What statistical analysis to compare these two groups? - USMLE Forums
 USMLE Forums         Your Reliable USMLE Online Community     Members     Posts
 Home USMLE Articles USMLE News USMLE Polls USMLE Books USMLE Apps
 USMLE Forums What statistical analysis to compare these two groups?
 Register FAQs Members List Search Today's Posts Mark Forums Read

 USMLE Step 1 Forum USMLE Step 1 Discussion Forum: Let's talk about anything related to USMLE Step 1 exam

#1
05-04-2012
 USMLE Forums Master Steps History: 1+CK+CS+3 Posts: 1,353 Threads: 94 Thanked 873 Times in 506 Posts Reputation: 886
What statistical analysis to compare these two groups?

=======

A researcher has designed a retrospective study, where patients undergoing laparoscopic and open operations for right hemicolecomy are compared. He has identified 30,000 patients records from National Inpatient Sample Database and created similar groups of patients treated by laparoscopic (N=15,000) or open (N=15,000) approach with matching demographical characteristics. Within the laparoscopic group, 5% died and within open 15% died. if he wants to compare the immediate perioperative outcomes for mortality within these two groups, what statistical test should he use?

a. Fisher's exact test
b. Paired T test
c. Chi-Square test
d. ANOVA
e. Two-Sample T test
f. Relative Risk
__________________
Ambition is a dream with a V8 engine. EP
 The above post was thanked by: bebix (05-04-2012), drhma (05-04-2012)

#2
05-04-2012
 USMLE Forums Master Steps History: 1 + CS Posts: 675 Threads: 84 Thanked 424 Times in 206 Posts Reputation: 434

chi square because of the percentages?
#3
05-04-2012
 USMLE Forums Addict Steps History: 1 + CS Posts: 193 Threads: 29 Thanked 134 Times in 71 Posts Reputation: 144

relative risk?
he wants to compare the same outcome [mortality] but in different procedures... open surgery is 3 x times more likely to die
although relative risk is associated with cohort, which is a PROspective study, dunno :S

whats the answer please , pm if u dont want to post it here now...
__________________
i'm so bad i make medicine sick,
Imma show you how great i am
#4
05-04-2012
 USMLE Forums Master Steps History: 1+CK+CS+3 Posts: 1,357 Threads: 194 Thanked 3,259 Times in 881 Posts Reputation: 3269

Dear DocSikorski

I know that you created this question...

a. Fisher's exact test No for matched pairs
b. Paired T test no for inference in proportions/rates
c. Chi-Square test NO for matched pairs
d. ANOVA no for inference in proportions/rates
e. Two-Sample T test no for inference in proportions/rates
f. Relative Risk THIS is NOT a TEST.

you should re-write the question not using the "matched" assumption

good luck to all!
 The above post was thanked by: drhma (05-04-2012)
#5
05-04-2012
 USMLE Forums Addict Steps History: 1 + CS Posts: 193 Threads: 29 Thanked 134 Times in 71 Posts Reputation: 144

Quote:
 Originally Posted by bebix Dear DocSikorski the correct answer is NONE! I know that you created this question... a. Fisher's exact test No for matched pairs b. Paired T test no for inference in proportions/rates c. Chi-Square test NO for matched pairs d. ANOVA no for inference in proportions/rates e. Two-Sample T test no for inference in proportions/rates f. Relative Risk THIS is NOT a TEST. you should re-write the question not using the "matched" assumption good luck to all!
bebix dear bebix... thank u for still hanging around the forum or even if suddenly seeing this
great explanation

good luck to u in whatever u are doing now these days
__________________
i'm so bad i make medicine sick,
Imma show you how great i am
 The above post was thanked by: bebix (05-04-2012)
#6
05-04-2012
 USMLE Forums Master Steps History: 1+CK+CS+3 Posts: 1,353 Threads: 94 Thanked 873 Times in 506 Posts Reputation: 886

Quote:
 Originally Posted by bebix Dear DocSikorski the correct answer is NONE! I know that you created this question... a. Fisher's exact test No for matched pairs b. Paired T test no for inference in proportions/rates c. Chi-Square test NO for matched pairs d. ANOVA no for inference in proportions/rates e. Two-Sample T test no for inference in proportions/rates f. Relative Risk THIS is NOT a TEST. you should re-write the question not using the "matched" assumption good luck to all!
uhm...
why not?

to compare apples to apples you have to have groups matched....
__________________
Ambition is a dream with a V8 engine. EP
#7
05-04-2012
 USMLE Forums Master Steps History: 1+CK+CS+3 Posts: 1,353 Threads: 94 Thanked 873 Times in 506 Posts Reputation: 886

I believe, that when you are able to write a good question - you really aced the topic.

Tests are my weak area, but I tried to make a good one
__________________
Ambition is a dream with a V8 engine. EP
#8
05-04-2012
 USMLE Forums Master Steps History: 1+CK+CS+3 Posts: 1,357 Threads: 194 Thanked 3,259 Times in 881 Posts Reputation: 3269

Quote:
 Originally Posted by DocSikorski uhm... why not? to compare apples to apples you have to have groups matched....
hahaha...yes, and for that you need to use

Mc Nemar test or Logistic regression for matched p. test with robust standard errors, hahahaha...

so, this is not only apple vs apple...this is apple vs apple´s sister!, hahaha

#9
05-04-2012
 USMLE Forums Master Steps History: 1+CK+CS+3 Posts: 1,353 Threads: 94 Thanked 873 Times in 506 Posts Reputation: 886

Quote:
 Originally Posted by bebix hahaha...yes, and for that you need to use Mc Nemar test or Logistic regression for matched p. test with robust standard errors, hahahaha... so, this is not only apple vs apple...this is apple vs apple´s sister!, hahaha
That's why reviewers of the papers still have their jobs =)
haha
=============

Multiple regression analysis will work for not matched groups, where all the comorbidites and risks, prior to surgery can be weighted individually to estimate the effect on the outcome...

For matched groups: e.g. young patients with 0-1 comorbidites (low illness severity) of same races/ethnicity, gender and diagnosis (e.g. colon cancer will be most appropriate for right hemicolectomy, as I used in the question) Chi-square will work just fine (i used it several times), if there were less than 10pts that died in each, we could use Fisher's exact test, but thats kinda sample size that you don't wont to make a paper of haha
__________________
Ambition is a dream with a V8 engine. EP

Last edited by DocSikorski; 05-04-2012 at 12:46 PM.
#10
05-04-2012
 USMLE Forums Addict Steps History: Not yet Posts: 125 Threads: 17 Thanked 66 Times in 38 Posts Reputation: 76

Note:
If you have only nominal variable do chi-square,if you have only interval variable go for correlation,one nominal and one interval go for t-test,but if you have more than 2 variables (at least one of each) go for anova
#11
05-04-2012
 USMLE Forums Master Steps History: 1+CK+CS+3 Posts: 1,357 Threads: 194 Thanked 3,259 Times in 881 Posts Reputation: 3269

Quote:
 Originally Posted by DocSikorski That's why reviewers of the papers still have their jobs =) haha ============= Multiple regression analysis will work for not matched groups, where all the comorbidites and risks, prior to surgery can be weighted individually to estimate the effect on the outcome... For matched groups: e.g. young patients with 0-1 comorbidites (low illness severity) of same races/ethnicity, gender and diagnosis (e.g. colon cancer will be most appropriate for right hemicolectomy, as I used in the question) Chi-square will work just fine (i used it several times), if there were less than 10pts that died in each, we could use Fisher's exact test, but thats kinda sample size that you don't wont to make a paper of haha
hahaha...multiple regression (in this case logistic regression, which is part of the family of GLM) and weighted = with means marginal regression with sandwich estimation of the variance (robust SE)

So...anyone of your alternatives wasnt correct...dont worry, Im a Biostatistician

and X2 does not work...maybe is time to go back to the books
this is the bible for categorical analysis:

http://www.amazon.com/Statistical-Me...6164393&sr=8-1

good luck!

#12
05-04-2012
 USMLE Forums Master Steps History: 1+CK+CS+3 Posts: 1,357 Threads: 194 Thanked 3,259 Times in 881 Posts Reputation: 3269

Quote:
 Originally Posted by irakly Note: If you have only nominal variable do chi-square,if you have only interval variable go for correlation,one nominal and one interval go for t-test,but if you have more than 2 variables (at least one of each) go for anova

chi2 works only for independent samples
in case of matched data, you must use McNamer test.

good luck!
#13
05-04-2012
 USMLE Forums Master Steps History: 1+CK+CS+3 Posts: 1,353 Threads: 94 Thanked 873 Times in 506 Posts Reputation: 886

Thanks, I might ask my supervisor to order it...
__________________
Ambition is a dream with a V8 engine. EP

 Tags Biostatistics-Epidemiology, Step-1-Questions

Message:
Options

## Register Now

In order to be able to post messages on the USMLE Forums forums, you must first register.
User Name:
Medical School
Choose "---" if you don't want to tell. AMG for US & Canadian medical schools. IMG for all other medical schools.
 AMG IMG ---
USMLE Steps History
What steps finished! Example: 1+CK+CS+3 = Passed Step 1, Step 2 CK, Step 2 CS, and Step 3.

Choose "---" if you don't want to tell.

 Not yet Step 1 Only CK Only CS Only 1 + CK 1 + CS 1+CK+CS CK+CS 1+CK+CS+3 ---
Favorite USMLE Books
 What USMLE books you really think are useful. Leave blank if you don't want to tell.
Location
 Where you live. Leave blank if you don't want to tell.

## Log-in

Human Verification

In order to verify that you are a human and not a spam bot, please enter the answer into the following box below based on the instructions contained in the graphic.