Methods for the Dartmouth Atlas of Health Care Report: Regional and Racial Variation in Health Care among Medicare Beneficiaries.

Size: px

Start display at page:

Download "Methods for the Dartmouth Atlas of Health Care Report: Regional and Racial Variation in Health Care among Medicare Beneficiaries."

Brendan York
5 years ago
Views:

1 Methods for the Dartmouth Atlas of Health Care Report: Regional and Racial Variation in Health Care among Medicare Beneficiaries prepared for The Robert Wood Johnson Foundation Aligning Forces for Quality program December 2008 Elliot Fisher, MD MPH David Goodman, MD MS Amitabh Chandra, PhD Kristen Bronner, Editor with the assistance of the Dartmouth Atlas of Health Care working group For more information, visit

2 Document description: This technical document provides a description of the methods used for the Dartmouth Atlas Report, Regional and Racial Variation in Health Care among Medicare Beneficiaries and the accompanying measures. The document is divided into three sections. In the first section, descriptions of each measure are provided, including the numerators and denominators. The second section gives the ZIP codes defining the Aligning Forces for Quality (AF4Q) regions. The third section provides a formal description of the methods used to calculate confidence intervals. Other resources can be found at These include Frequently Asked Questions and tables containing the five measures, with their corresponding confidence intervals. A word about the precision of Atlas measures: The Atlas team has taken care to prepare measures of reasonable precision and to provide the necessary supporting information that allows users to judge the likely importance of differences in rates and proportions. The denominators differ widely, depending on the measure and the region. For some measures (e.g. ambulatory care-sensitive condition discharges), the entire fee-for-service Medicare population is the denominator. For other indicators, such as those for diabetics, the denominators are obviously much smaller. Stratification by race leads to especially small denominators in some regions. To prevent misinterpretation of the significance of differences in rates and proportions, we have suppressed any measure with a denominator of less than 200. Even with this suppression, a rate that differs across regions or race cannot be assumed to be statistically significant or clinically important. Confidence intervals (95%) are provided at to guide users interpretation. Although these are not formal inferential tests, non-overlapping confidence intervals generally mean that it is unlikely that the differences occurred by chance. The reader must be cautious in making multiple comparisons where differences may be found even if occurring by chance. It is equally important to interpret the clinical or policy significance of differences. If an observed difference in rates is not important (particularly when including the range of the confidence intervals), then a statistical test of the difference is moot.

3 Section I. Definition of Measures

4 Measure: Leg amputation Last revised: 10/6/08 Faculty: Goodman Statistical analyst/programmer: Steph; Zhao Description: Defined as average rate of leg amputations. Time period: , , Population: Medicare, non-hmo, age (mid-year), 100% sample; Part A eligibility Event: ICD-9: procedures to Rates: Adjustment: Geographies: Overall, blacks, and non-blacks Age, sex, race indirect adjustment States, AF4Q regions, HRRs, HSAs

5 Measure: Ambulatory care-sensitive condition discharge rate Last revised: 10/6/08 Faculty: Goodman Statistical analyst/programmer: Steph; Zhao Description: Ambulatory care sensitive condition discharge rate. Note this uses the Dartmouth Atlas definition and may differ from other commonly used definitions. Time period: , , Population: Medicare, non-hmo, age (mid-year), 100% sample; Part A eligibility Event: Any of the following Primary diagnosis A01 Convulsions 780.3x A02 COPD 491xx, 492xx, 494xx, 496xx, 466.0x 466.0x only w/secondary dx 491xx, 492xx, 494xx, 496xx A03 Pneumonia 481xx, 482.2x, 482.3x, 482.9x, 483xx, 485xx, 486xx Excl. secondary dx 282.6x A04 Asthma 493xx A05 CHF 428xx, , , , 518.4x Excl. sx 36.01, 36.02, 36.05, 36.1x,37.5x, or 37.7x A06 Hypertension 401.0x, 401.9x, , , Excl. sx 36.01, 36.02, 36.05, 36.1x,37.5x, or 37.7x A07 Angina 411.1x, 411.8x, 413xx Excl. sx A08 Cellulitis 681xx, 682xx, 683xx, 686xx Excl. sx ,except if 86.0x is the first and only sx code A09 Diabetes 250.0x, 250.1x, 250.2x, 250.3x, 250.8x, 250.9x A10 Gastroent x A11 Kidney/Urinary 590xx, 599.0x, 599.9x A12 Dehydration 276.5x Rates: Adjustment: Geographies: Overall, blacks, and non-blacks Age, sex, race indirect adjustment States, AF4Q regions, HRRs, HSAs

6 Measure: Diabetic receiving HgbA1C test Last revised: 10/30/08 Faculty: Goodman Statistical analyst/programmer: Steph; Weiping Description: Defined as the occurrence of at least one HgbA1c event during the measurement year to a person with a diabetes diagnosis. Diabetes diagnosis: two face-to-face encounters with different dates of service in an ambulatory setting or nonacute inpatient setting or one face-to-face encounter in an acute inpatient or emergency room setting during measurement year or prior year. Time period: , , Population: Non-HMO Medicare, age (December 31) with Parts A & B eligibility, 20% sample. Diabetes definition: ICD-9 codes: 250, 357.2, 362.0, , 648.0: DRGs 294, 295. Outpatient/nonacute inpatient CPT codes: , , , , , , , , , , , , , , 99411, 99412, 99420, 99429, Acute inpatient/emergency department CPT codes: , , , , , , , Event: CPT code: Rates: Adjustment: Geographies: Overall, blacks, and non-blacks None States, AF4Q regions, HRRs, HSAs

7 Measure: Diabetic receiving blood lipids test Last revised: 10/30/08 Faculty: Goodman Statistical analyst/programmer: Steph; Weiping Description: Defined as the occurrence of at least one blood lipids (LDL_C) test during the year to a person with a diabetes diagnosis. Diabetes diagnosis: two face-to-face encounters with different dates of service in an ambulatory setting or nonacute inpatient setting or one face-to-face encounter in an acute inpatient or emergency room setting during measurement year or prior year. Time period: , , Population: Non-HMO Medicare, age (December 31) with Parts A & B eligibility, 20% sample. Diabetes definition: ICD-9 codes: 250, 357.2, 362.0, , 648.0: DRGs 294, 295. Outpatient/nonacute inpatient CPT codes: , , , , , , , , , , , , , , 99411, 99412, 99420, 99429, Acute inpatient/emergency department CPT codes: , , , , , , , Event: CPT codes: 80061, 83715, 83716, Rates: Adjustment: Geographies: Overall, blacks, and non-blacks None States, AF4Q regions, HRRs, HSAs

8 Measure: Diabetic receiving eye examination Last revised: 10/30/08 Faculty: Goodman Statistical analyst/programmer: Steph; Weiping Description: Defined as the occurrence of an eye exam during the measurement year. This includes diabetics who had one of the following: a retinal or dilated eye exam by an eye care professional (optometrist or ophthalmologist) in the measurement year a negative retinal exam (no evidence of retinopathy: ICD-9 codes 250.5, ) by an eye care professional in the year prior to the measurement year. Diabetes diagnosis: two face-to-face encounters with different dates of service in an ambulatory setting or nonacute inpatient setting or one face-to-face encounter in an acute inpatient or emergency room setting during measurement year or prior year. Time period: , , Population: Non-HMO Medicare, age (December 31) with Parts A & B eligibility, 20% sample. Diabetes definition: ICD-9 codes: 250, 357.2, 362.0, , 648.0: DRGs 294, 295. Outpatient/nonacute inpatient CPT codes: , , , , , , , , , , , , , , 99411, 99412, 99420, 99429, Acute inpatient/emergency department CPT codes: , , , , , , , Event: ICD-9 codes: , 14.9, , 95.11, 95.12, V code: V72.0 CPT codes: 67101, 67105, , 67110, 67112, 67141, 67145, 67208, 67210, 67218, 67227, 67228, 92002, 92004, 92012, 92014, 92018, 92019, 92225, 92226, 92230, 92235, 92240, 92250, 92260, 92287, , , Rates: Adjustment: Geographies: Overall, blacks, and non-blacks None States, AF4Q regions, HRRs, HSAs

9 Section II. Dartmouth Atlas Definition of AF4Q Regions The Dartmouth Atlas uses ZIP Codes as the fundamental geographic identifier of Medicare beneficiaries. We approximated AF4Q county-based regions by linking to ZIP Codes. NOTE: three AF4Q regions represent entire states, so a ZIP code list is not given. These are Maine (518 ZIP codes), Minnesota (1,037 ZIP codes) and Wisconsin (905 ZIP codes).

10 AF4Q Name Number of ZIP Codes ZIP Codes A01 CINCINNATI, OH

11 AF4Q Name Number of ZIP Codes ZIP Codes A02 CLEVELAND, OH

12 AF4Q Name Number of ZIP Codes ZIP Codes A03 DETROIT, MI

13 AF4Q A04 Name Number of ZIP Codes ZIP Codes HUMBOLDT COUNTY, CA

14 AF4Q Name Number of ZIP Codes ZIP Codes A05 KANSAS CITY, MO

15 AF4Q Name Number of ZIP Codes ZIP Codes A07 MEMPHIS, TN

16 AF4Q Name Number of ZIP Codes ZIP Codes A09 PUGET SOUND

17 AF4Q A10 Name Number of ZIP Codes ZIP Codes WILLAMETTE VALLEY, OR

18 AF4Q Name Number of ZIP Codes ZIP Codes A11 WESTERN MICHIGAN

19 AF4Q Name Number of ZIP Codes ZIP Codes A12 WESTERN NEW YORK

20 AF4Q A14 Name Number of ZIP Codes ZIP Codes SOUTH CENTRAL PENNSYLVANIA

21 Aligning Forces for Quality Communities Community Cincinnati, OH Cleveland, OH Detroit, MI Humboldt County, CA Counties In Ohio Hamilton County Clermont County Warren County Butler County Brown County Adams County Clinton County Highland County In Kentucky Boone County Kenton County Campbell County Grant County In Indiana Dearborn County Ripley County Cuyahoga County Wayne County Oakland County Macomb County St. Clair County Livingston County Washtenaw County Monroe County Humboldt County Community Puget Sound Willamette Valley, OR Western Michigan Counties King County Kitsap County Pierce County Snohomish County Thurston County Multnomah County Washington County Marion County Polk County Yamhill County Clackamas County Linn County Benton County Lane County Mason County Lake County Osceola County Oceana County Newaygo County Mecosta County Montcalm County Muskegon County Ottawa County Kent County Ionia County Allegan County Barry County Kansas City, MO In Kansas Johnson County Wyandotte County In Missouri Jackson County Platte County Clay County Western New York Cattaraugus County Alleghany County Erie County Genesee County Niagara County Orleans County Wyoming County Chautauqua County Maine Statewide (all counties in the state) Wisconsin Statewide (all counties in the state) Memphis, TN Shelby County South Central Pennsylvania Adams County York County Minnesota Statewide (all counties in the state)

22 Hospital Service Areas in the Cincinnati, Ohio AF4Q Region Cincinnati AF4Q counties Hospital service areas Principal cities

23 Hospital Service Areas in the Cleveland, Ohio AF4Q Region Cleveland AF4Q counties Hospital service areas Principal cities

24 Hospital Service Areas in the Detroit, Michigan AF4Q Region Detroit Toledo AF4Q counties Hospital service areas Principal cities

25 Hospital Service Areas in the Humboldt County, California AF4Q Region Eureka AF4Q counties Hospital service areas Principal cities

26 Hospital Service Areas in the Kansas City, Missouri AF4Q Region Kansas City, KS Kansas City, MO AF4Q counties Hospital service areas Principal cities

27 Hospital Service Areas in the Maine AF4Q Region Augusta Portland Hospital service areas Principal cities

28 Hospital Service Areas in the Memphis, Tennessee AF4Q Region Memphis AF4Q counties Hospital service areas Principal cities

29 Hospital Service Areas in the Minnesota AF4Q Region Minnea- polis St. Paul Hospital service areas Principal cities

30 Hospital Service Areas in the Puget Sound, Washington AF4Q Region Seattle Tacoma Olympia AF4Q counties Hospital service areas Principal cities

31 Hospital Service Areas in the Willamette Valley, Oregon AF4Q Region Portland Salem AF4Q counties Hospital service areas Principal cities

32 Hospital Service Areas in the Western Michigan AF4Q Region Grand Rapids Lansing AF4Q counties Hospital service areas Principal cities

33 Hospital Service Areas in the Western New York AF4Q Region Rochester Buffalo AF4Q counties Hospital service areas Principal cities

34 Hospital Service Areas in the Wisconsin AF4Q Region Madison Milwaukee Hospital service areas Principal cities

35 Hospital Service Areas in the South Central Pennsylvania AF4Q Region York AF4Q counties Hospital service areas Principal cities

36 Section III. Calculation of Confidence Intervals

37 Calculating Confidence Intervals for Atlas Measures Center for Health Policy Research, The Dartmouth Institute for Health Policy and Clinical Practice, Dartmouth College, Hanover, NH, USA May 6, 2008 Summary. This document provides a review of the methods used to calculate variances (and standard errors) of indirectly standardized rates for count variables and continuous variables. The methods may not be suitable for all situations for which confidence intervals are sought, for example, proportions. However, counts and continuous variables are the most common variables currently included in the Dartmouth Atlas for Health Care. Calculating Confidence Intervals for Count Variables The indirect standardized rate (ISR) for region i is calculated as: ISR i = m O i E i, where m is the crude national rate (total number of events divided by the total population), O i is the observed count in region i and E i is the expected count in region i. The expected count in region i is calculated as E i = j n ij crude rate j, where j is the stratum (say, age, sex or race). The variance of the ISR i is then written var(isr i ) = m2 var(o Ei 2 i ). 1

38 Table 1 Example of 3 Region, 2 Age group indirect standardization; included are the population counts (n), observed events (O) and expected events (E). Region Age Population Observed Expected Sum i j n ij Events O ij Events E ij Events E i In this calculation, we assume that var(o i ) = O i (Poisson assumption) and E i is constant. The standard error is then se(isr i ) = m E i Oi The normal distribution-based 95% confidence interval is then ( ISR i 1.96m O i Example E i, ISR i m O i E i An example with 3 regions (i = 1, 2, 3) and 2 age groups (j = 1, 2) is used to illustrate the principle. The counts are shown in Table 1. The population crude rate is m = = = The crude rate for crude rate 1200 j=1 = = 0.1 and the crude rate for crude rate 400 j=2 = = 90 = Using these calculations as a basis, ISR i=1 = = 0.067, ISR 32.5 i=2 = = and ISR i=3 = = ) = The standard errors are then se(isr 1 ) = = 0.015, se(isr 2) = = and se(isr 3 ) = = The 95% confidence intervals are then calculated as the ISR i plus and minus 1.96 times the standard error. 2

39 Calculating Confidence Intervals for Continuous Variables Continuous variables include Atlas variables such as patient days or expenditures. Using the same notation as above, the variance for continuous variables is written var(isr i ) = m2 var(o Ei 2 i ). However, we write the pooled variance (across all regions) var(o i ) as regions var(o i ) = n i i (n i 1)s 2 i i (n i 1) ( (n1 1)s 2 1 = n i i (n i 1) + (n 2 1)s2 2 i (n i 1) + (n 2 1)s2 3 i (n i 1) +... = n i s 2 P where n i is the number of people in the region i, s 2 i is the within region variance and where the second term is often written as s 2 P, simplifying the the representation of the pooled (weighted average) variance across regions. Then, the standard error is written se(isr i ) = ms P E i ni, and the normal distribution-based 95% confidence interval is then ( ISR i 1.96ms P ni, ISR i ms ) P ni. Example E i Expenditures for Region 1 have variance 6,000 and sample size 1,000. E i ) Meanwhile, expenditures for Region 2 have variance 3,000 and sample size 2,000. The pooled estimate of the variance is then written s 2 P = 1,000 6,000 3, ,000 3,000 3,000 = 4, 000. Now, if we assume the population crude mean expenditures (m) are 10,000, while the observed expenditures in Region 1 (O 1 ) are 8,000 and the expected expenditures for Region 1 (E 1 ) are 10,000, the adjusted expenditures for Region 1 are 10,000 8,000 10,000 = 8, 000. The variance of expenditures for Region 1 is then calculated as 4,000m2 1,000 E 2 = 126, 491. Alternatively, the standard error is 356 and the 95% confidence interval is then (7302, 8698). 3

Statewide and County Demographics

2017-2018 Statewide and County Demographics Statewide and County Membership Demographics 2017 Statewide Demographics All Active Members by US Residency... 1 All Active Members by Michigan Residency...