Quantifying the Risk of Infectious Diseases for People With Diabetes

The universal health care system in Ontario, Canada, allows a near-complete collection of data from all hospitalizations of and physician claims for Ontario residents (population 11,410,046 in 2001). The diabetic cohort was defined as the entire population of Ontario diagnosed with diabetes before 1 April 1999, using a validated algorithm (8). The nondiabetic cohort was chosen from the remaining Ontario population alive on that date, matched 1:1 for date of birth within 30 days, sex, region, and income quintile by ecological attribution of neighborhood-level census data (9,10).

A database of hospital separation abstracts prepared by the Canadian Institute of Health Information was searched for any records between 1 April 1999 and 31 March 2000 bearing a primary or complication diagnosis code for an infectious disease. A database of all claims by Ontario physicians to the health insurance program was searched for any claims between those dates for an infectious disease. The outcome of interest was at least one hospital separation or physician claim for an infectious disease. Multiple events in the same patient were not counted. The cohorts were compared using the χ² test, and a risk ratio with 99% CI was calculated. The analysis was repeated evaluating only hospital separations. Mortality attributable to infectious disease was compared between the diabetic and nondiabetic cohorts. Either of two events defined attributable mortality for an individual: death during an admission with an infectious disease diagnosis code or death within 5 days of a physician claim for an infectious disease. The risk ratio with 99% CI for attributable mortality between cohorts was determined.

In a secondary analysis, all individual infectious diseases with frequencies >50 per 100,000 people were evaluated. For infections requiring in-hospital treatment, only hospital separations were counted, whereas for those that could be managed as an outpatient, both hospital separations and physician claims were counted. Risk ratios and 99% CIs were calculated for each infectious disease.

To assess the stability of any measured effects, the analyses were repeated using diabetic and nondiabetic cohorts defined on 1 April 1996, evaluating discharges and claims between 1 April 1996 and 31 March 1997.

There were 516,494 people with diabetes in Ontario on 1 April 1999. Because of failure to match to a nondiabetic control, 0.5% were excluded. The remaining 513,749 people with diabetes were matched to an equal number of people without diabetes. In both groups, 51.7% of people were male, with a mean ± SD age of 61.0 ± 16.0 years. From the 1996 data, 0.7% of the 404,412 people with diabetes could not be matched, so each cohort included 401,661 people, of whom 51.8% were male with a mean ± SD age of 60.9 ± 16.0 years.

In the 1999 cohort, 46.0% of diabetic people had a hospitalization or physician claim for an infectious disease in the year evaluated. Only 38.0% of people in the matched nondiabetic population had such a hospitalization or physician claim. Therefore, the risk ratio for an infectious disease hospitalization or physician claim for the diabetic cohort versus the nondiabetic cohort was 1.21 (99% CI 1.20–1.22, P < 0.0001) (Table 1). In the 1996 cohort, the frequencies were 47.5 and 39.2%, respectively, giving a risk ratio of 1.21 (1.20–1.22, P < 0.0001). The risk ratios for the primary outcome were similar between men and women and across all income quintiles. Although there were small differences by age, no consistent pattern emerged.

The majority of these events were physician claims. In the 1999 cohort, 5.2% of the diabetic population and 2.6% of the nondiabetic population had hospital separations with infectious diseases among the discharge diagnoses. Therefore, the risk ratio for inpatient infections was 2.01 (99% CI 1.96–2.06, P < 0.0001). In the 1996 cohort, hospitalization rates were 5.4 and 2.5%, respectively, giving a risk ratio of 2.17 (2.10–2.23, P < 0.0001). There was a striking gradient by age; in 1999, the frequency of hospital separation with an infectious disease increased from 3.4% for people <30 years of age to 12.1% for those aged ≥80 years, whereas the risk ratio decreased from 4.33 (3.42–5.48, P < 0.0001) for people aged <30 years to 1.54 (1.47–1.62, P < 0.0001) for those aged ≥80 years. The risk ratio was slightly higher for women (2.12, 2.04–2.20; P < 0.0001) than men (1.91, 1.84–1.98; P < 0.0001). There was no consistent association between risk ratio and income quintile.

In 1999, 1.0% of the diabetic cohort had a death attributable to an infectious disease. The rate in the nondiabetic cohort was 0.6%, resulting in a risk ratio of 1.84 (99% CI 1.73–1.95, P < 0.0001) (Table 2). In 1996, the risk ratio was 1.92 (1.79–2.05, P < 0.0001). The risk ratio for mortality within 5 days of a physician claim for an infectious disease among those who had such claims was elevated comparing the diabetic and nondiabetic cohorts. However, the risk ratio for mortality during hospitalization with an infectious disease among those who had such hospitalizations was close to unity, occurring in ∼15% of patients hospitalized for infectious diseases.

The risk ratios for the individual infectious diseases evaluated in the secondary analyses are shown in Table 1, indicating that the risk of many individual infections was increased with diabetes. These findings were also consistent in both years analyzed.

Nearly half of all people with diabetes had at least one hospitalization or physician claim for an infectious disease in each year studied. Furthermore, the presence of diabetes compared with matched control subjects was associated with an absolute risk increase of ∼8% and a risk ratio of 1.21 for hospitalization or physician claim, confirming and quantifying a commonly held but previously unproven clinical belief. The risk was stable in both time periods studied. The risk ratio for hospitalization with an infectious disease diagnosis was even higher, up to 2.17, with a particularly high risk ratio for young people.

Confirming the findings of previous researchers who examined specific causes of death from death certificate data (6), we found that mortality attributable to infectious diseases for patients with diabetes exceeded that of those without diabetes. Death within 5 days of a physician claim for infection was elevated. However, mortality during hospitalization for infectious disease was not. Since the diabetic and nondiabetic cohorts were not matched for severity of disease or comorbidity, and since more diabetic patients than nondiabetic patients had hospitalizations with infectious diseases, these findings suggest that physicians have a lower threshold for admitting a diabetic patient with an infectious disease. However, this different hospitalization strategy appears to be appropriate, as it leads to similar in-hospital mortality rates.

The secondary analysis revealed increased risk ratios for those infections expected to be associated with diabetes, such as osteomyelitis and pyelonephritis, and did not demonstrate significant risk ratios for infections where no link with diabetes has been postulated, such as appendicitis and HIV infection. Certain infections may be more common among people with diabetes because of another factor associated with both diabetes and the infection, such as obesity with biliary tract infections or peritoneal dialysis with peritonitis. Diabetes appears to be protective against male genital infections, which may reflect the sexual dysfunction experienced by some diabetic men (11).

Some limitations of the study must be considered. Because the study was conducted using administrative data not primarily collected for research purposes, the accuracy of the coding of diagnoses is uncertain, particularly for uncommon diagnoses in the secondary analyses. However, it is unlikely that data-coding accuracy is associated with the presence or absence of diabetes in a manner that would change the true risk ratio. Furthermore, the validity of the data are supported by the consistency of the results across many different infections. Secondly, the observed differences in the risk of infection between the diabetic and nondiabetic cohorts may not be due to differences in the frequency of infectious diseases, but rather due to differences in care-seeking behavior, caregiver vigilance, or physicians’ threshold for hospital admission. However, this appears unlikely by the homogeneity of the results across diagnoses and the observation that many infections for which hospitalization is not discretionary also had increased risk ratios.

These findings support in vitro evidence of impaired immune function in people with diabetes. The risk of and mortality from infectious diseases is higher in people with diabetes, and so, in addition to microvascular and macrovascular sequelae, infection must be thought of as a complication of diabetes. Clinicians must maintain vigilance for infectious diseases among diabetic patients.

B.R.S. holds a postdoctoral fellowship from the Canadian Institute of Health Research, and J.E.H. is a career scientist of the Ontario Ministry of Health and Long-Term Care. The opinions, results, and conclusions are those of the authors and no endorsement by the Ontario Ministry of Health and Long-Term Care or by the Institute for Clinical Evaluative Sciences is intended or should be inferred.

We gratefully acknowledge the assistance of Mei Tang and Andreas Laupacis with the completion of this study.