Extended Data Fig. 7: Multivariate uncertainty analysis for the malaria transmission model.

The true model uncertainty was quantified by calculating the additional clinical malaria cases, malaria deaths, and years of life lost due to malaria, using an additional 20 draws from the joint posterior distribution of the fitted model parameters. These simulations were performed for all 37 administrative 1 units in Nigeria, and 40 other units across four countries—Zambia (all provinces included), Mozambique, Democratic Republic of the Congo and Burkina Faso—and for each COVID-19 and malaria scenario. We used the outcomes for the Nigeria administrative units to calculate the coefficient of variation (CoV) and tested the application of a Normal approximation to compare the uncertainty intervals (UI) for the other countries. a, Shows the 95% UI for each of the four countries estimated from the different model runs (pink and purple error bars) with values estimated from the Normal approximation and fitted CoV values (red bars). Results indicate that the uncertainty generated using both methods was broadly similar. b, Illustration of how malaria parameter uncertainty influences estimates of the additional weekly deaths due to malaria in Nigeria over the year May 2020–April 2021 for each of the four COVID-19 scenarios. Two different levels of malaria service interruption are considered for each scenario, the first where LLINs and SMC are ceased and case management is reduced by 50% (pink line, Supplementary Table 1 row 1), and the second when only case management is reduced by 50% (purple line, Supplementary Table 1, row 3). The solid dark lines represent best guess model predictions for the additional malaria deaths (difference between the levels of malaria service interruption and no COVID-19 induced disruption) whilst the shaded regions represent the 95% UIs generated by varying the input parameters within plausible ranges.