Stig Ejdrup Andersen

Background Machine learning can operationalize the rich and complex data in electronic patient records for exploratory pharmacovigilance endeavours. Objective The objective of this review is to identify applications of machine learning and big patient data in exploratory pharmacovigilance. Methods We searched PubMed and Embase and included original articles with an exploratory pharmacovigilance purpose, focusing on medicinal interventions and reporting the use of machine learning in electronic patient records with ≥1000 patients collected after market entry. Findings Of 2557 studies screened, seven were included. Those covered six countries and were published between 2015 and 2021. The most prominent machine learning methods were random forests, logistic regressions, and support vector machines. Two studies used artificial neural networks or naive Bayes classifiers. One study used formal concept analysis for association mining, and another used temporal difference learning. Five studies compared several methods against each other. The numbers of patients in most data sets were in the order of thousands; two studies used what can more reasonably be considered big data with >1 000 000 patients records. Conclusion Despite years of great aspirations for combining machine learning and clinical data for exploratory pharmacovigilance, only few studies still seem to deliver somewhat on these expectations.

Purpose While the beneficial effects of medications are numerous, drug–drug interactions may lead to adverse drug reactions that are preventable causes of morbidity and mortality. Our goal was to quantify the prevalence of potential drug–drug interactions in drug prescriptions at Danish hospitals, estimate the risk of adverse outcomes associated with discouraged drug combinations, and highlight the patient types (defined by the primary diagnosis of the admission) that appear to be more affected. Methods This cross‐sectional (descriptive part) and cohort study (adverse outcomes part) used hospital electronic health records from two Danish regions (~2.5 million people) from January 2008 through June 2016. We included all inpatients receiving two or more medications during their admission and considered concomitant prescriptions of potentially interacting drugs as per the Danish Drug Interaction Database. We measured the prevalence of potential drug–drug interactions in general and discouraged drug pairs in particular during admissions and associations with adverse outcomes: post‐discharge all‐cause mortality rate, readmission rate and length‐of‐stay. Results Among 2 886 227 hospital admissions (945 475 patients; median age 62 years [IQR: 41–74]; 54% female; median number of drugs 7 [IQR: 4–11]), patients in 1 836 170 admissions were exposed to at least one potential drug–drug interaction (659 525 patients; median age 65 years [IQR: 49–77]; 54% female; median number of drugs 9 [IQR: 6–13]) and in 27 605 admissions to a discouraged drug pair (18 192 patients; median age 68 years [IQR: 58–77]; female 46%; median number of drugs 16 [IQR: 11–22]). Meropenem‐valproic acid (HR: 1.5, 95% CI: 1.1–1.9), domperidone‐fluconazole (HR: 2.5, 95% CI: 2.1–3.1), imipramine‐terbinafine (HR: 3.8, 95% CI: 1.2–12), agomelatine‐ciprofloxacin (HR: 2.6, 95% CI: 1.3–5.5), clarithromycin‐quetiapine (HR: 1.7, 95% CI: 1.1–2.7) and piroxicam‐warfarin (HR: 3.4, 95% CI: 1–11.4) were associated with elevated mortality. Confidence interval bounds of pairs associated with readmission were close to 1; length‐of‐stay results were inconclusive. Conclusions Well‐described potential drug–drug interactions are still missed and alerts at point of prescription may reduce the risk of harming patients; prescribing clinicians should be alert when using strong inhibitor/inducer drugs (i.e. clarithromycin, valproic acid, terbinafine) and prevalent anticoagulants (i.e. warfarin and non‐steroidal anti‐inflammatory drugs ‐ NSAIDs) due to their great potential for dangerous interactions. The most prominent CYP isoenzyme involved in mortality and readmission rates was 3A4.