Probabilistic seasonal dengue forecasting in Vietnam: A modelling study using superensembles
PLOS Medicine·
- DOI
- 10.1371/journal.pmed.1003542
- PMID
- 33661904
- PMCID
- PMC7971894
- OpenAlex
- W3135866936
- Study type
- Mathematical modelling
- Publisher
- Public Library of Science (PLoS)
- Article type
- journal-article
- Integrity
- current
Publication version
This article has a linked preprint
The relationship comes from Crossref version metadata. Compare both records because findings and wording may have changed between versions.
Open linked preprint →Why this research matters now
The operational forecasting system demonstrates that dengue outbreaks can be predicted with sufficient lead time (up to three months) to inform dengue control activities and public health planning. Predictions align with key Vietnamese decision and planning deadlines, and the system showed added value compared to previous practice of not using forecasts.
Structured evidence summary
Research question
The study examined whether combining Earth observations, seasonal climate forecasts, and lagged dengue cases in a superensemble modeling approach could predict dengue outbreak risk in Vietnam up to six months in advance with sufficient lead time to inform public health decision-making.
Study design
Bayesian spatiotemporal models were fit to 19 years of province-level dengue data (2002-2020) across Vietnam. A superensemble of these probabilistic models was developed to predict dengue incidence at various future time horizons, with retrospective validation using historical data and prospective evaluation for May-October 2020.
Population and setting
The study covered all provinces in Vietnam over the period 2002-2020, with predictions evaluated across different transmission settings including southern Vietnam, which experiences semi-regular seasonal dengue transmission.
Main findings
The superensemble generated more accurate predictions than individual component models across multiple time horizons and settings. At lead times of 1-3 months, the superensemble performed slightly better than a baseline model (CRPS 66.8 vs 79.4) though with larger uncertainty, and demonstrated considerably higher outbreak detection capability (69% vs 54.5%). Predictions were most accurate in southern Vietnam. Prospective predictions for May-October 2020 were also slightly more accurate than the baseline (CRPS 110 vs 125) but with larger uncertainty.
Public-health relevance
The operational forecasting system demonstrates that dengue outbreaks can be predicted with sufficient lead time (up to three months) to inform dengue control activities and public health planning. Predictions align with key Vietnamese decision and planning deadlines, and the system showed added value compared to previous practice of not using forecasts.
Important limitations
The approach is limited by inconsistencies in dengue case data and the lack of publicly available, continuous, long-term data on mosquito control efforts and serotype-specific case information. The system omits important variables not currently available at subnational scale.
GIDS interpretation
This article would be discoverable in GIDS through queries combining dengue surveillance with climate forecasting, outbreak prediction models, or Vietnam-specific dengue research. The classifier links to transmission dynamics and climate-environment topics indicate relevance for users monitoring environmental drivers of vector-borne disease patterns.
Related GIDS surveillance
Literature context does not validate, explain, or change a surveillance signal. Exact and contextual relationships are shown separately.
Evidence relationships
This article has 9 auditable classifier relationships to diseases, places, topics, and study design.