Is fever control or improved survival the 'risk factor' for ventilator-associated pneumonia?

We read with interest the paper by Launey and colleagues [1] regarding the effect of fever control on the incidence of ventilator-associated pneumonia (VAP) in brain-injured patients. We commend the authors for addressing this relationship, given that fever and nosocomial infections are frequent in these patients. However, we have concerns regarding the interpretation of data.

Firstly, independent risk factors for VAP, namely disease severity and lung contusion [2,3], are unevenly distributed between groups. In addition, other risk factors, such as chronic obstructive pulmonary disease (COPD) and ICU readmission [2,4], were not taken into account, suggesting any effect of fever control is likely subject to confounding.

Secondly, mortality in the control group was higher (34% versus 23%), lowering the observation period if death occurred within 28 days. Thus, patients in the fever-control group were observed for a longer period of time, increasing their time at risk for VAP.

Thirdly, VAP incidence was higher in patients who were subject to fever control for longer than 3 days. However, duration of fever control was likely determined by factors also affecting the patients’ risk of acquiring VAP; that is, death. Not surprisingly, those who died within 3 or fewer days of initiating fever control did not develop VAP as frequently as those who did not.

In summary, the intervention group and historical control group in this study do not seem to be optimally matched on crucial parameters. Additionally, decreased mortality and longer follow-up in the intervention group likely resulted in uncontrolled lead time bias/attrition bias. Decreased mortality also seems contradictory to the claim that VAP, a deadly condition for ICU patients, is more frequent among fever-control patients.

We thanks Dr Harmon and colleagues for their comments, which we would like to nuance.

Indeed, COPD was not reported at admission; however, such disease is clearly unusual in the studied population. In a recent multicentre study performed in the same population, a history of respiratory disease (including COPD) was observed in only 3.6% [5]. Secondly, none of the included patients were readmitted in our ICU during the first 28 days of ICU stay, which was the endpoint period to assess VAP occurrence.

As reported in the statistical section, we used a competing risk multivariate model [6] which considers death within 28 days as a competing event. Consequently, this model takes into account only the periods of time during which the patients were exposed to the risk of VAP. Moreover, the benefit of this model is that the death event is not a confounding factor in the determination of VAP incidence between the patients receiving fever control for longer than 3 days and those who did not. The size of the study was not large enough to reach statistical significance.

Finally, we acknowledge that the matching of the two groups was not optimal in our observational study and obviously a prospective randomized study is required. Nevertheless, we would like to emphasize our study was not designed to assess the impact of fever-control strategy on mortality and, despite the apparent results, no conclusion can be drawn on mortality. Moreover, a recent meta-analysis showed VAP was not associated with over-mortality in medical and trauma patients [7].