Accounting for additional causes of absence

[wiktafesse]

Healthcare workers face different incentives which may impact their time at work and productive (patient facing) time at work. Effective time at work may not be adequately measured by asking a group of representative healthcare workers or by consulting official leave documentation and may be better measured using unannounced facility visits to observe health worker activity. Therefore, we aim to convert “potential supply” to “effective supply” of healthcare worker time - PFT, in the representation of the healthcare system using data on absences from the HHFA.

To do:

• Wiktoria to consult CHAI about methods to generate data on PFT.
• Wiktoria and Martin to generate an alternative approximation of PFT accounting for absences in the HHFA.
• Wiktoria, with support from @BinglingICL to incorporate this into the model

[BinglingICL]

Thanks for raising this issue @wiktafesse. Very happy to have any further discussion and to help if I could : )

[BinglingICL]

Thanks @wiktafesse. The way you raised the issue looks perfect to me. Once solutions found, we will then raise a pull request to solve this issue : )

[BinglingICL]

Hi @Tim Hallett, Wiktoria @wiktafesse and I just had a discussion on her analysis on HHFA patient facing time data considering absence. Considering that assumptions and collection methods of that HHFA data are very different to CHAI data, we have been thinking to create another scenario of HCW capabilities using HHFA patient facing time (and CHAI HCW allocation) in order to study impact of different patient facing time on health outcome (e.g. health loss due to absence) using TLO model.

However, the HHFA only has available data for a subset of cadres incl. Medical, Nursing and Midwifery and DCSA staff. We wonder if our TLO model can run with such a limited scenario with capabilities from only these cadres?

Otherwise, as we have just discussed, we might have two further ways to approach Wiktoria's research objectives:

(1) within HHFA data, to estimate patient facing time for a full set of cadres (that is requested by TLO model) by making assumptions of absence probability and relevant using limited cadres' data, and then to create the new scenario of HCW capabilities (e.g. let us say "HHFA scenario") to be run by TLO model (@wiktafesse, with this approach, I wonder if we will be able to tell the different health impacts of different HCW capabilities are due to absence, or generally due to different patient facing time estimated? Also, we could have two sub HHFA scenarios, one with absence and the other without absence, to be compared.)

(2) use HHFA data to estimate the (additional) absence probability for all cadres at all facility levels and transfer the probabilities into a parameter (set) in TLO model, which could be reasonably applied to our "current/actual" (and/or "funded", "funded_plus") HCW capabilities.

Would be grateful if you have any quick thought. Many thanks. @tbhallett

And Wiktoria @wiktafesse, please feel free to comment if I am not clear or correct and raise any question here or via slack.

[wiktafesse]

Thank you very much @BinglingICL for raising this question with @tbhallett ! Please see below for an overview of progress so far.

Wiktoria and @mjchalkley have created a template spreadsheet estimating PFT based on HCW absence and current activities in the HHFA using the same framework in the original TLO model PFT resource sheet created by @BinglingICL and the TLO team.

Meeting with @BinglingICL 13 April.

• We confirmed the aim of the proposed analysis which is to estimate health loss/gains comparing possible daily capabilities using CHAI data compared to accounting for observed absence and non-patient facing time using the HHFA.

• We discussed the methodology to calculate a new alternative PFT resource file based on the HHFA data which can then be used to generate a 4th scenario of daily (current) capabilities.

• We established that the data limitations and assumptions need to be explained and discussed in a document.

Next steps

• Investigate how to account for/impute missing HHFA data on a number of cadres with help from @tbhallett and rest of the TLO team.

• Wiktoria will generate an amended PFT resource file to be used by @BinglingICL when running the alternative scenario to calculate total daily capabilities based on the HHFA. The file will be stored in the data repo.

• Wiktoria will write a word document which explains how to generate the data used in the alternative PFT resource file, assumptions and limitations following the structure in existing TLO model documentation.

• Wiktoria will ask @timcolbourn about how being on call and doing outreach activities should be translated into PFT.

@BinglingICL feel free to comment/edit anything.

[tbhallett]

Hi @wiktafesse

I'd be happy to come on the next call with you, Martin and Bingling to discuss the approach and see if we can come up with ways to cope with the missing cadres in the HHFA data.

We could either:

• Impute to come up with the times for all the cadres; or,
• Redefine the officers that are needed for each APPOINTMENT_TYPE so that we only need a few number of cadres.

I think the first of these is preferable.

I think what we should do will depend on which cadres are present/missing. But, would one (very simple) approach be to just use the CHAI data for the "missing" HCWs....?

[tbhallett]

Hi @wiktafesse

I’m picking up the discussions we’ve about the estimation of ‘absence’ of healthcare workers (this issue).

Basically, I think we had reached the point where we realised that the assumptions used in the model for the numbers of HCW and their time spent at work had some differences in definitions with the data that you have been using to estimate absences of HCW (i.e., the HHFA). I believe the discussions with @marghe-molaro reached the conclusions that whilst these discrepancies were small, they did result in estimates that we could not use (as the headline ended up being that for some cadres/levels, “fewer than 0% of HCW were estimated as being absent”).

I think that to achieve our goal of having empirically-based estimates being used in the model, it would be perfectly reasonable to use the HHFA to estimate the proportion of time that is, ostensibly, lost from patient-facing time of HCW for reason of an (apparently) avoidable absence. (This is a clumsy definition and I’m sure you’ll have a better one, but I’m using it as a placeholder).

This would furnish us with estimates for each cadre and level of x_cadre_level, where x is the proportion of time “lost to absence” (per your definition).

We would then combine this with the model estimate of the total patient-facing time for each cadre and level that is contracted (should be available, not allowing for such absence), Y_cadre_level, to give an effective amount of patient_facing_time, accounting for absence of Y*(1-x).

Does this sound ok to you? If so, I would be very grateful if you could share such estimates...?

This is the ResouceFile to edit. It looks like the below (note that the issue referred to above is that some of these factors are greater than 1.0).

Many thanks Tim