Fig 1.
Schematic of revisions to the HIV Policy Model.
Shown is a simplified schematic of the HIV Policy Model. Each oval represents a main health state and the arrows represent transitions between health states. Dotted lines reflect new health states or transition probabilities, while colors indicate the implemented model revision. Red represents the Re-entry into Care revision, purple the Universal ART revision, and blue the Guideline-based ART revision. Death from AIDS- or non-AIDS-related causes is not shown.
Fig 2.
Example of unintentional errors detected by comparing model projections.
This figure is a snapshot of projections from all three model versions (Single Name, Single Cell, Matrix) for the Guideline-based ART revision. Each cell in this snapshot shows the number of people living with HIV projected to be in a given health state (Asymptomatic Early stage of disease, Not engaged in care) between 2010 and 2013, after standard debugging. The figure shows that model output for this health state is identical across the three model versions for 2010, 2011, and 2012, indicating no unintentional implementation errors. However, in 2013, the projected output for this health state differs across the three model versions (Single Name: 24,056, Single Cell: 30,025, Matrix: 24,056). In this example of an unintentional error, the principle error was due to an incorrect cell reference in the Single Cell model version, was propagated over an additional 10 health states, and took 10 minutes to identify and correct. Full model projections for the Guideline-based ART revision with unintentional errors are available in tabs Model (SN), Model (SC), and Model (Matrix) in S5 File.
Table 1.
Number, type, and rates of unintentional errors, by model version and revision.
Fig 3.
Distribution of the percentage difference in model output due to implementation errors, by model version and revision.
This figure shows the distribution of the size of unintentional errors for each cell in the spreadsheet model. The horizontal axis shows the percentage difference in model projections for models with unintentional errors compared to the revised, gold-standard model without unintentional errors. The vertical axis shows the percentage of spreadsheet cells. Panels are organized from least complex model revision to most complex (top to bottom). A percentage difference of 0% indicates no unintentional error. No unintentional errors occur in the Re-entry revision of the Single Cell–Re-entry model version. However, nearly half of cells in the more complex Universal revision of the Single Cell model version produced a >5% difference between model projections with versus without unintentional errors.
Table 2.
Frequency and percentage of error type.
Fig 4.
Impact of unintentional errors on model projections along the HIV care continuum.
This figure shows the impact of unintentional errors on policy-relevant model projections for a complex model revision (Guideline-based ART). On the horizontal axis are five key steps along the HIV care continuum, while on the vertical axis are the projected estimates of the number in each step along the care continuum by 2023. Projections made by the error-free model are shown in blue, while model revisions with unintentional errors are shown in different shades of red, by model version ( for Single Cell,
for Single Name, and
for Matrix).