Is climate a primary driver of Vietnam’s dengue, or a shared long-term trend? A five-method analysis over 36 years
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
Vietnam’s reported dengue burden has risen roughly five-fold since 1990. Multi-decadal studies linking climate indices to dengue rarely separate genuine year-to-year coupling from a long-term trend the two share.
Methods
We assembled a provenance-preserving national annual dengue series (1990-2025; OpenDengue plus Ministry of Health figures) and correlated it with annual and March-May means of eight tropical sea surface temperature (SST) indices at lags of zero and one year under five trend-correction lenses: raw, linear detrending, first-differencing, socio-demographic-index residualisation and AR(1) prewhitening.
Findings
Cases rose 3·5 percent annually. Seven predictors were significant at zero lag, led by the annual Indian Ocean Basin-Wide index (IOBW; Spearman +0·534), but linear detrending removed all. First-differencing preserved six, led by spring IOBW (+0·468), the annual Atlantic Multidecadal Oscillation (AMO; +0·462) and annual IOBW (+0·423); three survived AR(1) prewhitening - annual AMO and annual and spring IOBW. Spring AMO and a lag-1 Tropical North Atlantic signal (−0·452) did not, and are hypothesis-generating. El Niño-Southern Oscillation indices failed throughout.
Interpretation
Most of the apparent association reflects a trend shared by warming oceans and expanding surveillance; we could not demonstrate that climate is the primary driver at this scale. Trend is not the whole story: IOBW and annual AMO persist under trend- and persistence-removing transformations. Because transmission responds to climate over weeks to months, annual averaging smooths the lags through which El Niño acts; these nulls reflect temporal scale, not climate insensitivity; usable predictors will require monthly, province-level models.
Funding
Center for Environmental Intelligence, VinUniversity (project VUNI.CEI.FS_0001).
Research in context
Evidence before this study
We searched PubMed, Web of Science and Google Scholar for studies published up to May 2026 linking large-scale climate indices or sea surface temperature to dengue incidence, combining dengue, climate, sea surface temperature, ENSO, teleconnection and time-series terms with Vietnam, without language restriction. Many studies covering two or more decades reported strong correlations between basin-scale indices and national dengue counts. Most, however, relied on raw correlations or a single detrending choice, and rarely tested whether an apparent association reflected genuine year-to-year coupling or merely a shared long-term trend.
Added value of this study
Most long-term studies remove the shared upward trend in only one way, or not at all. To our knowledge this is the first study to compare five trend-correction methods on a multi-decadal national dengue record and to read their agreement or disagreement as a diagnostic of which climate signals are real. A signal that appears only before the trend is removed is probably following it; one that persists is more likely real. Applied to a record spanning more than three decades, this comparison separates the two: several widely reported raw correlations weakened once the shared trend was accounted for.
Implications of all the available evidence
Climate-informed analyses of multi-decadal data should report at least two trend-correction approaches alongside the raw correlation and treat their disagreement as evidence about where a signal sits, rather than operationalising raw long-span correlations. For Vietnam, the apparent national-scale association is dominated by a shared long-term trend but retains a smaller, robust inter-annual component led by the Indian Ocean and AMO signals; genuine coupling is more likely detectable at monthly resolution and provincial scale, where statistical power and physical mechanism are jointly available. Surveillance systems should retain explicit source provenance, so trend-corrected re-analysis remains possible as records grow.
