# Backing out into a p-value for clinical trial

#### executor320

##### New Member
I'm looking to analyze some clinical trial results but the company hasn't released a whole lot of information so far. All that I know are the hazard ratios and p values of the overall population and a subset of the population. What I'm trying to figure out is if it's possible to back into the p-value for the other subset of the population given such little information. To facilitate this (or confirm that it isn't possible, as I suspect), here are some sample values:

The treatment arm showed a 20% risk reduction in progression free survival (HR=0.80, p = 0.05) in the entire population (n = 500). A subset of these patients (n = 400), showed a 30% risk reduction in progression free survival (HR=0.70, p = 0.005).

I want to know as much as possible about the remaining 100 patients, but I'm shaky on the statistics. First, is it appropriate to use a weighted average to find the HR? If so, I'm assuming the calculation would just be: 0.80 = 0.70 (400/500) + x (100/500). Which would give an HR of 1.2 (meaning that the treatment arm was actually worse than the control). If this is right, is it possible to then determine the p value (given that I have none of the underlying data)?
Any help would be appreciated!

#### ondansetron

##### TS Contributor
I'm looking to analyze some clinical trial results but the company hasn't released a whole lot of information so far. All that I know are the hazard ratios and p values of the overall population and a subset of the population. What I'm trying to figure out is if it's possible to back into the p-value for the other subset of the population given such little information. To facilitate this (or confirm that it isn't possible, as I suspect), here are some sample values:

The treatment arm showed a 20% risk reduction in progression free survival (HR=0.80, p = 0.05) in the entire population (n = 500). A subset of these patients (n = 400), showed a 30% risk reduction in progression free survival (HR=0.70, p = 0.005).

I want to know as much as possible about the remaining 100 patients, but I'm shaky on the statistics. First, is it appropriate to use a weighted average to find the HR? If so, I'm assuming the calculation would just be: 0.80 = 0.70 (400/500) + x (100/500). Which would give an HR of 1.2 (meaning that the treatment arm was actually worse than the control). If this is right, is it possible to then determine the p value (given that I have none of the underlying data)?
Any help would be appreciated!
Need to clarify the terminology since this may frame the answer. You have the population (all possible participants) or a sample (some "subset" of a much larger possible total, generally)?

#### executor320

##### New Member
To clarify: the 500 patients are a sample of cancer patients and represent all of the participants enrolled in the trial. The 400 patients are a subset of the original sample and represent patients that have a particular biomarker. I want to know details on the subset of 100 patients who do not have this biomarker.