Patient's visits list is not ordered - solved it by adding two lines

[smessica]

I'm working with the MIMIC 3 dataset using your library, and there is a bug in the order of visits per patient that not only affects the logic when trying to create custom task functions but also creates a bug in your readmission task function (https://github.com/sunlabuiuc/PyHealth/blob/master/pyhealth/tasks/readmission_prediction.py) and maybe in other tasks too.

The issue is under https://github.com/sunlabuiuc/PyHealth/blob/master/pyhealth/datasets/mimic3 (mimic3.py file) in the basic_unit(p_id, p_info) function (line 105):

In the line 114 the groupby ruins the order of the visits by their date:
for v_id, v_info in p_info.groupby("HADM_ID"):

I fixed it by adding 2 lines of code: (marked in bold)

1. p_info = p_info.sort_values(by="ADMITTIME") (to make sure the visits are ordered according to their date, but it might work without this row as well)
2. for _, row in p_info.iterrows(): (instead of the groupby)

def basic_unit(p_id, p_info):
p_info = p_info.sort_values(by="ADMITTIME")

        patient = Patient(
            patient_id=p_id,
            birth_datetime=strptime(p_info["DOB"].values[0]),
            death_datetime=strptime(p_info["DOD_HOSP"].values[0]),
            gender=p_info["GENDER"].values[0],
            ethnicity=p_info["ETHNICITY"].values[0],
        )
        # load visits
        for _, row in p_info.iterrows():
            visit = Visit(
                visit_id=row["HADM_ID"],
                patient_id=p_id,
                encounter_time=strptime(row["ADMITTIME"]),
                discharge_time=strptime(row["DISCHTIME"]),
                discharge_status=row["HOSPITAL_EXPIRE_FLAG"],
                insurance=row["INSURANCE"],
                language=row["LANGUAGE"],
                religion=row["RELIGION"],
                marital_status=row["MARITAL_STATUS"],
                ethnicity=row["ETHNICITY"],
            )
            # add visit
            patient.add_visit(visit)
        
        return patient

If you want some visual proof, we can discuss it further via email or something; I did not want to upload here MIMIC 3 information since it is sensitive.