Development and pilot evaluation of an automated screening tool to support prioritised medication review in primary care patients with polypharmacy

Technical Summary: Development and Pilot Evaluation of VIGIA

Problem Statement
Polypharmacy in primary care is a growing challenge associated with increased risks of unplanned hospital admissions and pharmacotherapeutic inadequacy. While structured medication reviews are a core clinical intervention, systematically reviewing the entire population of patients with polypharmacy is not feasible due to resource constraints. Existing prioritization strategies often lack validation, fail to integrate with clinical workflows, or do not automatically link admission risk with specific medication-related problems. Furthermore, manual application of explicit criteria (e.g., Beers Criteria, STOPP/START) is time-consuming and impractical for large populations. There is a recognized need for automated, interoperable tools that can prioritize patients for review based on risk and specific medication issues within routine primary care settings.

Methodology
The study employed a mixed-design approach comprising two phases to develop and pilot VIGIA (VIsor de Grupos terapéuticos potencialmente InAdecuados), a web-based clinical decision support tool.

• Phase I: Tool Development

  • Hospital Admission Risk Component: A retrospective population-based prediction study was conducted on 1,607 adults with polypharmacy (defined as ≥15 active chronic prescriptions). Patients were classified based on unplanned hospital admissions in 2020 (excluding SARS-CoV-2). Logistic regression and random forest models were evaluated using demographic and pharmacological covariates. To address class imbalance (approx. 9% admission rate), class-balancing strategies (downsampling and upsampling) were applied. The model with the best internal performance was selected.
  • Pharmacotherapeutic Inadequacy Index: A quantitative indicator was constructed based on expert consensus and literature review. It detects five predefined categories of medication-related problems: therapeutic duplications, low therapeutic value medicines, prescribing cascades, QT-interval prolongation risk, and anticholinergic burden. Each detected item is weighted by a clinical coefficient, and a global index is calculated.
  • Integration: Both components were integrated into VIGIA, a modular web-based tool interoperable with the Andalusian Health Service electronic prescribing system. It features distinct modules for pharmacists (screening, detection, reporting) and physicians (review, decision-making).

• Phase II: Pilot Implementation and Evaluation

  • The prototype was implemented in five primary care centers with a reference population of 1,079 adults with polypharmacy.
  • Pharmacists reviewed prioritized patients and generated structured reports.
  • A professional utility survey was administered to 13 healthcare professionals (7 pharmacists, 6 physicians) who used the tool, assessing usability, time savings, decision-making confidence, and interprofessional communication.

Key Results

• Model Performance: The random forest model with downsampling demonstrated the highest internal performance, achieving a mean AUC of 0.998, sensitivity of 0.969, and specificity of 0.990 across 100 replicates. Age and the number of active medicines were identified as the primary predictors of unplanned admission.
• Prioritization and Detection: In the pilot, VIGIA prioritized 419 patients (38.8% of the reference population). Among these, 363 patients (86.6%) had at least one pharmacotherapeutic inadequacy, totaling 1,532 detected problems (mean of 4.2 per patient). Therapeutic duplications were the most frequent category (41.3%).
• Professional Feedback: All 13 surveyed professionals (100%) reported that the tool simplified decision-making and expressed intent to use it again. 92.3% reported time savings and improved confidence in decision-making.
  • Subgroup nuances: While all pharmacists reported easy access, only 33.3% of physicians did, citing a lack of direct integration with the Diraya electronic health record as a barrier. Physicians unanimously (100%) found the tool useful for improving prescribing quality.

Key Contributions

1. Dual-Index Prioritization: The development of a tool that combines a machine-learning-based hospital admission risk component with a rule-based pharmacotherapeutic inadequacy index, allowing for risk-based prioritization that targets specific medication problems.
2. Operational Integration: The creation of an interoperable, web-based tool that functions within routine primary care workflows, enabling daily automated prioritization and linking high-risk medicines to health problems.
3. Feasibility Evidence: Demonstration that automated screening can identify a high proportion of patients with multiple medication-related problems in a real-world setting, supporting the feasibility of targeted interventions.

Significance and Claims
The paper claims that VIGIA is a feasible and interoperable tool that enables automated, risk-based prioritization for medication review in primary care patients with polypharmacy. Its significance lies in its ability to:

• Shift medication review from an opportunistic approach to a proactive, population-based strategy.
• Focus limited clinical pharmacy resources on patients most likely to benefit from structured review.
• Provide pharmacists and physicians with actionable, patient-specific information by integrating predictive modeling with the detection of specific medication-related problems.
• Support prevention-oriented prioritization in routine care by combining admission risk with pharmacotherapeutic inadequacy.

The authors note that while the tool shows high internal performance and professional utility, the hospital admission risk component requires external validation before it can be considered generalizable. Additionally, the pilot evaluated feasibility and perceived utility but did not measure clinical outcomes (e.g., reduction in hospital admissions), which is the focus of the broader RINGPOL project.