[Skip to content]

Sign up for our daily newsletter
The Actuary The magazine of the Institute & Faculty of Actuaries

Critical illness

T en years after the original launch of the Continuous Mortality Investigation’s (CMI) critical illness (CI) investigation, the results for the first quadrennium (19992002) have landed on member offices’ desks. This article provides background on the methodology used; more detailed results will be published in a CMI report later this year. Full details of the methodology are contained in Working Paper 14, available on the profession’s website.

The 19992002 data
The investigation compiled a substantial set of data from 16 offices, comprising 7.4 million life-years’ exposure and almost 12 thousand claims. The split of exposure data on a lives basis illustrates the nature of the underlying business (see table 1).
The data in the investigation have grown considerably during the four years, with the exposure and number of claims both increasing by over 20% a year. This is the result of new offices joining the investigation, offices extending the range of products included, and underlying business growth.
The data remain immature, with a greater weighting towards younger ages and shorter durations than one would see in a more mature investigation. As a result, the investigation cannot yet be considered a sufficiently robust basis for graduation and the production of a recognised table.

Claim dates and methodology
The CMI requests four dates for each claim diagnosis, notification, admittance, and settlement in order to gain understanding of the claim delays. The delay patterns for some of the more common causes of claim are shown in figure 1. Death claims are subject to shorter delays than most of the CI events. Cancer and heart attack the two largest causes of CI claims exhibit very similar delay patterns. Stroke claims generally take longer to settle. Total and permanent disability (TPD) claims perhaps surprisingly actually show a significant proportion settled very quickly, with around 30% settled within two months of diagnosis.
The date of diagnosis is the most appropriate to use as the ‘date of claim’ as it matches the exposure and the risk incurred by the office. However, using this creates problems:
– analysis would involve long delays if we waited until all the claims have eventually been settled;
– claims are submitted by offices according to the year in which they are settled, not the year in which they are diagnosed;
– just over half of records gave the date of diagnosis;
– for some Cl events, the date of diagnosis may be unclear
An example of the last problem is for cancer: is it the date symptoms are detected by the GP, or the date that a diagnosis is confirmed by the consultant? For other events such as multiple sclerosis (MS) and TPD, some claims appear to be allocated a date of diagnosis based on the occurrence of the original event that gives rise to the disability, while for others it is recorded as the date the office establishes the total and permanent nature of the disability. Therefore ‘date of diagnosis’ may vary between offices or between assessors within an office.
Our methodology uses date of diagnosis to calculate the age and duration at the time of claim. However, ‘actual’ claims are those settled, not diagnosed, in the quadrennium. This would not cause a problem with a stable portfolio, but, as noted earlier, the number of claims in the investigation has increased rapidly year-on-year. The effect of this is an understatement of the results, which will depend, among other things, on the growth in the number of claims for an individual office. Overall the All Office results for the 19992002 quadrennium are under-stated by around 15% (the ‘grossing-up factor’).
Where offices were unable to supply dates of diagnosis, these had to be estimated using average delay patterns derived from other claims. When the 1998, 1999, and 2000 analyses were previously released, the observed average delay between diagnosis and settlement of 155 days was used to estimate the date of diagnosis where it was unknown. We expected the average delay to lengthen as the exposure to longer-delayed claims increased. This has happened; the simple average observed delay is now 176 days. We expect this to continue to lengthen until the number of claims per year stabilises and accordingly we have developed a more robust approach to estimating the missing dates of diagnosis. We have also developed a model of the underlying delays, which suggests an underlying average delay of around 260 days. The observed delays and our model of underlying delays are illustrated in figure 2.

19992002 results
We have used CIBT93 to calculate the expected claims. CIBT93 was derived from aggregate population data for 1993, so is not adjusted for insured lives or smoker/non-smoker status. As with other CMI investigations, the experience of individual offices differs significantly from the All Office experience individual office quadrennium results for male accelerated business vary between 50% and 125% of the All Office figure. The numbers shown make no allowance for grossing-up factors; these have not yet been calculated for subsets of the data. This could affect any apparent differences in the raw results.
The All Office experience is summarised in table 2. This overall level is approximately that for previous analyses conducted by the Critical Illness Healthcare Study Group for 199197. Amounts experience is lighter than lives experience for accelerated business. Stand-alone business has much smaller volumes of data, but experience appears heavier compared to CIBT93 than accelerated for lives, and amounts experience is heavier than lives.
There has been much debate about the existence of a selection effect in CI experience. Table 3 shows that the All Office results appear to indicate a weak selection effect. The length of the select duration is, as yet, unknown and may vary by cause of claim. Therefore, the duration 2+ results are not a reliable indicator of eventual ultimate experience.
It is likely that some CI events exhibit selection while others do not and this could explain the apparent difference between males and females. For example, stroke is certainly displaying a selection effect and is a more significant cause of claim for males than for females. TPD experience increases markedly with duration. This may be because of the substantial claim delays and the likelihood that, for many claims, a decision is deferred until the disability can be considered permanent. As a result, there could be a very long select period on TPD, with both select and ultimate experience substantially higher than is currently suggested.

Looking forwards
The good news is that the profession will not have to wait another ten years for the next set of results, now that the CMI’s data requirements are more flexible and the data volumes are more firmly established. The committee expects to be able to produce more refined analyses over time and, hopefully, a graduated table soon.