Identifying Plasmodium falciparum transmission patterns through parasite prevalence and entomological inoculation rate

  1. Benjamin Amoah
  2. Robert S McCann
  3. Alinune N Kabaghe
  4. Monicah Mburu
  5. Michael G Chipeta
  6. Paula Moraga
  7. Steven Gowelo
  8. Tinashe Tizifa
  9. Henk van den Berg
  10. Themba Mzilahowa
  11. Willem Takken
  12. Michele van Vugt
  13. Kamija S Phiri
  14. Peter J Diggle
  15. Dianne J Terlouw
  16. Emanuele Giorgi

  • Curated by eLife

    eLife logo

    Evaluation Summary:

    The association between metrics of malaria transmission based on prevalence of existing infections and the incidence of new infections has epidemiologically important consequences for malaria control and elimination efforts. The association between P. falciparum entomological inoculation rate and parasite prevalence has been previously characterized, and this report evaluates the added-value of spatio-temporal models to such analyses.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #1 and Reviewer #2 agreed to share their names with the authors.)

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

Monitoring malaria transmission is a critical component of efforts to achieve targets for elimination and eradication. Two commonly monitored metrics of transmission intensity are parasite prevalence (PR) and the entomological inoculation rate (EIR). Comparing the spatial and temporal variations in the PR and EIR of a given geographical region and modelling the relationship between the two metrics may provide a fuller picture of the malaria epidemiology of the region to inform control activities.

Methods:

Using geostatistical methods, we compare the spatial and temporal patterns of Plasmodium falciparum EIR and PR using data collected over 38 months in a rural area of Malawi. We then quantify the relationship between EIR and PR by using empirical and mechanistic statistical models.

Results:

Hotspots identified through the EIR and PR partly overlapped during high transmission seasons but not during low transmission seasons. The estimated relationship showed a 1-month delayed effect of EIR on PR such that at lower levels of EIR, increases in EIR are associated with rapid rise in PR, whereas at higher levels of EIR, changes in EIR do not translate into notable changes in PR.

Conclusions:

Our study emphasises the need for integrated malaria control strategies that combine vector and human host managements monitored by both entomological and parasitaemia indices.

Funding:

This work was supported by Stichting Dioraphte grant number 13050800.

Article activity feed

  1. Version published to 10.7554/elife.65682 on eLife
  2. eLife

    Reviewer #3 (Public Review):

    This study implements a secondary analysis of data collected as part of a randomized control trial of malaria vector control interventions in Malawi. The key outputs are statistical associations between two metrics of malaria transmission: P. falciparum parasite prevalence (PfPR) and P. falciparum entomological inoculation rate (PfEIR). There is a rich history of studies investigating this association, spanning a range of approaches: (i) meta-analyses (e.g. Smith et al Nature 2005); (ii) local epidemiological analyses (e.g. Beier et al. AJTMH 1999); (iii) large-scale geo-spatial mapping (e.g. Malaria Atlas Project); and (iv) mathematical transmission models (e.g. Griffin et al Nature Comms 2014). This paper promises to add to this literature using spatio-temporal modelling.

    I was excited by the abstract, and especially by the ambitious questions posed in the introduction (lines 112-117). However, upon reading the manuscript I was left a bit underwhelmed, as the results didn't have much to say in terms of either the spatial or temporal aspects of this relationship. Rather the best-fit model was simply a logit linear model between PfPR and PfEIR with a one month lag.

    Major comments:

    1. Spatial aspect of association. Geostatistical models are challenging to fit, but I have confidence in the authors' ability to do so. Rather, the authors have not demonstrated the extra value of using this approach. Indeed, no spatial results are presented in the manuscript, apart from estimates of model parameters in the appendix which will be uninterpretable to most readers. Points of interest would include, what does a hot spot look like? What does the overlap between different types of hotspot look like? What is the degree of spatial correlation? I appreciate some of this is provided in the separate online animation, but there's no interpretation of what we're seeing.

    2. Temporal aspect of association. The association between PfEIR and PfPR is clearly a temporally complex one as demonstrated by the data in Figure 2. I don't think this complexity has been fully accounted for, beyond simple time lags. For example, I'm quite skeptical of the following result:

    "From the estimated relationship for children, a decrease in PfEIR from 1 ib/person/month to 0.001 ib/person/month is associated with a reduction in PfPR from 37.2% to 20.7% on average (i.e., a 44.5% decrease in PfPR). When transmission has been driven almost to zero, PfPR remains consistently high in children."

    This is a 1000-fold reduction in PfEIR associated with a 44.5% decrease in PfPR. I find this hard to believe, and don't think such a generalizable statement should be made. Rather these are dynamic quantities that vary with each other, and with the time scale over which they are measured.

  3. eLife

    Reviewer #2 (Public Review):

    This study explains the motivation behind considering a spatio-temporal model for modelling malaria transmission and achieves it by using two metrics - Plasmodium falciparum entomological inoculation rate (PfEIR) and Plasmodium falciparum prevalence rate (PfPR), as they believe the two metrics together provide a better picture of transmission. The study modeled the spatial distribution of PfEIR and PfPR for children (0.5-5yr) and women (15-49) in rural Malawi. To estimate PfEIR which is a product of Human biting Rate (HBR) and P.f. sporozite rate (PfSR), HBR and PfSR are modelled as Poisson mixed model with log link and Binomial mixed model with logit link, respectively.

    The study then models the relationship between PfEIR and PfPR, where PfPR is modelled as a Binomial mixed model. Six different models were considered and compared for modelling the relationship between PfEIR and PfPR. Subsequently, the PfEIR and PfPR are then used for hotspot detection.It is satisfactory to note that separate models were used for different species of mosquitos, which eventually led to different set of covariates and random effects. We are also satisfied that the authors have provided the estimates of covariates, temporal trends, and spatial trends. The paper has a well-written discussion section.

    The following issues warrant further attention and clarification.

    1. It seems that a single model is fitted for all three focal regions. Please comment on why the authors believe that the parameter estimates should be common for the three regions (or is this a pragmatic decision)

    2. In the model for PfSR, no spatial random effect was included (formula 2), despite mentioning the spatial heterogeneity throughout the manuscript. Some justification for not including the space term is needed.

    3. In the six models for modelling the relationship between PfPR and PfEIR, do the results change when an overdispersion term (i.e. an independent Gaussian random effect) is included?

  4. eLife

    Reviewer #1 (Public Review):

    Using well-designed surveys, the authors collected mosquito samples and human data along with environmental variables to estimate parasite prevalence (PR) and the entomological inoculation rate (EIR) in three regions of Malawi. They developed advanced geostatistical models to estimate PR and EIR and illustrated the spatial-temporal variation. The online interactivity web-based application showing the spatial-temporal pattern of PR and EIR as well as hot spots in map is particularly useful for visual understandings. These estimations then allow to unveil the time-lagged relationship between PR and EIR. Their data and research approach add very useful information for improving vector-born disease control strategies. Certainly, the data and findings are very useful for malaria control in Malawi.

    Their conclusion seems largely supported by their statistical models and data. However, some outstanding research questions remain. In addition, some statistical issues need to be justified and clarified.

    1. While the spatial-temporal pattern of PR and EIR is illustrated, what are the mechanisms underlying those spatial-temporal variation? Specifically, I think environmental factors and spatial distribution of human population certainly play important roles. Indeed, environmental factors were included in their geostatistical models to estimate PR and EIR. However, the authors made no attempt to provide explanation and discussion for these results (results shown as tables in their appendix).

    2. Furthermore, if environmental factors are left out of focus, what is the additional value of using modelled PfSR and PfEIR for evaluation instead of empirical (observed) PfSR and PfEIR? What is the scientific motivation and justification of using modelled PfSR and PfEIR instead of empirical ones to make the spatial-temporal map and further statistical analyses and then to draw their conclusion on the relationship between modelled PfSR and PfEIR? Statistically, if the same environmental variable is used to fit PfSR and PfEIR, then there is potential spurious correlation (statistical artifact) between the modelled PfSR and PfEIR. The authors need to demonstrated this is NOT the case in their results and analyses.

    3. With A, B, C three regions separated by the national park in the middle (large spatial missing data), is the assumption of isotropic Gaussian process reasonable in their geostatistical model? Sites between A and B have very large distances, but there is no observation data in between. Alternatively, the authors can model the three regions separately?

    4. For hotspot detection, it is unclear whether the hotspots are decided: (1) when the point estimates of PfEIR and PfPR exceed the threshold; or, (2) when the lower 95% confidence bound of the estimates exceed the threshold? If it is the case (1), please justify. Statistically, case (2) is more appropriate. The uncertainty associated with estimates needs to be carefully addressed throughout the manuscript. In any case, please elaborate how the exceedance probability is obtained. My similar concern also appears in other analyses, for example the confidence interval shown in Figure 4.

  5. eLife

    Evaluation Summary:

    The association between metrics of malaria transmission based on prevalence of existing infections and the incidence of new infections has epidemiologically important consequences for malaria control and elimination efforts. The association between P. falciparum entomological inoculation rate and parasite prevalence has been previously characterized, and this report evaluates the added-value of spatio-temporal models to such analyses.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #1 and Reviewer #2 agreed to share their names with the authors.)

  6. Version published to 10.1101/2021.01.14.426709v1 on bioRxiv