Objective: To assess the technical feasibility of generating drug indication - ICD-10 linkage suggestions as a basis for developing a tool to support pharmacist medication-order review.
Subjects and methods: This cross-sectional descriptive study analysed secondary textual data comprising 15,844 ICD-10 codes and indications for 1163 active substances. Two clinical pharmacists independently assessed 150 randomly selected suggestions.
Results: Configuration 6 generated 503 suggestions for 260/597 substances (43.5%). Of 501 suggestions comparable with the preliminary reference, 65.3% matched the full code and 79.3% matched the three-character ICD-10 category. Compared with the initial method, the number of suggestions decreased from 880 to 503, while substance coverage increased from 33.7% to 43.5%. Retaining Vietnamese diacritics increased full-code agreement from 0% to 61.4%. For 112/150 evaluated indication- ICD-10 code pairs (74.7%; 95% CI, 67.2-81%), both pharmacists selected level 1. Exact agreement was 78% (95% CI: 70.7-83.9%); unweighted Kappa was 0.176 (95% bootstrap CI: 0.015-0.341).
Conclusion: Configuration 6 generated a suggestion list, most sampled pairs received a level 1 rating from both pharmacists. The low Kappa indicates that rating criteria and clinical context require further standardisation. The study did not assess completeness of indication coverage or clinical effectiveness.