Use of Xpert consumable is not correctly logged

[sakshimohan]

The use of Xpert consumable (item_code 187) is currently logged as 0 across all scenarios generated by job_id htm_and_hss_runs-2025-01-16T135243Z comparing horizontal versus vertical investments.

The tb module seems to request the use of Xpert correctly because it is logged as Not_Available under tlo.methods.healthsystem.summary, and the quantity not available correctly increases under TB scale up scenarios.

Xpert is delivered at level 2 facilities, with an average consumable availability recorded at 31%. The availability is 0% at level 1b facilities. The healthsystems.py module takes the average availability across levels 1b and level 2 (by district and month), so this is likely to bring the average availability down for the merged level 1b and 2 facilities used in the simulation, but not down to 0. So it's not clear what's happening here.

Additionally, there are major challenges with the disaggregated estimates of availability by month. For all districts, for 7 months, the availability recorded is close to 0 as per LMIS data. This is probably an artefact of OpenLMIS errors and needs to be corrected. See attached the availability at level 2 facilities by District (rows) and month (columns).

[tdm32]

I have a couple of first suggestions:

• set the availability of xpert to 1 and double-check that it is being recorded as Item_Used in the consumables log
• check the dates that xpert is being requested (Tara) to make sure the requests are not all clumping in the first half of the year
• simplify the tb diagnostic algorithm to make the debugging process more straightforward (Tara)
• check the NTP reports to see how many xpert tests are being administered and over-write the availability to match with the reports

[sakshimohan]

@tdm32 I have just discovered that this problem is arising because the healthsystem.py script sets the availability of consumables at level 2 equal to that at level 1b. See code snippet.

So the solution for this problem is simply on of the following -

1. updating this code snippet to reflect that availability should be equal to the average of availability at levels 1b and 2 (weighted by the number of facilities at each level)
2. updating the availability of Xpert cartridges at level 1b to match that at level 2.
   In both cases, we should update the availability of Xpert for the months during which it was recorded as 0 even at level 2.

[sakshimohan]

@tdm32 I have just discovered that this problem is arising because the healthsystem.py script sets the availability of consumables at level 2 equal to that at level 1b. See code snippet.

So the solution for this problem is simply on of the following -

1. updating this code snippet to reflect that availability should be equal to the average of availability at levels 1b and 2 (weighted by the number of facilities at each level)
2. updating the availability of Xpert cartridges at level 1b to match that at level 2.
   In both cases, we should update the availability of Xpert for the months during which it was recorded as 0 even at level 2.

Hi @tbhallett. It looks like Xpert availability is taken as 0 in the model because the availability at level 2 facilities is assumed to be equal to level 1b, as described above. I have proposed two alternative solutions above. Do you have a preference?
I would lean towards option 1 because there might be other services which are only available at level 2 and above and this would prevent the model from assuming the corresponding consumable availability to be equal to 0 for this, based on level 1b availability.

[tdm32]

Noting here the quick fix used in the ROI paper:
consumables availability at levels 1b and 2 are set to the mean at each facility level. For xpert, we only see availability from Sept 2018 onwards in OpenLMIS, this means many months are showing 0 availability at every level.

From NTP reports, we know that xpert is used from 2015:
2015 - 15,000 xpert tests
2016 - 20,000
2017 - 35,000
2018 - 60,600
approx 85,000 per year from 2019

This comprises 45-48% of all TB tests. As not all are eligible for xpert testing, this suggest a high availability for those requesting it. NTP reports state low chance of stockout but do not give a figure.

In TLO, all xpert is delivered at level 2. Taking the mean of the non-zero values recorded in OpenLMIS gives 0.857 and so for the ROI paper, we override consumables availability for xpert only to this. This is done in tb module, initialise_simulation.

We could introduce a stepped availability over time reflecting increased usage, or change diagnostic referrals to use more xpert over time. Importantly, by 2024, the availability and usage should be correct for the projections.