Another view on the prediction of outcomes in patients with community-acquired pneumonia

To the Editors:

At least two clinical rules for predicting short- and long-term mortality in patients with community-acquired pneumonia (CAP) have been successfully validated: the Pneumonia Severity Index (PSI), and the CURB-65 score (confusion, urea >7 mmol·L⁻¹, respiratory frequency ≥30 breaths·min⁻¹, systolic blood pressure <90 mmHg or diastolic blood pressure ≤60 mmHg and age ≥65 yrs) and its modifications. The biomarker procalcitonin (PCT) has received much attention as another tool to predict outcomes and, possibly, stratify patients according to treatment settings. Recently, we found that PCT might carry an additional predictive potential for mortality across the clinical prediction score CRB-65 (confusion, respiratory frequency ≥30 breaths·min⁻¹, systolic blood pressure <90 mmHg or diastolic blood pressure ≤60 mmHg, and age ≥65 yrs) [1]. Schuetz et al. [2] recently reported that PCT performs poorly as a predictor of mortality and does not increase mortality prediction levels of clinical scores.

There may be several reasons for these conflicting results. One of these reasons may be differences in the populations studied, particularly in the number and type of comorbidities. Another potential bias is the study design, since the study by Shuetz et al. [2] was primarily designed to guide treatment decisions and not to evaluate outcomes. Finally, the number of patients who received antimicrobial pre-treatment may affect the prognostic predictions. We previously showed that predictive values of PCT are considerably negatively affected by antimicrobial pre-treatment [3]. Unfortunately, Schuetz et al. [2] did not report the number of patients who received such pre-treatment.

In addition, we are tempted to ask whether a different analysis would not lead to divergent conclusions.

First, the authors decided to test predictions based on PCT levels using a four-level tool in order to make decisions on antimicrobial treatment. The use of only one threshold would have been statistically adequate, since deaths were very rare in the low-risk groups (four deaths in the group with <0.1 μg·L⁻¹ PCT and seven deaths in the group with 0.1–0.25 μg·L⁻¹ PCT). In fact, there was no linear trend across the four levels; however, figure 1 in that article demonstrates that mortality was twice as high in those with PCT levels >0.25 μg·L⁻¹, which is very close to our findings suggesting that a cut-off of 0.228 μg·L⁻¹ separates survivors from nonsurvivors quite confidently.

Secondly, from previous data, it appears that PCT level, alone or in conjunction with clinical scores, predicts mortality best in higher risk classes. This may also be a consequence of very low mortality in low-risk classes, leading to a high probability of missing a small, but possibly significant, difference.

Thirdly, mortality increasingly appears to be a problematic end-point of risk assessment. The authors found that the rate of complications and intensive care unit (ICU) admission, but not mortality, closely followed the four-level risk model of PCT. This may hint at the problem of treatment restrictions in elderly and/or severely disabled patients, also favouring death in lower and intermediate risk classes. Unfortunately, Schuetz et al. [2] did not report the number or mortality of patients at highest risk, i.e. those residing in nursing homes or being bedridden. In any case, complications and ICU admission appear to be the more robust end-points for the evaluation of predictive tools of outcome in patients with CAP.

Overall, receiver operating characteristics differed significantly from our findings, which showed PCT as a good predictor of mortality and CRB-65 plus PCT as the best. In the study by Schuetz et al. [2], the area under the curve (AUC) of initial PCT was very low (AUC 0.6, 95% CI 0.52–0.67). Potential reasons for this large difference are numerous, and include characteristics of the population studied, different study design and the number of patients with antimicrobial pre-treatment. Such variations challenge the general use of PCT as a predictive tool.

However, different conclusions may be drawn from the study by Schuetz et al. [2]. A PCT value of ∼0.25 μg·L⁻¹ identifies patients at increased risk of death. When PCT is used with clinical scores, it might give additional predictive value; in fact, whereas the predictive potential in low risk classes is currently unknown, it seems to be considerable in moderate- and high-risk classes [4]. Having said this, it appears from other studies that pro-adrenomedullin is superior in terms of prediction of prognosis [5–7].

In any case, mortality as end-point should be regarded with caution because of variations in populations studied, comorbidities and possible effects of treatment restrictions, and at least always be interpreted in the light of complication and ICU admission rates. In our view, this seems to be an important message of the study by Schuetz et al. [2].

• ©ERS 2011