Best Disclaimer Ever

Paul Alper sends this in, from the article, “Ovarian cancer screening and mortality in the UK Collaborative Trial of Ovarian Cancer Screening (UKCTOCS): a randomised controlled trial,” by Ian J Jacobs, Usha Menon, Andy Ryan, Aleksandra Gentry-Maharaj, Matthew Burnell, Jatinderpal K Kalsi, Nazar N Amso, Sophia Apostolidou, Elizabeth Benjamin, Derek Cruickshank, Danielle N Crump, Susan K Davies, Anne Dawnay, Stephen Dobbs, Gwendolen Fletcher, Jeremy Ford, Keith Godfrey, Richard Gunu, Mariam Habib, Rachel Hallett, Jonathan Herod, Howard Jenkins, Chloe Karpinskyj, Simon Leeson, Sara J Lewis, William R Liston, Alberto Lopes, Tim Mould, John Murdoch, David Oram, Dustin J Rabideau, Karina Reynolds, Ian Scott, Mourad W Seif, Aarti Sharma, Naveena Singh, Julie Taylor, Fiona Warburton, Martin Widschwendter, Karin Williamson, Robert Woolas, Lesley Fallowfield, Alistair J McGuire, Stuart Campbell, Mahesh Parmar, and Steven J Skates:

Declaration of interests IJJ reports personal fees from and stock ownership in Abcodia as the non-executive director and consultant. He reports personal fees from Women’s Health Specialists as the director. He has a patent for the Risk of Ovarian Cancer algorithm and an institutional licence to Abcodia with royalty agreement. He is a trustee (2012–14) and Emeritus Trustee (2015 to present) for The Eve Appeal. He has received grants from the Medical Research Council (MRC), Cancer Research UK, the National Institute for Health Research, and The Eve Appeal. UM has stock ownership in and research funding from Abcodia. She has received grants from the MRC, Cancer Research UK, the National Institute for Health Research, and The Eve Appeal. NNA is the founder of, owns stock in, and is a board member of MedaPhor, a spin-off company at Cardiff University. He has a patent for the ultrasound simulation training system MedaPhor. SA is funded by a research grant from Abcodia. AD reports personal fees from Abcodia. AL reports personal fees from Roche as a panel member and advisory board member and fees from Sanofi Pasteur Merck Sharp & Dohme (Gardasil) as an advisory board member. JM is involved in a private ovarian cancer screening programme after closure of this trial. MWS reports personal fees from Abcodia as a consultant. LF reports funding from the MRC for the UK Collaborative Trial of Ovarian Cancer Screening psychosocial study. She reports personal fees from GlaxoSmithKline, Amgen, AstraZeneca, Roche, Pfzier, Teva, Bristol-Myers Squibb, and Sanofi and grants from Boehringer Ingelheim and Roche. SJS reports personal fees from the LUNGevity Foundation and SISCAPA Assay Technologies as a member of their Scientifi c Advisory Boards. He reports personal fees from Abcodia as a consultant and AstraZeneca as a speaker honorarium. He has a patent for the Risk of Ovarian Cancer algorithm and an institutional license to Abcodia. All other authors declare no competing interests.

All right, then.

Oh, in case you were wondering, here’s the last part of the paper’s summary:

Although the mortality reduction was not significant in the primary analysis, we noted a significant mortality reduction with MMS when prevalent cases were excluded. We noted encouraging evidence of a mortality reduction in years 7–14, but further follow-up is needed before firm conclusions can be reached on the efficacy and cost-effectiveness of ovarian cancer screening.

N = 202638 and the effect wasn’t statistically significant. No problem, says the non-executive director and consultant of Abcodia, the director of Women’s Health Specialists, the trustee of the Eve Appeal, the owner of stock in Abcodia, the owner of stock in MedaPhor, the patenter of MedaPhor, the receiver of personal fees from Abcodia, the panel member of Roche and advisory board member of Sanofi Pasteur Merck Sharp & Dohme, the receiver of personal fees from GlaxoSmithKline, Amgen, AstraZeneca, Roche, Pfzier, Teva, Bristol-Myers Squibb, and Sanofi and grants from Boehringer Ingelheim and Roche, the receiver of personal fees from SISCAPA Assay Technologies, Abcodia, and AstraZeneca. No problem, says the patent holder for the Risk of Ovarian Cancer algorithm. No problem at all. Encouraging evidence, they say.

P.S. I get funding from Novartis and have also been paid by Merck, Procter & Gamble, Google, and lots of other companies that I can’t remember right now.

58 thoughts on “Best Disclaimer Ever

  1. Andrew – did you notice this bit?
    “We did all statistical analyses using Stata (version 14), R (version 3.2.1; packages fl exsurv and PwrGSD), and Stan (version 2.8.0).”

  2. I always find these disclaimers odd since the standard situation is that the researcher’s career depends upon publishing “positive” results. I don’t understand why that incentive does not have to be mentioned.

    • Agreed, for some reason maintaining multimillion dollar federal funding projects, and all that represents in terms of summer dollars and promotion etc., presents no disclosure or conflict of interest risks, but accepting a 5K consulting fee from a private firm might activate a requirement that a researcher’s interactions with grad students are monitored to show they aren’t coercing them. The incentives could be 40X when the funds are federal, but there are apparently no risks that researchers will coerce grad students to pursue their research angle, or spin their results to improve chances of renewal, as long as the funds come from NIH and not Acme Biotech. The assessment of conflicts of interest seems a bit incomplete to say the least.

      • Well, when the funds are Federal aren’t they automatically listed as “A part of this work was funded by a US-DoE Grant number XXXX for AAG” etc.

        The fedaral funding agencies don’t like it if you use their $$ and not list them on papers you publish.

  3. Author Foo declares that he spent 22 years in school and has $400,000 in non-bankruptcy-eligible student loans whose payment depends on him keeping his salary as an assistant professor of _____INSERT___FIELD___HERE_____ and that his two children are very cute, but need over $30,000 a year in day-care costs, and the house he bought with an initial incentives package from his university has a mortgage totalling $3000/mo, and with tenure coming, he really needs to publish at least 3 more papers in Nature, Science, PNAS, or Psychological Science in the next 2 years, as well as to get grants totaling at least $100,000 per year in order to pay a required portion of his own salary, as well as the salary of his postdoc, and the stipends for his graduate students, otherwise he’ll have to take on an additional teaching load, and this will lead to his wife filing for divorce. But otherwise has no competing interests.

  4. You just can’t spend that kind of money (we randomly allocated 202 638 women) and not show something!

    More seriously, the the health care cost that have not be spent (already) doing the apparently useless screening are likely much greater.

    Hopefully there has been some improvement in methodology and a third party can decided if screening can be dismissed as a bad idea rather than the trial just being dismissed as flawed/of limited value – http://statmodeling.stat.columbia.edu/2010/04/15/when_engineers/

    • I have a colleague whose father was an engineer in Poland. The custom there at that time was that after a bridge was built, but before it was open to use, the engineer and construction manager stood under the bridge while it was fully loaded.

  5. I find all these declarations of interest/funding completely idiotic, though I admit I’m in the minority.

    Assuming away fraud, we already know that the Forking Paths Guide to Inerrancy will yield significant results no matter what the source of funding, or insignificant ones if that’s what you’re trying to show (see Fisher on smoking and lung cancer.) The crudeness of this sort of declaration shouldn’t be helpful to the reader; if it is, the reader ought to get out in the world a little bit before reading any more scientific papers. And statements like these don’t help diagnose actual fraud, either… Only an audit of the study could possibly do that.

    Statements like these are a bureaucrat’s notion of action: meaningless hoops to jump through to “address” a problem. Hello, problem.

      • I’d say overall death rate maybe isnt the best metric to use to quantify the improvements in cancer care, instead perhaps compare the 5-year relative survival stats over the past 40 years- for many types of cancer the prognosis is far better for patients today. It was for sensible reasons that the Screening study used time-to-death rather than than overall death-rate as its endpoint. If improvements in treatment or procedure means a patient now dies of cancer after 12 years rather than 2 years, the death rate would be unchanged- so a lack of movement in overall death rate doesn’t mean there haven’t been real improvements to the lives and lifespans of patients.

        That said, the graph you link on your site still shows a 25% reduction in cancer deaths in 2012 from its peak 25 years before, a gradual but real difference (obviously its a pity its not as dramatic as the drop in Aids deaths). Without looking into the numbers I would guess that the main part of that reduction is actually due to screening! I’m particularly thinking of the effect of Prostate cancer screening, which now allows a disease with high incidence to be caught early enough for a 100% cure rate.

        • 5 year survival is really NOT the stat to use because increased screening simply moves the diagnosis earlier. if you start treating earlier and then you die at the same point in time that you would have if you didn’t do any treatment, it looks great, but it’s not an improvement at all, in fact you are sicker longer from the treatments!

          I’d guess that almost all of the decline in overall cancer rates is down to reduced smoking and a few other lifestyle factors, not primarily any treatment improvements. It’s hard to even know if something like the prostate cancer decline is from improved treatment or from secondary effects of reduced tobacco use without an analysis specifically for smoking.

          You can see the breakdown by cancer here:

          http://seer.cancer.gov/statfacts/

          stomach has a pretty good decline, lung goes up and then down, but oral cancers and etc go down in both deaths and incidence. To get the long term trends you have to click the “changes over time” section of each cancer’s summary otherwise you just see since the mid 90’s

          The best metric would be QALY loss per year actually. Especially if we could quantify how bad the effects of cancer treatments are. I think maybe reduced dosage etc has made people on chemo less sick, but earlier diagnosis may have increased the number of days of total treatment. Overall screening may be increasing QALY loss, or decreasing, it’s hard to say.

          Prostate cancer death rate/100k went from 39 at peak around 1993 to 20 in 2012. So, that’s good, but none of it is really what you’d call a breakthrough like in AIDS treatment or the development of a vaccine for Polio or something like that

          I’m not saying we’ve made *no* progress, but we’ve made no impressive progress.

        • Another thing in favor of QALY loss per year per capita (besides the fact that it’s nondimensional) is that eventually people die of something. If you live long enough you’re probably going to get some cancer. People who die at age 101 from “breast cancer” (really, at that point probably just a proxy for “multiple system failure but hey, one of those systems was breast cancer”) have a very different QALY loss count than someone who dies at say age 30 from aggressive breast cancer due to inherited genetic problems.

          An improvement in treatment might eliminate some early deaths, causing a dip in death rate, but eventually if you live long enough you’re likely to get cancer, and then death rates go up again because you’re not measuring the right thing (so yes, I agree raw death rates aren’t the right metric but they’re a lot better than 5 yr survival rate).

          So, I think a good dimensionless measure of the effectiveness of cancer treatments would be Quality Adjusted Life Years Lost Relative to a nominal 110 year life span per year per capita relative to the average QALY loss per year per capita experienced between 1950 and 1970.

          This would be a dimensionless ratio, and O(1) and is sensitive to life extension and quality of life improvement including improvements from reduced sickness due to less toxic treatments etc.

          Is anyone calculating that?

        • >”An improvement in treatment might eliminate some early deaths, causing a dip in death rate, but eventually if you live long enough you’re likely to get cancer, and then death rates go up again”

          At the top of those at your SEER link it looks like they took #cases/population at different age groups, did an age-adjustment[1] to get a standardized (to some “standard” population distribution) rate for each age group, then summed these standardized rates. This allows a clean comparison from year to year as the population changes, but the plotted rates don’t have a physical meaning. For simplicity, lets assume they didn’t do this standardization, ie instead we were looking at crude rates.

          According to the current theory, cancer is due to accumulation of genetic errors beyond some threshold. So maybe a cell gets the first mutation when you are 3, the second when you are 20, etc., then when n mutations have accumulated in the cell, it turns cancerous. IE, cancer is a word for a cell (or cell-line) having accumulated the n required mutations. So if the probability of getting one mutation in any given year is p, the probability of it happening by age a would follow the geometric distribution: 1-(1-p)^a. Assuming the mutations are independent and happen with equal probability*, getting n mutations by age a would be [1-(1-p)^a]^n.

          So the cumulative probability a cell turns cancerous, CMF=P(Cancer | age $\leq$ X yrs) should increase to 1, so I would agree with “if you live long enough you’re likely to get cancer at some point”. However, the slope of that curve, PMF=P(Cancer | age = X yrs)=diff(p0), should peak somewhere. This is what we see if you scroll down at the breast cancer link[2].

          To connect to the mortality rate data, consider that a person has B breast cells simultaneously undergoing this process and may develop multiple independent cancerous cells. Then the expected number of independent tumors (per person) forming at each age would be B*PMF. Also, not all tumors will be potentially deadly (or even detectable), only some proportion d. Then the expected number of deadly tumors (per person) forming at each age would be d*B*PMF.

          All that is just to get to the point that (according to this model) the mortality rate plots are of sum(d*B*PMF), where PMF is calculated for each age group each year. An improvement in treatment would lower d, which must lower the total mortality rate. How would you modify this model in order to see a dip but then increase in mortality rates? Or are you thinking a totally different way about how this data was generated?

          [1] http://seer.cancer.gov/seerstat/tutorials/aarates/definition.html
          [2] http://seer.cancer.gov/statfacts/html/breast.html

          *I think this basic idea would still work even if we relax those assumptions and say p refers to the average probability of getting any one of these mutations in a given year.

        • Errata: “diff(p0)” should be “diff(CMF)” and my attempt to use mathjax didn’t work out (is there a way to do that on this blog?), “$\leq$” should be “less than or equal to”.

        • Presumably a fantastic treatment would extend the time to death to say 110 years of age independent of the fact that you have cancer, but eventually, you’d die, and then be marked down as “died of cancer”.

          If we apply this fantastic treatment to everyone starting tomorrow… you’d have zero deaths due to “cancer” next year, but after a while some of the people with cancer would hit 110 years of age, and then die of the cancer, so the rate would increase again through time.

        • on the other hand, life years lost relative to a nominal 110 year lifespan would under this “fantastic” treatment drop to zero and stay there. Hence the importance of life years lost as a metric in this case.

        • >”The basic idea is that people are more susceptible as they get quite elderly, so treatments are likely to stop being effective.”

          I’m still having trouble picturing exactly what you think is going on here. As I understand it, the current standard theory of cancer predicts that essentially every dividing cell line in your body will eventually acquire a set of mutations making it become cancerous. The DNA can only be copied so many times before this is bound to happen.

          So the effect depends on the nature of the treatment. A treatment that 1) arrests the growth of cancer cells, or 2) kills them, would be much different from a treatment that 3) repairs the mutations, or 4) replaces the cells with “fresh” (lacking any of the required mutations) functional cells.

          In the former two cases you will eventually run out of functional cells and die from organ failure of one form or another. I’d guess this “treatment” occurs naturally due to immunosurveillance, etc and is a major contributor to the aging process.

          In the later cases, the process of generating the cancer cells would begin anew, shifting the same incidence and mortality patterns to older ages. This would be more similar to taking preventative measures that eg reduce the mutation rate. Since your population of cells is being maintained in a “fresh” state, I don’t see why the treatment would need to stop working after a certain age, just keep replacing/repairing the cells.

          It sounds like you are considering only scenario 1, where the cells are already in a non-functional cancerous state, yet lying in wait to resume the process of forming problematic tumors?

        • Anoneuoid:

          I just think that after some point you won’t be able to treat people effectively, I don’t have a specific model for why, but I can offer several thoughts:

          Immune system, heart, liver, kidneys, pancreas etc are all declining in function between say 90 and 110 pretty dramatically. Any bad effects from giving a medication which can be shrugged off by a 30 year old could be catastrophic to someone of seriously advanced age. So, at some point, it’s likely that there’s a tradeoff between risk of death from the treatment and risk of death from the cancer, which at some point becomes in favor of longer and better life without treatment, even if only by a few weeks or months. I just estimate that such a thing is almost certain to occur before age 110 years based on the behavior of the tail of life tables over the last 50 years. They’ve become steeper, but aren’t being extended out past 100 years.

          It really has nothing to do with the hypothesis on cancer generation and elimination, and more on the decline of overall health and robustness leading to failure to be able to effectively treat.

          If you imagine some magic treatment, like nano-bots running around continuously inside your body zapping cells magically with virtually no side effects… well maybe you’re talking a different kind of magic than what I’m thinking, but I think we’re a long way off from that. I’m more thinking about say targeted treatment with a combination of super-quality radiation, immunotherapy (inducing the immune system to attack the cancer selectively), targeted drug delivery, careful dosage control, less toxic drugs, tayloring the drugs and immune system response to the expression profile of the specific cancer, etc.

        • Thanks for explaining further, I think I’ve identified our differences:

          >”So, at some point, it’s likely that there’s a tradeoff between risk of death from the treatment and risk of death from the cancer, which at some point becomes in favor of longer and better life without treatment”

          Perhaps your body is already doing this near optimally.

          >”It really has nothing to do with the hypothesis on cancer generation and elimination, and more on the decline of overall health and robustness leading to failure to be able to effectively treat.”

          From my understanding of the current “hypothesis on cancer generation and elimination”, defending against it should be a primary driver of the “decline of overall health and robustness” we see with old age.

          >”I’m more thinking about say targeted treatment with a combination of super-quality radiation, immunotherapy (inducing the immune system to attack the cancer selectively), targeted drug delivery, careful dosage control, less toxic drugs, tayloring the drugs and immune system response to the expression profile of the specific cancer, etc.”

          Perhaps the law of diminishing returns is coming into play, ie your body is already removing cancerous cells so well that we can only expect minor improvements due to these approaches. There isn’t much room left for removing cancer that doesn’t lead to organ failure due to too many cells removed/arrested.

        • >”Presumably a fantastic treatment would extend the time to death to say 110 years of age independent of the fact that you have cancer, but eventually, you’d die, and then be marked down as “died of cancer”.

          If we apply this fantastic treatment to everyone starting tomorrow… you’d have zero deaths due to “cancer” next year, but after a while some of the people with cancer would hit 110 years of age, and then die of the cancer, so the rate would increase again through time.”

          It sounds like you are considering a much more specific scenario than I thought. Can you give an example of why such a treatment would stop working once people reach 110 years old? Why not just apply the same treatment to the new cancer?

        • I just assume that it’s not possible to extend people’s lives indefinitely. My very approximate model of “the best we can do” is something like p(live to more than age x) = 1 for x 110 yrs. If you look at life tables through time the life expectancy has been increasing by making the hazard curve closer and closer to something like that function (except it’s really more like age 100 rather than 110).

          The basic idea is that people are more susceptible as they get quite elderly, so treatments are likely to stop being effective.

        • Mikey:

          > Prostate cancer screening, which now allows a disease with high incidence to be caught early enough for a 100% cure rate.
          Sounds like a bus add ;-)

          Screening is very complicated stuff, for instance – “Whatever full mortality reductions emerge, those who might wish to ‘purchase’ them need to know how much they cost. Some may well consider that even if screening could achieve a sustained reduction of 67% (or even 97%), the very low prostate mortality rates in the control group means that the small absolute reductions will be achieved at an unacceptable cost” – http://www.medicine.mcgill.ca/epidemiology/hanley/Reprints/Hanley-JMS-2010.PDF

        • @Daniel

          So, I’m confused: Are you saying screening is ineffective? Or treatments are ineffective?

          Aren’t they different questions? I’d be more likely convinced of the first that the second.

        • I’m saying that “the treatment industry” is fairly ineffective (not completely). Treatment industry = diagnosis (including screening), and application of radiation, chemo, surgery, etc. The overall effect has been… suspicious and plausibly near zero when health behavior and environmental factors are taken into account.

          Anyone who lived in the LA area in the period between 1970 and 1995 or so will tell you that the environmental health impact of improved auto technology has been huge. and LA isn’t the only area like that. Driving through Benecia CA (where there are several refineries) in the 1980’s uesd to be a “try to hold your breath for 15 minutes) type event. Now you don’t even notice the refinery.

          what if cleaner air, cleaner water, lower smoking rates, and less red meat consumption is a large fraction of the cause of all the improvements we see in cancer?

          There are a few cancers where we see an obvious change-point (like Non-hodgkins lymphoma: http://seer.cancer.gov/statfacts/html/nhl.html ) but the change point is nothing like the AIDS one, it’s still a factor of 4 or 5 lower rate.

          My main point here is that if we measure the cancer industry success by what the cancer industry marketing department uses, things look rosy. If we measure it by some kind of external “what we really want from you guys” metric, it looks abysmal

        • Well yes, screening can do more harm than good, if you do it like crazy, as it has been done in the US. But the conclusions presented in that summary paper by USPSTF are wrong, because they are based on a trial (PLCO), where more than 50% of the people in the control group were screened, plus low compliance with the screening protocol. (No wonder the effect is zero…).

  6. From the NYT, http://www.nytimes.com/2015/12/18/health/early-detection-of-ovarian-cancer-may-become-possible.html?hpw&rref=health&action=click&pgtype=Homepage&module=well-region&region=bottom-well&WT.nav=bottom-well&_r=0

    “Dr. Menon and several other authors have financial interests in a British company, Abcodia, that will be marketing the new version of the CA125 test.”

    As far as I can tell, Abcodia is keeping the CA125 test algorithm secret.

  7. I am wondering what a good study design would be here. If I understand correctly, “prevalent case” means a woman who already had ovarian cancer at the beginning of the study. So if the intent is (as stated) to study “the effect of early detection by screening on ovarian cancer mortality,” then it seems that women who already have ovarian cancer at the time of first screening should indeed be excluded from the study. That would, I presume, require first screening all women recruited, then randomizing only those who had no evidence of ovarian cancer to screening or not screening. But the Methods section says that women who were discovered after randomization to have ovarian cancer were excluded from the analysis.
    Sounds like maybe the study design was not well thought out. Or possibly there was some ethical reason for not randomizing before initial screening for all participants?

    • ” then it seems that women who already have ovarian cancer at the time of first screening should indeed be excluded from the study. That would, I presume, require first screening all women recruited, then randomizing only those who had no evidence of ovarian cancer to screening or not screening.”

      And that, according to the full article, is what they did, or, more precisely, planned to do. The exclusions of prevalent cases, if you read all the details in the article, amount to 4 women who actually had a diagnosis of ovarian cancer prior to randomization, but, for whatever reason, that information was not known to the investigators who, therefore, went ahead and randomized them anyway. When the error came to light, they were excluded. In a study this size, it is, frankly, surprising there weren’t more errors of this nature. Even with good systems in place and everybody doing his or her best, there are just things that somehow get “lost in the shuffle” and then turn up later. I could imagine a dozen reasons why these four women slipped through the cracks. This sort of thing occurs all the time in clinical trials. And, evidently, in a trial of this size, the disposition of these four women could not meaningfully alter the outcome or conclusions anyway.

  8. I once worked on a similar study – an evaluation of effectiveness of screening which clearly didn’t work. Similarly to this story, the published paper ignored the obvious fact that it didn’t work and highlighted a sub-group where something was significant, concluding there were promising results.

    The interesting part was that none of the authors had a direct financial interest like they do in this example. Instead, all the authors worked in this particular area of medicine (they did not carry out screening in their professional capacity). They believed that their disease was important, and they should do everything they could to get money from the government to support it. This was enough, direct financial interests were not necessary.

    I think this is probably quite common in medical research when research teams are made up of specialists in one particular medical area, or researchers who only focus on this area.

      • Generally screening costs in the tens/hundreds of millions of dollars. For that sort of money, you want to see a very clear effect of health improvement. Even if ‘the screening was effective on women aged 25-35 outside of major cities’ wasn’t statistical noise, it ignores the overall picture – you spent tens/hundreds of millions of dollars and there was no significant effect on your overall target population.

        • I think Pre-registration is the obvious solution to the problem you describe of “ignored the obvious fact that it didn’t work and highlighted a sub-group where something was significant,”

          That, and ruthless referring & editorial policies that prevent post-hoc data dredging to be couched in language that oversells it as something stronger.

    • @sr

      I think people focus excessively on the financial interest angle.

      Much bigger is the problem of bad science being done for careers, fame, padding Resumes etc. The things some Prof.’s will do to scoop a competing lab must be seen to be believed.

  9. I want to criticize some of the points made here by Daniel.
    “Prostate cancer death rate/100k went from 39 at peak around 1993 to 20 in 2012. So, that’s good, but none of it is really what you’d call a breakthrough like in AIDS… ” A 50% reduction is not good? What you can ask is if all this screening is cost-effective and there you have a point, we (US) are screening way too much, at least in prostate cancer. Other countries took a much more cautious approach. (vested interests from pharmaceutical industry are indeed quite strong in the US).

    “Of course they are. Just as idiotic as running a 200k person study on cancer screening given that 40 years of improvements in screening has done NOTHING to the death rates of cancer overall (or in most sub-cancers).” It is not idiotic at all. The problem is that there is not enough power in most studies to detect this effect, we would need to include millions of patients. Remember prostate cancer will kill only about 2-4% of men, and studies showed a maximum of about 20-30% prostate cancer mortality reduction due to screening. I think i remember seeing an editorial somewhere saying we should start making trials with millions of patients but good luck finding enough funding for that.

    • me: “so that’s good, but …”
      you: “a 50% reduction is not good?”

      I just said it’s good. But it’s not like finding a combination of drug and radiation therapy that is 95% effective at eradicating this cancer permanently, in which case you’d have a reduction in deaths of 95% in just a year or two.

      or as Mikey above says “a disease with high incidence to be caught early enough for a 100% cure rate” which is the marketing version of what’s going on.

      But the marketing version is NOTHING LIKE true, it CAN’T BE given the deaths data. I haven’t done a massive study of this stuff (but neither have I ever been pointed to one by anyone else !), but if you just estimated the effect of smoking on risk for death due to prostate cancer:

      http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2836346/

      and then apply changes to smoking behavior:

      http://www.cdc.gov/tobacco/data_statistics/tables/trends/cig_smoking/

      do you get a 50% reduction in prostate cancer death rate over 20 years? I don’t know, but the point is that you know damn well what happened in the AIDS data, but with the cancer data, it’s not at all obvious that treatment is doing *anything* it would take a serious study which explicitly includes the possibility that the changes in cancer *death rates* is down to maybe healthier behavior overall in the population over the last 50 years to determine if treatment is really helping anyone in any way.

      Things I am pretty sure about:

      1) Reduced smoking is reducing cancer risk across many cancers.
      2) The population overall is getting older, and older people get cancer.
      3) General health consciousness is substantially increased since the 1970’s vs the period 1940-1970
      4) Air pollution, diet, and chemicals in the environment (DDT, PCB, etc) have changed in time, generally gotten better since 1970.

      All put together, it is plausible to me that those things might be a large portion of the causal factors for the changes in cancer death rates since 1970. It would take a serious study considering those types of things to carefully estimate whether there’s ANYTHING left for the medical *treatment* industry to take credit for (as opposed to preventative medicine, smoking cessation, automobile emissions engineering, etc).

      When the treatment for AIDS came out, it cut the death rate in half in about 2 years. Is there even one cancer where we have that? In 40 years of cancer research across all the possible cancers, it would be good if we’d come up with some combination treatment that really nailed one of those cancers to the floor. We haven’t.

      • But isn’t that a totally artificial benchmark: ” 95% effective at eradicating this cancer permanently, ”

        Why should that be the yardstick of anything? A 50% reduction sounds decent to me.

        I’d accept your hypothesis that a general reduction in smoking was responsible for some part of this, but till someone shows me the numbers as to how much, I’ll take the 50% reduction at face value. It is always possible to hypothesize many confounders: maybe men became less promiscuous or started using more statins? Without numbers it is all speculation.

        • I’m just trying to say that not only is total reduction important, but also RATE of reduction. The net relative rate here in prostate cancer is basically .02/yr (30-20)/30/(20yrs)

          The rate with AIDS was (50-20)/50/2 yrs = .3 or more than an order of magnitude larger. It’s actually better than that when you consider the that the number of cases grows linearly and the death rate states virtually constant.

          In 40 years of cancer research I’d hope that we’d have at least one kind of cancer where a really effective treatment was brought to market and gave us a reduction rate something like AIDS.

          Even with something like lymphoma where they have this supposedly “great” treatment based on immune system activation, the death rate hasn’t fallen that dramatically, and it basically made up for the increase that happened earlier in the period.

          What HAS happened is an enormous amount of patting on the back about 5 year survival rates and etc, and that almost entirely seems down to detecting things earlier, and then “curing” them even though they mostly weren’t a problem in the first place. See the prostate cancer paper linked above.

      • Daniel:
        1) In the case of prostate cancer, if the disease is caught early enough then the chance that you would die from is indeed almost zero. But you have to understand that in general cancer is quite heterogeneous, there are slow growing and fast growing cancers, and from what i understand, we dont know yet how to distinguish between them. This i think is very different from the AIDS situation. So what you want is the development of a biomarker that can accurately predict which cancers will become deadly. In the case of prostate cancer we are far away from that. For now it seems that using a panel combination of multiple urine and genetic markers is the way to go.

        2)Did cancers deaths decrease due to environmental factors or do to treatment? Woow this is would be super cool to research! About prostate cancer, I am not sure if smoking is a big deal , at least i hear little talk about that in the circle that i know. It should also be emphasized that the quality of treatment improved a lot, and that the frequency of side effects is much lower than 10-15 years ago and survival probably improved as well.

        • I think we agree then that the metric of QALY loss per year is likely to be a better one. It would help us see things like “cancers tend to be killing people later in life than before” and “cancer treatments are making people less miserable than before” and etc.

          I really think the hypothesis of environmental and health behavior needs to be looked at carefully. I’ve actually got a friend who is headed towards a public health masters and wants to get some data analysis experience, so I may spend some time on this problem with him. Look for posts on my blog with alternative metrics, better graphs etc.

        • 1) We are already doing that in fact. Not in the US, for political reasons, but in European countries, cost per QALY gained is certainly being taken into account when looking at screening. See this nice example: https://www.ncbi.nlm.nih.gov/pubmed/25505238 .

          2) Well about this environmental and health behaviour: It is really hard to show something like amount of environmental condition X or behaviour Y, has reduced mortality from cancer Z. Why? Well I think you would need a formidable amount of follow-up (maybe you should look at scandinavian data?) and a super large database in order to make such conclusions. And is it not confounded by improvements in treatment (and how do you measure them???? ) or earlier detection? I am not sure how you can disentangle so many effects, but I am open for suggestions :)

        • Since 1900 things like smoking, red meat consumption, and air pollution went up and then down, at various different times whereas presumably treatment since 1970 or so hopefully never got *worse* so should have at least made death rates stay flat, or decline. There is some chance to identify the various effects potentially.

        • This sounds like one of those situations where you choose the conclusion you want and one can cherry pick the theory and data to support it. Just too much leeway.

        • Rahul: every Bayesian analysis is contingent on the model being correct, or at least “good”. It’s perfectly possible to fit a variety of models and show how well or poorly the causality is identified, and also to compare predictive performance of the model say in the final 5 years of the data… You can put priors on 4 or 5 different basic “stories” (ie. screening and treatment really help a lot, treatment helps but screening doesn’t, neither one are very meaningful relative to environment and behavior, all three of screening, treatment, and environment/behavior are important… etc)

          The Bayesian machinery will tell you what the data says about the various hypotheses, downweighting those that don’t fit well.

      • I think some people already looked at what explained this reduction, and i think some part of it was due to screening and other due to treatment but most of it was unexplained. So the environmental and health behaviour hypothesis could be plausible.

  10. It seems at a first look that attendance is high and the degree of screening in the control group is fairly small, thus no source of bias from there. Strangely the incidence in both screening groups is almost the same as the incidence in the control group.

    Given that both tests seem to have a fairly high sensitivity, then this cancer must have a relatively small lead time so i wonder if screening would reduce mortality by much.

    On the other hand, if there is a significant number of prevalent cases, and IF there was NO screening going on before the trial then maybe the authors have a point in removing those cases, but reading all these conflicts of interest i am not sure if I would trust this analysis…

Leave a Reply

Your email address will not be published. Required fields are marked *