The final two days of the workshop were largely dedicated to allowing the participants to explore the modelled maps to derive programmatically-pertinent recommendations that could plausibly be applied in the context of their current positions. Outcomes of these group discussions are summarized here, illustrating different ways in which The DHS Program modelled surfaces may be easily and rapidly applied by NMCPs and fellow stakeholders. While more formal analyses were encouraged, these were outside the timeframe of the workshop.
Example 1
Strengthening access to intermittent preventative treatment in pregnancy.
Intermittent preventative treatment in pregnancy (IPTp) with sulfadoxine–pyrimethamine (SP) was brought into policy in Madagascar in 2004, and a progressive scale-up has led to this now being implemented in all control-phase districts. The WHO recommendation to increase the minimum number of doses from two (IPTp2+) to three (IPTp3+) was initiated in Madagascar in 2015. Given this, the IPTp3+ coverage indicator was not appropriate to examine given the temporality of its definition (having as denominator the “total number of women surveyed who delivered a live baby in the 2 years preceding the survey” [23], i.e. extending to before the IPTp3+ policy was locally implemented). Instead, IPTp2+ was evaluated. District-level aggregation of the modelled surface indicated that coverage in 2016 remained below 20% in 40% of target districts (Fig. 1a). This was consistent with relatively weak coverage of antenatal consultations with an estimated one third of pregnant women never attending an antenatal consultation (source: MoH, 2017), and health facilities regularly notifying SP stock-outs (43% did during 2016; source: NMCP, 2017).
The modelled surfaces made apparent a degree of spatial heterogeneity in the coverage of IPTp2+ (Fig. 1a), with high predictive uncertainty (> 20%) also widespread (Fig. 1b) likely associated with the relatively small sample sizes inherent to this indicator (N = 2786, relative to 10,816 respondents for other indicators in 2016).
Recommendations were made for increased sampling effort in high uncertainty coastal districts during future MIS, as well as strengthening the reporting of IPTp during antenatal consultations at health facilities. Reinforced collaboration between the NMCP and the National Family Health programme was also recommended, alongside renewed awareness campaigns targeted to areas of weakest coverage (Fig. 1a).
Example 2
Spatio-temporal trends in access to insecticide-treated nets and implications for future mass distribution campaigns.
Madagascar aims for universal coverage of insecticide-treated nets (ITN) across all control-phase districts. This is achieved primarily through mass distribution campaigns, the last having been in 2012–2013, 2015 and 2018 [18, 24]. The current objective is that at least 90% of households in the target districts should have at least one ITN per two residents. Several channels of continuous distribution supplement the mass campaigns, including antenatal consultations, community-health worker distributions, and subsidized sales in peri-urban communities. The MIS in 2013 and 2016 therefore assessed the overall impact of these activities, allowing changes in coverage during that time period to be quantified.
Interpretation of the survey outcomes must account for timings relative to mass distribution campaigns. The 2013 MIS took place part way through a distribution campaign, with 31 districts covered in the 6 months before the MIS, and 61 districts after the MIS. In contrast, all target districts were included in the mass ITN distribution during the 6 months preceding the 2016 MIS.
Several ITN coverage indicators based on different denominators (households vs. residents) were included in the spatial analyses, each representing different aspects of the programme’s impact (Table 1). Here, participants selected three of these indicators to assess ITN coverage across Madagascar in 2016 (Fig. 2a–c) and relative changes in these indicator levels since 2013 (Fig. 2d–f), with the aim of exploring what lessons could be derived from the modelled surfaces for future distribution campaigns. These included the spatial reach of the distribution campaigns using the indicator of presence of any ITNs in the household (Fig. 2a, d), the adequacy of coverage when accounting for the number of household residents (the target is for one net per two individuals; Fig. 2b, e), and finally usage of available nets (Fig. 2c, f). The mapped surfaces were aggregated to district units to reflect the level at which decision-making and logistics are coordinated during ITN campaigns.
The modelled surfaces revealed that while the reach of the ITNs was quite high, with 52 of 92 target districts predicted to have > 90% of households with at least one bed net, this coverage dropped dramatically when considering the adequacy of ITN availability according to the objective of one ITN per two people by household. No district met the national objective of 90% in 2016, though 39 (42%) had levels > 75%. Nevertheless, reported indicators of usage by household residents indicated that ITNs were being used at rates suggesting that even where ITN numbers were insufficient within households, residents were sleeping under the nets which were available. Coverage of these ITN indicators was generally better in coastal areas where transmission was higher [6], particularly along the northern districts of the west and east coasts. Coverage dropped into the highland areas where transmission was less intense. The maps suggested generally sustained or positive changes in coverage between 2013 and 2016, with greatest improvements in southern districts even though coverage remained among the lowest nationally in these same districts. In the east coast districts, where transmission is highest and ITN coverage still falls short of national targets, there was little reported change from 2013. However, there were extensive areas of high uncertainty in the model predictions, notably in the maps of ITN accessibility and usage where the map estimates need to be interpreted with caution (Fig. 2b, c, e, f). Spatial heterogeneity in the cluster-level results may explain uncertainty in these areas.
Recommendations for future campaigns that emerged from these modelled maps were to focus on increasing the numbers of nets distributed, with reinforced efforts particularly in east coast areas where coverage was low despite relatively high transmission. Results on usage were encouraging but still inadequate, indicating that further behaviour communication interventions would be important alongside the distributions, as per the NMCP’s guidelines.
Example 3
Treatment-seeking for febrile infants.
Seeking treatment during a febrile episode is the critical first step towards effective case management and reductions in malaria morbidity, as well as towards ensuring reliable reporting of malaria episodes for surveillance. Low treatment-seeking rates across much of Africa are a main reason for the WHO using data sources independent of routine surveillance in their estimations of clinical case burdens [15, 25]. The MIS indicator quantifying this is rates of treatment seeking by mothers for any children younger than 5 years having suffered a febrile episode in the 2 weeks preceding the survey.
The national-level MIS results from Madagascar suggest an increased, but nevertheless low, treatment-seeking rate from any type of health provider for febrile children from 38% in 2013 to 46% in 2016. Only 29% and 36%, respectively, sought treatment from the public health facilities likely to provide appropriate free case management and to report monthly case estimates to the centralized MoH database. These low rates of contact with recommended healthcare providers are a target of Madagascar’s current National Strategic Plan through behaviour-change communication activities. Better insight into this indicator’s tendencies would help focus future efforts based on current gaps and local infection risk levels.
The MIS treatment-seeking indicator was therefore evaluated to see what spatio-temporal trends could be derived beyond the national summary figures. Descriptive statistics are presented for 2013 (Fig. 3a–c) and 2016 (Fig. 3d–f). The cluster-level raw numbers showed a high level of spatial heterogeneity (Fig. 3a, d), likely associated with the variable and sometimes small samples sizes (overall n2013 = 633 and n2016 = 1096 eligible mothers across the country whose children had suffered a febrile episode in the 2 weeks preceding the survey, which become very small when considered at the cluster-level). The raw datapoint values (Fig. 3a, d) and the variograms (Fig. 3b, e) revealed that the dataset had limited spatial structure, also reflected by high relative uncertainty in the predicted maps. Consistent with these characteristics, the model predictions revealed low correlation with the raw cluster-level observed data (Fig. 3c, f). These warnings in the data suggest that the spatial predictions from this current dataset and model may not be appropriate to rely on.
These insights indicate to stakeholders interested in improving rates of appropriate malaria case management that the data and models examined here are insufficient to allow meaningful assessments of current treatment-seeking levels. Additional efforts will be necessary to strengthen the evidence base and allow this indicator’s subnational trends to be understood. The low sample sizes associated with this specific MIS indicator—owing to its opportunistic nature—limit the statistical power required for high-resolution analyses.
Alternative approaches, such as active case detection, or simply larger sample sizes could provide more robust insight into this important indicator. A strong message, therefore, emerges to advocate for reinforcement of this indicator, to allow important questions about variability in behaviour across the island and the impact of NMCP initiatives on behaviour change over time to be answered. This remains essential to improving treatment-seeking rates and appropriate case management: cornerstones of any control programme.