Utilizing Accurint to Inform HIV Surveillance Activities: AAAA: Assessing the Accuracy of Addresses found in Accurint Amy Bennett, MPH & Susan Buskin, MPH, PhD May 16 th, 2016
Background: Why use Accurint? Investigate people with no labs Consider 12 18 months Find relocations and deaths Verify local residence Update contact information for outreach One of multiple resources including CDC, MRA, Other databases Provider and Patient
Accurint Limitations 1. Age restrictions 2. Order of addresses may not be informative 3. Extraneous addresses may be included 4. May need to look at several PHI elements for an accurate match 5. May need to look at several addresses to find an accurate one 6. More than one person linked to a single SSN
Accurint Strengths 1. Relocations (but not date of relocation) 2. Deaths and dates of death 3. Incarcerations 4. Can find information to help de-duplicate, reassign
Will other jurisdictions have someone who has relocated to or from there? Not all states have comprehensive CD4/VL reporting (or may not have had these at time of relocation). As of July 2014, 9 US states and 6 territories did not have comprehensive reporting, including: Arizona Idaho Kansas Kentucky Nevada New Jersey Oklahoma Pennsylvania Vermont American Samoa Federated States of Micronesia Marshall Islands Northern Mariana Islands Puerto Rico Republic of Palau HIV Surveillance Supplemental Report. Vol 19 no 3. Table 12. Available at: http://www.cdc.gov/hiv/pdf/surveillance_report_vol_19_no_3.pdf, Accessed 3/9/15
Accurint Evaluation: Methods (1) People living in King County: 99 people with HIV related labs AND ADDRESSES reported Reports received January -- February 2013. Accurint reviewed February -- March 2013. 3 unmatched or unclear matches re-abstracted in June 2013. (2) People moving in to King County 93 individuals apparent in-migrants Identified February 2013 In-migrated over ~ 1 year Source was partner services interviews or case reports
Accurint Evaluation: Results (1 of 2) Verify King County residence: 99% (98/99) had a King County address in Accurint (within 3 most recent) Verify SPECIFIC address 81/99 (82%) exact match -- one of first two 88/99 (89%) exact match of ANY address OTHER addresses 7/99 (7%) also their TOP (probable current) address OOJ (6 OOS, 1 another WA county)
Accurint Evaluation: Results (2 of 2) IN-migration/ Relocations About HALF, 47 of 93 (51%) had King County as one of top two Accurint addresses. 41% had a King County address as the topmost address. 57% had a Washington State address as one of their top two addresses. 68% had a King County address within 5 years.
What, if anything, predicted finding a KC address versus not for in-migrants? Q. Length of time between First KC lab and abstraction? A.No. No correlation between length of time between first lab in King County and Accurint search date: Mean 245 days, median 252 when we did find a KC address Mean 294 days, median 254, when we didn t find KC address
What, if anything, predicted finding a KC address versus not for in-migrants? Q: Number of addresses (a marker of frequent relocation?) A1: No, if cut off is past 6 years (2008-2013) No. people Number of Accurint addresses 2008-2013 was not associated with the likelihood of finding a King County address as one of their top two addresses among relocators 15 10 5 0 1 3 5 7 9 11 13 KC Not KC 15 Number of addresses A2: Trend towards yes, if cut off is past year (2012). Of those WITHOUT KC in top 2: 49% had 4+ addresses past year relative those WITH KC in top 2: 34% had 4+ addresses past year (not statistically significant).
70 60 50 40 30 20 10 0 Could we verify prior residence for inmigrants? Q: How well did we do in finding an address at the previous state? A: 64/93 (69%) had the previous state as one of the top two Accurint addresses. 64 22 4 3 Yes got prior state Only WA Different state No match
Summary (1 of 2) Accurint did VERY well in confirming King County residence (99% -- 98/99 -- had a KC address as one of top 3) Accurint did less well in capturing presumptive inmigrants, of 93 41% had KC as the top address, 51% one of the top two addresses, and 68% as any address in the past 5 years. Length of time between investigation and identification of KC not predictive
Summary (2 of 2) Accurint limitations need to be kept in mind Lab data and surveillance data also could be inaccurate HIV cohorts have high degree of mobility Accurint s ability to produce a single best address is likely to be better thought of as a series of probable addresses
Questions?