JAMA Netw Open. 2025 Oct 1;8(10):e2539870. doi: 10.1001/jamanetworkopen.2025.39870.
ABSTRACT
IMPORTANCE: Incomplete electronic health record (EHR) documentation may limit the effectiveness of clinical decision support (CDS) algorithms designed to identify patients eligible for hereditary cancer genetic evaluation.
OBJECTIVES: To determine whether a CDS algorithm can identify patients who meet criteria for hereditary cancer genetic evaluation when family history data are incompletely documented in the EHR, and to examine whether data missingness is associated with identification patterns across patient subgroups.
DESIGN, SETTING, AND PARTICIPANTS: This cross-sectional study analyzed EHR data extracted in December 2020 from 2 large US health care systems: University of Utah Health (UHealth) and NYU Langone Health (NYULH). Eligible patients were adults aged 25 to 60 years who visited a primary care clinic within the previous 3 years and had some EHR documentation of cancer family history. Data analysis was conducted in August 2024.
EXPOSURES: Patient demographic factors (age, sex, race and ethnicity, and language preference) and cancer family history characteristics (number of cancer history records, number of affected first- and second-degree relatives, relatives with rising mortality cancers, presence of hereditary cancer-related terms in comments, and completeness of documentation).
MAIN OUTCOMES AND MEASURES: The primary outcome was meeting at least 1 CDS algorithm criterion for genetic evaluation of hereditary cancer risk based on National Comprehensive Cancer Network guidelines. Missing data patterns were assessed using the Little missing completely at random test, with analyses conducted using complete case analysis and multiple imputation.
RESULTS: This study included 157 207 patients: 55 918 from UHealth and 101 289 from NYULH. Their mean (SD) age was 43.5 (9.8) years, and most (65.7%) were female. A total of 5607 UHealth patients (10.0%) and 10 375 NYULH patients (10.2%) met CDS criteria for genetic evaluation. At UHealth, data appeared to be missing completely at random (χ239 = 39.09; P = .47), and complete case compared with multiple imputation analyses yielded similar results. At NYULH, data were not missing completely at random (χ255 = 914.89; P < .001). Compared with multiple imputation, complete case analysis produced different association magnitudes for older age and having relatives with rising mortality cancers, suggesting bias when excluding incomplete records.
CONCLUSIONS AND RELEVANCE: In this cross-sectional study, the magnitude of the association between incomplete family history documentation and identification of patients eligible for hereditary cancer genetic evaluation depended on whether data were missing randomly or systematically. These findings suggest that health care organizations implementing CDS algorithms should assess their specific missing data patterns and consider tailored approaches to handling incomplete family history information to ensure equitable identification of all patients who could benefit from genetic evaluation services.
PMID:41148137 | DOI:10.1001/jamanetworkopen.2025.39870
