Polypharmacy prescribing guidance - draft: consultation

24. Appendix L: AI and Polypharmacy

Research Questions

What is the role of AI in identifying and improving appropriate prescribing/polypharmacy?

[The context is: AI and appropriate prescribing]

What were we asked to look at and why?

Population People taking multiple medications one of which is a high-risk medication

Intervention Undertaking review of appropriateness of prescribing using AI to identify and inform decision-making for the use of medications as part of medication review.

Comparator Medication review without AI support

Outcomes Reduction in inappropriate prescribing, identification of patients and improvement in patient outcomes using AI

Study Design SR/GL

Key words

Polypharmacy and role of AI to identify high risk medication that will cause harm in an individual, medication appropriateness, medicines management. Economic impact of using AI

The importance of this topic

There is an increasing interest in the role of AI to identify patients at most risk of harm from multiple medicines and also use of AI in the decision-making on the actions to take from a medication review.

The methodological approach we took

We conducted a systematic literature search was carried out between 23/01/2024 – 24/01/2024 to identify guidelines and systematic reviews. Medline, Embase, and Cochrane databases were searched for systematic reviews. Medline, Embase, and a number of guideline specific databases and producers’ websites were searched for guidelines, including the Guidelines International Network, SIGN and NICE.

Results were limited to 2021 onwards and English language publications. Concepts used in all searches included: polypharmacy, deprescribing, medication review / artificial intelligence, machine learning, deep learning. One person sifted the titles and abstracts, and then the full papers.

In line with SIGN 50, this Topic Exploration aim was to identify guidelines, Health Technology Assessments (HTAs), Cochrane reviews and other systematic reviews that exist in the topic of interest by searching several different sources. We therefore did not perform a detailed analysis of the reports retrieved, instead conducting a high-level review.

Summary of the evidence

We retrieved one scoping review; two literature reviews; four systematic reviews (SRs); two protocols (one for a SR, and one for quantitative and qualitative work); and one study that conducted a mean-shift clustering analysis. While all were relevant to Pharmacy in general; only the following four focused to varying degrees on Polypharmacy.

• Bukhtiyarova et al. (2022) conducted a scoping review which explored the current research on the application of AI to health care administrative data, including those involving medications. One objective was to identify the main clinical and pharmacotherapeutic interests of AI-based research. Their eligibility criteria included data available in health care administrative databases, in particular prescribed medications.
• Only 6 of 343 (1.7%) studies, referred to Polypharmacy. They concluded that while the health effects of polypharmacy are of great importance for pharmacoepidemiology, they have tended to be under-represented. This paper suggests the need for more studies of AI Application in relation to Polypharmacy. It does not however illuminate the role of AI in identifying and improving appropriate prescribing/polypharmacy.
• Chalasani et al.’s (2022) literature review explored AI applications in the field of pharmacy practice; the research gaps and challenges; and highlighted the future directions for research within the field. This was relevant to a lot of areas within Pharmacy and while there was little on Polypharmacy, it made the following important point.
• Potentially inappropriate medications (PIMs) risks outweigh their benefits. Comorbid conditions and polypharmacy are prevalent among elderly patients leaving them exposed to potentially inappropriate prescribing (PIP). AI/machine learning (ML) algorithms are being used to develop predictive models for PIMs prescription. ML can improve prediction accuracy for high alert drug (HAD) medication treatment, which can reduce adverse events, and improve medication safety. ML has also supported screening and reducing errors in HAD prescriptions.
• Shirazibeheshti et al. (2023) conducted a mean-shift cluster to identify groups of patients at the highest risk of polypharmacy, focusing on the interaction between multiple medications of anticholinergic drugs, and the interaction between different medicine groups (regardless of one being an anticholinergic or not). They devised the following metrics: 1) drug–drug polypharmaceutical risk [Weighted Interaction Risk Score (WIRS)]; and 2) the risk between drugs of the anticholinergic drug group [Weighted Anticholinergic Risk Score (WARS)]. The metrics input to a mean-shift clustering (unsupervised learning) algorithm that clustered the data to different levels of polypharmaceutical risk. Groupings were based on the individual WIRS and WARS categories.
• This new method worked with big data, processing 300,000 patient records and identifying high risk groups, as few as tens of individuals in size. They concluded that it is safer and faster than manual inspection of patient records, necessitating up to date polypharmaceutical knowledge and time to run the searches.
• Walker et al. (2022) devised their protocol for qualitative and quantitative work. The Artificial Intelligence (AI) for dynamic prescribing optimisation and care integration in multimorbidity (DynAIRx) project addresses problematic polypharmacy in multimorbidity (co-existence of ≥2 long-term conditions).Their objectives are to extract how health and medications changed over time from clinical records, apply interpretable artificial intelligence (AI) approaches to predict risks of poor outcomes, and juxtapose with care records to inform Structured Medication Reviews (SMRs). Their study combines qualitative stakeholder engagement (Qualitative Phase 1, clinical needs analysis), large-scale health informatics (health data) and co-development/iterative analysis (Qualitative Phase 2). This would link data across primary, secondary and social care to visualise patient journeys, estimate risk-prediction, and prescribe dashboards to support SMRs. They aim to pilot this in primary care prescribing audit and feedback systems, and co-design future medicines optimisation decision support systems.
• The expectation is that DynAIRx will target polypharmacy in three key multimorbidity groups: a) people with mental and physical health conditions, b) people with four or more long-term conditions taking ten or more drugs and (c) older age and frailty.

Exploration summary and conclusions

Useful insights / points to note:

• Few studies have examined Polypharmacy and AI, highlighting the need for more research in this area. The role of AI in identifying and improving appropriate prescribing/polypharmacy is unclear and might be worth examining.
• R&D is developing AI/machine learning (ML) algorithms for predictive models for PIMs prescription. ML could improve prediction of high alert drug (HAD) medication treatment, thereby reducing adverse events, and improving medication safety. ML can also reduce errors in HAD prescriptions.
• Big data processing of 300,000 patient records is capable of identifying high risk groups, as few as tens of individuals in size. It is safer and faster than manual inspection of patient records.
• DynAIRx has the capability to target polypharmacy in three key multimorbidity groups: a) people with mental and physical health problems, b) people with four or more long-term conditions taking ten or more drugs and (c) older age and frailty.

Our research question was what is the role of AI in identifying and improving appropriate prescribing/polypharmacy? Our conclusions are tentative because of the lack of clarity around the evidence base – this is only a topic exploration, and not a systematic evidence synthesis. Based on the information retrieved, we can say roles include devising algorithms for predictive models for PIMs prescription and reducing errors in HAD prescriptions. There is evidence to suggest that AI could improve prediction of HAD medication treatment and reduce adverse events and improve medication safety. Lastly, it is hoped that in the future AI could target polypharmacy in people across three key multimorbidity groups: a) mental and physical health problems, b) four or more long-term conditions taking ten or more drugs and (c) older age and frailty.

Retrieved papers:

Bukhtiyarova, O., Abderrazak, A., Chiu, Y., Sparano, S., Simard, M. and Sirois, C., 2022. Major areas of interest of artificial intelligence research applied to health care administrative data: a scoping review. Frontiers in Pharmacology, 13, p.944516.

Chalasani, S.H., Syed, J., Ramesh, M., Patil, V. and Kumar, T.P., 2023. Artificial intelligence in the field of pharmacy practice: A literature review. Exploratory Research in Clinical and Social Pharmacy, 12, p.100346.

Shirazibeheshti, A., Ettefaghian, A., Khanizadeh, F., Wilson, G., Radwan, T. and Luca, C., 2023. Automated Detection of Patients at High Risk of Polypharmacy including Anticholinergic and Sedative Medications. International Journal of Environmental Research and Public Health, 20(12), p.6178.

Walker, L.E., Abuzour, A.S., Bollegala, D., Clegg, A., Gabbay, M., Griffiths, A., Kullu, C., Leeming, G., Mair, F.S., Maskell, S. and Relton, S., 2022. The DynAIRx Project Protocol: Artificial Intelligence for dynamic prescribing optimisation and care integration in multimorbidity. Journal of multimorbidity and comorbidity, 12, p.26335565221145493