[MA 2023 20] SHOWCASE PharmacoInformatics: utilizing medical informatics to optimise pharmacotherapy outcomes

Amsterdam UMC, location AMC, Department of Medical Informatics

Proposed by: Dr. Joanna E. Klopotowska, assistant professor leading PharmacoInformatics research line [j.e.klopotowska@amsterdamumc.nl ]

References: Please visit PharmacoInformatics Lab website for more information on ongoing various research projects you could participate in: www.pharmacoinformaticslab.nl

Introduction: Unsafe medication practices are a leading cause of injury and often avoidable patient harm in health care systems across the world. In the Netherlands, approximately 1.000 people die each year as a result of inappropriate pharmacotherapy, and nearly 50.000 patients a year are admitted to a hospital because of adverse drug events (ADEs). In addition, despite major efforts to improve medication safety, the number of patients who suffer from ADEs during hospital stay has not decreased since 2008, but even increased by 58% in 2019. The most frequent reasons for ADEs provided by healthcare professionals are: the increasing complexity of patients due to multimorbidity and lack of knowledge how to prescribe safely for such complex patients.

Usually multiple medications are needed to manage multiple diseases. This is called polypharmacy. Polypharmacy is unsafe for patients because of increased risk of adverse drug events. People with multimorbidity are largely underrepresented in clinical trials which form the gold standard for prescribing guidelines. Furthermore, these guidelines are also heavily focused on managing individual diseases instead of multiple diseases at the same time. Lastly, as medical knowledge doubles every two months, it is hard to stay on top of every new piece of information for each chronic disease. Hence, when managing pharmacotherapy in people with multimorbidity, healthcare professionals and patients are often "navigating in the dark".

Reuse of observational data registered during routine patient care; i.e. real-world data (RWD), from Electronic Health Record (EHR) systems (structured and unstructured data) represents a huge potential for optimizing and personalizing pharmacotherapy in patients with multimorbidity. To make this happen, various challenges related to RWD and RWD analytics need to be addressed. The PharmacoInformatics Lab (PIL) aims to address these challenges by developing digital tools (varying from simple but useful data-visualizations and rule-based algorithms to more sophisticated statistical and machine learning models for diagnosis and prognosis and natural language processing). A pre-requisite for reliable and robust digital tools is fit-for-purpose RWD, which requires that data are FAIR and or sufficient quality. Furthermore, the digital tools need to be embedded at point-of-care to be used in clinical practices. Here, use of clinical decision support systems is investigated to delivered the knowledge learned on data to-practice.

Description of the SRP Project/Problem: Depending on students’ interest and skills, within the PharmacoInformatics Lab we can offer various SRP projects pertaining to different aspects of pharmacotherapy and medical informatics. Examples of problems where medical informatics could come to the rescue are:

Drug nephrotoxicity: Drugs often cause drug-induced kidney diseases (DIKD). However, several drugs have an understudied and/or controversial effect on kidney function (e.g. morphine, norepinephrine or metamizol). We aim to study such cases with detailed EHR data and causal inference methods that utilize machine learning. This approach may provide a much needed insights on drug nephrotoxicity.

Adverse drug events: EHR data is becoming an essential source of information to quantify the occurrence of?adverse drug events?(ADEs). However, majority of ADE cases is documented as free text in the clinical notes (unstructured), making retrieval of this information very time consuming. Application of natural language processing may improve ADE detection efficiency.

Decision support for medication: at present up to 90% of drug alerts generated by clinical decision support systems are overridden by clinicians. Most often provided reason is the lack of clinical relevancy of the drug alert. Using more sophisticated algorithms behind the alerts that can account for patient and drug characteristics, may improve clinical relevancy of drug alerts.

Research questions: Depending on the pharmacoinformatics problem chosen by the student, specific research questions will be defined together with PIL researchers. Examples of research questions pertaining to above provided examples of problems are:

Drug nephrotoxicity: What is the average causal effect of morphine on kidney function?

Adverse drug events: What is the performance of BERT-like models on ADE detection in Dutch clinical notes?

Decision support for medication: Which patient and drug attributes improve clinical relevancy of drug-drug interaction alerts?

Expected results: Depending on the pharmacoinformatics problem chosen by the student and research questions defined, the following results can be expected:

Novel average causal effect estimates of understudied and/or controversial drugs on kidney function obtained through state of the art causal inference methods that utilize machine learning.

Performance on Dutch clinical notes of medical and non-medical BERT-like models, pre-trained or not pre-trained on Dutch text, in detecting medication and disease entities, and in detecting indication and ADE relationships between medication and disease entities.

An optimized alert logic for a specific drug-drug-interaction (e.g. drug-drug interactions that may cause arrhythmias, acute kidney injury), and validation and performance results of this algorithm on clinical data (e.g data of intensive care patients).

A master thesis written in a form of a scientific article.

Time period: 7 months

Contact: Dr. Joanna E. Klopotowska, assistant professor leading PharmacoInformatics research line, Department of Medical Informatics, Amsterdam UMC, location AMC, Amsterdam; e-mail: j.e.klopotowska@amsterdamumc.nl