Cancer risks: Are we getting them right?
Researching cancer mortality over the past few months has proved to be a bit of an eye-opener, and in three ways: firstly the level of excess mortality seen in a number of cancers, secondly the duration over which an extra risk persists, and thirdly that excess mortality may extend over a considerable period.
Some of the high excess death rates (EDRs) were noted stage II/III disease, where mortality was substantially higher than in localised. For example, from the US Surveillance, Epidemiology and End Results (SEER) Program1 relating to diagnoses between 2006 and 2012:
- Cervical cancer: five-year relative survival for localised and regional disease 91.3% and 65.7% respectively. Those survival percentages equate to EDRs of over 20 per mil and over 100 per mil.
- Kidney cancer: five-year relative survival for localised and regional disease 92.5% and 57.4% respectively, equating to around 15 per mil and 85 per mil.
- Ovarian cancer: five-year relative survival for localised disease was 93.7% at ages under 65 and 86.4% at ages 65 up, equating to about 12 per mil and 30 per mil respectively. For regional disease the figures were 80.4% (40 per mil) and 56.8% (over 100 per mil) respectively.
In some cancers mortality, even for localised disease, was sky-high; for example, and again from the SEER data:
- Small cell lung cancer: 27.5% relative survival at five years for localised disease; for non-small cell tumours 59.2%. Both percentages imply EDRs of well over 100 per mil.
- Liver cancer: relative survival at five years 30.9% for localised disease – again an EDR far above 100 per mil.
(The calculation of relative survival is based on deaths unrelated to the diagnosis being excluded; thus the deaths implied by a relative survival percentage are purely cancer deaths.)
Of course it is well understood that liver and lung cancer are associated with poor survival even in localised disease, often owing to the presence of latent tumour seedlings at the time of diagnosis. What may be less appreciated is that not all lung cancer deaths occur relatively early. The SEER data for all stages show deaths occurring 24 years after a diagnosis made in 1985-1989. The same is true at 15 years for cases diagnosed in 1998. The drop in relative survival at these durations is only a few percentage points but because of the small number of survivors (single percentage figures) the EDRs are high – over 40 per mil.
It is clear that liver cancer mortality is concentrated in the first ten years but the longer-term picture is unclear; deaths are few and far between but nevertheless could signify EDRs of over 20 per mil.
Ovarian cancer is another example of disease appearing to carry a long mortality tail. Dinkelspiel and colleagues2, in reviewing SEER data for up to 24 years from diagnosis, found that cumulative frequency of death curves flattened out after 11 to 15 years. Nevertheless deaths from ovarian cancer still occurred subsequently. In years 15 to 24, ovarian cancer deaths averaged approximately 30% of all deaths. Even among subjects with stage III/IV disease, deaths due to ovarian cancer persisted to 23 years from diagnosis.
Research studies also reveal that it is far from inevitable that survival curves flatten out significantly after two to five years, suggesting that a postponement period following diagnosis – to avoid heavy extra mortality in the early years and allow the better risks to manifest themselves – is not always appropriate. It is often the case, it seems, that either mortality levels are such that risks are manageable (ie acceptable) from after treatment and beyond or that risks are not insurable at all.
Our research covered much more than the SEER data and naturally outcomes varied from study to study and from country to country. But none of the other research material prompted us to doubt that some cancers – maybe not all but quite possibly more than one might think – are associated with very high mortality in the early years and/or a significant long mortality tail. Admittedly, long-term survival data is relatively rare but the SEER data are impressive on account of the sheer numbers involved, and none of the other material we examined provided a strong contradiction.
Findings such as these prompt one to reconsider underwriting philosophy for cancer risks. But first, can the statistics mislead the researcher? For example, could non-cancer deaths be under-reported in the SEER data? Possibly, but what to do about a mild suspicion? Could the treatment of cancer in more recent years have changed the picture for the better? Maybe, but there are three reasons to doubt there being grounds for great optimism: one, a lot of cancers have not seen any particular breakthroughs; two, in many instances survival data for more recently diagnosed cases point to persisting high mortality; and three, keeping more people alive for longer would push deaths forward to later durations, reinforcing the magnitude of any long mortality tail. As ever, there is no substitute for insured-life data but those are sparse and thus unreliable.
How indicative are the study outcomes of potential insured-life experience? The SEER figures represent deaths associated with the original diagnosis so the picture (faint doubt aside) is fairly clear. Where simple overall survival percentages were quoted we took into account background mortality from other causes of death. But clearly it is dangerous to take research findings from a clinical setting and apply them straightforwardly as life insurance ratings. Adjustment is needed to take account of the nature of an insured population (with its significantly lower mortality compared with the general population) and of the power of underwriting which can stratify risks that on the basis of statistics look the same, including those cases that have taken a turn for the worse.
But does the power of underwriting deal well enough with the patterns of mortality and EDRs illustrated above? The classic underwriting approach for cancer is a postponement period followed by a per mil extra for a fixed number of years, very occasionally accompanied by a permanent percentage extra mortality rating. This is predicated on there being a significant volume of early recurrences, either giving rise to deaths or revealing themselves to the underwriter as unfavourable risks; after the postponement period a cohort of identifiably better lives remains.
However, the research data suggest that in many instances:
- Survival curves, although flattening out in time, remain steep for a considerable period; nor is there exceptionally high mortality in the first few years
- EDRs are high, even in localised disease; regional involvement (let alone distant spread) means higher mortality still
- Extra mortality persists way beyond ten years; the notion that ‘if you’ve made ten or so years without a recurrence you’ve beaten the disease’ seems to be only partially true at best.
It looks as though postponement and temporary extras are not necessarily suited to the pattern of risk.
Can underwriting reliably weed out the risker cases? In some cases yes: medical evidence may reveal a recurrence. Ditto incomplete eradication, though one would imagine that the majority of better-risk cancers (stage IA/T1N0M0 for example) would be amenable to potentially curative treatment. But maybe underwriting cannot always be relied on; it is not unusual to find that overall survival curves and disease-free survival curves run very closely together, implying that death follows not long after a recurrence. And in any case, even if underwriting can select the better lives the mortality in that group may still be above the 20 per mil threshold that the industry (or is that just reinsurers?) finds comfortable.
Other worthwhile lessons seem to be:
- The importance of staging – apparently subtle increments in staging can make a big impact on survival; it may well be the case that a disease stage increase warrants declinature rather than one higher tumour rating classification. And within even a ‘localised’ group with high mortality there may be a stage sub-group that enjoys, on average a rather more favourable outcome
- The importance of tumour grade in determining prognosis; and grade does not always correlate with disease stage
- The effect of age on EDRs – not for all cancers but for some.
Rating cancer risks has come a long way over the last 40 years or so and the approach in underwriting manuals has become more sophisticated – so much so that risk calculators have been introduced. And with medical advances promising to make cancer more of a chronic disease than a killer, ratings in future will need to change in their severity and shape.
But even now some of the approaches to dealing with these risks look to be wanting. Risks we’ve traditionally thought of as insurable may not be necessarily so, and we may be under-rating risks at the later durations..
- Long-term mortality among women with epithelial ovarian cancer. Dinkelspiel HE et al. Gynecol Oncol 2015;138(2):421-8