Article Text
Statistics from Altmetric.com
We agree with Buising et al that severity scores for community acquired pneumonia (CAP) that are accurate for predicting 30 day mortality, such as the pneumonia severity index (PSI) and CURB-65, are unlikely to be as useful for predicting the need for admission to the intensive care unit (ICU).1 In our institution, as their paper describes, the patients who die from CAP are generally older, have unstable co-morbid conditions, and are made “not for resuscitation”, so are not suitable candidates for admission to the ICU. We also agree that the revised American Thoracic Society criteria are of very limited use as a tool for predicting severe CAP as the major criteria (need for mechanical ventilation or inotropic support) identify patients who should already be in the ICU rather than predict those who will need it.2
However, we have some concerns with the methodology used to validate the scoring systems in their study.1 The entry criteria based on the admission diagnosis of CAP could lead to errors since, in our experience, the admission diagnosis is frequently incorrect. Published reports suggest that the disparity between clinicians’ reports of chest radiographs for CAP and those reported by radiologists is high, with 20–50% of radiographs not confirmed as demonstrating pneumonia, despite the clinician’s assessment as such.3–5 In our own institution, a review of the radiographs of almost 1000 patients admitted with an emergency department admission diagnosis of “CAP” showed that it was confirmed in less than half of the cases (unpublished data). Buising et al believe that this admission diagnosis is appropriate to analyse, since it is used by the admitting medical staff who first assess the patient to develop the patient’s severity score. However, if that is the case, we are concerned by the fact that Buising et al calculated the severity scores based on the worst results in the first 24 hours after admission, rather than the initial set of observations and investigations as is the intended use of both the PSI and CURB-65 assessment tools.6,7 Using the worst results from the first 24 hours may falsely increase these scores so that they may, in fact, be even less accurate than those reported by Buising et al. Ideally, a scoring system for identifying patients who may need more aggressive treatment should detect these patients as early as possible, rather than requiring the initial 24 hour observations to make this assessment. Secondly, it would appear from table 5 in the paper by Buising et al, in which only patients with a discharge diagnosis of CAP were analysed, that since the 95% confidence intervals for sensitivity and specificity for PSI class IV/V and modified BTS in predicting death and/or need for ICU management overlap, both these scoring systems are statistically comparable in predicting these outcomes.
While the findings of Buising et al are interesting, we believe that clinicians are still without a useful tool for identifying patients with severe CAP who are likely to require ICU management.
References
Authors’ reply
Charles et al have raised two important issues concerning our study1:
(1) Should scoring systems intended to guide admission or the need for ICU assessment or empirical antibiotic choices be built using the data from patients who the examining clinicians believe to have pneumonia based upon their interpretation of the chest radiograph (that is, exactly the group of patients to whom the scoring system will be applied in practice), or should they be built only using data from those patients in whom the radiologist described consolidation on the chest radiograph? We chose the former in order to make the results as directly relevant and applicable as possible and note that, in 88% of our patients, the treating clinicians applied a discharge diagnosis (with knowledge of the radiologists report) of pneumonia.
(2) Having chosen the patient group, should we build the model with data available to the emergency department clinician during the patient’s stay in the emergency department or only with the “first pass” data available? We built our models using data available at any time during the patient’s emergency department stay (provided this did not exceed 24 hours)—once again, exactly the time that key management decisions are being made by the emergency department clinicians. In practice, the vast majority of patients in our study recorded their worst values during the first 4 hours in the emergency department, which seems a reasonable time period for making early management decisions.
However far we may still be from a prediction score that is 100% sensitive and specific, we believe that the tools we have tested may be useful to the busy clinicians in our emergency departments. Importantly, our study suggested that the more complex tools offer little more than the simpler tools in identifying patients with severe illness when applied in a way that reflects “real life” clinical practice.