Hi all, writing with a quick question regarding stata code for an analysis I'm carrying out. The brief background is that this is a secondary analysis of a case control study looking at risk factors for esophageal cancer. We have data on a number of exposures and we are trying to assess whether there are different risk factors for esophageal cancer across different age groups, ie 30-45, 45-60 and over 60 years. For our univariate analysis, I would like to compare OR for various exposures across age groups via the breslow day statistic test for homogeneity. The variables are defined as follows:
eventname = case vs. control (case = esophageal cancer, binary outcome)
riskfactor = exposure
agegroup = age group (categorical variable stratified across 3 groups 30-45 yrs, 45-60, > 60 yrs)
I've been using the following stata code:
cc eventname riskfactor, by(agegroup) bd
For BINARY risk factor variables, I've found that when using this code stata output gives the appropriate OR for each of the age-groups (confirmed using logistic regression command restricted to specific age stratifications ie "logistic eventname riskfactor if agegroup==1, logistic eventname riskfactor if agegroup==2, etc). The command above also gives a P value for the breslow day statistic.
. cc eventname riskfactor, by(agegroup) bd
agegroup | OR [95% Conf. Interval] M-H Weight
-----------------+-------------------------------------------------
30- | .6438152 .358623 1.154127 15.97619 (exact)
45- | 1.319615 .8166934 2.137096 16.19938 (exact)
60- | 1.089783 .6089814 1.946551 12.42308 (exact)
75- | 1.909091 .4977058 7.476565 1.98 (exact)
-----------------+-------------------------------------------------
Crude | 1.149634 .8692045 1.520809 (exact)
M-H combined | 1.051579 .7919205 1.396375
-------------------------------------------------------------------
Test of homogeneity (M-H) chi2(3) = 5.05 Pr>chi2 = 0.1684
Test of homogeneity (B-D) chi2(3) = 5.08 Pr>chi2 = 0.1658
Test that combined OR = 1:
Mantel-Haenszel chi2(1) = 0.12
However, for risk factor variables with multiple categories (3 or more), ie for my risk factor variable assessing region of patient's region (currentzone), I can calculate the OR for each of the zones within each of the age groups using the code below
logistic eventname i.currentzone if agegroup==0
logistic eventname i.currentzone if agegroup==1
logistic eventname i.currentzone if agegroup==2
logistic eventname i.currentzone if agegroup==3
But I'm wondering if the stata code for the breslow day statistic holds for these risk factors with multiple (3 or more) categorical variables, given the obvious difference in the output?
cc eventname currentzone, by(agegroup) bd
agegroup | OR [95% Conf. Interval] M-H Weight
-----------------+-------------------------------------------------
30- | 4.105477 2.057345 8.373765 4.672986 (exact)
45- | 2.739631 1.687973 4.450906 10.98734 (exact)
60- | 2.646592 1.620191 4.333867 10.98065 (exact)
75- | 2.357143 .9548818 5.895511 3.676768 (exact)
-----------------+-------------------------------------------------
Crude | 2.845181 2.151246 3.765284 (exact)
M-H combined | 2.870071 2.181033 3.77679
-------------------------------------------------------------------
Test of homogeneity (M-H) chi2(3) = 1.54 Pr>chi2 = 0.6727
Test of homogeneity (B-D) chi2(3) = 1.55 Pr>chi2 = 0.6713
Test that combined OR = 1:
Mantel-Haenszel chi2(1) = 58.10
Pr>chi2 = 0.0000
Any input would be greatly appreciated. Can also be reached by email. [email protected]. Thanks!
eventname = case vs. control (case = esophageal cancer, binary outcome)
riskfactor = exposure
agegroup = age group (categorical variable stratified across 3 groups 30-45 yrs, 45-60, > 60 yrs)
I've been using the following stata code:
cc eventname riskfactor, by(agegroup) bd
For BINARY risk factor variables, I've found that when using this code stata output gives the appropriate OR for each of the age-groups (confirmed using logistic regression command restricted to specific age stratifications ie "logistic eventname riskfactor if agegroup==1, logistic eventname riskfactor if agegroup==2, etc). The command above also gives a P value for the breslow day statistic.
. cc eventname riskfactor, by(agegroup) bd
agegroup | OR [95% Conf. Interval] M-H Weight
-----------------+-------------------------------------------------
30- | .6438152 .358623 1.154127 15.97619 (exact)
45- | 1.319615 .8166934 2.137096 16.19938 (exact)
60- | 1.089783 .6089814 1.946551 12.42308 (exact)
75- | 1.909091 .4977058 7.476565 1.98 (exact)
-----------------+-------------------------------------------------
Crude | 1.149634 .8692045 1.520809 (exact)
M-H combined | 1.051579 .7919205 1.396375
-------------------------------------------------------------------
Test of homogeneity (M-H) chi2(3) = 5.05 Pr>chi2 = 0.1684
Test of homogeneity (B-D) chi2(3) = 5.08 Pr>chi2 = 0.1658
Test that combined OR = 1:
Mantel-Haenszel chi2(1) = 0.12
However, for risk factor variables with multiple categories (3 or more), ie for my risk factor variable assessing region of patient's region (currentzone), I can calculate the OR for each of the zones within each of the age groups using the code below
logistic eventname i.currentzone if agegroup==0
logistic eventname i.currentzone if agegroup==1
logistic eventname i.currentzone if agegroup==2
logistic eventname i.currentzone if agegroup==3
But I'm wondering if the stata code for the breslow day statistic holds for these risk factors with multiple (3 or more) categorical variables, given the obvious difference in the output?
cc eventname currentzone, by(agegroup) bd
agegroup | OR [95% Conf. Interval] M-H Weight
-----------------+-------------------------------------------------
30- | 4.105477 2.057345 8.373765 4.672986 (exact)
45- | 2.739631 1.687973 4.450906 10.98734 (exact)
60- | 2.646592 1.620191 4.333867 10.98065 (exact)
75- | 2.357143 .9548818 5.895511 3.676768 (exact)
-----------------+-------------------------------------------------
Crude | 2.845181 2.151246 3.765284 (exact)
M-H combined | 2.870071 2.181033 3.77679
-------------------------------------------------------------------
Test of homogeneity (M-H) chi2(3) = 1.54 Pr>chi2 = 0.6727
Test of homogeneity (B-D) chi2(3) = 1.55 Pr>chi2 = 0.6713
Test that combined OR = 1:
Mantel-Haenszel chi2(1) = 58.10
Pr>chi2 = 0.0000
Any input would be greatly appreciated. Can also be reached by email. [email protected]. Thanks!
