Data Enrichment by Aggregating Patient Records From Different Hospitals-Pilot Study in Connecticut Tumor Registry Workflow

The care of patients with cancer generates vast amounts of data with enormous research potential. Unfortunately, these data are currently distributed across an array of different organizations and stored in different places. Medical discovery may be accelerated by a secure and privacy-preserving approach to linking these distributed data sets. The objective was to determine whether deidentified patient-level tumor registry data can be aggregated and enriched using a novel cryptographic technique of record tokenization called the Fischer-Hochman method.

This was a cross-sectional study of two Connecticut hospitals and the Connecticut state tumor registry. A total of 50 synthetic patient records were analyzed, followed by 133 breast and lung cancer authentic records of patients who were cared for at both hospitals. Performance parameters were defined as follows: identifier formation (patient record assigned a token), aggregation (patient record correctly merged), collision (patient record incorrectly merged), enrichment (patient record obtains a nonmissing value from the other hospital's data), and nonmissing discordance.

All patients were assigned a unique encrypted token (Cancer Research Identifier) at each of the two hospital tumor registry databases. In the synthetic data set, 98% of records were successfully aggregated, and 100% of eligible records were bidirectionally enriched. In the authentic data set, 95% of records were aggregated, and 78% of eligible records were bidirectionally enriched. There were no collision events in either data set.

Data aggregation from two hospitals that co-manage patients with cancer results in a more complete depiction of the patient journey than either individual hospital record. The Fischer-Hochman method for encrypted token generation is a promising approach to privacy-preserving record linkage, pending further evaluation.