Electronic vs manual approaches to identify patients from the EHR for cancer clinical trials–what’s feasible

Abstract

Objective: Electronic health records (EHRs) offer a platform to identify patients for clinical trials. We compared an electronic approach combining natural language processing (NLP) with query capabilities of Data Warehouse using structured and unstructured information against manual review to assess feasibility in identifying subjects for a breast cancer trial. Materials and methods: Study included women with new metastatic, ER-positive, HER2-negative breast cancer, treated with letrozole monotherapy between January 2012 and December 2015 who did not receive prior systemic therapy for advanced disease. Concordance between approaches was assessed using Cohen’s kappa statistic. Results: 826 breast cancer cases were identified; 83 were truly metastatic, ER-positive, HER2-negative. Manual review identified 77 (93%) patients compared to 51 (61%) by NLP. Cases missed by electronic approach were due to inaccessibility of data and variability in physician documentation. Cohen’s kappa was 0.36 (95% CI 0.27-0.45), indicating fair agreement. The final eligible study population included 30 women, 28 (93%) identified by manual review and 17 (57%) electronically. The electronic approach markedly reduced time spent: 44 vs. 280 hours. Discussion: While electronic approach offers substantial cost and time savings, variability in physician documentation and inaccessibility of unstructured key data requires manual support to redress misclassification and exclusion of patients by electronic review. Conclusion: Key common data elements need to be developed and incorporated into the clinical care process. Technological innovations are needed to lessen the pain of structured field entry. Whereas the ultimate cost savings can be substantial, there needs to be upfront investment to obtain such efficiencies.