Regression without time variable


I have a large dataset from a register based study. We are using Stata 17.
We have demographics, dates for hospitalisations and diagnosis codes from hospital visits in a given period. We want to do a regression on potential risk factors for myocardial infarctions, however, we only have the diagnoses codes registered at discharge or post-discharge outpatients visits. As far as I know we cannot do a cox regression without the time variable, which we are lacking as we don't know which day during admission the patient was diagnosed with a MI. I know there is a "new" command in Stata called Stintcox, but I'm not very familiar with it and cannot find much info on how to use it. We considered Poisson Regression, however, as far as I understand, the dependent variable should be a count, whereas here its a categorical variable (MI_90days yes/no).

Any good suggestions on appropriate statistical analyses to predict risk factors?

Thanks in advance


Less is more. Stay pure. Stay poor.
Please better describe your data. Perhaps provide a small sample which could be made up.

Yes in cox regression you need a time to event. This can be a continuous or discrete interval data, event at day 56 or event between 1-30 days.

The alternatives would be doing a logistic regression or generalized linear model with identity link and binomial distribution. The former would output odds ratios and later approach would output risk differences (if set-up properly - though if you have continuous predictors - bootstrapping is needed for standard errors if the model has convergence issues).