5 ms·

Part of the Stanford article is highly misleading. They say: "That study, published in September in The New England Journal of Medicine, revealed that prostate

by v0x 10y ago

Part of the Stanford article is highly misleading. They say: "That study, published in September in The New England Journal of Medicine, revealed that prostate cancer patients randomized to either active monitoring, surgery or radiation therapy all had the same risk of death from the cancer after 10 years. Ninety-nine percent of men in the study survived regardless of initial treatment. These startling results suggest that active monitoring of prostate cancer patients may be as good as early radical treatment and may cause fewer side effects." In the actual NEJM study they reference, though, the conclusions are the exact opposite: "...death from prostate cancer occurred in 8 of the 545 men assigned to active monitoring, 5 of the 553 men assigned to surgery, and 4 of the 545 men assigned to radiotherapy" "For today, we can conclude on the basis of level 1 evidence that PSA monitoring, as compared with treatment of early prostate cancer, leads to increased metastasis. Therefore, if a man wishes to avoid metastatic prostate cancer and the side effects of its treatment, monitoring should be considered only if he has life-shortening coexisting disease such that his life expectancy is less than the 10-year median follow-up of the current study..." http://www.nejm.org/doi/full/10.1056/NEJMe1610395#t=article http://www.nejm.org/doi/full/10.1056/NEJMe1610395#t=article

pcrh 10y ago
From the article: "causality between an increase in metastatic disease and the use of active monitoring versus treatment was established." The notion that monitoring PSA levels can per se increase the risk of metastasis is ridiculous. It's far more likely that some other factor caused the increased incidence of metastases.
- v0x 10y ago
  They are just saying that metastasis occurs more frequently with monitoring as compared to active treatment, not that PSA testing itself causes metastasis.
  - pcrh 10y ago
    Ah yes, you're correct...
- wavegeek 10y ago
  The difference is not that monitoring causes disease, but you are not doing surgery or radiotherapy, and it is not doing that which results in higher risks, in some patients. Unfortunately they mixed up Gleason 6 (low risk) and 7+ (intermediate and high risk) patients in the statistics. In the supplement they give information which allows you to see that the risk of dying is far higher (~10X higher) for the 7+ patients. But they do not provide this breakup for the risk of progression and metastases. So we cannot know what the risk is for low risk patients. And it is low risk patients who are usually recommended for active surveillance. People have a hard time believing this, but this kind of statistical fail is common, even normal in medical research. The average doctor, including medical researchers, has zero formal training in statistics. So they basically use cookbooks without really knowing what they are doing. So you get researchers saying that because they has a statistically insignificant result, they showed there is no effect. We had a study which had a slightly below significant result for fish oil reducing post natal depression in women with a 30% reduction(P=.11m where < .05 is arbitrarily counted as 'significant'). The lead researcher then gets on the media and tells the country <we showed fish oil is useless>. The Bayesian revolution in statistics has largely bypassed medicine even though it is well accepted in many other fields.
nradov 10y ago
I don't get it. How can they draw such a conclusion based on a very small sample? The difference between 4 and 8 deaths could easily be due to random luck. I wouldn't make healthcare decisions based on that study until it's replicated with ~10× as many patients.
- tgb 10y ago
  I got a p value of 0.19 for getting at least as extreme results in this direction under the nul hypothesis that there was no difference by the treatments. So I agree this is pretty weak evidence.
wavegeek 10y ago
I have been through this study and the supplements carefully. The major problem is that they mixed patients with low risk (Gleason score of 6) and those with moderate risk (Gleason 7) and even some with high risk (8+). When you look at the supplement the death rate for the Gleason 7's on active surveillance was 10X higher than for Gleason 6. Why on earth they combined the 7s and 6s together in the statistics would be a complete mystery to me if I had not been reading medical papers for 40 years and come to expect this sort of statistical idiocy as par for the course. http://slatestarcodex.com/2013/12/17/statistical-literacy-among-doctors-now-lower-than-chance/ http://slatestarcodex.com/2013/12/17/statistical-literacy-am...

by v0x 10y ago

pcrh 10y ago

From the article: "causality between an increase in metastatic disease and the use of active monitoring versus treatment was established." The notion that monitoring PSA levels can per se increase the risk of metastasis is ridiculous. It's far more likely that some other factor caused the increased incidence of metastases.

v0x 10y ago
They are just saying that metastasis occurs more frequently with monitoring as compared to active treatment, not that PSA testing itself causes metastasis.
- pcrh 10y ago
  Ah yes, you're correct...
wavegeek 10y ago
The difference is not that monitoring causes disease, but you are not doing surgery or radiotherapy, and it is not doing that which results in higher risks, in some patients. Unfortunately they mixed up Gleason 6 (low risk) and 7+ (intermediate and high risk) patients in the statistics. In the supplement they give information which allows you to see that the risk of dying is far higher (~10X higher) for the 7+ patients. But they do not provide this breakup for the risk of progression and metastases. So we cannot know what the risk is for low risk patients. And it is low risk patients who are usually recommended for active surveillance. People have a hard time believing this, but this kind of statistical fail is common, even normal in medical research. The average doctor, including medical researchers, has zero formal training in statistics. So they basically use cookbooks without really knowing what they are doing. So you get researchers saying that because they has a statistically insignificant result, they showed there is no effect. We had a study which had a slightly below significant result for fish oil reducing post natal depression in women with a 30% reduction(P=.11m where < .05 is arbitrarily counted as 'significant'). The lead researcher then gets on the media and tells the country <we showed fish oil is useless>. The Bayesian revolution in statistics has largely bypassed medicine even though it is well accepted in many other fields.

nradov 10y ago

I don't get it. How can they draw such a conclusion based on a very small sample? The difference between 4 and 8 deaths could easily be due to random luck. I wouldn't make healthcare decisions based on that study until it's replicated with ~10× as many patients.

tgb 10y ago
I got a p value of 0.19 for getting at least as extreme results in this direction under the nul hypothesis that there was no difference by the treatments. So I agree this is pretty weak evidence.

wavegeek 10y ago

I have been through this study and the supplements carefully. The major problem is that they mixed patients with low risk (Gleason score of 6) and those with moderate risk (Gleason 7) and even some with high risk (8+). When you look at the supplement the death rate for the Gleason 7's on active surveillance was 10X higher than for Gleason 6. Why on earth they combined the 7s and 6s together in the statistics would be a complete mystery to me if I had not been reading medical papers for 40 years and come to expect this sort of statistical idiocy as par for the course. http://slatestarcodex.com/2013/12/17/statistical-literacy-among-doctors-now-lower-than-chance/ http://slatestarcodex.com/2013/12/17/statistical-literacy-am...