mortality prediction

mortality prediction

by Aaron -

hi everyone,

I'd like to recreate the attached figure (oxygenation index on x axis, but conceptually can be any predictor) vs probability of death on y axis, with different lines for the trends when the predictor was measured at any time points. The figure is from Trachsel et al AJRCCM 2005, if you're interested.

I am not sure how to accomplish this and would appreciate any tips.

Thanks,

Aaron

Re: mortality prediction

by Elizabeth Uy-Smith -
Hello Aaron, From the figure, I'm assuming that you have both the predictor (oxygenation index) and outcome (probability of dying) as continuous variables. In order to start recreating the graph, you can use the command: twoway line death oxygenindex You can also access the twoway dialog box using Graphics--> twoway. Click "Create" and select "Line." We used the twoway command in Lab 6 to create a twoway scatter, and you can use the same basic instructions to tweak your graph. Please let us know if you have more questions, and the Friday office hours are also a great resource. ellie

Re: mortality prediction

by Aaron -

thanks Ellie.

I will need to also generate a probability of death, because I have the actual mortality numbers but not the mortality rate at a given oxygenation index. But you have given me a fantastic start.

Re: mortality prediction

by Elizabeth Uy-Smith -

Hello Aaron,

Though I’m not too familiar with probability models, I believe you need two separate commands to (1) generate a model for the probability of death [with the option of controlling for other variables] and (2) creating a new variable for the probability of death using the previous model:

probit death [confounder1 confounder2]

predict probdeath

Hope that helps, and please let us know if you have more questions.

Thanks!

ellie

Re: mortality prediction

by Valy Fontil -

Aaron

If that doesn't work, you may have to run a regression analysis first and then try just the predict command.