Propensity score Matched Clinical End Points

[Reinout Heijboer]

Hi,

For my propensity score matched cohort I want to calculate odds ratios for dichotomous and skewed categorical outcomes.
For dichotomous outcomes I use the McNemar test (using the mcc or mcci command in STATA). For categorical outcomes I'll either use the Wilcoxon signed-rank test or paired T-test (Austin; Propensity-score matching in the cardiovascular surgery literature from 2004 to 2006: A systematic review and suggestions for improvement; 2007).

How do I obtain odds ratios from these test?
Also, I'm not quite sure how the ranksum command is used.

Thanks,


Reinout

[Clyde Schechter]

-mcc- and -mcci- will give you the odds ratios directly in their output. Nothing more to do there.

Using the paired t-test or signed-rank test will not give you odds ratios and there is nothing you can do to calculate odds ratios for them. Moreover, it is simply inappropriate to use these tests for categorical variables. The signed-rank test is only valid for variables that are at least ordinal, and the t-test (paired or not) is only meaningful for interval-level data (or ordinal-level data that is thought to be equally spaced, hence treated as interval level.)

If you want to compare the distributions of a polychotomous categorical outcome between matched cases and controls you have to re-arrange your data set so that the observations are not individuals but the matched pairs, and the case and control outcomes are separate variables. This will typically be done by -reshape wide-. Once you have that you can use the -symmetry- command. See -help symmetry-. This will enable you to test whether the case and control outcome distributions are similar. It will not, however, give you odds ratios. I don't even see how an odds ratio could be defined for this situation, to be honest.

[Bree Shi]

Hi Clyde,

Thank you for your answer. I am just wondering whether I can use - symmetry- for matched cohort dataset (1:1) to test differences on categorical (k>2) baseline variable?

Many Thanks,
Bree

[Clyde Schechter]

Your question is a little vague and I'm not sure exactly what you want to do, but generally speaking, yes, -symmetry- is used for 1:1 matched data when you are comparing the distribution of a polytomous outcome.

[Bree Shi]

Hi Clyde,

Thanks for your reply. I have a 1:1 matched cohort data, matched on a few criteria. I need to compare whether there is significance differences between the baseline characteristics (i.e. age, gender, ethnicity etc.) between the two matched groups. I am currently using paired t-test for continuous variables and McNemar’s test for dichotomous variables. So just wondering whether I can use -symmetry- for comparing baseline characteristics with more than 2 categories. If I can, is it correct to call it transmission/disequilibrium test (TDT)? (I get the name from -help symmetry-)

Many Thanks,
Bree

[Clyde Schechter]

Yes and yes. But don't call it the TDT unless your audience will be familiar with the term.