Need help with medical study


I'm writing a medical paper on asthma and I'm afraid I 've reached a point where I am unconfident of my statistics knowledge.

I am hypothesizing that there is a relationship between elevated concentrations of a substance in the blood (we will call it X) and severity of asthma.

Several papers have grouped patients by severity: [mild], [moderate] or [severe]

And for each group they have provided n (number of patients) and mean (with S.D) concentration of X.

Now I could use standard one-way ANOVA and show that the means of X for each group are significantly "different". However, it is my hypothesis (and is the case from the data) that in fact X rises in accordance with asthma severity (i.e. mild < moderate < severe).

All of this is fine and the ANOVA still returns significance, but ANOVA would also do so for anything where the means are not the same (i.e. significance could occur if mild > moderate > severe or mild > moderate < severe).

What I would like to know: is there a more elegant way of testing for the "directionality" of X amongst the different groups? And specifically to only seek significance for mild < moderate < severe. In some ways it's slightly analogous to doing a one-tailed t-test, but of course this is not possible for ANOVA. Do I build up 2 t-tests (mild+moderate & moderate+severe) and then somehow combine them?

Or is what I am doing unnecessary or philosophically wrong and should I just stick with ANOVA?

All feedback very much appreciated.


Omega Contributor
What program are you using? In SAS you can use a contrast statement to show this positive trend. Another option you could have been eluding at would be doing pairwise comparisons using ttests. You would want to correct your level of significance if this is the approach you end up using. Less desirable (probably), may be looking at a correlation.
I have not used SAS, but will look into it. More at this stage, I want to get an understanding of the correct approach so that I can be sure I'm not making a mistake. I will investigate SAS. Thanks


Ambassador to the humans
You don't need SAS to do it and if you don't have access to SAS you definitely don't want to *pay* to get it. You should be able to do this in almost any stats package - what program are you using?
Excel is probably the most advanced tool I have immediate access to... of course I can switch to R, but have no familiarity with it. If you can provide tips on that it would be very helpful. However, I remain interested to get some kind of idea of how such an approach is working under the hood (given the inputs by group I mentioned: n, mean and sd). But I guess I can figure that out for myself, after I find an acceptable method. Thanks