Complex problem

#1
Hi everyone,
I'm doing a project where a have a group of patients with a disease X and they will have different treatments overtime.
There are 4 drugs (A, B, C, D) that I am analysing. The outcome is measured in a scale from 1 to 5 (according to how controlled the symptoms are).
Each patient can have a combination of the drugs for a period (and an outcome for this period) and then change for another drug (with a different outcome).

Example:
Patient 1 was treated with drug A + B for 2 years with an outcome of 2. Then he was treated with drug D for 1 year with an outcome of 5.
Patient 2 was treated with drug B for 4 years with an outcome of 1.

...

How is the best way to analyse this study?

How can I consider time and outcome at the same analyses? Is there a way to take advantage of the fact that the same patient might undergo different treatments and have some kind of "paired" analyses?

Thank you very much.
 

Karabiner

TS Contributor
#2
I'm doing a project where a have a group of patients with a disease X and they will have different treatments overtime.
There are 4 drugs (A, B, C, D) that I am analysing. The outcome is measured in a scale from 1 to 5 (according to how controlled the symptoms are).
So you recruit patients, patients undergo treatment for their health (?) problems,
you compare no less than 4 different drugs, and finally, at the end of this expensive,
laborious, elaborate experiment, your main outcome measurement consists of not
more than a crappy 5-point-scale? Reliablilty 0.5 at best, not to speak of validity?

I don't mean to be rude, I just want to point to a huge imbalance.

How is the best way to analyse this study?
Why not ask your biostatistican? If you perform a medical study on living creatures,
it is certainly not supposed that the data analysis is outlined by members
of an anonymus web forum. At least it would be suprising if the ethics committee
aggreed with this.

With kind regards

Karabiner
 

hlsmith

Less is more. Stay pure. Stay poor.
#3
Well, @Karabiner was a little more blunt than I will be today. I also saw all of these flaws, but hoped/figured you aren't actually doing this study but have observational data about patients taking varied treatments. Is this correct and if so, can you verify that treatment were not randomized? And yes, please elaborate on this 5 point outcome scale!
 

Karabiner

TS Contributor
#4
hoped/figured you aren't actually doing this study but have observational data about patients taking varied treatments.
Ok, I did not take this possibility into account.

So, hopefully "I'm doing a project where a have a group of patients with a disease X and they will have different treatments overtime." does not mean what it sounds like.

But if it's a study not yet carried out, I'd urgently ask to think about proper outcome measurements.
 
#5
So you recruit patients, patients undergo treatment for their health (?) problems,
you compare no less than 4 different drugs, and finally, at the end of this expensive,
laborious, elaborate experiment, your main outcome measurement consists of not
more than a crappy 5-point-scale? Reliablilty 0.5 at best, not to speak of validity?

I don't mean to be rude, I just want to point to a huge imbalance.


Why not ask your biostatistican? If you perform a medical study on living creatures,
it is certainly not supposed that the data analysis is outlined by members
of an anonymus web forum. At least it would be suprising if the ethics committee
aggreed with this.

With kind regards

Karabiner
Dear Karabiner,

First of all, thank you for your time.

This was supposed to be a retrospective analysis of patients medical record. Those four drugs are the most common drugs used for a disease X. The data that we can obtain is whether the patient was symptomatic during the period he was using the drugs (eg. the patient had 1 mild event and 3 severe events during 4 months he was using the drug B). If he remained symptomatic or experience any side effect from the drug, usually the doctor would change the medication and he would have a new period with new drugs. The idea was to create a scale to determine the degree of symptoms (mild, moderate or severe/ daily, weekly, monthly). It doesn't need to be 1 to 5.

This is merely an educational question. There is no such study being done yet and if it were to happen, a proper biostatistican would analyze it. Im just asking to know if it is possible to analyze this data and which one is the best way.

Again, thank you for your answer and if you could give your input about this I would appreciate.
 
#6
Well, @Karabiner was a little more blunt than I will be today. I also saw all of these flaws, but hoped/figured you aren't actually doing this study but have observational data about patients taking varied treatments. Is this correct and if so, can you verify that treatment were not randomized? And yes, please elaborate on this 5 point outcome scale!
Dear hlsmith, you are right. As I explained above, this is an retrospective analysis.

Regarding the scale, this was the way I was thinking to compare the efficiency of the drug.

Patients can mild, modera or severe symptoms. The number of time a patient shows a symptoms (an event) can be daily, weekly, monthly or no event.

The idea is to combine those 2 information and create a scale. If you have any better idea how to measure the efficacy, I would love to hear.