Param Revamp - HIV

[mmsuarezcosta]

Files

Files changes:

• parameters.csv
• scaleup_parameters.csv
  • It is important to review the modifications I have made to this file. I have updated to the standard parameters.csv format, thus the column target_value has been renamed value, and the max and mins have been changed to prior_min and prior_max. The logic is that anything within the scale_up file is a target by definition, so it is not required to label the value itself as target.

Files used in hiv module that require refactoring (request for @mnjowe to support in future iterations):

• hiv_prevalence.csv
• time_since_infection_at_baselin.csv
• MoH_numbers_tests.csv, spectrum_treatment_cascade.csv (they contain historical data, so refactoring may not be required)

Calibration Data Files

• LHC_samples.csv
• unaids_infections_art2021.csv
• unaids_mortality_dalys2021.csv
• children0_14_prev_AIDSinfo.csv
• unaids_program_perf.csv
• DHS_prevalence.csv
• MoH_number_art.csv
• MPHIA_incidence2020.csv
• MPHIA_prevalence_art2020.csv
• MoH_numbers_tests.csv (also used in HIV.py)

Files Not Used in Code (May be legacy - consider deleting)

• calibration_from_aids_info.csv
• MoH_CPT_IPT2020.csv
• unaids_pmtct2021.csv
• MPHIA_incidence2020.csv (output_plots.py references 2015 version that doesn't exist, consider updating reference)
• MPHIA_prevalence_art2020.csv (output_plots.py references 2015 version that doesn't exist, consider updating reference)

Parameters

Assumptions

• mean_months_between_aids_and_death
• mean_months_between_aids_and_death_infant
• rr_fsw (discrepancy in write-up value 20 and parameter.csv value of 25)
• rr_behaviour_change
• proportion_reduction_in_risk_of_hiv_aq_if_on_prep
• mean_months_between_aids_and_death_infant

Undetermined, calibration

I cross-checked the write-up and notes in the parameter CSV file to determine which I believed to be calibrated.

• hiv_testing_rate_adjustment, treatment_initiation_adjustment,vs_adjustment: believe all these are calibrated based on write-up and parameter documentation in code
• Circumcision parameters (prob_circ_after_hiv_test, increase_in_prob_circ_2019, prob_circ_for_child_before_2020, prob_circ_for_child_from_2020): Based on write-up, vmc was calibrated to national rates, so I have labeled all these as calibrated
• aids_tb_treatment_adjustment: believe it is calibrated because of adjustment label
• Other testing rate probabilities (e.g. prob_hiv_test_at_anc_or_delivery) I have marked as local, but they may have been adjusted for calibration.

Hard-Coding

• Any values the LinearModels I have left hard-coded. Example:

        predictors = [
            Predictor("age_years").when("<15", 0.0).when("<49", 1.0).otherwise(0.0),
            Predictor("sex").when("F", p["rr_sex_f"]),

• Consumables and equipment

Notes (Future development)

• More thoroughly review designation of local and universal (I did not have time to exhaustively review)

[marghe-molaro]

Thanks @mmsuarezcosta, this looks good to me, and I think is ready for @tdm32's review!

[tdm32]

this should be the max age of AGYW (age_max_agyw)

[tdm32]

prep_cons_notavailable_retry_days - should these parameters which describe repeat healthcare seeking behaviour if no cons available be standardised? One param for repeat ARV/PrEP/VMMC etc if no consumables on the day?

[marghe-molaro]

Thanks @tdm32 , I think you are in the best position to decide how disaggregated these should be. If currently there is one single assumption around repeated tries, regardless of the consumable type, then we could definitely standardise/unify it.

[tdm32]

I think the way it's done here is right, seeking PrEP and seeking VMMC are two very different processes so fine as it is. Thanks.

[tdm32]

On the sheet scaleup_parameters.csv columns max and min have been changed to prior_min and prior_max. This is not quite right, these values are hypothetical values for reducing program to min performance and scaling up to max performance for scenario analyses.

[marghe-molaro]

On the sheet scaleup_parameters.csv columns max and min have been changed to prior_min and prior_max. This is not quite right, these values are hypothetical values for reducing program to min performance and scaling up to max performance for scenario analyses.

Thanks @tdm32 , absolutely prior_min and prior_max should be N/A for scenario parameters so I will correct this.