I have a medical dataset that contains maternal and foetal data during pregnancy. There are some missing values in the dataset that I am unsure how to handle.
Here is a short example of my dataset:
id insulin ultrasound_AC 0 33 2651 1 2743 2 29
Patient 0 was prescribed insulin at 33 weeks gestation, patient 2 at 29 weeks. Whereas patient 1 was not prescribed insulin, hence the missing value. Similarly, patient 0's foetus had an ultrasound abdominal circumference measurement of 2651, patient 1 had a measurement of 2743, whereas patient 2 has a missing value for this feature, probably due to not attending this ultrasound appointment.
I am wondering how to handle these missing values.
In the case of the insulin feature, imputing missing values here would be incorrect, as the patients with missing values were never prescribed insulin. I could use SimpleImputer to fill all missing values with zeros, or would this be interpreted by a ML model as though the patient was prescribed insulin from the start of pregnancy?
As for the ultrasound abdominal circumference measurement, I could impute missing values using some imputation method, such as KNN Imputation, but with the data being medical, I am unsure if this is the best method as I do not want to modify the data too much.