In this population-based study investigating bacteraemia in the MED, we found that 7.6% of all MED patients who had blood cultures drawn, within the first 48 hours after admission, had bacteraemia, corresponding to 3.5% of all MED patients. Although fever, elevated CRP and SIRS all are associated with bacteraemia, they would separately overlook one third of all bacteraemic patients as individual tests. Among all bloodcultured patients admitted to the MED with a low CRP, no SIRS and no fever registered at arrival, only 1.6% had bacteraemia. However, 23% of all blood cultures were drawn among these low risk patients.
Other studies found a positivity rate of 4–12.6% for blood cultures drawn in the ED [
1,
5,
6,
9,
11,
23‐
25]. In the present study the three most frequently detected pathogens were
E. coli, S. pneumoniae and
S. aureus. Similarly, other studies have found
E. coli, S. aureus, and
Streptococcus species to be the most frequently detected pathogens in blood cultures drawn in the ED [
5,
7‐
9,
23,
26,
27]. We found that among the febrile bacteraemic patients the highest proportion of bacterial species were
E. coli and
S. pneumoniae. In addition,
S. pneumoniae was likely to have a CRP response >100 mg/L and a SIRS response. Except for
E. coli, S. aureus was most commonly isolated among the non-febrile bacteraemic patients (Table
2). This finding is interesting because it indicates that the lack of fever response may depend on the microorganism present in the blood culture. However, we found no other studies analysing this aspect within the emergency department.
Single parameters
In this study, a large proportion of the bacteraemic patients presented to the MED without fever. In parallel, other studies have found similar results, ranging between 24–37% [
6,
11,
28,
29]. A review by Coburn et al. found that a temperature cut-point of ≥38.5° gave a positive likelihood ratio for bacteraemia of 1.4 [
5]. In the present study the positive likelihood ratio for patients with a temperature >38.0 was 3.4 (3.1–3.6). A positive likelihood ratio higher than 10 has previously been proposed as an indication of an acceptable value of a diagnostic test [
30]. The combination of a large proportion of bacteraemic patients being non-febrile and the positive likelihood ratio of 3.4 for temperature as a diagnostic test for bacteraemia supports our research hypothesis that the validity of temperature as a single parameter for bacteraemia is only modest.
Many previous studies show that CRP is a difficult biomarker on which to rely solely in diagnosing different kinds of infections, due to the known delay in CRP response [
19]. Other studies have evaluated the effect of CRP as a predictor of bacteraemia and have found that CRP has limited validity as a diagnostic test for bacterial infections, because of the low positive predictive value and a poor discriminatory value [
7‐
9,
31,
32].
Tokuda et al. have shown that the presentation of shaking chills increases the likelihood of bacteraemia (positive likelihood ratio of 4.7) [
25]. In the present study we have no systematic information regarding chills, but find in parallel to other studies that a combination of fever with other clinical parameters improves the diagnostic validity.
Combination of different parameters
A population-based study of all first-time blood cultured patients by Leth et al. has recently proposed SIRS to be an adequate predictor of bacteraemia, and reports a crude odds ratio (OR) for bacteraemia of 7.25 (95% CI 1.75–30.1), and a sensitivity of 96.6%, compared to bloodcultured patients without SIRS [
11]. In contrast, the present study finds a sensitivity of SIRS of 64.3%. The discrepancy is probably related to the difference in patient population. Leth et al. studied an inpatient population, which involved both community-acquired and nosocomial bacteraemia. SIRS and temperature equally predicted bacteraemia, despite temperature being one of the SIRS criteria. Temperature had a slightly better specificity while SIRS had a better sensitivity (Table
3).
Although included in the SIRS criteria, we chose to separately analyze temperature as it is readily available, frequently measured and often decisive for the decision to draw blood cultures in daily clinical practice
If the decision to order blood cultures were based only on temperature, CRP or SIRS, (in our study population) one third of all bacteraemic patients would have been overlooked. For the clinician to minimize the risk of overlooking bacteraemic patients, one should use more than one predictor as diagnostic test. At the same time it is necessary to balance the use of resources. Studies on diagnostic strategies have previously focused on identifying low risk patients without the need of a blood culture drawn, thereby reducing healthcare costs without compromising patient care [
1].
In the present study 95% of the bacteraemic patients had either a CRP above 100 mg/L, a rectal temperature above 38.0°C, or fulfilled at least two SIRS criteria. In our population it resulted in a negative predictive value of 99.5% (95% CI 99.3-99.7). For unknown reasons 23% of all blood cultures in the present study were performed in the group of patients with low risk of bacteraemia.
Implementation of a combined test, where all patients with either a temperature >38.0°C, a CRP ≥100 mg/L or positive SIRS are blood cultured, presents a very high negative predictive value. This combined test would entail a large group (23%) of low risk patients to be withheld from blood cultures, and thereby has the potential to decrease blood cultures drawn in the MED, among patients without suspicion of infection.
Despite the possible decrease in blood cultures among patients in low risk of infection, a combination test would entail a net increase in blood cultures from 45.9% to 63.0% if all patients with either a temperature >38.0°C, a CRP ≥100 mg/L or positive SIRS had blood cultures drawn in the MED. However, if the knowledge of patients in very low risk of bacteraemia is used by the clinician to avoid unnecessary blood cultures, it might be possible to reduce the total number of blood cultures in the MED. But this remains to be confirmed in a prospective controlled trial.
In parallel, Shapiro et al. estimated that by implementing their prediction rule for bacteraemia, blood cultures drawn in the ED, could be reduced by 27%, reflecting a substantial financial saving per year and furthermore, a decreased quantity of false-positive results [
1]. In 1990, Bates et al. developed a clinical prediction model that allows the clinician to stratify patients according to their risk of bacteraemia and recommended blood cultures to be taken in all febrile patients (>38.3°C). For patients with a normal temperature and no other risk factors for bacteraemia (as they described) clinicians should consider to withhold blood cultures, which corresponds to this study’s conclusions of determining the risk of bacteraemia based on multiple risk factors and not solely rely on one single parameter [
33].
It is a challenge to identify a perfect fast diagnostic test indicating bacteraemia. The right combination of a diagnostic test depends on the basic prevalence of the condition and the associated morbidity or mortality of the disease. Furthermore, it requires a well-validated strategy with high internal as well as external validity and reproducibility across different populations. The present study is not a validated model and cannot serve as such. However, it provides basic information to the clinicians not to rely solely on single parameters such as temperature, CRP or SIRS when they decide to order a blood culture.
The strength of our study is the consecutive inclusion of all adult first-time admission patients arriving to the MED within the study period and the complete follow up on all included patients due to the unique personal identification number used by all Danish citizens.
We are aware of some potential limitations. As patients without blood cultures are classified as negative/non-bacteraemic, this might influence the predictive values for the presented results. However we were not able to take into account, whether or not patients without bloodcultures had undiagnosed bacteraemia. The MED does not receive obvious cardiological, chronic oncological, haematological, nephrological or acute haemorrhagic patients, parturient women and paediatric patients. This means that the results do not apply to all acute medical patients. Furthermore, this is a single-centre study, reflecting the standard care at Odense University Hospital within this period, and therefore the results may not be entirely generalizable to other wards and hospitals.