Prescription of oxycodone versus codeine after childbirth and risk of persistent opioid use: a population-based cohort study

Study setting

We conducted a population-based retrospective cohort study of people prescribed either oxycodone or codeine from Sept. 1, 2012, to June 30, 2020, in Ontario, Canada (population 14.7 million in 2020). We selected Sept. 1, 2012, as the study start date because the Narcotics Monitoring System (NMS) database began collecting data in April 2012. We did not examine the first few months after implementation of the database (AprilAugust 2012) to avoid concerns regarding accuracy and completeness of data early in its existence. Ontario residents have universal access to physician and hospital services, including prenatal and postpartum care. In addition, health care records and prescriptions for controlled medications can be tracked using established databases.

Data sources

We identified births using the ICES (formerly, the Institute for Clinical Evaluative Sciences) MOMBABY database, which identifies more than 98% of all Ontario births25 and links the health care records of birthing parents and their babies. We identified postpartum opioid prescriptions using the NMS database, which contains detailed information on prescriptions for opioids and other controlled drugs, regardless of payer. The NMS database commenced in April 2012 and contains information on drug name, dispensing date, and quantity and days of medication supplied. It is regularly used in studies examining prescription opioid use in Ontario.26^{29 We obtained hospital admission data from the Canadian Institute for Health Information Discharge Abstract Database, which contains diagnostic and procedural information for all hospital admissions in Ontario.30 We obtained basic demographic information from the Ontario Health Insurance Plan (OHIP) Registered Persons Database (RPDB), a registry of all people in Ontario eligible for health insurance.31,32 We obtained details on mental health-related hospital admissions using the Ontario Mental Health Reporting System database, and used the OHIP Claims History Database to identify claims for outpatient physician services. These databases are linked in an anonymous fashion using encrypted health insurance numbers, and have been used extensively to study drug safety.3337 ICES is an independent, nonprofit research institute whose legal status under Ontarios health information privacy law allows it to collect and analyze health care and demographic data, without patient consent, for health system evaluation and improvement. Similar coding has been used in studies by our group and others.3,6}

Study population

The study design is outlined in Figure 1. We established a cohort of people aged 1260 years who gave birth in hospital and filled a prescription for medications containing either oxycodone or codeine within 7 days of discharge. We chose a 7-day window after hospital discharge to allow enough time for patients to fill an outpatient opioid prescription. A shorter time period (used in some studies23) might exclude patients who filled a postpartum opioid prescription for pain, but a longer period (used in other studies38) would have risked including those who filled a prescription for an indication other than delivery pain. Studies by our group, and others, have used a similar time frame for ascertaining the first opioid prescription postpartum.3^,6^,39

Figure 1:

Schematic of study design. Note: *+1 to +365 days after the index date.

We chose codeine as the comparator because it has historically been perceived as a weak opioid and, until recently, was the preferred opioid prescribed postpartum.6 Owing in large part to concerns about neonatal safety with breastfeeding, codeine has been supplanted over the past decade by more potent opioids (including oxycodone).3^,6 In the last decade, in Ontario, oxycodone has become the most commonly prescribed opioid postpartum.3

We defined the index date as the date of maternal postpartum opioid prescription fill. We included only birthing parents who were discharged from hospital within 7 days of delivery, as extended hospital stays could reflect maternal illness or delivery complications that might influence prescribing decisions, and we limited the cohort to people who delivered between 20 and 42 weeks gestation. To identify patients newly treated with opioids postpartum, we excluded people who filled any opioid prescription, including buprenorphine or methadone, in the 100 days preceding the index date. Various definitions have been used in administrative data studies to classify people as opioid naive.23^,40^{42 Although no standard definition exists, we have used the preceding definition previously.3,6 For people with multiple deliveries during the study period, we included only the first.}

Statistical analysis

In the primary analysis, we conducted a complete case analysis and used inverse probability of treatment weighting to minimize potential confounders related to opioid selection.43 To generate weights, we estimated a propensity score using a logistic regression model in which the choice of opioid (oxycodone or codeine) was regressed on measured baseline factors, including maternal characteristics (age, income quintile, rurality, gravidity [primi-gravid and number of live births], documented history of addiction, history of mental illness, derived maternal aggregated diagnosis groups [ADG] score from the Johns Hopkins ACG System version 10) and delivery characteristics (mode of delivery, neonatal birth weight, gestational age, severe maternal morbidity, episiotomy, perineal tear, maternal hospital length of stay, year of delivery and stillbirth). Sample codes are found in Appendix 1 (available at www.cmaj.ca/lookup/doi/10.1503/cmaj.221351/tab-related-content). Opioid prescriptions were converted to approximate MMEs using a standard opioid conversion algorithm44 that has been used in other studies.6^,29

We calculated stabilized weights by multiplying the inverse probability of treatment by the marginal probability of receiving the drug that was given.45 We calculated standardized differences for baseline characteristics (both before and after weighting) to examine the balance between groups. Standardized differences less than 0.1 generally indicate good balance for a given covariate.46 We estimated adjusted relative risks (RRs) using weighted modified Poisson regression and a robust variance estimator.47 We also calculated weighted estimates of the absolute risk difference.

We performed prespecified subgroup analyses by mode of delivery (cesarean v. vaginal). We created unique propensity scores for the 2 subgroups (excluding mode of delivery from the propensity scores) and, after reweighting, we reran the analyses in each subgroup. We used z tests to formally test whether the 2 subgroup-specific relative and absolute risks were equal. We also performed unweighted multivariable logistic regression to identify other possible independent risk factors for persistent opioid use, stratified by mode of delivery and adjusting for maternal, delivery and prescription characteristics, including the amount of opioid dispensed (in MMEs) and the duration of the initial prescription (in days). In these analyses, we used logistic regression models to generate odds ratios (ORs), because modified Poisson regression models failed to converge. We used the Wilcoxon rank sum test to compare the median MMEs prescribed to those who filled a prescription for oxycodone and codeine.

We represented relative and absolute risks and ORs with 95% CIs, and considered a 2-tailed type I error rate of 0.05 as the threshold for statistical significance. All analyses were performed at ICES, using SAS software (v. 9.4; SAS Institute, Cary, NC).