Mental health diagnosis prediction chart – background and interpretation
The Finnish Institute of Occupational Health research project identified factors that predict a mental health diagnosis and which are related to employed people’s most common mental health issues. The extensive occupational health care data was analysed by utilising machine learning models. Among the over one hundred variables, the most important factors that predict a mental health diagnosis are young age, female gender, melancholy thoughts, intense feelings of stress and recurring feelings of daytime tiredness.
Work-related mental health
A wide variety of factors associated with poor mental health have been identified in work-related mental health research. But what is critical in terms of mental health? At the end of the 1960s, psychiatrists Thomas Holm and Richard Rahe developed a stress inventory that indicated the stressfulness of various life events. The inventory is based on the assessments by more than 5,000 patients of 43 different life events and their mental stressfulness. For example, the death of a loved one is the most stressful event in the inventory.
Occupational health research, on the other hand, has heavily relied on the thought that mental health is built over time and that it consists of a complex of factors that affect each other. Individual events are not as important as the environment and coping potential. A view of mental health emphasizing various aspects of life is based on the idea that, over time, an individual’s life may involve deterioration, stable development or improvement of mental well-being. At least to a certain degree, mental health fluctuates and changes.
Mental health research with machine learning methods
The chart of factors that predict a mental health diagnosis is based on a research project where information of a very large number of employees was collected anonymously from various sources in a data-driven manner. With this approach, we did not limit the factors to be studied in advance, but instead, we allowed the data collected automatically in the systems to speak for itself. The basic idea was to identify a large part of the working age population by monitoring the key factors that predict the deterioration of their mental health. A diagnosis related to a mental health disorder made by occupational health services indicated this deterioration. One of the most important prediction data sources was the occupational health services’ well-being survey, which included questions about work, lifestyle, dimension of well-being, illnesses and other factors that may potentially affect health and work ability.
(Text continues after graph)
As recently as ten years ago, we were not able to chart the most important factors related to the deterioration of mental health from a range of over a hundred factors in a longitudinal dataset of tens of thousands of employees, but today, machine learning methods and advanced computing performance have made it possible to use research approaches that apply big data.
Factors that predict a mental health diagnosis
The following chart displays a summary of the five key factors that predicted a diagnosis related to a mental health disorder in our data. Three of these factors were related to subjective well-being. We learned that melancholy thoughts, intense experiences of stress and recurring feelings of daytime tiredness predicted an increased likelihood of a mental health diagnosis in the monitoring period. This is not surprising in itself: the questions of the survey include dimensions that, when prolonged, can lead to depression, anxiety disorders and insomnia. Put more critically: sadness, stress and tiredness can be called depression or anxiety within the medical system. They can also lead to the prescription of mood stabilizers and other modes of treatment.
The other two key factors were being female and belonging to a young age group. A previous study also shows that women are at an increased risk of a mental health disorder diagnosis, especially depression and anxiety disorders, whereas more evidence of mental health challenges among young people of working age has been gathered in recent years. It can be said that, among young people and women, low subjective well-being indicators predict more accurately a future mental health diagnosis than among old people or men, although these indicators also suggest a risk among these groups.
Application of the chart and future guidelines
An interesting aspect of the study was that including the 20 next most important explanatory factors in the machine learning model does not materially improve the model’s predictive capability. When planning further actions, it is critical to determine where the sadness, tiredness and stress come from. Without this information, changing the employee’s situation is very difficult. It is essential to address the situation as early as possible.
With respect to practical applications, the project’s research group suggests that tools like this should be made available as a tool for managing the workload and HR resources of the health care sector. Instead, the use of machine learning is not recommended for the purpose of predicting individual mental health risks during appointments as the modelling is too uncertain at an individual level.
In the context of research, this study has something new and something old. The new is represented by experimental machine learning methods and exceptionally large datasets. Mental health research within the framework of an individual is more traditional. This is what the psychiatrists who classified the severeness of life events were involved with in the 1960s. However, it is clear that individuals of working age are part of work life and society in general. Because of this, it is important to understand that what is being predicted and what kinds of occupational health worries arise are parts of the whole economy and culture. The diagnoses reflect the prevailing social norms, conditions and work ability demands. Within a historical framework, it is important to consider why mental health indicators, classifications, prediction and management instruments are needed today. What could be developed alongside them or even to replace them?
Creative Commons License
The publication is licensed under Creative Commons 4.0 International -license.