The project team succeeded in automatically filtering treatment records and occupational health survey responses to identify factors that predict a mental health diagnosis or longer treatment period. Predictors of a diagnosis included intense stress, repeated fatigue, sadness and being female. The approach can be used as a new type of predictive tool for mental health problems at the population level.
Deterioration in employee mental health can be predicted with machine learning methods. In the research project of the Finnish Institute of Occupational Health, the first diagnosis of mental health and the continuation of treatment could be predicted quite well using machine learning.
In the Occupational Health Survey responses, seven factors were identified as increasing the likelihood of receiving a diagnosis related to a mental health problem or sleep disorder during the two-year monitoring period.
Predictors of deteriorating mental health included intense stress, repeated daytime fatigue and exhaustion, sadness and melancholy, and symptoms of anxiety. In addition to self-rated well-being, the gender of the person is also a key factor. Based on the project, women have an increased likelihood of receiving a diagnosis.
“We identified predictors of a diagnosis from more than a hundred questions. The predictive ability of these seven questions was almost as good as that of all the questions together,” explains Pekka Varjeresearch director at the Finnish Institute of Occupational Health.
Younger employees are more likely to receive one of the diagnoses covered in the study, but age was not among the strongest predictors.
Modeling not suitable for individual treatment planning
The results of the research project “Predictive methods to improve the sustainability of mental well-being at work” served as the basis for the publication of a predictive map for the diagnosis of mental health in the Work-Life Knowledge Service. A tool such as the predictive map could serve as an aid to resource planning in occupational medicine and prevention in mental health, for example.
“At the level of individuals, the modeling contains too many uncertainties. It is possible that this type of tool could help identify at-risk groups or development trends more broadly in the working-age population,” says the research professor. Ari Väänänen.
Prolongation of treatment can be predicted based on treatment records
In the second part of the project, the research topic was the extension of the treatment period in mental health. Treatment records written by physicians were automatically analyzed to identify the type of subjects predicting a treatment period of more than four physician visits.
The lengthening of the duration of treatment was predicted in particular by entries in the treatment record relating to depression, its medical treatment and exhaustion. However, the key predictor was a diagnosis received early in the treatment period. Certain diagnoses related to depression and anxiety disorders predicted a longer duration of treatment in occupational medicine.
Machine learning helps analyze massive datasets
The project’s analyzes were based on machine learning methods. Two very extensive research populations have been followed for several years. The populations, made up of people of working age, broadly represented different sectors and occupational groups.
“Machine learning methods and advanced computing performance have enabled the use of research approaches that apply big data. Barely ten years ago, this type of research could not be carried out,” says Pekka Varje.
Predictive methods to improve the sustainability of mental well-being at work
Source: Finnish Institute of Occupational Health (FIOH)