Qualitative changes in clinical records after implementation of pharmacist-led antimicrobial stewardship program: a text mining analysis

TL;DR

This study used text mining to show how pharmacist practices changed after an antimicrobial stewardship program was introduced in a hospital without infectious disease specialists.

Contribution

The paper introduces text mining as a novel method to evaluate qualitative changes in pharmacist-led antimicrobial stewardship practices.

Findings

Text mining revealed a shift from medication-focused to infection-focused documentation after ASP implementation.

Post-ASP, strong co-occurrence relationships emerged for lab parameters like c-reactive protein and white blood cell count.

The method successfully identified changes in documentation focus and could support future multi-center studies.

Abstract

Antimicrobial stewardship programs (ASPs) are essential for optimizing antimicrobial use, but many medium-sized hospitals lack infectious disease (ID) specialists. Ward pharmacists can contribute to ASPs, but the qualitative changes in their practice patterns after ASP implementation remains unclear. We aimed to explore the potential of text mining as a novel methodology to evaluate changes in ward pharmacist antimicrobial management practices after ASP implementation in a medium-sized hospital without ID physicians.

We conducted a retrospective observational analysis of data documented in clinical records by ward pharmacists in a 313-bed community hospital from April 2014 to March 2022. The ASP team conducted weekly reviews of targeted patients, provided feedback to physicians, and shared recommendations with ward pharmacists who then collaborated to optimize antimicrobial therapy. Using Python-based text mining with standardized technical terms and compound word extraction, we performed morphological analysis, co-occurrence network analysis, and hierarchical clustering to evaluate documentation patterns before and after ASP implementation in April 2018. Co-occurrence relationships were assessed using Dice coefficients (threshold, ≥ 0.3), and communities were detected using the Louvain algorithm. Changes in documentation patterns were compared using Fisher’s exact test.

The analysis included 1,353 pre-ASP and 5,155 post-ASP clinical records containing antimicrobial-related terms, which increased from 3.12 to 7.81% of the total pharmacy records. New strong co-occurrence relationships emerged in the post-ASP period for several laboratory parameters (c-reactive protein, 0.646; estimated glomerular filtration rate, 0.594; and white blood cell count, 0.582). Network analysis revealed a shift from medication-focused communities (Medication Review, Prescription Verification, and Patient Education) to infection-focused communities (Infection Assessment, Microbiological Review, and Severe Infection Management). Although Antimicrobial Management was consistently used in both periods (odds ratio [OR]: 0.70, 95% confidence interval [CI]: 0.38–1.20), cross-tabulation analysis increased significantly in Laboratory Monitoring (OR: 1.58, 95% CI: 1.39–1.78) and Infection Assessment (OR: 2.09, 95% CI: 1.85–2.36).

This pilot application of text mining demonstrated potential as a novel methodology for objectively evaluating qualitative changes in clinical practice patterns following ASP implementation, successfully identifying shifts in pharmacists’ documentation focus and providing a foundation for future multi-center validation studies across diverse healthcare settings.

The online version contains supplementary material available at 10.1186/s40780-025-00439-0.

Figures