Cancer survival is a poor indicator of quality of careBMJ 2015; 351 doi: https://doi.org/10.1136/bmj.h4059 (Published 29 July 2015) Cite this as: BMJ 2015;351:h4059
All rapid responses
Thank you for your comment.
I should have specified that the increase in survival for prostate cancer in France from 1990 to 2002 concerned the percentage of survivors at 5 years. Indicators currently used for cancer are survival at 5 years and at 10 years. You can imagine this indicators are under the influence of lead time bias.
In addition, cancer is deemed to be a chronic disease, and there is no end to it. Every person diagnosed with cancer is considered a survivor until he dies. So the formula about prevalence and incidence doesn’t apply.
In fact, the number and the percentage of survivors of prostate and breast cancer have skyrocketed in France like it has in other countries where screening has been recommended or put into practice (opportunisitic or individual screening)
Partial prevalence at 10 years for breast and prostate cancer in France in 2012 was respectively 47% and 45%. Meaning that in 2012, 383 310 surviving women more than 15 years old diagnosed with cancer during the past ten years or 47%of all the women diagnosed with cancer during this period had breast cancer. For men the figure is 402 480 and 45%. Breast cancer accounts for approximately 30% of incidental women cancers and prostate cancer for 25% of incidental men cancers.
In addition there are not such precancereous stages for prostate cancer as for breast cancer. All cancers detected even ductal carcinoma in situ are considered as potentially lethal and treated at least by surgery, radiotherapy and chemotherapy or hormonotherpay in the case of breast cancer. For prostate cancer, patients will have a biopsy that produces harms by itself. Some refinements have been introduced in diagnostic overtime trying to differentiate aggressive cancers and not aggressive ones but their influence on treatments is partial. This absence of precancerous condition and differences in growth speed between different tumors leading to length time bias and also metastatic potential make the difference with cervical cancer which is adapted to an effective mass screening.
There is no consensus untill now about this matters but it is noticeable that the tasks forces and the independent organizations of most countries have spoken against screening (PSA and prostate cancer) or extension of screening (breast cancer) and there are only academic or professional societies with major conflicts of interests or doctors with the same conflicts of interests still defending continuous extension of screening by highlighting the misleading argument of cancer survival.
Anyway it is essential that patients benefit from most accurate information.
Competing interests: No competing interests
It seems surprising that cancer survival would be a poor indicator of performance? Maybe one is surprised because survival is not only a poor indicator but is also misleading.
A statistician from the Oxford Nuffield department of primary care, Jason Oke, said it first a year ago in the BMJ,  although his letter didn’t prevent this indicator to be used more and more for cancer statistics.
Screening for cancers causes two mechanical and inevitable results: it increases survival duration by diagnosing cancers earlier even without changing age of death--that is, the lead time bias--and it increases the number of cancers diagnosed. Although the word “survival” sounds well, and therein lies its potential deception, if we think about it we will soon discover that living longer with a diagnosis of cancer is not good news.
It is difficult to compare countries because every country uses different indicators but, for instance, the median age at diagnosis in the United States where PSA (prostate specific antigen) has been most widely used for screening for prostate cancer is 66 years in 2012. That is really early since the median age at death for this particular cancer is 80 years in 2012 . In France, where opportunistic screening has been less widely used, the median age at diagnosis was 70 years and median age of death in 2009 was 83 years . The world age standardized incidental rate for prostate cancer has increased rapidly in the United States as in France, reaching a maximum of nearly 125/ 100 000 in the early 1990s in the United States and in 2005 in France . However, in 2012, the European age standardized mortality rate (ASMR Europe) was lower in Italy, 14,1, than in France 17,7, although Italy had a much lower incidental rate of prostate cancers than France because of less screening (100,4 and 144,7 respectively for Italy and France).
So, survival rates tell us nothing about outcomes of interest in view of an effective preventive public health intervention. From this perspective, and from an ethical perspective, desirable outcomes would be more people having longer lifetime without health impairments and a reduction in cancer mortality. Instead, earlier widespread diagnosis and longer survival time means more people live longer with health impairments.
It is quite clear when one observes the epidemiologic of cancer in France using age standardized figures which neutralize the effects of age structure of the population. The general trend in France, like in other rich countries, is an increasing of all cancers incidence and a reduction in all cancers mortality. In France the risk for a man to be diagnosed with cancer in terms of age standardized incidence from 1980 to 2005 increased by 113/100 000. The risk of being diagnosed with prostate cancer increased by 102/100 000 in the same period. That means the increase in the incidence of prostate cancer diagnosis accounted for 90% of all the increase in incidence of men cancers in France during this period. However the reduction in prostate cancer mortality was slower than the reduction for all cancer mortality (1,3 vs 1,5% a year).
The same calculation can be done for breast cancer. The incidence of breast cancer increased by 42,5/100 000 and this accounted for 59% of the increase of age standardized incidence for all women cancers in France. And the reduction of mortality was 0,6% every year while the reduction in all cancer mortality during this period for women was 1% .
Using cancer survival for assessing health systems performance also contributes to ignoring the downsides of widespread screening. Because screening tests, and also diagnostic tests, are not perfect. Screening a large population then implies the risk of false positives and false cancer diagnosis, not to mention overdiagnosis (overdiagnosis is a correct diagnosis of a disease that would never have had clinical consequences).
False positives, for instance, are the natural consequence of imperfect screening tests. If you screen a population of 100 000 persons for a disease that has a prevalence in this population of 1/1000 with a test that has a sensitivity of 90% and a specificity of 90% what would be the chances a person who is positive would have the disease? As the test has a sensitivity of 90% it will detect 90 persons among the 100 with the disease. As it has a specificity of 90% it will correctly detect 90% of the 99 900 who don’t have the disease, that is 89 910. But there will be 9990 false positives. So the chances you have the disease when you are positive to this test are only 90/ 10080, that is 0,9%.
That is why the chances of having a false positive among 1000 women screened for breast cancer every year for ten years is deemed to be 20 to 50% and the risk of having useless biopsy would then be of 5 to 20% .
To get the things worse the biopsy doesn’t guarantee a correct diagnosis because anatomopathologists disagree about diagnosis in as much as 25% of cases .
Most recent studies have also shown that there is no benefit of widespread screening for most cancers as for prostate and breast cancer. For breast cancer, there was the Canadian study from Miller in 2014, and the more recent study from Harding in the United States, the second published in the JAMA.
The Nordic Cochrane organization and Peter Götzsche have taken position about this issue.
Indicators like cancer survival contribute to misleading representations and misinformation of patients and doctors. For patients, a study has also shown that the decision of being screened is based on such widesperead misconceptions about the vertues of screening .
Despite these facts authorities continue promoting widespread screening and earlier cancer diagnosis  .
Who will benefit?
 Binder-Foucard F. Estimation nationale de l’incidence et de la mortalité par cancer en France entre 1980 et 2012, 2013, july.
Competing interests: No competing interests
Before discussing cancer survival, I wish to point out a flaw in the argument against cancer screening.
To use the figures quoted, screening for prostate cancer increased the age standardized incidence rate of prostate cancer from 24.8 per 100 000 in 1980 to 127.1 per 100 000 in 2005 (France). This was accompanied with “only a modest reduction in age standardized mortality rate (ASMR)”- from 16.3 to 13.3 (per 100 000). In addition, prostate cancer survival increased to 90% in 2002 (from 70% in 1990).1
The relationship between incidence, prevalence and duration of disease (in a stable population where incidence and duration are constant) is given as:
Prevalence = Incidence x Duration; or Incidence = Prevalence/ Duration.2
Prevalence includes both new (incident) and old cases. Assuming no change in duration of disease, it will remain constant in the presence of high incidence in one of two situations:
1. High cure rate
2. High death rate
In both situations, patients are ‘diseased’ for a short duration.
If the incidence increased five-fold, and there was only a “modest reduction” in death rate, the prevalence should have demonstrated a similar increase (assuming duration of disease did not change), unless there was a corresponding increase in cure rate. As the prevalence rate is not mentioned, I presume it remained largely unchanged.
We are told that cancer survival increased by 20% over a 12 year period. Clearly, this can only be due to better treatment outcomes- more persons with prostate cancer were receiving treatment early in the disease, hence had better chance of survival. The other possibility is that patients with advanced disease benefited from a vastly superior new treatment modality.
There is neither complete information nor consensus regarding prostate cancer screening, so the above argument suffers from conjecture.3
Cancer survival would refer to the number of patients who survived cancer.
The problem with this indicator is that it does not yield information regarding the duration of survival- like the Median Survival Time. Thus, two conditions with similar survival rates could have considerably different Median Survival Times.
Survival analysis uses the number of events (recurrence/ death) to estimate the Median Survival Time. The Median Survival Time is influenced more by the number of events than the survival time contributed by those who were censored- lost to follow up, etc.
If only a few cases of advanced cancer are detected, and the majority cases are censored (with long survival times), the Median Survival Time will decline because most events would occur to those diagnosed late.
Similarly, a good screening program will detect many cases in the precancerous stage. Typically, such patients will benefit from treatment administered during the lead time and many may not progress to overt cancer. These would be excluded from the calculation of survival rates because they would not satisfy the case definition- a person diagnosed with cancer. This, too, will inflate the survival rate.
Survival rates should not be interpreted in isolation, and even then, with great caution.
1. Michal-Teitelbaum C. Cancer survival is a poor indicator of quality of care. BMJ. 2015; 351:h4059.
2. Park K. Park's Textbook of Preventive and Social Medicine. 23rd ed. Jabalpur: M/s Banarsidas Bhanot Publishers; 2015.
3. Stark JR, Mucci L, Rothman KJ, Adami HO. Prostate cancer screening: the controversy continues. BMJ. 2009; 339:b3601.
Competing interests: No competing interests