After Evidence-Based Medicine: How Health Research Took Up the Logic Model — Research Problem and Study Protocol for a Sentence-Level Analysis of PubMed Abstracts, 2016–2025

Logic models have spread through evidence-based policymaking in Japanese central and local government, including prefectural health plans, but how health research itself uses the tool has not been measured. A PubMed search on 4 November 2026 found 2,099 records that mention logic models in the title, abstract, or author keywords. Relative to all PubMed records, mentions rose from 0.02 per 10,000 in 1991 to 1.40 in 2025. The first program-evaluation use appeared in 1993, and about three quarters of the records were published in 2016–2025. A check against Scopus shows that earlier records use the term in engineering and logic, and that the evaluation sense, like the term evidence-based medicine, dates from the early 1990s. This paper sets out the research problem and the protocol for a sentence-level analysis of that decade. Every abstract sentence that mentions a logic model will be coded with fixed rules for how it positions the model: constructed, referred to, applied, evaluated, reported, used as a container for findings, acting as a subject, listed, or discussed in general. Stated purposes, program stages, and related frameworks are coded as well. The rules will be validated against blind manual coding of 200 sentences, with Cohen’s kappa of 0.60 as the threshold, and the periods 2016–2020 and 2021–2025 will be compared. Version 1.1. Version 1.0 of this text is the study protocol deposited with the scripts before the main run (https://doi.org/10.5281/zenodo.23933473). Version 1.1 adds an abstract, a comparison with Scopus for logic models and evidence-based medicine, and a declaration on the use of generative AI.