Table 1.
Comparison of predictive performance for 30-day mortality between PSI-17 and PSI-20 using univariate, and risk-adjusted multivariable logistic regression models.
Fig 1.
The comparison of ROCs from (A) univariate logistic regression and (B) risk-adjusted multivariable logistic regression, adjusted for 11 risk factors for 30-day mortality.
Fig 2.
Calibration plots of mean observed rates versus predicted rates of 30-day mortality for univariate logistic regression on (A and B) and risk-adjusted multivariable logistic regression adjusted for 11 risk factors (C and D). Samples were equally divided into 8 groups, according to their predicted 30-day mortality probability. "Rug plot" below the main figures are histograms showing the number of observations for the corresponding predicted 30-day mortality probability.
Fig 3.
30-day mortality by PORT Class defined by Fine et al. (1997): PORT class II: ≤70; PORT class III: 71–90; PORT class IV: 91–130; PORT class V: >130.
30-day mortality rates for PORT Class I are 1.2% and 1.3% for PSI-17 and PSI-20, respectively. Individual patient PSI scores who survived or died are shown in green or yellow, respectively.
Fig 4.
Scatter plot of the components needed for the net reclassification index.
Open circles and crosses indicate alive and death, respectively. Blue symbols mean that PSI-20 improved PSI-17 since it had a higher predicted probability of 30-day mortality for those who died, or a lower predicted probability for those who were alive. Red symbols mean that PSI-20 didn’t improve PSI-17 since it had a lower predicted probability of 30-day mortality for those who died, or a higher predicted probability for those who were alive.