Topic Review - Newest First (only newest 5 are displayed)

rqayyum

Re: QUERY

Your response variable is dichotomous, i.e. there are only two answers to it. You can’t use any statistical test which will need continuous data. Your data is more complex than you realize (I think). You have outcome measurements at three different time inervals. You can model your data in three ways:

1. Use each time interval independent and separately. Thus you will run statistical test at Day 7, then at Day 14 and then at Day 28. You will have three different results and you will need to interpret these results separately. You may want to draw some conclusions based on the way results look but you will be unable to state (statistically) that one treatment results in earlier resolution than the other.

2. Pool all three time periods, either by stating that once there is a good response that subject will be labeled as ‘good response’ or vice versa, or you can pool all three results numerically. This is keeping in mind that outcomes can vary at each follow-up. I am assuming this based on repeated follow-up, otherwise there is no need to use repeated follow-up data and you can use Day 28 results alone.

3. Develop a model that fully incorporates the information-content and complexity of your data. Such a model should take into account the within-individual correlation (correlation between the three measurements) and then relationship with treatment assignment

Below are possible solutions to your problem (if I understand it correctly) ranked from simple to complex (there are other ways as well).

1. Probably the simplest approach will be to use chi-square test. If you have fewer numbers in one cell you may have to use Fischer’s exact test. You cannot model data as noted above under item 3.

2. A better approach will be to use regression models, such as logistic, in which you can control for confounding variables such as age, sex, skin photo-type etc. Again in a simple logistic regression model, you will not be able to model data as alluded to under item 3 above.

3. The best approach, and the one which I would prefer to use, will be to use multilevel modeling with mixed-effects. At the first level, you will model individual observations from each patient. The second level of model will be built from the first level and give you results of the treatment effect. Confounding variables (both at individual measurement and at patient level) can be easily adjusted in this model.

docosama

Re: QUERY

The statistical tests will usually depend on your variable types, sample size and objectives. If you are comparing two variables, then usual tests are t-test and chi-square. In t-test, you actually compare the means of both groups (variables), while in chi-square, you compare the proportions of two groups. So, for t-test you need continuous variables or one continous and other categorical. In chi-square you need both categorical variables. The result of the test will give you the p-value.

SHAHERYAR

QUERY

I am doing my FCPS in Dermatology.
I need some help regarding my dissertation.

The topic of my synopsis is ¡§ comparison of response of permethrin with ivermectin in the treatment of scabies¡¨
OBJECTIVES:
The objective of this study is to compare the response of Ivermectin with permethrin in the treatment of scabies.

OPERATIONAL DEF:
Response will be taken as absence of pruritis and disappearance of lesions and will also include presence or lack thereof of side effects.

HYPOTHESIS:
„« Null hypothesis: The response of Ivermectin is same as the standard treatment i.e., permethrin for scabies.
„« Alternate hypothesis: The response of Ivermectin is different to that of permethrin

Study design: Quasi- experimental

Response means absence of pruritis and papules. i.e., either it is yes or no. I have divided the patients in 2 treatment groups. I have evaluated the patients after 7, 14 and 28 days.

Now please tell me which statistical test should I use to compare the response between two groups?