BACKGROUND:
Electronic laboratory reporting (ELR) has improved the timeliness and sensitivity of communicable disease reporting but it has also increased the need for improved patient deduplication. To improve efficiency, the web-based Michigan Disease Surveillance System (MDSS) implemented an automatic patient merge function. The objectives of this analysis are to compare the proportion of automatically merged patients to manually merged patients and to identify reasons duplicate patient referrals were not automatically merged.METHODS:
Data were obtained for all reportable communicable disease laboratory referrals received by the MDSS from August 1, 2016 – August 31, 2016. Referrals for new patient records were excluded. Variables included patient first name, last name, sex, birthdate, phone number, and address. To be automatically merged, the incoming referral and existing record must match exactly by first name, last name, sex, date of birth, and either phone number or address.RESULTS: In one month, MDSS received 19,193 referrals that were merged with an existing patient record; 14,893 (78%) were ELRs. The majority (11,652; 61%) were manually merged; 7,541 (39%) were automatically merged. Incoming and existing data were compared for 6,562 manually merged referrals. Among these, 3,060 (47%) were not automatically merged because of a single variable discrepancy. In 1,997 (30%) instances, the existing record lacked a phone number or address to compare with the incoming referral. In 437 (7%) instances, either the first or last name in the referral did not match the existing record. Name discrepancies were most commonly due to small (≤2 character) misspellings (35%), inmate identification numbers being included in name (29%), middle name or initial included in first name (13%), and punctuation differences (6%).
CONCLUSIONS:
With ELR expansion and the anticipated implementation of electronic case reporting, the need for efficient patient deduplication is intensifying. Adjustments to the MDSS automatic merge criteria would decrease the burden of manual review. Recommendations to consider include reassessing how missing values are compared, allowing small (≤2 character) name differences when all other variables match, and implementing additional data quality measures that prevent numerals (eg. Inmate ID) from being entered in the name field.