Small Area Primary Care Geographic Needs Assessment: Target

1
Small Area Primary Care Geographic Needs
AssessmentTarget OvercomeUnmet Needs
Disparities

• by
• CDR Steven B. Auerbach, MD, MPH
• Health Resources Services Administration, OPR,
  Region 2
• Room 3337, 26 Federal Plaza, New York, NY 10278
• T 212-264-2550 E sauerbach_at_hrsa.gov

2
What Unit of Geography

• State
• Congressional District
• County
• Minor Civil District
• Zip Code
• Census Tract

3
Rational Service Area Issues

• Geographic size
• Population size
• Population density
• Travel distance
• Travel barriers
• Travel time (census commuting time urban not
  always close)
• Urban-Rural classifications
• Actual number of persons served
• Population not be too big (not a community or
  service area)
• or too small (no data available or too small N)
• Should have some internal homogeneity (is a
  community)
• Although do need to account for travel
  distance/barriers,
• ought to have limited range in size between them
• (treat all people equally)

4
Inherent Limits Biases of County Data

• Counties are commonly used politically relevant
  geographic unit
• Generally too big to represent a community or
  service area
• Smallest unit from most NCHS other CDC data
• But There are 3,200 Counties in the U.S.
• Population range of County is from just 556 to
  9,145,219!

5
Small Population Over-Represented

• Inherent over-representation of small population
  counties (county level) more of them so by
  chance alone events more likely to be detected
  among them.
• Inherent over-representation of the people in
  small population counties (person level)
• 7 of population is over-represented by living
  in the
• 50 of counties with the smallest population.
• 50 of population is under-represented by
  living in
• the 4 of Counties that are largest in
  population.

6
Cannot Detect Sub-Populations

• Cannot detect differences among sub-populations
  within county (heterogeneity).
• Large population counties are more likely than
  small counties to have distinct internal
  sub-populations, not detected or distinguished by
  county level data
• LA County with both Bel Air South Central
• Harlem and Park Avenue are both in New York
  County
• Fact of high rates in one community gets lost
  when averaged with the low rates of another
  community, when both communities are in the same
  county.

7
County Level Analysis Inherently Gives False
ResultsSmaller-Area Analysis More Equal
Accurate

• County level data will inherently detect outliers
  among the small population (generally rural
  and/or homogenous) counties, and miss them within
  the large-population (generally urban and/or
  heterogeneous) counties.
• Smaller area analysis allows us to treat identify
  and analyze communities more equally and
  accurately.
• Will still detect small and/or homogeneous and/or
  rural counties and communities that truly have
  high rates or are otherwise outliers but also
  detect true high rate communities within large
  population/heterogeneous counties.

8
Census Tracts

• Smaller more internally homogenous than
  Counties
• But even Census Tracts not equal in population
• e.g., In NY State the range in 1990 was 400 to
  8000 people per tract.
• Based on a rational service area what can be
  covered by census worker.
• Too many and too small for most data purposes
• but can make rationale aggregates.
• Not widely known by general public, not widely
  recorded or used.
• Ideal if known - Census data
• Or knowable can geocode from address data

9
Congressional District

• 1 person 1 vote 435 Congressional districts
• More or less equal in population and relatively
  homogenous
• (better than county)
• Vary greatly in geographic size, with small
  states having only 1.
• U.S. Population 273,000,000 so 630,000
  people/district.
• Too large for most service usage, but politically
  relevant.

10
Minor Civil Divisions

• Worthless nationally, since only some states or
  counties have them.
• Depends on local political history, not
  rationale, if they exist.
• Many small population counties have them.
• Many large population counties do not have them.
• Still left with same bias as at county level.

11
Zip Code

• There are 43,000 5-digit residential postal zip
  codes.
• Smaller and inherently more internally homogenous
  than counties.
• Based on a rational service delivery area (postal
  delivery route)
• Still vary widely in population and geographic
  size
• Sometimes a good compromise geographic unit of
  analysis
• But, postal zip codes are not constant new ones
  being added.
• New ZCTA from Census
• improves ability to map
• but dont exactly correspond to postal zip code
  as reported in other data sets (e.g., numerator
  may be ZCTA and population denominator may be
  ZCTA)

12
Measures
13
Health Other Social Measures

• Many possible health and social measures
• Must distinguish both purpose unit of measure
• Must identify smallest possible geography
• Many only available national or state
• Some down to county
• Only few to smaller units of geography
• Direct Measures vs. Proxy

14
Proxy

• Does not have to directly all that we want
• Does not have to measure ALL of what we want.
• Just has to track or trend geographically with
  it.
• If actual measure is high in given area then what
  we wish to measure is high (or low) in that area.
• Appropriate for intervention if intervention is
  very general (provide general primary health
  care) then general proxy okay.

15
What Can we / Do we Measure?

• General Sociodemographics (Census)
• Age, Race, Income Poverty, Educational
  attainment
• Population density, Travel time
• Are available to smallest are
• Are not directly health measures
• Health Services
• Insurance status, Provider availability
• Antenatal care from Birth certificates
• Usage Medicare, Medicaid, Hospital discharge,
  ER visits
• Morbidity
• Notifiable diseases (could be available to small
  area States have)
• Other disease morbidity NHIS, NHANE, BRFSS -
  mostly cluster survey and not directly available
  below State or County
• Mortality Natality Vital Records are
  available to small area from States

16
Limits of Health Service Measures

• Medicare
• Medicaid
• Hospital Discharge
• ER Visit
• Neither low service nor high service indicates
  need
• Only measures those already getting care
• Systematically misses those not being served
• Variation may be due to factors other than need
• High service use due to provider-driven causes
• High service use due to easy access and perceived
  need by non-needy population (psychiatry on upper
  east side)

17
Summary Indices of Population Health

• Mortality alone incomplete insensitive
• Population health measures that combine morbidity
  and mortality simultaneously in a single number.
• Overall estimates of burden of disease
• Differences, trends, inequalities in the health
  of populations
• Inform decisions about alternate uses of health
  dollars
• Comparisons of the relative impact of specific
  illnesses and conditions, and aggregates on
  communities
• Cost effectiveness and other economic analyses.
• Variety of indices, no gold standard, different
  choices, values, uses
• Health equivalent of GDP
• HALEs HRQLs

HRSA may wish to tap into this expertise as it
further develops HPSAs MUA/MUP and other
similar indices.
18
HALE HRQL

• Population Measures descriptive summary index -
  GDP for health GNHP
• HALE/HALY Health Adjusted Life
  Expectancy/Years
• Adds morbidity measure to adjust just mortality
  (MR, LE, YPLL).
• Add value preference utility judgment weighting
  for cost-effectiveness
• HRQL Health Related Quality of Life
• QALYS
• DALY
• YHL Years of Healthy Life (NCHS, CDC)
• and many others
• Morbidity measures may come from
• Expert opinion
• Self reported health
• Include experiential perception of symptoms and
  abilities
• But
• Typically require survey data for morbidity
  such as NHIS or BRFSS not currently available
  at small area level.

19
Mortality

• Death Certificate data is available down to small
  area
• NCHS does not have it, but States do
• Critique is that mortality
• only measuring extreme outcome i.e. death
• misses people suffering from full range of
  morbidity and disability
• Partial answer is more sophisticated use of
  Mortality
• Use clever aggregate of cause and/or age
• Preventable or premature causes of death
• Appropriate for HRSA intervention of generalist
  primary care
• Identifiable extremes as substitute proxy
• e.g. /- Suicide for depression and other mental
  illness?

20
Infant Mortality

• By itself very incomplete and limited measure
• Caused by wide range of specific issues
• Does not encompass full spectrum of issues
  (Adults)

21
Suggest Aggregate ICD Codes

• Need Aggregate
• Narrowly defined conditions or groups will have
  too small N at small area
• Our intervention is generalist primary care
  (Health Centers, NHSC),
• not conditions specific, cardiology or
  oncology, etc. per se.
• Ambulatory Sensitive Conditions
• Preventable Hospitalizations
• Premature Adult Disease Deaths
• Age 25-64 All Non-Injury (001-799.9)
• e.g., may AIDS ins 30 year olds in one place
  cardiovascular somewhere else
• Could also do for age 65-74 since still premature
  death, but separately since
• Do have Medicare
• May be different geography between working age
  and retirees
• Suicide (proxy for mental health)?

22
Mortality Rates

• Crude vs. Age Adjusted, when for service need
• Standardized mortality rates
• Life expectancy
• Excess deaths, Excess mortality
• Years of Potential Life Lost (before 65 or 75)
• a WHO method weights young adult deaths (years
  of peak economic productivity) higher then
  either elderly or infancy.

23
Natality

• Birth Certificate data also available to small
  area
• NCHS does not have but States do
• Antenatal Care
• None, started 3 trimester, etc
• Measure of health service availability usage
• Birth Outcome
• Low Birth weight lt 2500 grams (singleton, gt500
  gram)
• Premature birth before 37 weeks gestational age
• Tracts with infant mortality, but more directly
  related to service and larger counts number for
  small area

24
Health Professional

• Number of primary care practitioners per
  population
• Internal Medicine
• Pediatrics
• Family Practice
• ? Physician Assistant, Nurse Practitioner
• ? GP, General Surgeon, Ob-Gyn

25
SES

• Extensive how to literature and methods from
• Social sciences
• Health
• Other Government Agencies (dont reinvent!)
• Income probably best single indicator
• individual, family, household
• Income for persons gt30 (avoids student effect)
• Below Poverty
• Health Insurance
• Educational Attainment
• Job/Occupation
• Deprivation /or Disparity

26
(No Transcript)
27
New Jersey

• 21 Counties
• 141 Primary Care Service Areas
• 566 Municipalities (Townships, Boros, Cities)
• 760 Zip Codes

28
Effect of Geographic Level

• Deaths compare County-level to
  Municipality-level
• Births compare County to Primary Care Service
  Area to Zip Code
• Identify areas with high rates for indicators at
  the small area level that are systematically
  missed at the county-level.

29
Disease Deaths Ages 25-64, NJ 1994-1998 by
County
30
Disease Deaths Ages 25-64, NJ 1994-1998 by
Municipality
31
Disease Deaths Ages 65-74, NJ 1994-1998 by
County
32
Disease Deaths Ages 65-74, NJ 1994-98 by
Municipality
33
Low Birth Weight lt2500 gr, NJ 1994-98 by County
34
Low Birth Weight lt2500 gr, NJ 1994-98 by PCSA
35
Low Birth Weight lt2500 gr NJ 1994-98 by Zip Code
36
Prenatal Care 3rd trimester or None, NJ 1994-98
by County
37
Prenatal Care 3rd trimester or None, NJ 1994-98
by PCSA
38
Prenatal Care 3rd trimester or None, NJ 1994-98
by Zip Code
39
Can See Close-Up

• Where are Indicators the highest?
• Where are Health Center Sites currently?
• Where are there apparent Unmet Needs?

40
Disease Deaths Ages 25-64 by Municipality,
Health Center Site Locations, NJ
41
Disease Deaths Ages 65-74 by Municipality,
Health Center Site Locations, NJ
42
Disease Deaths Ages 25-64 Health Center
Location, Camden County
43
County vs. Small Area

• Out of 21 Counties
• 12 counties were in top 6 (30) for at least 1
  indicator.
• 9 counties were not in top 6 (30) for any of the
  4 indicators.
• Of those 9, 5 had small areas in top 30 for all
  4 indicators
• The other 4 had small areas in top 30 for 3 of
  the 4 indicators.
• and these Counties have no Health Centers!
• e.g., Monmouth County

44
Disease Deaths 25-64, Monmouth County by MCD
45
Disease Deaths Ages 65-74, Monmouth County by MCD
46
Low Birth Weight by Zip, Monmouth County
47
Low Birth Weight by PCSA, Monmouth County
48
Prenatal Care Start 3rd Tri or None by Zip,
Monmouth Cnty
49
Prenatal Care Start 3rd Tri or None by PCSA,
Monmouth Cnty
50
Minor Civil Divisions

• Sometimes used as sub-county geography
• Like county based on history and poltics, not
  service area or route.
• Typically may have one urban township, and
  remainder of county
• Suffers same problems for analysis as does county
• Zip Code, ZCTA and PCSA much better
• E.g., Essex county Newark.

51
Disease Deaths, Essex county by MCD
All of Newark treated as single area Dont
detect anything elsewhere
52
Late or No Prenatal care, Essex County by Zip
Code
At Zip code level can detect areas within Newark,
and find high areas elsewhere in county
53
Municipalities in highest 5 Disease Deaths Ages
25-64 that are gt10 Miles from a Health Center
Site
54
NJ MCDs with highest 20 Age 25-64 Disease
Mortality Per Capita Income less than 40 No
Health Center Within 10 Miles
55
Need is Relative NJ has relatively few of the
worst health status areas compared if use overall
U.S. rates
56
Potential Uses of these Analyses

• Identify and Target of areas of greatest need and
  unmet need.
• Community Assessment working with
  Community-grantees partners
• State Assessment working with State grantees
  partners
• Targeting New Start Expansions
• Measuring program impact/effectiveness
• Targeting and Improving Outreach / Scope of
  existing programs
• Supplement to Shortage Area Designation
• Supplement to Grant Scoring
• Other?