Adjusting for Race/Ethnicity

[BTAinRVA]

I'm rather new to the healthcare field and we are working on pediatric appendicitis outcomes. In the literature it is known that the racial/ethnic disparities in pediatric appendicitis outcomes are large. Unfortunately our database does not have reliable data on Race/Ethnicity. If we don't adjust for Race/Ethnicity, will our estimates for all other variables be highly suspect? Should we even move forward with the study?

Thanks,

Brian

[Reeza]

Your (All) study will have limitations - these are big ones. 

Is the data you're able to produce of value to others - for other research or for comparison even within your area?

[BTAinRVA]

Reeza,

Thanks for the quick reply. I'm just wondering if it's worthwhile to pursue this study further given that we can't adjust for race/ethnicity, knowing that it most likely a significant factor.

[Reeza]

Here's the thing - when you adjust for race - it means you account for the fact that race plays an impact - but it shouldn't. So doing analysis and learning where to target treatments, without race being a factor isn't wrong. If this is solely an academic exercise, then it may not be worth it. If you're trying to implement changes in your facility that's a different question.  

[ballardw]

Depending on which analysis is being performed "adjusted" often means summaries by race: i.e. summarize a factor for each level of the race variable.

You might provide a little more information about what you consider unreliable. Is the data not collected at all, collected occassionally or most of the time? If you are only missing 5% of records it likely is as good as many other data sources. Or do you have clients that somehow have different race or ethnicity recorded at different times? (Don't laugh, that is one the data quality checks I run on a routine basis and see a problem with about 2% of records from on data source.)

Is it a case of ethnicity being recorded as race?

And if it is your own data it may not hurt to get management attention to this lack of detail. It may be a symptom of other data issues not discovered such as recording only one complication with a procedure when there may have been multiples or improperly indicating the severity.

[BTAinRVA]

Reeza & BallardW,

Thanks again for your replies. They are GREATLY appreciated.

Essentially we are trying to replicate this study here: http://pophealthmetrics.biomedcentral.com/articles/10.1186/1478-7954-3-4

I am pretty sure there has to be a database that has the race/ethnicity data we need but the ones we usually work with has a field for that but it is only about 5% populated.

This is not an academic exercise; we definately are interested to see if other factors are significant. But can we have faith in the result if we can't control for race?

[Reeza]

Do you have individual names? It's obviously not very reliable but there at algorithms to match name to race. 

[BTAinRVA]

Reeza,

Yes we do have individual names and addresses. Do you have an algorithm in mind? I quickly googled it and found an article on Bayesian Surname and Geocoding (BSG) method at NCBI.

Thanks again!