Fig 1.
Overview of the structure of the codebase of the PopART-IBM model.
A) a metapopulation structure for modelling the PopART community (inside patch) and a surrounding area (outside patch); B) within each patch the codebase is partitioned into several processes related to demographics, sexual partnerships, HIV transmission and disease progression, and interventions; C) the nature of the model being individual-based allows detailed output from the model including the complete HIV transmission network.
Table 1.
Key partnership-related parameters used in PopART-IBM.
Fig 2.
Schematic of HIV progression in the model in the absence of ART.
Table 2.
Parameters estimated in the calibration framework.
Uniform priors were used for all parameters with ranges shown in the table.
Fig 3.
Runtime (left) and memory usage (right) of PopART-IBM for a single randomly chosen calibrated parameter set run for 120 years. Population size and PLHIV were measured in one patch at the beginning of 2020. The model is writing a typical set of files to disk in this experiment, including a time series of HIV prevalence, incidence, ART usage, etc. Shorter runtimes can be enabled during calibration by only writing essential indicators used within the calibration process.
Fig 4.
Calibration of PopART-IBM to PC, DHS and CHiPs datasets for women (columns 1 and 3) and men (columns 2 and 4).
For all graphs, the x-axis shows 5 year age groups, starting at 15–19 years. Panel shows calibration to HIV prevalence in PC0-PC36 (columns 1–2); HIV prevalence from 2002, 2007 and 2013 Zambia DHS surveys (top 3 rows, columns 3–4); HIV prevalence from CHiPs round 3 (bottom row, columns 3–4). In each graph the coloured bars represent the data used for calibration (observed proportions in each subgroup, with 95% confidence intervals), with corresponding colour in the Gantt chart in Fig 5; the grey lines represent the outputs from the 1,000 calibrated model runs; and the crosses represent the best model fit output.
Fig 5.
Calibration of PopART-IBM to PC and CHiPs datasets for women (columns 1 and 3) and men (columns 2 and 4). For all graphs, the x-axis shows 5 year age groups, starting at 15–19 years.
Panel shows percentage of PLHIV who are aware of status from CHiPs rounds 1–3 (columns 1–2), percentage of PLHIV aware of status who are on ART (top 3 rows, columns 3–4); viral suppression amongst all PLHIV from PC24 (bottom row, columns 3–4); and a Gantt chart showing the times that different data sources were collected (bottom left). In each graph the coloured bars represent the data used for calibration (observed proportions in each subgroup, with 95% confidence intervals), with corresponding colour in the Gantt chart; the grey lines represent the outputs from the 1,000 calibrated model runs; and the crosses represent the best model fit output.
Fig 6.
Projected outputs over time from 1,000 calibrated runs from PopART-IBM for a counterfactual scenario with PopART-related intervention HIV testing, linkage to care and VMMC switched off.
Left column shows normalized or percentage outputs, such as HIV incidence and prevalence, while right column shows absolute outputs, such as total number of PLHIV. Black line shows median and grey lines show 2.5% and 97.5% quantiles from model output. Clockwise from top left: HIV prevalence over time; total number of people living with HIV/AIDS over time (thousands); total people living with HIV/AIDS who are on ART over time (thousands); incident cases per year; HIV incidence rate; percentage of PLHIV who are on ART. py = person years.
Fig 7.
Transmission tree from a single calibrated run of the PopART-IBM over the period 1970–1980.
Nodes (circles) represent HIV infected individuals, and are coloured by sex. Edges (straight lines) represent transmission events, with infectors on the left and infectees on the right. Thicker lines represent where infections occurred when the infector was in the early stages of HIV infection.