Once again, the resulting decision tree chose from the presentation variables only the clinical score as the prediction rule, but now with a lower cut point of 0.65. This results in a slightly higher sensitivity (80 percent) but a lower specificity (72 percent). A clinical score of >0.65 means that at least four of the variables had been assigned a score of 1 (regardless of whether pulsus paradoxus was measured). Further increasing the prespecified penalty of discharging a patient who should have been hospitalized led to little change in this rule.
The CART method also provides a ranking of the relative importance of all variables for comparative purposes. The relative importance of variables obtained at presentation were clinical score (given a relative importance of 100 by convention), Sa02 (78), and number of previous admissions (33). buy levaquin online
Analysis 3 adds the variables available on disposition (after ER treatment) to the available list for the CART procedure. The clinical score at disposition is now the only variable chosen by CART, with a cutpoint of 0.65, and is more predictive than the clinical score on presentation. However, the score on disposition has lower sensitivity compared to the score at presentation, although the specificity is substantially increased (95 percent). Increasing the prespecified penalty of discharging a patient who should have been hospitalized (analysis 4) lowered the cutpoint for clinical score on disposition to 0.45, with a sensitivity of 80 percent and a specificity of 89 percent. A clinical score of >0.45 means that three or more of the variables were assigned a score of 1 regardless of whether pulsus paradoxus was measured or not.