Department of Computer Science and Engineering, University of Notre Dame, Notre Dame, IN 46556, USA.
Malar J. 2013 Aug 21;12:290. doi: 10.1186/1475-2875-12-290.
Agent-based models (ABMs) have been used to estimate the effects of malaria-control interventions. Early studies have shown the efficacy of larval source management (LSM) and insecticide-treated nets (ITNs) as vector-control interventions, applied both in isolation and in combination. However, the robustness of results can be affected by several important modelling assumptions, including the type of boundary used for landscapes, and the number of replicated simulation runs reported in results. Selection of the ITN coverage definition may also affect the predictive findings. Hence, by replication, independent verification of prior findings of published models bears special importance.
A spatially-explicit entomological ABM of Anopheles gambiae is used to simulate the resource-seeking process of mosquitoes in grid-based landscapes. To explore LSM and replicate results of an earlier LSM study, the original landscapes and scenarios are replicated by using a landscape generator tool, and 1,800 replicated simulations are run using absorbing and non-absorbing boundaries. To explore ITNs and evaluate the relative impacts of the different ITN coverage schemes, the settings of an earlier ITN study are replicated, the coverage schemes are defined and simulated, and 9,000 replicated simulations for three ITN parameters (coverage, repellence and mortality) are run. To evaluate LSM and ITNs in combination, landscapes with varying densities of houses and human populations are generated, and 12,000 simulations are run.
General agreement with an earlier LSM study is observed when an absorbing boundary is used. However, using a non-absorbing boundary produces significantly different results, which may be attributed to the unrealistic killing effect of an absorbing boundary. Abundance cannot be completely suppressed by removing aquatic habitats within 300 m of houses. Also, with density-dependent oviposition, removal of insufficient number of aquatic habitats may prove counter-productive. The importance of performing large number of simulation runs is also demonstrated. For ITNs, the choice of coverage scheme has important implications, and too high repellence yields detrimental effects. When LSM and ITNs are applied in combination, ITNs' mortality can play more important roles with higher densities of houses. With partial mortality, increasing ITN coverage is more effective than increasing LSM coverage, and integrating both interventions yields more synergy as the densities of houses increase.
Using a non-absorbing boundary and reporting average results from sufficiently large number of simulation runs are strongly recommended for malaria ABMs. Several guidelines (code and data sharing, relevant documentation, and standardized models) for future modellers are also recommended.
基于主体的模型（ABMs）已被用于估计疟疾控制干预措施的效果。早期的研究表明，幼虫源管理（LSM）和经杀虫剂处理的蚊帐（ITNs）作为控制媒介的干预措施是有效的，这些措施既可以单独使用，也可以联合使用。然而，结果的稳健性可能受到几个重要建模假设的影响，包括用于景观的边界类型，以及报告的重复模拟运行次数。ITN 覆盖范围定义的选择也可能影响预测结果。因此，通过复制，对已发表模型的先前发现进行独立验证具有特殊意义。
使用一种基于空间的冈比亚按蚊的昆虫学 ABM 来模拟蚊子在基于网格的景观中的觅食过程。为了探索 LSM 并复制早期 LSM 研究的结果，使用景观生成器工具复制原始景观和场景，并使用吸收和非吸收边界运行 1800 次重复模拟。为了探索 ITN 并评估不同 ITN 覆盖方案的相对影响，复制了早期 ITN 研究的设置，定义和模拟了覆盖方案，并为三个 ITN 参数（覆盖、驱避和死亡率）运行了 9000 次重复模拟。为了评估 LSM 和 ITN 的组合效果，生成了房屋和人口密度不同的景观，并运行了 12000 次模拟。
当使用吸收边界时，与早期的 LSM 研究结果一致。然而，使用非吸收边界会产生显著不同的结果，这可能归因于吸收边界不现实的杀灭效果。在 300 米范围内的房屋内去除水生栖息地并不能完全抑制丰度。此外，由于产卵与密度有关，去除不足数量的水生栖息地可能适得其反。还证明了进行大量模拟运行的重要性。对于 ITN，覆盖方案的选择具有重要意义，过高的驱避率会产生不利影响。当 LSM 和 ITN 联合使用时，随着房屋密度的增加，ITN 的死亡率可以发挥更重要的作用。在部分死亡率的情况下，增加 ITN 覆盖范围比增加 LSM 覆盖范围更有效，随着房屋密度的增加，整合这两种干预措施会产生更多的协同作用。
强烈建议在疟疾 ABM 中使用非吸收边界并报告来自足够大量模拟运行的平均结果。还为未来的建模者推荐了一些指南（代码和数据共享、相关文档和标准化模型）。
