What do machines know about our health?

Egészségügyi digitalizációját ábrázoló szimbolikus kép az emberi és a digitális kéz találkozásában mutatva. Michelangelo Ádám teremtése

The question investigated by a research group consisting of staff from the Health Economics Research Center (HECON) at the University Research and Innovation Center of Óbuda University, Eötvös Loránd University (ELTE), Budapest Corvinus University and foreign research institutions, published in a prestigious international journal, is how accurate and reliable modern machine learning methods are in providing an individual’s quality of life without actually measuring it. The researchers analyzed the anonymous data of more than 26,000 patients from 30 quality of life studies conducted in Hungary, Poland, and Slovenia using traditional and state-of-the-art machine learning methods. Most of the data was used to teach the algorithms, and a smaller part was used to verify the results. In four to five out of ten cases, the life quality was estimated with such accuracy that the differences were not noticeable from the individual’s perspective. However, in four out of ten cases, the system did not accurately predict whether the person felt completely healthy, moderately ill, or seriously ill. Although there were differences between the various methods in terms of providing more accurate estimates for healthy or sick individuals, overall, traditional and machine learning methods performed similarly. The research’s conclusion is that we still do not have enough high-quality data on people’s quality of life to use machine learning algorithms reliably and accurately. The first step is to collect good quality quality of life data from as many points in the healthcare system as possible, create patient registers, and establish a legal framework and secure infrastructure. To extract valuable information from the collected data in the future, data scientists, doctors, and health economists must work together. The digitization of healthcare and the proliferation of digital medical devices are generating an increasing amount of healthcare data worldwide. The volume and complexity of the resulting data mass are such that it can only be processed using machine learning and artificial intelligence methods. The advantage of machine learning methods over traditional data analysis techniques often represents a disadvantage. In the case of traditional data analysis methods, researchers formulate their hypotheses and obtain answers to their questions through analyses based on these assumptions. Since the question is known, the answer obtained is usually well interpreted. However, with machine learning methods, it is often not possible to understand how the result is reached, as these methods do not rely on a priori assumptions.

Compared to traditional data analysis techniques, the advantage of machine learning methods often comes with a disadvantage as well. In the application of traditional data analysis methods, researchers formulate their hypotheses and obtain answers based on these hypotheses. Since the question is known, the resulting answer is usually interpretable. However, in the case of machine learning methods, the results are often not explainable. In some areas of use, this is not a problem, for example, the quality of a photo can be visibly improved without precise knowledge of the calculations running in the background. However, it raises questions about relying on the results of analyses that are not or only partially interpretable for health-related decisions.

Today, algorithms are capable of providing more accurate diagnoses based on X-rays or photos of moles than doctors. Medical decisions are often influenced by how patients feel and their quality of life. Precise knowledge of individuals’ quality of life is fundamental in decisions related to healthcare financing.

Authors: Dr. Zsombor Zrubka, Prof. Márta Péntek, Prof. László Gulácsi, Jani Kinga, Óbuda University, Faculty of Economics, 1034 Budapest, Bécsi út 96/B

Source: https://hecon.uni-obuda.hu The original English-language article, “Predicting Patient-Level 3-Level Version of EQ-5D Index Scores From a Large International Database Using Machine Learning and Regression Methods” by Zsombor Zrubka, István Csabai, Zoltán Hermann, Dominik Golicki, Valentina Prevolnik-Rupel, Marko Ogorevc, László Gulácsi, and Márta Péntek, can be found at this link.