Milliman Irix® Risk Score 3.0 with Medical Data and Credit Data
Stratifying mortality risk by combining prescription drug history, medical claims records, and credit data
Global Inflation Tracker
© primeimages / Getty Images
    alt txt

    properties.trackTitle

    properties.trackSubtitle

    Predictive models and life insurance

    Munich Re Life US assessed Milliman Irix® Risk Score with Medical Data and Credit Data, a predictive modeling tool developed and owned by Milliman that assesses mortality risk through the combination of an individual’s prescription drug history, medical claims records, and credit data. Insurers interested in prescription-, medical- and credit-based mortality scores should perform a retrospective validation study on their experience data. Munich Re can assist carriers with the retrospective study, advise on changes to mortality assumptions, and recommend ways to incorporate the scores to streamline the underwriting process. 

    Executive summary

    Milliman Risk Score with Medical Data and Credit Data is a proprietary scoring algorithm that combines prescription drug history, medical claims records, and credit data to predict the mortality risk of individuals relative to other individuals of the same age and gender.   

    In 2020, Munich Re assessed Risk Score 2.2, a score based on prescription drug history with the option to include credit data, and concluded that the scores are predictive of mortality for the U.S. insurance applicant population.1 In 2022, Milliman upgraded its Risk Score to incorporate medical claims data to produce the Risk Score 3.0 with Medical Data.2 Since then, Milliman further enhanced its Risk Score 3.0 suite to incorporate credit data so that carriers now have the option to use all three data sources in the Risk Score 3.0.

    Milliman provided Munich Re with a dataset of 42 million lives sampled from the U.S. general insurance applicant population, including life, health, LTC, DI, final expense, and Medicare supplement lines of business. Life represents the majority of the dataset (64% of lives), followed by health (19% of lives). For each life in the study, risk scores were provided for the various data source combinations, including Prescription with Medical Data and Credit Data, Prescription with Medical Data, Prescription with Credit Data, and Prescription-only Data.

    Munich Re analyzed the Risk Score with Medical Data and Credit Data and confirmed it is predictive of mortality for the U.S. insurance applicant population. The combination of Prescription data, Medical data, and Credit data provides greater mortality risk stratification insights as compared to the Risk Score with Credit Data or Medical Data only. The Risk Score with Medical Data and Credit Data model yields higher number of total scored lives, as well as provides greater exposure in the lowest and highest risk score bands, compared to other versions of the model using fewer data sources.

    A company-specific insured population is not expected to have identical underlying characteristics as this insurance applicant population. Munich Re recommends that carriers interested in the Risk Score with Medical Data and Credit Data conduct a retrospective study on a company-specific applicant dataset to calibrate the tool to the carrier’s underwriting paradigm. This process will help carriers balance the score with the expected mortality of their unique target markets, distribution channels, and underwriting processes. For carriers already using a Prescription with Medical Data score or Prescription with Credit Data score, a Prescription with Medical Data and Credit Data combined score provides a more holistic assessment of mortality.

    Methodology

    Munich Re assessed the effectiveness of the new Risk Score with Medical Data and Credit Data in stratifying the mortality risk profile of a pool of 40 million3 applicants with entry ages 18 - 99. Juvenile lives were filtered out of our analysis as they were not scored under the Risk Score models incorporating credit data. The data provided included various versions of Risk Scores (prescription-only, prescription with medical, prescription with credit, and prescription with medical and credit combined scores) on a scale of 0 to over 100. Each life entered the study between the first quarter of 2005 and the last quarter of 2020. Deaths were sourced from a reliable third-party source of death data. The study population was comprised of 1.7 million deaths out of 224 million exposed life-years.

    The expected mortality basis was taken from the 2015 VBT primary select and ultimate age-nearest birthday tables split by age and gender with a 1 percent mortality improvement. Applicant smoking status was not provided, so an 85 percent non-smoker/15 percent smoker blend of the smoker distinct tables was used.

    Munich Re completed additional analyses of relative actual to expected mortality ratios (A/E) by various factors (e.g., hit type, age, and gender) to assess whether the mortality risk stratification by the Risk Score with Medical Data and Credit Data is influenced by those factors.

    Results

    Overall results

    Figure 1 is a high-level overview validating Risk Score with Medical Data and Credit Data’s ability to stratify mortality risk. Lives with higher scores have higher mortality risk, while lower scores correspond with lower mortality risk. 4.2% of the total exposure did not have a score from the Risk Score with Medical Data and Credit Data model, and are not shown in Figure 1.

    Exposure

    The Risk Score with Medical Data and Credit Data model produces a score for an individual with a prescription history (Rx) only, medical claims record (Dx) only, credit data (Cr) only, or any combination of the three data source hits. “Eligibility-only” prescription hits, which refers to applicants found in the prescription history database but with no prescription fills, were combined with “No Rx Hit” for the purposes of this analysis. As illustrated in Figure 2, a Prescription with Medical Data and Credit Data model yields a risk score for a larger subset of the same population exposure compared to a Risk Score model using fewer than all three data sources.
    Graphic chart of the study data exposure by database hit type
    Compared to a Prescription with Medical Data model, the use of a Prescription with Medical Data and Credit Data model increases the percent of total exposure with a risk score by 15.1% from 80.8% to 95.9%, based on the study data. Results in production may be different from the study population as hit rates evolve over time. A company-specific retro study will demonstrate the expected production hit rates for an individual carrier.

    Comparison between Risk Score Models

    Figure 3 - Distribution of Exposure
    Figure 3 compares the exposure distribution percent and mortality risk segmentation between the two versions of the current Risk Score 3.0: Prescription with Medical model and Prescription with Medical Data and Credit Data model (excluding lives with no score). We observe shifts in exposure percent to both ends of the score spectrum for the Prescription with Medical Data and Credit Data model. This means that in addition to an increase in total scored lives, a Prescription with Medical Data and Credit Data score allows carriers to identify a higher proportion of lives with low risk and high mortality risk than a Prescription with Medical Data only score. 
    Figure 4
    Figure 4 compares the exposure distribution percent and mortality risk segmentation between the various versions of Risk Score models with Credit Data: Prescription with Medical Data and Credit Data 3.0 and Prescription with Credit Data 2.2 and 3.0 scores (excluding lives with no score). We observe shifts in exposure percent in the highest and lowest score bands for the Prescription with Medical Data and Credit Data model. Additionally, we see more effective mortality stratification in the highest scores under the Prescription with Medical Data and Credit Data 3.0 model, suggesting that adding medical data allows for more effective segmentation of the worst mortality risks.

    Use Case – accelerated underwriting triage tool

    Figure 5 - Accelerated Underwriting Triage Example
    Figure 5 illustrates an example of using Risk Score 3.0 as a triage tool to segment lives into accelerated (AUW) and full underwriting (FUW) paths. All three scenarios were based on the same insurance applicant population and a kick-out threshold score of 1.6. Applying the same kick-out threshold score, the Prescription with Medical Data and Credit Data model can straight-through process (STP) more lives while maintaining slightly lower relative mortality for the accelerated cases compared to both the Prescription with Medical Data and Prescription-only models. The relative mortality for the lives triaged out of AUW is significantly higher under the combined score model with all three data sources. 

    Database hit type

    Graphic table of Relative A/E by Hit Type
    Table 1 summarizes the relative A/E by prescription (Rx), medical claims (Dx) and credit (Cr) data hit type within the Risk Score 3.0 model. The subsets with Cr hits have lower relative mortality than subsets without Cr hits. The subsets with Dx hits have higher relative mortality than subsets with no Dx hits. The subsets with Rx and Cr hits have lower relative mortality than subsets with Cr and no Rx / Eligibility-only hits. 
    Figure 6 illustrates the mortality risk segmentation of the Prescription with Medical Data and Credit Data score across Rx, Dx, and Cr hit types. While the score provides similar stratification of mortality across hit types, lives with a Dx hit demonstrate more effective segmentation at high mortality risks than lives with no Dx hit. Lives with a Cr hit show better relative mortality as compared to lives with no Cr hit at the same score bands. 

    Age

    Figure 7 shows the distribution percent of the exposure within each age group for the Prescription with Medical Data and Credit Data score (excluding lives with no score). The Risk Score is controlled for age. The distribution of scores has a slight bell-shaped curve, with a higher proportion of lives at both ends of the score spectrum for ages 40 to 69. We do not see material bias of a larger proportion of high scored lives in older ages.

    Figure 8 demonstrates the mortality risk segmentation of the Prescription with Medical Data and Credit Data score by age group. All age groups follow a similar mortality stratification pattern, where mortality risk increases as scores increase. The chart shows that the score is particularly effective at segmenting mortality for ages 30 to 59.

    When considering breakdowns by gender, the mortality segmentation pattern also looks similar across both males and females.

    Bias Testing

    Milliman tests the Risk Score with Medical Data and Credit Data for bias. Milliman tests for disparate impact by age and gender, defined as favoring or disfavoring certain demographics more than actuarially justified, by ensuring the impacts of the scores and score distribution track changes in the observed empirical mortality. For instance, for any given age or gender, Milliman tests that a higher score will have proportionally higher average relative mortality, consistent with the results shown by Munich Re in Figure 8. We encourage carriers to conduct bias testing as part of the retrospective study on the company-specific insured population.

    Summary

    Munich Re concludes that the Milliman Risk Score with Medical Data and Credit Data is an effective predictor of mortality for the U.S. insurance applicant population. Compared to earlier versions of the Risk Score based on fewer data sources, the Risk Score with Medical Data and Credit Data increases the volume of scored individuals and more effectively stratifies mortality risk. A score is provided for an individual with prescription history only, medical claims record only, credit data only, or any combination of the 3 data sources. While mortality segmentation is effective across all of these data combinations, the Risk Score is most effective at identifying high mortality risk when medical claims data hits are present. In the context of a fluid-less underwriting triage tool, the Prescription with Medical and Credit score provides better risk segmentation with more lives scored than earlier versions of the Risk Score, allowing carriers to achieve higher STP with better mortality.

    Finally, a company-specific insured population is not expected to have identical underlying characteristics to this insurance applicant population; we recommend that each carrier perform a retrospective study, including bias testing, to best assess the value and application of the Risk Score with Medical Data and Credit Data. Munich Re can assist carriers in structuring bias testing, evaluating the retrospective study and advise on ways to incorporate Milliman Risk Score with Medical Data and Credit Data into their underwriting process.

    References
    Contact the author
    Cherry Wang
    Cherry Wang
    Director & Actuary, Underwriting Risk Assessment & Research
    Munich Re Life US

    Newsletter

    Subscribe to our newsletter for the latest industry research, insights on life and health-related topics, and relevant Munich Re Life US updates. Join today and stay connected with our business.