Study Design
We conducted a retrospective cohort study between January 1, 2011, and December 31, 2015, in Ontario. All health services examined in this study (except prescription drug) are covered by a government single-payer insurance program called the Ontario Health Insurance Plan (OHIP). Therefore the study included all patients in Ontario who had accessed OAT within the study period. We used the date of the first opioid visit as the index date for all analyses. We designed the study around the index date, and follow-up times were fixed to eliminate bias related to the varying length of time in treatment for each patient. The last date of inclusion was December 31, 2015, and all dependent and independent variables were evaluated for one year after enrollment in OAT.
A treatment episode was defined as the time between the last and first OAT billing codes within a period of continuous retention in treatment (no interruptions care of more than 30 days). We used the first episode of OAT to identify patients, meaning that there was no previous history of OAT (including methadone or buprenorphine/naloxone) in the year before the first treatment episode. We chose to only include patients with no history of OAT in the year before the first episode during the study period to eliminate bias associated with cases involving multiple treatment attempts (23, 37, 38).
This study was approved by the Research Ethics Board of Laurentian University in Sudbury. This study is reported following the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines (39)
Study Population
The study cohort was created by extracting OAT patients from the Ontario Drug Benefit Plan (ODB) database, which captures data on publicly funded pharmacy billing using drug identification numbers (DIN) (see Appendix A). And with the (OHIP) database, which captures data on publicly funded physician-based health services using physician billing codes (see Appendix B). In previous published ICES studies (27, 35, 40), the ODB database was used as the primary source to identify OAT patients. However, in Ontario, residents are only eligible for ODB public drug coverage if they are aged 65 years or older, reside in a long-term care facility, are disabled, are receiving social benefits for income support, or have high prescription drug costs relative to their net household income. Since 2011, new billing codes have helped to identify OAT services (41) in administrative databases because OHIP coverage is available to all permanent residents of Ontario. Therefore, to avoid excluding a subset of the population and risking selection bias, we used both ODB and OHIP databases to identify the primary patient cohort.
A set of exclusion criteria was applied to define the main cohort. We excluded all patients under 15 years of age (n = 2,535 patients). Patients who were identified in the ODB with over 20% of their methadone dose in tablet formulation over one year were excluded because methadone in the tablet formulation is not approved for use as an addiction treatment in Canada (n = 5,560 patients). Additionally, the following patients were excluded from the study: patients who were not eligible for OHIP (n = 437 patients), and non-Ontario residents (n = 427 patients). ). Ontario residents may not be eligible for OHIP coverage if they have been absent for more than 30 days during the first six months of residence in the province and if they were not in Ontario for at least 153 days in 12-months).Patients identified from ODB (n = 1,383 patients), from OHIP (n = 30,124 patients) and patients who were identified in both databases (n = 24,417 patients) were combined to create the primary cohort (n = 55,924 patients). The steps used to create the primary cohort are outlined in Fig. 1.
Data Sources
We obtained anonymized patient-level data from Ontario publicly funded health services y submitting a formal requisition to ICES. ICES, previously known as the Institute for clinical evaluative sciences, is an independent research institute that collects and analyzes health care data for research. Patient information was linked anonymously across databases using encrypted ten-digit health card numbers. The linking protocol is used routinely for health system research in Ontario (42–44).
All diagnostic information from physician visits was determined using billing data from OHIP. ED visits were identified using the National Ambulatory Care Reporting System (NACRS). Hospital admissions were identified using the Discharge Abstract Database (DAD). We obtained patients’ location of residence and demographic information, including all-cause mortality from the Ontario Registered Persons Database (RPDB), which contains unique data for each resident who has ever received insured health.
Patients with a diagnosis of one or more mental disorders
Exposure to one or more mental disorders was assumed for all patients who had at least one of the OHIP diagnoses listed in Appendix D. We excluded any substance-related diagnoses from our mental disorder diagnosis definition, which means that all patients in the mental disorders group had OUD and one or more mental disorders other than substance use disorders. We used OHIP diagnosis codes to identify patients in the mental disorders group. In the database, there is one general substance use diagnosis code. All substance use disorders that fall under code 304 are listed in appendix D. All patients in the cohort had that code based on their opioid dependence. Therefore, there was no way to detect those who also had a dependence on other substances such as cocaine, benzodiazepines or other substances.
Patients were assigned to only one of the following groups: those diagnosed with a mental disorder other than OUD and those not diagnosed with a mental disorder other than OUD. The total time parameter to identify the diagnoses of mental disorders was two years for each patient. We identified patients who had a mental disorder diagnosis one-year before the index date to one-year after the index date (or the study end date). Mental disorders can be chronic, re-occurring conditions; therefore, the wide time frame was chosen to capture the condition accurately. Such wide time parameters are commonly used in other studies examining mental health comorbidities using administrative databases (45–49).
Outcomes
Study outcomes were defined based on the need to assess the association of all-cause mortality, frequent ED visits, and hospitalizations as a function of concurrent disorders and not as an exposure leading to an event. All-cause mortality, ED visits, and hospitalizations have been used as indicators of complexity in the OUD population in other studies (7, 27, 50, 51). Additionally, both frequent ED visits and hospitalizations are metrics used by health system planners and funders in Ontario to understand gaps in services in communities (50–52).
We requested the all-cause mortality variable from ICES as a dichotomized variable. At the time of the study, mortality specific data were not available for the entirety of our study period. We used data from the RPDB database to calculate the number of days to death date from the study index date for each patient in the cohort to create the variable. If the patient had a mortality event anytime between their index date and the end of the study period (December 31, 2016), we assigned a code of 1 (all-cause mortality) or 0 (no all-cause mortality).
We used data from the NACRS database to identify ED visits. We considered a patient as having frequent ED visits if 1) contact with ED was after the index date, and 2) a patient had ten or more ED events in a publicly funded Ontario hospital within one year. We used the diagnostic code that accompanied each ED event in the NACRS to categorize types of ED visits. Opioid-related, mental health-related and reasons other than mental health or opioids were included in the analysis. Cut-offs for dichotomization were based on the frequency and distribution of ED visits for the entire cohort. The median number of ED visits was 14 per year, the mode was 6, and the mean was 22 (SD = 30.9). We chose to dichotomize with a value between the mean and mode (code 1 for ten or more ED visits and 0 for less than ten visits per year).
We used the DAD database to identify hospitalization. Hospitalizations were captured in three groups: opioid-related, mental health-related, and for reasons other than mental health or opioids using the primary diagnosis code that accompanied the hospitalization event in the DAD database. Hospitalizations were dichotomized and counted if a hospitalization discharge record appeared after a patient’s index date in a publicly funded Ontario hospital. The cut-off of one hospitalization was decided based on the frequency distribution of the number of hospitalizations for the cohort. The mean number of hospitalization per year was 3.
We conducted a subgroup analysis of one-year treatment retention using the ODB database (n = 25,800). One-year treatment retention is correlated with a variety of positive health outcomes for patients, including reduced rates of drug use, criminal activity, and an increase in employment (53). One-year retention in OAT was assessed based on doses dispensed (from the ODB database). If the difference between the last and first days of dispensed medication within a period of continuous retention in treatment (no interruptions in prescribed doses > 30 days) was greater than 365 days, then the patient was considered to be retained for one year in OAT. Thirty days was chosen based on the use of this interval in previously published research (18, 53, 54). The database used for medication dispensing in this study might not capture doses administered in a hospital or provincial correctional settings. However, in Ontario, patients will typically continue to receive methadone or buprenorphine in these settings. Since most hospital admissions or provincial incarcerations are less than 30 days, this approach allowed us to conduct the analysis without misinterpreting such events as treatment interruption.
Baseline Covariates
The covariates available for the study included age, sex, location of residence, income quintile, human immunodeficiency virus infection (HIV), deep tissue infection including endocarditis (OHIP diagnostic code 429), osteomyelitis (OHIP diagnostic code 730) and septic arthritis (OHIP diagnostic code 711).
Statistical Analysis
We calculated descriptive statistics for exposure groups. We used chi-square statistics to compare categorical variables and the Wilcoxon rank-sum test to compare continuous variables between exposure groups. We applied logistic regression models to test the association between mental disorders and all-cause mortality, ED visits, hospitalizations and, on the subgroup, to study one-year treatment retention adjusting for patient covariates including age, sex, income, location of residence, HIV status and deep tissue infections. We calculated unadjusted and adjusted Odds ratios (OR) and 95% confidence intervals (CI) for each outcome. Results were considered statistically significant, where p < 0.05. All statistical analyses were conducted from the secure server using SAS Version 9.4 (55). Data was reviewed by ICES to insure privacy standards were met.