Impact of tclose on mode_appt_constraints = 2 #999

marghe-molaro · 2023-06-09T08:40:07Z

The variable “tclose” determines how long an hsi can be postponed after topen before a patient stops seeking care. Modules can either specify tclose independently, or rely on a “default” tclose specified by the healthsystem module, currently set to 7 days.

In mode_appt_constraints = 0 & 1, tclose has no impact on health outcomes, as care is never delayed.

In mode_appt_constraints = 2, where care is delayed if required medical officers have exhausted capabilities for the day, this assumption plays an important role in the simulation outcome. The probability of receiving care will in fact depend on the combination of healthsystem capacity, priority of individual hsis, the “patience” of patients waiting to receive care (i.e. tclose), and timeline of individual diseases resolution.

Notes on this issue:

Impact on simulation

Performance

Duration of simulations for default tclose = 2days, 7 days, 3 months in mode_appt_constraints = 2 is of 16hr, 1d 3hr, > 6d 4hr respectively. This (unsurprisingly) highlights that one of the biggest determinants of performance in mode_appt_constraints = 2 is length of queue. Margherita and Tim are working to address this.

Outcome

Below we show the total number of appointment ran, total number of appointments never ran, total number of appts requested (defined as sum the first two), and fraction of appointment delivered (count of appts ran over total number of appointments requested) for mode 1, mode=2 with default tclose 7 days, and mode=2 with default tclose 2 days, for a 5 year run/100,000 people. The priority policy assumed is Random (i.e. all hsis have equal priority).
Important note: this is based on a single run, so the following discussion should be tested with higher statistics.
From the plots, it appears that a decrease in the default tclose leads to an increase, in most cases, in ran “specialist” appointments. The overall number of hsis delivered, however, decreases if tclose=2days, as expected (see breakdown below). This could be explained as follows: a shorter tclose leads to a higher “expiration” of queued appointments for first attendance emergency and non-emergency (see breakdown below), and other high-volume appt types such as contraception and malaria testing; these hsis have a huge volume of requests (notice log scale) compared to many of the other specialist hsis, and therefore represent a significant bottleneck towards specialist appointments. By decreasing the bottleneck effect, capabilities can therefore be freed-up to to carry out specialist appointments.
We notice that several types of hsis don't appear at all in the "never_ran" plots; we notice that at least some of them (e.g. RTIs) specify a tclose longer than the default of 7 days (15days in the case of RTIs); this does suggest that tclose plays indeed an important role in determining which hsis ultimately will get ran, as a result not of their priority in the queue, but sheer patience on the healthseeking part. Unless motivated by actual data, this therefore represents a bias in favour of certain treatments in the current model.

This suggests that tclose plays an important role in determining which hsis are delivered and which aren’t under mode_appt_constraints = 2.

Calibration to data

Tara highlighted following reference (https://doi.org/10.1186/s13561-020-00271-2) about “patience” of patients in health-seeking in Malawi, and summarised main conclusions as follows:
The median time from first seeking care to receiving care was 59 days (interquartile range 26-108 days).
The time to care varied by location, with patients in rural areas waiting longer than those in urban areas.
The time to care varied by the type of care needed, with patients seeking chronic care waiting longer than those seeking acute care.
Mhango, I., Msiska, S., & Mkandawire, M. (2020). Determinants of healthcare seeking and out-of-pocket expenditures in a “free” healthcare system: evidence from rural Malawi. Health Economics Review, 20(1), 1-18. https://doi.org/10.1186/s13561-020-00271-2

This work suggests that existing assumptions about tclose - both in individual modules and the default healthsystem one - should be revisited to better reflect available data. Assumption of a default tclose that can be applied to all hsis equally may also need to be revisited, as behaviour seems to be specific to type of treatment sought.