If any dan-level statistician here has the inclination, I’ll post a link to the paper here for your perusal...
Is there any reason not to post the link immediately? You are creating an additional barrier (pretty steep one) that lessens your chances of getting any cooperation.
Some patients crossed between the two groups, but this does not matter, as they were testing the effects of the initial assignment.
They report p = 0.52, but they also give a 95% confidence interval for the difference in risk, which just barely contains zero; which is a dead giveaway that p should be around 0.05, right? Anyway, doing a chi-squared test on the above numbers, I got p = 0.053.
The relevant bit is at the top of page 289 (page 6 of the PDF). Also relevant are the Results section of the abstract, and Figures 1 and 2. Essentially the entire problem is this statement:
At 2 years, intention-to-treat analyses showed that pain interfering with activities developed in similar proportions in both groups (5.1% for watchful waiting vs 2.2% for surgical repair; difference 2.86%; 95% confidence interval, −0.04% to 5.77%; P=.52)
You are correct, and the pdf that you linked contains a correction on its last page:
On page 285, in the “Results” section of the Abstract, the value reported as P=.52 for pain limiting activities should instead have been reported as P=.06; the corresponding value should also have been reported as P=.06 in the first paragraph on page 289.
It does not say anything about whether this affects their conclusions.
Some patients crossed between the two groups, but this does not matter, as they were testing the effects of the initial assignment.
It matters to your case. I refuse to believe that writing a patient’s name on this list rather than that list has a direct causal influence upon their state in 2 years. The influence can only proceed via their actual treatment.
Assignment ---> Actual treatment ---> Outcome
The decision facing you is whether to have surgery early or not. That is the thing whose effect on the outcome you want to know. To the extent that in the study this differs from the initial assignment, the study is diminished; therefore it should matter to the people conducting the study also.
I see from the paper that 23% of those assigned to Watchful Waiting nevertheless had surgery within 2 years, and 17% of those assigned to surgery did not have surgery in 2 years. (Some others died of unrelated causes or left the study early.)
I’ll leave it to a dan-grade statistician to judge how to obtain the best conclusion from these data.
Is there any reason not to post the link immediately? You are creating an additional barrier (pretty steep one) that lessens your chances of getting any cooperation.
Well, I was only going to post all the minutiae if there was any interest...
http://jama.ama-assn.org/cgi/reprint/295/3/285.pdf
The two groups are as follows:
Assigned to “Watchful Waiting”:
336 patients
17 had problems after 2 years
Assigned to surgery:
317 patients
7 had problems after 2 years
Some patients crossed between the two groups, but this does not matter, as they were testing the effects of the initial assignment.
They report p = 0.52, but they also give a 95% confidence interval for the difference in risk, which just barely contains zero; which is a dead giveaway that p should be around 0.05, right? Anyway, doing a chi-squared test on the above numbers, I got p = 0.053.
The relevant bit is at the top of page 289 (page 6 of the PDF). Also relevant are the Results section of the abstract, and Figures 1 and 2. Essentially the entire problem is this statement:
You are correct, and the pdf that you linked contains a correction on its last page:
It does not say anything about whether this affects their conclusions.
Argh how silly of me not to see that. I stop reading at the references! Honestly though, it’s annoying that the abstract remains wrong.
It matters to your case. I refuse to believe that writing a patient’s name on this list rather than that list has a direct causal influence upon their state in 2 years. The influence can only proceed via their actual treatment.
The decision facing you is whether to have surgery early or not. That is the thing whose effect on the outcome you want to know. To the extent that in the study this differs from the initial assignment, the study is diminished; therefore it should matter to the people conducting the study also.
I see from the paper that 23% of those assigned to Watchful Waiting nevertheless had surgery within 2 years, and 17% of those assigned to surgery did not have surgery in 2 years. (Some others died of unrelated causes or left the study early.)
I’ll leave it to a dan-grade statistician to judge how to obtain the best conclusion from these data.
But the question is whether it’s safe to advise people to wait, knowing that they can have surgery later if needed.
Anyway my main question was whether I’d done the stats right.