MISPERCEPTION IN CHOOSING MEDICARE DRUG PLANS. National Predictive Modeling Summit September 23, 2008

Transcription

1 MISPERCEPTION IN CHOOSING MEDICARE DRUG PLANS Jeffrey R. Kling, Sendhil Mullainathan, Eldar Shafir, Lee Vermeulen, and Marian V. Wrobel Presented by Marian V. Wrobel, ideas42, Harvard University National Predictive Modeling Summit September 23, 2008

2 Author Affiliations & Acknowledgements Jeffrey R. Kling, The Brookings Institution Sendhil Mullainathan, Harvard Eldar Shafir, Princeton Lee Vermeulen, University of Wisconsin Marian V. Wrobel, Harvard We gratefully acknowledge support for this work provided by the John D. and Catherine T. MacArthur Foundation, the Charles Stuart Mott Foundation, the Robert Wood Johnson Foundation s Changes in Health Care Financing and Organization Initiative, and the National Institute on Aging (P01 AG005842). We also thank CVS Caremark Corporation and Experion Systems ( for sharing data.

3 INTRODUCTION Behavioral Economics Study of how real people make choices Draws on economics & psychology Today: Choice among Medicare drug plans Randomized experiment Show that a smart but small intervention has big effect on choices Suggest also that intervention made people better off

4 Choice among Public Services Policy interest in choice & competition for public benefits E.g. prescription drug insurance, schools, social security Rationale for choice Tastes & circumstances differ Choices lets people choose what they want Private provision & competition yield efficiency and innovation Concerns about choice People make dumb choices Sometimes we wish to protect or influence them

5 Approaches to Influencing Choices Laws & regulations You must eat your peas! Financial incentives If you eat your peas, you can play on the computer. Education Peas promote good health. Architecture / environment (today s focus) Peas are within easy reach & potato chips are far away.

6 Psychology of Choice Choice difficulty Proliferation of alternatives may be detrimental Tendency to delay, pronounced among elderly Tendency to focus on easy information Available Comparable Invariant E.g. focus on premium not out-of-pocket costs

7 Public Perception that Medicare Part D Confusing

8 Outline Practical background (Medicare Part D) Conceptual framework (misperceived prices) Original background re information environment Randomized field experiment Intervention = publicly available info on predicted drug plan costs Significant effect on behavior Suggestive evidence that people better off

9 PRACTICAL BACKGROUND Medicare Part D Free-standing private drug plans Voluntary, choice-based Launched Spring million enrolled in 2007 Open Enrollment For 2006: November 15, 2005 May 15, 2006 For 2007: Nov. 15 Dec. 31, 2006 Study timeframe.

10 Complexity of Choice for Typical Senior ~ 54 plans Differ by: Premium Deductible Cost-sharing schedule Formulary Pharmacy networks Utilization management tools Customer service Brand & financial stability of sponsor Plan difference interact with individual differences Drugs & other

11 THEORETICAL FRAMEWORK Misperceived Prices Standard economics Rational agents make decisions based on all available information Environment influences choices via costs of acquiring information Misperceived prices (behavioral economics) People make decisions based on costs & benefits as they perceive them Potential for error/variation in these perceptions Environment affects perceptions

12 Three Behavioral Predictions If we change environment Present predicted cost information Same information that is easily available from Medicare And use use psychological principles known to promote action Then 1. Choices will change. 2. Predicted costs will be lower. I.e., the changes will be consistent with the nature of the intervention 3. Intervention effect will be concentrated in out-of-pocket costs not premiums No judgment about underlying process or seniors welfare.

13 ORIGINAL EVIDENCE KNOWLEDGE & DEMAND FOR INFORMATION Phone & written surveys Phone: 351 seniors enrolled in Part D Written: 4646 seniors Conducted Spring 2007

14 Low Knowledge and Effort Majority knew basics Different plans are better for different people (86%) You can only change plans during open enrollment (74%) But few in command of specifics Some (not all) plans have deductible (37%) Some (not no) plans offer coverage in the gap (37%) Few had ever sought personalized information 34% had ever compared plans side-by-side 18% had ever reviewed personalized comparisons Most relied on passive, impersonal materials Mailings from plans & Medicare

15 Satisfaction & Complacency 86% rated their 2006 plan good or better 73% did not consider switching for % considered switching for 2007 but did not 10% switched plans from 2006 to 2007 Plans do not promote reviewing choices ANOCs highlight You do not need to do anything Consistent with notion of misperception

16 SUPPLY OF INFORMATION Drug Plan Info via Medicare.gov Medicare s website tool, the Prescription Drug Plan Finder Personalized, comparative information Allows input of information on prescriptions Also, preferences about pharmacy location & mail order use Generates an estimated annual out-of-pocket cost for each plan Predictions assume static drug profile Source of our study s information

17

18 Medicare Info by Phone Medicare (Audit data, 12 calls) Personalized comparative cost information Gathered inputs for personalized plan suggestions Read back two or three choices (sometimes with plan features) Then, offered to enroll the beneficiary Not an arduous process for the senior SHIPS (State Health Insurance Programs) (5 calls) Medicare s local outreach arm Similar to Medicare or referred to Medicare

19 Limited Assistance from Other Sources Pharmacies (88 in-person visits) Personalized in-store help at 10 percent General print materials at 78 percent Senior centers (8 in-person visits) Some help at some centers Other help-lines (web search & 12 calls) Generally not helpful A few exceptions Even Basic message of Choice among drug plans has significant cost implications. Help is available from Medicare not widely disseminated.

20 INFORMATION EXPERIMENT Concept A small intervention Information from Medicare website Not new information Not information that is hard to get Slight twist to create environment favorable to action We focused on costs because Medicare focuses on cost No implication that cost is optimal focus. Did not reduce effort Seniors still had to find Medicare phone number & make call to change plans Design = randomized experiment

21 Methods Participants: patients of University of Wisconsin Hospital, over 65 Enrolled in Part D Excluded if receiving subsidies or not resident of WI Baseline interview, Fall of 2006 (prior to randomization) Drug utilization etc. Conducted by UW pharmacy students Intervention letter, December 2006 First follow-up, Spring 2007 Switching, predicted cost N=406 Second follow-up, Spring 2008 Evaluation of own choices N=306

22 Information Intervention Both groups: Letter on university stationery Standard introductory and concluding paragraphs Brochure on using the Medicare website Comparison group: address of Medicare website only

23 Intervention Group: Personalized Information Simple, personalized, comparative information Current plan and estimated annual cost Lowest cost plan and its estimated annual cost Potential savings from the lowest-cost plan Also, Medicare website address & Plan Finder printout on all plans Behaviorally sensitive / favor action Challenge confirmation bias By showing available savings Alternative default The lowest cost plan Deadline

24 Distribution of Plan Costs Medications as of Share of sample Least expensive plan $623 $1417 $2580 $3556 Median plan $1053 $2010 $3383 $4789 Average cost of $937 $1883 $3142 $4279 plan selected Letters showed substantial savings Seniors not choosing lowest cost plans. Many lower-priced options available.

25 Prediction 1: Effect on Choices 28% of intervention group switched plans vs. 17% of comparison group Difference.12** Regression adjusted.12** 31% of intervention group switches were to lowest cost plan versus 12% of comparison group switches

26 Prediction 2: Effect on Predicted Savings Predicted savings Difference in cost between 2007 plan & 2006 plan Zero if don t change plans Computed by the Medicare Plan Finder Predicted savings (all) Intervention: $132 Comparison $16 Difference $116** Regression-adjusted $104** Regression-adjusted ln (Y 07 /Y 06 )=.063** Predicted savings, potentially affected Regression-adjusted $230**

27 Predicted Savings, Potentially Affected Upper bound Everyone who switched in comparison would also have switched in intervention Effect = 104/(.28)=$371 Lower bound Everyone who switched in comparison would not have switched in intervention Effect = 104/( ) = $230

28 Effect by Potential Savings Robustness check Mechanism of action was information, not suggestion Switching Probability Lower bound impact on estimated cost Comparison Intervention Difference Dollars (1) (2) (3) A. Dollar potential savings $ (44) > $ ~ * (106)

29 Prediction 3: Concentration of Effect in Out-of-Pocket Spending (Not Premiums) Total savings: $104 Decreases in premiums: $11 Decreases in OOP: $92 Out-of-pocket costs as share of total Costs in 2006 plan: 81% Potential savings from changing to least expensive plan: 80% Intervention effect: 91%

30 Welfare Effects: Realized Savings in 2007 Predicted savings (all) Intervention: $77 Comparison $-6 Difference $83** Regression-adjusted $83** Regression-adjusted ln (Y 07 /Y 06 )=.043** Predicted savings, potentially affected Regression-adjusted $193**

31 Welfare Effects: Experience in 2007 Switch rates in open enrollment 2008: no difference Plan ratings / satisfaction in 2007: no difference Reported issues with access in 2007: no difference Medicare quality ratings: no difference

32 Welfare Effects: Preference for 2007 Plan in Blinded Comparison Show two plans costs, market share, Medicare quality ratings Do not give plan names Ask Which plan would you choose? Chose 2007 plan over Least Expensive Plan Intervention 48 % (N=105) Comparison 37 % (N=102) P value.11 (almost statistically significant) Chose 2007 plan over 2006 plan Intervention 84 % (N=37) Comparison 39 % (N=18) P value.000 (highly statistically significant) Evidence on realized savings, experiences, & hypothetical choices consistent with improvements in well-being

33 Cost Benefit Analysis Average realized savings in first year= $83 Potential for persistence Respondent time: < one hour Interviewer time + materials ~=$40 Medicare and other organizations with drug hx could provide similar interventions at lower cost. Per person savings likely lower too. Population Intervention

34 Summary Randomized experiment of information re Medicare drug plan choice Accessible, publicly available information Presented in a format favorable to action In intervention presenting least expensive plan Switching increased by 11 percentage points Changes in predicted costs consistent with price misperception Surprising that our small intervention had such a substantial impact Consistent with choice errors & role for information environment

35 Potential Explanations (Standard Economics) Effect not big compensating differentials But, no differences in quality Intervention not small transactions costs But, easy to get plan information from Medicare And, effects not concentrated among the dissatisfied. Intervention not small more credible source Source was University Hospital (not Medicare) But we cited Medicare as source of information These explanations not trivial But seem small relative to predicted savings of $193

36 Preferred Explanation: Misperception In initial enrollment, misperception due to complexity & unfamiliarity In open enrollment for 2007 Misperception persists Compounded with confirmation bias & inertia Behaviorally sensitive intervention Alters perceptions Challenges confirmation bias Provides default Uses deadline

37 Policy Significance In Part D, additional efforts to distribute personalized, comparative information can affect choices & potentially improve welfare Publicize the Medicare help line Facilitate private information market?? Content important. Incorporate quality information? In general, policy-makers & other stakeholders should pay attention to design of choice environment Potential to address some concerns about choice via choice environment Rather than through regulation, incentives, education

38 Practical Significance Small, smart interventions can affect behavior in big ways Opportunities to alter environment to promote smart choices Interventions which reflect back personalized health information in actionable manner have potential.. Theories and models that rely exclusively on rational actors & costs may overlook key drivers on behavior and key levers/opportunities

39 Concluding Thought A randomized experiment is the strongest design to test a theory or intervention Significant interest among academic researchers in partnering with private sector organizations to conduct randomized experiments Potential benefits to you Collaborate with top researchers on program design and evaluation Be recognized as a thought leader in field

40 Thank you. Marian V. Wrobel Research Associate, ideas42, Harvard University Phone: Website: Ideas42.org