Episode-Based Price Indexes: Plans and Progress Ana Aizcorbe Nicole Nestoriak BEA Advisory Committee Meeting May 4, 2007.

Episode-Based Price Indexes:Plans and Progress

Ana AizcorbeNicole Nestoriak

BEA Advisory Committee MeetingMay 4, 2007

2www.bea.gov

There is a growing consensus that price indexes for health care should be based on treatment episodes.

Previous work for specific diseases shows that the issue is numerically important:

Heart attacks (Cutler et. al.)

Cataract (Shapiro/Wilcox)

Depression (Berndt et. al.)

National Academies Panel issued a recommendation for the construction of episode-based indexes.

3www.bea.gov

Preliminary work at BEA confirms the numerical importance of the issue in a dataset that includes a comprehensive list of diseases.

0

1

2

3

4

5

All Diseases

Comparison of Price Indexes for Medical Care, 2001-2003

(compound annual growth rates)

0

1

2

3

4

5

PrescriptionDrugs

Office Visits HospitalOutpatient

HospitalInpatient

Provider-Based

Disease-Based

Source: A. Aizcorbe and N. Nestoriak, “Using Commercially-Defined Episodes of Illness for the Measurement of Health Accounts: A Progress Report,” Paper presented at NBER/CRIW Summer Institute, July 2006

4www.bea.gov

Outline of talk

Provide a progress report on our ongoing work to construct these indexes for a health satellite account.

Provide an outline of next steps

Close talk with two important conceptual issues surrounding episode-based price indexes.

Progress Report

6www.bea.gov

Groupers are one way to identify treatment episodes.

Episode groupers are algorithms that sift through claims data and

Look at each claim and decide how the diagnoses fit together (comorbidities)

After a period of time without claims, subsequent care is a new episode (clean days)

We consider two commercial groupers (algorithms) Symmetry Health Medstat

7www.bea.gov

We apply these groupers to claims data from

Pharmetrics to explore implementation issues. Data contain a large number of claims:

40 million patients Over 70 health plans.

Our 10% sample contains $12 billion paid to providers, 22 million episodes of care (Symmetry Grouper),

and About 600 different types of episodes.

Price is the amount taken in by provider.

8www.bea.gov

What have we learned so far?

Groupers do not always yield clinically homogeneous episodes

Price indexes can be sensitive to: how expenditures are allocated over time the parameters used in the algorithm features of the underlying claims data

Bottom line: these choices need theoretical justification

9www.bea.gov

1. Assessing homogeneity of episodes using number of modes in distribution of episode lengths

0

.02

.04

.06

0 100 200 300epi_leng

episode length 2001 episode length 2002

episode length 2003

Minor Infectious Disease (006)

0

.001

.002

.003

.004

0 200 400 600 800epi_leng

episode length 2001 episode length 2002

episode length 2003

Breast Cancer (655)

•We take the presence of more than one mode as evidence of heterogeneity.

•This may not present problems if the distributions are stable.

10www.bea.gov

2. Sensitivity of price indexes to expenditure allocation

Price per day vs price per episode

0.8

0.9

1

1.1

1.2

1.3

1.4

2001 2002 2003

Price per day Price per episode

•Fluctuations in the average episode length accounts for measured differences in price/day vs. price/episode.

•We believe these fluctuations are an artifact of the data.

11www.bea.gov

3. Sensitivity of price indexes to choice of grouper

Quarterly Price Deflators, 2001:1-2003:4

0.6

0.7

0.8

0.9

1

1.1

2001:1 2002:1 2003:1

Symmetry Medstat

Both the trends and contours differ.

Odd seasonal pattern in the Medstat episodes

•Price per day declines with length of episode

•Symmetry’s definition for chronic episodes

Fisher Indexes of price per day

12www.bea.gov

4. Sensitivity of price indexes to underlying data

Type of Index Pharmetrics Ingenix

Episode-based 2.6% 4.9%

Treatment-based [3.6% – 4.3%] 6.8%

Fisher Indexes of price per episode

________________________________

Price growth is higher in the Ingenix data...

…one can not appeal to “law of large numbers.”

13www.bea.gov

Current thinking

One cannot take literal read of data or episodes.

Key is to find a way to use what is available to create a data set that is: representative of all US patients, with clinically homogeneous episodes, and a sensible way to deal with chronic

episodes

Next steps We’ve constructed standard errors for price indexes that we will use to address:

Homogeneity issue: Is there a tradeoff between granularity and precision of the price indexes?

Sensitivity of price indexes: To what extent are differences in price indexes “statistically significant?”

We will devise a plan for extracting a representative sample from the Pharmetrics database.

Looking ahead, we would like to construct price indexes for other patients as well (i.e., Medicare and Medicaid).

Two Important Conceptual Issues

16www.bea.gov

Issue 1. Reweighting treatment-based indexes to obtain price indexes by disease does not address the substitution issue.

Assume:

•no change in the costs of therapy or drug treatment

•Treatment-based indexes will show no price change regardless of weights (Berndt).

•But, substitution of drugs for therapy reduces the cost of treating depression.

•An episode-based index captures this price decline.

17www.bea.gov

Issue 2. Qualifications for episode-based price indexes.

Episode-based price indexes capture declines in cost from the substitution across treatment types, provided the disease is defined correctly.

These indexes implicitly assume that quality (the impact on health from treatment) is constant.

To the extent that quality is increasing, disease-based indexes provide an upper bound

on quality-adjusted price change.