possible inconsistencies in using get_blank_appt_footprint()

[tdm32]

Currently get_blank_appt_footprint() is used within some modules (HIV, TB etc) when a HSI has been scheduled but not resources are actually occupied, e.g. person is not eligible for test or is already on treatment. These HSIs are logged as having been delivered and therefore contribute to the total numbers of HSIs in outputs, even though no person-time / consumables are required.

When no resources are required, should a return statement be used so that this HSI is not logged? The current set-up may be skewing results where HSI numbers requested / delivered are used.

@tbhallett @marghe-molaro What do you think?

[tbhallett]

Thanks @tdm32

I agree with this being misleading if people are just looking at counts of HSI Event occurrences.

We could make a new logging key, which is for counts of appointment that incurred some HCW time (i.e. ran, and did not return a blank footprint). Would that help...?

[tbhallett]

@sakshimohan also notes here

Here is the script that I have used to get the total number of HSIs based on Number_By_Appt_Type_Code and TREATMENT_ID in the healthsystem.summary logger. Number_By_Appt_Type_Code gives me a 40% of the HSIs in total than what's recorded under TREATMENT_ID.

# Load outputs
results_folder = Path(outputspath / 'sakshi.mohan@york.ac.uk/impact_of_consumables_scenarios-2024-06-11T204007Z/')
log = load_pickled_dataframes(results_folder)

TARGET_PERIOD = (Date(2010, 1, 1), Date(2019, 12, 31))

def get_counts_of_hsi_by_treatment_id(_df):
    """Get the counts of the short TREATMENT_IDs occurring"""
    _counts_by_treatment_id = _df \
        .loc[pd.to_datetime(_df['date']).between(*TARGET_PERIOD), 'TREATMENT_ID'] \
        .apply(pd.Series) \
        .sum() \
        .astype(int)
    return _counts_by_treatment_id.groupby(level=0).sum()

counts_of_hsi_by_treatment_id = extract_results(
        results_folder,
        module='tlo.methods.healthsystem.summary',
        key='HSI_Event',
        custom_generate_series=get_counts_of_hsi_by_treatment_id,
        do_scaling=True
    )

def get_counts_of_hsis_by_level(_df):
    flattened_data = []
    for level, appt in _df['Number_By_Appt_Type_Code'].items():
        for appt, count in appt.items():
            flattened_data.append((level, appt, count))

    flattened_data = pd.DataFrame(flattened_data, columns=['Facility_Level', 'Appointment_Type', 'HSI_count'])
    flattened_data.set_index(['Facility_Level', 'Appointment_Type'], inplace = True)
    return pd.Series(flattened_data.stack()).fillna(0).astype(int)

counts_of_hsi_by_level = extract_results(
        results_folder,
        module='tlo.methods.healthsystem.summary',
        key='HSI_Event',
        custom_generate_series=get_counts_of_hsis_by_level,
        do_scaling=True
    )

total_hsis_using_log_by_facility_level = counts_of_hsi_by_level.sum()[0]
total_hsis_using_log_by_treatmentID = counts_of_hsi_by_treatment_id.sum()[0]

print(f"Baseline scenario (5 runs): \n HSIs based on 'Number_By_Appt_Type_Code' {total_hsis_using_log_by_facility_level} "
      f"\n HSIs based on 'TREATMENT_ID' {total_hsis_using_log_by_treatmentID}")

print(f"For run = 0, the total number HSIs as per 'Number_By_Appt_Type_Code_And_Level' key is {round(total_hsis_using_log_by_facility_level[0],2)},"
      f"whereas the total number of HSIs as per the 'TREATMENT_ID' key is {round(total_hsis_using_log_by_treatmentID[0],2)}")

@tbhallett responded:

I think this is to do with issue #1401 --- which notes many appointments that run have a blank_foot_print, meaning that that are counted in the HSI_Event logs used to generate the number of events by TREATMENT_ID.... but do not figure in 'Number_By_Appt_Type_Code' (because they do not have an Appt_Type).
It's a bit of a weird one! The top-line answer is that we're still thinking about but are leaning towards making an edit that avoids this by either; (i) making sure blank_appt_footprintgs are counted in ''Number_By_Appt_Type_Code'; or (ii) (more likely and easier) not counting as having run, any HSI_Event that returns a blank foot print.