CASE/06 / Insurance & Financial services
Florida, USALife expectancy modelling for life settlements
Standardised medical data feeds a model with visible contributing factors.
01 / THE ASK
The decision behind the brief.
Use individual medical information to support life-settlement policy valuation.
02 / WHAT WE FOUND
What the work revealed.
Existing models relied on broad demographic data while health information was scattered across handwritten notes, scans and structured files. The relevant medical factors first needed to become consistent data.
03 / WHAT WE BUILT
Medical records into an explainable estimate.
Extract the medical record
OCR and NLP extract and standardise medical information from varied sources.
Model contributing factors
A Cox Proportional Hazards model incorporates lifestyle and chronic condition factors. Dashboards show which factors contribute to each estimate.
Check against a baseline
Backtesting compared the model with industry benchmarks and the firm’s existing methods.
Placeholder for an approved, masked view of the delivered work.
04 / THE RESULT
Standardised medical data feeds a model with visible contributing factors.
Standardised medical information supports estimates with visible contributing factors. Backtesting compared the model with existing methods. Comparative accuracy remains unreported until the measure is defined.
[CONFIRM: accuracy metric, test period, sample size, benchmark and validation basis]
05 / ENGINEERING DETAILUnder the hood
- Medical notes, scans and structured files
- OCR and NLP standardisation
- Cox model and factor dashboard
OCR and NLP prepare the input for a Cox Proportional Hazards model. [CONFIRM: model assumptions, evaluation design, clinical review and data protection controls]
Workflow sketch. Unconfirmed components are marked explicitly.
- OCR
- NLP
- Cox Proportional Hazards
A GOOD PLACE TO START