A Multivariable Prediction Model for Pneumocystis jirovecii Pneumonia in Hematology Patients with Acute Respiratory...

The incidence of Pneumocystis jirovecii pneumonia (PjP) is rising. Longer time to treatment is associated with higher mortality.

To develop a multivariable risk prediction model for PjP diagnosis.

In a prospective multicenter cohort of ICU patients with hematological malignancies and acute respiratory failure, factors associated with documented PjP were identified. The risk prediction model was tested in an independent prospective multicenter cohort. We assessed discrimination (by areas under the receiver operating characteristic curves [AUCs]) and goodness of fit (by Hosmer-Lemeshow statistics). Model performance was assessed using 30 sets of imputed data sets.

Among the 1,330 patients, 134 of 1,092 (12.3%; 95% confidence interval [CI], 10.4-14.4%) had proven PjP in the derivation cohort, as did 15 of 238 (6.3%, 95% CI, 3.6-10.2%) in the validation cohort. The model included age, lymphoproliferative disease, anti-Pneumocystis prophylaxis, the number of days between respiratory symptom onset and ICU admission, shock, chest radiograph pattern, and pleural effusion. The median (interquartile range) score was 3.5 (1.5-5.0) (range, -3.5 to 8.5) in the derivation cohort and 1.0 (0-2.0) (range, -3.5 to 6.0) in the validation cohort. The best threshold was defined on the validation sample as 3, allowing us to reach 86.7% sensitivity and 67.7% specificity for PjP, with a negative predictive value of 97.9% in the case of 10% prevalence. The score had good calibration (goodness of fit, -0.75) and discrimination in the derivation cohort (mean AUC, 0.80; 95% CI, 0.76-0.84) and validation cohort (mean AUC, 0.83; 95% CI, 0.72-0.93).

The PjP score for hematology patients with acute respiratory failure can be computed at admission, based on readily available variables. Potential clinical benefits of using this score deserve assessment.