Analyzing insured mortality by cause of death (COD) often provides insights that a typical actual-to-expected study may not reveal. In Part 1 of this series we utilized COD statistics from the proprietary Transamerica Experience Database (TED) to gain insight into mortality trends. This article continues our analysis and compares the TED insured population experience against the National Center for Health Statistics U.S. population experience.
The current TED database includes reinsurance claims incurred between January 1, 2004 and July 31, 2008. This provides enough data to analyze patterns by incurred claim year, as shown in Table 1 below. Note that the percentage of deaths from the two leading causes, cancer and cardiovascular disease, have remained remarkably stable over the four and one-half year period. In fact, the distribution of most leading causes has remained relatively level. While the percentage of claims from suicide has increased slightly from 3.9 percent to 4.9 percent, the total from all non-disease causes (trauma, motor vehicle, and suicide) has actually declined from 15.5 percent to 12.8 percent.
|Table 1: Distribution of
Selected Causes by Incurred Claim Year|
|Cause of Death Category||2004||2005||2006||2007||2008|
|Cancer||37.6 %||36.9 %||36.2 %||36.0 %||37.4 %|
|Cardiovascular||25.9 %||26.4 %||27.3 %||28.3 %||25.8 %|
|Respiratory||4.8 %||5.8 %||7.2 %||6.3 %||6.0 %|
|Trauma/Injury/Poisoning||6.0 %||5.6 %||4.9 %||4.1 %||4.2 %|
|Motor vehicle accident||5.6 %||4.8 %||4.3 %||4.6 %||3.8 %|
|Suicide||3.9 %||3.7 %||4.2 %||4.3 %||4.9 %|
|Cerebrovascular||3.1 %||3.6 %||3.3 %||3.3 %||3.0 %|
|Influenza and pneumonia||2.7 %||2.2 %||2.6 %||3.0 %||3.3 %|
The two main causes of death have remained steady over the study period
By analyzing the distribution of causes by policy year, we can examine the long-term effect of industry underwriting practices on a cohort of newly issued insureds. To minimize the variability in cause distributions due to gender and issue age differences, Table 2 shows only the experience for male nonsmokers issue age 45-54.
|Table 2: Distribution of
Selected Causes by Policy Year Male Nonsmokers Issue Ages 45-54|
|Cause of Death Category||1-2||3-5||6-10||11-20||21+|
|Cardiovascular||32.8 %||27.6 %||26.9 %||30.1 %||35.3 %|
|Cancer||26.5 %||36.3 %||42.0 %||38.9 %||31.7 %|
|Motor vehicle accident||14.5 %||6.7 %||4.6 %||2.2 %||0.9 %|
|Trauma/Injury/Poisoning||10.1 %||6.4 %||4.6 %||3.1 %||1.7 %|
|Cerebrovascular||3.3 %||2.7 %||2.5 %||2.7 %||4.6 %|
|Homicide||2.6 %||0.9 %||0.6 %||0.1 %||0.0 %|
|Respiratory||2.8 %||3.5 %||4.0 %||6.5 %||9.3 %|
|Influenza and pneumonia||0.7 %||0.7 %||1.4 %||3.5 %||3.2 %|
Current testing is better at identifying cardiovascular disease than cancer
Notice how the proportion of deaths due to cardiovascular disease is relatively constant from policy year to policy year, whereas the proportion of cancer deaths increases dramatically over the first 20 years. Part of the cancer increase is due to the aging cohort. However, it also indicates that while underwriting is effective in determining current cancer risks, it is less useful in identifying long-term cancer profiles. On the other hand, the relative stability of the cardiovascular disease percentage suggests that the cholesterol and blood pressure criteria used to classify “preferred” risks remains effective over a long period of time.
A rather surprising pattern is the percentage of deaths from motor vehicle accident, trauma/ injury/poisoning and homicide. While the declining percentages in policy years 3 and up are certainly due to the aging of our 45-54 year old cohort, the very high proportions in policy years 1-2 are quite unexpected. This pattern repeats itself for many different age groups and to some extent for both genders. Because these deaths occur during the contestable period, one possible explanation is that some of them are misidentified suicides. At the very least, further investigation is needed.
Insured versus US Population
Using 2006 data from the National Vital Statistics System (NVSS), we compare COD statistics from the TED insured population against the general population of the United States. Because the two data sources have very different age and gender mixes, the experience in Table 3 shows only males age 35 to 44. Also, TED and NVSS use different cause of death classification systems, so a true apples-to-apples comparison is a bit obscured.
However, it is clear that the leading causes of death from TED are also the leading causes from NVSS. It is also clear that underwriting is very effective in identifying and reducing our exposure to risks that will eventually succumb to AIDS/HIV, diabetes mellitus or renal/genitourinary disease.
|Table 3: Distribution of Selected Causes for Males Age
at Death 35-44|
|Cause of Death Category||TED||NVSS*|
|Cardiovascular||20.8 %||16.5 %|
|Cancer||20.5 %||11.5 %|
|Suicide||14.9 %||9.7 %|
|Trauma/Injury/Poisoning||14.5 %||23.8 %|
|Motor vehicle accident||13.9 %|| |
|Homicide||3.3 %||4.4 %|
|Respiratory||2.7 %||0.8 %|
|Cerebrovascular||2.0 %||2.2 %|
|Septicemia||1.4 %||0.9 %|
|Influenza and pneumonia||0.9 %||0.9 %|
|Renal/Genitourinary||0.5 %||3.2 %|
|Diabetes mellitus||0.2 %||2.5 %|
|AIDS/HIV||0.2 %||5.4 %|
*Source: CDC/NCHS, National Vital Statistics System. LCWK2. Deaths, percent of total deaths and death rates for the 15 leading causes of death in 10-year age groups, by race and sex: United States, 2006.
Given similar leading causes of death between insured and general populations, we have some assurance that US general population mortality improvement may serve as a proxy for long-term trends in the insured population. If both populations are dying from the same causes, then medical and other advancements that reduce deaths from these causes should apply similarly to both.
Likewise, environmental factors that increase mortality in the general population, such as the recent rise in obesity, may also apply to insured populations to the extent that they are not detected during underwriting.
A cause of death analysis can be a useful adjunct to a traditional insured mortality experience study. Unexpected anomalies in COD distributions across age, gender, product or underwriting class may help identify potential anti-selection. Finally, comparison against general population experience can gauge the effectiveness of our underwriting practices.