Work Disability Risk Prediction Using Machine Learning

Forskningsoutput: Kapitel i bok/rapport/konferenshandlingKapitelVetenskapligPeer review

Sammanfattning

Virtually all developed countries share the problem of too many employees leaving the labor market permanently due to disability or health problems. Work disability means that a person is unable to work until her retirement age due to illness or disability. The risk factors for the disability should be identified promptly, as early intervention is known to be cost-effective, and more effective than later treatments. Thus, it is important to identify people who are at high risk of ending up in disability pension as early as possible. There are five stakeholders in the work disability risk prediction: an employee, an employer, occupational health care, a pension fund, and society. It is in the common interest of the stakeholders to keep the employees healthy and able to work as long as possible. There are only a few machine learning (ML) methods for this problem. However, ML-based methods are efficient and economical in the screening of potential risk cases. Thus, we developed an ML method MHealth for work disability risk prediction. MHealth uses textual data from occupational health care and Natural Language Processing (NLP) based deep learning algorithms. Training data is labeled by doctors into two or three risk classes. We compared MHealth with another work disability risk prediction method MPension. MPension uses structural data from pension decision registers and different ML algorithms such as decision trees. MHealth reached an accuracy of 72% by using neural networks and a two-class model. This tool is used in occupational health care to help in screening process of the patients. MPension achieved an accuracy of 69–78% depending on the algorithm used. The accuracy, sensitivity, and specificity of these methods are enough to use to support expert work but the decision-maker must still be a human and the responsibility for the decision cannot be on artificial intelligence but on the expert. Non-maleficence, accountability and responsibility, transparency and explainability, justice and fairness, and respect for various human rights are the most important aspects of ethical AI in work disability risk prediction. When estimating the ethicality of an AI method, we need also to consider stakeholders’ different interests, goals, and reasons for actions. We examined the AI ethics of prediction of work disability risk based on these criteria. When an ML method estimates a person’s ability to work, it is necessary to understand how the machine made the estimation. According to human rights, people are entitled to have explanations on how decisions were made so that they can maintain agency, freedom, and privacy. Thus, ML methods must be transparent and explainable for different stakeholders to trust the results. However, especially deep learning methods are typically black boxes, and their function is not well understood. To understand better the function of the methods, we formulated the visualizations for the methods MHealth and MPension and discussed how explainable they are. We can conclude that decision trees, which are used in MPension, are easier to explain than neural networks and deep learning algorithms, which are used in MHealth. However, we managed to formulate the visualizations for both methods. To conclude, MPension is both more accurate and more explainable than MHealth. However, MHealth can be used in an earlier stage of the process, which is important for early detection and preventive support for a person, who has increased work disability risk. It is important to produce ML methods for work disability prediction that are not only accurate, sensitive, and specific to support decision-making but also trustworthy, transparent, and explainable.

Originalspråkengelska
Titel på värdpublikationCurrent and Future Trends in Health and Medical Informatics
RedaktörerKevin Daimi, Abeer Alsadoon, Sara Seabra Dos Reis
Antal sidor15
UtgivningsortCham
FörlagSpringer
Utgivningsdatum1 okt. 2023
Sidor345-359
ISBN (tryckt)978-3-031-42111-2
ISBN (elektroniskt)978-3-031-42112-9
DOI
StatusPublicerad - 1 okt. 2023
MoE-publikationstypA3 Del av bok eller annan forskningsbok

Publikationsserier

NamnStudies in Computational Intelligence
FörlagSpringer
Volym1112
ISSN (tryckt)1860-949X
ISSN (elektroniskt)1860-9503

Bibliografisk information

Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.

Vetenskapsgrenar

  • 113 Data- och informationsvetenskap

Citera det här