Optimizing Predictive Modeling

Data analytics and predictive modeling are quickly becoming workers’ compensation industry staples as more employers recognize that the technology can positively impact their operations by improving claims management, risk assessments and loss costs. This session at the 24th Annual National Workers’ Compensation & Disability Conference illustrated key elements to include in workers’ compensation predictive models.

Speakers included:

• Frank Murray, Senior Vice President of Claims at ESIS
• Tim Starks, Director of Casualty at Georgia-Pacific

Predictive modeling is a commonly-used statistical technique to predict future behavior. Claims are defined by broad categories and treated with a common set of best practices. The claim process is based on analytic segmentation and claim categories are refined providing opportunities to have a greater impact on the subset of claims driving costs. In searching for meaningful patterns in data, the greatest value is when it forces us to notice what we never expected to see.

Steps to fully utilizing predictive models to their maximum potential, include:

• Develop a suite of sophisticated predictive models that accurately segment claims.
• Optimize claims technical expertise, systems, and a differentiated process that support action.
• Measure, connecting action to outcomes and quantifying the result.
• Execute a plan with consistent, focused and timely claim intervention that drives improved outcomes.

Data Mining & Variables

The Cross Industry Standard Process for Data Mining (CRISP-DM) is a useful method that can be utilized in predictive modeling. This process includes:

• Business understanding – What’s the problem?
• Data understanding – Where is the data and what shape is it in?
• Data preparation – Transforming and deriving
• Modeling – Find patterns
• Evaluation – Testing patterns
• Deployment – Fetting the model output into the right hands

In workers’ compensation, variables to record and measure typically include:

• Claimant – Age, job class, tenure, comorbidities
• Occurrence – Cause, body part and nature, responsible party, date and time
• Environment – State, facility location, culture

Workers’ Compensation Intake Model 

ESIS provided an overview of this model used for all workers’ compensation claims at first notice of loss.

Overview:

• Medical-only model evaluates potential severity relative to other medical-only claims and propensity for claims to convert to lost time.
• Lost-time model includes two sub-models (medical and indemnity) that create a composite claims severity score.
• Measures seven variables of interest – injury group, gender, claimant age, cause code, state, employer hazard group, report lag.
• Claims are routed to appropriate personnel based upon score and client service plan.

Output:

• Medical-only or lost-time claim severity score (1-100) plus top three severity indicators.
• Claim is re-scored within six months as information changes (e.g., injury) or becomes available.

Model Variation:

• Medical-only claims that scored 81-100 (worst 20% per model) were 90% more likely to convert to lost time and cost 160% more than average.
• Lost-time claims that scored 81-100 cost 60% more than average lost-time claim.