Fig 1.
Households in the Lake Kariba region of Southern Province, Zambia, are clustered into village-scale simulation constructs within twelve health facility catchment areas.
Adapted from [9].
Fig 2.
Spatial distribution of two vector species governs magnitude and seasonality of malaria transmission over the Lake Kariba region.
(A) Adult vector numbers are tuned by scaling larval habitat availability of An. arabiensis and An. funestus vectors. Relative scales of the two vector species govern the seasonality of human biting. Shown: arabiensis scale = 100, funestus scale = 30. (B) Best fit arabiensis and funestus larval habitat availabilities vary spatially over the study region, with more habitat available for both species in the lower-altitude regions closer to the lake front.
Fig 3.
Representative calibration of larval habitat availabilities in a single cluster located in Munyumbwe HFCA.
(A) Sampling over larval habitat availability scale factors for arabiensis and funestus identifies pairs of scale factors with good fits to: cluster-level prevalence by rapid diagnostic test (RDT) from surveillance data (blue, best fit to prevalence), HFCA-level seasonality of weekly reported clinical cases counts (yellow, best fit to clinical cases), and both prevalence and seasonality of clinical cases (red, best combined fit). Surface shows the residual error in the combined fit to both prevalence and clinical cases. (B) RDT prevalence between December 2011 and April 2014 observed during surveillance (black) and in simulation. Blue: Simulation of single pair of arabiensis and funestus habitat availability scales that result in the best fit to the surveilled prevalence without accounting for fit to seasonality of clinical case counts. Yellow: Simulation of single pair of habitat scales that result in the best fit to seasonality of clinical case counts without accounting for fit to prevalence. Red: Simulation of pairs of habitat scales that result in best (bold line) and top 10 (thin lines) fits when optimizing fit to both surveilled prevalence and seasonality of reported clinical case counts. (C) Normalized seasonality of clinical case counts observed in Munyumbwe health clinic reported data (black) and simulation of a single Munyumbwe cluster (blue, yellow, red, defined as in panel B).
Fig 4.
Simulations capture spatio-temporal variation in cluster-level prevalence of RDT-positive infections.
Top: survey data collected during MTAT rounds. Bottom: mean cluster prevalence in 100 samples from the joint posterior distribution of 10 best-fit habitat availability pairs for each cluster.
Fig 5.
Simulations capture the seasonality of HFCA-level weekly clinical case counts.
Simulated clinical and severe malaria cases are scaled by sampling treatment-seeking rates of clusters whose distance to their health facilities are within 2km of the target cluster’s distance to its health facility. Simulated case counts were also scaled by 150% in all HFCAs to account for underestimating catchment populations due to incomplete coverage during MTAT. Mean (line) and range (shaded area) of 100 samples from the joint posterior distribution of 10 best-fit habitat availability pairs for each cluster.
Fig 6.
Prevalence of RDT+ infections in the Lake Kariba region a decade after target elimination date of 2020 under various post-2015 intervention scenarios.
Clusters are divided into high-burden (n = 64) and low-burden (n = 51) groups as indicated in the inset map. Cluster prevalence in each scenario is the mean of 100 samples from the joint posterior distribution of 10 best-fit habitat availability pairs for each cluster. Post-2015 MDA: MDA is distributed in 2014 and 2015 to all HFCAs in all scenarios. Scenarios 1–4: MDAs discontinued after 2015. Scenarios 5–10, 12–13: MDAs continue annually between 2016 and 2020, a total of 5 additional distributions. Scenario 11: MDA is distributed only in 2017 and 2018. Scenarios 10–11, indicated with asterisk: post-2015 MDAs have 70% coverage. All other MDA distributions have coverage as indicated in S5 Fig. MDA region: MDAs are distributed to clusters in all HFCAs except in scenarios 8, 9, and 11, where only clusters in high-burden HFCAs receive MDAs after 2015. Case management: In scenario 1, case management is maintained at rates observed during 2012–13 surveillance. In all other scenarios, case management rates increase and plateau as shown in S4 Fig. ITN usage: In scenarios 1, 2, and 5 (“current”), ITNs are not distributed after 2015. In scenarios 3, 6, 8, and 10 (“ramp”), ITN usage after 2015 ramps up following historical rates. In all other scenarios (“aggr”), ITNs are distributed at an aggressive 80% coverage biannually between 2018–22. Importation: Infections imported from outside the 12-HFCA study area.
Fig 7.
Malaria elimination in the Lake Kariba region is possible under high levels of ITN usage even without distributing MDAs after 2015.
The fraction of total study area population living in clusters where no local transmission has occurred over a month-long period is plotted for each month between January 2012 and January 2030. Line indicates the mean and shaded area the range observed over 100 samples from the joint posterior distribution of 10 best-fit habitat availability pairs for each cluster. A simulation results in elimination if no new infections occur in all clusters over a 3-year period. The “elimination” row indicates the fraction of simulations where elimination was observed. (A) If ITN distributions stop after 2015, elimination is never observed to occur. (B) Under an aggressive ITN distribution scenario, elimination becomes likely even without additional MDAs after 2015.
Fig 8.
Administering five more years of MDA to all or a subset of HFCAs increases the likelihood of elimination under high levels of ITN usage but cannot achieve elimination on its own.
The fraction of total study area population living in clusters where no local transmission has occurred over a month-long period is plotted for each month between January 2012 and January 2030. Line indicates the mean and shaded area the range observed over 100 samples from the joint posterior distribution of 10 best-fit habitat availability pairs for each cluster. A simulation results in elimination if no new infections occur in all clusters over a 3-year period. The “elimination” row indicates the fraction of simulations where elimination was observed. (A) Five more years of MDA in all HFCAs increases the likelihood of elimination only when ITN usage is very high. Otherwise, elimination remains very unlikely. (B) Limiting the 2016–20 MDAs to high-burden areas yields similar results to simulations where MDA was distributed to all clusters.
Fig 9.
Higher MDA coverage cannot overcome insufficient vector control.
In scenarios 10 and 11, post-2015 MDAs have a higher coverage of 70%. The fraction of total study area population living in clusters where no local transmission has occurred over a month-long period is plotted for each month between January 2012 and January 2030. Line indicates the mean and shaded area the range observed over 100 samples from the joint posterior distribution of 10 best-fit habitat availability pairs for each cluster. A simulation results in elimination if no new infections occur in all clusters over a 3-year period. The “elimination” row indicates the fraction of simulations where elimination was observed.
Fig 10.
Human movement can disrupt the success of an elimination program.
Both increased movement within the 12 HFCAs as well as importation of infections into the region can drastically decrease the likelihood of elimination in the Lake Kariba region. The fraction of total study area population living in clusters where no local transmission has occurred over a month-long period is plotted for each month between January 2012 and January 2030. Line indicates the mean and shaded area the range observed over 100 samples from the joint posterior distribution of 10 best-fit habitat availability pairs for each cluster. A simulation results in elimination if no new infections occur in all clusters over a 3-year period. The “elimination” row indicates the fraction of simulations where elimination was observed.