Healthcare Utilization Due to Substance Abuse among Homeless and Non-Homeless Children and Young Adults in New York

Abstract

Objective

Substance abuse among youths is a significant public health challenge in the United States. Disproportionately higher substance use among homeless youths has been well documented, however, detailed data about healthcare utilization arising from this preventable health behavior remains sparse. The objective of this study was to assess healthcare utilization due to substance abuse among homeless youths compared to non-homeless youths.

Methods

Authors conducted a secondary data analysis using all homeless and non-homeless patients under 25 years of age with a primary diagnosis of substance abuse identified in the 2013 and 2014 New York Statewide Inpatient and Emergency Department (ED) Databases. Outcomes included ED visit rate, hospitalization rate, in-hospital mortality, cost, length of stay (LOS), intensive care unit (ICU) utilization, and revisit or readmission rate. Multivariable regression models with year fixed effects and facility random effects were used to evaluate the association between homelessness and each outcome.

Results

Authors identified 68,867 hospitalizations and ED visits related to substance abuse, yielding 9.38 and 4.96 ED visits and 10.53 and 1.01 hospitalizations per 1,000 homeless and non-homeless youths, respectively. Homeless patients had higher costs and LOS compared to non-homeless and were more likely to utilize the ICU and to revisit or be readmitted, (all ps < .01)

Conclusions

Homeless youths experience a two times higher ED visit rate and a 10 times higher hospitalization rate due to substance abuse than non-homeless youths. Detailed observation is needed to clarify whether homeless youths receive quality care for substance abuse when necessary.

INTRODUCTION

Substance abuse among youths is one of the most significant public health challenges in the United States. The 2017 National Survey on Drug Use and Health found that 21.9% of adolescents ages 12 to 17 reported past-year alcohol use and 16.3% reported past-year illicit drug use (1). Substance abuse during adolescence has been linked to negative outcomes, such as dropping out of school, suicide, risky sexual behavior, and underemployment (2–5).

Homeless youths are uniquely vulnerable to a variety of health problems, including those related to substance abuse (6). In 2018, more than 552,000 people in the United States experienced homelessness, including approximately 111,000 individuals under age 18 and 48,000 ages 18 to 24. Substantially higher substance abuse has been reported among homeless youths (7,8), compared with housed youths (9,10). Widespread substance abuse in this population is particularly concerning given that it is a major cause of death among homeless youths (11).

Although disproportionately higher substance abuse among homeless youths is well documented, detailed data about health care utilization resulting from this risky behavior among homeless youths remain sparse. Characteristics of health care utilization have been used as a proxy for overall health and well-being of patients with health conditions in general (12–14).

This study evaluated health care utilization related to substance abuse among homeless youths and assessed differences in basic characteristics of health care utilization related to substance abuse between homeless and nonhomeless youths. We hypothesized that health care utilization among homeless youths is higher than among nonhomeless youths.

Methods

Data Source

We used 2013–2014 discharge data from the New York State Inpatient Database (SID) and State Emergency Department Database (SEDD) from the Healthcare Cost and Utilization Project (HCUP), which is part of the Agency for Healthcare Research and Quality. The two data sets contain identical variables necessary for this study and were linked by using the patient’s ID number (VisitLink variable) and year of admission. At the time of the analyses, 2016 data were the most recent available, but because of the mandatory transition to ICD-10-CM on October 1, 2015, and a lack of validated diagnosis codes for substance abuse, we used patient records from 2013 to 2014. New York was selected because it represents almost 20% of the homeless youth population in the United States (15,16).

Identification of Patients

Emergency department (ED) visits and hospitalizations related to substance abuse for patients ages ≤24 were included because homeless population estimates within this age range were publicly availability (15,17) and there is a well-documented peak in substance abuse in the early 20s (18–20). ICD-9-CM codes were used to identify patients with a diagnosis of a substance use disorder on the basis of categorization in a previous publication that included alcohol- and drug- induced disorders, dependence, poisoning, abuse, and withdrawal (21). For patients with a primary diagnosis of these substance use disorders, we looked at secondary diagnosis codes to identify substance type, as previously done (21). Homelessness is a variable available in the database. We defined patients with a New York numeric zip code as nonhomeless. The institutional review board at the Lundquist Institute for Biomedical Innovation at Harbor-UCLA Medical Center approved this study.

Data Analysis

We calculated hospitalization and ED visit rates as the number of hospitalizations or ED visits per 1,000 population by using homeless population estimates from the Continuum of Care Homeless Populations and Subpopulations reports issued by the U.S. Department of Housing and Urban Development (HUD) (15,17). These reports provide demographic data and the number of homeless individuals in New York. (A detailed breakdown of the annual population of homeless youths is presented in a table as an online supplement to this article.) Corresponding population estimates for nonhomeless youths were obtained from the U.S. Census Bureau (22). ED visits and admissions analyses stratified by ages <18 and ≥18 and by opioid related or not were performed in response to the growing opioid epidemic in the United States (23–25).

The primary predictor of interest was homelessness. Outcomes of interest included in- hospital mortality, opioid-related admission, cost of ED visit or hospitalization, length of stay (LOS), intensive care unit (ICU) utilization, and ED revisit or hospital readmission. Encrypted personal identifiers provided by HCUP allowed us to track sequential visits for a patient across facilities, including inpatient and ED settings within the same calendar year, but not across calendar years. Therefore, we included only individuals discharged from an ED or hospital between January and June and evaluated their ED revisit or readmission within 6 months after discharge. Due to low death rates, ED mortality was not assessed.

To estimate health care utilization costs, we multiplied total hospital charges with HCUP hospital-level cost-to-charge ratios (26). To account for inflation, cost values were converted to 2015 U.S. dollars by using the medical care component of the Consumer Price Index (27).

Other variables of interest included age, race-ethnicity, insurance type, mental disorders, and year. Race-ethnicity was defined as non-Hispanic White, non-Hispanic Black, Hispanic, and other, and insurance type was categorized as public, private, self-pay, and other form of insurance. The category self-pay likely encompassed those with no insurance. Mental disorder diagnoses were identified by Clinical Classification Software (CCS) codes 650–659, 662, 663, or 670 (see table in online supplement for details). CCS code 663 includes substance abuse; therefore, CCS codes of 663 without a code for substance abuse were included in this category (28).

Descriptive statistics were tabulated and compared between homeless and nonhomeless youths. The type of substance used was examined across age groups and hospital settings (ED versus inpatient).

Logistic regression was used to assess in-hospital mortality, opioid-related admissions, ICU utilization, and ED revisits or readmissions. Linear regression was used to assess health care utilization costs after log-transformation of cost. Negative binomial regression was used to assess LOS. All models incorporated a year fixed effect and a hospital or ED random effect. Additionally, discharge against medical advice was included as a covariate for in-hospital mortality, ICU use, cost, and LOS models because it could affect each of these outcomes. All analyses used SAS, version 9.4. A two-sided p<0.05 was considered statistically significant.

Sensitivity Analyses

To check the robustness of our results, two sensitivity analyses were conducted. First, stratified analyses by opioid use for ICU utilization were performed because ICU utilization was possibly driven by use of opioids. Second, we repeated the main analyses with only patients under age 18 included.

Results

We identified a total of 68,867 cases in which individuals visited the ED or were hospitalized for substance abuse in New York between 2013 and 2014. Among those, 749 cases involved individuals who were homeless, and 68,118 involved individuals who were nonhomeless. Nearly 25% of these cases were opioid related, and more than 50% were alcohol related.

Compared with nonhomeless patients, homeless patients were less likely to be discharged from the ED (p<0.01) (Table 1). Health care utilization by homeless patients was more likely to be opioid related and less likely to be alcohol related, compared with health care utilization by nonhomeless patients (p<0.01). Compared with nonhomeless patients, homeless patients were significantly older; less likely to be White; more likely to be male; more likely to list public insurance; and less likely to list self-pay (all p<0.01).

TABLE 1.

Health care utilization related to substance abuse per among homeless and nonhomeless youths in New York, 2013–2014^a

	Total (N=68,867)		Nonhomeless (N=68,118)		Homeless (N=749)b
Variable	N	%	N	%	N	%	p
Emergency department (ED)	55,535	80.6	55,473	81.4	62	8.3	<.01
visit and discharged
Hospitalization	13,332	19.4	12,645	18.6	687	91.7	<.01
Cost ($)
Median	701.53		692.04		3,113.95		<.01
Interquartile range	1,085.72		1,058.46		2,676.90
Length of stay (days)							<.01
Median	0.00		0.00		3.00
Interquartile range	1.00		1.00		5.00
Intensive care unit	642	.9	618	.9	24	3.2	<.01
utilization
Discharged against	3,930	5.7	3,773	5.5	157	21.0	<.01
medical advice
ED was admission source	61,757	89.7	61,147	89.8	610	81.4	<.01
Died in hospital or ED	37	.1	>10	—	≤10	—	.34
Substance
Opioid related	16,900	24.5	16,552	24.3	348	46.5	<.01
Other substance
Alcohol	36,899	53.6	36,734	53.9	165	22.0	<.01
Amphetamine	279	.4	>10	—	≤10	—	.58
Cannabis (marijuana)	4,377	6.4	4,280	6.3	97	13.0	<.01
Cocaine	912	1.3	>10	—	>10	—	.10
Hallucinogen	1,005	1.5	971	1.4	34	4.5	<.01
Sedative	629	.9	601	.9	28	3.7	<.01
Other	6,41 6	9.3	6,37 6	9.4	40	5.3	<.01
Age
Mean±SD	20.55±2.93		20.54±2.93		21.37±2.59		<.01
Median	21.00		21.00		22.00		<.01
Interquartile range	4.0 0		4.0 0		3.0 0
Age group
0–9	344	.5	>10	—	≤10	—	.36
10–14	1,356	2.0	>10	—	≤10	—	.21
15–17	7,124	10.3	7,081	10.4	43	5.7	<.01
≥18	60,041	87.2	59,347	87.1	694	92.7	<.01
Race-ethnicity
Non-Hispanic white	38,539	56.0	38,237	56.1	302	40.3	<.01
Non-Hispanic Black	8,941	13.0	8,785	12.9	156	20.8	<.01
Hispanic	9,948	14.4	9,767	14.3	181	24.2	<.01
Other	11,435	16.6	11,325	16.6	110	14.7	<.01
Sex
Female	26,475	38.4	26,253	38.5	222	29.6	<.01
Male	42,387	61.5	41,860	61.5	527	70.4
Insurance
Public	21,934	31.8	21,244	31.2	690	92.1	<.01
Private	27,000	39.2	>10	—	>10	—	<.01
Self-pay	18,322	26.6	18,299	26.9	23	3.1	<.01
Other	1,611	2.3	>10	—	≤10	—	<.01
Mental disorder diagnosis	21,247	30.9	20,895	30.7	352	47.0	<.01
Year
2013	34,388	49.9	34,094	50.1	294	39.3	<.01
2014	34,479	50.1	34,024	50.0	455	60.8	<.01

^aGroups with 10 or fewer cases not reported in order to protect the identities of youths in these subgroups.

^bSome numbers are masked to prevent estimation of subgroups in neighboring cells with 10 or fewer cases

Homeless patients had substantially higher rates of ED visits and hospitalizations related to substance abuse (11.48/1,000), compared with nonhomeless youths (5.43/1,000) (Table 2). This difference was primarily driven by greater hospitalization rates among homeless youths ages 18 to 24 (64.48/1,000) versus nonhomeless youths (3.04/1,000). Stratified analyses for opioid- related health care utilization indicated that non-opioid-related hospitalization among patients ages 18 to 24 was more prominent with homeless youths (31.32/1,000) than nonhomeless youths (1.07/1,000).

TABLE 2.

Incidence rate of health care utilization related to substance abuse per 1,000 homeless and nonhomeless youths in New York, 2013–2014

				Age
	Total			<18			18–24
Variablea	Homeless	Nonhomeless	P	Homeless	Nonhomeless	P	Homeless	Nonhomeless	P
Total	11.48	5.43	<.01	.99	1.02	.81	70.58	14.92	<.01
ED visits	9.38	4.96	<.01	.74	1.00	.05	58.07	13.49	<.01
Hospitalizations	10. 53	1.01	<.01	.96	.06	<.01	64.48	3.04	<.01
Opioid related
Total	5.33	1.32	<.01	≤10b	.07	>.99	35.09	4.01	<.01
ED visits	4.08	1.01	<.01	≤10b	.06	>.99	26.75	3.06	<.01
Hospitalizations	5.04	.64	<.01	≤10b	.02	.11	33. 15	1.97	<.01
Other substance
Total	5.30	3.95	<.01	.69	.94	.051	31.32	10.91	<.01
ED visits	5.30	3.95	<.01	.69	.94	.051	31.32	10.43	<.01
Hospitalizations	5.49	.37	<.01	.90	.04	<.01	31.32	1.07	<.01

^aTotal includes emergency department (ED) visit and hospitalization rate. ED visits include patients who weredischarged directly home and patients who were hospitalized after an ED visit.

^bRates with 10 or fewer cases not reported to protect the identities of youths in these subgroups

Of all health care utilization related to substance abuse, opioid- and alcohol-related health care utilization were the most common (Table 3). Among patients ages 10 to 14 in both populations, alcohol was the primary substance involved in health care utilization. Opioid use increased with age among all youths, although alcohol remained the most common substance driving health care utilization among nonhomeless youths in the oldest age group. Stratified analyses by hospital setting showed that opioid use among homeless youths and alcohol use among nonhomeless youths were the most common substance-related drivers to ED visits, whereas opioid use drove the most substance-related hospitalizations for both homeless and nonhomeless youths.

TABLE 3.

Substances ranked by association with substance-related emergency department (ED) visits and hospitalizations among homeless and nonhomeless youths in New York, by age group, 2013–2014^a

	Most common substance				Second most common substance				Third most common substance
Group	Substance	N	%	P	Substance	N	%	p	Substance	N	%	p
Total
Overall
Homeless	Opioids	348	46.5	<.01	Alcohol	165	22.0	<.01	Marijuana	97	13.0	<.01
Nonhomeless	Alcohol	36,734	53.9	<.01	Opioids	16,552 24.3		<.01	Other	6,376	9.4	<.01
Age 0–9
Homeless	Cocaineb	≤10	—	.04 Hallucinogensb		≤10	—	<.01	NA	—	—	—
Nonhomeless	Opioids	147	42.9	<.01	Alcohol	97	28.3	<.01	Unidentified	33	9.6	—
Age 10–14
Homeless	Alcohol	≤10	—	≤.01	Opioidsb	≤10	—	<.01	Hallucinogensb	≤10	—	<.01
Nonhomeless	Alcohol	710	52.7	<.01	Marijuana	340	25.2	<.01	Other	135	10.0	<.01
Age 15–17
Homeless	Marijuana	15	34.9	<.01	Alcohol	>10	—	<.01	Hallucinogens	≤10	—	<.01
Nonhomeless	Alcohol	4,395	62.0	<.01	Marijuana	1,083	15.3	<.01	Other	737	10.4	<.01
Age 18–24
Homeless	Opioids	345	49.7	<.01	Alcohol	150	21.6	<.01	Marijuana	81	11.7	<.01
Nonhomeless	Alcohol	31,532	53.1	<.01	Opioids	15,962	26.9	<.01	Other	5,492	9.3	<.01
ED visitsc
Overall
Homeless	Opioids	266	43.5	.24	Alcohol	138	22.6	.12	Marijuana	84	13.7	.97
Nonhomeless	Alcohol	35,822	57.5	.12	Opioids	12,693	20.4	.24	Other	6,035	9.7	.94
Age 0–9
Homeless	Hallucinogens	≤10	—	.08	NA	—	—	—	NA	—	—	—
Nonhomeless	Opioids	133	42.0	.24	Alcohol	92	29.0	.12	Unidentified	31	9.8	—
Age 10–14
Homeless	Alcohol	≤10	—	.12	Opioidsb	≤10	—	.24	Hallucinogensb	≤10	—	.08
Nonhomeless	Alcohol	708	53.3	.12	Marijuana	340	25.6	.97	Other	135	10.2	.94
Age 15–17
Homeless	Alcoholb	≤10	—	.12	Marijuanab	≤10	—	.97	Hallucinogens	≤10	—	.08
Nonhomeless	Alcohol	4,369	62.7	.12	Marijuana	1,066	15.3	.97	Other	724	10.4	.94
Age 18–24
Homeless	Opioids	263	46.1	.24	Alcohol	125	21.9	.12	Marijuana	75	13.1	.97
Nonhomeless	Alcohol	30,653	57.1	.12	Opioids	12,165	22.6	.24	Other	5,165	9.6	.94
Hospitalizations
Overall
Homeless	Opioids	329	47.9	<.01	Alcohol	137	19.9	.06	Marijuana	93	13.5	.88
Nonhomeless	Opioids	8,018	63.4	<.01	Alcohol	2,304	18.2	.06	Other	646	6.0	.42
Age 0–9
Homeless	Cocaineb	≤10	—	.53	Hallucinogensb	≤10	—	<.01	NA	—	—	—
Nonhomeless	Opioids	41	59.4	<.01	Unidentified	12	17.4	—	Alcohol	≤10	—	.06
Age 10–14
Homeless	Alcohol	≤10	—	.06	Opioids	≤10	—	<.01	Hallucinogensb	<10	—	<.01
Nonhomeless	Alcohol	25	30.9	.06	Opioids	20	24.7	<.01	Marijuana	15	18.5	.88
Age 15–17
Homeless	Marijuana	>10	—	.88	Alcohol	≤10	—	.06	Hallucinogens	≤10	—	<.01
Nonhomeless	Alcohol	122	31.7	.06	Opioids	111	28.8	<.01	Marijuana	59	15.3	.88
Age 18–24
Homeless	Opioids	326	51.4	<.01	Alcohol	124	19.6	.06	Marijuana	77	12.2	.88
Nonhomeless	Opioids	7,845	64.8	<.01	Alcohol	2,150	17.8	.06	Marijuana	689	5.7	.88

^aGroups with 10 or fewer cases not reported to protect the identities of youths in these subgroups. Some numbers are masked to prevent estimation of subgroups in neighboring cells with 10 or fewer cases. NA, not applicable because zero cases were identified.

^bIndicates substances were equally ranked (i.e., the same number of cases were identified for each substance).

^cED visits include patients who were discharged directly home and patients who were hospitalized after an ED visit.

Adjusting for age, race-ethnicity, insurance type, and discharge against medical advice with year fixed effect and a hospital or ED random effect, regression results showed that, compared with nonhomeless youths, homeless youths had significantly higher costs for ED visits or hospitalizations (cost estimate=3.47) and longer LOS (incidence rate ratio=4.02) (Table 4). Homeless patients were significantly more likely than nonhomeless patients to be admitted to the ICU during hospitalization (odds ratio [OR]=4.75) and more likely to be readmitted to the hospital or to revisit the ED (OR=1.41).

TABLE 4.

Results of regression models to evaluate the association between homelessness and study outcomes among homeless and nonhomeless youths with substance-related health care utilization^a

Variable	Estimate	95% CI	P
Homeless in-hospital	3.13	.36–26.99	.30
mortalityb,c
Homeless opioid-related	.95	.78–1.17	.66
utilizationb
Homeless total costc,d	3.47	3.23–3.71	<.01
Homeless length of stayc,e	4.02	3.29–4.94	<.01
Homeless intensive care	4.75	2.84–7.93	<.01
unit utilizationb,c
Homeless readmission or	1.41	1.04–1.91	.03
ED revisitb,c,f

^aReference group for all comparisons: nonhomeless. Models were adjusted by age, race-ethnicity, insurance type, and mental disorder diagnosis, with year fixed effect and a hospital or emergency department (ED) random effect.

^bLogistic regression models were used.

^cDischarged against medical advice was added to the model as a covariate

^dLinear regression was used after log-transformation of cost.

^eNegative binomial regression was used.

^fOnly patients who had personal identifiers were included in the model.

Because the sample size was reduced in the stratified sensitivity analyses, one model that included only patients with opioid-related health care utilization did not detect significant differences in ICU utilization between homeless and nonhomeless patients. However, the direction of the point estimate did not change. This indicated a higher likelihood of ICU utilization among homeless patients compared with nonhomeless patients, regardless of opioid use (see table in online supplement). Results of analyses including only patients ages <18 mirrored the original analyses (see table in online supplement).

Discussion

Using the SID and SEDD from New York, this study demonstrated different characteristics of health care utilization, represented by hospitalizations and ED visits related to substance abuse among homeless and nonhomeless youths. We found that the rate of ED visits related to substance abuse was almost two times higher among homeless youths, compared with nonhomeless youths, and the hospitalization rate related to substance abuse was more than 10 times higher. This difference was mainly driven by a greater rate of non–opioid-related hospitalizations among homeless youths, especially among those ages 18 to 24.

Health care utilization could be used as a proxy for population well-being. For example, it could represent heavy substance abuse, because heavy use is more likely than casual use to result in health care utilization. However, differences in heavy use between homeless and nonhomeless youths could be greater in magnitude than the differences observed in this study because of disparities in access to care. Homeless youths face obstacles to receiving medical care, including a shortage of medical services in poor communities (29), lack of transportation, and social isolation. Our study showed that only 8.3% of homeless patients were directly discharged from the ED, whereas 81.4% of nonhomeless patients were directly discharged from the ED. Additionally, compared with nonhomeless youths, homeless youths had significantly higher ICU utilization, indicating that they did not utilize health care until their condition became severe.

Our models detected higher health care utilization cost and longer LOS among homeless youths, compared with nonhomeless youths. Health care cost and LOS are typically used as a proxy for severity. Specifically, if the condition is severe, the patient will require additional treatment, which will result in high costs and a long LOS. However, studies of patients experiencing homelessness have reported a pattern of a longer average LOS possibly not related to the severity of their condition, compared with their housed counterparts (30,31). Some doctors may refrain from discharging a patient until a shelter or other more stable form of housing is arranged (30,31). Therefore, our finding that homelessness was associated with higher costs and longer LOS does not necessarily mean that homeless youths experienced more severe medical conditions than nonhomeless patients, but their greater utilization of the ICU suggests that homeless patients did not utilize health care until their condition became severe (32). Detailed observations are needed to clarify whether homeless youths receive medical care for substance abuse when needed.

We demonstrated that, compared with nonhomeless patients, homeless patients were more likely to revisit the ED or be readmitted to the hospital. Several studies have cited high readmission rates and longer LOS among homeless individuals, compared with nonhomeless individuals (30,33). Both factors are largely affected by care transitions. If the individual does not have a suitable place to recover after discharge, the physician may be incentivized to keep the individual in the hospital until coordination with a shelter or other care has been established (30). If the recovering patient is released to the streets or unstable housing, the patient’s condition will likely worsen, and he or she will likely return to the ED or hospital within 30 days of discharge (30,33). Previous studies have reported high substance abuse rates among homeless youths, compared with nonhomeless youths (7–10), and developing programs to reduce substance abuse among homeless youths is particularly important. ED visits and admissions could be a good opportunity to provide prevention and education programs to these young homeless patients who would otherwise be difficult to identify and contact. Ensuring that these programs are available during their first visit to the hospital for substance abuse may help reduce the revisit and readmission rate as well as the mortality rate.

The opioid-related health care utilization rate among homeless patients ages 18 to 24 was much higher than the rate among their nonhomeless counterparts. Specifically, the opioid-related hospitalization rate among homeless patients was almost 17 times higher than the rate among nonhomeless patients in this age group. This result is consistent with previous studies for adults ages 18 and older (34–36). The magnitude of difference in the hospitalization rate was even greater for non-opioid-related hospitalizations. Alcohol accounted for the largest percentage of non-opioid-related ED visits and hospitalizations related to substance abuse, necessitating immediate prevention measures and education related to alcohol abuse among homeless youths.

Several limitations should be acknowledged. First, although 20% of homeless youths in the United States reside in New York, the results may not be generalizable to other parts of the country. New York provides various means of support to its homeless citizens, with 96.8% of those in our study utilizing public or private insurance—a likely result of the transition, beginning in 2012, into Medicaid managed care (37) by the state’s Office of Insurance Programs. Because New York law requires shelter for all who request it, 99% of youths ages 0–24 in New York were sheltered in 2013, contrasted with only 79% nationally (17,38). Although we included only a single state in our analyses, our findings are novel because health care utilization due to substance abuse among homeless youths has not previously been reliably assessed.

Second, this study is a serial cross-sectional analysis of an administrative database, and causality between health care utilization related to substance abuse and housing instability is difficult to establish. Additionally, the database lacks clinical detail beyond ICD-9-CM codes, and only patients who visited the ED or were hospitalized were included in the study. The 2013 National Survey on Drug Use and Health reported that only 10.9% of individuals ages 12 and older, including adults, who needed treatment for illicit drug or alcohol abuse received treatment (39). As a result, rates of ED visits and hospitalizations from this study could, therefore, be underestimated when used as a proxy for needed substance abuse care. Rates of ED visits and hospitalizations related to substance abuse from this study may also be biased because of the quality of coding. However, a previous study found that ICD-9-CM coding accuracy for opioid overdose was high (40). We used the primary diagnosis code to identify the primary reason for health service utilization, as previously practiced. Song (41) argued that the primary diagnosis code reflects the physician’s professional opinion as to the primary cause of admission. We believe our method identified health service utilization related to substance abuse as the primary condition; however, it is still possible that substance abuse was secondary to another condition.

Third, the quality of coding for homelessness is unknown. For example, some patients or medical staff may report a shelter zip code as the zip code of residence, some patients may choose not to disclose their living status for fear of being stigmatized, and some health care facilities do not code homeless status (42). Health care staff familiar with patients who frequently utilize services, especially those with behavioral conditions such as substance abuse, may be more likely to accurately list housing insecurity as a result of recurrent interactions with such patients. Therefore, our results indicating that those experiencing homelessness were more likely to revisit health care facilities could be a result of reverse causation (i.e., individuals who were more likely to revisit health care facilities were more likely to be coded as homeless).

Fourth, estimates of the homeless population based on point-in-time counts of sheltered and unsheltered persons are likely to be conservative because of the transient nature of homelessness. Nonsheltered youths may be a particularly undercounted population, because they may be out of sight during the counts (43). In 2014, children and youths represented less than 1% of New York’s total counted unsheltered homeless population (15). Given that most homeless children and youths in New York are sheltered, they are less likely than other homeless groups to be undercounted. Despite the biases discussed, these counts are the best estimates available. Each city’s homeless coalition provides HUD with local counts to ensure properly allocated federal funding for programs and services aimed at assisting those experiencing homelessness (44).

Fifth, all primary analyses were conducted at the level of the ED visit or the hospital admission, with each visit or admission included as a unique observation. The personal identifiers provided in the HCUP databases allow only for tracking of patient revisits and readmissions within a given data year. Because of this limitation, we were unable to conduct analyses at the patient level, and patients who were discharged and subsequently revisited or were readmitted may have been double-counted.

Sixth, ED and hospitalization rates for those ages 0–9 most likely are not representative of substance abuse by infants or children. It is more likely that substances were passed through breast milk or accidentally ingested (45,46).

Last, number of deaths due to opioid overdose increased steadily in 2013 (47). Our study period reflected only 2 years after the onset of this particular wave of the opioid epidemic. Further evaluation with additional years of data is needed.

Conclusions

This study demonstrated that, compared with nonhomeless youths, homeless youths experienced much higher rates of health care utilization related to substance abuse. Higher ICU utilization among homeless youths versus nonhomeless youths indicated that homeless youths did not utilize services until their condition had become severe. Detailed observation is needed to clarify whether homeless youths receive high-quality care for substance abuse when necessary.

Highlights:

• Nearly 18% of healthcare utilization cases (hospitalizations or ED visits) were opioid- related and more than 50% were alcohol-related.

• Alcohol was the most common substance among patients 10- to 14-years-old, in both populations.

• Opioid use increased with age among both homeless and non-homeless.

• Healthcare utilization by homeless patients was more likely to be opioid-related and less likely to be alcohol-related than healthcare utilization by non-homeless patients.”