Risk Factors for Hospitalization in People with Diabetes
Scot E. Moss, MA; Ronald Klein, MD, MPH; Barbara E. K. Klein, MD, MPH Objective To determine factors predicting hospitalization in people with diabetes. Methods Two population-based groups with diabetes were examined at baseline (1980-1982), 4 years (1984-1986), and 10 years (1990-1992). The younger-onset group (n=777) consisted of all persons diagnosed as having diabetes before age 30 years who were taking insulin, and the older-onset group (n=542) consisted of a sample of persons diagnosed after age 30 years. At the 10-year examination, participants were asked if they had been hospitalized in the previous year. Factors from the 4-year examination were examined for their ability to predict hospitalization at the 10-year examination. Results In the younger-onset group, 25.5% reported being hospitalized. In logistic models, glycosylated hemoglobin level (odds ratio [OR], 1.27; 95% confidence interval [CI], 1.16-1.39 for a 1% increment) and hypertension (OR, 1.60; 95% CI, 1.08-2.38) predicted hospitalization. Factors that were not significant included age, sex, systolic and diastolic blood pressures, body mass, smoking status, and alcohol consumption. In the older-onset group, 30.8% reported being hospitalized. In logistic models, only glycosylated hemoglobin level (OR, 1.16; 95% CI, 1.06-1.29 for a 1% increment) predicted hospitalization. Conclusions Glycemic control is subject to intervention. Better control may decrease hospitalization among people with diabetes. Thus, there is considerable potential for reducing health care costs. Arch Intern Med. 1999;159:2053-2057
PEOPLE WITH diabetes are hospitalized much more frequently than those without.1-3 Direct costs attributed to hospitalization of patients with diabetes were estimated to be $27 billion in 1997.4 Thus, even a modest reduction in hospitalizations among people with diabetes could result in major savings. If people at high risk for hospitalization could be identified along with risk factors that are subject to manipulation, it may be possible to reduce this burden. The purposes of this article are to examine the frequency of hospitalization in a diabetic population and to investigate which characteristics of this population are risk factors.
The Wisconsin Epidemiologic Study of Diabetic Retinopathy cohort was identified from a review of records of primary care physicians in southwest Wisconsin. Case identification methods and descriptions of the population appear in previous reports.5-10 Briefly, 10,135 patients with diabetes were identified from the records of primary care physicians in an 11-county area of southwest Wisconsin. A sample of 2990 of these persons was invited to participate in the baseline examination from 1980 to 1982 and was divided into 2 groups. The first group was composed of all subjects younger than 30 years at diagnosis and taking insulin (n=1210), referred to as the younger-onset group. The second group was a probability sample, stratified by duration of diabetes, of subjects who were older than 30 years at diagnosis (n=1780), referred to as the older-onset group. In the younger-onset group, 996 patients (82.3%) participated in the baseline examination; in the older-onset group, 1370 patients (77.0%) participated, 674 of whom were taking insulin and 696 of whom were not. The surviving persons were invited to participate in a 4-year follow-up examination in 1984 to 1986 and a 10-year follow-up examination in 1990 to 1992. These intervals were chosen to observe the incidence of an adequate number of complications for investigating relationships with risk factors. Both the baseline and follow-up examinations followed a similar protocol. This consisted of obtaining informed consent, measuring blood pressure, measuring refractive error and best-corrected visual acuity, performing a slitlamp examination for chamber depth and iris neovascularization, measuring intraocular pressure, dilating the pupils, administering a medical history questionnaire, examining the lenses for cataracts, performing an ophthalmoscopic examination, taking stereoscopic color fundus photographs of 7 standard fields, determining urine protein, and obtaining a venous blood sample for determining blood glucose and glycosylated hemoglobin levels. Hospitalization among the study participants was not determined at the baseline examination. At the follow-up examinations, hospitalization among the subjects was ascertained by history by asking subjects, "In the last year, how many times have you been hospitalized overnight or longer?" The response was recorded, and if the subject indicated a hospitalization in the previous year, he or she was asked questions pertaining to common reasons for hospitalization. The 1984-1986 examination occurred during a period of extensive change in hospital utilization. Therefore, to present results obtained in an environment more resembling the current state, we report results for the more recent follow-up examination in 1990-1992. Risk factors examined were those determined at the 1984-1986 examination to study their temporal relation to subsequent hospitalization. Current age was defined as age at the time of the 1984-1986 examination. Duration of diabetes was the time interval between the diagnosis of diabetes and the 1984-1986 examination. Glycosylated hemoglobin was measured by a microcolumn method.11 Systolic and diastolic blood pressures were the mean of 2 blood pressure determinations. Hypertension was defined as systolic blood pressure greater than or equal to 140 mm Hg, diastolic blood pressure greater than or equal to 90 mm Hg, or a history of hypertension with use of antihypertension medications. Body mass was defined as weight (in kilograms) divided by the square of height (in meters). A person was defined as a nonsmoker if smoked fewer than 100 cigarettes in his or her lifetime, an ex-smoker if smoked more than this amount but had stopped smoking before the 4-year follow-up examination, or a current smoker if had not stopped. Alcohol consumption was obtained from questions concerning average weekly amounts of beer, wine, and distilled alcoholic beverages consumed during the previous year. This information was summarized in terms of average daily consumption of absolute alcohol. A person was defined as a nondrinker if consumed no alcoholic beverages in the previous year, a light drinker if consumed up to 5.0 g daily, a moderate drinker if consumed 5.1 to 23.5 g daily, and a heavier drinker if consumed more than 23.5 g daily. Analysis of cigarette smoking and alcohol consumption was restricted to subjects 18 years or older. Physical activity was based on the number of times in a week a subject engaged in an activity long enough to work up a sweat. An active person regularly worked up a sweat 3 or more times in a week. Otherwise, the person was considered to be sedentary. Statistical analysis was performed with SAS.12, 13 Tests for trends in proportions were performed by Mantel-Haenszel procedures.14 Multivariable analysis of risk factors for hospitalization was accomplished with logistic regression.15 The end point was hospitalization in the year before the 1990-1992 follow-up examination. All values of risk factors were obtained at the 1984-1986 examination and so are antecedent to the end point.
Of 996 younger-onset participants at baseline, 765 (76.8%) were examined at the 4- and 10-year follow-up examinations and an additional 15 (1.5%) were interviewed. Information concerning hospitalization was available for 777 of these subjects. Of the 216 subjects who did not participate, 62 refused, 4 could not be located, and 150 had died. Of 1370 older-onset participants at baseline, 533 (38.9%) were examined at the 4- and 10-year follow-up examinations and 17 (1.2%) were interviewed. Information concerning hospitalization was available for 542 of these subjects. Of the 820 subjects who did not participate, 46 refused, 2 could not be located, and 772 had died. Comparisons between participants and nonparticipants at follow-up have been presented elsewhere.8-10 Briefly, younger-onset nonparticipants who had died were older, had diabetes longer, and had higher systolic and diastolic blood pressures. Older-onset nonparticipants who had died were older, had diabetes longer, had higher glycosylated hemoglobin levels, and had higher systolic and lower diastolic blood pressures. In both groups, living nonparticipants did not differ significantly from participants in these variables.10 Baseline characteristics of the 2 cohorts are presented in Table 1. In addition, 49% of the younger-onset group and 42% of the older-onset group were male, and 53% of the older-onset group were taking insulin. In the younger-onset group, 25.5% (n=198) reported having been hospitalized in the year before the 10-year examination. In the older-onset group, 30.8% (n=167) had been hospitalized, which is significantly higher (P = .03) than the younger-onset group. These percentages had changed little from the 4-year examination at which time they were 25.8% in younger-onset persons and 31.5% in older-onset persons. Table 2 presents the self-reported reasons for hospitalization in the 2 groups. The most common reason was glycemic control (6.3%) and heart disease (7.8%) in the younger- and older-onset groups, respectively. The reasons for hospitalization presented in Table 2 are neither mutually exclusive nor exhaustive. Among people who had been hospitalized in the previous year, 69% of younger-onset and 63% of older-onset subjects reported only one hospitalization. Therefore, for analysis, the incidence of at least one hospitalization in the previous year was examined. Figure 1 presents the frequency of hospitalization by age. There was little association between age and hospitalization in either cohort. Hospitalization increased modestly with longer duration of diabetes in both cohorts, although the relationship was significant only in the older-onset group (Figure 2). Table 3 presents the frequency of hospitalization by other characteristics in the younger cohort. There were significantly more hospitalizations of people with higher glycosylated hemoglobin levels and those with hypertension. Although subjects living a more active lifestyle were less likely to report being hospitalized, the result was not statistically significant. A similar univariate analysis for the older-onset group is shown in Table 4. In this group, people with higher glycosylated hemoglobin levels, higher systolic blood pressure, taking insulin, and who did not consume alcohol were more likely to be hospitalized. Because many of the factors presented in Table 3 and Table 4 are interrelated, multivariable logistic models were developed to determine which ones are significantly and independently associated with hospitalization. In the younger-onset cohort, persons with higher glycosylated hemoglobin levels (odds ratio [OR], 1.27; 95% confidence interval [CI], 1.16-1.39 for each 1% increment) and hypertension (OR, 1.60; 95% CI, 1.08-2.38) had higher odds of being hospitalized. A sedentary lifestyle (OR, 1.41; 95% CI, 0.99-2.01), although associated with hospitalization, did not reach statistical significance. In a model restricted to younger-onset subjects 18 years or older, smoking status and alcohol consumption were not related to hospitalization. In the older-onset group, only higher glycosylated hemoglobin level (OR, 1.16; 95% CI, 1.06-1.29 for each 1% increment) was associated with hospitalization. Combining the younger- and older-onset groups in one model with glycosylated hemoglobin and an indicator variable for group as risk factors shows both variables to be statistically significant (P<.005). Furthermore, the difference between the 2 groups is greater than indicated by the univariate analysis because of confounding by glycosylated hemoglobin level, which is higher in younger-onset persons. The addition of an interaction term between glycosylated hemoglobin and group was not statistically significant (P=.27), indicating that the effect of glycosylated hemoglobin did not differ between cohorts.
Diabetes is a disease associated with a host of complications, many of which may lead to frequent or extended hospitalization. Estimates from the National Health Interview Survey of 1991 are that 7.8% of the general population was hospitalized in the year before the interview.16 This compares to 25.5% and 30.8% in our younger-onset and older-onset groups, respectively. Thus, people with diabetes appear to be hospitalized more than 3 times more frequently than the general population. This agrees closely with results from the 1989 National Health Interview Survey1 that found a 3-fold increase, the study by Aro et al2 that reported a 4-fold increase in hospitalization among people with diabetes, and the study by Rosenthal et al3 that found a 2.2 times increase in hospitalization among older adults with diabetes. The direct costs attached to these hospitalizations are also high, estimated by a recent study4 to be $27 billion in 1997. Therefore, there is considerable potential for substantial savings in health care costs if the frequency of complications could be decreased or their appearance forestalled. Identification of modifiable risk factors is the first step toward prevention of hospitalization and reduction of attendant costs. However, to date, there have been few reports concerning risk factors for hospitalization. Charron-Prochownik et al17 and Kovacs et al18 were concerned with children 8 to 13 years of age, and Palta et al19 was concerned with children and young adults in the period after diagnosis of diabetes. All these investigators found a higher glycosylated hemoglobin level to be a risk factor for readmission to the hospital only in the early years after diagnosis. Rosenthal et al3 reported a higher glycosylated hemoglobin level to be a risk factor for hospitalization in a predominantly male cohort of older adults attending a geriatric diabetes clinic. Our cohorts are more representative of a general population of people with diabetes, in that they are beyond the initial postdiagnosis period when glycemic control is the most important cause of hospitalization and are approaching the time when other complications begin to exert an effect. From our results, it appears that metabolic control is also a significant risk factor for hospitalization later in the course of diabetes, in both younger- and older-onset persons. These results are consistent with those of the Diabetes Control and Complications Trial20 in persons with type 1 diabetes mellitus and the United Kingdom Prospective Diabetes Study21 in persons with type 2 diabetes mellitus. These studies showed the efficacy of glycemic control in reducing many microvascular complications of diabetes. What is noteworthy in the current study is the association of glycemia with any hospitalization, including many for reasons other than those usually associated with diabetes. These results have important implications for prevention. For instance, based on our models, reducing glycosylated hemoglobin level by just 1% could reduce hospitalization by 14% to 20% and save as much as $4 billion to $5 billion in direct costs alone. However, the manipulation of risk factors is not without its own hazards and costs. For example, caution is necessary when instituting intensive glycemic control because of the dangers of severe hypoglycemia. In the Diabetes Control and Complications Trial, insulin reactions in the intensive treatment group occurred 3 times higher than the standard treatment group, and the rate of hospitalizations was 1.5 times higher.20 Thus, severe hypoglycemic episodes could reduce the savings obtained, as could the expense of maintaining intensive glycemic control. Hypertension in younger-onset persons and systolic blood pressure in older-onset persons were also associated with hospitalization. We found no previous reports of these associations in diabetic populations. However, since they are associated in nondiabetic populations with numerous adverse health events such as cardiovascular disease, renal disease, and stroke that could result in hospitalization, it is not unexpected to encounter them here. These results suggest possible approaches for the development of preventive strategies. Indeed, a recent report from the United Kingdom Prospective Diabetes Study22 supports the role of blood pressure control in the prevention of diabetes complications. There are several limitations to our study. First, follow-up of the cohorts was incomplete, largely due to mortality. Since subjects who died could be expected to have been hospitalized more, the estimates of hospitalization in the surviving cohort may be understated. However, the cohort is also aging, which would be expected to increase hospitalization. The extent to which these processes balance is unknown. Second, our estimates of hospitalization rely on self-reports of the subjects. This may also bias the results toward underestimates. For this reason, subjects were queried only about hospitalizations in the previous year, similar to the National Health Interview Survey.16 The similarity of rates between the 1984-1986 and 1990-1992 examinations also gives us confidence that this approach produced useful information. Finally, some investigators may question the appropriateness of examining the associations of hospitalization with characteristics measured 6 years earlier. This may be a concern if the hospitalizations were only for acute conditions. However, many conditions for which people with diabetes are hospitalized are chronic and likely were initiated years earlier. Thus, the time chosen in our study is appropriate for understanding the temporal relation of risk factors to hospitalization. In summary, hospitalization among people with diabetes is substantially greater than in the general population. This adds a considerable cost to the health care system. Glycemic control was the most important factor related to hospitalization. Thus, there is potential for reducing hospitalization and great savings in medical and financial expenditure. However, these savings may be offset by the costs necessary to maintain the modifications and by increases in hospitalization for hypoglycemia.
From the Department of Ophthalmology and Visual Sciences, University of Wisconsin Medical School, Madison.
Corresponding author: Scot E. Moss, Department of Ophthalmology and Visual Sciences, University of Wisconsin-Madison, 610 N Walnut St, 460 WARF, Madison, WI 53705-2397 (e-mail: email@example.com). Accepted for publication January 25, 1999. This research is supported by grant EYO-3083 (Drs R. and B. E. K. Klein) from the National Institutes of Health, Bethesda, Md, and in part by the Research to Prevent Blindness, New York, NY (Dr R. Klein, Senior Scientific Investigator Award).
The authors have no proprietary interests.