Disentangling fine-scale effects of environment on malaria detection and infection to design risk-based disease surveillance systems in changing landscapes
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
Abstract
Landscape changes have complex effects on malaria transmission, disrupting social and ecological systems determining the spatial distribution of risk. Within Southeast Asia, forested landscapes are associated with both increased malaria transmission and reduced healthcare access. Here, we adapt an ecological modelling framework to identify how local environmental factors influence the spatial distributions of malaria infections, diagnostic sensitivity and detection probabilities in the Philippines. Using convenience sampling of health facility attendees and Bayesian latent process models, we demonstrate how risk-based surveillance incorporating forest data increases the probability of detecting malaria foci over three-fold and enables estimation of underlying distributions of malaria infections. We show the sensitivity of routine diagnostics varies spatially, with the decreased sensitivity in closed canopy forest areas limiting the utility of passive reporting to identify spatial patterns of transmission. By adjusting for diagnostic sensitivity and targeting spatial coverage of health systems, we develop a model approach for how to use landscape data within disease surveillance systems. Together, this illustrates the essential role of environmental data in designing risk-based surveillance to provide an operationally feasible and cost-effective method to characterise malaria transmission while accounting for imperfect detection.
Article activity feed
-
###Reviewer #3:
This study uses Bayesian inference to estimate the probability of detecting a malaria case and distribution of malaria cases using different surveillance methods in a district in Palawan, Philippines. The authors show that detection of malaria cases depends on household location and cannot be explained by distance to the health centre alone. They also argue that in low endemic settings it is economical to screen health care attendees stratified by their environmental risk (here, 100m proximity to closed canopy forest). The integration of unique high-quality spatial and molecular datasets is compelling. The authors argue that integrating remote sensing into triage for enhanced molecular detection of malaria could be economical in these settings.
Major comments:
The explanation of the modelling framework is, as written, …
###Reviewer #3:
This study uses Bayesian inference to estimate the probability of detecting a malaria case and distribution of malaria cases using different surveillance methods in a district in Palawan, Philippines. The authors show that detection of malaria cases depends on household location and cannot be explained by distance to the health centre alone. They also argue that in low endemic settings it is economical to screen health care attendees stratified by their environmental risk (here, 100m proximity to closed canopy forest). The integration of unique high-quality spatial and molecular datasets is compelling. The authors argue that integrating remote sensing into triage for enhanced molecular detection of malaria could be economical in these settings.
Major comments:
The explanation of the modelling framework is, as written, hard to follow and reproduce. Examples of where authors could improve clarity: the equations throughout use the same notation to mean very different things (si = patent infection (L380) or diagnostic sensitivity (L394)). The statement '𝑿𝑖𝜿 represents a vector of covariate effects' L383 does not make sense. Is X a specific location and 𝜿 the covariate estimate? It is difficult to understand how models were created and evaluated. The level of detail in the spatial data (Table S1) is insufficient for reproducibility, but could be easily amended to do so. Table 1- can authors list the actual range of these covariates before they are mean-centered and scale. Contextualizing the fixed effect estimates (i.e. distance to a closed canopy forest) is difficult to interpret given that no mean or sd of these distances are given (at least not that I could find).
Terminology changes throughout the manuscript, making things difficult to follow. For example, surveillance method 1 is referred to as passive case detection (Line 126), existing passive surveillance systems (Line 131), standard PCD (Line 137). Although one can assume these are all the same, it would help to use consistent terminology for this throughout. Convenience sampling is used throughout, but it's unclear if this is distinct from enhanced surveillance.
This is mentioned in the limitation section, but I don't think it gives a sufficient explanation. One benefit of the R-INLA framework is that it can account for spatio-temporal data - why was time of year and temporally relevant environmental characteristics not examined?
The authors don't provide convincing evidence that integrating remote sensing into this setting would actually add value. Could health care workers not ask residents if they live next to a big, closed forest? Wouldn't this achieve the same outcome? Wasn't it already known that frontier malaria was a problem here?
-
###Reviewer #2:
This is an interesting analysis and it is great to see a modelling analysis that has the potential to directly influence programmatic decisions. The idea of using remotely sensing data to stratify surveillance or diagnostic practices is interesting and scalable. The analyses are clearly described, and I found the use of the probability of detection metric particularly relevant to the types of decisions being made in pre-elimination settings. I have a few minor comments and would be curious if some discussion could be added to how this may be applicable to settings outside of SE Asia.
-
###Reviewer #1:
In 'Disentangling fine-scale effects of environment on malaria detection and infection to design risk-based surveillance' the authors analyze data from the Philippines to investigate the utility of landscape data to inform risk-based surveillance programs. The authors use occupancy modeling, a common approach in ecological studies, with health facility data (that combine both passive case detection via microscopy and RDTs with molecular approaches) to analyze the effectiveness of surveillance systems to detect malaria cases. Using cross-sectional surveys based at health facilities and the residence location of sampled individuals, the authors work to develop a method to detect locations with malaria infections. They find that in highly forested areas, there is a higher proportion of infections only detectable by …
###Reviewer #1:
In 'Disentangling fine-scale effects of environment on malaria detection and infection to design risk-based surveillance' the authors analyze data from the Philippines to investigate the utility of landscape data to inform risk-based surveillance programs. The authors use occupancy modeling, a common approach in ecological studies, with health facility data (that combine both passive case detection via microscopy and RDTs with molecular approaches) to analyze the effectiveness of surveillance systems to detect malaria cases. Using cross-sectional surveys based at health facilities and the residence location of sampled individuals, the authors work to develop a method to detect locations with malaria infections. They find that in highly forested areas, there is a higher proportion of infections only detectable by molecular methods.
In general, the authors provide a fine analysis. However, the novel aspects or new insights of this approach are unclear. The authors use a common standard statistical approach, although less common in epidemiology it is very common in ecology, to analyze fairly commonplace data. Their findings are in line with our existing knowledge of issues with enhanced (i.e. molecular) versus standard (RDT, PCR) and ability for ecological/landscape data to help improve surveillance systems. For example, it is not novel that enhanced surveillance would identify a wider spatial distribution than passive case detection since this method should identify more infections. Further, integrating landscape or geographic data to inform risk-prediction is commonly used for malaria or other vector-borne diseases that have an environmental component.
Major comments:
The authors do not provide adequate background on the setting, biases in the data used, and impact of health seeking behavior on their results. The authors find that the detection probability was negatively associated with travel time to the health facility. However, they do not elaborate upon whether this might be true or if health seeking biases from individuals who are from more forested areas and traveling to health clinics. In addition, the authors only analyze a single year of data which prevents any temporal trends to be analyzed or more robust analyses to be performed.
One of the key findings is that the cost per infection detected is less expensive using a risk-based surveillance. However, how do the authors suggest this would be actionable? What strategies would be done to follow-up these infections? Since these results are not about incidence or prevalence, just the presence or absence of at least one case of malaria in a location, how would this be translated into practice? In addition, is it reasonable to assume that molecular diagnostics would be deployed to these types of health facilities? It is already well known that passive case detection is less costly than molecular detection.
The authors do not elaborate on the implications of identifying additional locations where there is a larger proportion of sub-patent infections. Although the overall finding that infections only detected via molecular approaches are more common in forested areas, it is not clear how this would help the program. In addition, the primary outcome measure is the presence or absence of a malaria infection in a location. This is not a common outcome measure and further analyses of how this type of measure would be used and interpreted are needed.
-
##Preprint Review
This preprint was reviewed using eLife’s Preprint Review service, which provides public peer reviews of manuscripts posted on bioRxiv for the benefit of the authors, readers, potential readers, and others interested in our assessment of the work. This review applies only to version 2 of the manuscript.
-
-