2013-AMIA-Slides
- 1. Quantifying the Effect of Statin Use in Pre-Diabetic Phenotypes Discovered Through Association Rule Mining John Schrom, MPH; Pedro Caraballo, MD; M. Regina Castro, MD; György Simon, PhD
- 2. + Diabetes affect >25m people in US; Pre-diabetes affects >79m. + Statins are used to lower cholesterol, cardiovascular mortality. § Among most prescribed drugs in US § Multi-billion dollar industry + Statins may increase risk of diabetes in normal populations. + Statins seem to have no effect in pre-diabetic populations. + Is this because pre-diabetes is a heterogeneous condition? Statin Use & Pre-Diabetes: Background
- 3. + Association Rule Mining + Propensity Score Matching + Statin Use and Pre-Diabetic Patients Outline
- 4. Association Rule Mining Obese Hyperlipidemia DM Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
- 5. Association Rule Mining Obese Hyperlipidemia DM Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes P(DM | Hyperlip) = 0.8 {Hyperlip} -‐> {DM}
- 6. Association Rule Mining Hyperlipidemia DM On Sta4n? Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Phenotype: {Hyperlipidemia} On Sta4n Not on Sta4n DM 2 2 No DM 0 1 RR: 1.51
- 7. Association Rule Mining Hyperlipidemia DM On Sta4n? (Confounders) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Are the staGn and no-‐staGn populaGons really comparable?
- 8. + Statistical matching technique to compensate for confounders + Two Steps: § Build a model: quantify how sick a patient is through predicting how likely they are to receive treatment § Match patients based on this propensity (“sickness”) score + This method can adjust for the added sickness/confounding associated with being prescribed statins. Propensity Score Matching: Overview
- 9. Propensity Score Matching: Overview Hyperlip DM On Sta4n? (Confounders) Score Yes 0 Yes Yes Yes 0.66 Yes Yes Yes Yes 0.66 Yes Yes 0 Yes Yes Yes Yes 0.66 On Sta4n Not on Sta4n DM 2 1 No DM 0 0 RR: 1.0 Phenotype: {Hyperlipidemia}
- 10. + 18,958 pre-diabetic patients from Mayo Clinic between 1999 – 2004 § Outcome: Progression to DM within 5 years § Lab Values § Medication History § Diagnosis History + Analysis in R § Association Rule Mining § Propensity Score Matching w/in 1 percentage point caliper Statin Use & Pre-Diabetes: Methods
- 11. Statin Use & Pre-Diabetes: Results Predictor % Gender: Male 49 Hypertension 31 Hyperlipidemia 37 Obese 20 IHD 10 PVD 2 ACE/ARB 13 Predictor % Beta Blocker 17 Ca Chan Blocker 7 DiureGc 13 Fibrate 2 StaGn 18 Aspirin 29 High SBP 21 Predictor % High DBP 9 High t-‐chol 43 Low HDL 23 High LDL 63 High BMI 27 High TGL 30
- 12. Statin Use & Pre-Diabetes: Results 5-‐year risk of DM On StaGn 13.6% Not on StaGn 9.8%
- 13. + Built Diagnosis + Medication Analysis Variables § Hypertension w/ medication § Hypertension w/o medication § Elevated BP w/ medication § Elevated BP w/o medication § Hyperlipidemia w/ (non-statin) medication § Hyperlipidemia w/o (non-statin) medication Statin Use & Pre-Diabetes: Method
- 14. Statin Use & Pre-Diabetes: Results
- 15. Statin Use & Pre-Diabetes: Results Sup RR Obesity, elevated BP w/ meds 0.05 1.73* Hyperlipidemia w/o meds, hypertension w/ meds 0.23 1.35* Hyperlipidemia w/o meds, elevated BP w/ meds 0.10 1.62* Obesity, Hyperlipidemia w/o meds, elevated BP w/ meds 0.03 1.76* Obesity, Hypertension w/ meds, elevated BP w/ meds 0.05 1.76* Hyperlipidemia w/o meds, hypertension w/ meds, elevated BP w/ meds 0.10 1.61*
- 16. Statin Use & Pre-Diabetes: Results Sup RR Obesity, hyperlipidemia w/ meds 0.19 0.59 Hyperlipidemia w/ meds, hypertension w/ meds 0.02 0.59 Hyperlipidemia w/o meds 0.52 1.14 Elevated BP w/ meds 0.12 1.32 Obesity, hyperlipidemia w/o meds 0.19 1.15 Obesity, hypertension w/ meds 0.12 1.19 Hypertension w/ meds, elevated BP w/ meds 0.11 1.31 Obesity, hyperlipidemia w/o meds, hypertension w/ meds 0.11 1.41 Obesity, hyperlipidemia w/o meds, hypertension w/ meds, elevated BP w/ meds 0.04 1.72
- 17. + Unmatched: Statins increase DM risk from 13 – 41% + Matched: Statin Use & Pre-Diabetes: Summary Hyperlipidemia Hypertension DM Risk Requires cocktail Treated + Controlled 0.59 No cocktail Treated + Controlled 1.35 No cocktail Treated + Uncontrolled 1.62
- 18. + György Simon + Pedro Caraballo + M. Regina Castro + Connie Delaney + Stewart Speedie + David Pieczkiewicz Thank You!