Presentation
DH3 - AI-Driven Solution for Reducing Polypharmacy Risks Following Hospital Discharge: A Predictive Model for Patient Medication Review
SessionPoster Session 2
DescriptionPolypharmacy has become increasingly prevalent, especially among aging populations with multiple chronic conditions. While medications are crucial for managing complex diseases, the use of multiple medications can increase the risk of adverse drug effects, and drug-drug interactions, compromising patient safety. Additionally, the use of medications can present side effects which in turn are treated by other medications, situation known as prescribing cascade. Patients susceptible to polypharmacy are likely to experience consequences of multi-drug administration, exacerbating health risks, which can lead to hospital readmissions, prolonged hospital stays, or even mortality. These outcomes are especially prevalent during the transition of care from hospital to home environments, when medication schedules are often complex and less closely monitored. Considering the risks, effective intervention requires timely identification of patients at high risk of medication-related complications, however, it is common that clinicians face overwhelming workloads and cognitive overload, hindering their ability to perform thorough medication reviews. Moreover, deprescribing medications remains uncommon because medications are often prescribed by multiple specialists, each focusing on a specific aspect of the patient’s care. This can create reluctance among physicians to modify or discontinue prescriptions initiated by their peers, out of concern for overstepping professional boundaries or lacking full insight into the rationale behind another specialist’s treatment decisions.
With that, clinical decision support tools that streamline the identification of at-risk patients could significantly enhance both workflow efficiency and patient safety. This study presents the development of a predictive model aimed at flagging patients who would benefit from a comprehensive medication review, specifically targeting individuals in the post-hospital discharge period, aimed to bring attention to clinicians, and consequently reduce the risks associated with polypharmacy and medication related. In the development, the algorithm will also be useful to generate insights about the variables and relationships within the dataset, enhancing the clinical understanding of polypharmacy risks.
The primary goal of this work is to build an AI model that identifies patients during post-discharge phase that could benefit from a medication review due to medication-related risks while reducing clinical cognitive workload and enhancing decision-making within a fast-paced clinical environment. To ensure its usefulness and to ease adoption, a user-centered design approach will support the development of an explainable AI (XAI) model that provides clinicians access to more detailed information on the reasoning behind the model predictions. This deeper level of transparency is expected to support clinical decision-making and facilitate the investigation of the drivers for polypharmacy risk. Ultimately, it is expected that such systems would increase the frequency and quality of medication reviews, targeted to patients who are most vulnerable to adverse drug effects.
The AI model is built using a comprehensive longitudinal dataset with information from over 400,000 patients, tracked across multiple years. The dataset includes a wide range of variables, such as demographics (age, sex, race), social determinants of health (income level, education, housing stability), hospital admissions, primary and secondary diagnoses, and detailed medication histories. Advanced machine learning techniques are employed to identify patients at elevated risk of medication-related complications. The explainable AI model helps go beyond the prediction of high risk from polypharmacy complications, by clarifying why the predictions were made, either by identifying patterns of medication use or associated clinical factors such as recent diagnoses and hospitalizations, for example. Furthermore, early findings from the model’s predictions and XAI explanations will be shared with clinicians, allowing for the refinement of both the model and its explanations based on real-world feedback and clinical relevance.
A medication review predictive model has the potential to enhance how medication-related risks are identified post-hospital discharge and managed across healthcare settings. By flagging patients who would benefit from medication reviews in care transition, the model will not only enhance medication safety but also support clinicians in making more informed decisions under time constraints. One key benefit from a user-centered approach is that it can guide the design of the model's outputs, ensuring that they are not only accurate but also resonate with the practical experiences of healthcare professionals, facilitating their integration into clinical workflows. Another key benefit of the model is its ability to prioritize high-risk patients. Clinicians often struggle to deal with the high volume of patients and the complexity of their cases, by identifying patients who are at risk of the effects of medication-related problems and likely need a review of their prescribed medication, such systems can help by streamlining the identification of at-risk individuals and consequently reducing mental workload in clinicians.
With that, clinical decision support tools that streamline the identification of at-risk patients could significantly enhance both workflow efficiency and patient safety. This study presents the development of a predictive model aimed at flagging patients who would benefit from a comprehensive medication review, specifically targeting individuals in the post-hospital discharge period, aimed to bring attention to clinicians, and consequently reduce the risks associated with polypharmacy and medication related. In the development, the algorithm will also be useful to generate insights about the variables and relationships within the dataset, enhancing the clinical understanding of polypharmacy risks.
The primary goal of this work is to build an AI model that identifies patients during post-discharge phase that could benefit from a medication review due to medication-related risks while reducing clinical cognitive workload and enhancing decision-making within a fast-paced clinical environment. To ensure its usefulness and to ease adoption, a user-centered design approach will support the development of an explainable AI (XAI) model that provides clinicians access to more detailed information on the reasoning behind the model predictions. This deeper level of transparency is expected to support clinical decision-making and facilitate the investigation of the drivers for polypharmacy risk. Ultimately, it is expected that such systems would increase the frequency and quality of medication reviews, targeted to patients who are most vulnerable to adverse drug effects.
The AI model is built using a comprehensive longitudinal dataset with information from over 400,000 patients, tracked across multiple years. The dataset includes a wide range of variables, such as demographics (age, sex, race), social determinants of health (income level, education, housing stability), hospital admissions, primary and secondary diagnoses, and detailed medication histories. Advanced machine learning techniques are employed to identify patients at elevated risk of medication-related complications. The explainable AI model helps go beyond the prediction of high risk from polypharmacy complications, by clarifying why the predictions were made, either by identifying patterns of medication use or associated clinical factors such as recent diagnoses and hospitalizations, for example. Furthermore, early findings from the model’s predictions and XAI explanations will be shared with clinicians, allowing for the refinement of both the model and its explanations based on real-world feedback and clinical relevance.
A medication review predictive model has the potential to enhance how medication-related risks are identified post-hospital discharge and managed across healthcare settings. By flagging patients who would benefit from medication reviews in care transition, the model will not only enhance medication safety but also support clinicians in making more informed decisions under time constraints. One key benefit from a user-centered approach is that it can guide the design of the model's outputs, ensuring that they are not only accurate but also resonate with the practical experiences of healthcare professionals, facilitating their integration into clinical workflows. Another key benefit of the model is its ability to prioritize high-risk patients. Clinicians often struggle to deal with the high volume of patients and the complexity of their cases, by identifying patients who are at risk of the effects of medication-related problems and likely need a review of their prescribed medication, such systems can help by streamlining the identification of at-risk individuals and consequently reducing mental workload in clinicians.
Event Type
Poster Presentation
TimeTuesday, April 14:45pm - 6:15pm EDT
LocationFrontenac Foyer