Using Natural Language Processing to Characterize Early Steps in the Kidney Transplant Evaluation Process Documented...

Efforts to identify barriers and improve access to kidney transplantation in the United States are limited by a lack of population-level data on early steps in the transplant evaluation process.

We used a rule-based natural language processing (NLP) approach with clinical notes in the US Veterans Affairs Healthcare System (VA) electronic health record (EHR) and linkage with the United States Renal Data System registry to characterize sequential steps in the kidney transplant evaluation process. Adults with advanced kidney disease (estimated glomerular filtration rate ≤20 mL/min/1.73m²) from 1/1/2012-12/31/2019 who were receiving care within the VA were followed through 12/31/2021.

Among 45,174 cohort members, the median age was 71 (IQR 64, 80) years, and 97.2% were men. There was documentation of kidney transplant being mentioned as a treatment option for 46.3% of cohort members, 28.2% engaged in some type of evaluation for transplant, and 8.4% were referred to and 5.4% evaluated at a VA kidney transplant center. 6.9% of cohort members were added to the national deceased donor waitlist and 3.1% received a kidney transplant. Compared with events identified through EHR chart search and manual review by two clinicians, NLP identified events within 90 days with a precision of 0.82-0.94 and recall of 0.56-0.89.

These results illuminate the substantial proportion of patients who engage in early steps in the kidney transplant evaluation process. The work also demonstrates that NLP can accurately identify these key steps in the process as documented in patients' EHRs.