Drug Abuse Mortality in the Insured Population

We will focus our analysis at the substance abuse level, not specifically drilling down to opioids. Due to the way the cause of death is recorded in the databases, drilling down any further could lead to a higher level of misclassification risk. However, Figure 1 shows that almost all of the increase in drug-related mortality over the period being considered comes from opioids, so conclusions drawn at this level should reasonably translate to the subset of opioid abuse.

We used two main data sources in this analysis: first, the Mortality Multiple Cause-of-Death Public Use Records from the CDC, and second, Munich Re’s proprietary cause of death experience study. The CDC data were based on the general U.S. population and included more than 460,000 deaths from substance abuse between 2000 and 2014. Munich Re’s study was based on life insurance claims data and comprised more than 1,400 deaths from 2006 to 2015.

A socio-economic view of the crisis

We began our investigation by splitting the population experience to test the hypothesis that socio-economic class was a key differentiator in mortality rates. Since the CDC data did not include any measure of income or wealth, we used the highest educational level attained as a proxy. In Figure 2, we split the mortality rate per 100,000 into four main categories of educational attainment (high school or less, some college, BA/BS degree, graduate work). First, we saw that the mortality rate due to substance abuse decreased significantly the higher the academic attainment. Second, the subsets of the population with the highest academic attainment had the flattest mortality trend over time.

We have not attempted to explain the correlation between higher education and lower substance abuse-related deaths in this 

paper.⁴  We have, however, hypothesized that the college-educated sub-group⁵ was a good proxy for the insured population, and proceeded to test that hypothesis.

Testing the hypothesis

In order to test this hypothesis, we compared the mortality rates for the college educated population with the life insurance claims experience from our own block of reinsurance business split into two segments: first, claims that occurred in policy durations one to 15 (to approximate the select period) and second, claims that occurred in policy durations 16 and later (to approximate an ultimate period). When we looked at the mortality rate by calendar year of death (Figure 3) and then by attained age (Figure 4), we saw that the mortality rate for the college-educated population was a reasonable proxy for the insured mortality rate in the ultimate period. We also observed that the select rate was significantly lower than the ultimate rate, implying that good protective value was already being derived from the underwriting process. 

It was also interesting to note the significant change in the cause-of-death rankings between the general population and the college educated population for ages 20-59 (Table 3). Substance abuse moved from first in the general public to fifth in the college educated sub-population for ages 30-39 and was not even in the top five for ages over 40.

Risk selection considerations

This analysis provides evidence that there may be a way to approximate the impact on life insurance mortality using the college educated subset of the population data that is readily available. However, there are some key filters that will be missing in the population databases e.g., smoking class, face amount, issue age and duration. In this section, we will provide some insights into these views using Munich Re’s proprietary cause of death experience study. 

Smokers

Typically, the ratio of all-cause mortality of smokers to non-smokers is between 200 percent and 300 percent. When looking at the same ratio for substance abuse only mortality, we see a materially higher differential. By count the ratio for attained ages 30-59 is between 485 percent and 582 percent, and the ratio by amount is between 470 percent and 778 percent. In the Journal of Addictive Disease⁶, the authors posit that one explanation is that cigarette smoking may be a gateway drug to illegal drug use. Another hypothesis might be that drug users are more likely than not to smoke.

Whatever the reason, this observation might lead underwriters to consider additional drug screening for applicants who test positive for cotinine or lead actuaries to build in some additional conservatism into future mortality improvement for smoker classes. 

Duration

We analyzed the substance abuse mortality rate by policy duration to see if the select period showed any evidence of early duration adverse-selection. Adjusting for age, we would expect that the mortality rate would start off lowest in duration one and gradually increase in an exponential curve. The data show this pattern with a noticeable selection effect and no evidence of adverse-selection. We also looked at the percentage of claims that are due to substance abuse, which shows a slightly decreasing trend. This is as expected since as underwriting wears off, medical causes of death such as circulatory system and neoplasm represent a larger percentage of total claims, pushing down the share of substance abuse claims. 

Issue Age

We analyzed the mortality rate and the percentage of claims that are due to substance abuse by issue age to see if there were any unexpected spikes in the pattern. For issue ages 10-19 the data show evidence of adverse-selection since there was a jump in both the mortality rate and the ratio of substance abuse deaths to total deaths. Otherwise the pattern is as expected.

Face Amount

The pattern of drug-related mortality rates by face amount shows a declining trend as face amount increases. This aligns with the premise that the higher one’s socio-economic status, the smaller the impact of the substance abuse epidemic. The percentage of claims that are due to substance abuse was relatively constant by face amount. 

Moving into the middle market

As companies aim to grow their distribution in the middle market, they need to keep in mind that this segment of the population may not behave as the current insured population. We may find that a different subset of the population data serves as a better proxy for middle market mortality experience and the hypotheses above may not hold for this group. Additionally, as the industry continues toward automated underwriting programs and away from fluid testing, we may be more vulnerable to adverse selection in all areas, including substance abuse risks.