The research team set out to correct a critical bias that had affected earlier syntheses of the evidence. A previous meta-analysis published in 2015 had pooled studies of serotype and severity, but it included investigations that sampled only hospitalised patients. Because hospitalisation itself requires a certain threshold of illness severity, restricting the denominator to hospitalised cases artificially inflates the estimated probability that a symptomatic infection becomes severe. The new analysis therefore excluded studies that monitored severe disease only among hospitalised patients, and it also kept severe cases defined under the WHO 1997 guidelines, the WHO 2009 guidelines, and hospitalisation as three separate classification systems rather than merging them, since these definitions capture clinically distinct constructs.

The search strategy was exhaustive. The authors queried Medline, Global Health, and Embase Classic plus Embase for all relevant studies published up to 20 January 2025, generating 8,533 records. After de-duplication and screening, 73 studies passed the review criteria, of which 71 contributed data to the meta-analysis. To be included, studies had to report laboratory-confirmed symptomatic dengue cases with matched serotype and severity information for the same group of people, and prior exposure-serotype combinations involving fewer than five cases were excluded to avoid unstable estimates. Patients with non-dengue co-infections or chronic conditions such as diabetes and hypertension were also excluded, as these are known risk factors for severe outcomes that could confound the serotype-specific signals the team sought to isolate.

Methodologically, the study represents a substantial advance over conventional random-effects meta-analysis. The authors developed a Bayesian inferential framework implemented in the Stan probabilistic programming language, fitting a generalised linear mixed additive logit model in which each study outcome, the number of severe cases among a given number of symptomatic cases, was assumed to follow a binomial distribution. The model estimated an intercept, a geographic region effect distinguishing Asia from the Americas, and separate coefficients for every prior exposure-serotype combination, with Asia-secondary-DENV2 serving as the reference scenario. Crucially, the framework accounted for correlation between multiple outcomes reported within the same study through multivariate normal random effects with scenario-specific covariance matrices, allowing data from one region to inform estimates in the other where individual scenarios lacked sufficient studies for pooling.

The headline quantitative findings are sobering. Under the WHO 1997-type classification, secondary DENV3 infection in Asia emerged as the most severe scenario overall, with a pooled 37.61 percent of symptomatic cases developing dengue haemorrhagic fever or dengue shock syndrome, a credible interval spanning 24.19 to 52.32 percent. Secondary DENV4 produced the lowest proportion of DHF and DSS among secondary infections, at 26.00 percent in Asia and 15.80 percent in the Americas. Yet when the authors formally compared serotypes conditional on the same prior exposure status, the differences were not statistically significant. Even the contrast between the most and least severe secondary infections, DENV3 versus DENV4, yielded an odds ratio of 2.16 with a 95 percent credible interval of 0.56 to 6.05, crossing unity.

The picture changed decisively when the analysis turned to prior exposure. Under 1997-type severity, secondary DENV2 and secondary DENV3 were both significantly more severe than primary infections of the same serotype, and secondary DENV3 was significantly more severe than primary infections of any other serotype for which estimates were available. This pattern is consistent with the well-characterised immunopathology of dengue, in which a first infection confers long-lasting homotypic immunity but only transient heterotypic protection, and secondary infection with a different serotype can trigger antibody-dependent enhancement. In this mechanism, individuals carrying intermediate titres of pre-existing cross-reactive antibodies experience enhanced viral entry into Fc-receptor-bearing cells, elevating the risk of the vascular leakage that defines severe dengue relative to both dengue-naive individuals and those with high antibody titres.

Geography also left its mark on the estimates, though less robustly. Using 1997-type severity, the odds of a dengue case being severe in the Americas were on average roughly half those in Asia, an odds ratio of 0.53 with a credible interval of 0.27 to 0.91 that excluded one. However, this regional difference lost statistical significance in sensitivity analyses that excluded cases with unknown prior exposure status or applied stricter de-duplication criteria, suggesting that differences in population-level immunity profiles and underlying demographics, rather than an intrinsic regional virulence, may drive the apparent excess of severe dengue in Asia. Historically higher rates of DHF in Asia have also been attributed to variation in circulating viral genotypes and to differences in healthcare practice, including the clinical laboratory tests, such as haematocrit measurement, that the 1997 classification requires.

The authors subjected their results to an unusually thorough battery of validation and sensitivity checks. A simplified model fitted only to data from the long-running Nicaraguan and Thai cohort studies reproduced the central finding that DENV3 was the most severe primary and secondary infection, although this restricted dataset did reveal significant serotype differences within the same prior exposure status, with secondary DENV2 and DENV3 emerging as the most severe. Pooled estimates of hospitalisation proportions for DENV1, DENV2 and DENV4 with unknown prior exposure aligned well with observed hospitalisation rates in the placebo arm of the TAK-003 vaccine trial at 57 months of follow-up, though DENV3 estimates ran two- to threefold higher than trial observations, a discrepancy the authors attribute partly to country-specific hospitalisation practices, since most trial DENV3 cases came from the Philippines. Leave-one-out analyses, stricter de-duplication, and prior-distribution sensitivity analyses all left the core conclusions intact, and risk-of-bias assessment using an adapted Hoy tool classified no study as high risk.

Substantial statistical heterogeneity remained a persistent feature of many scenarios, with I-squared values exceeding 75 percent in all 1997-type scenarios where prior exposure was unknown, and prediction intervals were correspondingly wide; the prediction interval for primary DENV2 in the Americas stretched from 0.02 to 56.06 percent around a central estimate of just 4.08 percent. The authors identify several plausible contributors: regional differences in healthcare practice, genetic variation in circulating strains, varying proportions of primary and secondary infections among participants, patient age, study design differences, and exposure to other flaviviruses, notably Zika virus, which has been shown to enhance subsequent dengue disease risk. Data limitations also precluded modelling patient age or the sequence of infecting serotypes across a lifetime, both of which have been implicated in clinical outcomes.

The practical implications reach from the clinic to the modelling suite. Clinically, the absence of serotype-specific severity differences means that the infecting serotype alone, independent of other patient parameters, cannot be used to predict the course of illness or to guide triage. For outbreak response, knowledge of the dominant circulating serotype, considered separately from historical transmission dynamics, is unlikely to determine healthcare and hospital demand, which underscores the value of seroprevalence surveys in reconstructing population immunity profiles. The estimates generated here are intended to parameterise mathematical models of dengue transmission and burden, informing projections of the impact and cost-effectiveness of vaccination, antiviral drugs and vector control, including Wolbachia-based strategies, under current and changing climatic conditions. With the Qdenga vaccine deployed in Brazilian public campaigns since February 2024 and its efficacy known to depend on both pre-vaccination serostatus and infecting serotype, quantifying severe disease risk across the full matrix of exposure and serotype combinations has become a public health priority that this analysis now addresses with unprecedented rigour.

Subject of Research: Association between dengue virus serotypes, prior exposure and severe disease outcomes

Article Title: Association of dengue severity with the serotype of infection: a systematic review and meta-analysis

Article References: Ang, H. J., McCormack, C., Donnelly, C. A., & Dorigatti, I. (2026). Association of dengue severity with the serotype of infection: a systematic review and meta-analysis. eClinicalMedicine, 100, Article 104218. https://doi.org/10.1016/j.eclinm.2026.104218

Image Credits: AI Generated

DOI: 10.1016/j.eclinm.2026.104218

Keywords: dengue, DENV serotypes, severe dengue, meta-analysis, antibody-dependent enhancement, WHO classification, secondary infection, hospitalisation, Bayesian modelling, vaccination, epidemiology, global health