Association Between Opioids Prescribed to Medical Inpatients With Pain and Long-Term Opioid Use

Kaitlin E. Keenan, MD; Michael B. Rothberg, MD, MPH; Shoshana J. Herzig, MD, MPH; Simon Lam, PharmD, BCPS; Vicente Velez, MD; Kathryn A. Martinez, PhD, MPH

Disclosures

South Med J. 2021;114(10):623-629. 

In This Article

Methods

This was a retrospective cohort study using data from the Cleveland Clinic Health System, a large integrated health system including 10 academic and community hospitals in northeast Ohio. Eligible patients were hospitalized for an acute medical reason in 2016, based on admitting service specialty, including Internal Medicine, Hospital Medicine, and Family Medicine. Our sample included adult patients aged 18 years and older with a hospitalization of ≥2 days and who reported any pain score > 0 during their hospitalization. We excluded patients with prolonged hospitalizations (>31 days) because their experience is likely to be atypical. For patients who had multiple hospitalizations in 2016, we selected a random admission to use as their index hospitalization. Eligible patients had at least 1 outpatient encounter in the 12 months preceding the hospitalization and at least 1 primary care visit in the 6 to 12 months following the hospitalization to assess opioid exposure before and subsequent to the hospitalization. Because we assumed that opioid use patterns may differ for patients with readmissions, and because it would be difficult to account for opioids prescribed during the second admission, we excluded patients with readmissions within 30 days of their index hospitalization. We excluded surgical patients, critical care patients, cancer patients, pregnant patients, and patients transferred from outside hospitals. This study was approved by Cleveland Clinic's Institutional Review Board.

Measures

All of the information came from the electronic health record (EHR). Medication data in the EHR represent inpatient orders and outpatient prescriptions by physicians.

Opioid-related Measures. Prior Opioid Use: We defined opioid-naïve patients as those without any opioid exposure in the 12 months preceding admission,[20] according to their medication list in the EHR. This included both patient-reported medications and opioid prescriptions by physicians.

Hospital Opioid Exposure: We assessed opioid exposure related to the hospitalization in two ways. First, we assessed whether the patient received opioids during the hospitalization, defined as morphine milligram equivalents per day >0 at any point during the stay. Second, we assessed whether patients received outpatient opioid prescriptions at discharge.

Long-term use. We defined postdischarge long-term use as at least 2 prescriptions of ≥30 pills in at least 2 different months during the 6 to 12 months following hospitalization. If a patient had 2 prescriptions for opioids but both prescriptions occurred in the same month, then they were not considered a long-term user.

Covariates. From the EHR, we collected patient clinical characteristics, including self-reported pain scores (on a scale of 0–10, where 10 is greatest), history of chronic pain diagnosis or history of psychiatric diagnosis (independent dichotomous variables based on diagnosis codes in the medical record prehospitalization), and length of stay (LOS; days). We collected patient demographic data: age (years), sex (male, female), race (White, Black, and Other), and smoking status (current vs not). We also dichotomized patients by whether they had a single admission versus multiple admissions during the study period.

Statistical Analysis

We assessed the distribution of all of the measures and characterized them via means and standard deviations or proportions and frequencies, as appropriate. We described the proportion of patients who were opioid naïve versus opioid experienced at admission and the proportion in each group who received an opioid during their hospitalization. We conducted the rest of the analysis on opioid-naïve patients only. We examined bivariate comparisons between patient characteristics stratified by opioid receipt during the admission (during or at discharge) and opioid use 6 to 12 months later using χ 2 statistics or analysis of variance. We used mixed-effects logistic regression to assess differences in the odds of opioid receipt during the hospitalization by patient characteristics. For patients who received opioids during their hospitalization, we used mixed-effects logistic regression to assess differences in the odds of receiving an outpatient prescription for opioids at discharge. To assess difference in odds of long-term use, we used mixed-effects logistic regression, with opioid exposure associated with the hospitalization as a predictor (none, during the hospitalization but not at discharge, at discharge). To determine whether pain mediated the association between opioid receipt during a hospitalization and long-term use, we ran this model both excluding and including the last recorded patient-reported pain score during the hospitalization. All of the models accounted for clustering by hospital. The analysis was conducted in Stata 14.0 (StataCorp, College Station, TX).

Medical Record Review

To better understand why medical inpatients received opioids, we conducted a review of 100 randomly selected patient charts among those who received opioids during their hospitalization. We described documentation in the chart related to the opioid prescription.

processing....