Optimal Design of Multiple Medication Guidelines for Type 2 Diabetes Patients Jennifer E. Mason PhD Student Edward P. Fitts Department of Industrial and Systems Engineering NC State University, Raleigh, NC INFORMS Annual Meeting, Austin, TX November 7, 2010 1
Collaborators Brian Denton, PhD, NC State University Nilay Shah, PhD, Mayo Clinic Steve Smith, MD, Mayo Clinic Supported by the Agency for Health Care Research and Quality (R21HS017628) and the National Science Foundation (CMMI-844511) 2
Diabetes The American Diabetes Association estimates 23.6 million people have diabetes in the U.S. 8% of the population 90-95% have type 2 diabetes Two out of three people with diabetes will die from either stroke or coronary heart disease (CHD) 3
Treatment Managing a patient s cholesterol and blood pressure are important for preventing stroke and CHD events Numerous cholesterol medications (e.g., statins) and blood pressure medications (e.g., beta blockers) 4
Cost Projections Source: Huang et al., Projecting the Future Diabetes Population Size and Related Costs for the U.S., Diabetes Care, 32: 2225-2229, 2009 5
When and in what order should medications be initiated? 6
US Guidelines ATP III 1 : Diabetes patients now considered CHD risk equivalents. Treatment Goal: LDL < 100 mg/dl JNC 7 2 : Treatment Goal: SBP/DBP < 130/80 mmhg 1 Third report on the National Cholesterol Education Program Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III), NIH Publication No. 01-3670, 2001 2 The Seventh Report of the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure, NIH Publication No. 03-5233, 2003 7
Optimal Treatment Guidelines 8
Markov Decision Process Model Stages Ages 40 to 100 Decision Horizon: 40 to 80 States Actions Annual Decision Epochs TC, HDL, and SBP (each L, M, H, or V), HbA1c, smoking status, history of CHD event or stroke, medication status At each epoch, each medication is either initiated or delayed 9
MDP Model Actions A (l,mi ) = *I, W+ if m i = 0 *W+ if m i = 1 Rewards r l, m = R(l, m) C O C m C S l + C CHD l CF S l + CF CHD l where R l, m = R 0 1 d S l 1 d CHD l 1 d Med m 10
Optimality Equations Optimality Equations v t l, m = max α a (l,m) r l, m + λ p t α l, m l, m v t+1 (l, m ) (l,m ) for t, l, m 11
MDP Model 12
MDP Model 13
Complexity Total number of states: Health States Event States 4 3 2 6 2 2 40 = 655,360 Medication States Yearly Decisions Solved MDP model using backwards recursion Model instances solved in less than 18 minutes on a 2.83GHz PC with 8GB of RAM 14
Data Model Input Source R 0 = $100,000 Rascati (2006) Probabilities among health states Probability of death from other causes Probability of stroke and CHD events Mayo EMR and DEMS 1 CDC Mortality Tables 2 UKPDS Models 3 1 Gorman et al. 2000. 2 National Vital Statistics Reports, National Center for Health Statistics, 2007. 3 Stevens et al. 2001, Kothari et al. 2002. 15
Results 16
Meet Jack Age 55 Diabetes TC: 270 (V) HDL: 34 (L) SBP: 148 (H) 17
Jack s Treatment Plan US Guidelines Age 55 Age 56 Age 57 Age 71 statins + thiazides fibrates + ACE/ARBs beta blockers calcium channel blockers 22.48 expected QALYs and $32,592 expected treatment Optimal Treatment Age 55 Age 56 Age 57 Age 65 statins thiazides beta blockers ACE/ARBs 22.40 expected QALYs and $23,485 expected treatment 18
Meet Jill Age 40 Diabetes TC: 217 (H) HDL: 33 (L) SBP: 161 (V) 19
Jill s Treatment Plan US Guidelines Age 40 Age 41 Age 42 Age 43 statins + thiazides ACE/ARBs fibrates + beta blockers calcium channel blockers 37.53 expected QALYs and $30,875 expected treatment Optimal Treatment Age 40 statins Age 62 thiazides 37.66 expected QALYs and $15,051 expected treatment 20
Overall Tradeoff: Males 21
Overall Tradeoff: Males 22
Overall Tradeoff: Females 23
Overall Tradeoff: Females 24
Primary Prevention Results Objective: Maximize rewards for time until first event minus medication costs r l, m = R 0 C(m) if no events 0 if event or death 25
Males 26
Females 27
Limitations The patient cohort is from one health system Sparse clinical data to model other races or ethnicities Only stroke and CHD events are modeled 28
Conclusions Personalized treatment plans result in lower costs and greater expected QALYs Guidelines should manage cholesterol and blood pressure with coordinated treatment Use of optimal guidelines could result in large savings at the population level 29
Future Work Expand the model to include more medications (including glucose control medications) Prove threshold properties for initiation of treatment Provide sufficient conditions for the sequencing of treatment 30
Jennifer Mason jemason2@ncsu.edu THANK YOU 31