Identifying clinical features in primary care electronic health record studies: methods for codelist development.

Electronic health records are widely used in primary care research. However, defining clinical features from these records requires accurate codelist development. Prior research has shown that codelist creation often lacks transparency and reproducibility. This gap motivated the need for standardized methods to ensure scientific rigor. Existing approaches may not fully address the complexity of clinical coding. Researchers have explored various coding strategies, but consensus remains limited. No prior work had resolved how to systematically define and validate clinical codes. This paper introduces a reproducible framework to address these limitations.

The study aimed to develop a standardized method for clinical codelist creation. It focused on improving transparency and replicability in EHR research. The authors sought to provide a three-stage process for defining clinical features. They wanted to ensure that codelists could be shared and replicated. The motivation came from the need for consistent and auditable coding practices. The study also aimed to demonstrate the method using a real-world example. Shortness of breath was selected as the clinical feature for illustration. The goal was to show how the method could be applied in primary care EHR studies.

The first step involved defining the clinical feature using reliable resources. The second step used statistical software to search for all relevant codes. The third step employed a modified Delphi process with primary care practitioners. This process included generating an 'uncertainty' variable for sensitivity analysis. The authors demonstrated the method using shortness of breath as an example. They applied the method to a dataset from the Clinical Practice Research Datalink. The process included identifying candidate codes and reaching consensus on their relevance. The final step involved documenting the syntax for replication and sharing.

A codelist for shortness of breath was developed using the described method. Out of 78 candidate codes, 29 were excluded as inappropriate. Complete agreement was reached for 44 (90%) of the remaining codes. Partial disagreement occurred for 5 (10%) of the codes. The codelist was used to analyze a cohort of 28,216 patients with lung cancer. A total of 13,091 episodes of shortness of breath were identified. Sensitivity analysis showed that uncertain codes were rarely used in practice. The method enabled the creation of a reproducible and auditable codelist.

The authors propose that standardized codelist development improves EHR research quality. Their method includes a three-stage process for defining and validating clinical codes. The use of a modified Delphi process with primary care practitioners is recommended. The inclusion of an 'uncertainty' variable allows for sensitivity analysis. Published codelists should be badged by quality and include detailed methods. The method supports transparency and replication in primary care EHR studies. The authors suggest that this approach future-proofs findings and enhances auditability. They emphasize the importance of documenting definitions, syntax, and code categorization.