FAQs.RmdThe Global Plan specified broad intervention scale-up projections that were, in most instances (not all) at the country level. As such, there is space for an increase in efficiency, resulting in a greater impact for the same or less spending, with optimal sub-national targeting of interventions. This is what the optimised solutions are in the Global Fund runs.
There are a number of reasons why this could occur, Firstly the optimisation is set up to minimise and equally weighted sum of cases and deaths. Therefore an increase in one output, may be outweighed by a decrease in the other. In this case the aggregate weighted sum of cases and deaths will have increased with a smaller budget.
Secondly, One of the conditions of the budget optimisation is that treatment must be the last intervention for which coverage is reduced when reducing the available budget. This is implemented by removing any lower-coverage treatment options when the budget is high enough. When the budget is reduced below this threshold the lower-coverage treatment options become available again. For low incidence settings, due to model stochastic, it is possible for a run with lower tx coverage to have lower case numbers than a run with higher treatment coverage.
post = continue, whilst not always for the scenario post = gp?This seems counter intuitive as we wouldn’t expect the post-replenishement period to have any impact on the replenishment period! The optimisation is run on the post = continue scenario. When we apply this optimal solution to the post = gp scenario it is applied to a different set of model runs. These vary stochastically from the post = continue runs. This stochastic variation can lead to the weighted sum of cases and deaths (y) not always decreasing monotonically with b. This problem will be most pronounced in low transmission countries where the stochastic variation is more influential.
b scenario not always <= b * global plan budget?For the same reason as above, stochastic differences in the treatment costs for the post = gp can lead to small fluctuations in the total cost.
Filtering a specific scenario can be performed by selecting the relevant scenario options from the pre, replenishment and post scenario columns. For budget optimisation outputs, select the relevant pre and post scenarios and the required budget proportion (e.g. pre = Follow_GP, replenishement = 0.75 and post = Revert_to_GP). For fixed scenarions, select the relevant pre, replenishment and post scenarios (e.g. pre = Follow_GP, replenishment = Revert_to_GP and post = Revert_to_GP). Please see the companion output dictionary vignette for more help.
In most cases yes, however intervention trends over multiple budget steps may not be linear or monotonic. As a result it is suggested to always interpolate between single budget steps (e.g. 0.6 to 0.65), not across multiple (e.g. 0.6 to 0.8). Optimised output is provided with steps of 0.05 to facilitate this.
In a number of cases the solutions of lower budget levels are not possible as the minimum cost solution for the country exceeds these lower budget bounds. As no solution can be found in these instances, the output is omitted.
As the budget decreases we must remove interventions to decrease costs accordingly. Treatment is prioritised in the optimisation and not reduced before all other interventions have been removed. This means we must remove preventative interventions, such as LLINs first. Reducing these interventions can lead to increased transmission and, for a fixed treatment coverage, an increase in treatment costs. This creates a threshold condition whereby either very large decreases in coverage must be made in a single step (to offset both the decrease in budget and associated rise in treatment cost), or none at all. The point at which this big jump from high to low (or no) coverage occurs corresponds to a big jump in cases and deaths. The level of b where this happens is transmission dependent.