The Level and Risk of Out-of-Pocket Health Care Spending Michael D. - - PowerPoint PPT Presentation

The Level and Risk of Out-of-Pocket Health Care Spending Michael D. Hurd RAND and NBER Susann Rohwedder RAND

Financial support from the Social Security Administration via a research grant to the Michigan Retirement Research Center is gratefully acknowledged. Additional support from NIA for data development is gratefully acknowledged

Of many issues connected with health care spending here are two:

• 1. Health care spending and economic preparation for

retirement Center for Retirement Research at Boston College

• Nearly 45 percent of households are "at risk" of not

having enough to maintain their living standards in retirement.

• Explicitly including health care in the Index drives up the

share of households ‘at risk’ to 61 percent. Yet, Hurd and Rohwedder: actual spending in a life-cycle context find much higher rates of adequate economic preparation

Part of difference may be estimations and projections of out-

• f-pocket spending health care costs

• 2. Estimations of models of economic behavior that account

for risk (dynamic programming models) Entire distribution of costs, not just mean or median Depending on method, estimations quite sensitive to large

• utliers

Health and Retirement Study is only adequate vehicle for such studies Want to examine HRS measures of out-of-pocket spending on health care: mean etc but also “outliers,” …large enough to influence mean Do not include spending for health insurance which is predictable (just another life-cycle expense).

Out-of-pocket spending (1000s), mean and percentile points, 2003$. HRS 2004. N = 9089 mean p50 p90 p95 p99 max 65-69 2.1 0.7 3.8 5.9 25.6 420.0 70-74 2.4 0.8 4.5 7.2 28.8 218.3 75-79 2.6 0.9 4.2 6.4 30.1 268.3 80-84 3.0 1.0 5.2 11.3 36.4 180.8 85+ 4.4 1.0 9.6 24.5 60.6 127.2 Total 2.7 0.8 4.8 8.1 36.0 420.0 Can these large values be valid?

Two-year average out-of-pocket spending by households between years t-2 and t and income and wealth (1000s 2003$). HRS 2004. N= 9089 top 1% top 10 obs Mean Median Mean Median OOP spending 115.9 90.2 477.3 434.2 Household income 38.7 24.2 48.9 13.6 Household wealth at t-2 407.1 145 282.9 113.9 Household wealth at t 383.8 134.9 328.8 78.3

Two-year average out-of-pocket spending by households between years t-2 and t and income and wealth top 1% top 10 obs Mean Median Mean Median OOP spending 115.9 90.2 477.3 434.2 Household income 38.7 24.2 48.9 13.6 Household wealth at t-2 407.1 145 282.9 113.9 Household wealth at t 383.8 134.9 328.8 78.3 2*income – ΔW 100.7 58.5 51.9 62.8 Large values cannot be correct: spending could not be financed.

Issues to be investigated

• Imputation for missing values
• Spending on drugs…hard to measure
• Comparison with other surveys
• Moment-in-time spending versus panel spending

Role of imputation HRS method 1. Ask whether particular service used (doctor visit). 2. Ask about out-of-pocket spending for that service. 3. If nonresponse with respect to amount spent, amount is bracketed 4. Amount imputed using covariates and bracket Is imputation responsible for outliers?

HRS rate of imputation is 23% in middle of spending

• distribution. (Any imputation among number of

spending categories)(

Two-year out-of-pocket spending, income and wealth of households by top 1% of spenders by whether any spending was imputed in $1000s. Age 65 or older N spending income wealth t-1 wealth t Means no imputations 213 120.4 46.6 537.3 537.0 some imputations 244 90.2 32.1 333.7 282.3 Medians no imputations 213 86.9 34.0 215.1 236.3 some imputations 244 77.3 19.5 87.8 86.3 Large outliers whether imputations or not. Imputations associated with low income and wealth

Spending on drugs particularly difficult to measure. Episodic for some One-year recall best… but recall error Very regular for others Monthly recall best HRS question (hopes to do both) On average, about how much have you paid

• ut-of-pocket per month for these

prescriptions in the last two years?

At median most of spending is from drugs

Median annual spending, total and excluding drugs, HRS 2004

200 400 600 800 1000 1200 65-69 70-74 75-79 80-84 85+ Total total exclude drugs

Smaller differences as percent, but large differences in absolute value

95th percentile of spending

5000 10000 15000 20000 25000 30000 65-69 70-74 75-79 80-84 85+ Total total exclude drugs

99th percentile of spending

5000 10000 15000 20000 25000 30000 65-69 70-74 75-79 80-84 85+ Total total exclude drugs

Out-of-pocket spending on drugs responsible for (some) large values

Compare with other data Medical Expenditure Panel Survey, but noninstitutionalized population only Medicare Current Beneficiary Survey, but age 65 or

• lder only

However, both focus on health and health care spending Use greater survey effort Comparison with HRS

Mean HRS about twice as large

Spending by non-nursing home populaton. 65-69

5000 10000 15000 20000 25000 mean p90 p95 p99 HRS MEPS MCBS

70-74

5000 10000 15000 20000 25000 30000 mean p90 p95 p99 HRS MEPS MCBS

75-79

5000 10000 15000 20000 25000 mean p90 p95 p99 HRS MEPS MCBS

80-84

5000 10000 15000 20000 25000 30000 35000 mean p90 p95 p99 HRS MEPS MCBS

85+

5000 10000 15000 20000 25000 30000 mean p90 p95 p99 HRS MEPS MCBS

HRS consistently higher values than either MEPS or

• MCBS. Comparison between MEPS and MCBS shows

no particular pattern

Non-drug spending by non-nursing home population. 65-69

1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 mean p90 p95 p99 HRS MEPS MCBS

70-74

1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 mean p90 p95 p99 HRS MEPS MCBS

75-79

2000 4000 6000 8000 10000 12000 mean p90 p95 p99 HRS MEPS MCBS

80-84

2000 4000 6000 8000 10000 12000 mean p90 p95 p99 HRS MEPS MCBS

85+

2000 4000 6000 8000 10000 12000 14000 16000 18000 20000 mean p90 p95 p99 HRS MEPS MCBS

Now HRS and MCBS mostly in agreement and higher than MEPS. Same populations? Samples recruited in different ways MEPS much smaller sample sizes N = 367 age 85 or older MCBS: N= 1600 (non-nursing home) HRS : N = 1200 (non-nursing home)

But spending by nursing home population is important Cannot use MEPS

Annual spending including nursing home residents

1000 2000 3000 4000 5000 6000 65-69 70-74 75-79 80-84 85+ Total 65-69 70-74 75-79 80-84 85+ Total HRS MCBS

Medians Means

Total spending including nursing home population. Age 65-69

5000 10000 15000 20000 25000 30000 p50 p90 p95 p99 HRS MCBS

70-74

5000 10000 15000 20000 25000 30000 35000 p50 p90 p95 p99 HRS MCBS

75-79

5000 10000 15000 20000 25000 30000 35000 p50 p90 p95 p99 HRS MCBS

80-84

5000 10000 15000 20000 25000 30000 35000 40000 p50 p90 p95 p99 HRS MCBS

85 or older

10000 20000 30000 40000 50000 60000 70000 p50 p90 p95 p99 HRS MCBS

HRS higher than MEPS at the upper percentiles in lower age groups. Leads to differences in means but not medians. At higher ages spending on nursing homes relatively more important.

Spending over time Transitions between spending quartiles Use spending transitions to get a qualitative idea of stability of spending

Percent distribution of spending in wave t conditional on spending quartile in wave t-1, HRS waves 1998-2004. Single persons quartile in wave t quartile in wave t-1 lowest 2nd 3rd highest all lowest 58.8 20.8 11.8 8.7 100.0 2nd 19.9 41.2 24.7 14.1 100.0 3rd 9.3 23.9 39.9 26.9 100.0 highest 8.6 12.3 24.7 54.5 100.0

Percent distribution of spending in wave t conditional

• n spending quartile in wave t-1, HRS waves 1998-
• 2004. Married persons

quartile in wave t quartile in wave t-1 lowest 2nd 3rd highest all lowest 47.1 26.4 15.6 11.0 100.0 2nd 22.2 33.0 26.1 18.8 100.0 3rd 13.3 24.1 34.3 28.2 100.0 highest 10.9 17.4 26.5 45.1 100.0 Moderate stability at lower and upper quartiles

Conclusions Compared with MEPS and MCBS spending on drugs

• verstated in HRS.
• median, mean and upper percentiles
• reason can be traced to survey methods

Other types of spending consistent with those surveys HRS modified questions about drugs in 2006, apparently reducing values.

Conclusions (cont.) Impact of spending needs to be put in life-cycle perspective: some wave-to-wave persistence but not

• complete. Life-cycle risk

Conclusions (cont.) In future data

• For over 65
• link to Part D data. But would lack data on

Medicare Advantage plans or employer provided insurance)

• high end spending reduced by Part D insurance

(but not all persons covered)

• Under 65
• Further improvements in HRS questionnaire

Conclusions (cont.) Use of past data

• Pay attention to outliers
• Bayesian shrinking
• But not necessarily case that MEPS or MCBS is

accurate at high end

How large is risk? Smaller than observed in household data sets. How large is median? $500-$1000 per year (in 2003) not including health insurance premiums. Any way to get reach $300,000 in out-of-pocket spending in retirement?

The end