Opioid Overdose Risk Following Hospital Discharge Among Individuals Prescribed Long-Term Opioid Therapy: a Risk Interval Analysis

Abstract

Background

Individuals prescribed long-term opioid therapy (LTOT) have increased risk of readmission and death after hospital discharge. The risk of opioid overdose during the immediate post-discharge time period is unknown.

Objective

To examine the association between time since hospital discharge and opioid overdose among individuals prescribed LTOT.

Design

Self-controlled risk interval analysis.

Participants

Adults prescribed LTOT with at least one hospital discharge at a safety-net health system and a non-profit healthcare organization in Colorado.

Main Measures

We identified individuals prescribed LTOT who were discharged from January 2006 through June 2019. The outcome was a composite of fatal and non-fatal opioid overdoses during a 90-day post-discharge observation period, identified using electronic health record (EHR) and vital statistics data. Risk intervals included days 0–6 after index and subsequent hospital discharges. Control intervals ranged from days 7 to 89 after index discharge and included all other time during the observation period that did not fall within a risk interval or time readmitted during a subsequent hospitalization, which was excluded. Poisson regression was used to estimate incidence rate ratios (IRR) and 95% confidence intervals (CI) for overdose events during risk in comparison to control intervals.

Key Results

We identified 7695 adults (63.3% over 55 years, 59.4% female, 20.3% Hispanic) who experienced 9499 total discharges during the study period. Twenty-one overdoses occurred during their observation periods (1174 per 100,000 person-years [9 in risk, 12 in control]). Overdose risk was significantly higher during the risk interval in comparison to the control interval (IRR 6.92; 95% CI 2.92–16.43).

Conclusion

During the first 7 days after hospital discharge, individuals prescribed LTOT appear to be at elevated risk for opioid overdose. Clarifying mechanisms of overdose risk may help inform in-hospital and post-discharge prevention strategies.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11606-022-08014-1.

INTRODUCTION

In the USA in 2020, over 142 million opioid prescriptions were dispensed and 16,416 Americans died from an overdose event involving a pharmaceutical opioid, a 16% increase from 2019.^1–3 That number does not include the tens of thousands of individuals estimated to suffer non-fatal overdoses each year.^4,5 Individuals prescribed long-term opioid therapy (LTOT), defined as daily use of pharmaceutical opioids for 90 days or more, are at an increased risk of opioid overdose and death.^6–8 While several practice guidelines have been developed aiming to improve the safety of opioid prescribing and reduce the risks of LTOT in outpatient settings,^9–13 less attention has focused on risks during and after hospitalization.

Inpatient settings are frequent and important points of contact for individuals prescribed LTOT. Those prescribed LTOT utilize emergent healthcare services and require hospitalization at higher rates than those not prescribed LTOT.^14–16 In addition, individuals who are prescribed LTOT experience poor health outcomes after hospital discharge with an increased risk of readmission and all-cause death in comparison with individuals not prescribed LTOT.^16–18 Whether these outcomes are in part attributable to opioid overdose, the overall risk of opioid overdose in the post-discharge period and whether there is a temporal association between hospital discharge and overdose is unknown.

The aim of this study was to estimate the risk of opioid overdose in the post-discharge period and assess the association between pre-defined time intervals after hospital discharge and opioid overdose risk using a self-controlled risk interval design. We hypothesized that the immediate post-discharge period would be associated with an increased risk of opioid overdose when compared to more distant post-discharge time periods.

METHODS

Study Design

We used a self-controlled risk interval analysis (RIA) to compare incidence rates of opioid overdose in pre-defined risk and control intervals. In this approach, only exposed individuals are included in the study and individuals serve as their own controls with comparisons made between different time periods of hypothesized risk (e.g., risk and control intervals). A self-matched design implicitly controls for time-invariant characteristics^19–21 such as sex, race, ethnicity, and other unidentified factors that might otherwise contribute to residual confounding often experienced in cohort and case control studies.^{22, 23} This approach is particularly useful when an appropriate control group is not readily available due to significant differences in confounders such as the case with individuals prescribed LTOT.^{15, 24, 25} Additionally, because only individuals with a discharge exposure were included, selection bias that may be introduced when comparing hospitalized to non-hospitalized individuals was limited^19–21

Study Setting and Population

This retrospective study was conducted using data from a safety-net health system and a non-profit healthcare organization in Colorado. Both systems offer a comprehensive and integrated healthcare plan and delivery system, jointly serving approximately 730,000 Coloradans. We included individuals 18 years and older who were prescribed LTOT and had at least one hospital discharge between January 1, 2006, and June 30, 2019. Data from the COVID-19 pandemic era were not included due to numerous changes in hospital admission and practice that occurred. The Kaiser Permanente Colorado (KPCO) Institutional Review Board (IRB) approved this study with Colorado Multiple Institutional Review Board ceding to KPCO’s for IRB oversight. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines were followed.

Data Sources and Definitions

Demographic and clinical characteristics were identified via electronic health record (EHR) and health plan claims data, including age, gender, race, ethnicity, and medical and mental health comorbidities. Comorbidities were assessed in the year prior to the admission of index hospitalization and were identified using International Classification of Diseases, Ninth Revision (ICD-9) and International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) diagnosis codes. Charlson Comorbidity Index scores were calculated according to established methods.²⁶

LTOT was defined as three or more opioid dispensings in a 90-day period with at least and 80-day supply and less than a 5-day gap between prescription fills.²⁷ Methadone dispensed through a pharmacy and transdermal buprenorphine products were included as they are typically used for pain. Methadone dosed through an opioid treatment program and non-transdermal buprenorphine products were excluded because they are primarily used to treat addiction disorders at our study sites. Pharmacy dispensings were identified using National Drug Codes from outpatient pharmacy dispensing records and billing claims. Morphine milliequivalents (MME) were calculated for each opioid dispensing and the average daily MME was determined for the 90-days preceding index hospital admission using an established conversion factor (Supplementary Table S3).^{7, 28, 29}

The index hospital discharge for individuals was identified using EHR data and defined as the first hospital discharge during the study period for which the individual was (1) prescribed LTOT in the 90-day period immediately preceding the date of admission, (2) admitted for ≤ 32 days, (3) alive at the time of discharge, and (4) discharged to an independent living environment (e.g., home, shelter).

Primary Outcome

The primary outcome was a composite of fatal and non-fatal opioid overdoses during a 90-day post-discharge observation period. Non-fatal overdoses due to opioids were identified using ICD-9 and ICD-10 codes from ensuing emergency department and inpatient encounters (Supplementary Table S4). Fatal opioid overdoses were identified using underlying cause of death ICD-10 codes for drug poisoning and a contributing cause of death code indicating opioid involvement. To ensure completeness, vital status was also linked to the National Death Index Plus using patient identifiers.³⁰ Fatal and non-fatal opioid overdoses included overdoses related to pharmaceutical opioids, heroin, and synthetic opioids; some may have involved additional substances.

Risk and Control Intervals

The date of discharge from the index hospitalization defined the start of individuals’ 90-day observation periods. The observation period represents the overall person-time available to identify exposures (hospital discharge) and events (overdose) and was partitioned into two person-time intervals: (1) risk interval and (2) control interval. We defined the risk interval as post-discharge days 0–6 after the index and any subsequent hospital discharges during the observation period. The control interval ranged from post-discharge days 7–89 and included all other time during the observation period that did not fall within a risk interval or subsequent hospitalization. Time hospitalized was excluded from the analysis. Because eligible individuals could experience more than one hospital discharge, data on all subsequent hospital discharges and overdose events during the observation period were included (Fig. 1). We defined these time intervals based on periods of increased risk observed among other populations.^31–33 Follow-up was censored at health plan disenrollment, death, disqualifying hospitalization (i.e., hospitalization >32 days or with discharge disposition to a structured living environment) or the end of the observation period, whichever occurred first.

Figure 1.

Risk interval design schematic (not to scale): 90-day observation period by different hypothetical patient scenarios. Patient A: One overdose event in the 0–6-day risk interval after hospital discharge. No other hospitalizations or overdose events during the observation period. Patient B: Two overdose events in the control interval. Patient C: Subsequent hospitalization discharge exposure during the observation period with an overdose event in the 0–6-day risk interval after subsequent discharge. *End of the observation period: Follow-up censored at health plan disenrollment, death, disqualifying hospitalization (i.e., hospitalization >32 days or with discharge disposition to a structured living environment) or the end of the observation period, whichever occurred first. ^†Time hospitalized: Subsequent hospitalization during observation period; time spent hospitalized excluded from analysis.

Statistical Analysis

We first estimated the risk of overdose events during the entire observation period. A conditional Poisson regression model was then used to estimate the incidence rate ratios (IRR) and corresponding 95% confidence interval (CI) for opioid overdose events during the risk interval as compared to the control interval. Secondary analyses used 0–13 days and 0–29 days risk intervals.

RESULTS

Of 33,624 individuals who were prescribed LTOT from January 2006 through June 2019, 7695 had an index hospital discharge. Among those individuals, an additional 1804 subsequent hospital discharges occurred during the observation period and were included in the final analysis (n = 9499 total discharges) (Fig. 2).

Figure 2.

Hospital discharge exposures. *Index hospital discharge: Corresponding hospitalization where patient (1) prescribed LTOT in the 90-day period preceding admission, (2) admitted for ≤32 days, (3) alive at the time of discharge, and (4) discharged to an independent living environment. ^†Subsequent hospital discharges: Hospital discharges that occurred within 0–89 days after the index hospital discharge and prior to the censoring date.

Table 1 shows demographic and clinical characteristics of the population by study site and combined. Overall, most were female (59.4%) and over 55 years of age (63.3%). One-fifth (20.3%) of the sample identified as of Hispanic ethnicity; 7.9% were Black, and 77.0% were White. Mean length of hospital stay was 3.67 (3.27 SD) days. In the year preceding index hospital discharge, 474 (6.1%) had an opioid use disorder diagnosis. The mean daily MME for individuals in the 90-days preceding the index hospitalization was 76.2 mg.

Table 1.

Demographic and clinical characteristics of patients with an index hospital discharge

Characteristics	Patients with an index hospital discharge
	Site 1 n = 6007	Site 2 n = 1688	Total n = 7695
Sex, n (%)
Female	3715 (61.8)	856 (50.7)	4571 (59.4)
Age* (years)
<=45	784 (13.1)	344 (20.4)	1128 (14.7)
46–55	1081 (18.0)	612 (36.3)	1693 (22.0)
56–65	1572 (26.2)	509 (30.2)	2081 (27.0)
>65	2570 (42.8)	223 (13.2)	2793 (36.3)
Race, n (%)
White	4589 (76.4)	1336 (79.2)	5925 (77.0)
Black	302 (5.0)	308 (18.3)	610 (7.9)
Other	411 (6.8)	27 (1.6)	438 (5.7)
Unknown	705 (11.7)	17 (1.0)	722 (9.4)
Hispanic ethnicity, n (%)	794 (13.2)	768 (45.5)	1562 (20.3)
Insurance†, n (%)
Medicaid	626 (10.4)	466 (27.6)	1092 (14.2)
Charlson Comorbidity Index‡
Mean (SD)	2.43 (2.5)	2.28 (2.4)	2.40 (2.5)
Median	2.0	2.0	2.0
Morphine equivalents, mg§
Mean (SD)	77.90 (111.2)	70.04 (143.0)	76.18 (118.9)
Median	43.0	35.7	41.3
Length of hospital stay, days
Mean (SD)	3.65 (3.2)	3.75 (3.6)	3.67 (3.3)
Diagnosis of, n (%)
Substance use disorder‡
Alcohol	424 (7.1)	352 (20.9)	776 (10.1)
Opioid	369 (6.1)	103 (6.1)	472 (6.1)
Stimulant	54 (0.9)	125 (7.4)	179 (2.3)
Tobacco	1713 (28.5)	982 (58.2)	2695 (35.0)
Mental health disorder‡
Schizophrenia	208 (3.5)	78 (4.6)	286 (3.7)
Mood disorders including bipolar	367 (6.1)	256 (15.2)	623 (8.1)
Anxiety	1965 (32.7)	492 (29.2)	2457 (31.9)
Depression	2560 (42.6)	753 (44.6)	3313 (43.1)
Any mental health disorder	3371 (56.1)	987 (58.5)	4358 (56.6)

*At index hospital discharge date

^†At the start of prescription LTOT

^‡Assessed year prior to index hospital discharge

^§Assessed 90 days prior to admission for corresponding index hospital discharge

We identified 21 overdose events that occurred during the observation period (1,174 per 100,000 person-years), two of which were fatal. Nine occurred in the risk interval of 0–6 days after discharge (5143 per 100,000 person-years) and 12 occurred in the control interval (744 per 100,000 person-years). The risk of overdose was significantly higher during the risk in comparison to the control interval of 7–89 days after discharge (RR 6.92; 95% CI 2.92–16.43). Secondary analyses examining 0–13 and 0–29-day risk intervals demonstrated similar findings (0–13 day [RR 3.89; 95% CI 1.65–9.16]; 0–29 day [RR 4.00; 95% CI 1.55–10.32]) (Table 2).

Table 2.

Number of opioid overdoses (OD), OD rate per 100,000 person-years (p-y), and incidence rate of opioid overdose (OD) comparing risk interval to control interval during up to 90 days after index hospital discharge

	Risk interval			Control interval			Incidence rate ratio (95% CI)
	OD events, n	Person-years (p-y)	OD rate per 100,000 p-y	OD events, n	Person-years (p-y)	OD rate per 100,000 p-y
Primary analysis
0–6 days	9	175	5143	12	1614	744	6.92 (2.92 16.43)
Secondary analyses
0–13 days	10	339	2950	11	1450	759	3.89 (1.65 9.16)
0–29 days	15	688	2180	6	1101	545	4.00 (1.55 10.32)

DISCUSSION

In this large, diverse multisite sample of individuals prescribed LTOT, we conducted a self-controlled risk interval analysis and identified an increased risk of opioid overdose in the immediate period after hospital discharge. The risk of overdose was highest in the first 7 days after hospital discharge and remained elevated for at least 30 days. To our knowledge, this is one of the first studies to provide information on this underrecognized, high-risk time period specific to individuals prescribed LTOT, demonstrating that poor health outcomes after hospital discharge may be in part attributable to overdose events.

Although the relative risk of overdose was nearly 7-fold higher in the first 7 days after discharge, the absolute risk we observed may appear small. For comparison, however, a study conducted among patients prescribed LTOT who were receiving care within an integrated healthcare system in Washington State in the early 2000s found that their annualized overdose rate was 148 per 100,000 person-years — an absolute risk nearly 35-fold lower than what we observed during the first week after discharge (5143 per 100,000 person-years) among our study population and 8-fold lower than the overall overdose rate (1174 per 100,000 person-years) observed across the observation period.⁷

To reduce risk, a better understanding of the pathway between hospital discharge and overdose is needed. Prior research involving opioid-exposed populations suggests that overdose risk is increased after periods of restricted access to opioids or dose fluctuations, including after periods of incarceration,^{31, 34, 35} hospitalization in drug treatment clients,^{32, 33} after dose titration in outpatient settings,^{36, 37} and after initiating or discontinuing medications for opioid use disorder.^{38, 39} Direct comparisons between these study groups and our study population are difficult because of the inherent differences among them, including in access to opioids and other risk factors for overdose. However, these studies offer insights into reasons why risk may be elevated in the post-discharge period among patients prescribed LTOT and suggest fluctuations in opioid dependence and tolerance are important for understanding mechanisms involved in overdose events. For example, reductions in opioid dose prior to, during, or after hospitalization could reduce opioid tolerance and thus increase overdose risk if individuals return to their previously tolerated regimens, or if they seek additional or higher risk opioids (e.g., illicit) due to the onset of withdrawal symptoms. Alternatively, opioid dosing could be increased or medications that potentiate their sedating effects, such as benzodiazepines, could be added during hospitalization. The onset or worsening of clinical conditions might impact opioid metabolism (e.g., liver or renal failure) or make previously tolerated regimens riskier, such as in the case of respiratory conditions that may make one more vulnerable to the respiratory depressant effects of opioids. Finally, the psychosocial stressors related to hospital discharge, such as worsened health status, loss of housing, or negative impacts on finances and employment, could lead individuals to engage in greater risk behaviors.^{27, 40, 41} These pathways could impact the susceptibility to overdose from both prescribed and non-prescribed opioids, such as heroin or fentanyl. Future research is needed to examine the specific mechanisms that contribute to overdose among these patients to inform risk reduction strategies that might be employed during or soon after discharge.

Most efforts to improve safe opioid prescribing have focused on outpatient settings.^9–13 However, prior studies have demonstrated wide variation in in-hospital opioid prescribing practices^42–44 and that hospital prescribing decisions may be influenced by a variety of factors, including provider discomfort, prior negative prescribing experiences, concerns about patient satisfaction scores, or to facilitate discharge or prevent readmission.^{45, 46} Along with this prior work, our findings suggest that inpatient recommendations or guidelines for LTOT management, especially around the time of discharge, are needed.⁴⁷

To date, there is little research on interventions that may be implemented during hospitalization or at the time of discharge to prevent opioid overdoses among patients who are prescribed LTOT. Areas for future study could include investigating the effectiveness of patient education initiatives on known risk factors for overdose such as dose variability and polypharmacy, screening for opioid use disorder prior to discharge and initiating treatment if indicated, providing take-home naloxone, and avoiding co-prescribing with benzodiazepines. Studies could also investigate improving care transitions by attending to immediate post-discharge needs. This could include direct communication between inpatient and outpatient provider teams, access to expedited follow-up care including telemedicine options, pharmacy medication reconciliation, and transportation services.

This study has several limitations. First, despite access to a large population of individuals prescribed LTOT, because overdose events were rare, the confidence intervals associated with our estimates were wide. Even so, the lower limit of our confidence interval for our primary analysis suggests nearly a 3-fold higher risk of opioid overdose in the 7-day post-discharge time period, a clinically relevant and statistically significant result. Additionally, because we used a self-controlled study design, all fixed intrapersonal covariates that did not vary over time (e.g., race, gender, time-invariant medical comorbidities) were implicitly controlled; limiting bias despite the small number of events. Sociodemographic and clinical correlates of opioid overdose among similar populations prescribed LTOT have been well described in prior studies and include age, gender, opioid dose and formulation, co-prescription with other sedating medications, concurrent substance use disorders, and other mental health and chronic medical comorbidities.^48–52 Future work is needed to consider acute illness, hospitalization and discharge characteristics to now identify risk factors of opioid overdose specific to the time period after hospital discharge and to develop targeted interventions to reduce the risk.

Second, the overdose outcomes were identified using EHR and National Death Index data; thus, we were able to identify fatal and non-fatal events that came to the attention of the healthcare system and were coded as overdoses. Only two fatal overdose events were identified and 19 non-fatal overdoses, likely a reflection of the known lower rates of fatal of events than non-fatal events.^{4, 5} This case identification approach risks missing non-fatal events that were not recognized in the formal healthcare setting, for example those treated in the community setting with take home naloxone. We do not have reason to believe that these false negative results were distributed differentially between risk and control intervals. Importantly, community-based naloxone distribution is a legitimate, effective intervention that was made available to that population. The chief contribution of our study is to highlight a risk period for which no interventions were available and thus an overdose event led to an Emergency Department visit, hospitalization or death; further distribution of naloxone to our study population may prove to be an effective means to reduce rates of readmission and death among individuals prescribed LTOT.

Third, by utilizing pharmacy and claims data, we were able to identify individuals prescribed LTOT and calculate average daily MME in the 3-month period preceding index hospitalization. During that time, individuals had at least an 80-day supply of opioids and the mean daily dose was 76.2 MME, suggesting daily access to pharmaceutical opioids and a relatively high level of tolerance, but the available data could not be used to measure continuous use or tolerance. Data were also lacking to identify changes in opioid dose that may have occurred immediately prior to, during or just after hospitalization or changes to co-prescriptions of other sedating medications such as benzodiazepines, which may be relevant to overdose risk in the discharge period.

Lastly, we used risk interval analysis to control for time-invariant characteristic among patients. However, this approach does not eliminate the potential for residual confounding from time-varying characteristics.

In summary, individuals prescribed LTOT have an increased risk of overdose in the 30-day time period after hospital discharge and are most vulnerable to overdose in the 7 days after hospital discharge. Future work is needed to better understand the mechanisms underlying this risk, and whether and how in-hospital and post-discharge clinical management might influence that risk.

Funding

Funding for this study was provided by the National Institute on Drug Abuse of the National Institutes of Health under Award Number R01DA047537. The content is solely the authors’ responsibility and does not necessarily represent the official views of the National Institutes of Health.