Health care system costs related to potentially inappropriate medication use involving opioids in older adults in Canada

The Étude sur la Santé des Aînés (ESA)-Services longitudinal study recruited community-dwelling older adults consulting in primary care clinics between 2011 and 2013 in Québec, Canada. Older adults aged 65 years and older interested in participating in the study were contacted for an at-home interview. Of the 1,811 older adults interviewed in the ESA-Services study, health administrative data in the 3 years prior and following the baseline interview (T1) were successfully merged to health survey data for 1,657 consenting participants. Further details of the longitudinal study, which aimed to assess health service utilization and inappropriate prescribing in a primary care population of older adults, can be found in a previous article [26]. The analytic sample for this study was limited to 1201 older adults with no malignant tumor diagnosis (ICD-9: 140.X-165.X, 170.X-176.X, 179.X-209.X, ICD-10: C00.X-C26.X, C30.X-C41.X, C43.X-C58.X, C60.X-C97.X) and no opioid use in the year before the at-home interview (T1). Older adults with cancer were excluded given that their health service use is based on specific treatment protocols that are not well documented in administrative databases (ex. chemotherapy, radiotherapy). Individuals with opioid use in the year before the interview were excluded to implement a wash-out period. The sample was also limited to older adults who were covered by the provincial public drug plan over the follow-up, allowing the study of opioid use over time. The province of Québec has a universal health program that covers physician and hospital services for the entire population, as well as a public drug plan that covers virtually all individuals aged 65 years and older for their medications, with more than 90% of older adults covered. As such, data on physician services and drugs dispensed are available from the Régie de l’assurance maladie du Québec (RAMQ) and data on hospitalizations are available from the Maintenance et exploitation des données pour l’étude de la clientèle hospitalière (MED-ÉCHO). Ethics approval for this secondary data analysis was obtained from the CIUSSS Estrie-Centre Hospitalier de Sherbrooke (CHUS) (# 2021–3807).

Opioid use was identified over a 3-year period following interviews using medication claims data from the RAMQ. PIOU was identified over 3 years using the 2019 Beers criteria [23]. The Beers criteria are a list of medications that should be avoided or used with caution in older adults due to a high risk of side effects, adverse reactions, or drug interactions [23]. The most recent Beers criteria, published in 2019, were used to reflect the current impact of PIOU in terms of costs. PIOU included exposure to pentazocine use, meperidine use, concomitant opioid and benzodiazepine use, concomitant opioid and pregabalin or gabapentin use, opioid use in those with a history of falls or fractures, or opioid use combined with two central nervous system (CNS)-active medications. A 1-day overlap in the days’ supply was considered potentially inappropriate for cases of concomitant use. That is, an overlap of an opioid supply by at least one day with the supply of a benzodiazepine, gabapentin, pregabalin, or two other CNS-active medications was deemed potentially inappropriate. The operationalization of this variable has been described previously [15]. Individuals were grouped based on their exposure as no opioid use, opioid use (not potentially inappropriate), or PIOU over 3 years.

Observation time for the descriptive phase-based approach was divided into 3 phases: (i) no opioid use, (ii) opioid use, and (iii) PIOU. Individuals who did not use an opioid in the follow-up period had their entire observation period assigned to phase (i), opioid users contributed a portion of their time to phase (i) until the date that the first opioid was dispensed, after which the remainder of the follow-up period was attributed to phase (ii). Finally, individuals with PIOU contributed to all 3 phases, unless their first opioid use was inappropriate, in which case they would only contribute to phases (i) and (iii). Observation time ended at the time of death or the end of the 3-year follow-up. In this way, exposure was defined using a hierarchical approach such that an individual cannot return to a previous phase once they have entered a new phase under the assumption that returning to baseline health care utilization within the study period after having experienced a PIOU is unlikely [25].

Costs from the health system perspective were calculated according to economic guidelines [27, 28] and published methods [29, 30]. Health system costs considered were those incurred due to a hospitalization (including per diem costs), day surgery, outpatient visit, and emergency department visit. Costs for physician services and the proportion of medication costs covered by the public drug plan (RAMQ) were also considered. Data on hospitalizations and day surgeries were extracted from MED-ÉCHO. Ambulatory visits and drug utilization data were identified using RAMQ medical and pharmaceutical data. Unit costs were calculated for the 2013–2014 fiscal year using budget and activity reports from the Ministry of Health and Social Services (MSSS) that are submitted annually by health care institutions in Québec (AS-471) [31, 32]. Costs were converted to 2022 Canadian dollars using the Consumer Price Index (CPI) for health and personal care for the province of Québec [33]. To consider the opportunity costs associated with land and facilities being used during health care utilization, 6% was added to the unit costs for hospitalizations and surgeries and 4% for outpatient and emergency room visits [34]. Additional costs associated with security, maintenance (9.7%) [30] and building depreciation (10%) [35] were also added to all visits. The costs considered and the unit costs used (in Canadian $) are described in Additional Table 1 [32]. Unit costs were then multiplied by the number of visits (and the number of days for hospitalizations) to obtain the total costs over 3 years and the total costs per phase. The costs used in the main analyses were presented over 3 years to align with the duration of follow-up for the sample. Moreover, the phase-based costing approach was performed by first defining inflection points to represent clinically important phases of medication use among older adults [25]. Next, observation time and costs were allocated to each phase to determine the costs per phase [25]. The total costs per phase were added and divided by the duration of each phase to obtain average daily costs. The mean phase-based costs were then standardized to costs per 365 days (per annum). This novel phase-based approach has been shown to be particularly useful in dealing with censoring given that cost data are not required for all participants for the entire period of interest [25].

The current study analyses were guided by the conceptual framework on opioid use in older adults developed by the Agency for Healthcare Research and Quality (AHRQ), which highlights the relation between problematic opioid use and increased health service use such as emergency department visits and hospitalizations related to adverse events [36]. Descriptive analyses were conducted to describe the characteristics of the study sample. Variables from the conceptual framework that were significantly associated with both the exposure (opioid use group) and the outcome of interest (total health system costs over the 3-year follow-up period) were identified as potential confounding variables. Based on these analyses (which are not presented in full), the following potential confounding variables were used to create a probability of receiving an opioid or a potentially inappropriate prescription involving an opioid: age, physical and psychiatric multimorbidity, presence of a substance use disorder, number of daily hassles, and number of prescribers in the 3 years before opioid use or PIOU. The number of daily hassles was measured using the Daily Hassles Scale (score of 0 to 22), which includes items pertaining to family, health, finances, transportation, household chores, and social activities [37]. Substance use disorder presence was identified based on a diagnosis in the 3 years before the interview [38] or the presence of a benzodiazepine dependence based on the DSM-IV criteria [15]. Analyses were weighted using IPTW to reduce the effects of confounding and selection bias associated with receiving an opioid or an inappropriate prescription involving an opioid, or not receiving one [39]. Analyses were further adjusted for variables from the conceptual framework that had a standardized difference ≥ 10% after weights were applied [40]. These included age, sex, education level, country of origin, psychological distress, presence of a traumatic event (including an accident, natural disaster, or life-threatening disease), physical and psychiatric multimorbidity, pain severity, type of pain (inflammatory, nociceptive, neuropathic, nociplastic, other), presence of a substance use disorder, number of prescribers, continuity of care, and type of primary care practice during recruitment. Standardized differences are presented in Additional Table 2, in addition to the sample characteristics after weighting.

The Kolmogorov–Smirnov test showed that costs over the 3-year period and per phase were not normally distributed. Levene’s test showed that variances were not equal across each group (no use, opioid use, PIOU). As such, the condition of homoscedasticity was not respected. Generalized linear models (GLMs) with a gamma distribution and a log link function were chosen to study the costs associated with opioid use group (no use, opioid use, or PIOU), while accounting for the non-normal distribution of such costs [41]. In addition, several other studies have successfully modeled costs using this distribution [4, 12, 30, 42]. GLMs were conducted on a subset of 1124 individuals who had complete data. In GLMs, interactions were tested between three-level exposure (no use, opioid use, PIOU) and sex. All analyses were conducted using SPSS 27.0.