1. Clinical Decision Support: Using Logistic Regression to Diagnose COPD and CHF ©2012 Wayne G. Fischer, PhD 1 COPD patient inclusion criteria: Discharged 01Feb  31Oct2011 > 40 years of age Primary dx of COPD, or any one or more of the 11 “indicator” variables All multiple encounters, no matter when COPD first diagnosed

2. What is Logistic Regression? 2 ©2012 Wayne G. Fischer, PhD Start with the sigmoid…the “s-curve”: Probability X (predictor or indicator variable) 0 = absent 1 = present Condition / Event f

3. ©2012 Wayne G. Fischer, PhD 3 What is Logistic Regression? (cont’d)

4. ©2012 Wayne G. Fischer, PhD 4 Stepwise LR  7 of 15 predictors significant

5. ©2012 Wayne G. Fischer, PhD Diagnostics: 7-term model 5

6. ©2012 Wayne G. Fischer, PhD (more) Diagnostics: 7-term model 6

7. ©2012 Wayne G. Fischer, PhD Receiver Operating Characteristic (ROC) Curve Explained (sensitivity vs. specificity + cutoff) 7 A B

8. ©2012 Wayne G. Fischer, PhD Receiver Operating Characteristic (ROC) Curve Explained (cont’d) 8 ROC curve generated using various cutoff points (e.g., A and B are two different cutoff points)

9. ©2012 Wayne G. Fischer, PhD ROC Curve: 7-term model 9 Power = 1 - β α False Positive (1 – Specificity)

10. ©2012 Wayne G. Fischer, PhD ROC Table: 7-term model (see Word file) Choosing a Cutoff Point  10 False + True +

11. ©2012 Wayne G. Fischer, PhD 3-term model: CC + Bronch + Methylpred 11

12. ©2012 Wayne G. Fischer, PhD But, significant Lack of Fit (LOF) 12

13. ©2012 Wayne G. Fischer, PhD ROC Curve: 3-term model 13 Power = 1 - β α False Positive (1 – Specificity)

14. ©2012 Wayne G. Fischer, PhD ROC Table: 3-term model  Choosing a Cutoff Point 14 False + True +

15. CHF Prediction using Logistic Regression ©2012 Wayne G. Fischer, PhD 15 CHF patient inclusion criteria: Discharged 01Feb  30Nov2011 Primary dx of CHF, or any one or more of the “indicator” variables Age > 40 years, up to 100 years

16. ©2012 Wayne G. Fischer, PhD “All In” Model – 11 predictors 16

17. ©2012 Wayne G. Fischer, PhD Parameter Estimates – “All In” model 17

18. ©2012 Wayne G. Fischer, PhD ROC Curve: “All In” model 18 Power = 1 - β False Positive (1 – Specificity) α

19. ©2012 Wayne G. Fischer, PhD ROC Table: “All In” model (see Word file) 19 False + True + Choosing a Cutoff Point 

20. ©2012 Wayne G. Fischer, PhD 3-term Model: CC + Lasix + NT-proBNP 20

21. ©2012 Wayne G. Fischer, PhD LoF and Param Estimates: 3-Term Model 21

22. ©2012 Wayne G. Fischer, PhD ROC Curve: 3-Term Model 22 Power = 1 - β False Positive (1 – Specificity) α

23. ©2012 Wayne G. Fischer, PhD ROC Table Values: 3-Term Model Choosing a cutoff point 23 False + True +