By Claire Manneh

The COVID-19 pandemic will go down in history as one of the largest modern-day global health problems. When COVID-19 vaccines were introduced to the public in 2021, the world saw death rates dwindle, and researchers rushed to examine variations of populations in the United States. Political affiliation wasn’t a top area of study at the time, but recently, researchers from the Yale School of Public Health and the Yale School of Management sought to understand whether there is a link between party affiliation and COVID-19 deaths.

Doctors Jacob Wallace, Jason L. Schwartz, and Paul Goldsmith-Pinkham accessed public profile data for Florida and Ohio voters and linked it to mortality data in a privacy preserving way. The results indicated a growing mortality gap during the post-COVID-19 vaccine era, with 76% more Republicans dying post-vaccine, April 2021-December 2021, than Democrats in Florida and Ohio.
The importance of linking these disparate datasets by the Yale team is three-fold:

1. Privacy Preserving Record Linkage (PPRL) software satisfies regulatory requirements. The HIPAA Privacy Rule includes two methods to de-identify data: Safe Harbor and Expert Determination. Researchers using the Safe Harbor method accomplish the de-identification of patient data by removing 18 identifiers; however, once de-identified, that dataset cannot be linked to other datasets. Expert Determination, the PPRL method, applies statistical and scientific principles to mitigate risk of identification. A unique de-identified and site-specific key for an individual, or a token, was created by the Yale team for each record and used to link across disparate datasets. The Yale team installed the PPRL software behind their firewall to de-identify patient data where patient identifiers such as names, date of birth, zip codes, and gender, were used in the creation of an irreversible, site-specific key. The benefit of using PPRL is to link to other disparate records while maintaining patient privacy.

The Yale team worked with two data types: voter profile data from Florida and Ohio, and mortality data from the Social Security Administration’s Death Master File along with obituarydata.com, a data source that includes online obituary and newspaper feeds. When the Yale researchers tokenized the data, they created site-specific keys for each dataset. Each token was encrypted so that they were site-specific and protected partners from security breaches. The tokens were also hashed, making them irreversible so they did not compromise patient privacy.

Health IT Homepage