By Tom McCarthy* and Doug Ingle†
Life is finite. That is one certainty we all understand. The final resolution… death. In many basic ways, this simple but universal concept is what fuels the life insurance industry. In most instances the unknown is when, during an individual’s life span, death will rear its ‘ugly head’. The concept of risk selection or underwriting exists in order to allow insurance carriers to try to ‘predict’ what a given individual’s remaining life expectancy might be. This comes from close review of medical and non-medical history either in detail (in traditional underwriting) or via data calls (‘accelerated’ underwriting US-style). Underwriting also tries to prevent anti-selection; when the proposed insured has any knowledge that might impact their life expectancy, an insurance carrier attempts to investigate and protect against this being used against them and, if necessary, makes it part of their assessment.
Let’s zero in on medical underwriting. Underwriters study whatever medical information they can acquire on a risk. They then make an assessment of the risk for all factors that might have an impact on life expectancy (mortality basically). This assessment is not one-sided of course. Review often includes medical information that might lead to better life expectancy (better mortality), such factors being acknowledged in ‘preferred’ underwriting.
In this process, underwriters typically use underwriting manuals which allow them both to leverage prior research on thousands of impairments and their severity, as well as apply consistent criteria to their assessments of risk. This sets a fair playing field. Underwriting is sometimes called an art rather than a science. This can be misleading as there is plenty of science in underwriting. The art may come into play in reviewing reams of information on a person’s medical history and their various impairments. Any underwriter will tell you that they rarely encounter a perfectly defined impairment in a person’s medical history. This is where seasoned underwriting judgement comes into play. The underwriter relies on their expertise and judgement and the guidelines in the underwriting manual they are using, to assign debits for an impairment. Judgement comes into play in assigning severity and timing and other factors in the impairment’s history and current testing results.
So, once an underwriter has made this assessment for impairments on an individual proposed insured, what happens next in underwriting? In a straightforward well defined case with three impairments, the underwriter might simply total the debits and assign an overall rating. +100 and +50 and +50 is 200 debits, to become a +200 or ‘Table 8’ or ‘Table H’ decision. One debit equals a 1% increase in mortality over baseline. Thus, +200 is a 200% increase in mortality. This increase in mortality is over a standard rating or 100% baseline and, when described as a multiple of standard, means adding the baseline to the rating: 100% (standard) plus 200% increase over standard, is three times standard mortality, or 300% of standard. Of course, underwriting is complex, and judgement is applied by the underwriter. Such a case might be assessed overall as a Table 6 or Table F based on the totality of the case, or it might end up a Table 10 or Table J based on the interaction of the impairments.
But that is where the discussion becomes interesting. The authors have observed a possible philosophical shift in the industry which, from a simple standpoint, might be termed: “You can only die of one thing.” This thinking looks at cases with multiple impairments and allows the worst rate to drive the overall assessment, rather than the additive approach noted above. In that particular example, an underwriter might rate the case Table 4 or Table D based on the impairment with the largest debits. This thinking is tied to the statement above; rate only the worst as you cannot die from all three impairments.
The authors will attempt to point out some of the fallacies of this type of approach using their varied yet distinct backgrounds: Tom from a more generalist, logical standpoint and Doug from a more analytical, actuarial-based one.
The biggest fallacy to start with is the idea that underwriters really underwrite individuals. This is tricky as it certainly appears that way. But the truth is that the underwriter is actually underwriting a group of like individuals. It is not really possible for an underwriter to predict completely an individual’s life expectancy: there are too many variables. What they can do is underwrite for a group with similar characteristics. In the three-impairment example above, a group of similar individuals will have mortality that is consistent with all of the impairments. The individuals in the group may die of any of the impairments or wholly different causes. Underwriting manuals can help the underwriter place the risk in the cumulative risk class that factors in all the impairments. It is the expected mortality of these like groups that underwriters work towards.
Underwriters that have reviewed claims through their career often see the cause of death not necessarily correlate with the original underwriting and the impairments noted. That happens with regularity. Another key factor is the complex relationship between impairments and cause of death. When a person has diabetes and hypertension and is overweight and dies of a heart attack or even from kidney failure, what exactly is the ‘one thing’ they actually died from? A difficult call in most situations. In many cases it is a combination of all of the above.
For many years, our industry was seduced by the concept of ‘table shaving’. Table shave programmes came into vogue years ago as a way to facilitate the issue of cases with low substandard ratings. These cases were ‘shaved’ to standard (residual standard) with the thought they were ‘close enough’. The reasons were myriad: competitive edges, ease of issue of standard cases, aggressive pricing. In time however, table shaving was exposed as not a very viable concept. It took a while, but the industry gradually has moved away from it. One interesting criticism was always that table shave programmes had a key fault in the idea that exact underwriting was not truly critical if you frequently ignored a true rate to place a case. For an industry that prides itself on accuracy of risk selection, evidence-based ratings and detailed underwriting guidelines, it has always seemed counterintuitive to ‘drop’ a rated case to standard without actual evidence to support it. It falls under the heading of ‘talking from both sides of your mouth’ and in effect making underwriting less scientific. Assessing cases at a rate for only their worst rated impairment smacks of the same unscientific, naïve approach and can hurt the ability to underwrite in the long term.
And of course, there is the ‘lure’ of being competitive. No underwriter is immune to this of course, but degrees vary hugely by carrier. The problem is and always will be the impact on mortality. Individual case impact can sometimes seem ‘insignificant’ to an underwriter reviewing case after case. But that is where the overall impact can be huge, particularly with any blanket approaches to ‘shaving’ mortality off cases.
There’s a significant body of literature on this subject that seems to have been disregarded as companies focus merely on ways for underwriters to make more aggressive offers without considering the science of underwriting. For example, the Society of Actuaries, in conjunction with HOLUA-IHOU, the erstwhile Mortality and Morbidity Liaison Committee and the Medical Insurance Bureau, published a text in 1998 called the Multiple Medical Impairment Study. In 2007 the Journal of Insurance Medicine published two articles on Mortality in Co-morbidity. The results were a bit mixed and equally distributed, meaning debit results were additive, more than additive, and less than additive roughly one-third of the time for each category. There was no definitive approach to be gleaned but that meant two-thirds of the time results were additive or more than additive and only one-third of the time less than additive.
In 2017 Fasano Associates published an article in their Newsletter by Dr Jochen Russ reporting an analysis of multiple impairments and mortality. Overall, when they added debits for all impairments, the actual mortality rate came in at 96% of expected. When they applied debits only for the largest impairment and ignored the rest, the overall actual to expected came in at 127%.
During the 2022 AHOU meeting I presented some original findings of my own. I set the stage with a slide quoting W. Edwards Deming: “Without data, you’re just another person with an opinion.”
To adhere to this insightful quote, I used the NHANES Continuous datasets spanning the years 1999 to 2013 with exposure through 31 December 2015. There were 47,279 individuals, 6,314 deaths and 383,082 exposure years in the study population. The analysis looked at three impairments admitted to at time of exam. They were coronary artery disease (CAD), diabetes (DM), and chronic obstructive pulmonary disease (COPD). Applying the 2015 Valuation Basic Table (published by the Society of Actuaries) as a standardised expected, the overall mortality for each impairment was:
- CAD: 53 debits
- DM: 70 debits
- COPD: 94 debits.
Subsets of individuals in the population with more than one diagnosis allowed for a study of results for multiple impairments. The mortality rates, described as debits, for the combinations, were as follows:
- CAD + DM: 116 debits
- COPD + DM: 187 debits
- COPD + CAD: 120 debits.
With the mortality studies completed we can look at what would happen if we added debits:
- CAD + DM: Expected debits: 53 + 70 = 123. Actual debits 116; off by -7
- COPD + DM: Expected debits: 94 + 70 = 164. Actual debits 184; off by +20
- COPD + CAD: Expected debits: 94 + 53 = 147. Actual debits 120; off by -27.
If the underwriter ignored the lowest debit finding the results would be:
- CAD + DM: Expected debits 70. Actual debits 116; off by +46
- COPD + DM: Expected debits 94. Actual debits 184; off by +90
- COPD + CAD: Expected debits 94. Actual debits 120; off by +26.
Every instance where the underwriter rates just for the worst impairment results in significant under-weighting of the true mortality risk. When adding debits, results are either about right or slightly high or low within about a one table deviation in findings.
As stated at the outset, the life insurance industry is grounded in statistics and probability theory. We can find what we should do using these fundamentals. The probability of rolling a one with a six-sided die is one possibility out of six options (1/6). So long as events are independent, additional probability outcomes, which are a subset of the universe of outcomes, are additive. Let’s say now a hit occurs if the person rolls a one or a two. These outcomes are independent. You can’t roll a one and a two on one throw of the die. You can roll a one, two, three, four, five or six. Thus, the probability of rolling a one or a two is 2/6.
Let’s say a standard risk death rate is one death per one thousand individuals, thus a probability of 1/1000, which is like using a one-thousand-sided die where one side has a dot. There is a one in a thousand chance of a hit. If there is a second risk, such as private flying, where we assign a $3 per thousand flat extra (which means we expect a death rate of three deaths per thousand for that risk), then three other surfaces out of the one thousand also need a dot. That means the new grand total probability of death is four deaths per thousand (1/1000 + 3/1000 = 4/1000). Underwriters understand that is the expected mortality.
But what should the underwriter do if that person is also a scuba diver with an underwriting rating of $2/1000? That person can’t be flying and scuba diving at the same time. Risks are independent and thus additive. In other words, probability theory tells us not to blend flat extras. To Tom’s point above, remember this is now a population of 1,000 individuals that each fly enough to warrant a three-dollar flat extra, and scuba dive enough to warrant a two-dollar flat extra. All one thousand individuals partake in both avocations to this degree and thus the population, according to probability theory, should produce on average, 1 + 3 + 2 = 6 deaths per thousand per year. It’s understandable there could be some overlap in risk for medical impairments, but flat extras for avocations are truly independent risks and additive.
It is also important to understand the medical literature. In an ideal world, researchers would compare two identical populations where the only difference is the disease being studied. That is unrealistic and virtually impossible to do, because every person with a unique comorbidity would have to be thrown out of the study, making the study population incredibly small. What medical researchers do is keep all subjects in the study and adjust for comorbidities using multivariate regression.
The model I am most familiar with is the multivariate Cox proportional hazard model, ‘multivariate’ meaning multiple variables are accounted for in the study findings. By adjusting for the influence of comorbidities, the authors can parse out the unique mortality risk associated with each variable being studied. Think of an unadjusted model, a univariate study of BMI, exhibiting 100 debits, a hazard ratio of 2.0, for a BMI of 38. We know that, as BMI increases, so does the likelihood of hyperlipidaemia and high blood pressure. If, after conducting a univariate study, the researchers then conduct research that controls for high blood pressure and hyperlipidaemia using a multivariate Cox proportional hazard model, they have subtracted out the overlapping debits that should be assigned to lipids and blood pressure.
Often the BMI debits for a fully adjusted (multivariate) model will be less than those 100 debits found in the univariate model. It is incredibly important to understand this fact and look at how the authors conducted their study and what they controlled for. If they control for BP and lipids, then if ratings are present for BP and lipids, the underwriter should, at a minimum, add the ratings for these items to the rating for BMI. In reality, when more than one risk is present in a Cox model, the hazard ratios are multiplicative according to the fundamentals of the Cox formula, and that story is now taking us down a path that goes beyond the scope of this article.
The authors primarily wish to remind underwriters of the significant power intrinsically embedded in everything they do and every decision they make. In a world where, simplistically, an insurance carrier or reinsurer makes a profit based on the relationship between mortality, expenses and investments, mortality (ie underwriting) continues to play a very significant role. To that end, underwriters should continue to be vigilant in protecting their individual-case risk selections. ‘Holistic’ case review will always be important; just don’t sacrifice them for quick blanket approaches that might impact profitability.
And SelectX’s take:
Thank you, Tom and Doug, for a highly informative and thought-provoking article. It is always worth being reminded that corporate and even industry memories can be short; there is a wealth of information out there supporting the additive or even multiplicative effects of multiple risk factors. Cardiovascular disease risk calculators used in primary healthcare take account of multiple risk factors: the more adverse risk factors you have, the greater the risk. Do the ‘only die of one thing’ practitioners reject those calculations and the underlying data? This is patently a nonsensical idea.
And finally, the point about individual underwriting decisions in the portfolio overall is well made. First, in general, one decision is neither here nor there, but decisions in their thousands materially influence the experience of the portfolio. Second, think not about the risk for the individual applicant (you can’t possibly know enough about him or her plus, unless you have special powers, you can’t foretell the future) but about that for a thousand or more similar lives based on the evidence at hand.
*Tom McCarthy is a career underwriting executive now retired/consulting. The majority of his career involved reinsurance and involved working with most of the carriers in the US and Canada on traditional and accelerated underwriting and technology.
†Doug Ingle FALU, FLMI, is a former Vice-President of Underwriting for several reinsurers and direct carriers in the US. Currently semi-retired and the President of Doug Ingle Underwriting Research LLC, he is also an assistant basic mortality methodology instructor helping doctors learn how to translate the clinical literature into underwriting guidelines.
Both Tom and Doug have been inducted into the Association of Home Office Underwriters’ ‘Hall of Fame’.