Statistical Analysis Plan and Brief Protocol: NLP Framework for Automated Symptom Severity Staging in Heart Failure and COPD Clinical Notes Using Ontology Integration

This deposit contains two pre-results documents for a study developing and evaluating an ontology-integrated natural language processing framework that extracts heart failure and chronic obstructive pulmonary disease symptom information from de-identified clinical narrative and assigns severity categories. The Statistical Analysis Plan specifies, before any analysis is run and before the reference corpus is opened, the estimands, analysis populations, estimators and interval methods, handling of abstention and insufficient evidence, subgroup and fairness analyses, prespecified sensitivity analyses, and the prespecification and multiplicity policy. It includes 21 dummy tables in APA format giving the complete structure of the results to be reported. The Brief Protocol is a condensed statement of the study covering design, data sources, sample, framework architecture, staging targets, reference standard, outcomes and limitations, deposited so that the analysis plan can be read and assessed on its own. Severity targets are the New York Heart Association functional classification, the GOLD ABE assessment group, and the GOLD spirometric grade where a documented FEV1 permits. Evaluation is against a 500-note reference standard annotated independently by two clinicians and adjudicated by a third, held out from all training and model selection. Reporting follows TRIPOD+AI and the CINEX guideline. No results exist at the time of deposit. Ethics approval: Kibabii University Scientific and Ethics Review Committee, KUSERC/011/26. Research licence: National Commission for Science, Technology and Innovation, NACOSTI/P/26/4194726.