BACKGROUND: To target highly connected healthcare facilities for infection prevention, we created a network of hospitals in Tennessee. The Tennessee Hospital Discharge Data System (HDDS) provides claims data from hospitals licensed by the Tennessee Department of Health (TDH). Protected health information (PHI) in the HDDS is accessible for internal TDH use; less is available for approved research conducted by investigators outside of TDH. Using HDDS data, we evaluated two person-matching strategies to identify individuals with single and multiple hospital admissions to perform social network analyses.
METHODS: Patient date of birth (DOB), sex, and zip code of residence are available for external and internal use. Social security number (SSN) and full-name are available only for internal TDH use. In an attempt to create a unique ID for each person across facilities we generated ID variables for each hospital admission by concatenation of values from aforementioned identifiers. Individuals were identified using two methods, simple matching, for external use, and enhanced matching, for internal use. We conducted multilevel matching on DOB, sex, SSN, and subsequently DOB, sex, and full name to identify individuals with multiple and single admissions for our enhanced matching strategy. The simple strategy matched on DOB, sex, and zip Code. Both were compared to measure the concordance of matches.
RESULTS: There were 1,521,005 inpatient admissions from January 1, 2014 to September 30, 2015. Patient’s DOB, sex, zip code, and full names were available in 99.9% of admissions while SSN was only available in 89.8%. Enhanced matching identified 986,561 unique individuals including 26% with multiple admissions. Simple matching identified 934,066 unique individuals including 28% with multiple admissions. However, 25,898 (3%) individuals were falsely identified by simple matching as a different person due to a changed zip code. The two matching strategies had a moderately-high interrater agreement (weighted Kappa=0.759, 95% CI=0.758 to 0.76). Simple matching had a sensitivity of 87.4% and specificity 96.4% to identify individuals with multiple admissions (positive and negative predictive value= 91.1 % and 94.8%, respectively) compared to enhanced matching.
CONCLUSIONS: Simple matching on DOB, sex and ZIP code can identify unique individuals with few limitations. This simple strategy is useful when lacking sensitive, protected health information such as SSN and patient names.