Hi, Gaurav,

I think you can as far as you find possible neurofunctional differences that might be responsible for differences between responders and non-responders.

check this paper, for example:

ihttp://www.nature.com/npp/journal/vaop/ncurrent/full/npp2015341a.html

or this much older paper:

http://www.ncbi.nlm.nih.gov/pubmed/6146084

Hi 

In addition to my agreement to Silvana, also, the following paper indicate that you can do.

 Prednisolone response in patients with chronic obstructive pulmonary disease: results from the ISOLDE study.

Burge PS1, Calverley PM, Jones PW, Spencer S, Anderson JA.

Author information

Abstract

BACKGROUND:

A trial of corticosteroids has been recommended for all patients with chronic obstructive pulmonary disease (COPD), with the subsequent "response" determining the treatment selected. This approach assumes that patients can be reliably divided into responder and non-responder groups. We have assessed whether such a separation is statistically valid, which factors influence the change in forced expiratory volume in 1 second (FEV(1)) after prednisolone, and whether the prednisolone response predicts 3 year changes in FEV(1), health status, or number of exacerbations during placebo or fluticasone propionate treatment.

METHODS:

Oral prednisolone 0.6 mg/kg was given for 14 days to 524 patients with COPD before randomised treatment for 3 years with fluticasone propionate or placebo. Factors relating to change in FEV(1) after prednisolone were investigated using multiple regression. The response to prednisolone was entered into separate mixed effects models of decline in FEV(1) and health status during the 3 years of the study.

RESULTS:

The post-bronchodilator FEV(1) increased by a mean 60 ml (CI 46 to 74) after prednisolone with a wide unimodal distribution. Current smoking was the factor most strongly associated with the change in FEV(1) after prednisolone, with an increase of 35 ml in current smokers and 74 ml in confirmed ex-smokers (p

It depends on your hypothesis. But: I've done this twice in my publications. Once for animal behavioral data, and once for analyzing the proportion of different types of responsive neurons to a drug stimulus (i.e., non-responsive neurons, neurons that showed increased activity in response to stimulus, and neurons that showed decreased activity to a stimulus). In this latter case, in my analyses, I eliminated the non-responders and I did a binomial exact test to see if the proportion of neurons showing an increased response were significantly greater/fewer in proportion to neurons showing a decreased response. 

Behavioral example:

http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3957330/pdf/nihms538444.pdf

Physiology example: (Attached). 

http://www.ncbi.nlm.nih.gov/pubmed/26691962

Thank you very much Tom for sharing your experience. I will go through these papers and come back to you if I have questions. Gaurav

Thank you Silvana and Dr. Ali for the guidance. I really appreciated it. 

Yes, I absolutely agree with approuch of Tom Morrison.You can divide your cases to subgroupes. But you can also add some more cases to your experimental group and again to count up all cases and receive the mean (m+-SEM). And after that you can compare the mean values in control and experimental groupes. If you have enough number of cases the difference between experimental and control groups ( as a whole, not divided) may be statisticaly significant. Finaly, if you have not too many cases in your subgroups,  preferablу to choose nonparametric tests for statistic analysis.

Dear Tom!  Сertainly, it is possible, especially if isolated groups differ sharply qualitatively. Naturally, the more the number of subgroups with different signs, the greater must be the whole sample in this experiment. 

I have a follow up question: how do you decide where the cutoff of the responders and non responders? Is it done arbitrary, or is there a statistical test that sets a cutoff? E.g., Tom, in your publication you set the cutoff to 15%. How did you come up with that number?