Methods: Using data from 6 pneumonia cohorts, we designed PSI-HRs a 6-class prediction rule using the original prognostic weights of all PSI variables and modifying the risk score thresholds to define risk classes. We calculated the proportion of low-risk and high-risk patients using CURB-65 and PSI-HR and 30-day mortality in these subgroups. We compared the rules’ sensitivity, specificity, positive and negative predictive values for mortality all risk class thresholds and assessed discriminatory power using areas under their receiver operating characteristic curves (AUROCs).
Results: Among 13,874 patients with pneumonia, 1036 (7.5%) died. For PSI-HR versus CURB-65, aggregate mortality was lower in low-risk patients (1.6% vs. 2.2 %, p=0.005) and higher in high-risk patients (36.5% vs. 32.2%, p=0.27). PSI-HR had higher sensitivities than CURB-65 at all thresholds; PSI-HR also had higher specificities at the 3 lowest thresholds and specificities within 0.5 percentage points of CURB-65 at the 2 highest thresholds. The AUROC was larger for PSI-HR than CURB- 65 (0.82 vs. 0.77, p<0.0001).
Conclusions: PSI-HR demonstrated superior prognostic accuracy to CURB-65 at the lower end of the severity spectrum and identified high-risk patients with nonsignificant higher short-term mortality at the higher end.
Bibliographical noteAcknowledgments Author contributions: MJF and RSB are the guarantors of the content of the manuscript, including the data and analysis. RSB and MJF were responsible for writing and revision of the manuscript. All authors contributed to the conceptualization of the study design, had access to the study data and review and synthesis of the final version of this manuscript. The authors wish to especially thank Dr. Thomas J. Marrie (Dalhousie University) for his expertise and critical review of the manuscript.
Funding information: The authors received no financial support for the research, authorship, and/or publication of this article.
Data Availability StatementSupplementary data to this article can be found online at https://doi.org/10.1016/j.rmed.2022.106884.
- severity of illness
- prediction rules