The time series trends based on yearly data on malaria incidence per 1000 population for all the 5 regions in the Lake Zone are presented in Figs. 3–6. The two scenarios outlined above are first investigated through visual inspection. However, the time series plots represent several intervention statuses (color-coded) for each district from one year to the next. These scenarios are a result of intervention status change from one year to the next across different districts. As indicated in the legend and of interest to this analysis, the observed intervention status with variation across the districts are the deployment of a) IRS with LLINs, and b) withdrawing IRS and maintaining only LLINs (IRS w/LLINs m). It is important to note that other changes in the intervention particularly an introduction of Piperonyl Butoxide (PBO) nets in area with or without standard LLIN or IRS is indicated using the legend as well but the analysis for such scenarios are not included as more data is required, thus, it is out of scope and the focus is on IRS withdrawal. In all the Figs. 3–6, the primary axis represents yearly incidence per 1000 population and secondary axis the percentage population protected by IRS. The IRS spray month for each year is presented by a numeric value (i.e., 1 = Jan … 12 = Dec).
The impact of IRS withdrawal based on yearly data
As indicated in Fig. 3 for Mwanza region, the impact of withdrawing IRS while maintaining LLINs after three rounds (Buchosa DC) or two rounds (Kwimba DC and Sengerema DC) or one round (Misungwi) is observed by an increase in malaria incidence per 1000 population.
As indicated in Fig. 4 for Geita region, in Nyang’hwale DC the impact of IRS is observed in 2017 where IRS was implemented in February followed by increase in maria incidence perhaps due to late spray in 2018 followed by IRS withdrawal - though a slight decrease is observed in 2020 but not below LLINs levels.
As indicated in Fig. 5 for Kagera region, the impact of withdrawing IRS after one round (Biharamulo DC and Muleba DC), three rounds (Bukoba DC and Missenyi DC) and after four rounds (Ngara DC) is clearly observed by the increase of malaria incidence.
As indicated in Fig. 6 for Mara Regions, the impact of withdrawing IRS and maintaining LLINs after one round in Rorya DC (2016) and three rounds in Butiama DC (2018) and Musoma DC (2018) is observed by an increase of malaria incidence over time following the IRS withdrawal.
Overall, key observations are noted from the yearly time series plots (Fig. 2 to 6) the data presented indicates that a) withdrawing IRS after one, two, three, or even four rounds result into resurgence of malaria incidence as indicated in Mara (Rorya DC, Butiama DC, Musoma DC). The districts where IRS was implemented followed by withdrawal is investigated further using monthly time series plots.
Monthly – Time series Plots
The monthly time series plots are also used to provide visual inspection of the impact of IRS withdrawal. Although the visual inspections maybe used to show the difference between shifting of interventions in different districts, it is important to investigate whether the difference observed is statistically significant or not. Therefore, the statistical analysis is performed based on monthly data to provide more observations suitable for running a robust statistical model. Therefore, monthly time series trends are presented under each accompanied by statistical model results. The data for 2021 from January to August is also included in the monthly based analysis.
Implementation of IRS combined with standard LLIN followed by IRS withdrawal; to show the effect of IRS withdrawal.
Figure 7 presents malaria monthly incidence data per 1000 population before and after IRS is withdrawn for different IRS rounds. The impact of withdrawing IRS (vertical line) after one (A), two (B), and three (C) round(s) is clearly observed in Kagera - Muleba DC, Mwanza - Kwimba DC, and Mara - Butiama DC respectively – based on the observed resurgence of malaria incidence.
The monthly time series plot presented in Fig. 7 provides a clear picture of the impact of withdrawing IRS. However, the interrupted time series statistical model was deployed to assess the difference during IRS implementation and after withdrawal. The model results indicate that the difference between the trend during IRS and after withdrawal is negative and statistically significant (p < 0.001) - indicating that monthly malaria incidence decreases over time during IRS. For each month that passes, the malaria incidence decreases at a log mean of -0.30 (-0.34, -0.27), 95% Confidence Interval (CI), -0.08 (-0.09, -0.07), 95% CI, and − 0.02 (-0.03, -0.02), 95% CI on the index in Muleba, Kwimba, and Butiama DC respectively.
The immediate effect after the IRS is withdrawn is positive and statistically significant (p < 0.001) - indicating that withdrawing IRS increased the malaria incidence at log of 1.38 (1.20, 1.58), 95% CI, 0.41 (0.07, 0.74), 95% CI, and 0.51 (0.36, 0.66), 95% CI in Muleba, Kwimba, and Butiama DC respectively.
The sustained effect after IRS is withdrawn is positive and statistically significant (p < 0.001) - indicating that each day that passes after IRS is withdrawn, the malaria incidence increases at a log mean of 0.32 (0.29, 0.36), 95% CI, 0.13 (0.08, 0.17), 95% CI, 0.04 (0.03, 0.06), 95% CI and points on the index in Muleba, Kwimba, and Butiama DC respectively. Additional figures with almost similar trends exhibiting the same scenario are provided in Supplementary file 2.