Data structure for p-value analysis of medical data I have a good-sized data set containing details and outcomes of various types of injury. I've worked through it and cleaned the data as best I can - it's now consistent and complete.
At this point I'm focusing on 3 main types of injury: gunshot, stab wound and blunt force trauma (variable values of 1, 2 and 3). Outcomes are binary: survival and death (0 and 1).
The hypothesis I'd like to test is "Gunshot victims are more likely to die from their wounds". (The mortality rates are already well known, so this analysis is mostly just an exercise)
I'm not a statistician, so will likely have the analysis done by someone else. My question is this: how should I structure the data for analysis? I currently have a spreadsheet containing all the data with many more points (ICU duration, etc). Should I submit just 2 columns pertaining to the aforementioned variables? Note I'm not testing in relation to other variables such as injury location etc.
Alternatively, if testing the hypothesis / calculating p-score is something I can do myself with reasonable reading and effort that would be great - I'm always happy to learn something new.
 A: It sounds like you have one categorical predictor/independent variable (injury) and one binary/categorical response/dependent variable (survival outcome). While many analysts have preferences for data formatting, the most versatile format for sharing your data is probably a spreadsheet (in .csv or a variation of excel format) that has 5 columns:




Gunshot
Stab
Blunt Force
Survived
Died




1
0
0
0
1


0
1
0
1
0


0
0
1
0
1




A value of '1' in the column indicates the categorical event occurred for the experimental unit (a victim in your case). I would be surprised if any analysts can't easily manipulate this format into their more preferred formatting should choose to.
As for performing this analysis yourself, I would highly recommend taking 15-30 minutes to read about the capabilities of Excel to help in performing this analysis. For people of all backgrounds, learning this task for "simple" analyses requires minimal effort in return for a useful skill. Some test options are Chi-Squared test for association or a linear regression.
A: To answer your main question, if you are certain that all you want is to test the association between injury type and death, then in the interests of only sharing data when necessary you should just send those two variables to the analyst.  You could even just send a cross-tabulation of the two columns, this would be sufficient for the analysis.  Your interpretation would be strictly limited to measuring this association.  The analyst will also want to know how the data was sampled.
(This assumes that 'survived' and 'died' are well enough defined, eg as survived or died within 28 days or something like that, you could imagine somebody dying from an unrelated cause or dying many months after the initial injury).
You should be able to do this analysis yourself.  The most suitable method will depend on your sample size, the numbers of people with each type of injury and the number who died, as well as how the data was sampled.  That is, whether it a case-control study whereby individuals were sampled on the basis of whether they died or not, were they sampled on the basis of their injury or are they a randomly selected cross-section of a population?  Technically the analysis shouldn't be difficult.  You could explore chi-squared tests if you have enough data and all you want is a p-value, or a suitable regression analysis to estimate rates and associations.  Depending on how much data you have your study might not be powerful enough to detect reasonable differences in rates between groups, so I would tend to prefer a method such as regression analysis that also gave confidence intervals for what the actual differences in survival rates might be, even if not statistically significantly different.
