Regular Outpatient Medical and Drug Abuse Care and Subsequent Hospitalization of Persons Who Use Illicit Drugs FREE

Users of illicit drugs have complex health care needs.^1- 5 Yet, many drug users receive medical care only when crises arise that require emergency department care or hospitalization.^1,6- 9 Drug users are twice as likely to visit an emergency department and nearly 7 times more likely to be hospitalized than comparably aged persons who do not use illicit drugs.¹⁰ Even in Canada, where citizens have access to universal health insurance, substance abuse accounts for approximately 8% of hospitalizations.¹¹ In the United States, public payers are responsible for much of the hospitalization costs for persons with substance abuse.^9,12 Once hospitalized, drug users also appear to have longer lengths of stay. In a cohort of human immunodeficiency virus (HIV)–infected Medicaid enrollees, we reported that drug users averaged 2 weeks longer in the hospital than nonusers following initial acquired immunodeficiency syndrome (AIDS) diagnosis.¹³ Patients and the public stand to benefit from the identification of factors that prevent drug users' heavy reliance on inpatient care.

Medical care and drug abuse treatment in ambulatory settings may reduce use of inpatient services by providing timely management of medical conditions and by preventing drug abuse–related complications. Having a regular source of medical care has been associated with improved delivery of preventive care and reduced substance abuse in a poor population.¹⁴ Having a regular source of care has also been related to lower use of hospital services by HIV-positive persons.¹⁵ Outpatient drug abuse treatment frequently offers an array of services that address not only drug abuse but also the social, psychological, and legal needs of this population.¹⁶ Regular ambulatory substance abuse treatment has been found to improve survival of drug users¹⁷ and to reduce HIV risk behaviors.¹⁸

We hypothesized that a combination of regular medical care with regular drug abuse treatment would show a protective association with hospital use. To explore this hypothesis, we examined the association of medical and drug abuse care with hospitalization in a cohort of HIV-positive and HIV-negative drug users enrolled in the New York State (NYS) Medicaid program.

We conducted a retrospective cohort study of drug users enrolled in the NYS Medicaid program from federal fiscal years 1996 through 1997. Data were from longitudinally linked claims for all ambulatory services from physicians and clinics and drug abuse services covered by the Medicaid program. These files provide information on various covered services: inpatient, pharmacy, home health care, selected case management, and laboratory. The institutional review boards of Thomas Jefferson University, the New York State Department of Health, and Montefiore Medical Center approved the study.

We identified Medicaid claims associated with illicit drug use, HIV infection, and AIDS using previously tested algorithms developed by our group.^19- 21 The algorithm to screen claims for illicit drug use searches for International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes for heroin, cocaine, amphetamine, or unspecified illicit drug dependence or abuse; for Medicaid services (or rate codes) for methadone treatment or medically supervised drug abuse treatment; and for New York State diagnosis related groups (DRGs) for inpatient drug detoxification. Persons infected with HIV were identified by another case-finding method that searches for at least 1 of the following: (1) 1 or more pharmacy claims for an antiretroviral drug, (2) 1 or more inpatient stays with a diagnosis of an AIDS-defining condition or HIV-related DRG, or (3) at least 2 occurrences separated by more than 30 days of: an outpatient physician or clinic visit with an HIV-related or AIDS-defining diagnosis, an HIV-specific outpatient care rate code, or inpatient stay for an HIV-related condition such as pneumonia. Our initial longitudinal Medicaid analysis file included 78 943 persons who met the illicit drug use algorithm, were 13 through 59 years of age in 1996, and were enrolled continuously in Medicaid for 10 months or more in 1996. Of this group, 77 618 had any outpatient visit(s) in 1996 to permit analysis of patterns of care. Of these, 59 104 were enrolled in Medicaid for 10 months or more in 1997 and 59 092 had demographic information. After excluding women with evidence of pregnancy or delivery in 1997, the final study population totaled 58 243 persons.

To gauge the operating characteristics of these Medicaid claims case-finding algorithms, we conducted a validation study using a sample of patients treated at Montefiore Medical Center in the Bronx, NY. From outpatient chart review supplemented by self-report data from research questionnaires for some drug users, 207 Medicaid enrollees in 1996 were classified into 4 groups: using illicit drugs, never using illicit drugs, HIV-positive, and HIV-negative. According to these classifications, the NYS Medicaid claims algorithm had a sensitivity of 0.86 and a specificity of 0.87 for identifying drug abuse, and these characteristics were 0.96 and 1.0, respectively, for detecting HIV infection. We conducted a detailed examination of Medicaid claims for the 14 individuals who were classified as drug users according to Medicaid data but classified as nonusers according to Montefiore data sources. This review showed that 11 had numerous (ie, >10) outpatient visits for drug abuse treatment and/or at least 1 inpatient stay with the diagnosis of illicit drug dependence, all from providers outside of the Montefiore system. Thus, they were most likely misclassified by the Montefiore-only data sources.

We categorized each drug user's pattern of outpatient care based on Medicaid claims in 1996 as regular drug abuse care only, regular medical care only, both, or neither. Regular drug abuse care was defined as care from a single methadone or drug-free treatment program for at least 6 contiguous calendar months in 1996. We applied a 6-month criterion because a minimum length of stay for effective methadone treatment according to Ball and Ross²² is 6 months and studies of treatment for cocaine abuse similarly define treatment periods as 6 to 12 months.²³ At least 3 weekly claims for drug abuse treatment per month were required to fulfill criteria for regular drug abuse care; this allows for at most a 1-week lapse in treatment. During hospital stays, patients were classified as having continued their drug treatment if they had already started drug treatment before the hospitalization. In the rare instances when outpatient drug abuse treatment claims are submitted daily, we considered 15 or more such claims from the same provider in each 30-day interval of nonhospitalized days within a 6-month calendar period as indicating regular drug abuse care.

We studied 2 types of ambulatory drug abuse treatment: methadone maintenance treatment clinics and medically supervised (Title 1035 or "medically supervised drug-free") clinics. These clinics can be part of large programs with multiple sites but we considered care within the same program as continuous, in part because Medicaid claims often do not distinguish separate providers within multisite practices. We did not consider clinics that provided only alcohol abuse treatment. We also did not consider detoxification, residential, and non–medically supervised ambulatory programs in this analysis because we were interested in longitudinal ambulatory services supervised by a professional. Furthermore, detoxification treatment for opiate-addicted persons has been reported to be less successful in reducing drug-related behaviors than standard methadone treatment.²⁴ We also did not consider physicians who provide drug abuse treatment in private offices because they delivered less than 10% of all drug abuse care reimbursed by NYS Medicaid in 1998 (Peter Gallagher, NYS Department of Health, oral communication, January 2001).

A regular source of medical care was defined as a clinic or physician visited by a study subject at least twice as an outpatient during 1996 and delivering more than 35% of all outpatient encounters from primary care physicians, obstetrician-gynecologists, or medical subspecialists in that year.¹⁵ These providers could be clinics, group practices, or individual physicians. When a patient saw 2 providers for the same proportion of visits, we used a previously developed hierarchy of specialties to select the regular medical provider.¹⁵ In a previous study of HIV-positive Medicaid enrollees, we found no significant differences in outcomes from 2 cutpoints (>35% and >50%) for defining regular medical care.¹⁵ Using the "greater than 35%" cutpoint avoids misclassification of patients with more than one third of their care from a given provider as having no regular medical care.

The study-dependent variable was 1 or more hospitalization during federal fiscal year 1997. Hospitalizations were determined from Medicaid claims for inpatient stays of at least 1 day. Using ICD-9-CM diagnosis codes (up to 5 per hospital claim), we also classified hospitalizations into 4 groups: drug abuse related (eg, illicit drug poisoning or medical complications of drug abuse such as endocarditis), alcohol related (eg, pancreatitis, cirrhosis, Mallory-Weiss tear), mental health–related (eg, depression, schizophrenia), and medical (eg, HIV-related and general medical conditions such as pneumonia, sickle cell crisis, diabetes, congestive heart failure). A hospital stay categorized into more than 1 of these clinical types was considered in the analysis of each relevant hospitalization type. Unclassified types of hospitalizations such as injuries were placed in an "other" category. A complete list is available from the authors.

Patient demographic characteristics were obtained from Medicaid claims and eligibility files including those for age, sex, and NYS region of residence. Residence was classified from ZIP codes as rural, small city, upstate urban, New York City suburb, or New York City. New York State Medicaid files do not contain reliable data on ethnicity. Data on comorbid conditions were also obtained from ICD-9-CM codes on inpatient and outpatient claims files in 1996. We used only 1996 data to measure comorbid conditions to avoid analytic difficulties associated with concurrent measurement of comorbidity and hospitalization outcomes. Using diagnoses on Medicaid claims of AIDS-defining conditions (eg, Pneumocystis carinii pneumonia or tuberculosis) or HIV-related complications such as an episode of pneumonia, we determined the HIV clinical stage of each patient. We also identified specific types of drug abuse, psychiatric disease (eg, depression, non–drug-related psychoses, anxiety disorders), and chronic diseases other than HIV (eg, diabetes, hypertension, asthma, sickle cell anemia) (Table 1). Lastly, we included a variable for the total number of hospital days in 1996 in our models as a proxy for unmeasured health status.

All analyses were conducted separately for HIV-positive and HIV-negative patients. After examining bivariate associations between covariates and hospitalization end points, we estimated multivariable regression models with any hospitalization (logistic regression) in 1997 as the dependent variable. We also estimated separate logistic regression models for each clinical type of hospitalization.

To assess potential selection effects for the models with any hospitalization as the dependent variable, we also conducted a propensity score analysis that considered each patient's propensity of being in a particular outpatient care group.²⁵ To determine the propensity scores, we conducted a polychotomous logistic regression analysis with a variable indicating pattern-of-care group membership as the dependent variable and the variables listed in Table 1 as the independent variables. Since there are 4 pattern-of-care groups, we derived 3 propensity scores. We then repeated the multivariable analyses using any hospitalization as the outcome with the 3 propensity scores, including indicators for quintiles of each propensity score and the pattern-of-care grouping variable. We compared these results to those of the logistic regression model including separate patient characteristics. Analyses were performed using SAS version 8.0 (SAS, Cary, NC).

Table 1 shows the baseline characteristics of the 11 556 HIV-positive and 46 687 HIV-negative drug users in the analysis. Approximately 30% of the study population had regular drug abuse care alone or with regular medical care in 1996 but a higher proportion of HIV-positive drug users had regular care of both types. More than half of the study population had regular medical care.

In 1997, more than half of the HIV-positive and one third of the HIV-negative group had at least 1 hospitalization (Table 2). Those who were hospitalized spent nearly 1 month as an inpatient over the course of 1997. In both groups, those with regular drug abuse care only or with regular medical care had the lowest proportions of any hospitalization in 1997. Similar relationships of the patterns of ambulatory care with average number of inpatient days in 1997 were observed in both groups.

The proportions of persons in both groups who were at least 50 years old or who had another chronic disease such as diabetes were higher among those with regular drug abuse care and regular medical care than for those in the other patterns of care (data not shown). Among HIV-negative persons, only medical care was more likely for those with heroin or cocaine abuse or dependence, drug dependence of unspecified type, acute drug-related complication (eg, cellulitis, drug withdrawal, phlebitis), alcohol abuse, and alcohol-related complication (eg, pancreatitis, acute alcoholic hepatitis). The distribution of these characteristics by pattern of care for HIV-positive drug users was less skewed.

Among HIV-positive drug users, regular drug abuse care with regular medical care was associated with nearly a 25% reduction in the adjusted odds of hospitalization vs neither type of care, while the associations were somewhat weaker for either type of care alone (Table 3). Among HIV-negative drug users, regular drug abuse treatment alone or with regular medical care was associated with a more than 25% reduction in the adjusted odds of hospitalization (0.71 and 0.73, respectively). However, we observed a greater benefit associated with regular medical care alone for the HIV-positive drug users (ie, adjusted odds of hospitalization reduced by 18%, vs 9% for HIV-negative).

In both groups, the adjusted odds of hospitalization were greater for persons with drug abuse or medical complications during 1996, the baseline year. Of these factors, acute alcohol-related complications were associated with the greatest increase in the likelihood of any hospitalization. As expected, hospitalization in 1996 was strongly related to the risk of hospitalization in 1997.

For both HIV-positive and HIV-negative groups, we estimated models predicting the probability of an individual receiving each of the patterns of care and included the propensity scores grouped in quintiles from these models as independent variables in models predicting any hospitalization (data not shown). Among HIV-positive persons, the adjusted odds ratio (AOR) of any hospitalization from the propensity score analysis was also lowest for persons with both types of care (AOR, 0.76; 95% confidence interval [CI], 0.68-0.86), followed by only regular medical care (AOR, 0.82; 95% CI, 0.74-0.91) and only drug abuse care (AOR, 0.84; 95% CI, 0.75-0.95), compared with neither type of care. In the model including the propensity score variables for the HIV-negative group, the adjusted odds were similar for persons with both regular drug abuse care and regular medical care (AOR, 0.67; 95% CI, 0.63-0.71) and those with only drug abuse treatment (AOR, 0.66; 95% CI, 0.61-0.71) vs persons with neither form of care. Only regular medical care was less protective (AOR, 0.90; 95% CI, 0.86-0.94). These results closely resemble those for the model that did not include the propensity scores (Table 3).

Hospitalizations for the study population in 1997 were categorized by discharge diagnoses on claims into 4 non–mutually exclusive clinical categories (Table 4). All but 2843 (8.5%) hospitalizations were in 1 or more of these categories. Medical hospitalizations were the most frequent type. Regardless of HIV status, regular drug abuse care with or without regular medical care was associated with a reduction in the adjusted odds of alcohol-related and mental health–related hospitalizations. Regular drug abuse care with or without regular medical care was associated with an approximately 15% reduction in medical hospitalization in both groups, while in the HIV-positive group only medical care was similarly protective.

The only significant associations with drug abuse–related hospitalizations were found for drug abuse care alone, which was associated with higher adjusted odds for this type of hospitalization in HIV-negative persons. Because a drug abuse–related hospitalization could be for detoxification (a potentially desirable reason for hospitalization) we repeated the analysis excluding hospitalizations for which detoxification was indicated as a procedure during the hospitalization (n = 456). In both groups, this model showed more protective associations of regular outpatient care with drug abuse–related hospitalization. In the HIV-positive group, the patterns of care were associated with the following AORs for this outcome: regular drug abuse care alone (AOR, 0.93; 95% CI, 0.78-1.11), regular medical care alone (AOR, 0.87; 95% CI, 0.77-1.00), and both patterns of care together (AOR, 0.76; 95% CI, 0.63-0.91). In the HIV-negative group, the patterns of care were associated with the following results: regular drug abuse care alone (AOR, 0.89; 95% CI, 0.79-1.01), regular medical care alone (AOR, 0.96; 95% CI, 0.90-1.03), and both patterns of care together (AOR, 0.89; 95% CI, 0.79-0.91).

Our study affirms the enormous demand by users of illicit drugs for hospital care but also sheds light on possible solutions to this problem. In our population-based cohort of nearly 60 000 drug users, more than half of the HIV-positive group and one third of the HIV-negative group were hospitalized in 1997 and those hospitalized spent nearly 1 month as an inpatient over the course of the year. This high use of hospital care not only reflects substantial morbidity in this Medicaid-enrolled population but is also extraordinarily costly.

French et al²⁶ reported that out-of-treatment chronic drug users generated about $1000 annually in excess service use per individual compared with nonusers, with most of this cost due to greater inpatient care and emergency department use. That study also found that drug users used fewer ambulatory services than nonusers. Nearly 50% of our cohort did not have a regular source of medical care. In addition, only 30% of our study population received regular drug abuse treatment. Engaging illicit drug abusers in treatment for at least 6 months has been associated with significant improvement in both health and social indicators such as keeping outpatient appointments and improving housing situations.²⁷ A reduction in hospitalization of HIV-positive and HIV-negative patients has been associated with participation in methadone treatment at 1 institution.²⁸

Regardless of HIV status, we observed at least a one-quarter reduction in the adjusted odds of any hospitalization in 1997 associated with receipt of both regular medical care and regular drug abuse care in 1996. While the strongest protective association in the HIV-positive group was observed for persons with both forms of care, HIV-negative drug users with drug abuse treatment alone showed an equally favorable association with lower hospitalization rates.

These data support the benefits of linking persons who abuse illicit drugs to these 2 complementary types of ambulatory care to reduce subsequent use of inpatient care. However, the observational nature of this study prohibits conclusions about causal relationships. Although our models adjusted for several demographic characteristics associated with entry and retention in substance abuse treatment,^29- 30 other factors such as history of a criminal record and drug use during treatment²⁹ were unavailable for this claims-based analysis. Following the recommendations of D'Agostino,²⁵ we used propensity score methods to reduce potential selection bias by categorizing patients into quintiles of similar risk of having specific patterns of care. Models including propensity scores showed little change in the adjusted odds of hospitalization for persons with both regular medical and regular drug abuse care. Furthermore, our adjusted analyses included as a potential confounder an indicator for each patient's total hospital days during the baseline year. Among HIV-negative persons, Neimcryk and colleagues³¹ demonstrated that previous hospitalization predicts future hospitalization. In that study, 58% of persons hospitalized in the baseline time period had another hospitalization in a subsequent time period. By including this variable as a covariate, we biased our analysis against finding an effect related to care type. We believe that these results are sufficiently robust to support approaches to promote continuity of medical and drug abuse treatment in drug users.

Regular drug abuse care alone or with regular medical care was most strongly associated with lower likelihood of hospitalization for alcohol- or mental health–related complications. As expected, hospitalization for medical complications such as pneumonia was less likely for persons with medical care alone whether or not they also had regular drug abuse care. However, receipt of regular medical care and/or regular drug abuse care showed a less favorable effect on the probability of subsequent hospitalization for drug abuse–related diagnoses. Stein¹ reported that most hospitalizations for a sample of HIV-positive persons without AIDS were for users of injected drugs with drug abuse–related complications. Since drug users in treatment often persist in using illicit drugs, albeit often at reduced levels, it is possible that drug abuse–related conditions that warrant hospitalization may be more likely to be detected while in drug treatment, leading to greater inpatient care. When we excluded hospitalizations for detoxification, regular medical and regular drug abuse care were associated with significantly lower adjusted odds of drug abuse–related hospitalization. The role of inpatient detoxification is controversial and considered by some as not warranting inpatient care.³² Nonetheless, these additional analyses suggest that a substantial proportion of the drug use–related hospitalizations might have been for detoxification.

Our claims-based analysis also did not distinguish specific services delivered at the drug abuse treatment sites or whether medical care services were available on site. Samet and colleagues³³ postulated that important benefits would accrue from linking drug abuse care to medical care in the same setting. New York State has fostered co-located medical care for HIV-positive drug users in drug abuse treatment, which may explain the higher proportion of these patients with both forms of care than among HIV-negative drug users. Others have found that on-site medical services have several effects for HIV-positive patients.^34- 35 For other diseases such as tuberculosis, medical care linked to drug abuse treatment appears to be cost effective.³⁶ To improve the likelihood that drug users receive both forms of care, the linked-care model should be explored for both HIV-positive and HIV-negative persons on a broader scale.

Specific characteristics of medical care settings may also influence the probability of hospitalization. For HIV-positive drug users with a clinic as their usual source of medical care, we have shown previously that the adjusted odds of hospitalization were at least 50% lower for patients in clinics with case managers or with a high degree of coordination of care according to the clinic director.³⁷ To reduce hospitalization of drug users, additional studies should assess alternative drug abuse treatment models ranging from the most basic (eg, striving to retain drug users in drug treatment and in medical care) to more specific efforts (eg, offering targeted services in these settings).

Our study years witnessed the introduction of highly active antiretroviral therapy (HAART) that can reduce the need for inpatient care of HIV-positive persons.³⁸ We did not consider the use of HAART in this analysis of both HIV-positive and HIV-negative drug users but expect that improved access to these medications may contribute to the benefit associated with regular medical care for HIV-positive persons. In a sample of nearly 300 HIV-positive women in their first year postpartum (of whom 29% had evidence of current drug abuse), we found that visits to NYS medical providers offering an array of HIV-specific services were associated with increased adjusted odds of antiretroviral therapy treatment and adherence.³⁹

Several other limitations of our study deserve mention. We used claims data to identify our study population. Yet, our validation study suggests that our algorithms performed well in identifying drug users and persons with HIV infection. We did not specifically investigate the impact of mental health and alcohol treatment, but focused more narrowly on the relationship of medical and drug abuse care and hospitalization. Use of psychiatric and alcohol abuse services may further contribute to variation in hospitalization.⁴⁰

The study's strengths are its large population-based sample and empirically defined patterns of care. Our data offer compelling evidence of an association between drug users' outpatient care patterns and future hospitalization. Efforts to promote access to and retention in medical care and drug abuse treatment appear to be an attractive strategy for improving the health of this medically complex population.