We used physician and prescription drug claims, hospital discharge data, and demographic records from January 1997 to March 2012 from the provincial health services administrative databases administered by the Régie de l’assurane maladie du Québec (RAMQ). In this Canadian province, coverage for outpatient and inpatient physician services is provided for the entire population (about 7.5 million people). Individuals aged 65 years or older (1,106,428 individuals in 2011; 90 % of that population) and those under 65 years (2,261,786 individuals in 2011; 32 % of that population including those who receive social assistance, 493 212 individuals, and those who do not have collective private drug insurance, 1,768,574 individuals, such as the self-employed), have their prescription drugs covered by the provincial government. Medication dispensations have been demonstrated to be accurately and reliably recorded in the RAMQ prescription claims database . The Quebec hospital discharge abstract database provides information on all hospital admissions including primary and secondary discharge diagnoses [International Classification of Diseases (ICD), 9th revision (ICD-9) codes until April 2006 and 10th revision (ICD-10) codes thereafter] and admission/discharge dates. The databases are linkable by a unique patient identifier. Permission to link the data was obtained from the Provincial Ethics Board, the Commission d’accès à l’information. Approval to conduct the study was obtained from the McGill University Health Centre Ethics Review Board.
As anti-TNF users were expected to have failed at least two DMARDs and to have a more severe RA profile in the earlier years of marketing, we conducted our main analyses over two separate time-periods January 2002 - December 2006 and January 2007 - December 2011 and results were compared. To further understand the effect of anti-TNF on the study outcomes, we also ran our analyses over the time-period immediately preceding the introduction of anti-TNF medications on the Quebec drug formulary, January 1998 - December 2001.
We describe the study cohort in 2002–2006. Study cohorts of 1998–2001 and 2007–2011 are similarly constructed. An RA cohort was first constructed including patients 20 years of age and older who in 2002–2006 had two outpatient ICD-9 codes for RA (714.x) at least 30 days and no more than 2 years apart or one diagnosis (principal or secondary) from the hospital abstract database (ICD-9 code 714.x until April 2006 and ICD-10 codes M05.x, M06.x, M08.x, and M09.x afterwards) . Individuals who received at least one dispensed prescription for either a DMARD or anti-TNF during the study period were considered in the study. The anti-TNF agents available in Quebec during 2002–2006 were: infliximab, etanercept, adalimumab. Certolizumab, rituximab and golimumab became available in the period 2007–2011 and were also included. The DMARDs included in the study were: chloroquine, hydroxychloroquine, cyclophosphamide, cyclosporine, methotrexate, minocycline, penicillamine, auranofine, sulfasalazine, leflunomide, aurothioglucose, aurothiomalate. Other non-anti-TNF biologic drugs were not included in this study because of the very small number of users.
Each study cohort consisted of three groups, anti-TNF, DMARDs and non-users. Patients in the RA cohort dispensed an anti-TNF in 2002–2006 were identified at the first dispensation date (index date). Those covered by the provincial drug plan for at least 1 year prior and 3 months post the index date were included and formed the anti-TNF group (anti-TNF naïve in the previous year). Two other groups, the DMARD and non-user groups, were formed to match the anti-TNF group on age (±2 years), sex, date of RA diagnosis (±90 days) and high-dimensional propensity score (±0.05) . To form the DMARD group, for each individual in the anti-TNF group, three anti-TNF naïve individuals (for at least 1 year before and 90 days after the date of anti-TNF dispensing) who used a DMARD within 90 days of the index date of the anti-TNF individual were selected at random from those eligible at the date of the DMARD use (index date; for those who used more than one DMARD within the 90-day-period, the closest date to the anti-TNF index date was chosen). The non-user group was constructed similarly. For each individual in the anti-TNF group, three anti-TNF and DMARD naïve individuals were selected at random to match the anti-TNF individual on age, sex, date of RA and high-dimensional propensity score as described above. The index date of the anti-TNF individual was assigned as the index date of the non-user individuals. Of note, patients in the non-user group did not use an anti-TNF or a DMARD for at least 1 year before and 90 days after the index date, but have used at least one of these drugs at another time during the study period as described above. The non-users could be under therapy with other drugs, such as corticosteroids and nonsteroidal anti-inflammatory drugs (NSAIDs). The high-dimensional propensity score has been proposed to adjust for indication and confounding biases caused by missing or misclassified information. In this study, the high-dimensional propensity score (probability of receiving an anti-TNF) was calculated in a multivariate logistic regression model adjusted for a large number of covariates (500), as suggested in the algorithm, assessed based on all diagnoses, procedures, services and drug-dispensations recorded in the databases and selected according to their potential to bias the exposure-outcome relationship under study . A high–dimensional propensity score algorithm is available as downloadable SAS software files from the Brigham and Women’s Hospital . All included patients were required to be alive for at least 90 days past their index date and not to have had any MSD-related hospitalization in the prior year.
All study patients were followed from index date until the first date of death, end of drug coverage, switch/discontinuation of their index treatment or a maximum of 1 year. Discontinuation of treatment was defined as at least 90 days without the treatment and treatment switch was a switch between the three exposure groups (a switch between two DMARDs or two anti-TNFs was not considered to be a treatment switch).
The first hospitalization with a principal diagnosis for any MSD reason, ICD-9 and 10 codes included in the chapter XIII - diseases of the MSD system and connective tissue, during follow-up was the principal outcome (Appendix 1 and Appendix 2– diagnoses found in the study).
Patient baseline characteristics
Patient characteristics assessed at index date included: age, sex, type of insurance plan (based on patient eligibility for premium subsidies; low income patients were those receiving premium or partial subsidies), region of residence (urban or rural), visits to a rheumatologist in the prior year, comorbidity (cancer, ischemic heart disease, congestive heart failure (CHF), peptic ulcer disease, cerebrovascular disease (CVD), atrial fibrillation, and hematologic disorders), medication used in the prior year [corticosteroids, gastroprotective agents (proton pump inhibitors, misoprostol and histamine-2 receptor blocker), serotonin reuptake inhibitors (SSRI), anticoagulants, antidiabetics, antihypertensives and NSAIDs]. These factors were chosen because of their potential association with the choice of RA treatment and the outcome (MSD-related hospitalizations). In addition, our data included an index of socioeconomic status (SES), with sub-indices of social and material deprivation, that was developed by the Institut National de Santé Publique du Québec on the basis of census enumeration area data on education level, employment/population ratio, and average income [14, 15].
In secondary analyses, patient selection, variable assessment and statistical analyses described above were repeated for patients on DMARDs and patients on neither drug, in the three periods (1998–2001, 2002–2006 and 2007–2011). The DMARD group in this analysis consisted of all patients who used a DMARD, but had never been on an anti-TNF prior to the DMARD use, as opposed to the previous analysis where the DMARD group was selected to match the anti-TNF group.
The following analyses were conducted separately in each of the study periods. Descriptive analyses [mean and standard deviation (SD) or proportion] were used as appropriate to report baseline patient characteristics by treatment group. Polytomous logistic regression models were used to compare patient baseline characteristics between the three treatment groups. The crude rates/100 patient-years (py) of MSD-related hospitalizations were assessed. Kaplan Meier curve displayed time to the first MSD-related hospitalization in the three treatment groups. Multivariable Cox proportional hazard models were used to compare the hazard ratios of MSD-related hospitalizations between treatment groups adjusting for patient baseline characteristics.
All statistical analyses described above were repeated in secondary analyses to compare MSD-related hospitalizations in the DMARD users versus non-users in the three time-periods. All statistical analyses were performed using SAS version 9.4 for UNIX (SAS Institute Inc., Cary, NC).