In this study of 29,273 patients hospitalized in internal medicine wards, we found a two-fold higher mortality during the first 2 days following admission among patients with community-acquired bacteremia compared with patients with negative blood cultures. Thereafter, mortality among patients with bacteremia was only slightly higher than among patients with negative cultures. Stratifying by type of bacteremia showed that mortality was highest during the first week after admission among patients with Gram-positive and polymicrobial bacteremia, but patients with polymicrobial bacteremia had an increased mortality for at least 180 days. After the first 2 days following admission, mortality in patients with Gram-negative bacteremia was similar to patients with negative blood cultures. The level of comorbidity did not greatly influence the association between blood culture status and mortality.
Although our use of data from population-based registries with complete follow-up and the universal health care coverage in Denmark minimized selection bias, we have to consider potential weaknesses when interpreting our results. We had no data on the indications for performing blood cultures. Therefore, to define a population of patients with suspected community-acquired bacteremia, we restricted the study to patients with blood cultures performed within 2 days of admission to an internal medicine ward and with no hospital contact during the preceding 30 days. Because we defined our condition of community-acquired bacteremia as any positive blood culture obtained within the first 2 days following hospital admission, the patient's exposure status could vary over the first 2 days of admission. Thus, patients with negative blood cultures who died shortly after admission may have had bacteremia that was undiagnosed. Compared with patients with bacteremia, culture-negative patients were less likely to have their blood cultures taken on the day of admission, and as a result, our relative estimates of mortality in the first 2 days following admission may have been biased because of misclassification of exposure status. Some patients with negative cultures may have had undetected bacteremia, particularly if they had received antimicrobial therapy ; this lack of detection could have led to underestimation of bacteremia-related mortality and consequently to more conservative estimates of relative mortality. Because we used administrative data lacking clinical detail, we had no information for evaluating a patient's clinical state at the time of blood culturing or about potential empirical antibiotic treatment in patients with negative blood cultures, which could have contributed to a favourable prognosis.
Measurement of disease severity in patients with community-acquired bacteremia poses a particular challenge because it is difficult to assess disease severity at the optimal time point immediately before the onset of bacteremia . Therefore, some of the clinical data used to calculate acute severity scores and the clinical presentation actually may be influenced by the bacteremia. Finally, we computed the CCI based on all previous diagnoses recorded in the Danish National Patient Registry. The CCI is widely used but cannot assess the existence of comorbidity as accurately as clinical data . Thus, inaccuracy of discharge data and diagnoses not included in the index may have reduced our ability to estimate comorbidity completely, and misclassification may have led to residual confounding that could have affected our mortality estimates. The validity of discharge diagnoses registered in the Danish National Patient Registry is variable but is generally high for the diseases included in the CCI . In any case, the results of blood cultures obtained during a current admission were unlikely to have affected the accuracy of the diagnoses from previous admissions. Thus, any misclassification should bias the observed mortality estimates toward unity.
By focusing on patients with community-acquired bacteremia and stratifying by type of bacteremia, our data extend previous studies comparing the prognosis of patients with positive and negative blood cultures [4, 5]. A Canadian study found that patients with bacteremia had higher in-hospital mortality in the 30 days after a positive blood culture than patients with a negative blood culture (27.3% vs. 7.3%) . In line with that study, we observed the highest mortality among patients with polymicrobial bacteremia; however, the Canadian study, unlike the current work, identified no difference in 30-day mortality between bacteremia caused by Gram-positive cocci and Gram-negative bacilli. Nonetheless, our result of a better prognosis with Gram-negative bacteremia is corroborated by the finding of a low mortality from E. coli bacteremia with urinary tract foci in the Canadian study. In a single hospital study from the US, Bates et al.  reported a higher short-term mortality (adjusted 30-day MRR 2.3, 95% CI: 1.2-4.4) but less-affected long-term mortality (adjusted 1-year MRR 1.3, 95% CI: 0.76-2.1) among 142 bacteremia patients compared with 142 patients with negative blood cultures (matched by age, gender, severity of underlying disease, and the presence of major comorbidity). Our study results are in accord with this result and further suggest that the increased mortality associated with community-acquired bacteremia occurs predominantly within the first days after admission. Conversely, an Israeli study reported an increased long-term mortality in patients with bacteremia when compared to controls without infectious diseases, matched by age, gender, underlying diseases, hospital department, and admission date (43% vs. 19% 180-day mortality rate) . The use of different comparison cohorts most likely explains this discrepancy.
Recently, Laupland et al.  reported that bacteremia was associated with a 60% increase in in-hospital mortality (crude OR 1.6, 95% CI: 1.1-2.2) among Canadian ICU patients with sepsis. When they controlled for variables that reflected the acute systemic response to infection, the adjusted OR was 1.1 (95% CI: 0.7-1.8), suggesting that the effect of bacteremia is mediated to a large extent by the severity of the acute systemic inflammatory response. Similarly, a study based on data from 170 French ICUs showed that bacteremia in patients with severe sepsis or septic shock was associated with mortality within 3 days of ICU admission (adjusted OR 1.7, 95% CI: 1.1-2.8) but not at 28 days after admission . We used overall mortality as our outcome measure because of the difficulty of distinguishing the contributions to mortality of the septic process and of underlying disorders. However, the inclusion of patients with negative blood cultures as a reference in our study is important when interpreting our findings because we surmise that patients with negative cultures had presented with several essential signs of sepsis, not only pyrexia. Our MRRs therefore may reflect the effect of bacteremia per se on mortality. Although underlying comorbidities may have influenced the physician's decision to obtain blood cultures, the association between blood culture status and mortality remained robust in analyses restricted to patients with infectious disease registry diagnoses and in different comorbidity strata.