Data Analytics and Underwriting: Case Studies in Applied Risk Aggregation

With the increasing amount of data available, the underwriting process continues to reach new levels of complexity. At the same time, innovative tools and techniques are hard at work to combat this complexity and generate practical criteria on which to base underwriting decisions. 

Through it all, the ability to combine various sources of information into meaningful risk assessment is paramount. For cases with multiple impairments, for example, life insurers must define underwriting guidelines through the process of risk aggregation, which incorporates risk from each individual impairment into an overall risk assessment.  

Underwriters have a variety of risk aggregation techniques from which to choose. For example, the additive principle suggests the risk from one impairment be added on top of the other and may be applied in various ways, such as 1+1>2, 1+1=2 or 1+1<2. Another common approach, the knock-out principle, suggests that out of multiple impairments only the worst impairment matters.

Risk aggregation techniques such as these are useful for a range of underwriting applications; however, a significant degree of subjectivity may still be required in making a decision. Even the most experienced underwriters can find this challenging. In such cases, data analytics offers a potential solution.

The case studies below provide examples of how insurers can better leverage risk aggregation techniques using insights from data analytics to achieve more accurate risk assessment.

Case Study One: Multiple impairments

Challenge: For cases with multiple impairments, the knock-out principle, which focuses only on the worst impairment, can help simplify the underwriting process. The simple rules needed to apply the knock-out principle also make it a favored approach for automated underwriting engines. However, the knock-out principle is likely to underestimate risk for cases where available data makes the additive principle more insightful.

Data Analytics Application:

• Available data: a mortality experience study dataset with medical claim history
• Method:
  1. Assign a mortality risk score to each of the individual’s medical codes and pick the worst score as the “max score.”
  2. Define the individual’s medical conditions (impairment category).
  3. Count the number of medical conditions having the max score.
  4. For each condition with the max score, count the number of codes having the max score.
  5. Perform multivariate analysis to answer the following two questions:
     • Does the number of medical conditions (from Step 3) impact mortality beyond the max score (from Step 1)?
     • Does the number of medical codes (from Step 4) impact mortality beyond the max score for that condition?
• Results: In this hypothetical case, the answers were mostly YES to the first question and NO to the second question.

Conclusion: Across multiple medical conditions, the knock-out principle may be insufficient, suggesting the additive principle is still needed. However, within a given medical condition, knock-out principle may be sufficient.

Case Study Two: Comorbidity impact of impairments A and B

Challenge: The risks of Impairments A and B, individually, are well documented in medical literature, and underwriting rules for both are well established. Also well understood is that having Impairment A increases the risk of having B. However, little is certain about how the two impairments jointly impact mortality if they co-exist. Two controversial and competing hypotheses are at play:

1. Hypothesis 1: The total risk, or total debits, should be less than the sum of debits from each, or 1+1<2. The reasoning is that two impairments are biologically related; therefore, the risk is more likely overlapped.
2. Hypothesis 2: The total risk should be higher than the sum of the debits from each because having both impairments implies the underlying disease is more severe. Disease severity is often associated with mortality risk exponentially.

The reasoning behind both hypotheses seems biologically sound.

Data Analytics Application:

• Available data: A mortality experience study dataset with medical claim history, divided into the following sub-populations: having condition A only, condition B only, both A and B, and neither A nor B
• Analysis:
  1. Compute the mortality actual/expected (AE) ratios for all four sub-populations.
  2. Use the sub-population of neither A nor B as reference to calculate the relative risk (RR) for the other three sub-populations, which is the ratio between AE of a given sub-population and AE of the reference.
  3. Use the formula of 100*(RR-1) to convert RR into debits.
  4. Compare the debits of both A and B with debits of A only + debits of B only.
• Results: In this hypothetical case, debits of comorbidity are much higher than the sum of debits from A and B alone.

Conclusion: While underwriting cases with both Impairments A and B, additional debits can be justified after the summation of debits from each.

As these case studies demonstrate, the convergence of underwriting and data analytics is well underway and essential to the future of insurance. Progress will be incremental, with underwriters focusing on the subjective factors requiring human expertise while data analytics provides objective criteria on which to base their judgements. The amount of available data to evaluate will only increase, and insurers’ ability to combine various data sources using all tools, techniques, and technologies at their disposal will be the key to success.