Diabetes and Outcome of Community-Acquired Pneumococcal Bacteremia: A 10-year population-based cohort study

The study was conducted during 1992–2001 in North Jutland County, Denmark, with a population of 496,000 inhabitants, which is ∼9% of the total Danish population. The entire population in the county was provided with free, tax-supported health care. All patients hospitalized with acute conditions, including bacteremia, were treated in one of seven public hospitals, of which one (Aalborg University Hospital) served as both district and referral hospital.

Patients with pneumococcal bacteremia were identified in the microbiological County Bacteremia Registry, which is described in detail elsewhere (15). The registry is maintained by the Department of Clinical Microbiology at Aalborg University Hospital, which provides diagnostic bacteriology for the entire county. A case of pneumococcal bacteremia was defined as a clinical episode with Streptococcus pneumoniae detected by blood culture. The infection had to be present or incubating at admission to the hospital (16). Only the patient’s first episode of monomicrobial bacteremia was included. We excluded patients with either regular contact with hospitals or hospitalization within 30 days before admission with bacteremia (n = 68) (17), and pediatric patients aged ≤15 years (n = 61) because we consider these cases to constitute distinct groups.

Three different systems for blood culture were used during the study period: inoculation of blood into multiple tubes of bacteriological media in the laboratory (1992), the Colorbact system (Statens Serum Institut, Copenhagen, Denmark) (18) (1992–1995), and the BacT/Alert system (bioMèrieux, Durham, NC) (1996–2001). The nominal volume per blood culture for the three systems was 16–18, 20–22, and 28–32 ml for adult patients, respectively. Pneumococcal isolates were identified directly by Quellung reaction (Omniserum; Statens Serum Institut) or latex agglutination for pneumococcal antigen (Slidex pneumo-Kit; bioMèrieux) (19). Optochin susceptibility was confirmed by subculture on 5% horse blood agar. All isolates, including occasional ones with ambiguous results, were referred to the national reference laboratory at Statens Serum Institut for definitive identification and serotyping. We obtained an indirect measure of bacterial density at the time the first positive blood culture was drawn from 1) the hours of incubation until first indication of growth (>24 h = low density) and 2) the number of positive culture bottles per set, each comprising three bottles (only one positive culture bottle = low density). A 1-μg oxacillin disc was used to screen for the presence of penicillin resistance, and this was confirmed by penicillin E-test (Biodisk, Solna, Sweden). Pneumococcal isolates with a minimum inhibitory concentration of penicillin ≥2 μg/ml were classified as resistant. Ongoing antibiotic therapy at first notification of a positive blood culture was categorized as therapy including a β-lactam/macrolide antibiotic, any other antibiotic therapy, or no antibiotic therapy. The probable focus of infection was assessed based on microbiological and clinical findings.

A 10-digit civil registry number has been assigned to every Danish citizen since 1968 in the Central Population Registry, making valid record linkage for each individual possible. We identified diabetic patients by record linkage with two population-based registries. The Prescription Database of North Jutland County (20), initiated on 1 January 1991, retains key information on prescriptions for refundable drugs dispensed from all pharmacies in the county. The database includes insulin (Anatomical Therapeutical Chemical [ATC] classification system code A10B) and all oral antidiabetic drugs (ATC code A10A), which are available by prescription only. Among our patients with pneumococcal bacteremia, we identified 59 users of antidiabetic drugs. The County Hospital Discharge Registry (21) is an administrative public registry that is truly population-based and covers all nonpsychiatric hospitalizations in the county from 1 January 1977. The registry includes civil registry numbers and up to 20 discharge diagnoses coded by medical doctors at discharge according to the Danish versions of the International Classification of Diseases (ICD). From this registry, we identified six additional patients with a discharge diagnosis of diabetes before or upon admission with pneumococcal bacteremia. We used the diagnosis of diabetes according to the ICD, 8th revision (ICD-8) (codes 249–250), until the end of 1993, and subsequently, according to the 10th revision (code E10–E11). (The 9th revision has not been used in Denmark.)

Hospital records were reviewed for 65 patients with potential diabetes using a detailed standardized form. To confirm the diagnosis of diabetes, we searched previous laboratory reports, using the World Health Organization diagnostic criteria from 1985 (22), with certain modifications followed in most epidemiological studies (23). We thus classified patients as having diabetes if they had a single, fasting venous blood glucose value >6.7 mmol/l or a random venous blood glucose value >10.0 mmol/l with a concurrent medical history of diabetes. Patients were categorized as having newly diagnosed diabetes if they maintained the diagnostic criteria after recovering from the pneumococcal infection (10 patients). Two patients with impaired glucose tolerance were excluded.

The patients were separated into type 1 and type 2 diabetes according to conventional clinical details such as age and weight at onset of diabetes, treatment modality, or propensity to ketoacidosis. In case of uncertainty, consensus was reached with a senior diabetologist. HbA_1c was evaluated from the value closest to the day of admission. Ketoacidosis was defined as presence of hyperglycemia, massive ketonuria, and metabolic acidosis with P-bicarbonate <18 meq/l. Nonketotic hyperglycemic-hyperosmolar coma was defined as presence of hyperglycemia >35 mmol/l, hyperosmolarity, and absence of significant amounts of ketone bodies in the urine. Diabetic microvascular complications included nephropathy, retinopathy, or neuropathy, as documented in the hospital record and with no other suspected etiology than diabetes. Macrovascular complications included history of transient ischemic attack, stroke, angina pectoris, myocardial infarction, coronary- or peripheral-artery revascularization, claudication, and ulcer or amputation on the lower extremities as a result of ischemia.

The following patient groups were separated according to age: adults (>15 to 65 years), elderly (>65 to 80 years), and aged (>80 years). To adjust for coexisting morbidity, we calculated a comorbidity index developed by Charlson et al. (24–26), which has been validated for the prediction of short- and long-term mortality. The Charlson index for each patient, diabetic and nondiabetic, was calculated based on previous discharge diagnoses. Thus, at the time of the patient’s hospital admission with bacteremia, we did a computerized linkage with the County Hospital Discharge Registry to identify ICD codes from all of the patient’s previous hospitalizations in the county. Three levels of the index were defined: 0 (low), which corresponded to patients with no recorded underlying diseases implemented in the Charlson index; 1–2 (medium); and >2 (high). Diagnosis of diabetes was excluded from the index.

We further examined variables reflecting bacteremic disease severity at the time of hospital admission in a subsample of the study population. This was done to assess the possibility of surveillance bias, i.e., a situation wherein the proportion of patients with mild or less advanced bacteremic disease is higher among diabetic compared with nondiabetic patients due to closer medical surveillance. Hospital records were reviewed for the 63 diabetic patients and 63 nondiabetic patients matched within sex and age-groups. Sepsis on admission was defined in accordance with the criteria of Bone et al. (27) as two or more of the following: body rectal temperature >38°C or <36°C, heart rate >90 bpm, tachypnea >20 breaths per minute or PaCO₂ < 32 mmHg, and leukocyte count >12 × 10⁹/l or <4 × 10⁹/l. Severe sepsis was defined as sepsis with one or more of the following: acute alteration of mental state, sepsis-induced hypotension with a systolic blood pressure <90 mmHg, or a serum creatinine of >140 μmol/l on admission. We also assessed the serum concentration of C-reactive protein, the presence of a correct diagnosis of infection by the attending physician, and if antibiotic therapy was instituted immediately after physical examination without any delay. Finally, we linked our cohort to the Central Population Registry, which is electronically updated daily, to obtain date of death or migration within 90 days after the episode of bacteremia.

We analyzed data by obtaining contingency tables for the main study variables: diabetes, sex, age-group, comorbidity, focus of infection, and death after 30 and 90 days based on Kaplan-Meier curve estimates. These were constructed from the date of the patient’s first positive blood culture. For each prognostic factor, the category with lowest risk for death was used as reference category.

Cox proportional hazards regression analyses were used to compare the mortality among the diabetic patients with that among nondiabetic patients, with estimation of the hazard ratios (mortality rate ratios [MRRs]) and associated 95% CIs, adjusted for sex, age, and level of comorbidity as assessed by the Charlson index. The Cox proportional hazards regression analyses were conducted both including and excluding the focus of infection. The assumptions for the model were assessed graphically. Statistical analyses were performed with the use of STATA software (version 8.0; STATA, College Station, TX).

The study was conducted according to guidelines of the regional scientific ethics committee for use of clinical and laboratory data and approved by the Danish Registry Board (J.nr 2002-611-0060).

During the study period, 628 patients aged >15 years were hospitalized with a first episode of community-acquired pneumococcal bacteremia. Of these, 63 (10.0%) patients had a known diagnosis of diabetes. Table 1 summarizes the characteristics of diabetic and nondiabetic patients. Diabetic patients were slightly older (median age 71.9 vs. 67.0 years) and were found to have more prior myocardial infarctions (11 vs. 4%), congestive heart disease (25 vs. 8%), and any heart disease (43 vs. 17%). Details about the patients’ infection including bacteremia density, focus, penicillin resistance, and choice of antibiotic therapy are also shown in Table 1. Pneumococcal serotype distribution was similar between groups, the most frequent serotypes being types 1 (18% of diabetic vs. 21% of nondiabetic cases), 4 (8 vs. 13%), 14 (11 vs. 7%), 12F (8% in both), 7F (10 vs. 7%), 3 (10 vs. 6%), and 9V (5 vs. 7%). The serotype coverage rates of the 23-valent pneumococcal vaccine were 89 and 94% for diabetic and nondiabetic patients, respectively (data not shown).

Table 2 shows the characteristics of the 63 patients with diabetes. Ninety-four percent of these patients had type 2 diabetes, and micro- and macrovascular complications were present in 17 and 37%, respectively. In five patients (8%), ketoacidosis was present (median P-bicarbonate 12.2 meq/l, range 9.7–16.6; reference interval 21.3–26.5 meq/l); three of these had been formerly diagnosed as having type 2 diabetes. Only one patient had nonketotic hyperglycemic-hyperosmolar coma. Ten diabetic patients were newly diagnosed on admission with bacteremia; of these, 1 patient was admitted with ketoacidosis.

In our subsample study of the 63 diabetic and 63 sex- and age-matched nondiabetic patients, we found the initial C-reactive protein to be higher in the diabetic group (median 277 vs. 204 mg/l), and a higher proportion of diabetic patients fulfilled the criteria of severe sepsis (56 vs. 40%). The proportion of patients with a correct diagnosis of infection on admission (81 vs. 76%) and with institution of antibiotic therapy without delay (46 vs. 48%) was comparable. Only one (diabetic) patient who had been splenectomized was known to have a prior pneumococcal vaccination.

Table 3 shows diabetes and several other factors associated with mortality after 30 and 90 days of follow-up in 628 episodes of community-acquired pneumococcal bacteremia. The mortality in diabetic compared with nondiabetic patients was 11.1 vs. 16.5% after 30 days, resulting in a crude MRR of 0.7 (95% CI 0.3–1.4). Of the seven diabetic patients who died within 30 days, five were >80 years of age, and none of these deaths were associated with ketoacidosis or nonketotic hyperglycemic-hyperosmolar coma. After 90 days, 16.0% of diabetic patients had died, as compared with 19.5% of nondiabetic patients (crude MRR = 0.8, 0.4–1.5). Kaplan-Meier curves for 90 days of follow-up for diabetic and nondiabetic patients are shown in Fig. 1. After adjustment for sex, age-group, and comorbidity, the association between diabetes and mortality after 30 and 90 days of follow-up was almost unchanged (adjusted MRR = 0.6, 0.3–1.2). The risk estimates were similar if age was included as a continuous variable in the model instead of categorized age-groups.

Patients with a high bacteremia density had a poorer prognosis than those with a low density (mortality 16.7 vs. 13.0% after 30 days and 19.8 vs. 16.0% after 90 days), as was the case for patients with a focus of infection within the central nervous system (30-day mortality 20.0%) or an undetermined focus (30-day mortality 42.5%) compared with a respiratory focus of infection (30-day mortality 13.8%) (data not shown). If focus of infection was included in the analysis, the adjusted MRR remained 0.6 (95% CI 0.3–1.3) after 30 days and was 0.7 (0.4–1.4) after 90 days (not shown). Excluding the 10 patients with newly diagnosed diabetes, in whom diagnosis may have required survival of the acute stage of infection, did not significantly change the estimate (adjusted MRR = 0.6 [0.3–1.4] after 30 days).

In this analysis of 628 patients with pneumococcal bacteremia, those who had diabetes did not have a higher case fatality compared with their nondiabetic counterparts. On the contrary, the estimated MRR seems to indicate a slightly better prognosis among patients with diabetes.

Our findings agree with the limited data available on the outcome of pneumococcal bacteremia specifically in patients with diabetes (1–3,11–14). Watanakunakorn et al. (1) showed that an association between diabetes and increased mortality in 385 patients with pneumococcal bacteremia disappeared after adjustment for higher age and comorbidity in the diabetic group. Previous case series, however, included few patients with diabetes, making it difficult to assess the impact of diabetes on outcome in pneumococcal bacteremia. Several studies (28–30) on unspecified bacteremia in diabetic patients have reported similar case fatality in patients with and without diabetes. Unlike our study, most reports have determined diabetes status based on interviews or reviews without strict criteria, and few studies have adjusted for confounding by comorbidity. Our data, combined with previous results, give strong evidence that diabetes is not associated with a worse prognosis in pneumococcal bacteremia.

Our study has both strengths and limitations. We used prospective population-based registries with complete follow-up data. Diabetes and outcome data were recorded independently of this study, which makes bias due to differential diagnostic effort by the study hypothesis unlikely. We used two data sources to identify patients with diabetes. Combined, these data sources are nearly complete (31) and proved to be of high quality regarding diabetes. On the other hand, the control subjects may have included some diabetic patients who were never previously hospitalized or were treated with drugs. This may have led us to underestimate the differences in outcome between patients with diabetes and those without. Further, diabetic subjects and their physicians may be more alert to possible infections and thus, milder cases of bacteremia may have been admitted to the hospital among diabetic patients. However, our findings of a comparable measure of bacteremia density and greater severity of inflammation and sepsis in diabetic patients speak against a more meticulous case ascertainment related to diabetes status. Focus of infection did not explain the differences in outcome.

Numerous in vitro studies have shown altered immunity in diabetes, including decreased leukocyte function (8). It is possible that patients with diabetes have a lower threshold of bacterial translocation into the bloodstream during less severe pneumococcal infection due to vascular changes or leukocyte dysfunction. Moreover, as hypothesized by Moss et al. (32), a less active inflammatory cascade may protect diabetic patients with sepsis against the development of acute respiratory distress syndrome. They found that in patients with septic shock, diabetes was associated with a lower risk of developing acute respiratory distress syndrome as compared with that of nondiabetic patients.

Coding and diagnosis of comorbidity (for example, heart disease) in our study may have been more complete for patients with known diabetes due to more frequent hospitalizations. However, adjustment for comorbidity and other possible confounding factors had only a very minor influence on the risk estimates, indicating that none of the factors including comorbidity were strong confounders of the association between diabetes and outcome of pneumococcal bacteremia.

In this observational study, other unmeasured factors may have had an impact on the outcome. If patients with diabetes were hospitalized at an earlier stage of infection, we would have expected an overrepresentation of patients with low bacteremia density, low C-reactive protein, and lesser severity of sepsis, which was not the case. We found no difference in timing and choice of antibiotic therapy. Previous pneumococcal vaccination was probably underreported in the hospital records. However, based on sales statistics from the Danish Medicines Agency, the uptake of pneumococcal vaccine in North Jutland County has been as low as 2 of 1,000 people per year since 1997, when national vaccine recommendations were proposed (Statens Serum Institut, December 1996), and according to our clinical experience, patients are rarely vaccinated because of diabetes. Further, it is uncertain if prior pneumococcal vaccination can influence the outcome of bacteremic pneumococcal disease. Hence, it does not have major impact on our estimate. Intensive insulin therapy (33) and statins (34) have recently been associated with reduced case fatality of bacteremia. We may only speculate about a benefit on survival in the 46% of diabetic patients who received insulin during hospitalization in this study. Statin use in our diabetic patients was unknown, but the prevalence of patients on lipid-lowering drug therapy was generally very low in our county until the end of the study period, increasing from 2.8 to 9.4 per 1,000 women and from 3.8 to 12.5 per 1,000 men from 1994 to 1998 (35).

In conclusion, this study provides strong evidence against the hypothesis that patients with diabetes carry a higher case fatality of community-acquired pneumococcal bacteremia compared with nondiabetic patients.

This study was supported by the Western Danish Research Forum for Health Sciences (Vestdansk Forskningsforum) and by grants from the Medical Research Council of North Jutland, the North Jutland County Medical Association, the Heinrich Kopp’s Legat, and the A. P. Møller Foundation for the Advancement of Medical Science. Research at the Department of Clinical Microbiology, Aalborg Hospital, is supported by a grant from Det Obelske Familiefond.

We thank the staff of the Hospital Discharge Registries in North Jutland County (Amtsgaarden) for preparing the data. We also thank the staff of the seven hospitals in the county for their great help with retrieving hospital records.

Results of this study have been presented in part at the 13th European Congress of Clinical Microbiology and Infectious Diseases, Glasgow, U.K., 10–13 May 2003.