Blog

Opioid Dispensing and Group Life Mortality Experience

By Jonathan Polon on 2/13/2020

Key Findings

1.   Per-capita levels of opioid dispensing are predictive of county-level group life mortality experience.
2.   Area-specific group life mortality risk can vary by age. As a result, carriers may misprice group life insurance for cases where the age distribution differs from average.

Background
From 2010-2017, midlife all-cause mortality rates increased by 6%. A major contributor has been an increase in mortality from specific causes (eg, drug overdoses, suicides, organ system diseases) among young and middle-aged adults. How does this trend affect group life mortality?

Methdology

  1. Determine the average annual opioid dosage units per-capita for each county by averaging the number of units dispensed for each county from 2006-2012, as specified in the dataset published by the Washington Post, and dividing by the county population from the 2010 US census.
  2. Classify each county into one of five groupings, by the amount of per-capita opioid dosage units.
  3. Calculate actual vs. expected mortality for each of the five county groupings, using two different bases for expected mortality:
    1. Determine expected mortality without regard to area risk. Expected mortality is an age/gender table with multiplicative risk factors for 8 salary bands and 12 industry groupings. Multiplicative risk factors were determined using a Generalized Linear Model.
    2. Determine expected mortality with the inclusion of area risk factors. Expected mortality is an age/gender table with multiplicative risk factors for 8 salary bands, 12 industry groupings and 102 area groupings. The area groupings included the 51 Metropolitan Statistical Areas (MSA) with more than 300 deaths in the database, as well as the remaining areas for each state, Puerto Rico and International. Multiplicative risk factors were determined using a Generalized Linear Model.
  4. If the actual vs. expected mortality ratio increases (decreases) as opioid dispensing rates increase, then the conclusion is that opioid dispensing rates are predictive of group life mortality rates – even after accounting for the other risk factors that are included in the expected basis – and that mortality rates increase (decrease) as dispensing rates increase.

Results

  1. Group Life mortality is correlated to county-level rates of opioid dispensing. The chart below compares actual to expected mortality rates for counties, grouped by the per-capita rate of opioid dispensing. For this chart, the expected mortality basis is a function of age, gender, salary and industry. The chart clearly demonstrates that opioid dispensing correlates to a significant amount of residual mortality risk that is not correlated to the other rating factors – age, gender, salary, industry:
  2. Area risk factors will capture most of the county-level mortality variations that are correlated to opioid dispensing rates. Like the previous chart, the chart below compares actual to expected mortality rates for counties, grouped by the per-capita rate of opioid dispensing. For this chart, the expected mortality basis has been refined to also include multiplicative risk factors for area. As compared to the prior chart, this chart shows only a small correlation between mortality experience and opioid dispensing rates. Ie, most of the correlation between opioid dispensing rates and mortality experience is reflected in the calculated area risk factors:
  3. The area risk factors are most successful at capturing mortality risk correlated to opioid dispensing for individuals aged 45 or greater. Thus, there is only a small correlation between mortality experience and opioid dispensing rates for ages 45 or greater. All-cause mortality is higher at older ages and so it may be that any additional deaths correlated to opioid dispensing have proportionally less impact on the overall mortality rate than for younger ages.
  4. The area risk factors do not capture a significant amount of the mortality risk correlated to opioid dispensing for individuals aged less than 45. Thus, the chart below shows a high degree of correlation between mortality experience and opioid dispensing rates. All-cause mortality is lower at younger ages and so it may be that any additional deaths correlated to opioid dispensing have greater proportional impact on the overall mortality rate for younger ages.
  5. The calculated area risk factors are highly correlated to per-capita opioid dispensing rates. The chart below shows the weighted-average area risk factor by per capita dosage rates:

Conclusion

Existing methodologies for calculating area risk factors will quantify opioid-correlated mortality risk for Group Life insurance. However, the nature of this risk varies by age; thus, area risk factors may misestimate the risk for groups with age distributions that deviate from average.

Data Sources

  1. In July 2019, the Washington Post made publicly accessible part of a Drug Enforcement Administration database that tracked the path of every pain pill sold in the United States between 2006 and 2012. From this dataset, we were able to get a summary of dosage units by county for the years 2006-2012.
  2. As part of its 2016 Group Term Life Experience Study, the Society of Actuaries made publicly accessible the underlying data – representing the years 2010-2013 – at an aggregated, but sufficiently granular, level.
  3. 2010 US census data, such as population by county.

Acknowledgements

  1. Thank you to the Washington Post for making the opioid data publicly accessible. Here is a link to their original article: https://www.washingtonpost.com/graphics/2019/investigations/dea-pain-pill-database/
  2. Thank you to the Society of Actuaries and its Group Life Insurance Experience Committee for making the Group Life experience study data publicly accessible.

Next Steps

We would like to study trends in Group Life mortality experience, to determine if the mortality rate increases observed in population mortality have also occurred in Group Life mortality. Unfortunately, the SOA Group Life experience study only covered a four-year horizon, 2010-2013. Not only is this time period insufficiently long for a trend analysis but it is also out-of-date. We will work with the insurance industry to collect the data required for this additional analysis.