Patient Outcomes After Opioid Dose Reduction Among Patients With Chronic Opioid Therapy

Sara E. Hallvik; Sanae El Ibrahimi; Kirbee Johnston; Jonah Geddes; Gillian Leichtling; P. Todd Korthuis; Daniel M. Hartung


Pain. 2022;163(1):83-90. 

In This Article


Data Sources

We assembled a data set of Oregon Medicaid beneficiaries' claims linked with Prescription Drug Monitoring Program (PDMP) pharmacy data and death certificate data from 2014 to 2017. The linked data set included Medicaid beneficiaries with any eligibility in the 4-year period and at least one opioid prescription or an opioid-related diagnosis (eg, poisoning, dependence, and adverse event). Analysts in the Oregon Public Health Division linked and deidentified the data sets.[19] All activities were approved by the Oregon Health and Science University and Oregon Public Health Division Institutional Review Boards.

Cohort Development and Exposures

Within this linked data set, we first identified patients with chronic opioid therapy (COT).[18] We identified opioids in PDMP data using the First Databank national drug code file's opioid therapeutic class.[32] COT was defined as 84 or more consecutive days with an opioid available (excluding buprenorphine). We used the prescription fill date and days' supply variable to estimate opioid availability for each day of the study period, then estimated each patient's average daily morphine milligram equivalent dose (MME) on every day of their consecutive use period using drug strength, MME conversion factor,[37] fill dates, and days' supply. Next, we applied a 28-day rolling average to smooth out day-to-day variation in MME because of overlapping medication fills. This calculation rolled forward, so for each day, we estimated the average daily dose of the 28 preceding days. Patients with at least 84 consecutive days (3 months) of opioid availability with an average daily MME of 50 or greater on each of those days were retained. We selected each patient's first episode of high-dose COT.

We retained only patients whose COT episode ended in 2014 or 2015 to allow sufficient follow-up. We considered the first day that the average daily dose dropped below 50 MME after at least 84 consecutive days as the end of COT and the index date with which we tracked subsequent dose changes. Although dose may vary over time, we only attempted to identify dose trajectories after the average daily dose dropped below 50 MME.

Following this index date, we categorized patients into 4 mutually exclusive groups (Figure 1). We first identified patients who discontinued opioid prescriptions in the year after high-dose COT, defined as 56 consecutive days without any opioid availability. Discontinuations were identified using the exact daily dosage because any immediate and prolonged stoppage would seem gradual using a rolling average approach. Among this group, we further categorized patients as having either an abrupt discontinuation or a dose reduction before discontinuation. The dose change calculation compared 2 28-day average MME values: the day immediately preceding the exact discontinuation date and day 29 days prior. Abrupt discontinuation was defined as a dose increase, stable dose, or reduction <50% in the 4 weeks before discontinuation. A dose reduction before discontinuation was defined as a reduction in average daily MME of ≥50% in the 4 weeks preceding discontinuation.

Figure 1.

Cohort development of Medicaid patients with abrupt discontinuation, dose reduction and discontinuation, dose reduction without discontinuation, or stable or increasing dose after the end of high-dose chronic opioid therapy.

Next, we identified patients who did not discontinue opioid prescriptions in the year after the end of their COT episode. These patients may have had days without opioids totaling less than 56 consecutive days. We categorized COT patients who did not discontinue opioid prescriptions as either having a dose reduction without discontinuation or a stable or increasing dose in the year after the end of their COT episode. A dose reduction without discontinuation was defined as a ≥50% reduction in average daily MME in any 4-week period after the end of the COT period but without ever discontinuing opioid prescriptions for 56 or more consecutive days. A stable or increasing dose without discontinuation was defined as a dose increase, stable dose, or reduction <50% in the year after the end of the COT episode without ever discontinuing opioid prescriptions for 56 or more consecutive days. Deceased patients with a discontinuation date on or after their date of death were placed in this last group.


Outcomes were assessed in a 12-month period starting 28 days before discontinuation for those who discontinued (abrupt discontinuation or dose reduction before discontinuation) and after the start of the dose reduction for those with a dose reduction who did not discontinue (Figure 2). For patients with a stable or increasing dose, the 12-month follow-up started at end of their COT episode (ie, the first day they fell below 50 MME per day). We measured outcomes using death certificate and Medicaid encounter data. To ensure complete capture of nonfatal suicide, overdose, and other related adverse events from claims data, patients were required to have continuous Medicaid enrollment throughout their follow-up period. Patients who died during the follow-up were included until their date of death.

Figure 2.

Timeline, from chronic opioid therapy to dose trajectory group to follow-up period.

We characterized the occurrence of 3 potential opioid-related events in the follow-up period: suicide, opioid overdose, and other opioid-related adverse events. We identified both fatal[11] and nonfatal[21] suicides and opioid overdose events using ICD9 and ICD10 diagnostic codes. Other opioid-related adverse events included a diagnosis indicative of adverse effects, opioid abuse, opioid dependence, and opioid use, unspecified as identified by ICD9 and ICD10 diagnostic codes in any setting (Table A, Appendix A, available at[22]

Because opioid discontinuation can be implemented for the purpose of clinically indicated buprenorphine induction, we also identified patients who filled at least one buprenorphine prescription during the 12-month follow-up period.

Analyses and Covariable Adjustment

We tabulated the number of patients with each event type according to the dose change group and used chi-square tests to identify statistically significant differences. We used Cox proportional hazard models to evaluate the differences in time-to-event risk of any opioid-related event (fatal or nonfatal suicide, fatal or nonfatal overdose, and other adverse events) between the dose change groups. The proportional hazards assumption was tested and met based on the graphed Schoenfeld residuals for predictors and covariables. We adjusted tied data using the Efron approximation. We also modeled the odds of filling a buprenorphine prescription after COT using logistic regression. All models were adjusted patient demographics, baseline COT characteristics, and comorbidities.

Demographics included age, race/ethnicity, and rural/urban status. Patient zip code was used to define rural or urban residence according to the Oregon Office of Rural Health.[20] We characterized baseline COT for patients filling opioid and benzodiazepine prescriptions during their COT episode. Specifically, we computed the average MME per day during COT, identified those with concurrent benzodiazepine use, and having multiple providers (4 or more prescribers or 4 or more pharmacies in any six-month period during the COT episode). These variables have been associated with increased risk of poorer opioid-related outcomes.[2,12,36]

We used diagnostic codes appearing in Medicaid claims in the 3 months before the end of the index COT episode to characterize any drug abuse, depression, alcohol abuse,[14] or chronic pain[30] comorbidities (Table B, Appendix A, available at

P values were considered significant at P < 0.05. Data management and analyses were performed using SAS 9.2 (SAS Institute Inc, Cary, NC).