Physician Beliefs and Patient Preferences: A New Look at Regional Variation in Spending

Transcription

1 1 Funding from the National Institute on Aging (T32-AG to the National Bureau of Economic Research and P01-AG to Dartmouth) and LEAP at Harvard University. Survey data collected under P01-AG and available at intensity.dartmouth.edu. Cutler, Skinner, Stern, Wennberg April 2, Physician Beliefs and Patient Preferences: A New Look at Regional Variation in Spending David Cutler, Jonathan Skinner, Ariel Dora Stern, and David Wennberg 1 April 2, 2014

2 Introduction Source: Skinner, Gottlieb & Carmichael (2011); age Cutler, Skinner, Stern, Wennberg April 2, Price-Adjusted per capita Medicare expenditures, by HRR, 2008

3 Introduction Source: Skinner, Gottlieb & Carmichael (2011); age Cutler, Skinner, Stern, Wennberg April 2, Differences in health, demand for care, or supply of care?

4 Introduction Cutler, Skinner, Stern, Wennberg April 2, Outline Explanations for regional variations Patient preferences Supplier response Augmentation of the basic model - strategic surveys Patients, primary care physicians, cardiologists Results Define cowboys and comforters in physician population Differing physician beliefs about what is right Supply, but not necessarily supplier-induced demand

5 Model Cutler, Skinner, Stern, Wennberg April 2,

6 Model Cutler, Skinner, Stern, Wennberg April 2, Framework

7 Model Cutler, Skinner, Stern, Wennberg April 2, When λ higher (lower), incentives to do less (more)

8 Model Cutler, Skinner, Stern, Wennberg April 2, When λ higher (lower), incentives to do less (more)

9 Model Cutler, Skinner, Stern, Wennberg April 2, Expand to different productivity across doctors: g(x) = αs(x)

10 Model Cutler, Skinner, Stern, Wennberg April 2, An augmented supplier-induced demand model where: L = Ψs(x) + Ω(W + πx R) φ( x x D ) ϕ( x x 0 ) saving lives increases utility positive utility from income: π is profit per procedure x, R is fixed cost, W salary deviation from patient s desired demand decreases utility deviation from organizational expectations decreases utility and... Ψs (x) = Ω π φ ϕ λ

11 Model Cutler, Skinner, Stern, Wennberg April 2, Definitions of Supplier Induced Demand 1 Any equilibrium where x > x (economist s definition) physicians concern for a patients best interest can be traded off against self interest, concern acts as a ceiling on inducement. 2 Harm: x > x (conditional on doctors beliefs about s(x)) not traditional induced demand when doctors don t think they are trading off patient s best interest.

12 Model Cutler, Skinner, Stern, Wennberg April 2, Return to augmented model; take linear approximation of FOC: supply and demand factors x i = x + Z D i + Z S i + ɛ i where... Zi D = φ M (xd i x D i ) Zi S = 1 M (ω π i + π ω i + φ(x o i xo ) + Ψs ( x) α i ) and M = Ψs ( x) + φ + ϕ

13 Model Cutler, Skinner, Stern, Wennberg April 2, Identification Approach Can t use price (change) Use strategic surveys to identify difficult-to-identify models of physician and patient behavior. Used elsewhere (e.g. Ameriks et. al., 2011) with evidence that: questions are taken seriously questions answered in a fashion that is internally consistent consistent with intuition and actual behavior

14 Cutler, Skinner, Stern, Wennberg April 2, Patient 2005 Survey of Medicare Patients N = 2,718; 65% Response Rate (of which 1,413 data points that matched geography of physician surveys) Patient level data: HRR of residence, vignettes/hypothetical questions about demand for tests and visits, demographic information

15 Cutler, Skinner, Stern, Wennberg April 2, Patient survey: demand for unneeded tests Suppose you noticed a mild but definite chest pain when walking up stairs... a) Suppose you went to your regular doctor for chest pain and your doctor did not think you needed any special tests but you could have some tests if you wanted. If the tests did not have any health risks, do you think you would probably have the tests or probably not have them? b) Suppose your doctor told you he or she did not think you needed to see a heart specialist, but you could see one if you wanted. Do you think you would probably ask to see a specialist, or probably not see a specialist?

16 Cutler, Skinner, Stern, Wennberg April 2, Patient survey: demand for unneeded tests Suppose you noticed a mild but definite chest pain when walking up stairs... a) Suppose you went to your regular doctor for chest pain and your doctor did not think you needed any special tests but you could have some tests if you wanted. If the tests did not 73% have any health risks, do you think you would probably have the tests or probably not have them? b) Suppose your doctor told you he or she did not think you needed to see a heart specialist, but you could see one if you wanted. Do you think you would probably ask to see a 56% specialist, or probably not see a specialist?

17 Cutler, Skinner, Stern, Wennberg April 2,

18 Cutler, Skinner, Stern, Wennberg April 2,

19 Cutler, Skinner, Stern, Wennberg April 2, Patient survey: end of life care preferences The next set of questions are about care a patient may receive during the last months of life. Suppose that you had a very serious illness. Imagine that no one knew exactly how long you would live, but your doctors said you almost certainly would live less than 1 year... a) If you reached the point at which you were feeling bad all the time, would you want drugs that would make you feel better, even if they might shorten your life? b1) If you needed a respirator to stay alive, and it would extend your life for a week, would you want to be put on a respirator? and/or b2) If it would extend your life for a month, would you want to be put on a respirator?

20 Cutler, Skinner, Stern, Wennberg April 2, Patient survey: end of life care preferences The next set of questions are about care a patient may receive during the last months of life. Suppose that you had a very serious illness. Imagine that no one knew exactly how long you would live, but your doctors said you almost certainly would live less than 1 year... a) If you reached the point at which you were feeling bad all the time, would you want drugs that would make you feel 48% better, even if they might shorten your life? b1) If you needed a respirator to stay alive, and it would extend your life for a week, would you want to be put on a respirator? and/or b2) If it would extend your life for a month, would you want to be put on a respirator? 8%

21 Cutler, Skinner, Stern, Wennberg April 2,

22 Cutler, Skinner, Stern, Wennberg April 2,

23 2 Both surveys restricted to 64 HRRs with sufficient physician data (N 3) Cutler, Skinner, Stern, Wennberg April 2, Physician 2005 Surveys (2) of Physicians 2 Cardiologists: N = 516, (61% Response Rate) Primary Care Physicians: N = 807, (73% Response Rate) Physician level data: HRR of practice, vignettes/hypothetical questions about supply of tests and visits, demographic information Patient and physician level linked by HRR of residence/practice

24 Cutler, Skinner, Stern, Wennberg April 2, Utilization 2005 Dartmouth Atlas of Health Care Medicare spending per beneficiary (price, age, sex, race adjusted) Risk-adjusted end-of-life spending 2-year 6-month 1-year risk and price-adjusted expenditures for Medicare enrollees following hip fracture

25 Cutler, Skinner, Stern, Wennberg April 2, Cardiologists Vignette A Think about a patient with stable angina whose symptoms and cardiac risk factors are now well controlled on current medical therapy. In general, how frequently do you schedule routine follow-up visits for a patient like this?

26 Cutler, Skinner, Stern, Wennberg April 2, The committee believes that the patient with successfully treated chronic stable angina should have a follow-up evaluation every 4-12 months. Additionally, the report notes that a more precise interval cannot be determined based on clinical evidence.

27 Cutler, Skinner, Stern, Wennberg April 2,

28 Cutler, Skinner, Stern, Wennberg April 2,

29 Congestive heart failure Cutler, Skinner, Stern, Wennberg April 2,

30 Cutler, Skinner, Stern, Wennberg April 2, NIH Definition Heart failure is a condition in which the heart can t pump enough blood to meet the body s needs. In some cases, the heart can t fill with enough blood. In other cases, the heart can t pump blood to the rest of the body with enough force. Some people have both problems....currently, heart failure has no cure. However, treatments - such as medicines and lifestyle changes - can help people who have the condition live longer and more active lives. 5.8 million people with CHF vs. 1.2 million heart attacks

31 Cutler, Skinner, Stern, Wennberg April 2, Cardiologists Vignette B A 75 year old man with severe (Class IV) congestive heart failure from ischemic heart disease, is on maximal medications and has effective disease management counseling. His symptoms did not improve after recent angioplasty and stent placement and CABG is not an option. He is uncomfortable at rest. He is noted to have frequent, asymptomatic nonsustained VT on cardiac monitoring. He has adequate health insurance to cover tests and medications. At this point, for a patient presenting like this, how often would you arrange for each of the following?

32 Cutler, Skinner, Stern, Wennberg April 2, Cardiologists Vignette B: options a - Repeat angiography b - Initiate antiarrhythmic therapy c - Recommend an Implantable Cardiac Defibrilator (ICD) d - Recommend biventricular pacemaker for cardiac resynchronization e - Initiate or continue discussions about palliative care

33 Cutler, Skinner, Stern, Wennberg April 2, Cardiologists Vignette B: Comforters a - Repeat angiography b - Initiate antiarrhythmic therapy c - Recommend an Implantable Cardiac Defibrilator (ICD) d - Recommend biventricular pacemaker for cardiac resynchronization e - Initiate or continue discussions about palliative care

34 Cutler, Skinner, Stern, Wennberg April 2,

35 Cutler, Skinner, Stern, Wennberg April 2, Cardiologists Vignette B: Cowboys a - Repeat angiography b - Initiate antiarrhythmic therapy c - Recommend an Implantable Cardiac Defibrilator (ICD) d - Recommend biventricular pacemaker for cardiac resynchronization e - Initiate or continue discussions about palliative care

36 Cutler, Skinner, Stern, Wennberg April 2, Cardiologists Vignette B: results (frequently or always/almost always) a - Repeat angiography 6% b - Initiate antiarrythmic therapy 21% c - Recommend an Implantable Cardiac Defibrilator (ICD) 64% d - Recommend biventricular pacemaker for cardiac resynchronization 46% e - Initiate or continue discussions about palliative care 30% Note: none of the interventions in red are supported by the medical literature or guidelines.

37 Cutler, Skinner, Stern, Wennberg April 2, Cardiologists Vignette C An 85 year old male patient has severe (Class IV) congestive heart failure from ischemic heart disease, is on maximal medications, and is not a candidate for coronary revascularization. He is on 2 liters per minute of supplemental oxygen at home. He presents to your office with worsening shortness of breath and difficulty sleeping due to orthopnea. Office chest x- ray confirms severe congestive heart failure. Oxygen saturation was 85% and increased to 94% on 4 liters and the patient is more comfortable. He has adequate health insurance to cover tests and medications. At this point, for a patient presenting like this, how often would you arrange for each of the following?

38 Cutler, Skinner, Stern, Wennberg April 2, Cardiologists Vignette C: options a - Allow the patient to return home on increased oxygen and diuretics b - Admit to the hospital for aggressive diuresis (not to the ICU/CCU) c - Admit to the ICU/CCU for intensive therapy and monitoring d - Place a pulmonary artery catheter for hemodynamic optimization e - Recommend biventricular pacemaker for cardiac resynchronization f - Initiate or continue discussions about palliative care

39 Cutler, Skinner, Stern, Wennberg April 2, Cardiologists Vignette C: Comforters a - Allow the patient to return home on increased oxygen and diuretics b - Admit to the hospital for aggressive diuresis (not to the ICU/CCU) c - Admit to the ICU/CCU for intensive therapy and monitoring d - Place a pulmonary artery catheter for hemodynamic optimization e - Recommend biventricular pacemaker for cardiac resynchronization f - Initiate or continue discussions about palliative care

40 Cutler, Skinner, Stern, Wennberg April 2, Vignette C: Palliative Care Not Always in the Conversation

41 Cutler, Skinner, Stern, Wennberg April 2, Cardiologists Vignette C: Cowboys a - Allow the patient to return home on increased oxygen and diuretics b - Admit to the hospital for aggressive diuresis (not to the ICU/CCU) c - Admit to the ICU/CCU for intensive therapy and monitoring d - Place a pulmonary artery catheter for hemodynamic optimization e - Recommend biventricular pacemaker for cardiac resynchronization f - Initiate or continue discussions about palliative care

42 Cutler, Skinner, Stern, Wennberg April 2, Cardiologists Vignette C: results (frequently or always/almost always) a - Allow the patient to return home on increased oxygen and diuretics 18% b - Admit to the hospital for aggressive diuresis (not to the ICU/CCU) 54% c - Admit to the ICU/CCU for intensive therapy and monitoring 17% d - Place a pulmonary artery catheter for hemodynamic optimization 2% e - Recommend biventricular pacemaker for cardiac resynchronization 15% f - Initiate or continue discussions about palliative care 43% Note: none of the interventions in red are supported by the medical literature or guidelines.

43 Cutler, Skinner, Stern, Wennberg April 2, Categorization: cowboys and comforters Comforters: Recommend palliative care frequently or always/almost always for both patients B and C Cowboys: Cardiologists responses highly correlated across patients B and C Of 28 percent (N=143) who frequently or always/almost always recommend 1 high-intensity procedure for patient C, 93 percent (N=133) frequently or always/almost always recommend 1 high-intensity intervention for patient B Use this overlap (highest in our data) to define a cowboy : recommends at least one of the three possible intensive treatments for both patients B and C at least frequently

44 Cutler, Skinner, Stern, Wennberg April 2, Related physician categories Cowboy Comforter Follow-up frequency Yes No Yes No Low 2% 10% 12% 5% 7% 12% Medium 13% 48% 61% 28% 33% 61% High 8% 19% 27% 10% 17% 27% 23% 77% 42% 58% Comforter Cowboy Yes No Yes 8% 15% 23% No 33% 44% 77% 42% 58% Conditional on being a cowboy, 35% are comforters; 65% are not While among non-cowobys, 43% are comforters; 57% are not Conditional on being a comforter, 21% are cowboys; 78% are not While among non-comforters, 26% are cowboys; 74% are not

45 Cutler, Skinner, Stern, Wennberg April 2, Related physician categories, cont.

46 Cutler, Skinner, Stern, Wennberg April 2, Patient averages Variable Mean Area Average SD Patient Variables Have unneeded tests 73% 10% See unneeded cardiologist 56% 10% Aggressive patient prefs. ratio 8% 5% Comfort patient prefs. ratio 48% 12%

47 Cutler, Skinner, Stern, Wennberg April 2, Physician averages Variable Mean Area Average SD Primary care physician variables Cowboy ratio 19% 19% Comforter ratio 44% 20% Followup low 9% 11% Followup high 4% 7% Cardiologist variables Cowboy ratio 27% 19% Comforter ratio 29% 20% Followup low 0% 3% Followup high 23% 21%

48 Cutler, Skinner, Stern, Wennberg April 2, Medicare spending averages (by HRR) Spending and utilization Mean SD 2-year EOL spending $56,219 $10,715 total per patient spending $7,837 $1,032 hip fracture patient spending $52,574 $4,996

49 Cutler, Skinner, Stern, Wennberg April 2, Log of Inpatient 2-year End-of-Life Regional Spending vs. Various Independent Variables

50 Empirical Specification Cutler, Skinner, Stern, Wennberg April 2, Overview of empirical strategy Use linear approximation of first order conditions: supply and demand factors: x i = x + Z D i + Z S i + ɛ i where... Z D i = φ M (xd i x D i ) Z S i = 1 M (ω π i + π ω i + φ(x o i xo ) + Ψs ( x) α i ) and M = Ψs ( x) + φ + ϕ Use patient preferences and physician beliefs to explain utilization

51 (1) (2) (3) (4) (5) (6) Estimates for Explaining Ln 2-Year End-of-Life Expenditures Combined Sample of PCPs and Cardiologists Cowboy Ratio, All Doctors *** *** *** *** *** (0.1626) (0.1385) (0.1173) (0.1446) (0.1221) Comforter Ratio, All Doctors ** *** ** *** ** (0.1681) (0.1109) (0.1103) (0.1065) (0.1044) Follow-Up Low, All Doctors (0.2755) (0.2849) (0.3299) (0.3215) Follow-Up High, All Doctors *** *** *** *** (0.2053) (0.1963) (0.2026) (0.1910) Have Unneeded Tests (0.2062) (0.2251) (0.3400) See Unneeded Cardiologist * * * (0.1549) (0.1679) (0.2855) Aggressive Preferences Patient Ratio (0.4607) (0.4409) (0.7526) Comfortable Preferences Patient Ratio (0.1584) (0.2015) (0.2499) N R p<0.10, * p<0.05, ** p<0.01, *** p< year EOL Spending is price, age, sex and race adjusted spending; results for 64 HRRs in which we have at least three patients and cardiologists surveyed; all regressions include a constant and controls for respondent patient race, age & heart disease history; sampling weights take into account differences in the number of observations per HRR

52 (1) (2) (3) (4) (5) (6) Estimates for Explaining Ln 2-Year End-of-Life Expenditures Combined Sample of PCPs and Cardiologists Cowboy Ratio, All Doctors *** *** *** *** *** (0.1626) (0.1385) (0.1173) (0.1446) (0.1221) Comforter Ratio, All Doctors ** *** ** *** ** (0.1681) (0.1109) (0.1103) (0.1065) (0.1044) Follow-Up Low, All Doctors (0.2755) (0.2849) (0.3299) (0.3215) Follow-Up High, All Doctors *** *** *** *** (0.2053) (0.1963) (0.2026) (0.1910) Have Unneeded Tests (0.2062) (0.2251) (0.3400) See Unneeded Cardiologist * * * (0.1549) (0.1679) (0.2855) Aggressive Preferences Patient Ratio (0.4607) (0.4409) (0.7526) Comfortable Preferences Patient Ratio (0.1584) (0.2015) (0.2499) N R p<0.10, * p<0.05, ** p<0.01, *** p< year EOL Spending is price, age, sex and race adjusted spending; results for 64 HRRs in which we have at least three patients and cardiologists surveyed; all regressions include a constant and controls for respondent patient race, age & heart disease history; sampling weights take into account differences in the number of observations per HRR

53 (1) (2) (3) (4) (5) (6) Estimates for Explaining Ln 2-Year End-of-Life Expenditures Combined Sample of PCPs and Cardiologists Cowboy Ratio, All Doctors *** *** *** *** *** (0.1626) (0.1385) (0.1173) (0.1446) (0.1221) Comforter Ratio, All Doctors ** *** ** *** ** (0.1681) (0.1109) (0.1103) (0.1065) (0.1044) Follow-Up Low, All Doctors (0.2755) (0.2849) (0.3299) (0.3215) Follow-Up High, All Doctors *** *** *** *** (0.2053) (0.1963) (0.2026) (0.1910) Have Unneeded Tests (0.2062) (0.2251) (0.3400) See Unneeded Cardiologist * * * (0.1549) (0.1679) (0.2855) Aggressive Preferences Patient Ratio (0.4607) (0.4409) (0.7526) Comfortable Preferences Patient Ratio (0.1584) (0.2015) (0.2499) N R p<0.10, * p<0.05, ** p<0.01, *** p< year EOL Spending is price, age, sex and race adjusted spending; results for 64 HRRs in which we have at least three patients and cardiologists surveyed; all regressions include a constant and controls for respondent patient race, age & heart disease history; sampling weights take into account differences in the number of observations per HRR

54 Empirical Specification Cutler, Skinner, Stern, Wennberg April 2, Can we explain physician behavior as a function of organizational and financial factors? Dependent variable: Are you a cowboy/comforter/high follow-up doctor? Independent variables: practice structure, organizational factors, financial issues, individual characteristics

55 Empirical Specification Cutler, Skinner, Stern, Wennberg April 2, What cardiologists worry about Now we d like you to think about your own cardiac catheterization recommendations. Sometimes a cardiologist will recommend cardiac catheterization for other than purely clinical reasons. During the past 12 months, how often, if ever, have each of the following led you to recommend cardiac catheterization for a patient? a - The patient expected to undergo the procedure b - Your colleagues would do so in the same situation c - You wanted to satisfy the expectations of the referring physicians d - You wanted to protect against a possible malpractice suit e - Doing so would enhance the financial stability of your practice

56 Empirical Specification Cutler, Skinner, Stern, Wennberg April 2,

57 Empirical Specification Cutler, Skinner, Stern, Wennberg April 2,

58 Empirical Specification Cutler, Skinner, Stern, Wennberg April 2,

59 Empirical Specification Cutler, Skinner, Stern, Wennberg April 2,

60 Empirical Specification Cutler, Skinner, Stern, Wennberg April 2,

61 Empirical Specification Cutler, Skinner, Stern, Wennberg April 2, Individual : Predicting Physician Types Consider Controls: age, gender, weekly patient days, board certification, etc. Reimbursement/financial incentives Practice structure Other types of organizational incentives

62 (1) (2) (3) Cowboy Comforter High Follow-Up Age *** *** (0.0013) (0.0015) (0.0012) Male * * (0.0315) (0.0370) (0.0314) Weekly Patient Days (0.0076) (0.0090) (0.0076) Board Certified * *** (0.0379) (0.0445) (0.0378) Cardiologists per 100k *** *** *** (0.0076) (0.0079) (0.0061) Cardiologist Dummy *** * (0.0363) (0.0426) (0.0361) Fraction Capitated Patients ** ** (0.0462) (0.0540) (0.0457) Fraction Medicaid Patients *** *** (0.0931) (0.1090) (0.0924) (Baseline = Solo or 2-person Practice) Single/Multi Speciality Group Practice ** *** (0.0265) (0.0310) (0.0262) Group/Staff HMO or Hospital-Based Practice *** *** (0.0429) (0.0502) (0.0426) Responds to Patient Expectations (0.0313) (0.0368) (0.0313) Responds to Colleague Expectations (0.0247) (0.0291) (0.0247) Responds to Referrer Expectations *** (0.0419) (0.0493) (0.0420) Responds to Malpractice Concerns (0.0247) (0.0290) (0.0247) N * p<0.10, ** p<0.05, *** p<0.01 All logit regressions include a constant, and HRR-level random effects as well as general physician-level controls.

63 (1) (2) (3) Cowboy Comforter High Follow-Up Age *** *** (0.0013) (0.0015) (0.0012) Male * * (0.0315) (0.0370) (0.0314) Weekly Patient Days (0.0076) (0.0090) (0.0076) Board Certified * *** (0.0379) (0.0445) (0.0378) Cardiologists per 100k *** *** *** (0.0076) (0.0079) (0.0061) Cardiologist Dummy *** * (0.0363) (0.0426) (0.0361) Fraction Capitated Patients ** ** (0.0462) (0.0540) (0.0457) Fraction Medicaid Patients *** *** (0.0931) (0.1090) (0.0924) (Baseline = Solo or 2-person Practice) Single/Multi Speciality Group Practice ** *** (0.0265) (0.0310) (0.0262) Group/Staff HMO or Hospital-Based Practice *** *** (0.0429) (0.0502) (0.0426) Responds to Patient Expectations (0.0313) (0.0368) (0.0313) Responds to Colleague Expectations (0.0247) (0.0291) (0.0247) Responds to Referrer Expectations *** (0.0419) (0.0493) (0.0420) Responds to Malpractice Concerns (0.0247) (0.0290) (0.0247) N * p<0.10, ** p<0.05, *** p<0.01 All logit regressions include a constant, and HRR-level random effects as well as general physician-level controls.

64 (1) (2) (3) Cowboy Comforter High Follow-Up Age *** *** (0.0013) (0.0015) (0.0012) Male * * (0.0315) (0.0370) (0.0314) Weekly Patient Days (0.0076) (0.0090) (0.0076) Board Certified * *** (0.0379) (0.0445) (0.0378) Cardiologists per 100k *** *** *** (0.0076) (0.0079) (0.0061) Cardiologist Dummy *** * (0.0363) (0.0426) (0.0361) Fraction Capitated Patients ** ** (0.0462) (0.0540) (0.0457) Fraction Medicaid Patients *** *** (0.0931) (0.1090) (0.0924) (Baseline = Solo or 2-person Practice) Single/Multi Speciality Group Practice ** *** (0.0265) (0.0310) (0.0262) Group/Staff HMO or Hospital-Based Practice *** *** (0.0429) (0.0502) (0.0426) Responds to Patient Expectations (0.0313) (0.0368) (0.0313) Responds to Colleague Expectations (0.0247) (0.0291) (0.0247) Responds to Referrer Expectations *** (0.0419) (0.0493) (0.0420) Responds to Malpractice Concerns (0.0247) (0.0290) (0.0247) N * p<0.10, ** p<0.05, *** p<0.01 All logit regressions include a constant, and HRR-level random effects as well as general physician-level controls.

65 (1) (2) (3) Cowboy Comforter High Follow-Up Age *** *** (0.0013) (0.0015) (0.0012) Male * * (0.0315) (0.0370) (0.0314) Weekly Patient Days (0.0076) (0.0090) (0.0076) Board Certified * *** (0.0379) (0.0445) (0.0378) Cardiologists per 100k *** *** *** (0.0076) (0.0079) (0.0061) Cardiologist Dummy *** * (0.0363) (0.0426) (0.0361) Fraction Capitated Patients ** ** (0.0462) (0.0540) (0.0457) Fraction Medicaid Patients *** *** (0.0931) (0.1090) (0.0924) (Baseline = Solo or 2-person Practice) Single/Multi Speciality Group Practice ** *** (0.0265) (0.0310) (0.0262) Group/Staff HMO or Hospital-Based Practice *** *** (0.0429) (0.0502) (0.0426) Responds to Patient Expectations (0.0313) (0.0368) (0.0313) Responds to Colleague Expectations (0.0247) (0.0291) (0.0247) Responds to Referrer Expectations *** (0.0419) (0.0493) (0.0420) Responds to Malpractice Concerns (0.0247) (0.0290) (0.0247) N * p<0.10, ** p<0.05, *** p<0.01 All logit regressions include a constant, and HRR-level random effects as well as general physician-level controls.

66 Conclusion Cutler, Skinner, Stern, Wennberg April 2, Conclusions Patient preferences have a limited role in explaining Medicare spending variations Prices and income matter for provider behavior, but organizational factors and especially beliefs are most important in explaining equilibrium utilization differences Don t know if some physicians really are better (as in Chandra and Staiger (2007)) High-utilization areas appear to rely on beliefs that are not supported by clinical evidence; waste estimated to be >35% in end-of-life spending, >15% in Medicare overall

67 Conclusion Cutler, Skinner, Stern, Wennberg April 2, Thanks Thank you