Please help with study design

[Sir Gillies]

Hello everyone,

This is my first post and I really hope someone can help me. I have decided to compare three surgical pathways by collecting data retrospectively from notes of patients who have already undergone these procedures. These are the three groups:
1) IMMEDIATE (200 patients) - two types of surgeries performed on the same day followed by radiation. This would be the "control" group.
2) DELAYED/IMMEDIATE (uncertain number) - brief gap between two types of surgeries, with radiotherapy after the second surgery
3) DELAYED (35 patients) - first type of surgery, then radiotherapy, then second surgery many months down the line.

The outcomes I am measuring are:
1) Recurrence of cancer
2) Complications
3) Risk factors
4) Aesthetics - this will be assessed by selecting 35 patients from each arm (selected according to similar characteristics to the third arm) and asking a panel of experts to answer a standardised questionnaire (including factors such as size, shape, etc) about each patient
5) Satisfaction - this will be assessed by sending out to the above 105 patients a standardised questionnaire.

Questions from a very inexperienced statistician : ( :

a. is it a case-control or a retrospective cohort study? Are parts 4 and 5 considered a prospective study? Can I mix it all up in one single study?
b. can you see any ways I can improve my methodology?
c. how can I try to reduce bias and confounding factors?
d. is it ok if I am the sole person collecting and analysing the data? Would it be better if the statistical analysis was performed by someone else?
e. unfortunately I only have 35 patients in the third group. Is it still ok to use such a large number in the first group?
f. what statistical analysis should I get ready (i.e. learn) to perform?
g. how can I improve the statistical power on this study? how do I calculate it?

I apologise for the many questions, but any input would be really really really appreciated : )
Thanks in advance for your help!

[victorxstc]

Hi and welcome to this community

Briefly put, in case-control, we split the sample according to the outcome (for example, the result of smoking [cancerous/non-cancerous patients]), and in retrospective cohort, we split the sample according to the exposure (for example smoking/not smoking]).

But your design is none of the above. It is a clinical trial, in which you are assessing the effect of an intervention. It is similar to a retrospective cohort, however, instead of exposure (like smoking, sunlight, etc.), there is an experiment. So your case seems quite a clinical trial.

[victorxstc]

can you see any ways I can improve my methodology?

If possible, increase your sample size; match your patients, collect more risk factors. I recommend gathering every possible data. if possible, you can also switch to a prospective design, since a real clinical trial is a prospective one.

how can I try to reduce bias and confounding factors?

There are different types of bias, each of which can be reduced by taking some actions. Blinding the observers in a prospective design is a solution. Reducing the number of categories and drawing more vivid cut-offs will reduce your bias at the cost of losing some data.

is it ok if I am the sole person collecting and analysing the data? Would it be better if the statistical analysis was performed by someone else?

As far as you are not inclined to manipulate the methods, it is quite OK. When someone is as caring as you, I don't think cheating will be the case. But does it introduce bias to your study? No, since selecting statistical approaches are based on our design and data, so a well-informed or knowledgeable researcher can do the stats himself/herself too.

But that was for "analyzing" the data. When it comes to collecting the data, it is a little bit controversial, and each design has its own advantages. The "only one observer" method removes interobserver bias when there is a need to judge something as the observation (not when the task is simply collecting the standardized questionnaires by various collectors), while two or more observers with a high or excellent interobserver agreement can collect more data and can reduce the subjective bias.

unfortunately I only have 35 patients in the third group. Is it still ok to use such a large number in the first group?

The asymmetry is OK. The problem might arise from the low power this small subsample can cause. If the variations are low, there is chance. Otherwise, a larger sample is needed. It is strongly recommended to calculate necessary sample size before your study (based on the current results, as a pilot study).

what statistical analysis should I get ready (i.e. learn) to perform?

For determining this one, we both need the type and nature of your variables, all of them, and your data distribution. It seems that multivariate statistics are appropriate, but more detail is needed for better suggestions.

how can I improve the statistical power on this study? how do I calculate it?

By increasing the sample size and decreasing the variations. Since the latter is not quite under your control, you have to do the former.

[Sir Gillies]

Thank you very much for yr thorough reply!

It might be a silly question, but I thought clinical trial were only prospective. Do you have any links where I can learn about retrospective clinical trials?

Re power calculation: I have results for the first group (ie the control) but not the third. Can I use results from another paper? To be honest I'm not even sure where to start : (

Also is there such a thing as blinding in retrospective studies?

Thanksa lot for your help.

[victorxstc]

No that's not silly at all, as we had rather long discussions over this issue before. Yes, although a real clinical trial is prospective, but when you assess the effect of some intervention, even retrospectively, you still can't categorize it in a cohort study. I don't have a link, but a search will show good links to you.

Can I use results from another paper? To be honest I'm not even sure where to start : (

Yes, that is another accepted and correct way of power calculation.

Also is there such a thing as blinding in retrospective studies?

No! But I think not all parts of your study are retrospective, as your judge panel is going to rate some aesthetic factors etc., so it can't be purely retrospective.