Discussion
The main new finding of the present study is that a reduction in time to antibiotics was not found to be associated with an improvement of relevant clinical outcomes in our cohort of ED patients with mild to severe stages of sepsis. A reduction in time to antibiotics was also not found to be associated with an increase in number of surviving days outside the hospital at day 28 after ED presentation. Corresponding to the number of ventilator-free days in ICU-related studies, we chose this primary endpoint instead of hospital LOS to avoid confounding of the association between time to antibiotics and hospital LOS by mortality.
Our findings are in contrast to a study by Houck et al. of approximately 14,000 patients with a pneumonia, which found an increase in hospital LOS of 0.4 days with administration of antibiotics after four hours. Several differences may explain the discrepancy. The retrospective design of the study is prone to information bias, especially with regard to the timing of antibiotics. Furthermore, it consisted of a population with only elderly patients with pneumonia, while in our study all adults with all sources of infection were included. Also, there was no control for appropriateness of the initial choice of antibiotics and amount of fluids. Several smaller retrospective studies that found an increased hospital LOS with administration of antibiotics beyond two to eight hours had similar methodological drawbacks.
Unexpectedly, we found that in the PIRO 1 to 7 group, administration of antibiotics after three hours was significantly associated with a larger number of surviving days outside the hospital. It is unlikely that delaying antibiotic treatment is beneficial for patients with infections. The most likely explanation for the unexpected association of delayed administration of antibiotics with shorter hospital stay is that administration of antibiotics is more often delayed in patients with less severe infections. In contrast, severely ill patients may be administered antibiotics shortly after admission to the ED. This confounding effect of severity of illness may not be completely adjusted for by including the PIRO score in the multivariable analysis, because in the low illness severity range (PIRO 1 to 7) the PIRO score has limited discriminative value, as shown in the study by Howell et al.. In contrast to the study by Hranjec et al. in surgical ICU patients, the percentage of patients who received appropriate antibiotics in the ED was lower in the group with delayed administration of antibiotics. This could therefore not explain the association of delayed administration of antibiotics and the larger number of surviving days outside the hospital in PIRO category 1 to 7. Finally, in the sensitivity analyses the primary association of interest did not change, but the aforementioned unexpected finding was no longer significant. It cannot be excluded, however, that unknown confounders may be responsible for this unexpected finding.
In this prospective study in ED patients, a reduction in time to antibiotics was not found to be associated with a reduction in mortality. The most likely explanation for the discrepancy with the three large retrospective studies that showed that delayed administration of antibiotics is associated with increased mortality is the much lower overall mortality, that is, disease severity, in the present study. This suggests that, in less severely ill patients, timing of antibiotics is less important than other aspects of ED treatment, including appropriateness of antibiotics and initial resuscitation with fluids and oxygen. Although it should be emphasized that the present study was not designed to estimate the effect of other treatment variables on relevant clinical outcomes, it should be noted that several confounders of the primary association of interest, such as appropriateness of antibiotics, supplemental oxygen, and fluids, were significantly associated with the clinical endpoints of the present study. Interestingly, appropriateness of initially administered antibiotics was associated with lower mortality, but possibly at the expense of less surviving days outside the hospital. It may be hypothesized that these variables represent more important aspects of ED treatment compared to time to antibiotics, at least in study cohorts with relatively low mortality. Correspondingly, in a study by Rivers et al., early goal-directed therapy (EGDT) had an enormous impact on outcome in a study cohort with 47% mortality, while the ProCESS trial (Protocol-based Care for Early Septic Shock) showed no benefit of EGDT in a study cohort with a much lower mortality of around 20%, which is in line with the previous observations that treatment benefits depend on disease severity. Indeed, some effective therapies were shown to be less relevant or even harmful in low risk populations.
Study Strengths and Weaknesses
The strengths of the present study include the prospective design, the adjustment for illness severity, the quantification of the accuracy of the registration of time to antibiotics, and excellent control for predefined potential confounders, including ED treatment and relevant medication. However, there are also some limitations.
First, in the most severely ill group (PIRO >14), there were fewer than 400 patients, therefore the power in this group might be relatively low. However, although we wanted to explicitly investigate the effect of illness severity on the association between time to antibiotics and relevant outcomes, in retrospect the illness severity (PIRO category) did not affect the association. Therefore, we also investigated all 1,168 included patients in a model with PIRO score as an interaction term with time to antibiotics in SPSS. In this model, the interaction term was not significant (P = 0.209), indicating that in retrospect we indeed could have analyzed all patients in one group. After having performed this analysis the main conclusion of the study still stands: no association between time to antibiotics and relevant clinical outcomes was found. In this analysis approximately three times more patients were analyzed than needed (a priori calculated sample size of 400). It is therefore unlikely that the absence of an association is attributed to limited power of the study. The relatively small number of deaths (secondary outcome) in PIRO group 1 to 7 may have resulted in less accurate effect size estimations. The HRs in this group with very mild disease severity should therefore be interpreted with caution.
Second, because many patients had negative cultures, the estimation of appropriateness of the initial choice of antibiotics may include inaccuracies. We believe, however, that the predefined flow diagram to decide if an antimicrobial agent was appropriate was the most objective way in culture-negative patients. Furthermore, similar strategies to interpret appropriateness of antibiotics were used in the most relevant previous studies, facilitating comparison of our results with those found in the literature. In addition, a realistic association between efficacy of antibiotics and relevant endpoints is also obtained by taking into account potential side effects related to antibiotic administration to patients who may not have needed antibiotics, such as those suffering from viral infections. More importantly, in clinical practice, many patients receive antibiotics while being culture-negative, either reflecting more localized infections like a pneumonia without bacteremia, or viral infections. These patients should be included to assess the overall benefit of antibiotics.