View the most recent version.

Findings

In recent years, diabetes-related mortality has increased, an increase that has been linked to an upturn in the prevalence of obesity.^1,2  Diabetes is currently the sixth leading cause of death in Canada.³  In 2005, approximately 1.3 million Canadians aged 12 or older (5% of the population in that age range) reported that they had been diagnosed with the disease.⁴  The complications of diabetes can attack every major organ. Because of its wide-ranging impact on the health of individuals and the economic burden it places on the health care system,^5,6 diabetes is recognized as a major public health problem. Yet, to some extent, diabetes is also “ambulatory care sensitive.”  That is, it can be managed with appropriate care in the community.⁷ But if not well controlled, diabetes can result in multiple co-morbidities that may require extensive care,⁸  including hospitalization.^8-10

With the rise of health expenditures,¹¹ emphasis is being placed on the effectiveness of services.  A conceptual framework, developed jointly by Statistics Canada and the Canadian Institute for Health Information,^12,13 highlights the relationship between patient outcomes and three sets of possible determinants―health care system, care path and patient characteristics―in the context of place and environment. Through a linkage of survey and hospitalization data, the current study applies this framework to diabetes. 

Specifically, this analysis examines risk factors for acute care hospitalization in a cohort with type 2 diabetes (T2DM), the predominant form of the disease, accounting for 95% of cases.  Acute care hospitalization (for any reason) is used as a proxy for negative outcome.  Associations between hospitalization and patients’ characteristics, care path factors, and health system characteristics are analysed.

With regard to care path, relatively little is known about the roles of general practitioners (GPs) and specialists in the risk of hospitalization of people with diabetes.  Previous research suggests that the more aggressive care offered by specialists^14-18 may simply reflect more advanced disease.

With regard to characteristics of the health care system, the use of services has been shown to vary by region.^19-22 It is possible, then, that the likelihood of hospital admission may, in part, reflect where an individual lives; all else being equal, residents of “high-use” areas may be more likely to be hospitalized. 

Of course, the role of risk factors like smoking, drinking, physical activity and diet must also be taken into account.^23,24

This article assesses a range of factors associated with hospitalization of individuals with type 2 diabetes, using linked survey and administrative data.

Methods

The dataset used in this analysis links the 2000/2001 Canadian Community Health Survey (CCHS) to the Hospital Morbidity Data Base (HMDB).  The CCHS collects information about the health and well-being of the household population.  For cycle 1.1 (2000/2001), about 130,000 Canadians aged 12 or older were interviewed.²⁵  In addition to questions about socio-demographic characteristics, risk factors, and health care services use, they were asked about diabetes and related treatment.  The HMDB is a national administrative database of acute inpatient hospital stays from 1992/1993 to 2003/2004.  It can be used to analyse trends in causes of hospitalization,^26-28 but it lacks detailed patient information such as socio-economic background and risk factors.  However, linkage of the CCHS and the HMDB yields a dataset with information about hospital patients’ socio-economic status and risk factors.²⁹ 

The linked dataset used in this analysis consisted of a preliminary cohort of 6,361 CCHS 1.1 respondents who self-reported diabetes.  Of these, 1,003 Quebec residents were excluded because Statistics Canada did not have health insurance numbers with which to link them to the HMDB.  An additional 467 non-Quebec respondents who refused permission to link their survey results to health services administrative data were excluded, along with 37 who reported only gestational diabetes.  An algorithm³⁰ was used to exclude another 243 respondents identified as having type 1 diabetes.³¹  The final T2DM cohort consisted of 4,611 CCHS respondents.  Statistics Canada’s Policy Committee approved the data linkage .

A hospital episode is defined as any record of acute hospital discharge obtained from the HMDB.  Hospital episodes that occurred within one year before and two years after the survey date were included in the analyses. The primary outcome of interest was hospitalization during the two years after the CCHS interview, excluding hospital stays related to pregnancy/birth.  All other acute hospitalizations, defined as all-cause hospitalization, were included in the study,

Univariate and bivariate statistics were calculated to describe the data.  With bootstrap methods^32,33 and special linked weights, the descriptive statistics were adjusted for the complex survey design and to account for non-response when permission to link survey data to hospital records had been denied.  Predictors of hospitalization were identified through multivariate logistic regression.  Normalized weights were used to ensure proportional representation of the provincial, age and gender distributions in the sample with diabetes, and to report the 95% confidence intervals and significance levels (0.01 and 0.05). 

Most variables were derived directly from the CCHS or the HMDB (Table 1).  For example, among the CCHS derived variables, “impact of health problem” is a measure of the effect of long-term physical and mental conditions on  home, work or school, and other activities:  often, sometimes or never.  GP and specialist consultations were based on separate questions about the number of contacts with doctors in the past 12 months.

Table 1
Canadian Community Health Survey (CCHS) and Hospital Morbidity Database (HMDB) variables used in analyses of hospitalization of CCHS respondents with type 2 diabetes

Two additional variables were derived from the linked CCHS/HMDB file.  The first attempts to capture regional hospital utilization patterns as an indicator of regional variations in the use of health care services.  This variable, based on the full CCHS sample linked to all acute hospitalizations (excluding pregnancy/delivery) in the two years after the survey is the ratio of the observed number of hospitalizations to the expected number in each health region, controlling for characteristics of the region’s population that represent need:  age, sex, functional health status as measured by the Health Utility Index Mark 3 (HUI3), self-reported health compared with a year earlier, prior hospitalization, co-morbidities, smoking status, alcohol use and physical activity.  The second derived variable―prior hospitalization―is a marker for disease severity indicating if the respondent had been admitted to hospital in the year before the CCHS interview.  

The role of risk factors, care path and health system characteristics on all-cause hospitalization of the T2DM cohort was examined while controlling for age, sex and household income.^34,35 

Results

The characteristics of the T2DM cohort differed substantially from those of other Canadians (Table 2).  The T2DM cohort was much older, more likely to be male, and tended to have a lower household income.  For instance, almost half (45%) the cohort were aged 65 or older, compared with 16% of the rest of the population.  The cohort was more likely to have chronic conditions other than diabetes and to have been hospitalized in the past year.  They were more likely than other Canadians to be overweight or obese and physically inactive, but less likely to be current smokers  or regular drinkers.  Higher percentages of T2DM had consulted a GP or specialist in the year before their CCHS interview.

Table 2
Selected characteristics of type 2 diabetes cohort compared with other Canadians, household population aged 12 or older, Canada excluding Quebec and territories, 2000/2001

Given their generally less favourable health status, it is not surprising that in the two years after their CCHS interview, almost a quarter (24%) of the T2DM cohort were admitted to hospital (Table 3).  But not all members of the cohort were equally likely to have been hospitalized. 

Table 3
Percentage hospitalized within two years of 2000/2001 Canadian Community Health Survey interview, by selected characteristics, household population aged 12 or older with type 2 diabetes, Canada excluding Quebec and territories

As might be expected, hospitalization rates were higher among those who were older, lived in lower-income households, had co-morbidities, or reported having been hospitalized in the year before the CCHS interview.  Around a third (34%) of cohort members aged 65 or older were hospitalized, compared with 12% of those aged 14 to 44.  Similarly, about  third of the T2DM cohort who lived in lower-income households were hospitalized versus 16% of those in the highest household income quintile.  And fully 54% of the T2DM cohort who had been hospitalized in the year before the CCHS interview were admitted to hospital in the two years after the interview, compared with 19% who had no prior hospitalization.  T2DM cohort members who never smoked were less likely than former or current smokers (19%, 27% and 25%, respectively) to be hospitalized.  Regular drinkers had lower hospitalization rates than did occasional or non-drinkers.  In terms of care path, equal percentages of those who did and did not have a recent consultation with GP were hospitalized (24%).  However, 30% who had consulted a specialist were hospitalized, compared with 19% who had not done so. 

Of course, many characteristics associated with high hospitalization rates are related to each other.  For instance, older people with diabetes are more likely than their younger counterparts to have other chronic conditions and to have had a prior hospitalization.  Smoking tends to be more prevalent among low- than high-income groups.  When the potentially confounding effects of other variables were taken into account, the factors significantly associated with hosptialization among the T2DM cohort were:  older age, male, lower reported health utility, presence of other chronic condition(s), impact of health problems, physical inactivity, smoking, alcohol consumption, insulin use, doctor consultations, and system response (high- or low-hospitalization region) (Table 4).

Table 4
Adjusted odds ratios relating selected characteristics to hospitalization within two years of 2000/2001 Canadian Community Health Survey, household population aged 12 or older with type 2 diabetes, Canada excluding Quebec and territories

The strongest predictor of hospitalization in the two-year follow-up period was prior hospitalization (OR=3.0, 95% CI: 2.5 to 3.7). T2DM cohort members who had contacted a specialist in the year before their CCHS interview were more likely to be admitted, compared with those who had not (OR=1.4, 95% CI: 1.2 to 1.6).  By contrast, those who had contacted a GP were less likely to be admitted to hospital in the next two years (OR=0.7, 95% CI 0.5-0.9).  Cohort members who lived in health regions with generally higher hospitalization rates had significantly higher odds of hospitalization in the next two years (OR=2.6, 95% CI: 1.8 to 3.7).

Discussion

Linkage of the CCHS and the HMDB made it possible to identify a number of factors significantly related to all-cause hospitalization of people with type 2 diabetes.  Because diabetes is, to a considerable extent, an ambulatory care sensitive condition, it was assumed that hospitalization is an indirect indicator of poor outcome. 

Age was obviously important.  As well, females with diabetes had a lower risk of hospitalization than did males. Smoking, whether former and current, was a strong predictor of hospitalization; regular alcohol consumption had a protective effect.³⁶  As expected, having other chronic disease(s), prior hospitalization, and the impact of long-term physical and mental conditions on daily life were strong predictors of hospitalization. 

T2DM cohort members who had consulted a specialist in the 12 months before their CCHS interview had a significantly higher risk of hospitalization over the next two years.  Of course, the specialist consultation did not “cause” the hospitalization; rather, consulting a specialist was likely a reflection of disease severity. 

At the health care system level, T2DM cohort members in high-hospital-use health regions had signficantly high odds of hospital admission.  

Limitations

This analysis has several limitations, foremost among them, in the case of the CCHS, reliance on self-reports.  An earlier study found that only about 75% of people with physician-diagnosed diabetes self-reported the condition to the CCHS, and their characteristics differ from those of people who do report diabetes.³⁷ 

The analyses are limited to acute care hospitalizations.  Information is not provided about the use of emergency rooms, where diabetes-related events such as hyperglycemia are often treated.  As a result, the full extent of diabetes patients’ use of hospitals is not represented. 

Because the CCHS does not include residents of institutions, the linkage with hospital data is necessarily confined to the household population.  An evaluation study of the linked data showed a high undercoverage rate among people aged 75 or older, many of whom live in institutions.³⁸  Consequently, the data presented here likely underestimate the strength of the relationship between diabetes and hospitalization.  As well, Quebec residents were excluded from the analyses. 

The analysis would have been stronger had it been possible to include the nature of the care respondents were receiving as a potential factor in their odds of hospitalization.³⁹  However, these data were not collected by the 2000/2001 CCHS.   Such information (for example, use of haemoglobin A1C testing, foot care, eye exam) was collected in 2005 by the CCHS 3.1, but the data needed to examine subsequent hospital use are not yet available.

Sample size is an issue.  Although the CCHS sample was constructed to allow the reporting of various conditions at the health region level, the study pertains to a relatively small group―people with type 2 diabetes―and a low-probability outcome―hospitalization.  This combination makes it impossible to conduct analyses even at the provincial level, let alone the health region level.  To overcome the problem of small sample size, combining surveys may be an option in the future.

Clinical variables related to hospitalization, such as physiologic characteristics, diagnoses and treatments, could not be considered in this study.  

Conclusion

When the effects of demographic, socio-economic and health status characteristics were taken into account, physical inactivity and former or current smoking were significantly associated with an increased likelihood of all-cause hospitalization of people with type 2 diabetes.  Specialist visits were positively related to hospitalization, but the relationship with general practitioner visits was negative.  However, the fact of having seen a specialist is unlikely to be a risk factor for hospitalization, but rather, a marker for disease severity.  Regional hospital use patterns were also significantly associated with all-cause hospitalization.  Whether these factors would remain important if the focus was limited to diabetes-specific hospitalization can be a topic for future analyses.