Epidemiology of Substance-Related Admissions to Intensive Care Units in the United States

Abstract

Objective:

To determine the prevalence, demographic characteristics, and predictors of in-hospital mortality for substance-related ICU admissions in the United States.

Design:

Multicenter, retrospective cohort study.

Setting:

U.S. ICUs reporting data to the nationally representative Premier Healthcare Database between 2016 and 2019.

Patients:

Adult ICU admissions with an ICD-10 diagnosis of a substance-related disorder not in remission.

Interventions:

None.

Measurements and Main Results:

Among 4,740,799 ICU admissions, a substance-related diagnosis was present in 760,153 (mean age 51.8 years, 65.5% male, 73.7% White, 74.5% non-Hispanic), representing 16.0% (95% CI 16.00%-16.07%) of all ICU admissions or approximately one out of every six ICU admissions. Alcohol was the most common substance associated with ICU admission (8.9% of all encounters, 95% CI 8.87%-8.92%), followed by opioids and stimulants (4.0%, 95% CI 3.97%-4.01%, and 2.9%, 95% CI 2.91%-2.94%, respectively). Rates of nearly all substance-related ICU admissions were higher in patients aged 55-64 years, and in patients who identified as male, non-Hispanic, and “Other” race (not identified as White, Black, or Asian). In comparing White and Black patients, the two largest racial groups within our cohort, opioid-related ICU admission rates were higher in White patients while stimulant-related ICU admission rates were higher in Black patients. Only 6.5% (95% CI 6.37%-6.60%) of opioid-related ICU admissions identified heroin use. In multivariable analysis adjusting for relevant covariates, odds of in-hospital mortality following a substance-related ICU admission were higher for encounters which included alcohol-related diagnoses (aOR 1.12, 95% CI 1.06-1.19), female sex (aOR 1.07, 95% CI 1.04-1.09), and non-Hispanic ethnicity (aOR 1.08, 95% CI 1.01-1.16).

Conclusions:

A substantial burden of substance-related diagnoses exists in patients admitted to ICUs. Alcohol use was most common and associated with increased hospital mortality relative to other substances. Rates of substance-related ICU admission differed by age, sex, race, and ethnicity. These findings may have implications for effective allocation of resources towards addiction-related diagnoses, treatment, and secondary prevention for ICU patients.

INTRODUCTION

In 2023, 46.3 million adults in the United States had a substance use disorder, a medical condition characterized by a problematic pattern of substance use despite adverse consequences [1, 2]. Substance use disorders are a leading cause of preventable death in the United States despite often being treatable [3, 4]. With the substantial rise in substance use disorders over the last decade, the burden of substance-related disease, injuries, and deaths have stretched our healthcare system with increased Emergency Department visits and hospital admissions [5].

Yet, little is known about the impact of substance misuse on critical illness or the prevalence of substance-related intensive care unit (ICU) admissions in the United States. Existing literature is limited to single, urban centers with brief study periods, reporting varied prevalence in substance-related ICU admissions from 13% to 23% [6–9]. Even less is known about patient demographics and the relationship between substance use and in-hospital mortality.

A better understanding of the prevalence, demographics, and mortality associated with substance-related ICU admissions is necessary to improve resource allocation, target treatments, and improve outcomes in the critical care setting. We therefore sought to describe the national prevalence and patient characteristics of substance-related ICU admissions, and to determine risk-factors for in-hospital mortality associated with these admissions.

MATERIALS AND METHODS

We conducted a multicenter, retrospective cohort study using the Premier Healthcare Database, with records from over 800 non-governmental, community and teaching hospitals in the United States representing approximately 11 million annual hospitalizations (approximately 25% of national inpatient discharges) [10]. The database includes patient characteristics, disease state classifications (International Classification of Diseases, Tenth Revision [ICD-10]), and discharge location, including whether a patient had died during the hospital encounter. To calculate population-based rates of substance-related ICU admissions across demographics, we used U.S. Census Bureau annual population estimates by sex, age, race and ethnicity [11].

We included adult ICU admissions (≥18 years) from January 1, 2016 to December 31, 2019, defined by ICU billing charges (Table E1). We excluded data collected during the COVID-19 pandemic due to observed differences in hospital data reporting and concerns for pandemic related alterations in care capacity and treatment seeking for substance use disorders. Encounters were defined as substance-related if they included an ICD-10 diagnosis related to alcohol or drug use. Alcohol-related diagnoses included those related to alcohol use (intoxication, abuse, withdrawal, and dependence), in addition to alcohol-related organ dysfunction (e.g., alcoholic cardiomyopathy, alcoholic hepatitis, alcoholic pancreatitis) [12, 13]. Drug-related diagnoses included those related to opioids, stimulants, cannabis, sedatives, inhalants, and hallucinogens, among others [12]. A full list of ICD-10 diagnosis codes is included in Table E2.

We collected patient demographics for each encounter, including age, sex, ethnicity (Hispanic, non-Hispanic) and race. Race was categorized by the Premier Healthcare Database using a pre-defined variable with four groups: White, Black, Asian, and “Other”. The “Other” race category included racial groups recognized by the U.S. Census which are non-White, non-Black, and non-Asian such as American Indian/Alaskan Native, Native Hawaiian/Pacific Islander, and individuals reporting two or more races. Additional covariates included year of admission, ICU type (medical or surgical), hospital length of stay, indicators of ICU severity (defined by billing codes for mechanical ventilation and vasopressor use), in-hospital mortality, insurance type, and hospital location, size, academic affiliation and urban or rural service area (Table E3). The database was not missing any variables of interest apart from “unknown” categories for gender, race, and ethnicity. Prior to modeling, the “unknown” observations for gender (n=107, 0.01%) were removed from the dataset. The “unknown” values for race and ethnicity were left in the dataset, and an effect was estimated for each of these “unknown” categories. The “Asian” (n=84,443, 1.8%) category for race was combined with the “Other” category. Hospitals with fewer than 10 encounters meeting inclusion criteria were combined and treated as a single hospital when modeling.

Statistical Analysis

We used summary statistics to describe the prevalence of substance-related ICU admissions. Chi-square tests compared baseline characteristics of non-substance-related and substance-related ICU admissions. Using Census Bureau annual population estimates by age, sex, race, and Hispanic ethnicity, we calculated ICU admission rates per 100,000 people for the three most common substances in our data (alcohol, opioids, and stimulants), as well as substance-related ICU admission rates for specific demographic groups. We excluded ICU admission rates for “unknown” gender, race, or ethnicity groups due to a lack of clinical relevance in comparing substance-use prevalence between known demographic groups. Inference on per-capita rates was conducted via Poisson generalized linear models. Due to the large sample size, all group-based comparisons are statistically significant, but not necessarily clinically relevant. Therefore, we focused only on differences in magnitude of ≥10% rather than p-values. To evaluate predictors of in-hospital mortality, we used binomial-family logit-link generalized estimating equations (GEE), adjusting for a priori defined confounders including characteristics of the patient (age, sex, race, ethnicity, Elixhauser score), admission (type of substance-related diagnosis, length of hospital stay, receipt of vasopressors or mechanical ventilation, medical vs surgical ICU, admission year, insurance payor) and hospital (geographic location, urban vs rural service area, number of beds, academic affiliation). The GEE used an exchangeable working correlation matrix to account for the non-independence of patients admitted to the same hospital. Unadjusted odds ratios (ORs) fit via logistic regression, as well as GEE-based adjusted odds ratios (aORs) and their 95% confidence intervals, were calculated for every variable in our model. Analyses were performed using R version 4.4.0 and SAS version 9.4 [14, 15]. This study was deemed exempt by the Colorado Multiple Institutional Review Board (COMIRB# 21-4056).

RESULTS

We identified 40,631,065 unique encounters occurring from 2016 through 2019. After excluding non-ICU encounters (n=33,706,399) and encounters for patients under 18 years of age (n=2,183,867), we further narrowed to 4,740,799 adult ICU admissions (Table 1). Patients admitted to an ICU with substance-related diagnoses were younger (mean age [SD]: 51.8 [15.4] vs. 65.1[16.4]) and more frequently identified as male (65.5% vs 51.7%) compared to those admitted without a substance-related diagnosis.

Table 1.

Baseline characteristics of ICU admissions

Characteristics	All ICU admissions (N = 4,740,799)	Non-substance-related ICU admissions (N = 3,980,646)	Substance-related ICU admissions (N = 760,153)
Age, years
18-24	2.7% (128,671)	2.3% (93,414)	4.6% (35,257)
25-34	5.3% (251,993)	4.0% (160,941)	12.0% (91,052)
35-44	6.9% (326,731)	5.5% (218,033)	14.3% (108,698)
45-54	12.4% (586,735)	10.6% (423,690)	21.4% (163,045)
55-64	21.3% (1,009,014)	20.2% (804,099)	27.0% (204,915)
65-74	24.0% (1,138,495)	25.8% (1,027,322)	14.6% (111,173)
>75	27.4% (1,299,160)	31.5% (1,253,147)	6.1% (46,013)
Sex
Male	53.9% (2,555,342)	51.7% (2,057,367)	65.5% (497,975)
Female	46.1% (2,185,035)	48.3% (1,922,964)	34.5% (262,071)
Race
White	75.0% (3,554,095)	75.2% (2,994,158)	73.7% (559,937)
Black	14.0% (661,501)	13.7% (547,140)	15.0% (114,361)
Asian	1.8% (84,443)	2.0% (77,648)	0.9% (6,795)
Other	7.3% (347,371)	7.1% (283,362)	8.4% (64,009)
Unknown	2.0% (93,389)	2.0% (78,338)	2.0% (15,051)
Ethnicity
Non-Hispanic	74.9% (3,550,610)	75.0% (2,984,225)	74.5% (566,385)
Hispanic	6.9% (327,109)	6.9% (275,987)	6.7% (51,122)
Unknown	18.2% (863,080)	18.1% (720,434)	18.8% (142,646)

^aAll differences between non-substance-related and substance-related ICU admissions were found to be statistically significant.

Overall, 16.0% (n=760,153) of ICU admissions included a substance-related diagnosis. Alcohol-related diagnoses were the most common at 8.9% (n=421,704), contributing to more than twice the ICU admissions compared to any other substance. In a sensitivity analysis limited to ICD-10 diagnosis codes for alcohol use (intoxication, abuse, withdrawal or dependence), thus mirroring the ICD-10 diagnosis codes for other substance categories, the prevalence of alcohol-related diagnoses was 8.0% (n=381,140) (Table E4). Opioid-related diagnoses were included in 4.0% (n=189,185) of ICU admissions, and stimulant-related diagnoses followed with 2.9% (n=138,834) of ICU admissions. Collectively, the remaining substance categories, including cannabis, sedatives, inhalants, hallucinogens, and unspecified substances, constituted 4.8% (n=228,220) of ICU admissions. Diagnosis codes for more than one substance were included in 3.5% (n=168,080) of ICU admissions.

Demographic differences among patients admitted to an ICU with a substance use diagnosis for the three most common substances (alcohol, opioids, and stimulants) are summarized in Figures 1–3 with ICU admission rates per 100,000 people per demographic group. Rates of substance-related ICU admissions differed by age group and were highest for patients aged 55-64 years (Figure 1). For patients 35 years of age or older, ICU admission rates were primarily associated with alcohol, which was two to three times the rate of ICU admissions for opioids or stimulants. Among younger patients (<35 years), differences in ICU admission rates for alcohol, opioids, and stimulants were less prominent. Substance-related ICU admission rates also differed by sex (Figure 2). Compared to women, men experienced twice the rate of overall substance-related ICU admissions, including two times higher stimulant-related and three times higher alcohol-related ICU admission rates.

Figure 1.

Substance-related ICU admission rates by age group

Figure 3.

Substance-related ICU admission rates by race and Hispanic ethnicity

Figure 2.

Substance-related ICU admission rates by sex

Substance-related ICU admission rates differed by Hispanic ethnicity and the racial groups identified within our cohort (Figure 3). Overall and across substance types, non-Hispanic patients had higher substance-related ICU admission rates than Hispanic patients. This difference was greatest for opioid-related ICU admissions, for which non-Hispanic patients were admitted at nearly three times the rate of Hispanic patients. Patients identified as “Other” race, which includes patients identified as a racial group other than White, Black, or Asian, had the highest rates of ICU admission across all substance types compared to other racial groups. The lowest rates of substance-related ICU admission compared to other racial groups were experienced by patients identified as Asian.

In comparing patients identified as Black or White, the two largest racial groups within our cohort, differences in ICU admission rates varied by substance type (Figure 3). While Black patients had higher rates of substance-related ICU admissions overall, White patients had higher opioid-related ICU admission rates. Notably, heroin diagnoses accounted for only 6.5% of all opioid-related ICU admissions, which was consistent among White and Black patients (6.4% vs 6.8%). Stimulant-related ICU admission rates, however, were the most discrepant between Black and White patients, with Black patients admitted at rates two times greater than White patients. This difference was largely due to cocaine, which accounted for 87.8% of stimulant-related ICU admissions in Black patients compared to 39.4% for White patients (Table 2).

Table 2.

Percentage of heroin and cocaine-related ICU admissions for White and Black patients

Substance	All races	White	Black
Opioids
Number of Heroin Related Admissions	12,268	9,601	1,425
Number of Total Opioid Related Admissions	189,185	151,036	20,863
% of Opioid Related Admissions due to Heroin	6.5%	6.4%	6.8%
Stimulants
Number of Cocaine Related Admissions	70,469	35,096	27,577
Number of Total Stimulant Related Admissions	138,834	89,065	31,395
% of Stimulant Related Admissions due to Cocaine	50.8%	39.4%	87.8%

Adjusting for patient, admission, and hospital characteristics, we found several significant predictors for in-hospital mortality among those with substance-related diagnoses during an ICU admission (Table 3). Of the three most common substances related to ICU admission, patients with alcohol-related ICU admissions had higher odds of in-hospital mortality compared to patients with stimulant-related ICU admissions (aOR 1.12, 95% CI 1.06-1.19), while opioid and stimulant-related admissions had similar odds of in-hospital mortality (opioid-related aOR: 1.05, 95% CI 0.99-1.12). Patients with a substance-related ICU admission had higher odds of death if identified as female sex (aOR 1.07, 95% CI 1.04-1.09) or non-Hispanic ethnicity (aOR 1.08, 95% CI 1.01-1.16), and lower odds of death if identified as Black race (aOR 0.82, 95% CI 0.79-0.85).

Table 3.

Adjusted odds ratios from GEE and unadjusted odds ratios from logistic regression for predictors of in-hospital mortality for substance-related ICU admissions

Predictor	Adjusted Odds Ratio	95% Confidence Interval	Unadjusted Odds Ratio	95% Confidence Interval
Substance Type (Ref. Stimulants)
Alcohol	1.12	1.06 - 1.19	1.37	1.32 - 1.42
Opioids	1.05	0.99 - 1.12	1.03	0.99 - 1.07
Age (decades since 18 years)	1.13	1.11 - 1.14	1.25	1.24 - 1.25
Female sex (Ref. Male)	1.07	1.04 - 1.09	0.90	0.88 - 0.91
Race (Ref. White)
Black	0.82	0.79 - 0.85	0.89	0.87 - 0.92
Other	1.05	1.00 - 1.10	1.26	1.23 - 1.29
Non-Hispanic ethnicity (Ref. Hispanic)	1.08	1.01 - 1.16	1.06	1.03 - 1.10

^a

Variables not listed in table 3:

• Substance type: polysubstance, cannabis, sedatives, other
• Hospital variables: U.S. census division, urban vs rural service area, size by number of beds, teaching vs non-teaching affiliation
• Encounter variables: Severity of illness, length of stay, insurance payor, medical vs surgical ICU, years since 2016
• Patient variables: elixhauser index, chronic pain index, psych index, drug use index

DISCUSSION

In a cohort of U.S. adults admitted to an ICU between 2016 and 2019, we found that 16%, or one in every six ICU admissions included a diagnosis related to substance use. The leading substance associated with these admissions was alcohol. Substance-related ICU admission rates were highest among patients identified as middle-aged, male, non-Hispanic, and “Other” race (not White, Black, or Asian). Black patients experienced higher rates of overall substance and stimulant-related admissions, and White patients experienced higher rates of opioid-related admissions. In-hospital mortality after a substance-related ICU admission was associated with alcohol-related diagnoses, female sex, and non-Hispanic ethnicity.

To our knowledge, this is the first study to describe the national prevalence of substance-related ICU admissions in the United States. Prior studies investigating substance use rates in patients presenting to the ICU were limited to single, urban hospitals over short time periods (e.g., months) [6–9]. These studies, which utilized diagnosis coding in addition to patient history and toxicology screens, reported slightly higher prevalence of substance-related ICU admissions at their respective centers compared to our national results. Given that our data is based on administrative coding which relies on provider recognition and recording of substance-related diagnoses, our 16% prevalence of substance-related ICU admissions is likely an underestimate. Studies show that ICD-10 coding consistently underestimates the true prevalence of substance use disorders, particularly alcohol use disorder (AUD). One study found that only 17.5% of hospitalized patients screening positive for AUD had an alcohol-related ICD-10 diagnosis code [16]. Another found that nearly one-third of AUD cases were missed by ICD coding [17]. In contrast, a study of hospitalized patients with opioid use disorder found that nearly 90% had an opioid-related discharge diagnosis, suggesting that underdocumentation may vary by substance [18]. These findings suggest that our results likely reflect a lower bound of the true prevalence of substance use disorders among ICU admissions.

Moreover, the ICU may represent a peak in substance use disorder prevalence across healthcare settings. National data show that 9.4% of Emergency Department visits and 11.9% of hospitalizations include a diagnosis of a substance use disorder, substantially lower than the prevalence we and others have observed in the ICU [5–9]. This supports a continuum of increasing substance use disorder prevalence with rising clinical acuity, emphasizing the ICU as a key setting for potential interventions.

Previous single center studies of substance-related ICU admissions largely represent urban, safety net hospitals with potentially higher prevalence of substance use disorders. Further research is needed to determine how the prevalence of substance-related ICU admissions differ by hospital type, including urban or rural service area and geographic location. A key strength of our study is the use of data from over 4 million ICU admissions across all hospital types to establish the best estimates for prevalence. Of these substance-related ICU admissions, we found that alcohol was the most commonly coded substance, constituting more ICU admissions than opioids and stimulants combined and remaining the most common substance across all age, sex, and racial and ethnic groups. As the substance use disorder epidemics continue to garner much needed attention from healthcare institutions, policy makers and the media, these results highlight the impact of substance use, particularly alcohol use, in our ICUs and opportunities for intervention.

Our findings reveal key differences in the impact of substance use on critical illness by age, sex, race, and ethnicity, adding to a growing body of literature that describes demographic differences in substance use patterns. Although substance use disorders are most common among young adults, we found that ICU prevalence of substance use diagnoses was highest among adults aged 55 to 64 years [1]. This aligns with data demonstrating the steepest rise in alcohol-related hospitalizations and drug overdose deaths among older adults over the past two decades [19, 20]. These findings emphasize the need to focus critical care and substance use resources on older adults who may bear the most severe substance-related illness.

Additionally, the stark difference in substance-related ICU admissions rates for patients identified as “Other” races compared to White, Black and Asian patients demonstrates the disproportionate impact of substance use on critical illness within groups such as American Indian/Alaskan Natives and people reporting two or more races. Although these two racial groups report higher rates of substance use and substance use disorders compared to other races, the health effects of substance use and implications for ICU care for these populations are understudied [21–23]. Our analysis was constrained by the structure of the Premier Healthcare Database, which groups these individuals into a single “Other” race category, limiting our ability to analyze each distinct population. Further research is crucial to understand the risk factors and health impacts of substance use leading to an ICU admission for these specific racial groups.

Results from our GEE model on risk factors for in-hospital mortality after a substance-related ICU admission further illustrate important differences related to the impact of substance use. Although a substantial portion of drug-overdose deaths occur outside of the hospital and are therefore not included in our analysis, our findings that in-hospital mortality is associated with alcohol-related ICU admissions compared to opioids and stimulants reflects broader U.S. mortality data which shows higher alcohol-related deaths compared to other substances [3, 24, 25]. Despite this increased mortality, alcohol receives proportionately less media attention, policy proposals and government funding compared to other substances [26]. Interestingly, after controlling for substance type, Black and male patients had decreased odds of in-hospital mortality compared to White and female patients, respectively. This lower mortality contrasts with the increased rates of substance-related ICU admission for both demographic groups. Given that all substance-related deaths are inherently preventable, identifying specific risk factors for in-hospital mortality may offer opportunities for focused healthcare interventions.

Our study has several strengths and limitations. A key strength is the use of a large and nationally representative database over a multi-year period, with data from multiple hospital types and service areas, as well as pooled medical, trauma and surgical ICUs. We believe this enhances the generalizability of our findings to many ICUs across the country. Additionally, our study provides novel information regarding the impact of specific substance use on demographic groups within the ICU, which is critical for future surveillance and research. However, substance use was not necessarily the primary diagnosis or reason for the ICU admission. While a substance use disorder is often an upstream cause or exacerbating factor in critical illness, it can also be a secondary or concurrent condition which is not directly related to the acute cause of ICU admission. Another limitation is the use of a pre-defined race variable in the Premier Healthcare Database that aggregates American Indian/Alaska Native, Native Hawaiian/Pacific Islander, and multiracial individuals into a single “Other” race category. This restricts our ability to examine these groups individually, despite known disparities in substance use prevalence and outcomes among them. Our study is also limited by the reliance on ICD-10 coding within our dataset, which requires not only provider recognition of substance use during the ICU admission, but also inclusion of the appropriate diagnosis code. More specifically, we are unable to analyze the specific impact of increasing fentanyl use on ICU admissions due to the lack of ICD-10 coding for this substance distinct from other opioids prior to October 2020 [27]. Although universal screening and lab-based toxicology tests may reflect the true prevalence and demographic differences in substance-related ICU admissions and would thereby be preferred to coding which is susceptible to provider bias, these practices are not widely adopted throughout the United States. Another limitation of using ICD-10 codes to identify substance-related diagnoses in ICU admissions is the inconsistency in classification of diagnoses between alcohol and other substances. Alcohol-related ICD-10 codes include not only alcohol use (intoxication, withdrawal, dependence) but also health conditions directly attributable to alcohol (e.g., alcoholic cardiomyopathy, alcoholic hepatitis). In contrast, ICD-10 codes related to other substances are limited to use. Although this discrepancy may lead to differential capture of harm related to alcohol compared to other substances, our sensitivity analysis found the notable majority (>90%) of alcohol-related ICU admissions included ICD-10 codes related to alcohol use. Regardless, there is a need to expand ICD-10 codes to encompass the multitude of severe health consequences related to substances other than alcohol, including opioids and stimulants, in order to better understand the impact of substance use on human health.

The ICU admission may represent a “teachable moment” for patients with substance use disorders [28]. For ICU patients with alcohol misuse, an alcohol-related medical crisis has been identified as a facilitator for change in drinking behaviors, and a greater severity of acute illness further increases readiness to change [29, 30]. ICU-based interventions such as proactive management of withdrawal, involvement of addiction consult services, initiation and continuation of medications for addiction treatment, and effective linkage to long-term outpatient addiction care may help capitalize on this opportunity. Emerging literature suggests that addiction counseling reduces readmissions and 1-year mortality for ICU survivors with alcohol misuse yet remains underutilized [31, 32]. With a substantial burden of substance use disorders among ICU patients, and the potential to use substance-related ICU admissions as critical opportunities for change, our findings emphasize the growing need for effective substance use disorder screening, counseling and treatment resources among critically ill patients.

CONCLUSION

A high proportion of ICU admissions in the United States are related to substance use, with alcohol leading all other substances in both prevalence of ICU admission and in-hospital mortality. Furthermore, rates of ICU admission differed dramatically by age, sex, race, and ethnicity. To our knowledge, this is the first study to describe the national prevalence of and demographic differences within substance-related ICU admissions. These findings highlight the impact of substance use within our healthcare system and emphasize the urgent need for targeted interventions, support, and resources to address this devastating epidemic.

KEY POINTS.

Question:

What is the national prevalence of ICU admissions related to substance use? Do patient demographics differ among substance-related ICU admissions? What are the risk factors for in-hospital mortality associated with substance-related ICU admissions?

Findings:

Among 4,740,799 U.S. adult ICU admissions, 16.0% were substance-related, with alcohol most identified. ICU admission rates differed by age, sex, and race/ethnicity with varied patterns across substances. Alcohol use, female sex, and non-Hispanic ethnicity were independent predictors of in-hospital mortality.

Meaning:

Accurate national estimates of substance-related ICU admissions epidemiology are necessary for health care resource allocation and improvements in health care delivery.