and the National Ambulatory Care Reporting System (same day surgery).13-15 Both databases identify the procedures performed during the hospital stay, using Canadian Classification of Health Interventions (CCI) codes, and patient comorbidities and complications, using International Statistical Classification of Diseases and Related Health Problems 10th revision (ICD-10) codes. Study protocols were approved by IC/ES. Use of the data in this study was authorized under section 45 of Ontario’s Personal Health Information Protection Act, which does not require review by a Research Ethics Board.
The Ontario Drug Benefit (ODB) funds prescription medication for all patients aged 65 years and older. The ODB database contains a record for each prescription filled including the date, physician, number of days supplied, and drug identification number (DIN).
Baseline and post-discharge covariates for each patient were obtained from the following databases: the Registered Persons Database (RPDB), for basic demographic information on each individual; the ICES Physician Database, for surgeon demographic information and specialty; the Continuing Care Reporting System, for identifying patients treated in complex continuing care facilities; and the National Rehabilitation System database, for identifying stays at inpatient rehabilitation institutions.
Participants
We selected patients receiving elective primary total joint replacements (hip, knee and shoulder) between April 1, 2002 and March 31, 2016 from physician and hospital records. We excluded patients who were younger than 67 years (to allow for a look-back period for pre-operative opioid use), bilateral procedures, and patients from out of province. Only the first elective joint replacement for each patient was retained. Individuals were followed for 12 months after surgery. For additional exclusions, please refer to Figure 1.
Primary exposure
The exposure of interest was opioid use in the year immediately preceding the surgery. Opioid medications included codeine, oxycodone, hydrocodone, hydromorphone, meperidine and fentanyl. Each patient was categorized as a ‘non-user’, an ‘intermittent-user’ or a ‘chronic user’ based on their opioid use during the year before surgery. ‘Non-users’ were individuals who did not fill a prescription for opioids in the year prior to their joint replacement. ‘Chronic use’ was determined using a well-established definition, and these were individuals who had at least 90 days of continuous use of opioids in this period.16, 17 “Continuous” in this context refers to patients that received one or multiple prescriptions in succession, where the days dispensed added up to 90 days or more. Individuals who had some use but who did not meet the criteria for chronic use were categorized as ‘intermittent-users’.
Outcome of interest
The primary outcome of interest was the occurrence of a composite complication of deep infection requiring surgery, dislocation or revision arthroplasty within one year. These complications were joint-specific – i.e. a shoulder infection did not count as a complication for someone who had a hip replacement. We used a composite outcome, as the rates of these complications at one year are generally quite low (<0.5%).18-20 Additionally, we analyzed each complication individually. We additionally looked at a composite of readmission to the hospital and return to an emergency department within 30 days. These complications were identified using ICD-10 diagnostic, OHIP billing and CCI procedure codes.19 Infections were identified by the occurrence of a hospital code for intra-articular infection with a confirmatory procedure code or physician claim for an irrigation and debridement, or a spacer insertion.18 Revision procedures were identified using CCI codes accompanied by the supplementary status attribute “R.”18
Covariates of interest
Patient age, sex and neighbourhood income quintile were obtained from the RPDB.21 Co-morbidities in the four years before surgery were categorized using the Deyo-Charlson Index22 and the Elixhauser scale.23 Frailty was defined using the Adjusted Clinical Groups (ACGs) indicator (The Johns Hopkins ACG® System Version 10.0).24, 25 Rheumatoid arthritis was identified using a previously validated algorithm.26 Income quintile and the Ontario Marginalization Index were used as surrogates for socioeconomic status.27-30 Obesity was determined from surgeon billing records. Surgeon volume was defined as the number of arthroplasty procedures performed by the surgeon in the previous year.19 Hospital-volume was similarly defined. Hospitals were also categorized as either ‘academic’ or ‘community’ (www.cahohospitals.com).31
atching and statistical analysis
Intermittent opioid users were matched to non-users by the joint being replaced (hip/knee/shoulder) and a propensity score incorporating socio-demographics (age, sex, income quintile, Ontario Marginalization Index, rurality), pre-existing health status (Charlson score, Elixhauser score, frailty, obesity, rheumatoid arthritis), provider characteristics (teaching hospital, surgeon volume, hospital volume) and the year of surgery. These patients were matched using calipers of width equal to 0.2 of the standard deviation of the logit of the propensity score19, 32 via the greedy (or “nearest neighbor without replacement”) matching method.33 A matching ratio of 1:1 was used.34 Patients were specifically matched by the joint replaced, such that knee replacements in intermittent opioid-users were only being compare to knee replacements in non-users, and so on.
This process was then repeated to match chronic-users to non-users. We estimated standardized differences for all covariates before and after matching, with a standardized difference of 10% or more considered indicative of imbalance35. Complications were compared between the two groups using proportional hazards survival analyses adjusted for matching All analyses were performed using SAS software (version 9·3 and SAS EG 6·1, SAS Institute, Cary, NC). The two-tailed type I error probability was set to 0·05 for all analyses.