Derivation and validation of a prediction rule for mortality of patients with respiratory virus-related pneumonia (RV-p score)

Abstract

Background: Respiratory viruses are important etiologies of community-acquired pneumonia. However, current knowledge on the prognosis of respiratory virus-related pneumonia (RV-p) is limited. Thus, here we aimed to establish a clinical predictive model for mortality of patients with RV-p. Methods: A total of 1431 laboratory-confirmed patients with RV-p, including 1169 and 262 patients from respective derivation and validation cohorts from five teaching hospitals in China were assessed between January 2010 and December 2019. A prediction rule was established on the basis of risk factors for 30-day mortality of patients with RV-p from the derivation cohort using a multivariate logistic regression model. Results: The 30-day mortality of patients with RV-p was 16.8% (241/1431). The RV-p score was composed of nine predictors (including respective points of mortality risk): (a) age ⩾65 years (1 point); (b) chronic obstructive pulmonary disease (1 point); (c) mental confusion (1 point); (d) blood urea nitrogen (1 point); (e) cardiovascular disease (2 points); (f) smoking history (2 points); (g) arterial pressure of oxygen/fraction of inspiration oxygen (PaO2/FiO2) < 250 mmHg (2 points); (h) lymphocyte counts <0.8 × 109/L (2 points); (i) arterial PH < 7.35 (3 points). A total of six points was used as the cut-off value for mortality risk stratification. Our model showed a sensitivity of 0.831 and a specificity of 0.783. The area under the receiver operating characteristic curve was more prominent for RV-p scoring [0.867, 95% confidence interval (CI)0.846–0.886] when compared with both pneumonia severity index risk (0.595, 95% CI 0.566–0.624, p < 0.001) and CURB-65 scoring (0.739, 95% CI 0.713–0.765, p < 0.001). Conclusion: RV-p scoring was able to provide a good predictive accuracy for 30-day mortality, which accounted for a more effective stratification of patients with RV-p into relevant risk categories and, consequently, help physicians to make more rational clinical decisions. The reviews of this paper are available via the supplemental material section.

Incorporation of Suppression of Tumorigenicity 2 into Random Survival Forests for Enhancing Prediction of Short-Term Prognosis in Community-ACQUIRED Pneumonia

“…To date, the clinical application of index-based prognosis remains very limited, with the CURB-65 score and the Pneumonia Severity Index (PSI) score remaining the most commonly used indicators to determine the prognosis of CAP patients [ 9 , 10 , 11 ]. However, several studies indicate that these two scores do not perform well in predicting the risk of ICU admission and death, so new severity scores are needed to predict the prognosis of CAP patients [ 12 , 13 , 14 , 15 ]. Over time, machine learning based methods are nowadays increasingly being used to predict CAP patients [ 16 , 17 , 18 , 19 ].…”