Annals of Internal Medicine | 2019
Patterns of Opioid Administration Among Opioid-Naive Inpatients and Associations With Postdischarge Opioid Use
Abstract
The opioid epidemic places a significant burden on families, communities, and health systems across the United States (1, 2). Prescription and illicit opioids are responsible for the highest drug overdose mortality rates ever recorded, accounting for 63600 deaths in 2016 (3). Health system and policy interventions have largely focused on reducing and monitoring high-risk opioid prescribing (4). Prescription drug monitoring programs seek to mitigate aberrant prescribing and patient misuse (4). The 2016 Centers for Disease Control and Prevention guidelines for opioid prescribing for chronic pain (5) have reduced prescribing volume and some high-risk prescribing (6), and payers have instituted measures to promote better prescribing (7, 8). These initiatives have largely focused on outpatient prescribing (9). Despite the Joint Commission s standards for inpatient pain management being implicated as a driver of increased opioid use, opioid prescribing in inpatient settings has received far less empirical evaluation (10). Recent studies reported that up to 13% of opioid-naive patients hospitalized for surgical procedures use opioids as outpatients for extended periods after surgery (1119). Research on opioid prescribing after nonsurgical hospitalization reported similar rates of persistent use (20). One study of more than 1 million nonsurgical admissions to 286 U.S. hospitals found that opioids were used 51% of the time (21). However, little is known about the timing, duration, and inpatient setting of opioid administration or whether particular inpatient patterns of use correlate with long-term use after discharge. To fill these knowledge gaps, we linked inpatient and outpatient electronic medical record data from a large health system between 2010 and 2014 to examine inpatient opioid administration among opioid-naive patients. We addressed 3 questions. First, who is most likely to receive opioids during a hospital stay? Second, when, where, and for how long are opioids used during hospital stays? Third, which patterns of opioid prescribing are associated with continued use after discharge? Methods Data Source We obtained data from the UPMC Health System, an integrated delivery and financing system that includes academic and community hospitals and accounts for 41% of hospital admissions in western Pennsylvania. We obtained hospital discharge data from 2010 to 2014 from 12 UPMC-affiliated community and academic hospitals. System-wide implementation of outpatient and inpatient electronic health records was complete before our study period. We linked inpatient electronic health record data (CERNER) to outpatient records (Epic Systems) using the health system s enterprise master patient index. We obtained the following data on inpatient encounters: demographic characteristics (age, sex, and race); diagnosis codes; admission type (medical vs. surgical); length of stay; intensive care unit (ICU) stay; source of insurance coverage; in-hospital mortality; and medication administration data, including drug name, route of administration, setting, and day and hour of administration. Information on prior home medications was collected by clinical personnel at the time of admission through patient or caregiver report. When a home medication list was not recorded, medication lists recorded in the discharge summary of a prior hospital stay were used when available. From the outpatient encounter database, we obtained information on outpatient visits before and after the index hospital stay along with outpatient medication lists. Medication data in the outpatient database represented the reconciled medication information that providers review during each visit and captured both active medications reported by patients and new prescriptions ordered by providers at outpatient visits. Study Sample Our sample included adults (aged 18 years) admitted to study hospitals from the emergency department, home, or transfers between 2010 and 2014 (Appendix Figure). We included admissions for opioid-naive patients, defined as those with no documented opioid use in the inpatient and outpatient encounter databases in the prior 12 months. We excluded admissions that were for deliveries, those occurring less than 90 days after a previous admission (we included only the first admission in such cases to allow for complete measurement of outpatient medication use), those that were missing complete inpatient medication records, and those for patients with no outpatient encounter in the 12 months before and after the admission (to ensure that we could measure outpatient opioid use after discharge). We included admissions for patients with diagnoses that typically prompt opioid treatment (such as burns, major trauma, and advanced cancer) because our objective was not to adjudicate appropriateness of opioid prescription but to analyze associations between inpatient use patterns and long-term use. Appendix Figure. Study flow diagram. Inpatient and Postdischarge Opioid Use Opioids were identified in inpatient and outpatient encounter databases using the same medication list (Appendix Table 1). We excluded partial opioid agonists that are approved for opioid use disorder treatment, such as buprenorphinenaloxone, from our definition of opioid use. Appendix Table 1. Types We constructed 3 outcomes related to opioid use. We measured the number of days on which any opioid was administered during the hospital stay, excluding perioperative use (24 hours after surgery). Almost all patients admitted for surgery received perioperative opioids, but use varied thereafter. The other 2 measures captured postdischarge use. Using outpatient records, we constructed dichotomous indicators for any opioid use recorded within 90 and 365 days after discharge. Opioid use recorded in the outpatient encounter database could represent self-reported use or prescriptions ordered during the encounter. We could not observe prescription fills in pharmacies by using the outpatient database. Patterns of Inpatient Opioid Administration To understand associations between patterns of inpatient analgesia (opioid and nonopioid) and subsequent outpatient use, we created measures reflecting the presence, timing, duration, and location of opioid and nonopioid analgesia administration during each hospital stay. These indicators included the number of calendar days of a stay with any opioid administration, timing of the last opioid administration relative to discharge (in hours), location of first use (for example, emergency department, ICU, or ward), and use of nonopioid analgesics (such as nonsteroidal anti-inflammatory drugs or acetaminophen) (Appendix Table 1). Covariates On the basis of prior research and clinical judgment, we selected several patient- and discharge-level characteristics to include as covariates in our multivariable regression models. These included patient sex, age (as an ordinal variable), and race (white, black, or other); calendar year of admission (to account for time trends in opioid prescribing); hospital indicators (to account for differences in case mix and prescribing differences by hospital); payment source (commercial, Medicare, Medicaid, other, or uninsured) as a proxy for socioeconomic status; a 4-category variable that combined type of admission (medical or surgical) with whether the stay involved an ICU stay; Healthcare Cost and Utilization Project Clinical Classifications Software categories based on the International Classification of Diseases, Ninth Revision (22), which we combined into 9 variables to capture reason for admission; Elixhauser Comorbidity Index score (range, 0 to 30); several comorbid conditions previously found to be associated with opioid use, including musculoskeletal pain, depression, other mental disorders, alcohol use disorders, drug dependence, and opioid poisoning (Supplement Table 1) (23); and an indicator for whether the patient had used benzodiazepines at home in the 12 months before admission. Length of hospital stay was included in statistical models as either an offset term or a covariate depending on the outcomes of interest. In-hospital mortality was obtained from inpatient records, and postdischarge mortality (treated as a competing event) was measured using monthly data that UPMC obtains from the Social Security Administration. Supplement. Supplementary Material Statistical Analysis Admission characteristics were summarized with descriptive statistics (mean and SD or median and interquartile range for continuous variables and frequency and percentage for categorical variables). We then conducted 3 analyses among opioid-naive hospitalized patients. First, we examined factors associated with inpatient opioid use (excluding the perioperative period) using multivariable Poisson models, where the dependent variable was the number of days an opioid was administered in the hospital, length of stay (in days) was log-transformed and treated as an offset term, and all covariates described earlier were included to adjust for case mix. We assessed model fit and found no evidence of overdispersion of the data and thus considered a Poisson model to be appropriate. We included hospital fixed effects to account for hospital-level confounding and robust SEs with clustering by patient to account for those with multiple hospital stays. We report adjusted incidence rate ratios (IRRs), 95% CIs, and P values for each covariate. Second, we fit 2 multinomial logistic regression models to examine the association between any inpatient opioid administration (excluding perioperative use for surgical patients) and any outpatient use within 90 or 365 days after discharge. To account for censoring of the outcome due to postdischarge death or readmission, we constructed the outcome variable as 4 mutually exclusive categories: presence of outpatient opioid use, absence of outpatient opioid use (without death or readmission), readmission without outpatient op