Predicted probabilities of insomnia at baseline for physician-identified chronic conditions and selected patient-reported comorbidities, adjusted for sociodemographic variables, health habits, other conditions and comorbidities. Results are shown separately for patients without depressive symptoms (left) and those with depressive symptoms (right). Sleep status was missing for 177 patients, who were excluded from analysis. DM indicates diabetes mellitus; MI, myocardial infarction; CHF, congestive heart failure; ANG, angina pectoris; HIP, hip impairment; PRO, prostate problems; and OAD, obstructive airway disease.
Proportion of Medical Outcomes Study patients in each sleep category at 2-year follow-up, grouped by insomnia status at baseline. Sleep data were missing for 116 patients, who were excluded from analysis.
Katz DA, McHorney CA. Clinical Correlates of Insomnia in Patients With Chronic Illness. Arch Intern Med. 1998;158(10):1099-1107. doi:10.1001/archinte.158.10.1099
Copyright 1998 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.1998
Patients with chronic insomnia are more likely to develop affective disorders, cardiac morbidity, and other adverse health outcomes, yet many clinicians tend to trivialize the complaint of insomnia or to attribute it only to psychiatric causes.
To estimate the prevalence and longitudinal course of insomnia in patients with documented chronic medical illness and/or depression and to quantify the associations between specific chronic conditions and insomnia.
The presence of mild or severe insomnia was based on responses to a sleep questionnaire completed by 3445 patients with at least 1 of 5 physician-identified chronic conditions (hypertension, diabetes, congestive heart failure, myocardial infarction, or depression) at baseline; a subsample of 1814 patients completed follow-up questionnaires at 2 years. Using multivariate techniques, we evaluated the relationship between chronic conditions, patient-reported comorbidities, and insomnia (complaints of initiating and maintaining sleep), adjusting for sociodemographics and health habits.
Sixteen percent of study patients had severe and 34% had mild insomnia at baseline. At 2-year follow-up, 59% (95% confidence interval, 55%-63%) of patients with mild insomnia and 83% (95% confidence interval, 78%-88%) of patients with severe insomnia at baseline still had sleep problems. Odds ratios corresponding to mild and severe insomnia for key risk factors were as follows: current depressive disorder, 2.6 and 8.2; subthreshold depression, 2.2 and 3.4; congestive heart failure, 1.6 and 2.5; obstructive airway disease, 1.6 and 1.5; back problems, 1.4 and 1.5; hip impairment, 2.2 and 2.7; and prostate problems, 1.6 and 1.4. The majority of insomnia-comorbidity associations observed at baseline persisted at 2-year follow-up.
Patients with insomnia require follow-up, as the majority continue to be bothered by difficulty initiating and maintaining sleep. In addition to detecting affective disorders in patients with insomnia, clinicians should focus on medical conditions that disturb sleep, especially cardiopulmonary disease, painful musculoskeletal conditions, and prostate problems.
INSOMNIA AFFECTS approximately one third of the general population and has a major impact on health.1,2 In the Alameda County study, the number of hours of sleep, combined with data on 6 other health practices, was a significant predictor of subsequent health and mortality.3,4 Patients with chronic insomnia are more likely to develop affective disorders,5,6 and sleep disturbance is associated with the persistence of depressive symptoms in older persons with depression.7 In addition to its psychological consequences, insomnia may worsen somatic symptoms.8 Insomnia has also been associated with clinically significant decrements in key quality-of-life domains, such as physical functioning and mental health in generally healthy, working adults9 and in patients with chronic conditions.10 Finally, insomnia (and associated daytime sleepiness) is associated with increased risk of automobile and work-related accidents11,12 and increased use of medical services.13,14 Thus, there is great need for the development of effective prevention and treatment strategies for this disorder.
The prevalence of insomnia tends to be higher in clinical populations than in the general population. Numerous investigators have demonstrated increased prevalence of insomnia in subjects with somatic diseases2,9,15,16 and in subjects with depressive symptoms.2,5,14,17,18 Whereas a fair amount is known about sleep complaints in the general population, less is known about the epidemiology of insomnia and its longitudinal course among patients with chronic illness.19 In particular, previous studies have not examined the relationships between specific somatic conditions and insomnia in this population, while controlling for the presence of depressive symptoms.
To address these key issues, this study was designed (1) to estimate the prevalence and longitudinal course of insomnia in patients with documented chronic medical illness and/or depression and (2) to identify clinical risk factors associated with insomnia. The identification of risk factors and high-risk groups is an important strategy for the evaluation and management of chronic insomnia and can assist clinicians in targeting conditions that are most likely to promote or aggravate sleep problems.
The data for these analyses were derived from the Medical Outcomes Study (MOS), a 4-year observational study of health outcomes for patients with chronic medical and psychiatric conditions. Details on study design and objectives have been extensively reported20- 25 and are briefly summarized here. The MOS was conducted in 3 cities: Boston, Mass; Chicago, Ill; and Los Angeles, Calif. In each city, patients and physicians were sampled from 5 practice settings differing in organization, physician specialty mix, and payment arrangement. From these health care systems, 523 clinicians trained in family practice, general internal medicine, cardiology, endocrinology, psychiatry, and clinical psychology were sampled. Study participants were English-speaking adults who had had an office visit with an enrolled MOS clinician during 9-day screening periods in February to November 1986. Patients were asked to complete a brief, standardized, self-report questionnaire that gathered information on chronic disease, depressive symptoms, sociodemographic characteristics, and general health status.
Data from standardized physician-completed forms were used to identify patients with 5 medical tracer conditions (hypertension, diabetes, congestive heart failure, recent myocardial infarction, and depression).20- 25 A 2-stage process, involving a short form of the Center for Epidemiological Studies Depression Scale26 included in the patient questionnaire and the National Institute of Mental Health Diagnostic Interview Schedule,25 was used to identify patients with depression and to stage the severity.25,26 Current depressive disorder was defined on the basis of criteria from the Diagnostic and Statistical Manual of Mental Disorders, Third Edition,27 for lifetime major depression or dysthymia and having had an unremitted episode of major depression or period of dysthymia during the last 12 months; patients with depressive symptoms who did not satisfy this definition were considered to have subthreshold depression.28
Inclusion in the longitudinal panel component of the MOS was determined by a combination of criteria based on relative severity of the chronic medical and psychiatric tracer conditions and active participation in baseline data collection.20 Enrollment criteria for the longitudinal component of the study were met by 4842 patients. From among the 4842 eligible patients, 2349 were selected for longitudinal follow-up in the MOS. The sample for the cross-sectional analysis reported herein is 3445 panel-eligible patients with chronic medical and psychiatric conditions who completed a self-administered patient assessment questionnaire by mail, after enrollment into the study.29- 31 The sample for the longitudinal analyses reported herein is 1814 panel patients who completed both the baseline patient assessment questionnaire and the 2-year follow-up questionnaire.29
The MOS compiled 12 sleep items that represented a synthesis of previous measures and the addition of some new items.32 Based on the results of 2 pilot studies, the final sleep battery included items that assessed the following 5 sleep domains: initiation and maintenance of sleep, respiratory problems during sleep, quantity of sleep, perceived adequacy of sleep, and daytime somnolence. Sleep items were collected by self-administered mail survey; all items used a 6-point Likert scale, with responses ranging from "all of the time" to "never." For the purpose of this analysis, insomnia was defined as the complaint of difficulty initiating or maintaining sleep, a definition used in other epidemiological investigations of insomnia.16,33 Mild insomnia was defined by responses indicating difficulty in initiating or maintaining sleep during the preceding 4 weeks "some" or "a good bit" of the time; severe insomnia was defined by difficulty in initiating or maintaining sleep "most" or "all" of the time. The MOS sleep questions inquired into sleep during the preceding 4 weeks and thus were suitable for detecting chronic insomnia.34
In addition to the 5 MOS medical and psychiatric tracer conditions, 19 medical conditions comorbid to the tracer conditions were identified by means of data from the MOS health examination. The health examination (standardized medical history and clinical examination) was independently conducted by specially trained MOS medical staff.21,29,35,36 These comorbidities included anemia, coronary artery disease (ie, angina, history of myocardial infarction that occurred longer than 12 months ago), obstructive airway disease (ie, asthma, chronic obstructive pulmonary disease), gastrointestinal tract disorders (ie, inflammatory bowel disease, functional bowel disturbance, peptic ulcer), genitourinary tract disorders (ie, prostate disease, kidney or bladder infections within the preceding 6 months), joint disorders (ie, hip impairment, osteoarthritis, rheumatoid arthritis), acute back pain or sciatica (within the preceding 6 months), and general musculoskeletal complaints. Patients with angina who had had a myocardial infarction or congestive heart failure were grouped with the latter tracer conditions.
All data on sociodemographic variables (ie, age, sex, race, education, income, and marital status) and health habit variables (ie, smoking, alcohol use, exercise frequency, and body mass index) variables were obtained from questionnaire responses elicited on entry into the study. Alcohol and smoking status were assessed by a 3-point scale (eg, no history of use, past user, current user). Frequency of exercise was assessed with the question "How often do you exercise?" with the use of a 6-point Likert scale, with responses ranging from "daily or almost daily" to "almost never or never." Body mass index was computed from patient-reported weight and height.
We used logistic regression to identify chronic conditions and comorbidities associated with insomnia at baseline and at 2-year follow-up. Because we suspected that some conditions or comorbidities might be associated only with mild but not severe insomnia, our model included mild and severe insomnia as dependent variables (vs no insomnia). The baseline data allow for cross-sectional analysis of the relationship between clinical conditions and insomnia, whereas the follow-up data enable assessment of whether these relationships persist over time. The independent variables of interest were physician-identified conditions and patient-reported comorbidities, which were assessed at baseline. Of the latter, we selected comorbidities that were suspected of influencing sleep on the basis of previous knowledge 8,37 and that were significantly associated with insomnia at the P≤.25 level in bivariate analysis.38 Patients with chronic comorbidities of similar cause were grouped together (eg, asthma and chronic obstructive pulmonary disease). As all patients in the current study had 1 of the 5 physician-identified conditions, we used mild hypertension39 as the reference group in our multivariate analysis.
We adjusted for sociodemographic characteristics, health habits, and study location. To account for nonlinear effects, we included 3 dummy variables for age: 40 to 55, 56 to 65, and older than 65 years (age younger than 40 years was the holdout category). Similarly, we included 2 dummy variables for education: less than 12 years and exactly 12 years (greater than 12 years was the holdout category). Categories with sparse data and those judged to be similar, on the basis of logistic regression coefficients obtained from bivariate analyses, were collapsed. In addition, we checked 2-way interaction terms between depression (clinical and subthreshold) and chronic medical conditions to determine whether the effects of depression were different for those with vs those without chronic medical conditions.
We present the regression results in 2 ways. First, we computed odds ratios and corresponding 95% confidence intervals for each chronic condition and comorbidity group. Second, we generated predicted probabilities of mild and severe insomnia, adjusting for all covariates in the model, and averaged them for each chronic condition and comorbidity group. All regression analyses were performed with the MLOGIT procedure in STATA 4.0.40
The demographic and clinical characteristics of the MOS study sample that forms the basis of this analysis are shown in Table 1. Fourteen percent had 2 or more clinician-diagnosed medical conditions, and 37% had a concomitant depressive disorder (or subthreshold depression). On the basis of our working definition of insomnia, 16% of study patients had severe and 34% had mild insomnia at baseline. Of the remaining 50% of patients, a substantial fraction had symptoms of sleep disturbance but did not satisfy our definition of insomnia. Table 2 lends support to the validity of the definition of insomnia used in this study. Patients in the mild and severe categories reported sleeping 6.9 and 6.0 hours per night on average, compared with 7.3 hours per night in those without insomnia. Similarly, patients with increasing severity of insomnia were more likely to report daytime somnolence and taking longer than 30 minutes to fall asleep, and were less likely to report adequate sleep.
Bivariate analyses of sociodemographic variables showed that advanced age (ie, >65 years), female sex, nonwhite race, less than 12 years of education, being unmarried (ie, single, separated, or divorced), and low income (ie, <200% of poverty level) were highly associated with insomnia (P≤.001). Analysis of health habit variables showed that patients with mild or severe insomnia were more likely to be infrequent exercisers than were patients without insomnia (P=.05). The associations between insomnia and sex, race, and educational status remained significant in multivariate analysis; advanced age approached statistical significance after adjusting for chronic conditions (P=.07). No significant trends were observed for smoking status, alcohol use, or body mass index.
Multivariate regression analyses examining the association between physician-identified conditions and patient-reported comorbidities and insomnia at baseline are summarized in Table 3. The main results of the baseline analysis are as follows: (1) current depressive disorder and subthreshold depression were most strongly associated with insomnia; (2) among cardiopulmonary disorders, congestive heart failure and obstructive airway disease were significantly associated with both mild and severe insomnia; angina pectoris and myocardial infarction were associated only with mild insomnia; (3) among rheumatological disorders, back pain and hip impairment were associated with mild and severe insomnia at baseline; osteoarthritis was associated only with mild insomnia (and approached statistical significance for severe insomnia); (4) among gastrointestinal tract disorders, peptic ulcer disease was significantly associated with severe insomnia only; and (5) among genitourinary disorders, prostate problems were associated with mild insomnia. Two-way interaction terms between particular medical conditions and depression were not significant, suggesting that the effect of depression was similar across conditions.
To evaluate the magnitude of any observed associations between specific conditions and each grade of insomnia, we computed predicted probabilities of insomnia. Even in the absence of a depressive disorder, a substantial proportion of patients with specific medical comorbidities had mild or severe insomnia (Figure 1, left). Patients with depression, however, were more likely to have severe insomnia than were nondepressed patients (Figure 1, right).
At 2-year follow-up, 59% (95% confidence interval, 55%-63%) of patients with mild insomnia and 83% (95% confidence interval, 78%-88%) of patients with severe insomnia at baseline still had sleep problems, whereas a substantial proportion showed improvement (Figure 2). Among patients without insomnia at baseline, 21% and 3% developed mild and severe insomnia at 2-year follow-up, respectively. As shown in Table 4 (columns 2 and 3), the majority of associations observed at baseline persisted at 2-year follow-up; for some comorbidities (ie, obstructive airway disease and peptic ulcer disease), the odds ratios were similar to those at baseline but were no longer statistically significant. When we repeated the above analysis controlling for insomnia at baseline, however, the disease-insomnia associations that we observed in the baseline analysis were largely attenuated by the effects of previous sleep status (Table 4, columns 4 and 5).
The relationship between change in comorbid conditions and change in insomnia status is less clear than that between comorbid conditions and insomnia at baseline. Table 5 shows that a greater percentage of patients who developed hip impairment, osteoarthritis, and peptic ulcer disease reported new or worsened insomnia, compared with those who did not develop these conditions (columns 2 and 3). Similarly, a greater percentage of patients who had resolution of a given comorbid condition reported improvement of insomnia compared with patients who reported no resolution of that condition (columns 4 and 5), although none of these differences was statistically significant. For example, 42% of patients who reported resolution of depression, vs 32% of patients who reported persistent depression, experienced improvement of insomnia at 2-year follow-up.
Some attrition was observed in the longitudinal patient panel (n=2304). Patients in the panel who completed the 2-year follow-up questionnaire were more likely to be male, to be white, to have greater education and income, and to be more likely to be married and employed (Table 1). Clinically, patients in the longitudinal analytic sample were less likely to have congestive heart failure (8% vs 14%) or clinical depression (18% vs 25%). Thus, the sample on which longitudinal analyses are based is biased toward patients with higher socioeconomic status and with a lower burden of comorbidity. The effect of this bias on results reported in Table 4 would be to underestimate the impact of clinical risk factors associated with insomnia at 2-year follow-up.
Our findings extend the results of other investigators by demonstrating that several chronic medical conditions are associated with insomnia after controlling for depression. In addition, there was no significant association between age and insomnia after controlling for the presence of chronic conditions, as reported by other investigators.15,41 As in previous population-based surveys, female sex14,16,18 and low educational status9,18 were significantly associated with insomnia; in contrast to other studies,18 our analysis suggested that nonwhites were more likely than whites to report insomnia. About half of the patients in our study sample, which included patients with chronic medical and/or psychiatric conditions, reported frequent problems initiating or maintaining sleep; this proportion exceeds community-based prevalence estimates in the 15% to 35% range, using various definitions.5,42
It is noteworthy that the majority of patients with insomnia at baseline still had sleep problems at follow-up, which has been reported by others.33,43 For example, Hohagen et al43 showed that 75% and 52% of patients with severe insomnia at baseline had moderate to severe insomnia at 4 months and 2 years of follow-up. Moreover, previous insomnia has been shown to be an important predictor of current insomnia, which we confirmed in our longitudinal data.16 In patients with chronic conditions, the need for vigilance in detecting insomnia is demonstrated by the finding that 23% of patients developed new-onset mild or severe insomnia at 2-year follow-up; this compares with a 6% 1-year rate of incident insomnia in a population-based sample.5
Our analysis confirms the strong association between poor sleep and depression, which has also been documented in clinical studies19 and in population studies.2,5,7,41,44 Although affective disorders play a prominent role in patients with insomnia, we have demonstrated that multiple medical conditions are independently associated with insomnia, especially those characterized by cardiopulmonary symptoms, pain during the hours of sleep, and frequent nocturnal micturition.16,45 Because insomnia often stems from a complex mix of contributory causes, it is essential for clinicians to target both psychological and medical risk factors that are most strongly associated with insomnia and those that are potentially modifiable. We will describe the relationships between specific medical conditions and insomnia in more detail below.
Our finding that patients with cardiovascular disease (ie, angina, congestive heart failure, or myocardial infarction) are more likely to have insomnia can be explained by several pathophysiological mechanisms. Patients with angina may suddenly awaken with pain during sleep,46 often during rapid eye movement sleep, on account of fluctuations in cardiac rate47; in addition, an increased frequency of sleep-disturbed breathing has been reported in patients with angina pectoris.48 Severe congestive heart failure has been associated with sleep-disturbed breathing (eg, Cheyne-Stokes respiration, obstructive sleep apnea) and may cause disruption of sleep on account of paroxysmal nocturnal dyspnea. Alternatively, patients with insomnia tend to be predisposed to cardiovascular problems. Specifically, people who sleep less than 6 hours per night have an increased occurrence of ischemic chest pain.49
Our results confirm the importance of coexisting obstructive airway disease as a risk factor for insomnia. Others have shown that patients with chronic obstructive pulmonary disease are more likely to have sleep problems than population-based controls50; in addition, wheezing and increased sputum production are significant predictors of the complaint of difficulty initiating or maintaining sleep.51 In patients with chronic obstructive pulmonary disease, arousals during sleep may be increased because of elevated arterial carbon dioxide concentration; moreover, oxygen desaturation typically worsens during rapid eye movement sleep and during snoring periods.42,52 In persons with asthma, sleep efficiency tends to improve after symptoms are well controlled.53 However, medications used to treat obstructive airway disease (eg, inhaled β-agonists, theophylline, and systemic corticosteroids) may also worsen sleep patterns.
AMONG THE other disorders that we examined, painful musculoskeletal conditions (ie, osteoarthritis, hip impairment, back pain)9,15,16,54 and prostate problems were associated with insomnia. Elderly people with osteoarthritis have been shown to experience more superficial sleep than age- and sex-matched controls.55 We suspect that the negative effect of prostate disorders (eg, benign prostatic hypertrophy) on sleep was mediated by the occurrence of nocturia, which may cause frequent awakening and difficulty in getting back to sleep.45,56 Alternatively, others have suggested that a large proportion of the awakenings that patients attribute to nocturia were directly related to sleep apnea or other primary sleep disorders.57 In addition, peptic ulcer disease was associated with a nearly 2-fold increased risk of severe insomnia and may be a marker for concomitant gastroesophageal reflux disease, which increases arousals from sleep.42,54
Insomnia has been linked to a host of other conditions, including irritable bowel syndrome,58 chronic renal failure,59 severe liver disease, dermatitis, and neurological disease (eg, stroke, Parkinson disease, epilepsy)47 and psychiatric disorders (eg, schizophrenia, personality disorders, dementia),60 which were either not sufficiently prevalent in our study sample or not specifically captured by MOS health questionnaires. Although these latter conditions were not implicated in our analysis, clinicians should be alert for their presence in patients with insomnia.
The limitations of this study deserve comment. Sleep problems were assessed by self-report and were not confirmed with polysomnographic measurements. For the time frame of the MOS sleep questions (4 weeks), it is doubtful that polysomnography would have provided a true criterion standard, as patients with insomnia have high night-to-night variability in the quality of their sleep.61 Moreover, the magnitude of differences between persons with insomnia and controls is often unremarkable, and many persons with self-identified insomnia do not demonstrate objective sleep abnormalities on polysomnography.62 Although no attempt was made to validate our findings in a sleep laboratory, data obtained from the patients regarding other measures of sleep appear to verify the existence of sleep problems. It should be noted, however, that some measures of self-reported sleep (ie, amount of sleep, number of arousals) tend to be underestimated, and other measures (ie, amount of time required to get to sleep) tend to be overestimated by patients, in comparison with laboratory data.63,64
Another limitation is the absence of data on drug use and its timing in relationship to sleep complaints. Many of the drugs used for treatment of the chronic conditions in MOS participants are associated with sleep disturbance and thus could confound the relationship between specific diagnoses and insomnia. For example, antihypertensives (eg, β-blockers, clonidine, reserpine, diuretics), sympathomimetics (eg, terbutaline sulfate, albuterol sulfate), xanthine derivatives (theophylline, caffeine), and antidepressants (eg, serotonin-selective reuptake inhibitors) have all been implicated as causes of insomnia.54,65,66 Moreover, occult caffeine intake from over-the-counter products has been associated with increased difficulty falling asleep in the elderly.54 Another consideration is the presence of residual confounding with respect to smoking and alcohol drinking status, both of which have been associated with snoring67 and sleep disturbance.65,68 However, our results did not change when we used more detailed, 11-category versions of smoking and alcohol history variables in the MOS database.
In contrast to the significant associations between comorbid conditions and insomnia at baseline, the associations between change in comorbid conditions and change in insomnia status at 2-year follow-up were weak. This finding may be attributable to several factors. First, most of the questions on comorbidity asked whether the patient had experienced that condition at any time during the 2-year follow-up period, which did not necessarily coincide with the period to which the questions on sleep applied. Furthermore, the MOS was not explicitly designed to measure change in chronic disease, and the assessments were taken at too long an interval to prove ideal for the purpose at hand. Second, the power to detect significant associations was weak, as the number of patients who reported a change in each comorbid condition during follow-up was much smaller than the number reporting the presence of each condition at baseline. However, we observed trends that supported expected relationships (for all conditions except back pain): (1) patients with resolved comorbid conditions reported more improvement in insomnia than those without resolution and (2) patients who developed comorbid conditions reported new or worsened insomnia more often than those who did not develop the comorbid condition.
Interpretation of the above findings should account for the nature of the study sample. Because all patients in the study sample had chronic conditions, condition-specific odds ratios were computed relative to mild hypertension, which was selected as the reference group. It is noteworthy that persons with sleep apnea are more likely to be hypertensive69 and sleep complaints may be more common in hypertensive subjects than in nonhypertensive subjects15; however, others have not observed this latter association.9 Thus, it is likely that the associations between chronic conditions (other than hypertension) and insomnia in the current study would have been even stronger had the reference group been composed of patients without any chronic conditions. Finally, that the MOS only sampled patients with 5 chronic conditions who were insured and had a continuous relationship with a provider in 3 large urban areas limits generalizability.
Physician assessment of sleep history has been demonstrated to be poor compared with the assessment of other health behaviors,70 and there is a tendency for many clinicians to trivialize the complaint of insomnia.12 A possible explanation for this finding is that asking about sleep often discloses a panoply of potential contributory causes that many busy clinicians feel ill-prepared to handle. The consequences of ignoring insomnia can be serious, however, as patients with chronic insomnia are more likely to develop affective disorders,5 cardiac morbidity,49 and other adverse health outcomes.10
Our analysis demonstrates that several medical conditions have an independent association with insomnia, after controlling for depression. In addition to detecting affective disorders, we recommend that clinicians focus their evaluation of insomniac patients on target medical conditions that disturb sleep, especially cardiopulmonary disease, painful musculoskeletal conditions, and prostate disorders. Underlying conditions should be treated with medications not known to disturb sleep, or, if this is not possible, such medications should be administered at times that are likely to cause minimal sleep disruption. Finally, clinicians should be alert for sleep problems in patients with chronic conditions, as a sizable fraction develop new-onset insomnia. Future research should address whether more aggressive treatment of underlying diseases associated with insomnia leads to improved sleep and reduced need for hypnotic medications.
Accepted for publication November 17, 1997.
Presented at the Society of General Internal Medicine Annual Meeting, Washington, DC, May 3, 1996.
We thank Ruth Benca, MD, PhD, and Khin Mae Hla, MD, MHS, for their critical review of the manuscript.
Reprints: David A. Katz, MD, MSc, Section of General Internal Medicine, Suite 100, 2870 University Ave, Madison, WI 53705.