# Analyze treatment effect, which test to use?

#### tonip

##### New Member
The heartrate of patients was measured at two times before and after they were given a treatment. If the healthy range for heartrate were 60-90, which statistical test should be used to analyze if the treatment was effective? Paired t-test? Which statistical analysis handles the improvement in two directions (a patient with lower than 60 moving to the range is improving, and same for higher than 90 moving to the range).
What sample size is the required minimum?

#### obh

##### Active Member
Hi Tonip,

Just to be sure, do you really have a treatment that improves both high blood pressure and low blood pressure? or do you mix 2 treatments?

The paired t-t test is equivalent to running the one sample t-test over the differences.
Technically you can try to calculate the improvement, and run one sample t-test over the improvement. (if meets the t-test assumptions)

before after improvement
30 40 10
40 38 -2
170 140 30
180 182 -2

But it doesn't feel right. Even if the same treatment does both... It is probably won't be normally distributed and not symmetric.

You try to achieve a sample size that will gain the test required power, usually 0.8
http://www.statskingdom.com/sample_size_all.html

Last edited:

#### tonip

##### New Member
Hi obh,

Thanks. Yeah, it's same treatment to improve heartrate.

#### Karabiner

##### TS Contributor
How large is your sample size?

Were ALL treated patients outside the 60-90 range before treatment?

And what is you clinical end point? Being in the 60-90 range after treatement (regardless of whether 60, 61, 70, 80, 89, or 90, for example)? Or perhaps something else, like "increase of heart rate by 10 or more, if the patient was < 60 before treatment".

With kind regards

Karabiner

#### obh

##### Active Member
Hi obh,

Thanks. Yeah, it's same treatment to improve heartrate.
I assume you would need to draw histogram to see the data distribution. I guess it won't distribute normally or symmetrically.

Regard the sample size, I think Karabiner that this was the question? . I assume you can start to calculate it base on the t distribution, and later check if it okay.