Adverse drug reactions (ADRs) present a significant challenge to healthcare systems, with major clinical and economic consequences. Clinical pharmacists play a vital role in preventing these reactions by conducting structured medication reviews and proposing pharmacist interventions (PIs) where necessary. However, the growing number of hospitalised patients and the associated workload necessitate more efficient strategies. Integrating a patient risk score into a clinical decision support system (CDSS) offers a potential solution. A study at Lille University Hospital evaluated the effectiveness of such a score—adapted from a Canadian model—in predicting hospital stays that would benefit from pharmacist review and intervention.
Development and Integration of the Risk Score
The PharmaScore was designed to calculate a patient's risk level based on age, number of prescribed drugs and presence of specific drug classes, such as anticoagulants or antiepileptics. The scoring ranged from 0 to 21, with higher scores indicating greater risk. This module was integrated into the PharmaClass CDSS, which supports clinical pharmacists by identifying at-risk prescriptions using pre-coded alert rules. Notably, pharmacists were blinded to the PharmaScore during the study period, ensuring unbiased medication reviews.
Must Read: Reducing Hospital Length of Stay Through Innovation
Score calculation occurred between 12 and 36 hours after hospital admission to allow for accurate data collection on drug prescriptions. Only hospital stays where patients were over 18 years old and admitted to wards attended by clinical pharmacists were included. Intravenous medications and certain prophylactic anticoagulants were excluded to reflect the home medication regimen accurately. The aim was to identify a threshold that could reliably distinguish between hospital stays likely to result in PIs and those that would not.
Assessment of Score Thresholds and Discriminative Ability
During the study period, 1717 scores were calculated, and 973 medication reviews were conducted, of which 248 led to a PI. Analysis revealed that a score threshold of 4 yielded the best balance between sensitivity and specificity, with a sensitivity of 80.6%, specificity of 44.8% and a positive predictive value (PPV) of 33.3%. At this threshold, 600 hospital stays were flagged, and 33.3% of those led to a PI. Nevertheless, this threshold failed to detect 48 stays that also triggered PIs.
Thresholds of 3 and 5 were also considered. A threshold of 3 identified more at-risk stays but had a slightly lower PPV and risked over-alerting. Conversely, a threshold of 5 improved specificity and PPV but missed more true positives. The area under the ROC curve was 0.66, indicating poor discriminative performance. This result underscores the complexity of identifying an optimal threshold that avoids alert fatigue while still capturing significant cases. Ultimately, the score can be a useful prioritisation tool rather than a definitive filter.
Clinical Implications and Limitations
The study revealed that hospital stays involving older patients and those with polymedication were more likely to result in PIs. In particular, patients over 85 years old or those taking seven or more medications often exceeded the threshold. Cardiovascular drugs and anticoagulants were the most frequent contributors to higher scores. These findings are consistent with broader literature linking age and polypharmacy with increased ADR risk.
However, the study had several limitations. Some medication data might have been missed due to the 12–36-hour delay in score calculation, or due to medications being prescribed later during the hospital stay. The score excluded intravenous drugs and certain anticoagulants, which might have influenced accuracy. Additionally, the score's design was tailored to local clinical practices and CDSS architecture, limiting its generalisability to other settings. Nevertheless, the high PI acceptance rate (94%) highlights the clinical relevance and impact of pharmacist interventions supported by this system.
Incorporating a patient risk score into a CDSS can support clinical pharmacists in identifying hospital stays likely to require intervention, helping to manage time and resources effectively. Although determining a universally optimal threshold remains challenging, a score of 4 offers a workable compromise for prioritisation. With further refinement and faster integration into pharmacists’ workflows, this tool can enhance medication safety practices. Continued collaboration between clinical pharmacy teams and CDSS developers will be essential to realise the full potential of this approach.
Source: JAMIA Open
Image Credit: iStock
