How useful are lifetime risks of disease?

Margaret McCartney asks "how useful are lifetime risks of disease?"
(BMJ 2011; 342:d1046) Implicitly, she considers them to be confusing and
uninformative. Whilst I agree that lifetime risk lacks the detail and
precision required for scientific publication or clinical consultation, it
does provide a useful and understandable summary of risk in a population.

The distinguished statistician, David Spiegelhalter, is quoted in Dr
McCartney's article. He is seemingly opposed to the use of lifetime risk
in all situations, but I expect that like me he would not recommend
lifetime risk for communicating an individual's risk, but would use them
to describe the population burden of disease. Indeed, in a December 2009
article he wrote "One in nine women will get breast cancer in her
lifetime" (http://plus.maths.org/issue53/risk/index.html). Dr McCartney
writes lifetime risk is "a hot topic in cardiovascular risk", but omits to
inform us that it is being used increasingly in the field (Eg Hippisley-
Cox et al BMJ 2010;341:c6624) and may replace the 10-year risk in the
Joint British Societies cardiovascular risk calculator (Pulse 21 April
2010). The lifetime risk of breast cancer is a measure of the proportion
of the women you know who (have had or) will get breast cancer.
Lifetime risk can be criticised, but so can every (univariate) measure of
risk. The 10-year risk, seemingly favoured by Dr McCartney, only makes
sense if one has a table of risks at different ages. Averaging the age-
specific risk (as is done in the age-standardised rate) is less
informative because it mixes the risk of a child (essentially zero for
many diseases) with that of a 75 year old. It reminds one of the
statistician who, with one foot in a bucket of ice water and the other in
a bucket of scalding water, said that on average the water is pleasantly
warm!

Dr McCartney quotes from Professor Spiegelhalter: "The crucial thing is
early preventable death". On the surface that is something about which we
can all agree, but an otherwise healthy 78 year old woman may feel quite
differently particularly given the poor survival in elderly breast cancer
patients in the UK compared with other countries. Further, one may ask
what constitutes an early preventable death. My view of what is an early
death has certainly changed since I was a child, and what is preventable
has changed dramatically in the last 70 years ago and many diseases that
are not preventable or curable today will be by 2075.
We are told that there is interest in "how a disease impacts on lifespan".

Saying that on average a smoker dies 10 years earlier than a non-smoker is
useful, but to say (for instance) that on average women die 7 months
earlier than they would if there was no breast cancer is, in my opinion,
not. Both are population averages, but it is easy to relate to a
population of non-smokers and a non-smoker can think about how starting
smoking would affect her lifespan. By contrast there are no acceptable
options to enable a population to gain those 7 months lost to breast
cancer. Most (~96%) of women don't die of breast cancer. A very few
tragically die in their thirties perhaps fifty years earlier than they
would have in the absence of the disease. That is why both risk and
expectation are important. The average "winnings" from the EuroMillions
lottery (per ?2 bet) is about minus ?1 (i.e., a loss). People take part
because of the very small chance (about 1 in 76 million) of winning the
Jackpot.

Even at a population level, lifetime risk is unhelpful when applied to
people with a particular risk factor. Heavy smokers have a smaller
lifetime risk of breast cancer (than non-smokers) - their lifetime risk is
less because their lifetime is shorter (smoking is not a risk factor for
breast cancer). Similarly, the increased breast cancer risk in (post-
menopausal) obese women will be counterbalanced by their shorter life
expectancy.

I agree that various factors including increasing longevity and breast
screening will have had much to do with the increase in estimated lifetime
risk of breast cancer over the last decade. Breast screening works by
diagnosing cancers that would not have become symptomatic for some time.
As a result, the incidence of breast cancer in the days following
screening is extremely high, but the incidence thereafter is very low,
gradually rising to that of unscreened women over a number of years. The
population effect is to increase the incidence in the age-groups in which
screening is offered. A reduced incidence in older age groups will take
some years to manifest.

Dr McCartney cites a statistic regarding harms and benefits of breast
screening from the Cochrane Reviews as if it were gospel. There is a huge
(25-fold) variation in the estimated ratio of harms to benefits in the
published literature - and the Cochrane estimate is at one extreme. It is
ironic that such a contentious and media-grabbing statistic should be used
as fact in an article on informative statistics.

I do not claim to speak on behalf of all "who deal with the concerns of
normal people", but it is my view that the fact-sheets provided by many of
the cancer charities in this country are excellent and that they provide
clear, balanced and up-to-date information that is useful for both the
public and GPs.