OBJECTIVE: To generate and validate an initial version of the predisposition, insult/infection, response, and organ dysfunction (PIRO) staging model for risk stratification in severe sepsis. The goal was to create distinct levels of mortality risk within each of the four categories (P, I, R, and O), and that these risk levels would be meaningful in terms of prediction independent of the other categories. DESIGN: : Retrospective analysis using a statistical model utilizing two large, global databases of patients with severe sepsis. SETTING AND PATIENTS: Database #1: Placebo-treated patients from a phase III clinical trial of patients with severe sepsis (PROtein C Worldwide Evaluation in Severe Sepsis [PROWESS], 840 patients). Database #2: Global severe sepsis registry performed in 276 intensive care units in 37 countries (PROmoting Global Research Excellence in Severe Sepsis [PROGRESS], 10,610 patients). INTERVENTIONS: None. METHODS: Classification and regression trees were used to classify patients and derive a scoring system from the PROWESS and PROGRESS databases with internal validation. Regression tree parameters included Chi-square tests and a minimum of five patients per node. The risk levels were done in a stepwise manner, adjusting for the previous categories. Initially, the predisposition scoring was developed, and subsequently, the infection scoring was then developed after adjusting for the predisposition levels, and so on. Logistic regression analyses, odds ratios, and area under the receiver operator characteristic curve were used to evaluate the scoring systems. MEASUREMENTS AND MAIN RESULTS: Each of the four PIRO components had similar odds ratios in multivariable logistic regressions. In PROWESS, the correlation of the PIRO total score and in-hospital mortality rates was 0.974 (p < 0.0001), and in PROGRESS, the correlation of the PIRO total score and hospital mortality rates was 0.998 (p < 0.0001). CONCLUSIONS: PIRO can develop into an effective model for staging severe sepsis, seems to be predictive of mortality, and may be useful in future sepsis research.
OBJECTIVE: To generate and validate an initial version of the predisposition, insult/infection, response, and organ dysfunction (PIRO) staging model for risk stratification in severe sepsis. The goal was to create distinct levels of mortality risk within each of the four categories (P, I, R, and O), and that these risk levels would be meaningful in terms of prediction independent of the other categories. DESIGN: : Retrospective analysis using a statistical model utilizing two large, global databases of patients with severe sepsis. SETTING AND PATIENTS: Database #1: Placebo-treated patients from a phase III clinical trial of patients with severe sepsis (PROtein C Worldwide Evaluation in Severe Sepsis [PROWESS], 840 patients). Database #2: Global severe sepsis registry performed in 276 intensive care units in 37 countries (PROmoting Global Research Excellence in Severe Sepsis [PROGRESS], 10,610 patients). INTERVENTIONS: None. METHODS: Classification and regression trees were used to classify patients and derive a scoring system from the PROWESS and PROGRESS databases with internal validation. Regression tree parameters included Chi-square tests and a minimum of five patients per node. The risk levels were done in a stepwise manner, adjusting for the previous categories. Initially, the predisposition scoring was developed, and subsequently, the infection scoring was then developed after adjusting for the predisposition levels, and so on. Logistic regression analyses, odds ratios, and area under the receiver operator characteristic curve were used to evaluate the scoring systems. MEASUREMENTS AND MAIN RESULTS: Each of the four PIRO components had similar odds ratios in multivariable logistic regressions. In PROWESS, the correlation of the PIRO total score and in-hospital mortality rates was 0.974 (p < 0.0001), and in PROGRESS, the correlation of the PIRO total score and hospital mortality rates was 0.998 (p < 0.0001). CONCLUSIONS: PIRO can develop into an effective model for staging severe sepsis, seems to be predictive of mortality, and may be useful in future sepsis research.