Quantitative Approaches to Improve Healthcare Access and ...

1 1 Rocky Mountain INFORMS, March 17, 2011 Quantitative Approaches to Improve Healthcare Access and quality Rocky Mountain INFORMS Chapter MeetingA panel presentation, featuring the work of:Linda LaGanga, ,3 Steve Lawrence, McKinney, Candidate1,4 Antonio Olmos, Samorani, Health Center of of of of Northern Colorado2 Rocky Mountain INFORMS, March 17, 2011 Healthcare Issues we address To overbook or not? If we schedule them, will they come? What would Deming do to Improve Healthcare ? To achieve efficiency and effectiveness of healthcare3 Rocky Mountain INFORMS, March 17, 2011 Where is our work developed and documented? Experience and data from the Mental Health Center of Denver Community mental health center serving over 14,000 people per year Surveys and interviews of other Healthcare providers/systems Presented at INFORMS annual conferences Other conferences.

2 Production & Operations Management Society Decision Sciences Institute Mayo Clinic Conference on OR/Systems Engineering in Healthcare American Evaluation Association4 Rocky Mountain INFORMS, March 17, 2011 Read more about Decision Science Journal (May, 2007) Journal of Operations Management (2010, in press) Conference presentations and proceedings Research posters on the wall opposite this room5 Rocky Mountain INFORMS, March 17, 2011 Appointment Scheduling and Overbooking Clinic Overbooking to Improve Patient Access and Increase Provider Productivity LaGanga, L. R., & Lawrence, S. R. (2007).Clinic overbooking to Improve patient Access and provider Sciences, 38(2), 251 Mountain INFORMS, March 17, 2011 Simple Overbooking Example7 Rocky Mountain INFORMS, March 17, 2011 Model Assumptions Number of patients booked, K: E(K) = SK= N S = Show rate, N = target n of patients K = N/S Patients scheduled at even intervals throughout the day T= N/K = S Inter-appointment times compressed by the show rate Patients arrive on time with probability S Patient service times deterministic Added variability in final version8 Rocky Mountain INFORMS, March 17, 2011 Overbooking.

3 Best CaseNo patients waitWaitingNo overtimeD9D7D5D3D1 ServiceX10A9X8A7X6A5X4A3X2A1 ArrivalsExpected number of patients (5) arrive, evenly spacedBest Time13121110987654321 Time SlotOvertimeRegular Time10 appointment slots / session; 50% show rate5 patients seen; no provider idle time; no patients wait; no clinic overtime9 Rocky Mountain INFORMS, March 17, 2011 Overbooking: Bunched EarlyW7W5W3W2 WaitingNo overtimeD7D5D3D2D1 ServiceX10X9X8A7X6A5X4A3A2A1 ArrivalsExpected number of patients (5)arrive, bunched earlyCase Time13121110987654321 Time SlotOvertimeRegular Time10 appointment slots / session; 50% show rate5 patients seen; no provider idle time; 4 patients wait; no clinic overtime10 Rocky Mountain INFORMS, March 17, 2011 Overbooking: Late ArrivalNo patients waitWaitingOT D10ID7D5D3D1 ServiceA10X9X8A7X6A5X4A3X2A1 ArrivalsExpected number of patients (5) arrive, one late arrivalCase Time13121110987654321 Time SlotOvertimeRegular Time10 appointment slots / session; 50% show rate5 patients seen; 10% provider idle time; no patients wait; 10% clinic overtime11 Rocky Mountain INFORMS, March 17, 2011 Overbooking: Bunched LateW9W8 WaitingOTD9D8D7 IID3D1 ServiceX10A9A8A7X6X5X4A3X2A1 ArrivalsExpected number of patients(5) arrive, bunched lateCase Time13121110987654321 Time SlotOvertimeRegular Time10 appointment slots / session; 50% show rate5 patients seen; 20% provider idle time; 2 patients waiting.

4 20% clinic overtime12 Rocky Mountain INFORMS, March 17, 2011 Overbooking: Extra ArrivalW9W7W5W3W2 WaitingOTD9D7D5D3D2D1 ServiceX10A9X8A7X6A5X4A3A2A1 ArrivalsMore patients arrive(6)than expected (5)Case Time13121110987654321 Time SlotOvertimeRegular Time10 appointment slots / session; 50% show rate6 patients seen; no provider idle time; 5 patients waiting; 20% clinic overtime13 Rocky Mountain INFORMS, March 17, 2011 Overbooking Utility Model14 Rocky Mountain INFORMS, March 17, 2011 Overbooking Utility Model Maximize clinic utility Trade-off Patient Access (number of patients seen) Average patient waiting times Expected clinic overtime Note that provider productivity is implicit in this model15 Rocky Mountain INFORMS, March 17, 2011 Relative Benefits and Penalties = Benefit of seeing additional patient = Penalty for patient waiting = Penalty for clinic overtime The values of , , and don t matter Just their ratios or relative importance16 Rocky Mountain INFORMS, March 17, 2011 Utility FunctionUSN ()

5 NOUUU ASN WO Expected utility withoutoverbookingExpected utility withoverbookingOUAWO Expected netutility withoverbooking17 Rocky Mountain INFORMS, March 17, 2011 Utility Function Described()NUASNWO Utility Benefit ofPatients that Show Less Utility Benefitw/o OverbookingLess PatientWaiting PenaltyLess ClinicOvertime PenaltyNet Utility Benefit from Overbooking (could be negative)18 Rocky Mountain INFORMS, March 17, 2011 Simulation Experiments Five clinic size levels N N= {10, 20, 30, 40, 50} Ten show rates S S = {100%, 90%, .. , 10%} Full factorial experiment SN = 5 100 = 500 factor levels 10,000 replications per factor 500,000 observations19 Rocky Mountain INFORMS, March 17, 2011 Regression Analysis Results from simulation analyzed using regression analysis Regression equations obtained Expected patient wait times Expected clinic overtime Expected provider productivity All coefficients significant R2= 98%+20 Rocky Mountain INFORMS, March 17, 2011 Sensitivity to Service Time VariabilityAverage Net UtilityN50R90N30R90N50R50N30R50N10R90N10 R50N10R10N30R10N50R10 Average of net utility UNwith overbooking as a function of service time variabilitycs , with and ( =1, = , = )

6 21 Rocky Mountain INFORMS, March 17, 2011 Conclusions Overbooking is one solution for appointment no-shows Can significantly Improve performance Patient Access (more patients seen) Clinic utility But with a cost Increased patient waiting & clinic overtime Good for some clinics, not for others22 Rocky Mountain INFORMS, March 17, 2011 Directions for Future Work Scheduling policies Double booking Wave scheduling Optimal overbooking policies Current overbooking policy is not optimal Dynamic programming Nonlinear waiting & overtime functions Long waits muchworse than short waits23 Rocky Mountain INFORMS, March 17, 2011 Lean Options for Walk-In, Open Access , and Traditional Appointment Scheduling in Outpatient Health Care Clinics 2008 Linda LaGanga and Stephen LawrenceLinda R.

7 LaGanga, of quality SystemsMental Health Center of DenverDenver, CO USAS tephen R. Lawrence, School of BusinessUniversity of ColoradoBoulder, CO USAMayo Clinic Conference on Systems Engineering & Operations Research in Health CareRochester, Minnesota August 17, 2009 Additional information available at: Mountain INFORMS, March 17, 2011 Data Mining in Appointment SchedulingMichele SamoraniPhD CandidateLeeds School of Business, University of Colorado at Boulder25 Rocky Mountain INFORMS, March 17, 2011 Finding Patterns with Data Mining26 Rocky Mountain INFORMS, March 17, 2011 Young clients are more likely to keep appointments with no reminder callDECISION TREE27 Rocky Mountain INFORMS, March 17, 2011If clients are under the age of years old and have low average CRM (<.)

8 5), then they are more likely to keep their appointmentsCLUSTERING28 Rocky Mountain INFORMS, March 17, 2011 Using Data Mining to Schedule Appointments29 Rocky Mountain INFORMS, March 17, 2011 Overbooking ShortcomingsSuppose service time = 30 minutesLittle waiting time and no overtimeSome waiting time and a high overtimeIf we could predict which patients show up and which don t, we could obtain a more controllable schedule101109:009:209:4010:0010:2010:40 11:0011:2011:40111012:00011109:009:209:4 010:0010:2010:4011:0011:2011:40011112:00 30 Rocky Mountain INFORMS, March 17, 2011 The methodEvery time a visit request arrives:1)A classifier is used to predict if it shows or not (for each day)2)The visit request is scheduled by solving a stochastic program through column generationNon controllable parameters Service time Revenue from seeing a patient Clinic overtime cost Waiting time costControllable parameters Number of slots K Scheduling horizon h Classification performance: Sensitivity (sn) Specificity (sp)How good we are at retrieving showing patientsHow good we are at retrieving non showing patients31 Rocky Mountain INFORMS, March 17, 2011 Productivity vs Punctuality Productivity: number of patients seen.

9 It is increased by: Punctuality: 1/(overtime + waiting time). It is increased by:32 Rocky Mountain INFORMS, March 17, 2011 Real world case: MHCD After playing for a few hours with the MHCD data set, I can achieve any of the following classification performances: sn= , sp = sn= , sp = sn= , sp = rate Same day 1 day 2 days 3 days 4 days R Low .74 .64 .65 .62 .61 .65 MHCD .87 .74 .75 .72 .71 .76 Goal: Find the best policy for MHCD in terms of: Overbooking Open Access Data Mining33 Rocky Mountain INFORMS, March 17, 2011 Policy DM OB OA.

10 (min) (min) 1 No No No 8 4 2 No No Yes 8 1 3 No Yes No 12 4 4 No Yes Yes 12 1 5 .6, .8 No No 8 5 .7, .7 No No 8 4 .9, .5 No No 8 5 6 .6, .8 No Yes 8 1 .7, .7 No Yes 8 1 .9, .5 No Yes 8 1 7.