Effects of Drinking on Hospital Stays and Emergency Room Visits Among Older Adults

Abstract

Objective: To evaluate whether alcohol drinking influences emergency room (ER) visits or hospital admissions among adults aged 65 and older. Method: Data from two independent national surveys are used to estimate multivariate logit models that quantify the relationship between drinking and ER visits and hospital admissions. The authors distinguish between ER visits linked to a hospital admission for that individual and ER visits not linked to an admission. Results: The authors find no significant effects of alcohol consumption on either ER visits or hospital admissions among older adults. These findings occur in both data sets, and for both men and women. Distinguishing between different types of ER visits makes no difference. Discussion: Analysis of two large and nationally representative data sets suggests that among older adults drinking alcohol, or even heavily drinking alcohol, does not raise or lower the risk of a hospital admission or the risk of an ER visit.

Keywords

alcohol drinking hospital emergency room elderly

Introduction

The use of alcohol among older adults in the United States is not uncommon. About half of all men aged 65 and older describe themselves as current drinkers, 27% to 39% report they drink one or fewer drinks per day, and 10% report they are heavier drinkers, meaning they usually drink more than one drink daily. Among women aged 65 and older one quarter to a third are current drinkers, 22% to 32% drink one or less drinks per day, and 2% to 3% drink more than one drink daily (Breslow, Faden, & Smothers, 2003). Research suggests that though light or moderate drinking may confer some protection from heart disease, heavy drinking contributes to a host of health problems, including the risk of sudden injuries, such as those resulting from car crashes or falls, diabetes, heart disease and strokes, malignant neoplasms, osteoporosis, behavioral problems, and chronic liver disease (Finlayson, 1995; Klatsky, 1999; Mukamal et al., 2005).

Because certain drinking patterns influence health, it is natural to ask whether they also contribute to the health care utilization of older adults, and if so, in what ways? This article addresses this question for two major categories of hospital care: inpatient stays and emergency room (ER) visits. Specifically, we empirically analyze how drinking patterns and other factors influence the occurrence of hospital stays and ER visits among community-dwelling adults aged 65 and older in the United States. We have nationally representative data from two separate and independent surveys to use in this analysis, and we examine the evidence from both of them. If our econometric findings are similar across the two data sets, then we can be more confident in our conclusions about the effects of drinking patterns on the use of hospital services. Our first data source is the 2001-2002 National Epidemiologic Survey on Alcohol and Related Conditions (NESARC). The second is the 2003-2004 Medical Expenditure Panel Survey (MEPS).

Understanding the effects of drinking patterns on hospital utilization among older adults is important for at least three reasons. First, there is widespread interest in knowing how drinking relates to one’s health, and hospitalizations and ER visits are often a direct result of adverse health events, such as sudden serious injuries or an acute medical problem, such as a heart attack or stroke. They are fundamentally different than physician office visits in this sense. Second, hospital care is expensive and its cost continues to grow rapidly. For seniors hospital costs are bourn primarily by the publically financed program, Medicare. Annual Medicare spending on hospital care, including ER visits and inpatient stays, averaged US$2,641 per senior in 2004 and comprised 44% of all Medicare spending per senior (Machlin, 2006). If heavy alcohol consumption increases Medicare spending by spurring hospital utilization, there may be a good financial argument for greater public investment in approaches to reduce problem drinking. Finally, with the aging of the large Baby Boomer cohort, the older population is growing rapidly, and this too makes alcohol’s consequences among older adults an important issue.

A number of previous studies have examined the association between alcohol consumption and health care utilization among adults generally, but their findings have tended to be mixed, making it difficult to generalize. Using samples drawn from hospital settings, several studies have found that in both hospital ERs and hospital inpatient facilities patients are more frequently heavy drinkers, and they report more alcohol-related problems, when compared to the general population (Adams, Yuan, Barboriak, & Rimm, 1993; Cherpitel, 1993); however, other studies have not found this to be the case (Cherpitel, 1994, 1995; Girard, Partridge, Becker, & Bock, 2004).

Studies based on community-drawn samples or on samples of enrollees in managed care organizations have also shown mixed findings, with most reporting a negative, if not insignificant, effect of alcohol consumption on health care utilization, even for heavy drinkers. For example, using data from the 1995 National Alcohol Survey, Cherpitel (1999) examined the effects of alcohol use on ER visits and primary care services but could find no consistent effects of drinking patterns on the use of health care. Rice et al. (2000) examined how hospitalizations and physician outpatient visits varied with drinking patterns in a large HMO and found that for both types of services, current drinkers, irrespective of drinking levels, had lower rates of health care utilization than nondrinkers. Polen et al. (2001) found that among enrollees in a large Kaiser Permanente health maintenance organization (HMO) drinking patterns were not important in predicting either the use of health care services or annual health care costs. Both measures were a little higher for former drinkers than for very light drinkers; however, life-time abstainers and heavy drinkers were comparable to very light drinkers in their use of health care. Zarkin, Bray, Babor, and Higgins-Biddle (2004) investigated the effects of alcohol use in a large managed care organization in the Western United States. Looking at inpatient days, outpatient visits, and ER visits, they too found little evidence that alcohol use was associated with increases in use of services. On the contrary, alcohol use, regardless of drinking patterns, was associated with decreases in utilization.

Salome, French, Matzger, and Weisner (2005) examined the effect of alcohol use on the probability of an inpatient stay and the occurrence of ER visits within a large sample of adults that included both the elderly and nonelderly. The sample was recruited from a county in northern California. Using several alternative measures of alcohol consumption, including number of drinks, heavy drinking days, and a problem drinking indicator, they found that most measures were unrelated to health care utilization. However, there was one exception: Problem drinking significantly increased the probability of ER use.

There are a few studies, however, which have reported a strong positive relationship between alcohol consumption and service utilization, again using community-based samples and a multiple regression framework. Alexandre, Roebuck, French, Chitwood, and McCoy (2001) and Balsa, French, Maclean, and Norton (2009), for example, found that problematic alcohol users had significantly more ER visits, more outpatient visits, and more inpatient days. In the United Kingdom, Cryer et al. (1999) found that heavy drinkers and abstainers were significantly more likely to use emergency services than were light drinkers.

There are at least two reasons to expect that the relationship between alcohol and health care utilization may be quantitatively different for older adults. First, older adults tend to be more vulnerable to the effects of alcohol due to biological changes that occur with aging (Dufour & Fuller, 1995; Kumar, 2003). For example, given the very same does of alcohol, older adults tend to experience higher blood alcohol levels than do younger adults, and consequently, their ability to perform routine activities are often more impaired than that of younger adults. Second, many older adults have one or more chronic diseases and are taking multiple medications. Alcohol interacts with some prescription drugs, and it can certainly complicate chronic medical problems, such as diabetes, hypertension, gastritis, heart failure, and liver disease. Thus, compared to younger adults, an excessive use of alcohol among older adults could well result in more frequent use of medical services, ER visits, and hospitalization, in particular due to accidental injuries and because alcohol complicates their chronic conditions.

To our knowledge, only one study has examined the effects of alcohol consumption on health care utilization among older adults. Balsa, Homer, Fleming and French (2008) used the 2001-2002 NESARC data to examine the effects of alcohol consumption on various indicators of health status and hospital utilization among the older adults. Their findings were mixed. Upon carefully estimating a series of multivariate models, they found no significant positive or negative effects of alcohol consumption among older men. Among older women, however, they found partial evidence of benefits associated with light-to-moderate alcohol consumption, specifically slightly better self-reported health, a reduced likelihood of heart problems, and a lower risk of hospitalization.

The present article extends earlier work on the effects of alcohol on health care utilization in three ways. First, like Balsa et al. (2008), we focus on older adults, who have been largely understudied in this subject area, yet who deserve more specific attention because of their growing presence in the population. Second, we analyze data from two recent large-scale surveys, the 2003-2004 MEPS, and the 2001-2002 NESARC, both of which enable us to generalize our findings to the older adults nationwide in the United States. The MEPS has not been utilized in this research area. Third, we pay special attention to the nature of ER visits, distinguishing between visits that are linked to a hospital stay for that same patient and visits that are not, and in so doing, explore whether there are subtle effects of alcohol that might otherwise go unnoticed.

Data

We first describe the MEPS data and then describe the NESARC data used for these analyses. The 2003-2004 MEPS is a large and nationally representative survey of individuals and their medical providers (doctors, hospitals, pharmacies, etc.) conducted by the Agency for Healthcare Research and Quality, which is part of the U.S. Department of Health and Human Services. The MEPS sampling frame is drawn from a nationally representative subsample of households that participated in the prior year’s National Health Interview Survey (NHIS), conducted by the National Center for Health Statistics (NCHS). The 2003-2004 MEPS gathered data from 16,681 respondents and achieved an overall response rate of 66%. For this study we limit our focus to the 3,778 respondents who were aged 65 or older at the time of interview, and we draw on the Household Component (HC) of MEPS, which is based on questionnaires fielded to individual household members and their medical providers. Household interview data were supplemented by information obtained during surveys of medical providers including hospitals, physicians, home health care providers, and pharmacies identified by MEPS-HC respondents. Together, these surveys gathered detailed information for each person in the household on demographic characteristics, health conditions, health status, use of medical services, access to care, satisfaction with care, health insurance coverage, income, and employment.

Of particular value for this study is the rich hospital-encounter data in MEPS, which was obtained from the hospitals that provided services to MEPS respondents. The survey collected detailed information for each reported hospital inpatient stay, including whether it began with an ER visit, allowing us to very precisely measure each of our three dependent variables.

The NESARC, our second data set, is a large and nationally representative survey jointly conducted in 2001 and 2002 by the National Institute on Alcohol Abuse and Alcoholism (NIAAA) and the National Institute on Drug Abuse (NIDA) of the National Institute of Health (NIH). Nationwide, more than 43,000 individuals participated in the survey and provided extensive information on their use of alcohol, tobacco, and drugs, as well a wealth of information about their demographic characteristics, health status, medical conditions, and use of health care services, among other things. Our focus here is limited to the 7,717 NESARC respondents who were 65 years old or older at the time of their interview.

Like many large-scale national surveys, both MEPS and NESARC adopted multistage complex survey design and oversampled various subpopulations. Consequently, the estimates reported in this article have been weighted, using the sampling weights provided on the data files. In the multivariate regression analysis, we also include stratum and primary sampling unit (PSU) to produce consistent estimates of the standard errors. All statistical analyses are carried out using STATA v10.

Method

We first use the MEPS and NESARC data to report on how the use of hospital services varies with drinking patterns among adults aged 65 and older. We then estimate a series of logit regressions to quantify more carefully the relationship between drinking and the use of hospital services. These logit models relate various types of hospital encounters to an individual’s drinking habits, their demographic characteristics, their health status, the nature of any chronic medical conditions they have, and their smoking habits. The general form for the models we estimate is:

Ln (P_{i} / (1 - P_{i})) = β_{0} + β_{1} DRIN K_{i} + β_{2} D_{i} + β_{3} HLT {Hi}_{i} + β_{4} CHRONI C_{i} + β_{5} SMOK E_{i} + ε_{i}

where P_i is the probability that individual i has a particular type of hospital utilization over the period, DRINK_i is a set of indicator variables describing individual i’s drinking patterns, D_i is a vector of his or her sociodemographic characteristics, HLTH_i is a vector of variables measuring the physical and mental health status of individual i, CHRONIC_i is a set of indicator variables describing whether individual i has various chronic health conditions, SMOKE_i measures individual i’s smoking status, and ϵ_i is a random error term. Our hospital utilization models are first estimated using the MEPS and then estimated using the NESARC data.

Hospital Utilization Measures

Hospital utilization is measured as uniformly as possible across the two surveys. It is described by three variables, each dichotomous. The first is whether the individual was ever hospitalized overnight during the past year, but none of the hospitalizations began in the ER. The variable equals 1 if yes, 0 otherwise. The second is whether the individual visited the ER during the past year and was hospitalized as a result. It equals 1 if yes, 0 otherwise. We refer to an ER visit that is linked with a hospitalization as “ER with hospitalization.” The third measure is whether the individual had any “ER without hospitalization” visits during the past year. It equals 1 if the respondent reported he or she visited an ER but never stayed in a hospital overnight during the past year.

Notice we are distinguishing between ER encounters that result in an inpatient stay and ER encounters in which the patient is sent home after treatment. This is because these types of visits may be very different from each other: different in terms of the nature and severity of conditions precipitating the visits, different in terms of individual’s total subsequent treatment costs, and possibly different in terms of alcohol’s influence on the individual’s risk for that type of visit. ER visits where the patient is sent home, for example, would certainly include visits for nonurgent medical conditions that arguably could have been treated in physician offices; instead, the patient is treated in the ER because that is where the he or she chose to go. On the other hand, ER visits that result in a hospitalization are likely much more serious. Examples include a patient having a heart attack or one suffering a hip fracture, which clearly necessitate a hospital stay. Some examples of nonurgent visits might include a patient with a sore throat or upper respiratory tract infection, a sprain, headache or backache, or a skin rash, none of which would likely require admission. Hospital charges can vary dramatically by type of ER visit. Average total charges for urgent ER visits have been shown to be roughly five times the charges for nonurgent ER visits (Williams, 1996). Nationally among adults aged 65 and older, 6% of ER visits are considered nonurgent, 16% semiurgent, 63% urgent/emergent, and the remaining 15% are visits for which the triage status is unknown or no triage system was used (McCaig & Nawar, 2006).

The MEPS contains detailed hospital encounter records for each respondent that record all of their ER and hospital utilization, and for every ER visit whether that visit resulted in immediate hospitalization. We used this information to construct the three utilization measures described earlier for each MEPS respondent.

In the NESARC two separate questions were used to ask each respondent whether he or she had any ER visits during the past year and whether he or she had any overnight hospital stays during the past year. The ER and hospital stay data are not directly linked, as they are in the MEPS. If a NESARC respondent said yes to both questions, we assume the respondent had an ER visit that resulted in hospitalization. We then were able to construct for each NESARC respondent the three hospital utilization measures described earlier.

How accurate is it to define an ER visit and a hospital stay to be linked if a respondent simply reported he or she had both during the past year? For older adults it is accurate the vast majority of the time. We say this on the basis of tabulations we conducted using the MEPS data (described previously), which precisely records the sequence and timing of all hospital encounters of respondents. Among adults aged 65 and older MEPS data show that among those who report both an ER visit and a hospitalization during the past year 86% are actually describing a hospitalization that began in the ER, that is, ER with hospitalization. Although this may not apply to the general population, it makes sense for older adults because ER visits often serve as the front door to other hospital services. Thus, though there is some measurement error in our NESARC classification of inpatient ER visits versus noninpatient ER visits, it is very infrequent. We will come back to this issue of measurement error in the discussion section.

Explanatory Variables

An individual’s drinking habits are characterized similarly across the two surveys, by including five dummy variables that correspond to the following mutually exclusive drinking categories: (1) life-time abstainers: those who never drank; (2) former drinkers: those who used to drink but who have not in the past year; (3) light drinkers: those who drank 3 or fewer drinks per week; (4) moderate drinkers: men who drank 3 to 14 drinks per week or women who drank 3 to 7 drinks per week, and (5) heavy drinkers: men who drank more than 2 drinks per day, or women who drank more than 1 drink per day.

Although the MEPS did not ask about alcohol use directly, the NHIS did for its sample adults, which includes MEPS respondents as a subset. Therefore, we merged 2003 and 2004 MEPS data with the corresponding NHIS data (2001-2003), using the linkage files provided by AHRQ, to obtain the relevant information on each MEPS respondent’s drinking patterns. In its health behavior section, the NHIS asked a series of questions to assess the life-time drinking status for all sample adults. Drinking patterns were determined based on the responses to these questions. In the MEPS analyses, those who did not drink in the past year are classified as nondrinkers. Among nondrinkers, life-time abstainers are those who had 12 or less drinks in life-time, and former drinkers are those had more than 12. Among current drinkers, light drinkers are those who drank no more than 3 drinks per week, moderate drinkers are men who drank no more than 14 drinks per week or women who drank no more than 7, and heavy drinkers are men who drank more than 14 per week or women who drank more than 7 per week.

Although NESARC asked about drinking patterns differently than did the NHIS, it is still possible to categorize respondents into the five drinking categories described earlier. Specifically, NESARC asked a detailed series of questions concerning both the frequency and quantity of consumption of each of four different types of alcoholic beverages: coolers, beer, wine, and liquor. The recall period for each was the past 12 months. For each type of beverage, each NESARC respondent’s frequency and quantity responses were then converted to accurately reflect the individual’s ethanol intake, based on the ethanol content in each beverage. Adding these ethanol amounts across the four types of beverages and dividing by 365 provides an average daily volume of ethanol intake for each respondent. For more specific details regarding the algorithms used for calculating average daily ethanol intake, readers are referred to NESARC’s Data Reference Manual (National Institute of Health & National Institute on Alcohol Abuse and Alcoholism, 2006). The NESARC data on average daily ethanol intake together with a question on current drinking status were then used to assign each respondent into one of the five distinct drinking categories defined earlier (assuming one standard drink contains 0.6 ounces of ethanol).

Sociodemographic characteristics in all of the estimated models include age, gender, marital status, race, education, and income. With regard to marital status, a respondent was considered “married” if he or she reported being married or living with someone as if married, and “not married” if he or she reported being widowed, separated, divorced, or never married. Race was categorized as White, Black, or Other. Education was coded into three categories based on the respondent’s highest grade or year of school completed. The education categories include less than high school, high school graduate or GED, or college graduate and above. Personal income is also included in the models, and it was coded into income categories due to the presence of right-sided outliers in the distribution. The income categories are less than US$10,000, US$10,000 to US$24,999, US$25,000 to US$49,999, or US$50,000 or more.

Measures of health status and chronic medical conditions are included in all of the models, as well. In both surveys, respondents were asked to rate their health on a 5-point scale ranging from poor, fair, good, very good, to excellent. We include a dummy variable in the models that reflects whether the respondent assessed his or her own health as fair to poor. There are also two summary scales included in the models to measure the respondent’s overall physical and mental functioning, respectively. For the former we use the norm-based physical component summary Scale (NBPCS) calculated from Version 2 of the Short Form 12 Health Survey (SF12-V2), and for the latter, the norm-based mental component summary Scale (NBMCS). For both scales, scores were standardized with a mean of 50 and standard deviations of 10, and ranged from 0 to 100. A scale score below 50 indicates the respondent’s physical/mental ability was below average and vice versa. For details on the construction of the NBPCS and NBMCS using the SF12-V2, and their reliability as measures of physical and mental functioning, readers are referred to Ware, Kosinski, Turner-Bowker, and Gandek (2002).

In models based on the MEPS data we control for seven specific health conditions: (1) high blood pressure or hypertension, (2) chest pain or angina pectoris, (3) heart problems, (4) arthritis, (5) diabetes, (6) asthma, and (7) joint pain. With NESARC we are able to measure the first four of these but not the last three. NESARC contains measures of three other serious conditions, so these were included instead in the NESARC models: (8) hardening of the arteries or arteriosclerosis, (9) liver problems, and (10) stomach ulcer or gastritis.

Respondents in both surveys were asked about their tobacco use. We include three variables that indicate whether the respondent is a current smoker, a former smoker, or a life-time nonsmoker.

Finally, because many ER visits are due to sudden injuries, each of the estimated models also controls for the occurrence of injuries, using an indicator variable derived from responses to the following question: “How many injuries have you had in the past year that caused you to seek medical help or cut down on your usual activities for more than half a day?” We include an indicator for whether the respondent reported a positive number.

Table 1 provides brief variable definitions and descriptive statistics from each data set for all variables used in the analysis, including their means or proportions, and for continuous variables, their standard deviations, as well.

Table 1.

Variable Definitions and Descriptive Statistics

		Mean or percentage
		MEPS	NESARC
Variable	Definition	n = 3,778	n = 7,717
Hospital utilization
Hosponly	1 if ever stayed in hospital overnight but never treated in an ER in the past year; 0 otherwise	7.88%	7.95%
Erhosp	1 if ever treated in an ER AND stayed in hospital overnight in the past year; 0 otherwise	10.79%	12.49%
Eronly	1 if ever treated in an ER but NEVER stayed in hospital overnight in the past year; 0 otherwise	9.25%	10.96%
Sociodemographics
Age	Age at interview	74.96 (6.38)	74.60 (6.9)
Female	1 if female; 0 if male	65.43%	57.96%
Married	1 if married; 0 if single	36.87%	57.82%
White	1 if White; 0 otherwise	73.85%	82.34%
Black	1 if Black; 0 otherwise	12.76%	7.87%
Other	1 if race is other than White and Black; 0 otherwise	13.39%	9.79%
Educ1	1 if education is less than high school; 0 otherwise	35.05%	27.48%
Educ2	1 if education is high school or GED; 0 otherwise	45.37%	55.47%
Educ3	1 if education is college or above; 0 otherwise	19.59%	17.05%
Income1	1 if income is less than US$10,000; 0 otherwise	32.85%	31.81%
Income2	1 if income is US$10,000-US$25,000; 0 otherwise	39.78%	40.18%
Income3	1 if income is US$25,000-US$50,000; 0 otherwise	19.35%	19.50%
Income4	1 if income is US$50,000 or more; 0 otherwise	8.02%	8.50%
Health
SPH	1 if self-perceived health is fair or poor; 0 if good, very good, or excellent	27.90%	31.48%
NBPCS	Norm-Based Physical Disability Scale (SF12-V2R)	40.56 (12.22)	42.46(13.07)
NBMCS	Norm-Based Mental Disability Scale (SF12-V2R)	50.93 (10.51)	52.9 (10.33)
Injury	1 if had any injuries that resulted in seeking medical help in the past year; 0 otherwise	35.28%	12.95%
Chronic conditions
Joint pain (arteries for NESARC)	1 if had join pain (hardening of arteries) in the past year; 0 otherwise	58.36%	7.57%
Highblood	1 if had high blood pressure or hypertension in the past year; 0 otherwise	63.10%	4.87%
Diabetes (liver for NESARC)	1 if had diabetes (liver problems) in the past year; 0 otherwise	19.69%	0.89%
Chest	1 if had chest pain or angina pectoris in the past year; 0 otherwise	8.97%	12.48%
Heart	1 if had heart problems in the past year; 0 otherwise	24.38%	18.41%
Asthma (ulcer_gastr for NESARC)	1 if had asthma (stomach ulcer or gastritis) in the past year; 0 otherwise	9.71%	12.48%
Arthritis	1 if had arthritis in the past year; 0 otherwise	53.44%	55.41%
Smoking status
nonsmoker	1 if life-time nonsmoker; 0 otherwise	50.58%	47.79%
exsmoker	1 if former smoker; 0 otherwise	37.19%	38.67%
currsmoker	1 if smoked in the past year; 0 otherwise	12.23%	13.54%
Drinking status
Abstainer	1 if life-time abstainer; 0 otherwise	32.03%	26.12%
exdrinker	1 if former drinker; 0 otherwise	27.34%	28.39%
Light	1 if light drinker in the past year; 0 otherwise	28.03%	29.74%
Moderate	1 if moderate drinker in the past year; 0 otherwise	9.71%	9.48%
Heavy	1 if heavy drinker in the past year; 0 otherwise	2.89%	6.27%
Smoking status
Nonsmoker	1 if life-time nonsmoker; 0 otherwise	50.58%	47.79%
Ex-smoker	1 if former smoker; 0 otherwise	37.19%	38.67%
Currsmoker	1 if smoked in the past year; 0 otherwise	12.23%	13.54%
Drinking status
Abstainer	1 if life-time abstainer; 0 otherwise	32.03%	26.12%
Ex-drinker	1 if former drinker; 0 otherwise	27.34%	28.39%
Light	1 if light drinker in the past year; 0 otherwise	28.03%	29.74%
Moderate	1 if moderate drinker in the past year; 0 otherwise	9.71%	9.48%
Heavy	1 if heavy drinker in the past year; 0 otherwise	2.89%	6.27%

Note. MEPS = Medical Expenditure Panel Survey; NESARC = National Epidemiologic Survey on Alcohol and Related Conditions; SPH = Self-Perceived Health; NBPCS = Norm-Based Physical Component Scale; NBMCS = Norm-Based Mental Component Scale. Standard deviation for continuous variables in parentheses.

Results

As shown in Table 1, among adults aged 65 and older the prevalence of drinkers is 41% based on MEPS data and 45% based on NESARC data. The prevalence of light drinking is 28% in MEPS and 30% in NESARC, and the prevalence of moderate drinking is 10% in MEPS and 9% in NESARC. Heavy drinking is uncommon among older adults; its prevalence is only 3% in MEPS and 6% in NESARC. Heavy drinking is clearly not as prevalent as it is among younger cohorts. These distributions reflects the fact that many older adults used to drink but have since quit (27% in MEPS and 28% in NESARC), and many of those who still drink are apparently limiting their alcohol intake; 32% of older adults in MEPS and 26% of older adults in NESARC describe themselves are life-time abstainers.

Figure 1 summarizes the prevalence of hospital stays and ER visits among adults aged 65 and older, based on the MEPS and the NESARC data. Based on the MEPS data, about 18% of seniors have been hospitalized over the past 12 months and 20% had an ER visit. On the basis of NESARC data, we found the corresponding percentages are 20% and 23%, respectively. There is substantial overlap in these two forms of utilization, however. About 11% of seniors in the MEPS report that their hospital stay began as an ER visit, and in the NESARC, 12% of seniors report both a hospital stay and an ER visit over the past 12 months. To express this overlap in other ways, in the MEPS data, just more than half (54%) of all ER visits among older adults turn into inpatient stays, and 58% of all their hospital stays begin as an ER visit.

Figure 1.

Use of hospital inpatient and emergency room (ER) services among older adults aged 65 and older

Table 2 reports how hospital utilization varies with drinking patterns among older adults. With either data set, these statistics, which do not adjust for the effects of other factors influencing hospital utilization, suggest that among older adults drinking patterns do not alter the overall pattern of hospital utilization by much. The broad picture here is that unadjusted rates of hospital utilization among seniors vary little, if at all, by drinking patterns. Notable exceptions are that heavy drinkers (based on the MEPS) and former drinkers (based on the NESARC) seem to have more-than-average hospitalizations that begin in the ER (15% vs. 11% and 16% vs. 12%, respectively). However, the same is not true based on the alternative dataset. Pearson’s chi-square statistics, adjusted for survey design effect, suggest that all three types of hospital utilization are unrelated to drinking status, with the only exception being hospitalizations that begin in the ER based on the NESARC data.

Table 2.

Hospitalizations and Emergency Room (ER) Use Among Adults Aged 65 and Older by Drinking Status

MEPS	Hospitalization only	ER visit with hospitalization	ER visit without hospitalization
All persons	7.88%	10.79%	9.25%
By drinking status
Life-time abstainer	6.39%	12.32%	10.29%
Former drinker	9.23%	11.20%	10.09%
Light drinker	7.82%	9.23%	8.60%
Moderate drinker	7.95%	9.01%	7.31%
Heavy drinker	9.80%	15.20%	7.39%
NESARC	Hospitalization only	ER visit with hospitalization	ER visit without hospitalization
All persons	7.95%	12.49%	10.96%
By drinking status
Life-time abstainer	9.75%	13.55%	10.31%
Former drinker	7.48%	16.06%	10.57%
Light drinker	7.07%	9.89%	11.56%
Moderate drinker	8.04%	9.68%	10.76%
Heavy drinker	6.65%	8.52%	12.85%

Note. MEPS = Medical Expenditure Panel Survey; NESARC = National Epidemiologic Survey on Alcohol and Related Conditions. All percentages are weighted to reflect population prevalence rates.

Do these results change if we control for the multitude of other factors that can influence utilization? Tables 3 and 4 address this question. They report the estimated multivariate logit models for various types of hospital utilization. Table 3 reports the MEPS models, whereas Table 4 reports models based on NESARC. It is clear from both data sets that once one controls for age, gender, race, marital status, income, health and functioning, smoking habits, chronic conditions, and the occurrence of injuries, there are no statistically significant effects of drinking patterns on hospital utilization among older adults. This statement applies to all three hospital utilization measures for which models were estimated.

Table 3.

Logit Regression Results for Hospitalization and Emergency Room (ER) Visit, MEPS Analysis

	Hospitalization only		ER with hospitalization		ER without hospitalization
	Odds ratio (SE)		Odds ratio (SE)		Odds ratio (SE)
Drinking status
Abstainer	0.654	(0.195)	0.929	(0.253)	1.154	(0.346)
Ex-drinker	0.904	(0.263)	0.734	(0.185)	1.195	(0.348)
Light	0.952	(0.271)	0.916	(0.236)	1.127	(0.309)
Heavy	1.076	(0.461)	1.678	(0.555)	0.950	(0.439)
Sociodemographics
Age	1.013	(0.014)	1.031^b	(0.012)	1.009	(0.013)
Female	1.017	(0.176)	0.902	(0.163)	1.079	(0.168)
Married	1.217	(0.198)	0.699^b	(0.108)	1.045	(0.174)
Black	1.362	(0.257)	0.885	(0.158)	1.207	(0.304)
Other	1.062	(0.249)	0.542^b	(0.152)	0.908	(0.223)
Educ2	1.075	(0.148)	1.159	(0.224)	0.735^a	(0.124)
Educ3	0.934	(0.212)	1.234	(0.299)	0.592^b	(0.126)
Income2	1.013	(0.158)	0.905	(0.131)	1.111	(0.174)
Income3	1.026	(0.237)	0.779	(0.157)	1.148	(0.221)
Income4	1.221	(0.337)	0.695	(0.219)	1.435	(0.421)
Health
SPH	0.985	(0.197)	1.469^a	(0.287)	1.368^a	(0.222)
NBPCS	0.966^c	(0.008)	0.984^b	(0.008)	1.013^b	(0.007)
NBMCSnbmcs	0.991	(0.007)	0.991	(0.006)	0.992	(0.006)
Injury	1.825^c	(0.291)	5.019^c	(0.761)	4.682^c	(0.665)
Chronic conditions
Joint pain	0.990	(0.166)	0.867	(0.139)	1.362^a	(0.244)
Highblood	0.866	(0.147)	0.959	(0.149)	1.064	(0.168)
Diabetes	1.009	(0.185)	1.199	(0.187)	0.775	(0.140)
Chest	0.770	(0.184)	1.262	(0.216)	0.965	(0.226)
Heart	1.077	(0.194)	2.358^c	(0.353)	1.042	(0.167)
Asthma	0.498^b	(0.161)	1.513^a	(0.333)	1.230	(0.306)
Arthritis	0.988	(0.180)	0.888	(0.146)	0.819	(0.151)
Smoking status
Ex-smoker	1.198	(0.213)	0.818	(0.143)	0.878	(0.129)
Currsmoker	1.055	(0.269)	1.215	(0.275)	1.023	(0.240)
Intercept	0.162	(0.215)	0.016^c	(0.018)	0.017^c	(0.022)

Note. MEPS = Medical Expenditure Panel Survey; SPH = Self-Perceived Health; NBPCS = Norm-Based Physical Component Scale; NBMCS = Norm-Based Mental Component Scale.

Significant at the 10% level.

Significant at the 5% level.

Significant at the 1% level.

Table 4.

Logit Regression Results for Hospitalization and Emergency Room (ER) Visit, NESARC Analysis

	Hospitalization only		ER with hospitalization		ER without hospitalization
	Odds ratio (SE)		Odds ratio (SE)		Odds ratio (SE)
Drinking status
Abstainer	1.016	(0.237)	0.952	(0.222)	0.770	(0.139)
Ex-drinker	0.737	(0.147)	1.051	(0.265)	0.775	(0.138)
Light	0.843	(0.161)	0.847	(0.205)	0.889	(0.135)
Heavy	0.844	(0.260)	0.723	(0.224)	1.094	(0.212)
Sociodemographics
Age	1.005	(0.009)	1.017^b	(0.008)	1.007	(0.007)
Female	0.777^b	(0.094)	0.800^a	(0.103)	1.374^b	(0.189)
Married	0.964	(0.126)	0.765^b	(0.101)	1.193	(0.146)
Black	0.923	(0.133)	1.166	(0.151)	1.307^a	(0.197)
Other	0.817	(0.175)	1.107	(0.175)	1.226	(0.210)
Educ2	0.953	(0.143)	1.050	(0.153)	0.973	(0.126)
Educ3	1.228	(0.211)	1.053	(0.199)	1.075	(0.188)
Income2	1.074	(0.173)	1.221	(0.149)	1.099	(0.164)
Income3	1.010	(0.194)	1.166	(0.194)	1.176	(0.193)
Income4	0.759	(0.151)	0.935	(0.215)	1.256	(0.239)
Health
SPH	0.896	(0.138)	1.151	(0.174)	1.304^a	(0.200)
NBPCS	0.971^c	(0.006)	0.971^c	(0.006)	1.003	(0.005)
NBMCS	1.002	(0.006)	0.985^c	(0.005)	0.993	(0.006)
Injury	0.746	(0.143)	3.349^c	(0.425)	5.604^c	(0.585)
Chronic conditions
Arteries	1.004	(0.245)	1.301	(0.243)	0.992	(0.213)
Highblood	1.067	(0.135)	1.137	(0.139)	1.123	(0.145)
Liver	0.326^c	(0.069)	1.535	(0.779)	0.573^c	(0.117)
Chest	1.088	(0.212)	2.182^c	(0.309)	0.914	(0.176)
Heart	1.586^b	(0.298)	2.281^c	(0.331)	1.009	(0.132)
Ulcer_gastr	0.946	(0.173)	1.238	(0.198)	1.234	(0.173)
Arthritis	1.040	(0.153)	0.729^b	(0.094)	1.328^b	(0.174)
Smoking status
Ex-smoker	0.907	(0.146)	1.371^b	(0.190)	1.178	(0.153)
Currsmoker	0.730	(0.186)	1.184	(0.222)	1.251	(0.276)
Intercept	0.238	(0.260)	0.153^a	(0.145)	0.033^c	(0.028)

Note. NESARC = National Epidemiologic Survey on Alcohol and Related Conditions; SPH = Self-Perceived Health; NBPCS = Norm-Based Physical Component Scale; NBMCS = Norm-Based Mental Component Scale.

Significant at the 10% level.

Significant at the 5% level.

Significant at the 1% level.

A few other interesting findings emerge from Tables 3 and 4 and deserve mention. Perhaps not surprisingly, the occurrence of injuries is one of the strongest predictors of all three types of hospital utilization, especially ER visits. Depending on the data set used, these models imply that older adults who have suffered an injury are 3.3 to 5 times more likely (p < .0001 for both estimates) to have an ER visit linked to a hospitalization and 4.7 to 5.6 times more likely (p < .0001 for both estimates) to have an ER visit that does not result in a hospitalization, compared to older adults who have not had an injury. Also in the NESARC, seniors who are ex-smokers are 1.4 times more likely (p = .003) to report ER use linked to hospitalization. In the MEPS, smoking does not have effects on hospital utilization.

Among the demographics in the models, age and marital status are important predictors of hospital utilization, whereas education and income are not. In both the MEPS and NESARC models, age is a significant predictor of having a hospitalization that begins with an ER visit. Interestingly, in both sets of models, married older adults are only 0.7 to 0.77 times as likely (p < .0001) to have a hospitalization that begins in the ER, compared to older adults who are not married. It seems that marriage provides older adults some protection against emergency hospitalizations. The effects of gender vary across the two data sets. In our NESARC models women have fewer hospitalizations than do men, and this applies to both hospitalizations that begin in the ER and hospitalizations that begin in the admissions department; however, in the MEPS models, there are no differences by gender.

A senior’s physical and mental health status and his or her chronic conditions are also predictors of hospital utilization, as one would expect. Considering all of the health measures we included, the norm-based physical component summary score (NBPCS) is the strongest predictor of an older adult’s hospital utilization. It is highly significant in five of the six models we estimated. As expected, the worse the senior’s overall physical health (i.e., the lower their NBPCS score), the higher their risk of hospital encounters, with the exception of nonurgent ER visits that do not result in hospitalization. Having fair or poor self-rated health is also predictive of ER utilization. Generally speaking, the risk of all forms of hospitalization tends to increase with the presence of physical or mental problems, particularly heart problems.

Sensitivity Analyses

We also estimated these hospital utilization models a number of other ways to check the robustness of our findings. Our substantive findings remained unchanged. First, we reestimated all of the models in Tables 3 and 4 without controlling for the occurrence of injuries, given the possibility that our injury variable might be picking up some of the effects of drinking since drinking can obviously lead to injuries. Even with this change in model specification, however, not much changed. All of the coefficients on the drinking variables remained statistically insignificant. In fact, the magnitudes of the coefficient estimates were very similar.

Recognizing that the pathways from alcohol consumption to service use could also include other risk factors, that is, alcohol consumption could lead to worse physical/mental health (as measured by self-rated health and SF-36 component scores) or the onset of chronic conditions, we further dropped all health and disease variables to estimate the total indirect effects of drinking on the use of inpatient and ER services. Even with this specification, our conclusion of “no relationship” between alcohol consumption and service use still held true in most models. The only exception was that the effect of heavy drinking on ER visits linked to subsequent hospitalization became statistically significant based on the MEPS data (odds ratio = 1.847, p = .057). However, this effect remained statistically insignificant in the NESARC data. In addition, the magnitudes of most point estimates only increased slightly in both data sets.

We then reestimated the hospital utilization models separately for men and women, given the possibility that the effects of drinking habits on health care utilization differ by gender. Although the coefficient estimates generally differed between men and women, they remained statistically insignificant across all models in both data sets.

Finally, we reestimated the models treating the drinking status variables as endogenous. Drinking variables may be endogenous because unobserved factors such as risk aversion and genetics may affect both alcohol use and hospital utilization. Endogeneity may also result from reverse causality; that is, those who have been hospitalized or treated in the ER may decide to cut back on alcohol use due to health concerns. To address the potential endogeneity of drinking status, we reestimated the models using an instrumental variable (IV) approach. In this analysis, we restricted our attention to older adults in the MEPS sample who were married, so that we could implement an array of health behavior variables of the spouse as potential instruments. We ended up choosing the frequencies of dental checkups, routine physical checkups, and flu shots (all categorical variables) because of their relatively strong predictive power in the reduced-form equations for drinking status. In carrying out the IV estimation, we used the two-stage residual inclusion method instead of the more commonly used two-stage predictor substitution method because our regression models are nonlinear (Terza, Basu, & Rathouz, 2008). Our substantive results, namely that drinking status has no effect on hospital utilization, remained largely unchanged. Incidentally, our exogeneity tests of drinking habits using the regression-based procedure described in Wooldridge (2002) failed to reject the hypothesis that drinking habits were exogenous in the hospital utilization models. This finding suggests that the single-equation estimation is preferred because it should yield more precise coefficient estimates in the absence of endogeneity.

Discussion

Using two independent, nationally representative, and large samples of adults aged 65 and older, this article finds consistent and clear evidence that alcohol-drinking patterns are unrelated to hospital utilization. Upon controlling for other determinants of hospital use, such as sociodemographics, injury occurrence, and the presence of physical disabilities or chronic conditions, such as diabetes and heart disease, drinking patterns are found to be unrelated to the probability of both inpatient utilization and ER visits among seniors.

These results are generally consistent with many prior studies on the relationship between drinking and the use of health services among the nonelderly. The present study shows that this pattern remains the same among the elderly. We concur with Balsa et al. (2008) that there is no strong evidence on the beneficial health effects of moderate drinking among the older adults. Several studies have reported that relatively low levels of alcohol consumption may still lead to adverse consequences and that even moderate drinking may entail risks for older adults, such as stroke caused by bleeding or interactions with their medications, and these risks could very well offset the benefits of moderate drinking (Camargo, 1989; Gunzerath, Faden, Zakhari, & Warren, 2004; Werch, 1987). Compared to younger people, older adults are at higher risk because many of them have chronic conditions and regularly take multiple medications. Any clinical advice to older adults about drinking should clearly acknowledge the trade-offs involved.

What is puzzling is the finding that heavy drinking also does not have statistically significant effects on hospital utilization, though it is again in line with most previous studies for the nonelderly. One plausible explanation is that older adults who drink excessively may experience worse health but choose not to seek medical help until a later time when their medical conditions become catastrophic. In fact, our sensitivity analysis rendered some support for this explanation because heavy drinking turned out to have a fairly strong indirect effect on ER visits that are more urgent in nature, namely, those that result in hospitalization. However, our evidence is far from convincing (this finding does not hold true in the NESARC data). We conclude that this type of “delayed” effect on hospital use is difficult, if not impossible, to detect with cross-sectional data.

Although the findings of the present study generally coincide with those of Balsa et al. (2008), our analysis has brought an additional and important data source to bear on this question, namely, the MEPS. Our findings also demonstrate that even if one distinguishes between ER visits linked to a hospitalization and ER visits not linked to a hospitalization, there is still no apparent relationship between drinking patterns and ER utilization among older adults. We have also demonstrated that this conclusion of “no relationship” applies, even with the adoption of a detailed five-group characterization of drinking patterns and the estimation of models that also control for the occurrence of injuries and the health and functioning of the individual. This too has not been shown before.

This study is not without its limitations, however. First, it is conceivable that our characterization of drinkers according to their average daily volume is masking the effects of specific harmful patterns of drinking, such as heavy episodic alcohol consumption. For example, heavy drinkers in our data include individuals who drink 2 drinks daily as well as those who drink 14 drinks every Saturday night but who otherwise abstain from alcohol. It is possible that the latter pattern has implications for hospital utilization; our analysis did not address this possibility. Second, when using the NESARC to break up all ER visits into inpatient ER and noninpatient ER, we are aware of potential measurement error. It arises because NESARC asked about whether there were any hospitalizations and ER use in two separate questions, as opposed to asking for each reported ER visit whether a hospitalization occurred in conjunction with the visit. The econometric literature suggests that when a dependent variable is measured in error, one’s coefficient estimates are generally unbiased; however, the estimated standard errors will be larger, thus making it more difficult to identify significant effects (Maddala, 2001). Given the lack of significant findings in our analysis of the MEPS—however, where potential measurement error is definitely not a concern—we suggest that even if our NESARC measures were precisely measured we would have still not found a relationship between drinking and hospital use. Third, the effects of alcohol consumption on the use of ER may vary with the severity of the patient’s condition and the intensity of services provided to the patient. With medical claims data, it is possible to classify the severity and intensity of an ER encounter based on the ICD9-CM diagnostic codes (i.e., the “NYU” algorithm) and CPT codes (Billings, 2003; Kaskie et al., 2010; Wharam et al., 2007). Unfortunately, we are unable to undertake this approach because neither the MEPS nor the NESARC provide fully specified ICD-9 codes or CPT codes. The approach we took, namely, distinguishing between ED with and without subsequent hospitalization, while allowing us to roughly separate out urgent and nonurgent ER use, cannot fully account for the severity and intensity of the service. Finally, we have limited our focus in this article to modeling hospitalizations and ER utilization and have not examined the effects of drinking on the use of other types of health care, such as physician office visits, laboratory services, and prescription drug use. Correcting each of these limitations represents fruitful endeavors for future research.

Footnotes

Acknowledgements

The authors would like to thank Dr. Allen Goodman of Wayne State University for his helpful comments on earlier versions of the article.

This study was completed while Dr. Li was an assistant professor of health policy and management at Indiana University School of Medicine.

The authors declared no potential conflicts of interests with respect to the authorship and/or publication of this article.

The authors received no financial support for the research and/or authorship of this article.

References

Adams

W. L.

Yuan

Barboriak

J. J.

Rimm

A. A.

(1993). Alcohol-related hospitalizations of elderly people. Prevalence and geographic variation in the United States. Jama, 270(10), 1222-1225.

Alexandre

P. K.

Roebuck

M. C.

French

M. T.

Chitwood

D. D.

McCoy

C. B.

(2001). Problem drinking, health services utilization, and the cost of medical care. Recent Developments in Alcoholism, 15, 285-306.

Balsa

French

Maclean

Norton

(2009). From pubs to scrubs: Alcohol misuse and health care use. Health Services Research, 44, 1480-1503.

Balsa

A. I.

Homer

J. F.

Fleming

M. F.

French

M. T.

(2008). Alcohol consumption and health among elders. The Gerontologist, 48(5), 622.

Billings

(2003). Using administrative data to monitor access, identify disparities, and assess performance of the safety net. In Billings

Weinick

(Eds.), A took kit for monitoring the local safety net (pp. 305-311). Rockville, MD: Agency for Healthcare Research and Quality.

Breslow

R. A.

Faden

V. B.

Smothers

(2003). Alcohol consumption by elderly Americans. Journal of studies on alcohol, 64(6), 884-893.

Camargo

Jr. (1989). Moderate alcohol consumption and stroke. The epidemiologic evidence. Stroke, 20(12), 1611.

Cherpitel

C. J.

(1993). Alcohol consumption among emergency room patients: Comparison of county/community hospitals and an HMO. Journal of studies on alcohol, 54(4), 432.

Cherpitel

C. J.

(1994). Alcohol use among primary care patients: Comparing and HMO with county clinics and the general population. Drug and Alcohol Dependence, 36(3), 167-173.

10.

Cherpitel

C. J.

(1995). Alcohol use among HMO patients in the emergency room, primary care and the general population. Journal of Studies on Alcohol, 56(3), 272-276.

11.

Cherpitel

C. J.

(1999). Emergency room and primary care services utilization and associated alcohol and drug use in the United States general population. Alcohol and Alcoholism, 34(4), 581-589.

12.

Cryer

P. C.

Jenkins

L. M.

Cook

A. C.

Ditchburn

J. S.

Harris

C. K.

Davis

A. R.

Peters

T. J.

(1999). The use of acute and preventative medical services by a general population: relationship to alcohol consumption. Addiction, 94(10), 1523-1532.

13.

Dufour

Fuller

R. K.

(1995). Alcohol in the elderly. Annual Review of Medicine, 46, 123.

14.

Finlayson

R. E.

(1995). Comorbidity in elderly alcoholics. In Beresford

Gomberg

(Eds.), Alcohol and aging (pp. 56-69). New York, NY: Oxford University Press.

15.

Girard

D. D.

Partridge

R. A.

Becker

Bock

(2004). Alcohol and tobacco use in the elder emergency department patient: Assessment of rates and medical care utilization. Academic Emergency Medicine, 11(4), 378-382.

16.

Gunzerath

Faden

Zakhari

Warren

(2004). National Institute on Alcohol Abuse and Alcoholism report on moderate drinking. Alcoholism: Clinical and Experimental Research, 28(6), 829-847.

17.

Kaskie

Obrizan

Cook

Jones

Liu

Bentler

. . . Chrischilles

E. A.

(2010). Defining emergency department episodes by severity and intensity: A 15-year study of Medicare beneficiaries. BMC Health Services Research, 10(1), 173.

18.

Klatsky

A. L.

(1999). Moderate drinking and reduced risk of heart disease. Alcohol Research & Health, 23(1), 15-23.

19.

Kumar

(2003). General topics in geriatric psychiatry. In Hazzard

W. R.

Blass

J. P.

Halter

J. B.

Ouslander

J. G.

Tinetti

M. E.

(Eds.), Principles of geriatric medicine and gerontology (5th ed., pp. 1465-1478). New York, NY: McGraw-Hill.

20.

Machlin

S. R.

(2006). Expenses for a hospital emergency room visit, 2003. Rockville, MD: Agency for Healthcare Research and Quality.

21.

Maddala

G. S.

(2001). Introduction to Econometrics. New York, NY: John Wiley.

22.

McCaig

L. F.

Nawar

E. W.

(2006). National Hospital Ambulatory Medical Care Survey: 2004 emergency department summary. Advance Data, (372), 1-29.

23.

Mukamal

Ascherio

Mittleman

Conigrave

Camargo

Kawachi

. . . Rimm

E. B.

(2005). Alcohol and risk for ischemic stroke in men: The role of drinking patterns and usual beverage. Annals of internal medicine, 142(1), 11.

24.

National Institute of Health & National Institute on Alcohol Abuse and Alcoholism. (2006). U.S. alcohol epidemiologic data reference manual (Vol. 5, No. 1; NIH Publication No. 05-5737). Bethesda, MD: Authors.

25.

Polen

M. R.

Green

C. A.

Freeborn

D. K.

Mullooly

J. P.

Lynch

(2001). Drinking patterns, health care utilization, and costs among HMO primary care patients. The Journal of Behavioral Health Services and Research, 28(4), 378-399.

26.

Rice

D. P.

Conell

Weisner

Hunkeler

E. M.

Fireman

(2000). Alcohol drinking patterns and medical care use in an HMO setting. The Journal of Behavioral Health Services and Research, 27(1), 3-16.

27.

Salome

H. J.

French

M. T.

Matzger

Weisner

(2005). Alcohol consumption, risk of injury, and high-cost medical care. The Journal of Behavioral Health Services & Research, 32(4), 368.

28.

Terza

J. V.

Basu

Rathouz

P. J.

(2008). Two-stage residual inclusion estimation: Addressing endogeneity in health econometric modeling. Journal of Health Economics, 27(3), 531-543.

29.

Ware

J. E.

Kosinski

Turner-Bowker

D. M.

Gandek

(2002). How to score version 2 of the SF-12 Health Survey. Lincoln, RI: QualityMetric Incorporated.

30.

Werch

(1987). Relationship between alcohol consumption and alcohol problems in young adults. Journal of drug education, 17(3), 261-276.

31.

Wharam

Landon

Galbraith

Kleinman

Soumerai

Ross-Degnan

(2007). Emergency department use and subsequent hospitalizations among members of a high-deductible health plan. JAMA, 297(10), 1093.

32.

Williams

R. M.

(1996). The costs of visits to emergency departments. New England Journal of Medicine, 334(10), 642-646.

33.

Wooldridge

J. M.

(2002). Econometric analysis of cross section and panel data. Boston: MIT Press.

34.

Zarkin

G. A.

Bray

J. W.

Babor

T. F.

Higgins-Biddle

J. C.

(2004). Alcohol drinking patterns and health care utilization in a managed care organization. Health Services Research, 39(3), 553-570.