overview of general prevalence estimates
DESCRIPTION
CDC Site Visit at Emory CHD Surveillance Cooperative Agreement Prevalence Estimates September 25, 2013 Carol Hogue, PhD, MPH Cheryl Raskind-Hood, MS, MPH. Overview of General Prevalence Estimates. 5 counties* within Metropolitan Atlanta Population in 2010, ages 18 – 64: ~ 3 million** - PowerPoint PPT PresentationTRANSCRIPT
CDC Site Visit at EmoryCHD Surveillance
Cooperative Agreement
Prevalence Estimates
September 25, 2013
Carol Hogue, PhD, MPHCheryl Raskind-Hood, MS, MPH
Overview of General Prevalence Estimates
• 5 counties* within Metropolitan Atlanta– Population in 2010, ages 18 – 64: ~ 3 million**– 51% non-Hispanic White, 40% non-Hispanic Black– Living in same house > 1 year ~ 80%
• Resident who meets case definition with at least one encounter , 2008-2010– Hospital– Outpatient clinic– Medicaid (ResDAC)
• Presumed alive on January 1, 2010 and in age range (11-64)– GA Vital Records (Mortality 2008 – 2010)
* The five counties are within the Metropolitan Atlanta area and include Fulton, Cobb, Dekalb, Gwinnett, & Clayton.** U.S. Census 2010.
Map of Five Counties
Planned Strategy1: Population-Based Estimate
MACDP Not Linked to Data Sources includes GA Vital Records
• Hospitals and Clinics Included– Emory Healthcare: Emory HC (EHC)– Children’s Healthcare of Atlanta: CHOA– Sibley Heart Center: Sibley– Grady Health– Selected private providers:
• Pediatric Cardiology Services (PCS)
• To-date: Deduplicated CHOA & Sibley (2008-2010)– Merged by Last Name, First Name, DOB, Sex & County – Retained all CHD Dx Codes– ~4700 after adolescent merge– Few county differences
• Use of LexisNexis for validation
Data Sources for Population-based Prevalence
Variables Used to Link Various Data Sources
VARIABLES MACDP MEDICAIDCLINICAL
DBsVITAL
RECORDS
SSN - encrypted* x -
DOB x x x x
GENDER x x x x
RACE x x x xPROVIDER ID - x x x
Dx CODES ICD-9 ICD-9 ICD-9 ICD-10 NAME(LAST & FIRST)
withheld x x x
* May receive SSN to Bene_ID crosswalk from ResDAC
DATA SOURCES
Linkage for Population-based CHD Adolescents & Adults
GA Death Records
2008-2010
Emory HC2008-2010
Constrain to specific vars and remove
PHI
Sibley 2008-2018
Grady Health2008-2010
ResDAC2008-2009
Select Private Providers
2008-2018
Adult CHD Dataset
Step 4. Upload limited dataset to CDC via SAMS.
Step 3. Create limited datasets with specific core vars & PHI removed.
Step 2. Create datasets & identify survivors & non-survivors.
Adolescent& Adult
CHD Datasets
Step 1. Link GA death records to sources to determine survivors. Deduplicate.
CHOA2008-2018 Adolescent
CHD Dataset
Adolescent& Adult
CHD Datasets
Adolescent& Adult
CHD Datasets
Deduplicating & Matching Protocol
• Sibley• CHOA• Emory Clinics*• Emory Hospitals• Medicaid (ResDAC)• Others as Available
* Including archived Heart Failure database at Emory HC.
Used as example for all datasets
Deduplicated patients
SIBLEY
Flowchart: Deduplicating Adolescent EHC Clinic Datasets
Sibley extracted Name, DOB, Gender
& CHD ICD-9s for 2008-2010
CHOA 1 CHOA 2
Cross-walked county in by zip;
Concatenated multiple rows into a single row
for each patient
Sibley extracted Name, DOB, Gender
& CHD ICD-9s for 2008-2010
CHOA extracted Name, DOB, Gender
& CHD ICD-9s for 1/1/2008-4/30/2009
Concatenated multiple rows into a single row
for each patient
De-duplicate patients
CHOA extracted Name, DOB, Gender
& CHD ICD-9s for 5/1/2009-12/31/2010
Concatenated multiple rows into a single row
for each patient
De-duplicate patients
Merge & de-duplicate both datasets
Merge & de-duplicate patients
Five-county EstimatesPossible Over-estimate Because:
• Multiple data sources may not be completely deduplicated
• Name changes• Inaccurate county of residence coded (e.g.,
Grady free clinics serve only Fulton & DeKalb residents)
• Imprecision of some ICD-9CM codes / diagnoses
• Persons living with CHDs may not access the healthcare system in time frame (2008-2010) (clinical data supports this)
• You would think that • those with mild to moderate CHD defects would
be less likely to access care, BUT not so• Those with unsuccessful surgeries would be less
likely to access care, BUT according to clinical data, the complex severe defects are equally at risk (25%)
• Imprecision of some ICD9-CM codes / diagnoses
• Missing the uninsured
Five-county EstimatesPossible Under-estimate Because:
Population-based EstimateAdolescent & Adult CHD in Five-Counties:
Issues and Limitations
1. Not an issue -> Moved out of state before 20082. Data errors leading to non-match, e.g.,
incomplete dates, misspelled names, changed names (less an issue if MACDP is included in matching process)
3. Didn’t seek care in Georgia during 2008-20104. Received care outside of data sources
Should We Add 41 Counties?
Map of 46 counties
Augusta
Atlanta
Should We Add 41 Counties?
• Arguments in favor of including them– Additional data collection effort is minimal– Clinical Penetration in these counties are high
for our Network Consortium – Healthcare settings for the 5-county area also
serve the vast majority of the 46-county area– ResDAC will be reported for all of Georgia– Population of rural and semi-urban areas in the
South may have different CHD prevalence (e.g., because of migration closer to care)
– Different demographics for urban vs. rural outside and inside metro Atlanta limits
Should We Add 41 Counties?
• Arguments against including them– 46-county prevalence estimates will be more
affected by missing cases that are seen outside of Georgia
• ResDAC only for GA – GA residents can’t go out of state (out-of-pocket cost)
• Some specialized care in border areas in Alabama, Florida, and South Carolina
And the Answer Is. . . .?
Yes
Planned Strategy2:MACDP-based Estimate
Link MACDP to Data Sources includes NDI & GA Vital Records
Presumed ALIVE,
But NOT Found in
Lexis/Nexis
Seen in clinics & living within
5 counties
Seen in clinicsbut living outside
5 counties
Seen in MCAID within 5 counties
Seen in MCAID but living outside
the 5 counties
Seen in clinics & MCAID
& living within 5 counties
Seen in clinics & MCAID
but living outside 5 counties
CL
INIC
S
2008-2010
ME
DIC
AID
2008-2009
Relationship of Presumed MACDP Survivors to Clinical & Medicaid Datasets
Note. Encounters occurred between 2008-2010 for EHC & between 2008-2009 for Medicaid; 2010 Medicaid data available soon.
Summary of Data Sources
MACDP Medicaid Clinic records
GA death certificates
Maternal mortality
review
Ages in 2010
Born 1979-1999 (ages 11-31) Y Y Y Y Y
Born 1968-1978 (ages 32-42) Y Y Y Y Y
Born 1946-1967 (ages 43-64) Y Y Y
NO
YES
YES
NO
Remove from MACDP Survivors
Remove from MACDP Survivors
PRESUMED SURVIVORS
Original MACDP Dataset
Flowchart: To Obtain Dataset for Presumed MACDP Survivors
FOUND INGA DC ?
FOUND IN NDI ?
(To be completed by CDC contractors?)
Emory HC Deduplicated Dataset
FLOWCHART: Bringing Merged Sibley & CHOA Adolescent Dataset into CDC
for MACDP MATCHING
MACDP 1967-1999
Exact MatchWith Last Name, First Name, DOB,
& Gender
Add to repository matched file
& deduplicate
YES
NO
Matches with Additional
Approaches *
Add to repository matched file
& deduplicateYES
NO
Residual MACDP temporary file
* Protocol for additional matching approaches to be discussed with CDC & consortium partners.
DATA SOURCES
Linkage with MACDP: CHD Adolescents & Adults
GA Death Records
2008-2010
Emory HC2008-2010
Constrain to specific vars and remove
PHI
Sibley 2008-2018
Grady Health2008-2010
ResDAC2008-2009
Select Private Providers
2008-2018
MACDPwith or
without NDI
Linked MACDP Adult
CHD Dataset
Step 4. Upload limited dataset to CDC via SAMS.
Step 3. Create limited datasets with specific core vars & PHI removed.
Step 2. Create datasets & identify survivors & non-survivors.
Adolescent& Adult
CHD Datasets
Step 1. Link MACDP with GA death records & NDI to determine survivors. Merge with adult & adolescent sources. Deduplicate.
CHOA2008-2018
Linked MACDP Adolescent CHD Dataset
Adolescent& Adult
CHD Datasets
Adolescent& Adult
CHD Datasets
Remove PHI from MACDP not found in other sources.
Assumptions & Issues of MACDP Matching Process:
Multiple Reasons for Lack of Matching
1. Moved out of state before 20082. Data errors leading to non-match, e.g., incomplete
dates, misspelled names, changed names3. Didn’t seek care in Georgia during 2008-20104. Received care outside of data sources 5. Underestimate number of uninsured patients who
may have sought care elsewhere•Fulton & DeKalb covered through Grady Health•Emory HC does not turn anyone away
Note. 50-64 year olds will not be matched as they were born before the MACDP began collecting data.
What Does the Cooperative Agreement Gain with MACDP in the Emory Project?
• This is NOT a population-based prevalence estimate. It’s an estimate of MACDP survivors who are living in the 5-county area sometime in 2008-2010 and who accessed the healthcare system during that period.
• Can also be estimate of MACDP survivors who are living in GA (but outside the 5-county area) who accessed care through consortium or Medicaid.– And an estimate of survivors who didn’t “hit the system”
in Georgia.– Can also serve to help develop a ‘correction factor’ for
those who are still alive and who did not hit the HC system
Benefits of Linking MACDP Data to the Population-based Estimate
A population-based estimate allows for:• Estimating age-specific prevalence in five-county
area• Modeling the dispersion by age to estimate
MACDP movement out of area and out of the state of Georgia
• Modeling missed care between the MACDP and the population-based estimate by differences in prevalence rates by severity of diagnosis(there will be MACDP cases that did not match & some will be due to lack of accessing care during the 2008-2010 period; hypothesis - less severe cases are less likely to seek health care)
Prevalence Measures for Comparisons
• Determine age-specific prevalence (by decade) of those living inside the 5-county area sometime between 1/1/08-12/31/10
• Determine age-specific prevalence (by decade) of those living outside the 5-county area sometime between 1/1/08-12/31/10
GA Population-based• Five-County Prevalence (inside)
• 41-County Prevalence (outside)
• 46-County Prevalence
MACDP Survivors• Five-County Prevalence (inside)
• 41-County Prevalence (outside)
• 46-County Prevalence
Note. GMH & Pediatric Cardiology Services data will not be linked to MACDP per recent DUA agreement, October 2013.