This was a retrospective cohort study of all adult patients (20 years or older) admitted as inpatients from January 2009-September 2015 to 373 well-distributed U.S. academic and community hospitals. Patients’ electronic health record (EHR) data was curated from 3 datasets: Cerner HealthFacts, Institute of Health Metrics, and HCA Healthcare, which have previously been described in detail (17). The study was approved by the institutional review boards at Harvard Pilgrim Health Care Institute (1301443).

We identified hospitalizations with sepsis using the Centers for Disease Control and Prevention’s (CDC’s) Adult Sepsis Event definition, which requires clinical indicators of presumed serious infection (blood culture order and new antibiotics continued for ≥4 days or until ≤1 day prior to death, discharge to hospice, or transfer to another hospital) and concurrent organ dysfunction (initiation of vasopressors or mechanical ventilation, elevated lactate, increase in baseline creatinine or total bilirubin, or decrease in baseline platelets) (17). The Adult Sepsis Event is modeled after Sepsis-3 criteria but is optimized for automated implementation using routine EHR data. This validated definition has previously been shown to have higher sensitivity than explicit sepsis diagnosis codes for chart-review confirmed Sepsis-3 with comparable positive predictive value, higher positive predictive value than implicit sepsis codes (i.e., infection and organ dysfunction codes) with comparable sensitivity, and is less susceptible than administrative definitions to variable and changing diagnosis and coding practices over time (17, 18).

We identified patients with opioid-related (e.g., opioid use disorder and/or overdose) diagnoses using International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes as per the Injury Surveillance Workgroup (ISW7) consensus document (Table E1, Supplemental Digital Content 1) (19). We used the broadest definition, drug poisonings associated with opium, heroin, and opioid analgesics or opioid adverse effect, or abuse/dependence to identify acute adverse events from opioid use or the presence of underlying opioid use disorder. Both principal and secondary diagnosis codes were included to maximize sensitivity (20). We augmented diagnosis-based case-finding by including inpatient administrations of buprenorphine as an additional indicator of opioid-related hospitalizations. We did not include methadone due to the difficulty in distinguishing its use for treatment of chronic pain vs opioid dependence.

We conducted descriptive analyses of patients with opioid-related hospitalizations, including demographics, comorbidities (using the Elixhauser method) (21), ICU admission, length-of-stay, and discharge disposition. Among patients with sepsis, we characterized types of sepsis-associated organ dysfunction (based on the Adult Sepsis Event “eSOFA” criteria) (22), source of infection (based on previously published ICD-9-CM algorithms) (23), and positive blood cultures (based on microbiology records, and excluding common skin contaminants). All bivariate group comparisons were conducted using chi-square test and student t-test for categorical and continuous variables, respectively.

We assessed the association between opioid-related hospitalization and all-cause, in-hospital mortality in patients with sepsis using mixed-effects logistic models to adjust for baseline characteristics and severity of illness. Covariates included demographics (age, sex, and race), comorbidities, source of infection, hospital-onset sepsis (defined by the presence of all CDC criteria on hospital day 3 or later) (24), and indicators of severity-of-illness within the first 2 days of hospitalization including admission to the ICU, vasopressors, mechanical ventilation, and patients’ worst values for creatinine, anion gap, bilirubin, liver function tests, platelet count, hematocrit, and albumin. All covariates were included in the model as they were deemed a priori to be important potential confounders influencing the risk of sepsis-associated death. Missing laboratory data were assumed to be normal, as per commonly used calculations of severity-of-illness scores (25, 26). Hospitals were treated as random effects in the models. Regression analyses were run in each dataset separately and results were compiled using study-level meta-analysis (27). A sensitivity analysis was also conducted to evaluate in-hospital death or discharge to hospice as a combined outcome, and to evaluate the association between opioid-related hospitalization and in-hospital death stratified by specific ISW7 case definitions (all acute opioid poisonings, acute poisoning from analgesics, all acute and chronic poisonings, and buprenorphine).

Finally, we calculated trends in the annual proportion of cases associated with opioid-related hospitalization among sepsis patients and their in-hospital mortality rates, and for all hospitalizations from 2009–2015 (data from 2015 were limited to January-September). Different hospitals contributed data in different years; therefore, temporal trends were modeled using Poisson regression models, adjusting for hospital characteristics (institution, region, teaching status, bed count, and annual admissions) and case mix (median age of hospitalized patients, sex and race/ethnicity distributions, and proportion of ICU versus total admissions) (17). Generalized estimating equations were used to account for hospital-level clustering and correlations in the data over time. Adjusted rates were generated by creating binary indicators for each year in the model, with 2009 as the reference year. Percentages were presented as relative annual changes.

All analyses were conducted in R version 3.6 (r-project.org). For all analyses, p<0.05 was considered statistically significant and two-tail tests were used.

The study cohort consisted of 6,715,286 hospitalized adults, of whom 375,479 (5.6%) had sepsis, 130,399 (1.9%) had opioid-related hospitalization, and 8,764 (0.1%) had both sepsis and an opioid-related hospitalization (Figure E1, Supplemental Digital Content 1). Compared to patients with opioid-related hospitalization without sepsis (n=121,635/130,399; 93.3%), those with opioid-related hospitalization and sepsis (n=8,764; 7.2%) were older (mean 52.3 years vs 46.9 years), had more comorbidities (mean Elixhauser score 5.4 vs −0.5), higher rates of ICU admission (64.0% vs 14.3%), longer length of stay in the ICU (mean 6.7 vs 2.9 days) and hospital (mean 13.4 vs 5.9 days), and higher in-hospital mortality rates (10.6% vs 0.7%; p<0.001 for all comparisons) (Table E2, Supplemental Digital Content 1). Patients with sepsis met the opioid-related hospitalization case definition more frequently from any poisonings (28.5% vs 12%) and poisoning from analgesics (19.6% vs 8.6%), but less frequently from inpatient buprenorphine prescription (2.9% vs 6.1%) compared to opioid-related hospitalization patients without sepsis (p<0.001 for all comparisons).

Overall, amongst the 1,803 hospitalized patients with opioid-related hospitalization who died in-hospital, 928 (51.5%) had sepsis. This was higher than the prevalence of sepsis in patients without opioid-related hospitalization who died (n=59,433/163,278, 36.4%, p<.001).

Compared to sepsis patients without opioid-related hospitalization (n=366,715/375,479, 97.7%), sepsis patients with opioid-related hospitalization (n=8,764, 2.3%) were younger (mean age 52.3 vs 66.9 years), more likely to be white (80% vs 70.7%), and had fewer comorbidities (mean Elixhauser score 5.4 vs 10.5) (p<0.001 for all comparisons) (Table 1). The distribution of dysfunctional organs at sepsis onset was similar, except that mechanical ventilation and acute kidney injury were more common in sepsis patients with opioid-related hospitalization (35.9% vs 24.0% and 56.4% vs 46.8%, respectively).

Sepsis patients with opioid-related hospitalization had higher rates of pulmonary infections (54.2% vs 47.8%), skin and soft tissue infections (13.5% vs 10.1%), bone/joint infections (4% vs 2.8%), and endocarditis (4.7% vs 0.8%), but less frequently had genitourinary infections (23.5% vs 32.4%) and intra-abdominal infections (9.5% vs 14%) compared to sepsis patients without opioid-related hospitalization. Positive blood cultures were present in 1,484 patients with sepsis and opioid-related hospitalization (16.9%). The most common pathogens isolated were Staphylococcus aureus (45.6%), Streptococcus species (20.9%), and yeast species (9.8%); this distribution was different from sepsis patients without opioid-related hospitalization, in whom Escherichia species (23.8%) were most common, followed by Staphylococcus aureus (21.8%) and Streptococcus species (19.1%) (Table 2). Overall, percentages of gram-positive bloodstream infections and fungal infections were higher for sepsis patients with opioid-related hospitalization (68.9% vs 47.0% and 10.6% vs 6.4%, respectively), while gram-negative organisms were less common (28.3% vs 49.3%).

Sepsis patients with opioid-related hospitalization had higher rates of ICU admission (64.0% vs 51.7%) and longer hospital length of stay (13.4 vs 11.9 days) compared to those without OUD/overdose. However, they were more often discharged home (61.3% vs 43.7%) and more likely to leave against medical advice (3.7% vs 0.6%). Conversely, they were less often discharged to subacute facilities (21.3% vs 30.9%) or hospice (3.2% vs 6.4%) and had lower in-hospital mortality rates (10.6% vs 16.2%, crude odds ratio [OR] 0.60, 95% confidence interval [CI] 0.56–0.65). After adjusting for baseline characteristics and severity of illness, opioid-related hospitalizations remained associated with a lower risk of in-hospital mortality in patients with sepsis (adjusted OR 0.73, 95% CI: 0.60–0.79) (Table E3, Supplemental Digital Content 1). Results were similar when using a combined outcome of in-hospital death or discharge to hospice (adjusted OR 0.81, 95% CI: 0.75–0.85), and across the different ISW7 administrative definitions (Table E4, Supplemental Digital Content 1).

The overall prevalence of opioid-related hospitalization among patients with sepsis who died was low (n=928/60,361, 1.5%). However, the prevalence of opioid-related hospitalization was relatively high in sepsis patients under age 50 years who died (5.7% vs 1.0% in patients under vs over 50 years, p<.001).

From 2009–2015, the proportion of all hospitalizations associated with opioid-related hospitalization increased from 1.7% to 2.4% (relative increase 41.4%, 95% CI: 6.1%−88.4%, p=.02). Increases were also seen using a case-finding definition based only on diagnosis codes (without requiring inpatient administrations of buprenorphine) for acute opioid poisoning and both acute and chronic poisonings from illicit opioids and opioid analgesics (Figure 1A). The proportion of sepsis hospitalizations associated with opioid-related hospitalization using the primary case-finding definition also increased from 1.9% to 3.3% (77.3% relative increase, 95% CI 40.7%−123.5%, p<.001). Increases were also seen using case-finding definitions of opioid-related hospitalization based only on diagnosis codes (Figure 1B). Figure 1A and 1B also demonstrate that only a fraction of opioid-related hospitalizations were associated with buprenorphine use and that this fraction did not increase over time. Adjusted in-hospital mortality in sepsis associated with opioid-related hospitalizations decreased from 14.1% to 12.9%, but this relative decrease was not statistically significant (8.5%, 95% CI −38.8–36.8%, p=.67).