Approaches to addressing missing and incomplete information in primary care electronic health records: a rapid scoping review of statistical, machine learning, and information recovery methods (Preprint)

Authors:
Preprint
Publisher website
Google Scholar
Abstract
BACKGROUND Primary care electronic health records (EHRs) are increasingly used in observational and clinical research. However, EHR data are frequently affected by information gaps arising from missing values, incomplete coding, under-ascertainment of clinical conditions, information recorded only within unstructured text, and under-representation of population groups. These information gaps can compromise the validity, interpretability, and generalisability of research findings. OBJECTIVE To identify and describe approaches used in primary care EHR research to assess, characterise, and address missing and incomplete information. METHODS A rapid scoping review was conducted following the Arksey and O'Malley framework and PRISMA-ScR guidance. Systematic searches of MEDLINE, Embase, CINAHL, the Cochrane Database of Systematic Reviews, Web of Science, Scopus, and IEEE Xplore were undertaken from January 2000 to April 2026. Quantitative studies using primary care EHR data that applied statistical, computational, or algorithmic approaches to address missing or incomplete information were eligible. Data on study characteristics, information gaps addressed, analytical approaches, and reported impacts on study findings were extracted. Findings were synthesised narratively. RESULTS Seventy-five studies met the inclusion criteria. Studies addressed three broad types of information gaps: missing values in structured variables (n=40), incomplete coding or under-ascertainment (n=36), and under-representation of population groups (n=2). Missing values were primarily addressed using missingness-modelling approaches, whereas incomplete coding and under-ascertainment were predominantly addressed using information-recovery approaches such as natural language processing and classification algorithms. Only eight studies reported patterns of missingness. Methodological approaches also varied according to study purpose, with causal and epidemiological studies more commonly using missingness-modelling approaches and prediction-focused studies more frequently applying information-recovery approaches. CONCLUSIONS Different information gaps in primary care EHRs were addressed using different categories of methodological approaches. Method selection should therefore be informed by both the nature of the information gap and the purpose of the study. However, reporting of missingness patterns, assumptions regarding missingness mechanisms, and rationale for method selection remains limited. Greater methodological transparency may improve the robustness and interpretability of future primary care EHR research. CLINICALTRIAL N/A

This publication has 106 references indexed in Scilit: