Can Machine Learning aid in finding key factors to improve the Finnish healthcare system?
Published: 23.03.2023 / Publication / Blog
Finland is in the process of change in our health care system. The Nordic well-fare system is challenged in Finland, for instance, due to difficulties in attracting nurses, changing demographics in Finland, and a general pressure to reduce costs in the whole public sector. This poses severe challenges for the entire healthcare sector. Can Machine Learning (i.e., the subfield of Artificial Intelligence, which focuses on having a machine imitate intelligent human behavior) be used to understand relationships between different critical properties of our healthcare system? Yes, it can! An excellent example of how this can be done is found in a scientific paper by Hu et al. (2020), where the authors investigated nurses' willingness to report errors in a specific geographical area of the US.
Background and analysis of the study by HU et al. (2020)
Hu et al. showed through questionnaires and several Machine Learning techniques what affects the willingness of nurses to report medication errors committed. A study shows that only 5 % of the errors are reported to their managers, medical doctors, or other relevant personnel in the hospital (Cohen et al., 2003). This study was done with 38 variables and 328 nurses' answers to the questionnaire. These nurses belonged to different departments and hospitals in a limited geographical area in the US. The main questions concerned willingness to report errors concerning other critical variables like trust in the organization, how long they have worked in the organization, how long they have worked in the current unit, their view of their own managers, and other factors. With these 328 answers, a pretty comprehensive analysis was made. The analysis consisted of two major parts. The first part uses the algorithm Extreme Learning Machines, aka ELM (Huang, 2014; Hu et al., 2019; Cambria, 2013), to find the top prediction variables. ELM is a high-speed neural network that can run many times and be retrained with little computational effort, thus suitable for finding the best variables for prediction in an approach called "wrapper selection with elm" (for more information, see Hu et al. 2020). When these predictors were found, another Machine Learning technique was used to reduce the dimensionality of the data to find patterns in the nurses’ answers. This process made it possible to identify and cluster certain variables such that you could see relationships between, for instance, willingness to report errors and trust in the organization.
How can similar techniques be used in Finland?
The method and approach of Hu et al. (2020) were novel and, from a research setting point of view, very interesting. For instance, this approach could also be used in Finland to answer questions like "How likely are you to quit your job?" or "Is the salary even partly satisfactory?". Moreover, when these questions are answered, they could be linked similarly to other critical values (not only the willingness to report medication errors but, for instance, the willingness to do a good job in general or other interesting factors). In fact, the possibilities seem endless, but the approach used by Hu et al. (2020) and other Machine Learning methods are like to bring exciting and valuable information in a situation where decisions cannot go wrong. We would claim that the reformation of our health care system in Finland cannot afford to fail, or we are facing a considerable challenge, both humanitarian and financial, to correct it.
Kaj-Mikael Björk, Principle Lecturer, Arcada
Leonardo Espinosa-Leal, Principle Lecturer, Arcada
Cambria E, et al. 2013. Extreme learning machines [trends controversies]. IEEE Intell. Syst. 28(6). pp. 30–59.
Cohen H, Robinson ES, Mandrack M. (2003). Getting to the root of medication errors: survey results. Nursing 2003,33(9) pp.36–45.
Hu R, Farag A, Björk K-M, Lendasse A, (2020). Using machine learning to identify top predictors for nurses' willingness to report medication errors, Array, 8, pp. 100049, https://doi.org/10.1016/j.array.2020.100049. External link
Hu R, Ratner K, Ratner E, Miche Y, Bjork K-M, Lendasse A. (2019). ELM-SOM+: a continuous mapping for visualization. Neurocomputing. 365 pp.147–56.
Huang G-B. (2014). An insight into extreme learning machines: random neurons, random features and kernels. Cognitive Computation, 6(3) pp. 376–90.