Predicting Disability Pension Risk for Finnish Public Sector Employees Using Sickness Absence Data

Core Concepts

This research paper introduces a statistical model using sickness absence data to predict the risk of disability pension among Finnish public sector employees and proposes a "critical duration" indicator for early intervention.

Abstract

• Bibliographic Information: Sohlman, P., Louhi, R., & Salonen, J. (Year). New evaluation tool for predicting disability pension risk among Finnish public sector employees. Journal Name, Volume(Issue), Page range.
• Research Objective: To develop a statistical model for predicting disability pension (DP) risk among Finnish public sector employees and to identify a "critical duration" of sickness absence (SA) that indicates elevated risk.
• Methodology: The study used administrative data from Keva, the Finnish Public Sector Pension Provider, covering 940,021 observations from 2016 to 2019. A logistic regression model was developed to predict DP retirement based on individual employee data, including SA spells, age, gender, occupation, and previous DP history.
• Key Findings: The study found significant variation in DP risk across occupations and age groups. A "critical duration" of SA days was identified, beyond which the risk of DP application within three years significantly increased. This critical duration varied by occupation, highlighting the need for occupation-specific early intervention strategies.
• Main Conclusions: The proposed model can be a valuable tool for employers and occupational health professionals to identify individuals at high risk of DP and to implement timely interventions. The "critical duration" indicator provides a concrete metric for guiding early intervention efforts.
• Significance: This research contributes to the field of occupational health by providing a practical tool for predicting and potentially mitigating the risk of disability pension in the public sector.
• Limitations and Future Research: The study was limited to Finnish public sector employees and data from larger employers. Future research could explore the model's applicability in other sectors and countries, as well as incorporate additional factors influencing DP risk.

Stats

The number of disability pension recipients in Finland has decreased by 32% since 2017.
The incidence rate of disability pensions in the overall statutory earnings-related pension system has halved since 2002.
The study's dataset included 940,021 observations from 2016 to 2019, encompassing 340,816 individual employees.
The prevalence of disability pension within the dataset was 2.1% (19,816 observations).
The logistic regression model achieved an AUC score of 0.84 on the test sample.
The average critical duration of sickness absence days among study participants was 15 days.
In the case study of a major local government employer, the estimated number of new disability pensions within three years was 556.
The direct cost of sickness absences for the case study employer in 2021 was estimated at 50 million euros.
The employer's disability pension payments totaled 20.3 million euros in 2021.

Quotes

Key Insights Distilled From

New evaluation tool for predicting disability pension risk among Finnish public sector employees

by Petra Sohlma... at arxiv.org 10-29-2024

https://arxiv.org/pdf/2410.19890.pdf

Deeper Inquiries

How can the proposed model be adapted and implemented in other countries with different social security systems and labor market structures?

Adapting the Finnish disability risk prediction model for other countries requires careful consideration of contextual differences. Here's a breakdown of the key adaptations and implementation steps:
1. Data Harmonization and Availability:

Social Security Systems: Different countries have varying definitions of disability, eligibility criteria for disability pensions, and structures for sickness benefits. The model's outcome variable ("DP retirement") needs to be redefined according to the target country's system.
Labor Market Structures: Occupational classifications (like ISCO-08 used in Finland) might differ. Mapping occupations to a common framework or adapting the risk classification based on local data is crucial.
Data Collection: Ensure the availability of comparable data on sickness absence, employment history, demographics, and ideally, pension records. Data privacy regulations need to be strictly adhered to.
2. Model Recalibration and Validation:

Variable Selection: The predictors (age, gender, occupation, sickness absence patterns) might have different impacts in other contexts. New variables relevant to the local context (e.g., type of employment contract, economic indicators) could be explored.
Model Training: The model's parameters need to be re-estimated using data from the target country. This ensures the model's predictions are accurate for the specific population and labor market.
Rigorous Validation: Thoroughly test the model's performance using independent datasets from the target country. Metrics like AUC (Area Under the Curve), sensitivity, and specificity should be assessed.
3. Implementation and Ethical Considerations:

Phased Rollout: Start with pilot implementations in specific sectors or regions to assess real-world effectiveness and address any unforeseen challenges.
Transparency and Explainability: The model's logic and predictions should be transparent to employers and employees. This builds trust and allows for informed decision-making.
Ethical Oversight: Establish clear guidelines for using the model's predictions. Safeguards against discrimination based on predicted risk are essential. Focus should be on prevention and support, not penalization.
4. Continuous Monitoring and Improvement:

Regular Review: The model's performance and the social security landscape can change over time. Periodic reviews and updates are necessary to maintain accuracy and relevance.
Feedback Mechanisms: Gather feedback from employers, employees, and healthcare providers to identify areas for improvement and ensure the model aligns with evolving needs.

Could factors beyond sickness absence data, such as workplace environment, job demands, and access to healthcare, improve the model's predictive accuracy?

Absolutely, incorporating factors beyond sickness absence data can significantly enhance the model's predictive accuracy and provide a more holistic understanding of disability risk. Here are some key factors to consider:

Workplace Environment:

Psychosocial Factors: Job strain (high demands, low control), effort-reward imbalance, workplace bullying, and social support at work are strong predictors of both sickness absence and disability.
Physical Work Environment: Ergonomic factors, exposure to hazardous substances, and physically demanding work can contribute to musculoskeletal disorders and other health problems.
Organizational Factors: Workplace culture, leadership styles, and opportunities for skill development and career progression can influence employee well-being and work ability.

Job Demands:

Cognitive Demands: High workload, time pressure, and complex decision-making can lead to stress and burnout.
Emotional Demands: Jobs requiring emotional labor (e.g., healthcare, social work) can take a toll on mental health.
Physical Demands: Heavy lifting, repetitive motions, and prolonged standing or sitting can increase the risk of musculoskeletal injuries.

Access to Healthcare:

Timely and Quality Care: Access to early interventions, specialist services, and effective treatments can prevent minor health issues from escalating into long-term disabilities.
Mental Health Services: Stigma and limited access to mental health care can delay diagnosis and treatment, leading to prolonged sickness absence and disability.
How to Integrate These Factors:

Data Collection: Employers can collect data through employee surveys, health risk assessments, workplace inspections, and by analyzing job descriptions.
Data Linkage: Linking existing data sources (e.g., occupational health records, employee assistance program utilization) with sickness absence and pension data can provide valuable insights.
Advanced Analytics: Machine learning techniques can handle complex datasets and identify patterns that traditional statistical models might miss.
Benefits of a More Comprehensive Model:

Improved Prediction: A more nuanced understanding of individual risk factors leads to more accurate predictions of disability.
Targeted Interventions: Employers can develop tailored interventions and support programs based on the specific risk factors identified.
Proactive Approach: Shift from reactive case management to a proactive approach that promotes workability and prevents disability.

What are the ethical implications of using predictive models in occupational health, and how can potential biases be addressed to ensure fair and equitable treatment for all employees?

Using predictive models in occupational health presents significant ethical challenges. Here's a breakdown of the key concerns and strategies to mitigate bias:
Ethical Implications:

Discrimination and Fairness: Models trained on historical data can perpetuate existing biases. For example, if women have historically been overrepresented in certain lower-paying occupations with higher disability rates, the model might unfairly assign them a higher risk, even if their individual circumstances don't warrant it.
Privacy and Confidentiality: Collecting and analyzing sensitive health and work-related data raises privacy concerns. Employees need to be fully informed about how their data is being used and have control over their information.
Transparency and Explainability: "Black box" models that lack transparency in their decision-making processes can erode trust and make it difficult to challenge potentially unfair outcomes.
Stigmatization and Labeling: Employees identified as high-risk might experience stigma, anxiety, and reduced job opportunities, even if they never develop a disability.
Addressing Bias and Ensuring Fairness:

Data Quality and Bias Auditing:

Data Collection: Ensure data is collected using unbiased methods and that it represents the diversity of the workforce.
Bias Detection: Regularly audit the data and the model for potential biases related to gender, race, ethnicity, age, disability status, or other protected characteristics.

Model Development and Validation:

Fairness-Aware Algorithms: Explore the use of algorithms specifically designed to mitigate bias and promote fairness in predictions.
Diverse Development Teams: Include individuals from diverse backgrounds in the model development process to challenge assumptions and identify potential blind spots.

Implementation and Use:

Transparency and Explainability: Provide clear explanations of the model's logic, factors considered, and how predictions are made.
Human Oversight: Ensure human judgment and ethical considerations are part of the decision-making process. Never rely solely on model predictions.
Focus on Prevention and Support: Use the model to identify opportunities for early intervention, workplace improvements, and support programs that benefit all employees, not just those deemed high-risk.

Ongoing Monitoring and Evaluation:

Impact Assessments: Regularly assess the model's impact on different employee groups to identify and mitigate any unintended consequences.
Continuous Improvement: Continuously monitor and refine the model to address emerging ethical concerns and ensure fairness.
Key Principle: The goal of using predictive models in occupational health should be to promote well-being, prevent disability, and create a more inclusive and supportive work environment for all employees.