We found that heart failure, absence of chest pain, contracting the disease in autumn, lower natremia and younger age were all independently associated with an atypical aetiology of pneumonia in hospitalized, adult, immunocompetent patients with moderately severe CAP.
A younger age, a higher body temperature and a lower leukocytes count have been associated with the presence of an atypical pathogen in previous studies . Hyponatremia [19,20,21,22,23,24,25,26], absence of chest pain [23, 24, 27, 28], and contracting the disease in summer or autumn [29,30,31] are well-described predicting factors for Legionella sp infection.
Chronic heart failure as a risk factor for AP was an unexpected finding. Heart failure has been associated with a significantly lower risk of Legionella sp infection in some, [20, 25] but not all studies. [32, 33] The relations between pneumonia and heart failure are complex and multidirectional. Heart failure is a known risk factor for pneumonia, and pneumonia frequently triggers acute heart failure and other cardiovascular events. Whether the observed association between chronic heart failure and increased risk of AP is causal or a chance finding should be investigated in other populations.
Confusion was significantly associated with the presence of an atypical pathogen in univariate, as extensively described in other studies [22, 34, 35]. We chose to exclude confusion form the multivariate analysis because its prevalence was lower in our population than in observational studies, probably because our patients were included in an interventional trial and had to sign an informed consent form. This low incidence meant that the impact of confusion on the multivariate model was expected to be low.
With an AUC of 0.78 (95% CI = 0.71–0.85), the accuracy of the CASH-75 score is comparable with other scores used for predicting the aetiology of pneumonia. The Legionella score proposed by Fiumefreddo et al. has an AUC of 0.86 in the derivation study and of 0.73 and 0.91 in validation studies [25, 36]. Unfortunately, we could not attempt to validate this score in our population, because LDH and C-reactive protein (two of its 6 criteria) were not measured in all patients. The Winthrop-University Hospital (WUH) criteria by Cunha predicts the presence of L. pneumophila with 21 different clinical features and a weighted point system. While the original study does not state accuracy, a validation study has described an AUC between 0.68 and 0.72 . The CBPIS scoring system is a weighted point system described by Keller et al. with an AUC calculated at 0.76 . The New Score proposed by Saraya et al. was derived on a cohort of only 102 patients, with an AUC between 0.62 and 0.68 . We did not find any score or clinical rule predicting the presence of any atypical pathogens, i.e. not only L. pneumophila but also M. pneumoniae or C. pneumoniae.
We aimed to build a score easy to use at the bedside. All the information needed to compute the CASH-75 score can be obtained through readily available information and routine laboratory tests. The absence of weighting of the different items confers simplicity to this score, enhancing its potential usefulness in a busy clinical setting.
We chose to compare patients with AP to both patients with another aetiology (eg. typical bacterial pathogens) and patients without any identified pathogen. It can be argued that some patients without identified pathogens had viral pneumonia and as such would not need any antibiotic treatment. Though this assumption is probably true, some of these patients may as well have infection with an undetected bacterial pathogen [38, 39]. At present, international guidelines do not recommend withholding antibiotics in patients with viral pneumonia, as associated bacterial infection cannot be reliably ruled-out . As our main aim was to allow safe withholding of atypical coverage in all patients presenting with pneumonia, we thought that it was more appropriate to include in the analysis patients without detected pathogens.
The main usefulness of the CASH-75 score is to exclude an atypical bacteria as the causative pathogen in order to safely withdraw atypical coverage from the empiric antibiotic treatment. For this purpose, there are two candidate cut-offs. The first one, < 2, has a sensitivity and negative predictive value of 100%. Using this cut-off, antibiotic coverage for atypical pathogens would have been avoided in 193/578 (33%) patients with a score of 0 or 1, without missing any cases of atypical pathogen. Based on the highest diagnostic odds-ratio, the ideal cut-off is < 3. Using this cut-off, atypical antibiotic coverage would be withheld in 400/ 578 (69%) patients, at the cost of 9 CAP caused by atypical pathogens not being covered by an adequate antibiotic. Both cut-offs may help reduce the prescription of atypical coverage before obtaining the results of any microbiological investigation. This may lead to less bacterial resistance, less drug adverse effects and drug interactions.
This study has several strengths; it was conducted using a prospective multicentric cohort with thorough adjudication of the presence of pneumonia. The protocol mandated search for typical and atypical pathogens in all patients, hence minimizing the risks of misclassification. Patients were representative of older individuals admitted at the hospital in Switzerland.
Nevertheless, some limitations must be recognized. Not all variables of interest described in the literature were available in our patients. For example, the smoker status remained unknown and the LDH were not measured, both being described in other studies as significant predictive factors for the presence of L. pneumophila. Because no systematic attempt was made to detect L.pneumophila or M.pneumoniae by culture or PCR in the sputum, we cannot completely exclude the presence of these pathogens in the NAP group creating a possible classification bias. However, all patients were tested for the presence of either pathogens with PCR on an oropharyngeal swab (M. pneumoniae) or urinary antigen detection (L. pneumophila), making significant misclassification unlikely. The diagnosis was confirmed by chest-X ray, which has lower accuracy than CT-scan for pneumonia . However, chest X-ray remains a frequently used tool to confirm pneumonia in clinical studies.
Finally, despite the model fulfilling the goodness-of-fit hypothesis, the total number of patients in the AP group was low, and overfitting of the score is a possibility, reinforcing the need for external validation.