As long as you have not published your study in a peer-reviewed journal, you can increase the sample size, add any new parameters or redesign and so on. Even you can collaborate with other people doing similar studies and you can publish together. 

Assume that the prior studied was not published, this issue could be resolved in the following manner:

(i) PRIOR SAMPLE: Use the prior sample n = 35 as sample 1. Leave it as it is. Obtain the needed information as required, i.e. descriptive, inferential, modeling, etc. Use it as a comparative group to sample 2.

(ii) SUBSEQUENT SAMPLE: Assume that no additional factor or condition had been introduced to change the condition or observations, take a second sample as sample 2. Take a larger size if sample size is a concern, but n > 30 is adequate for purposes of statistical test. In this manner, sample 2 does not have to have the same patients from the sample 1.

It is not recommended that sample 1 and sample 2 be combined unless you can reproduce the same condition, same instrument (sample placement of the kineso tape) and "time lapse" is not a factor. The interaction with the patients may influence the result; thus, time may be a factor. If they need to be combined, it is necessary to verify that there are no significant difference between the two groups. See attached file for relevant test for homogeneity and group comparison.

WHERE COMBINED SAMPLE IS ALLOWED: If it has been found that the sample does not meet minimum sample size, the sample size may be increased. In such case, the "sampling" is considered one time sampling. The time between the initial and subsequent samples has not significantly lapsed to allow change condition or other stimulus to intervene. Both samples are part of one study. This is not the case at hand.

In a case where there is a significant time lapse, if the underlying condition under which the two samples are taken had changed, these are considered two separate samples in time and space. Thus, they should be treated as independent samples. In order to combine them they must be homogeneous. See relevant test in attached file.

IS IT LEGAL: There is no issue of legality here. However, there is ethical issue. It would be unethical if you do not disclose that the combined-samples were taken in two time periods and how long was the time lapse. Invariably, unless you can support by some test statistic to show that the two samples are the same, i.e. homogeneous, this study would fail on methodology review.

ALTERNATIVE TREATMENT OF SAMPLE 1: Assume you defined sample 1 where n = 35 as a test sample collected for purpose of reading descriptive and inferential statistics to arrive at a minimum sample size. THIS would be different. From the reading of this test sample of 35 counts, a minimum sample is determined for the study and this test sample is combined with the minimum sample. This approach would solve the problem so long as the time lapse issue could be explained away with appropriate statistical test that the test sample and the minimum sample are not significantly different.

Thanks alot Paul..........really very clear and helpful ! :)

the study hasnt been published yet.

Eliardo, actually the person was saying OA is relatively common disorder and most international journals ask for large sample size for common disorders.

You really need to perform a "power" analysis to actually determine the sample size required. This should have been performed prior to the start of the investigation. When you have completed enrollment according to that design - you should stop and report the findings. An independent monitoring group may perform interval analyses for reasons of patient safety - but this is not usually disclosed or reported. If you have this independent group and you need to increase your sample size to meet power requirement from a-priori analyses, then complete your study. Given the apparent interval analysis, unbliniding and reporting - rigorous experimental investigators would suggest that you perhaps use your current results to help design a better study and then publish those results after complying with the best in experimental study designs. 

Thank you David!!

I believe that Dr Heck's answer is the most accurate one. You could publish the results of your first study as a pilot study. You can use the results of the first study for power calculation (it's the same as sample size calculation) for your next study. Thus, the results of the first study will not be used again for the second study exception made of the sample size and power calculation. 

Dr. Ali,

Of course you can do it. My proposal is to follow the recommendations of Dr. Heck and Dr. Berjano.   

Thank you everyone .......i am going to publish current study as pilot study then.

I have a somewhat different approach to this problem.

Basically, approximately 15 to 20 cases in a test is sufficient to get a  trustfully answer, that is to assess the mean/median and dispersion (sd/var, range, interquartile range and so on) of your results. If you want to get a more trustfully answer you have to increase your  numbers substantially, perhaps to 200 cases.In most studies approximately 20 cases is a acceptable number. You have 35 cases, if you compare your results to historical data 35 is more than enough, if you compare two groups 35/2 is sufficient. There is one prerequisite for the above considerations, all cases must reach your primary end point(s). If you measure blood pressure, you get a blood pressure from all cases, so they all reach the primary end point. It your primary end point is reduction in new heart infarctions you need large samples  for the simple reason that only a few cases will reach the primary end point. The general rule is that you need a sufficient number of cases to expect to get approximately 20 new infarction in each treatment group. Basically, calculation of statistical strength and powering an analysis, is to calculate  the number to evaluate to get approximately 20 endpoint-cases in each treatment group.

Studies requiring large samples (>1000 cases) are typically studies where only few cases reach a primary end point. A study with large samples is not more reliable that a study with a enough  cases. whether the results are reliable depends on the number having each the primary end point..

The next problem is related to our bondage to statistical p-values (significances). I propose that the number 20 is sufficient for you to estimate the clinical implications of your findings. However, due to some mysterious reason, we like to have significant results (p