Motion Blur of People Walking in the City
© Marco_Piunti / Getty Images

Predictive Analytics and BMI Misrepresentation

Leveraging predictive analytics to mitigate the mortality impacts of BMI misrepresentation

    alt txt

    properties.trackTitle

    properties.trackSubtitle

    February 2022

    The U.S. life insurance marketplace is evolving rapidly toward automation and smarter allocation of resources to provide a better experience for policy holders and issuers. Many insurers are implementing various approaches for accelerated underwriting where a subset of applications for fully-underwritten products qualify to have their medical exams and fluid testing waived. This process is simpler, less invasive, and faster, resulting in an improved customer experience. These accelerated underwriting approaches are often coupled with data-driven predictive models that enable high straight through processing rates and help manage the extra mortality risk.

    Without medical exams and verification of disclosures, there is increased mortality risk. When applicants know their disclosures will not be verified, it becomes easy to misrepresent health status and obtain a better underwriting class. Additionally, there is increased mortality from loss of the sentinel effect and self-selection as less healthy individuals are more likely to apply for coverage when medical exams are eliminated. Additional mortality costs must be balanced with the cost-savings of the accelerated program.

    Body mass index is an important driver of risk class and stands as the second largest concern for misrepresentation in accelerated underwriting after smoking non-disclosure.

    Body mass index (BMI) is an important driver of risk class and stands as the second largest concern for misrepresentation in accelerated underwriting after smoking non-disclosure. BMI misrepresentation occurs when the BMI as measured by a medical professional differs from the BMI self-reported by the applicant, typically in the direction of the applicant underreporting the true BMI. With research confirming BMI misrepresentation rates of 20% or higher in fully underwritten programs, these rates are expected to increase in accelerated programs, imposing a material impact on mortality risk. Anticipating the likelihood and magnitude of misrepresentation is key for identifying high risk applicants and understanding the extra mortality impact. This poses a unique challenge compared to smoker misrepresentation, which doesn’t incorporate degree of misrepresentation. BMI misrepresentation prediction models address this challenge, as these models can proactively identify applicants who are likely to significantly understate their BMI and risk can be mitigated by selectively re-routing them to full underwriting.

    This article summarizes the extra mortality impact from BMI misrepresentation, Munich Re’s approach to modelling misrepresentation and how these models can be used to better manage mortality risk in accelerated underwriting programs.

    Mortality Impact

    Without verification of all applicants’ build values through medical exams as conducted in fully underwritten programs, BMI misrepresentation in accelerated underwriting programs will result in mortality increases since some applicants will be assigned better rate classes than they would otherwise. The insurer is generally willing to accept some mortality increase at the gain of higher straight through processing rates; BMI misrepresentation models help achieve this balance. 

    The dataset used in this study of BMI misrepresentation represents typical life insurance applicants to full underwritten programs, with an average BMI misrepresentation rate of 20%. However, not all individual applicants have the same likelihood of misrepresenting their BMI, and they also vary in the magnitude of their underreporting, producing the cumulative extra mortality cost chart below. The cases are ordered by their probability to misrepresent as assessed by a predictive model, and then split into 10 deciles with equal number of observations in each bin. Decile 1 has the lowest predicted BMI misrepresentation rates, and decile 10 the highest. If all cases are accepted through accelerated underwriting, there is a 2% mortality increase due to BMI misrepresentation with a 7% increase in the 10th decile. However, if the top 20% of applicants most likely to misrepresent (deciles 9 and 10) are rerouted to full underwriting, the additional mortality increase is nearly halved, from 2.2% to 1.4%, while maintaining a straight through processing rate of 80%.
    BMI Misrepresentation Data
    © Munich Re
    Figure 1: Cost benefit analysis of re-routing the top 20% of applicants most likely to misrepresent BMI for medical exams to verify BMI

    Data

    Our approach for modeling BMI misrepresentation uses underwriting data (basic demographics and medical disclosures from application data), geospatial features, and selected lifestyle attributes to predict the likelihood of misrepresentation. Multiple data sources are preferred to create robust models, though baseline models using only underwriting data can be constructed. Research suggests that the major identifiers of BMI misrepresentation are largely demographic and geographic attributes, such as age, gender, wealth, and county. Although model results will be applied in accelerated programs, data from pre-existing fully underwritten cases is needed to begin with so that models can be initially trained and evaluated using applicants’ reported and measured build values.

    In this study, approximately six years of fully underwritten applicant data was used to build the model. Exploratory analysis of the study data, as shown in the figure below, highlights that the likelihood of mispresenting BMI increases with true BMI (gray bars), which will cause an increase in mortality if these cases were underwritten without BMI verification through medical exams. Importantly, this trend does not hold when looking at disclosed BMI alone (teal bars); individuals have roughly the same rate of BMI misrepresentation at all levels of self-reported BMI, so it is important to use additional variables available at the time of underwriting to effectively stratify risk.

    BMI Misrepresentation Data
    © Munich Re
    Figure 2: Applicants with higher measured BMI are more likely to misrepresent (gray bars), whereas misrepresentation rates are fairly uniform across disclosed (self-reported) BMI bins (teal bars).

    Methodology

    Definition of misrepresentation

    Applicants misrepresent their BMI in different ways, reporting BMI that is either greater than or less than their actual measured values, with varying degrees of disparity. For an insurer, this difference between an applicant’s self-reported and measured BMI can cause a shift in underwriting class within a carrier’s build guidelines. For a specific carrier, the degree of misrepresentation between self-disclosed and measured BMI that should be flagged as significant is best determined through discussions between underwriters, actuaries, and data scientists, and may depend on the details of those guidelines.

    For the generalized industry-level model implemented in this study, we established a standard definition of BMI misrepresentation, given as absolute difference or percent change, which is an alternative to using a carrier’s specific build rules. We found that a 6% percent difference between self-reported and measured BMI is an appropriate misrepresentation flag, as it accurately segments misrepresenting cases, is typically large enough to cause a shift in assigned underwriting class, and significantly impacts mortality. Additionally, only BMI that was understated by 6% or more is considered misrepresentation, since overstating BMI is usually less impactful to underwriting.

     

    Modeling Approach

    Machine learning models use large, information-rich datasets to learn patterns and predict on new data. This BMI misrepresentation predictive model, developed to identify potentially misrepresenting applicants, leverages Munich Re’s underwriting and third party datasets. The model is calibrated on traditional demographic, medical, behavioral, geospatial, and selected lifestyle data — all available at the time of underwriting to provide instant recommendations. It flags applicants likely to misrepresent their BMI as well as providing the top reasons for why the model arrived at the prediction. When trained on sufficient data, the model is able to successfully segment potentially misrepresenting applicants.

    Implementation and Monitoring

    A cost of implementing accelerated underwriting programs is the additional mortality risk imposed when application disclosures aren’t verified through fluids or other exams. This cost is managed by flagging individuals at high risk of BMI misrepresentation (and high extra mortality cost) and re-routing to full underwriting. BMI misrepresentation rates vary between carriers and will change over time as applicants and advisors learn that build will not be verified in accelerated underwriting programs. We recommend that carriers implement a monitoring program with audits (i.e. post-issue medical underwriting) or a random hold out where a percentage of applicants are randomly selected to go through medical exams. This allows for tracking misrepresentation rates as well as model performance over time, revealing when it is time to adjust the model in order to continue to mitigate mortality risk. Last, monitoring provides valuable insights into the driving factors behind misrepresentation, contributing to better segmentation of applicants.  

    Summary

    Accelerated underwriting programs offer a subset of applicants a faster and less intrusive path to purchasing life insurance. Carriers benefit from lower underwriting costs and higher take-up rates. However, these programs pose some additional risk as the elimination of traditional underwriting factors may result in a less accurate risk assessment. When applicants know they are likely to avoid medical exams and fluid testing, some will misrepresent their health status. One way to mitigate the additional mortality costs from BMI misrepresentation is to apply BMI misrepresentation prediction models and route likely misrepresenting applicants through full underwriting. It is recommended that clients begin with their internal data to model likely misrepresenting applicants and add external data sources to improve the models in an iterative fashion. Munich Re can advise insurers on data sources for these models, provide modeling services and assist with implementation and monitoring model performance over time.

    Contact Author:
    Patrick Sullivan
    Patrick Sullivan
    Senior Vice President, Munich Re Life US