Purpose: Although medication-related adverse events (MRAEs) in health care are vastly studied, high heterogeneity in study results complicates the interpretations of the current situation. The main objective of this study was to form an up-to-date overview of the current knowledge of the prevalence, risk factors, and surveillance of MRAEs in health care. Methods: Electronic databases (PubMed, MEDLINE, Web of Science, and Scopus) were searched with applicable search terms to collect information on medication-related adverse events. In order to obtain an up-to-date view of MRAEs, only studies published after 2000 were accepted. Results: The prevalence rates of different MRAEs vary greatly between individual studies and meta-analyses. Study setting, patient population, and detection methods play an important role in determining detection rates, which should be regarded while interpreting the results. Medication-related adverse events are more common in elderly patients and patients with lowered liver or kidney function, polypharmacy, and a large number of additional comorbidities. However, the risk of MRAEs is also significantly increased by the use of high-risk medicines but also in certain care situations. Preventing MRAEs is important as it will decrease patient mortality and morbidity but also reduce costs and functional challenges related to them. Conclusions: Medication-related adverse events are highly common and have both immediate and long-term effects to patients and healthcare systems worldwide. Conclusive solutions for prevention of all medication-related harm are impossible to create. In the future, however, the development of efficient real-time detection methods can provide significant improvements for event prevention and forecasting.
Background: Hospital-acquired medication errors (MEs) are common in health care. Although voluntary reporting is criticized for not producing reliable estimates on ME frequency, it provides valuable knowledge on errors occurring in the medication process. Objective: The purpose of this study was to analyze and determine the risks and outcomes resulting from MEs related to the TOP15 medicines in the Finnish tertiary care units from July 2016 to July 2017. Methods: The data consisting of 1,447 ME reports was organized according to ATC classification, after which TOP15 medicines involved in the reports were selected. Inductive content analysis was performed to the reports. After this, the reports were categorized by ME outcome into five categories and further analyzed accordingly. Results: The most common ME outcome in the reports was “omitted medicine” (33.9%). More than a quarter (27.1%) of ME reports were estimated to cause moderate or severe risk to the patient. When compared with each other, none of the outcome groups were more susceptible to high-risk events (p = 0.71). Of the TOP15 medicines, only Norepinephrine had significantly higher risk of being involved in high-risk events (OR 2.43, 95%CI 1.35–4.61). Conclusion: Voluntary reporting has an important role in the development of medication safety and the overall medication process within organizations. Although majority of the TOP15 medicines were involved in MEs resulting in seemingly high-risk outcomes, they were estimated to be insignificant or minor within the reporting unit. In the future, more emphasis will be needed for the assessment and analysis of the reports for more efficient, real-time detection and response to signals from health care units.
In the era of digital healthcare, electronic health records generate vast amounts of data, much of which is unstructured, and therefore, not in a usable format for conventional machine learning and artificial intelligence applications. This study investigates how to extract meaningful insights from unstructured radiology reports written in Finnish, a minority language, using machine learning techniques for text analysis. With this approach, unstructured information could be transformed into a structured format. The results of this research show that relevant information can be effectively extracted from Finnish medical reports using classification algorithms with default parameter values. For the detection of breast tumour mentions from medical texts, classifiers achieved high accuracy, almost 90%. Detection of metastasis mentions, however, proved more challenging, with the best-performing models Support Vector Machine (SVM) and logistic regression achieving an F1-score of 81%. The lower performance in metastasis detection is likely due to the more complex problem, ambiguous labeling, and the smaller dataset size. The results of classical classifiers were also compared with FinBERT, a domain-adapted Finnish BERT model. However, classical classifiers outperformed FinBERT. This highlights the challenge of medical language processing when working with minority languages. Moreover, it was noted that parameter tuning based on translated English reports did not significantly improve the detection rates, likely due to linguistic differences between the datasets. This larger translated dataset used for tuning comes from a different clinical domain and employs noticeably simpler, less nuanced language than the Finnish breast cancer reports, which are written by native Finnish-speaking medical experts. This underscores the need for localised datasets and models, particularly for minority languages with unique grammatical structures.
10 August 2022 in Diabetes Epidemiology and Management
Finland’s nationwide health registers and universal healthcare system provide comprehensive, longitudinal electronic health record data with strong potential for artificial intelligence-driven clinical research. Despite their coverage and richness, raw electronic health record datasets are not directly usable for advanced analytics due to fragmentation, heterogeneity, missingness, and inconsistencies in documentation and coding. This study describes data quality challenges encountered while preparing a breast cancer cohort (diagnosed 2012–2022) from the Wellbeing Services County of North Ostrobothnia as a preparatory step for artificial intelligence-based survival modelling. The initial dataset included 8074 patients across multiple domain-specific data files capturing diagnoses, laboratory results, pathology reports, medications, and procedural information. Following predefined cohort restrictions and preprocessing steps, 1967 patients remained for analysis. Identified challenges were grouped into four categories: 1. human-generated errors 2. decentralization-generated issues 3. time or system life-cycle–generated inconsistencies and 4. data governance/coding issues. The study adopts a qualitative research design aimed at systematically characterizing data quality challenges relevant to artificial intelligence application, and no predictive model was developed at this stage. The findings demonstrate that substantial preprocessing and data loss are often unavoidable in real-world electronic health record research and emphasize the need for standardized documentation, harmonized data structures, and closer collaboration between healthcare professionals, system developers, and data scientists to fully realize the potential of register-based electronic health record data in reliable artificial intelligence-driven research.