Refining the suitable conditions index to predict dengue fever transmission in Bangladesh and Sri Lanka

Abstract

Background

We developed a Suitable Conditions Index (SCI) to predict dengue transmission in our prior work. However, the initial SCI was not refined with other important abiotic parameters. Therefore, in this study we refined the index by calculating three variants: temperature-based baseline daily average SCI (BDA-SCI), precipitation-weighted daily average SCI (PWDA-SCI), and waterbody-weighted daily average SCI (WWDA-SCI).

Methods

We used the district-wise data for two South Asian dengue-endemic countries: Bangladesh and Sri Lanka. Temperature-suitable days specific to Aedes aegypti (17.05–34.61 ℃) and Aedes albopictus (15.84–31.51 ℃) were averaged (BDA-SCI) and weighted by district-level precipitation (PWDA-SCI) and waterbody data (WWDA-SCI). We assessed the association between dengue incidence and each SCI, along with other covariates using negative binomial regression models. Furthermore, a binomial logistic regression model (BLR) was used to measure the predictive accuracy of each SCI.

Results

The BDA-SCI for Ae. aegypti was highest in Sri Lanka at 0.96 (Standard deviation [SD] 0.04, range 0.85–1.00), compared to Bangladesh 0.68 (SD 0.06, range 0.61–0.87). For Ae. aegypti, WWDA-SCI (Relative risk [RR]_aegypti = 1.06, p = 0.056, Akaike Information Criteria [AIC] 1218.6) and BDA-SCI (RR_aegypti = 1.05, p = 0.008, AIC 1214.2) had a stronger association with dengue incidence in Bangladesh than PWDA-SCI (RR_aegypti = 1.06, p = 0.056, AIC 1232.2), whereas in Sri Lanka, PWDA-SCI (RR_aegypti = 1.06, p = 0.056, AIC 472.63) performed better (AIC_BDA-SCI: 481.36, AIC_WWDA-SCI: 475.89) in the multivariable model, similar to the findings for Ae. albopictus. The BLR model predicted districts with above-median dengue incidence, and model performance indicated that BDA-SCI achieved highest accuracy for Bangladesh, while WWDA-SCI performed best for Sri Lanka, based on higher sensitivity and the Area Under the Curve value.

Conclusions

Overall, the SCI method demonstrated a practical approach for identifying dengue vector suitability and transmission risk. Refining this index with location-specific climatic and environmental variables may enhance the model accuracy and may be used for future predictions under climate change scenarios. Thus, our refined SCI will assist in creating a reliable early warning system and inform the policymakers to initiate vector control strategies, including monitoring and eliminating dengue breeding sites and implementing biocontrol strategies within hotspots.

Supplementary Information

The online version contains supplementary material available at 10.1186/s41182-026-00908-8.

Background

Throughout the twentieth century, periodic large-scale outbreaks of dengue incidence were reported globally [1], with Southeast Asia showing an increased disease burden due to the emergence of dengue hemorrhagic fever and dengue shock syndrome, caused by the atiological agent, the dengue virus [2]. Currently, approximately half of the total population globally is at high risk of dengue transmission [3]. Despite of few ambitious advancements in dengue vaccine development [4] and biocontrol measures [5], the global burden of dengue is anticipated to transcend any historical levels in the near future [6–8].

Aedes aegypti and Aedes albopictus are two primary vectors which spread dengue virus [9]. As ectotherms, they exhibit a great extent of climatic sensitivity, and temperature, being an important abiotic determinant, influences their vectorial capacity. Vectorial capacity is an aggregate measurement incorporating environmental, behavioral, physiological, and biochemical factors affecting the dynamic interactions between the arthropod vector, the pathogen it transfers, and the vertebrate host that harbors the pathogen [10]. A key component of vectorial capacity is vector competence, which is influenced by genetic characteristics that regulate a vector’s capability to acquire, replicate, and transmit a pathogen. Heritable traits, including host-feeding preferences and susceptibility to infection, critically shape vector competence. Besides, being an ectothermic organism, their demographic parameters are strongly influenced by seasonal fluctuations, which can lead to sporadic population surges and increased transmission potential when environmental conditions favor successful interactions with human hosts [11]. Both vector competence and vector population density are principal determinants of vectorial capacity, a concept originally modeled by Ronald Ross and mathematically expanded by George Macdonald [12, 13].

The physiological and behavioral function of dengue vectors are directly influenced by ambient temperature, while the biotic environment is indirectly regulated by larval habitat suitability and vector growth dynamics [14]. Ecosystems shifts, including urbanization, deforestation, and changes in land use, may affect climatic variability, vector habitats, and population dynamics, thereby play a major role in disease transmission patterns [15]. Identifying those patterns are important in this era of changing climate to enable a robust early warning system. Therefore, high resolution geospatial dengue databases are extremely important, but there is a notable gap due to insufficient health care resources in many countries. This puts a barrier to not only collecting strong epidemiological surveillance data but also routine entomological surveys, as suggested by the World Health Organization [9].

We have a solution to address such epidemiological data-related limitations within the predictive modeling by using mechanistic methods and then validate the model performance by adjusting the surveillance data to understand the presence of any random effect. Based on this idea, our previous study defined a suitable temperature range for the two major dengue vectors (Ae. aegypti and Ae. albopictus), known as Suitable Conditions Index (SCI). We used that index to predict the suitable geographical distribution of dengue vectors across six Asia Pacific countries: Australia, China, Indonesia, The Philippines, Thailand, and Vietnam [16]. The established temperature ranges were 17.05–34.61 °C for Ae. aegypti and 15.84–31.51 °C for Ae. albopictus, previously developed on an experimental study that explored temperature-dependent traits for these vectors [17]. These traits directly influence transmission dynamics and are able to define biologically plausible thermal thresholds within the SCI model. However, the previous SCI was not refined by other environmental factors, e.g., the presence of waterbody.

Ecologically, the availability of the appropriate breeding sites is fundamental for vector development, influencing the population density and the likelihood of prolonged transmission cycle [18]. By shipment carriers and routes, the vector or host may travel across the international border, thus facilitating virus transmission in different locations [19]. Therefore, we considered the presence of a waterbody as a key component refining the SCI, in addition to other climatic variables. To achieve this aim, we investigated the relationship between dengue incidence and the refined SCI across two highly dengue-endemic South Asian countries: Bangladesh and Sri Lanka. We calculated the baseline SCI (temperature dependent) and incorporated precipitation and waterbody data to evaluate their predictive accuracy for identifying high-risk areas for dengue transmission. To overcome potential confounding, we used additional variable including elevation [8]. We subsequently examined which version of the SCI had better predictive performance in capturing spatial heterogeneity in dengue risk. We hypothesized that location-specific adjustments to the SCI are necessary to reflect local climatic variability and to address the impact on vectorial capacity by affecting vector survival, breeding site stability, and pathogen development rates.

Methods

Study area

Bangladesh and Sri Lanka are situated in South Asia, at about 23.6850° N, 90.3563° E and 7.8731° N, 80.7718° E, respectively (Figure S1 in appendix 1). The total land area of Bangladesh is 147,630 km², while Sri Lanka covers 65,610 km². Administratively, Bangladesh is divided into 64 districts and Sri Lanka into 25 districts (Figure S2 in appendix 1). We selected Bangladesh and Sri Lanka due to their high dengue endemicity, diverse climatic and ecological conditions, and the availability of relevant environmental and epidemiological data, making them ideal for evaluating regionally adapted climate-based risk indicators.

Historical dengue surveillance and population data

In Bangladesh, dengue cases are classified according to three categories: suspected, probable, and confirmed. All case classifications are incorporated into the National Notifiable Disease Reporting System maintained by the Directorate General of Health Services, Ministry of Health and Family Welfare [20]. Sri Lanka’s National Dengue Control Unit (NDCU) conducts dengue surveillance by a multi-level decentralized system at the district level by means of passive case finding, sentinel site monitoring, and intensified surveillance processes. The activities are coordinated and published by the NDCU [21]. The district-wise population data were obtained in accordance with the most recently published population census for both Bangladesh [22] and Sri Lanka [23].

As the systematic district-level dengue case reporting in Bangladesh began in 2019, the yearly dengue incidence data ranged from 2019 to 2024. For Sri Lanka, yearly dengue case data were retrieved for the period 2018–2021. After integrating dengue case data with district-level population estimates, we calculated the total incidence rate (IR) per 100,000 population.

Climate and environmental data

Remotely sensed climate and environmental datasets, including temperature and precipitation, were acquired via the Google Earth Engine platform (https://developers.google.com/earth-engine/). Minimum and maximum daily temperatures (°C) and total daily precipitation (mm) were extracted from the ERA5-Land reanalysis dataset at an approximate spatial resolution of 11.13 km (Figure S3 in appendix 1). Due to the large geographical heterogeneity in precipitation and its non-linear, inconsistent effects on the dengue incidence, we categorized the precipitation data using percentile-based cutoffs: low to moderate precipitation (average precipitation < 75th percentile), high precipitation (75th ≤ average precipitation < 90th percentile), and very high precipitation (90th percentile ≤ average precipitation). We believe that using percentile-based categorization was a more suitable approach for representing variability without imposing assumptions not supported by the data. The Normalized Difference Water Index (NDWI) was computed from the Sentinel-2 surface reflectance product (COPERNICUS/S2_SR_HARMONIZED) using 10-m resolution bands (Green: B3, Near-Infrared: B8) (Figure S4 in appendix 1). Elevation data were downloaded from WorldClim, with a spatial resolution of 30 arc seconds (~ 1 km). More detailed information related to these variables is given in the supplementary appendix 1 (p. 5–8).

Defining SCI and calculation

To assess vector suitability at the district level, we applied temperature-dependent thresholds previously established for Ae. aegypti and Ae. albopictus [16]. The optimal thermal range for Ae. aegypti was defined as 17.05–34.61 ℃, and for Ae. albopictus as 15.84–31.51 ℃. In this study, we refined SCI using high resolution, daily gridded climate datasets to calculate three variants: the Baseline Daily Average SCI (BDA-SCI), Precipitation-Weighted Daily Average SCI (PWDA-SCI), and the Waterbody-Weighted Daily Average SCI (WWDA-SCI), calculated as follows:

$$BDA-SCI=\frac{1}{N}\sum _{\mathcal{i}=1}^N\delta \left(T_{min,\mathcal{i}},\ge ,T_{suitable,min},\wedge ,T_{max,\mathcal{i}},\le ,T_{suitable,max}\right),$$

$$PWDA-SCI=\left\{\frac{1}{N},\sum _{\mathcal{i}=1}^N,\delta ,\left(T_{min,\mathcal{i}},\ge ,T_{suitable,min},\wedge ,T_{max,\mathcal{i}},\le ,T_{suitable,max}\right)\right\}\times {\mathit{Dailyaverageprecipitation}}_{\mathcal{i}},$$

$$WWDA-SCI=\left\{\frac{1}{N},\sum _{\mathcal{i}=1}^N,\delta ,\left(T_{min,\mathcal{i}},\ge ,T_{suitable,min},\wedge ,T_{max,\mathcal{i}},\le ,T_{suitable,max}\right)\right\}\times {Waterbodyproportion(persquarekm}_{\mathcal{i}}),$$

$$\delta \left(c,o,n,d,i,t,i,o,n\right)=\left\{\begin{array}{c}1,\\ 0,\end{array}\right)\begin{array}{c}\mathit{iftheconditionistrue}\\ Otherwise\end{array}.$$

Here, $T_{\text{min,i}}$ and $T_{\text{max,i}}$ represent daily minimum and maximum temperature and $T_{\text{suitable,min}}$ and $T_{\text{suitable,max}}$ indicate the minimum and maximum temperature threshold for dengue vector suitability, respectively. $\delta$ returns 1 if the daily minimum and maximum temperatures fall within the defined suitability range and 0 otherwise and then was averaged for the whole period to calculate the daily average SCI (Tables S1–4 in appendix 1). The index was then weighted by daily average sum of the precipitation per district $i$ to calculate the PWDA-SCI or NDWI-derived waterbody proportion (for the whole period of study per country) for WWDA-SCI, and $N$ is the total number of days (2192 for Bangladesh and 1461 for Sri Lanka). The spatial distribution of SCI was subsequently mapped, and statistical analyses were performed to evaluate its association with district-wise total dengue incidence in both countries.

Statistical analysis

The district-level total dengue incidence was aggregated for both Bangladesh and Sri Lanka and was modeled using a negative binomial distribution for the dependent variable (dengue incidence) which was important to account for overdispersion of the dengue data:

$$Y_{\mathcal{i}}\sim NegativeBinomial\left({\mu }_{\mathcal{i}},,,\theta \right),$$

where ${\text{Y}}_{\text{i}}$ denotes the observed counts of dengue incidence in district $i$. ${\mathrm{\mu }}_i$ is the expected mean count, and $\theta$ is the dispersion parameter. We used different types of SCI to fit univariable and multivariable models to compare the direction and strength of associations. We further included precipitation category as environmental covariates and elevation data as a fixed effect, and the log of the census population as an offset to account for differing population sizes across subdistricts. We input elevation as a log-transformed variable $log{E}_i$, because the raw elevation values were highly skewed and caused convergence in the negative binomial model. For example, the formula was defined while considering BDA-SCI in a multivariable model:

$$\mathit{log}\left({\mu }_{\mathcal{i}}\right)={\beta }_0+{\beta }_1{(BDA-SCI}_{\mathcal{i}})+{\beta }_2\left({\mathit{Prec}}_{\mathcal{i}}\right)+{\beta }_3\left(\text{log},\left(E_{\mathcal{i}}\right)\right)+\text{offset}\left({\mathit{logPop}}_{\mathcal{i}}\right),$$

where $Pre{c}_i$ represents precipitation category; elevation is represented by $E_i$, and $Po{p}_i$ indicates the population which was used as an offset. The $Pre{c}_i$ was only used as a categorical covariate within the BDA-SCI and WWDA-SCI-based models.

For secondary analysis, the binary outcome variable was calculated using the median dengue IR defined as

$${\mathit{IR}}_{binary,\mathcal{i}}=\left\{\begin{array}{c}1,\\ 0,\end{array}\right)\begin{array}{c}if{\mathit{IR}}_{median,\mathcal{i}}>median(IR)\\ Otherwise\end{array}.$$

The median dengue incidence rate was taken as the criterion for binary classification to allow relatively balanced dividing of the districts into high- and low-risk categories and reduce class imbalance to account for proper comparison across the regions with varying incidence rates. The resulting binary measure was further modeled by a Binomial Logistic Regression (BLR) model with a logit link function to investigate the predictive accuracy of all types of SCIs:

$$\mathit{log}\left(\frac{P({\mathit{IR}}_{binary,\mathcal{i}}=1)}{1-P\left({\mathit{IR}}_{binary,\mathcal{i}},=,1\right)}\right)={\beta }_0+{\beta }_1{\mathit{SCI}}_{\mathcal{i}}+{\beta }_2{\mathit{Prec}}_{\mathcal{i}}+{\beta }_3{E}_{\mathcal{i}}.$$

Here, the response variable represents the log-odds of a district having dengue IR above the median. In this BLR model, average precipitation was modeled as a continuous covariate, which assumes a linear effect on the log-odds and results in a more simplified model compared to using categorized precipitation. Positive coefficients in the logistic model show a greater likelihood of having dengue IR above the median. Using the fitted BLR model, district-level probability estimates were calculated, and high-risk areas were classified using a cutoff value of > 0.5. We evaluated the predictive performance of the models by Akaike Information Criterion (AIC), Area Under the Curve (AUC), crude accuracy rate (CAR), and sensitivity and specificity test.

All the analyses were conducted in RStudio (version 2024.12.0).

Results

In total, 615,963 dengue cases in Bangladesh, and 218,322 in Sri Lanka were reported during the study period. The mean dengue incidence was 287.14 (Standard deviation [SD] 311.52, range 5.97–1851.86) across districts in Bangladesh and 896.9 (SD 644.11, range 120.1–2579.6) in Sri Lanka. During the study period, we observed heterogeneity in the distribution of dengue incidence and environmental data between these two countries, as illustrated in Fig. 1.

Fig. 1.

Box and whisker plots of variability in the dengue incidence and major environmental factors in Bangladesh and Sri Lanka. A Dengue incidence rate, B maximum temperature (℃), C minimum temperature (℃) D total precipitation (mm), and E waterbody proportion, in Bangladesh (n = 64) and Sri Lanka (n = 25). All the data except waterbody ranged between 2019 and 2024 for Bangladesh, and 2018 and 2021 for Sri Lanka. We aggregated the yearly dengue incidence to calculate the total dengue incidence observed during the study period. The waterbody data represent a multi-year median waterbody map derived by identifying all waterbody pixels detected during the study period and calculating their proportion per square kilometer for each district. Total dengue incidence rate was calculated as 100,000 population for both countries. District-wise population data were obtained from the Preliminary Report on Population and Housing Census 2022, Bangladesh Bureau of Statistics for Bangladesh [22], and from Vital Statistics: Estimates on Mid-year Population 2014–2024 By District and Sex (Department of Census and Statistics, 2025) for Sri Lanka [23]

In Bangladesh, the BDA-SCI for Ae. aegypti and Ae. albopictus was 0.68 (SD 0.06, range 0.61–0.87) and 0.46 (SD 0.09, range 0.33–0.78), the PWDA-SCI for Ae. aegypti and Ae. albopictus was 3.75 (SD 0.85, range 2.64–6.71) and 2.56 (SD 0.86, range 1.46–5.22) across districts. The average WWDA-SCI for Ae. aegypti and Ae. albopictus was 5.09 (SD 3.57, range 0.24–16.91), and 5.10 (SD 3.57, range 0.24–16.91). In Sri Lanka, the BDA-SCI for Ae. aegypti and Ae. albopictus was 0.96 (SD 0.04, range 0.85–1.00) and 0.76 (SD 0.17, range 0.49–0.99), the PWDA-SCI for Ae. aegypti and Ae. albopictus was 5.14 (SD 1.95, range 3.15–9.03) and 4.29 (SD 2.37, range 1.77–8.59) across districts. The average WWDA-SCI for Ae. aegypti and Ae. albopictus was 0.89 (SD 0.42, range 0.14–1.68) and 0.70 (SD 0.40, range 0.12–1.58). We mapped all the SCIs along with total dengue IR to illustrate the spatial heterogeneity (Fig. 2).

Fig. 2.

Spatial variation of dengue incidence and Aedes spp. mosquito suitability index in Bangladesh (2019–2024) and Sri Lanka (2018–2021). The figure illustrates the aggregated dengue incidence rate observed during the study period, calculated at 100,000 population. Two study locations A Bangladesh and B Sri Lanka were visualized, and the averages of all types of SCI were calculated for the whole study period. The administrative level 2 maps for both Bangladesh and Sri Lanka were obtained from The Global Administrative Unit Layers (GAUL) dataset [Source of Administrative boundaries: The GAUL dataset implemented by FAO within the CountrySTAT and Agricultural Market Information System projects, available at: https://developers.google.com/earth-engine/datasets/catalog/FAO_GAUL_SIMPLIFIED_500m_2015_level0#terms-of-use, accessed by Google Earth Engine]. SCI suitable conditions index, BDA-SCI baseline daily average suitable conditions index, PWDA-SCI precipitation-weighted daily average suitable conditions index, WWDA-SCI waterbody-weighted daily average suitable conditions index, FAO Food and Agriculture Organization of the United Nations

For Bangladesh, although the relative risk of BDA-SCI for Ae. aegypti (Relative risk [RR] = 1.05, 95% CI 1.01–1.1, p = 0.012, AIC 1241.2) remains similar, the level of significance increased when adjusted with the covariates (RR = 1.05, CI 1.01–1.09, p = 0.008, AIC 1214.2), and the AIC was comparatively lower. Lower and medium precipitation had a stronger association with dengue incidence (RR = 6.36, CI 3.25–11.61, p < 0.001), while the uncategorized PWDA-SCI-based models showed a negative association (RR = 0.66, CI 0.51–0.88, p = 0.001) with both vectors. Moreover, the importance of waterbodies as a covariate was not established in those models (RR = 0.99, CI 0.93–1.06, p = 0.72). Although all WWDA-SCIs did not exhibit the lowest AIC (compared to BDA-SCI 1218.6 versus 1214.2), they exhibited a stronger association in both univariable (RR = 1.13, CI 1.05–1.21, p < 0.001) or multivariable models (RR = 1.06, CI 0.99–1.14, p = 0.056). This suggests better ecological relevance of waterbody metrics in refining the SCI (Table 1). Similarly, WWDA-SCI in Sri Lanka did not demonstrate lowest AIC, comparing to PWDA-SCI, but rather showed stronger associations with dengue incidence. In contrast to Bangladesh, PWDA-SCI had a positive association in Sri Lanka (Table 2). In both countries, higher elevation had a negative association with dengue incidence.

Table 1.

Association of SCI with dengue incidence in Bangladesh

SCI Type	Mosquito type	Model	Variable	RR (95% CI)	p value	AIC
BDA-SCI	Aedes aegypti	Model 1	SCI	1.05 (1.01–1.1)	0.012	1241.2
		Model 2	SCI	1.05 (1.01–1.09)	0.008	1214.2
			Low to moderate precipitation	6.36 (3.25–11.61)	< 0.001
			High precipitation	2.36 (1.07–5.10)	0.025
			Elevation	0.83 (0.66–1.06)	0.16
	Aedes albopictus	Model 3	SCI	1.01 (0.99–1.04)	0.25	1245.6
		Model 4	SCI	1.03 (1.01–1.05)	0.010	1214.7
			Low to moderate precipitation	6.75 (3.43–12.41)	< 0.001
			High precipitation	2.62 (1.20–5.60)	0.011
			Elevation	0.74 (0.59–0.94)	0.012
PWDA-SCI	Aedes aegypti	Model 5	PWDA-SCI	0.64 (0.49–0.87)	0.001	1239.6
		Model 6	PWDA-SCI	0.66 (0.51–0.88)	0.001	1232.2
			Waterbody proportion	0.99 (0.93–1.06)	0.72
			Elevation	0.65 (0.52–0.83)	0.002
	Aedes albopictus	Model 7	PWDA-SCI	0.81 (0.62–1.08)	0.12	1244.9
		Model 8	PWDA-SCI	0.86 (0.66–1.16)	0.27	1238.7
			Waterbody proportion	1.01 (0.94–1.08)	0.86
			Elevation	0.68 (0.54–9.88)	0.007
WWDA-SCI	Aedes aegypti	Model 9	WWDA-SCI	1.13 (1.05–1.21)	< 0.001	1236.5
		Model 10	WWDA-SCI	1.06 (0.99–1.14)	0.056	1218.6
			Low to moderate precipitation	6.13 (3.09–11.22)	< 0.001
			High precipitation	3.91 (1.75–8.51)	< 0.001
			Elevation	0.81 (0.64–1.03)	0.11
	Aedes albopictus	Model 11	WWDA-SCI	1.21 (1.09–1.36)	< 0.001	1235.7
		Model 12	WWDA-SCI	1.11 (1.00–1.24)	0.02	1217.6
			Low to moderate precipitation	6.24 (3.17–11.33)	< 0.001
			High precipitation	4.02 (1.81–8.68)	< 0.001
			Elevation	0.82 (0.66–1.04)	0.13

Relative risks were estimated from univariable and multivariable negative binomial regression models, with 95% confidence intervals

The BDA-SCI originally ranged between 0 and 1; therefore, we normalized the index by calculating the percentage (multiplying by 100). Only the BDA-SCI and WWDA-SCI models included precipitation as a categorical variable in the multivariable analysis, with very high precipitation used as the reference group

SCI suitable conditions index, RR relative risk, CI confidence interval, AIC Akaike Information Criterion, BDA-SCI baseline daily average suitable conditions index, PWDA-SCI precipitation-weighted daily average suitable conditions index, WWDA-SCI waterbody-weighted daily average suitable conditions index

Table 2.

Association of SCI with dengue incidence in Sri Lanka

SCI Type	Mosquito type	Model	Variable	RR (95% CI)	p value	AIC
BDA-SCI	Aedes aegypti	Model 1	SCI	1.00 (0.94–1.06)	0.91	483.45
		Model 2	SCI	1.01 (0.95–11.07)	0.78	481.36
			Low to moderate precipitation	0.90 (0.40–1.81)	0.80
			High precipitation	1.19 (0.47–2.93)	0.72
			Elevation	0.76 (0.63–0.91)	0.007
	Aedes albopictus	Model 3	SCI	1.00 (0.98–1.02)	0.96	483.46
		Model 4	SCI	1.01 (1.00–1.03)	0.12	479.07
			Low to moderate precipitation	1.13 (0.49–2.38)	0.77
			High precipitation	1.15 (0.47–2.72)	0.76
			Elevation	0.68 (0.53–0.85)	< 0.001
PWDA-SCI	Aedes aegypti	Model 5	PWDA-SCI	1.02 (0.89–.17)	0.81	483.4
		Model 6	PWDA-SCI	1.16 (0.03–1.32)	0.018	472.63
			Waterbody proportion	1.92 (1.15–3.19)	0.010
			Elevation	0.70 (0.59–0.84)	< 0.001
	Aedes albopictus	Model 7	PWDA-SCI	1.00 (0.89–1.13)	0.94	483.46
		Model 8	PWDA-SCI	1.14 (1.03–1.27)	0.015	472.63
			Waterbody proportion	1.86 (1.11–3.08)	0.013
			Elevation	0.68 (0.56–0.83)	< 0.001
WWDA-SCI	Aedes aegypti	Model 9	WWDA-SCI	1.71 (0.91–3.17)	0.095	480.78
		Model 10	WWDA-SCI	2.03 (1.16–3.48)	0.010	475.89
			Low to moderate precipitation	0.74 (0.35–1.40)	0.40
			High precipitation	1.36 (0.58–3.06)	0.48
			Elevation	0.75 (0.63–0.90)	0.002
	Aedes albopictus	Model 11	WWDA-SCI	1.56 (0.75–3.36)	0.19	482.1
		Model 12	WWDA-SCI	2.12 (1.16–3.94)	0.010	476.06
			Low to moderate precipitation	0.81 (0.39–1.53)	0.56
			High precipitation	1.31 (0.56–2.95)	0.53
			Elevation	0.71 (0.58–0.85)	< 0.001

Relative risks were estimated from univariable and multivariable negative binomial regression models, with 95% confidence intervals

The BDA-SCI originally ranged between 0 and 1; therefore, we normalized the index by calculating the percentage (multiplying by 100). Only the BDA-SCI and WWDA-SCI models included precipitation as a categorical variable in the multivariable analysis, with very high precipitation used as the reference group

SCI suitable conditions index, RR relative risk, CI confidence interval, AIC Akaike Information Criterion, BDA-SCI baseline daily average suitable conditions index, PWDA-SCI precipitation-weighted daily average suitable conditions index, WWDA-SCI waterbody-weighted daily average suitable conditions index

For Bangladesh, the BLR model for Ae. aegypti showed that WWDA-SCI in the univariable model (OR_aegypti = 1.30, 95% CI 1.10–1.60; AIC = 82.36) and BDA-SCI in the multivariable model (OR_aegypti = 1.32, 95% CI 1.12–1.62; AIC = 63.46) performed better. Similar findings were observed for Ae. albopictus. In Sri Lanka, for Ae. aegypti, PWDA-SCI and WWDA-SCI in the multivariable models showed comparatively better performance, and similar patterns were observed for Ae. albopictus (Table 3). The results of the full multivariable models, including all covariates, are provided in supplemental appendix 1 (Table S5).

Table 3.

Odds ratio for predictors, assessing the association of dengue incidence rate with suitable conditions index using the binomial logistic regression model, with 95% confidence interval

	OR (95% CI)	p value	AIC	Adjusted OR (95% CI)	Adjusted p value	AIC
Bangladesh: Aedes aegypti
BDA-SCI	1.14 (1.03–1.23)	0.02	86.24	1.32 (1.12–1.62)	0.002	63.46
PWDA-SCI	0.45 (0.20–0.88)	0.03	87.05	0.53 (0.23–1.06)	0.10	82.07
WWDA-SCI	1.30 (1.10–1.60)	0.005	82.36	1.13 (0.93–1.42)	0.24	72.72
Bangladesh: Aedes albopictus
BDA-SCI	1.05 (0.99–1.12)	0.09	89.71	1.18 (1.08–1.32)	0.001	61.16
PWDA-SCI	0.71 (0.37–1.27)	0.26	91.38	0.86 (0.44–1.61)	0.65	85.03
WWDA-SCI	1.57 (1.20–2.19)	0.003	80.79	1.34 (0.98–1.97)	0.09	72.94
Sri Lanka: Aedes aegypti
BDA-SCI	1.07 (0.89–1.32)	0.47	35.06	1.03 (0.84–1.27)	0.79	39.34
PWDA-SCI	1.29 (0.85–2.09)	0.26	37.25	1.82 (1.04–3.65)	0.06	35.45
WWDA-SCI	4.55 (0.63–46.02)	0.16	36.40	8.98 (0.92–187.12)	0.09	35.86
Sri Lanka: Aedes albopictus
BDA-SCI	1.02 (0.98–1.08)	0.30	37.50	1.03 (0.96–1.13)	0.39	38.65
PWDA-SCI	1.21 (0.86–1.78)	0.29	37.42	1.61 (1.02–2.85)	0.06	35.77
WWDA-SCI	5.59 (0.68–71.51)	0.14	36.11	12.89 (1.00–394.67)	0.08	35.57

Data are odds ratios (95% CIs), unless otherwise stated. For the binary classification, 1 denotes the reference level for districts with dengue incidence rates higher than median. The BDA-SCI originally ranged between 0 and 1; therefore, we normalized the index by calculating the percentage. Details of the multivariable model performance are shown in appendix 1 (Table S5). Adjusted models included a few environmental covariates including uncategorized average precipitation and elevation. Categorized precipitation variable was not used in the PWDA-SCI model; instead, waterbody proportion was used as a covariate

OR odds ratio, CI confidence intervals, AIC Akaike Information Criterion, BDA-SCI baseline daily average suitable conditions index, PWDA-SCI precipitation-weighted daily average suitable conditions index, WWDA-SCI waterbody-weighted daily average suitable conditions index

The highest AUC values (0.87 and 0.88) were achieved by the multivariable BDA-SCI model for Ae. aegypti (CAR: 77%, sensitivity: 84%, specificity: 69%) and Ae. albopictus (CAR: 81%, sensitivity: 88%, specificity: 75%) in Bangladesh. In both Bangladesh and Sri Lanka, models incorporating covariates achieved the maximum AUC (Table 4). However, in Sri Lanka, the highest accuracy was observed in the WWDA-SCI model for Ae. albopictus (CAR: 77%, sensitivity: 75%, specificity: 77%).

Table 4.

Sensitivity, specificity, and crude accuracy rate, for the binomial logistic regression models. All BDA-SCIs were normalized as percentages for analytical purposes

SCI type	Mosquito type	Model	Sensitivity	%	Specificity	%	Precision	%	Crude accuracy rate	%	F1 score	AUC
			TP/ (TP + FN)		TN/ (TN + FP)		TP/ (TP + FP)		(TP + TN)/ (TP + TN + FP + FN)		a (%)
Bangladesh
BDA-SCI	Aedes aegypti	†	18/ (18 + 14)	56	21/ (21 + 11)	66	18/ (18 + 11)	62	(18 + 21)/ (18 + 21 + 11 + 14)	61	59	0.66
	Aedes albopictus	†	17/ (17 + 15)	53	20/ (20 + 12)	63	17/ (17 + 12)	59	(17 + 20)/ (17 + 20 + 12 + 15)	58	56	0.59
PWDA-SCI	Aedes aegypti	†	26/ (26 + 6)	81	17/ (17 + 15)	53	26/ (26 + 15)	63	(26 + 17)/ (26 + 17 + 15 + 6)	67	71	0.65
	Aedes albopictus	†	23/ (23 + 9)	72	15/ (15 + 17)	47	23/ (23 + 17)	58	(23 + 15)/ (23 + 15 + 17 + 9)	59	64	0.60
WWDA-SCI	Aedes aegypti	†	18/ (18 + 14)	56	23/ (23 + 9)	72	18/ (18 + 9)	67	(18 + 23)/ (18 + 23 + 9 + 14)	64	61	0.72
	Aedes albopictus	†	19/ (19 + 13)	59	24/ (24 + 8)	75	19/ (19 + 8)	70	(19 + 24)/ (19 + 24 + 8 + 13)	67	64	0.74
Bangladesh
BDA-SCI	Aedes aegypti	‡	27/ (27 + 5)	84	27/ (27 + 5)	69	27/ (27 + 10)	73	(24 + 27)/ (24 + 27 + 10 + 5)	77	78	0.87
	Aedes albopictus	‡	28/ (28 + 4)	88	24/ (24 + 8)	75	28/ (28 + 8)	78	(28 + 24)/ (23 + 24 + 8 + 4)	81	82	0.88
PWDA-SCI	Aedes aegypti	‡	26/ (26 + 6)	81	23/ (23 + 9)	72	26/ (26 + 9)	74	(26 + 23)/ (26 + 23 + 9 + 6)	77	78	0.80
	Aedes albopictus	‡	25/ (25 + 7)	78	24/ (24 + 8)	75	25/ (25 + 8)	76	(25 + 24)/ (25 + 24 + 8 + 7)	77	77	0.80
WWDA-SCI	Aedes aegypti	‡	26/ (26 + 6)	81	21/ (21 + 11)	66	26/ (26 + 11)	70	(26 + 21)/ (26 + 21 + 11 + 6)	73	75	0.83
	Aedes albopictus	‡	26/ (26 + 6)	81	22/ (22 + 10)	69	26/ (26 + 10)	72	(26 + 22)/ (26 + 22 + 10 + 6)	75	76	0.84
Sri Lanka
BDA-SCI	Aedes aegypti	†	10/ (10 + 3)	77	4/ (4 + 8)	33	10/ (10 + 8)	56	(10 + 4)/ (10 + 4 + 8 + 3)	56	65	0.67
	Aedes albopictus	†	9/ (9 + 4)	69	6/ (6 + 6)	50	9/ (9 + 6)	60	(9 + 6)/ (9 + 6 + 6 + 4)	60	64	0.62
PWDA-SCI	Aedes aegypti	†	7/ (7 + 6)	54	9/ (9 + 3)	75	7/ (7 + 3)	70	(7 + 9)/ (7 + 9 + 3 + 6)	64	61	0.60
	Aedes albopictus	†	7/ (7 + 6)	54	8/ (8 + 4)	67	7/ (7 + 4)	64	(7 + 8)/ (7 + 8 + 4 + 6)	60	58	0.60
WWDA-SCI	Aedes aegypti	†	8/ (8 + 5)	62	7/ (7 + 5)	58	8/ (8 + 5)	62	(8 + 7)/ (8 + 7 + 5 + 5)	60	62	0.65
	Aedes albopictus	†	8/ (8 + 5)	62	8/ (8 + 4)	67	8/ (8 + 4)	67	(8 + 8)/ (8 + 8 + 4 + 5)	64	64	0.71
Sri Lanka
BDA-SCI	Aedes aegypti	‡	9/ (9 + 5)	62	9/ (9 + 3)	75	8/ (8 + 3)	73	(8 + 9)/ (8 + 9 + 3 + 5)	68	67	0.69
	Aedes albopictus	‡	7/ (7 + 6)	54	7/ (7 + 5)	58	7/ (7 + 5)	58	(7 + 7)/ (7 + 7 + 5 + 6)	56	56	0.71
PWDA-SCI	Aedes aegypti	‡	9/ (9 + 4)	69	8/ (8 + 4)	67	9/ (9 + 4)	69	(9 + 8)/ (9 + 8 + 4 + 4)	68	69	0.79
	Aedes albopictus	‡	9/ (9 + 5)	69	8/ (8 + 4)	67	9/ (9 + 4)	69	(9 + 8)/ (9 + 8 + 4 + 4)	68	69	0.78
WWDA-SCI	Aedes aegypti	‡	9/ (9 + 4)	69	8/ (8 + 4)	67	9 / (9 + 4)	69	(9 + 8)/ (9 + 8 + 4 + 4)	68	69	0.78
	Aedes albopictus	‡	10/ (10 + 3)	77	9/ (9 + 3)	75	10 / (10 + 3)	77	(10 + 9)/ (10 + 9 + 3 + 3)	76	77	0.78

Details of the dengue transmission probability for each district are provided in Supplementary Appendix 2

^a2 * (precision * sensitivity) / (precision + sensitivity)

^†Univariable model $\left(log\left(\frac{P\left(I{R}_{binary,i}=1\right)}{1-P\left(I{R}_{binary,i}=1\right)}\right)\right)={\beta }_0+{\beta }_{1}SC{I}_i$

^‡Multivariable model $\left(log\left(\frac{P\left(I{R}_{binary,i}=1\right)}{1-P\left(I{R}_{binary,i}=1\right)}\right)\right)={\beta }_0+{\beta }_{1}SC{I}_i+{\beta }_{2}Pre{c}_i+{\beta }_3{E}_i$. Here, ${\mathit{SCI}}_{\mathcal{i}}$ indicates different types of SCI, while for the BDA-SCI and WWDA-SCI, average precipitation data were used. Within the PWDA-SCI multivariable model, waterbody proportion was used as a covariate

SCI suitable conditions index, TP true positive, TN true negative, FN false negative, FP false positive, AUC area under the curve, BDA-SCI baseline daily average suitable conditions index, PWDA-SCI precipitation-weighted daily average suitable conditions index, WWDA-SCI waterbody-weighted daily average suitable conditions index, IR incidence rate, E elevation

Overall, the presence of waterbodies showed a strong positive association with dengue incidence in Sri Lanka. District-wise predictive probabilities for all models are provided in supplementary appendix 2.

Discussion

Our study investigated whether refining SCI improves the model predictive accuracy which substantially broadens the scope of previously developed SCI. The results revealed that, compared to the precipitation-based SCI, the temperature and waterbody-based SCI provided lower AIC, higher AUC, and crude accuracy rates. The findings will be particularly useful to create an early warning system for tropical or sub-tropical countries sharing similar climates and experiencing severe climatic extremes, whereby the suitability indices are highly likely to be influenced. Overall, this study contributes novel insights into refining the SCI framework for Aedes spp. vectors and illustrates how a mechanistic, climate-informed ecological approach can better align with observed dengue incidence patterns.

Previously we used SCI to predict dengue incidence in Guangdong Province, China, as well as across six countries in the Asia–Pacific region including Australia, China, Indonesia, the Philippines, Thailand, and Vietnam [16, 24]. The initial SCI used the number of suitable months per year for each dengue vector and then was weighted by precipitation to understand risk locations. We provided a novel approach to refining SCI and used higher temporal resolution for the climatic and environmental data, which will allow us to further investigate different environmental covariates for more refinement, leading to country-specific, more reliable future predictions of dengue vector distribution based on the climate change scenarios.

With global warming intensifying, temperature-based prediction modeling is critical for identifying ecologically suitable locations for vector establishment and informing preventive public health interventions before the occurrence of autochthonous outbreaks [25–28]. Although vector suitability is primarily regulated by climatic variables, particularly temperature, the inadequacy of breeding habitats can stop the chain of pathogen–vector–host transmission even when the vector has high competency to harbor the pathogen [10]. Moreover, excessive rainfall can devastate larval habitats and further reduce the population size of the vector and the dissemination of the virus, as was further validated by our study [29]. Therefore, there is a need for creating different thresholds for important climatic variables which may efficiently capture the extremes and increase the prediction accuracy.

Traditionally, reactive public health responses are adopted to control vector-borne disease outbreaks, but may be exacerbated by high population mobility [30], overcrowded living conditions (such as during the pandemic-related lockdown conditions) [31], and limited access to vector control resources or to evidence-based policy implementation, factors that collectively contribute to prolonged infection [32]. Besides, entomological surveillance may provide inconsistent estimates of actual dengue incidence in a given time and space [33]; is subject to operational constrains [34], and reflect underlying geographic and infrastructural inequities, eventually hindering the ability to estimate the local vertebrate host susceptibility.

Therefore, understanding vector ecology is important to address the environmentally driven dynamics of arboviral transmission, in which temperature is found to be the most used parameter [8]. Although the critical thermal minimum, maximum, and optimum temperature-driven response of Ae. aegypti and Ae. albopictus was identified before [16], there is limited understanding of precipitation thresholds for dengue vector suitability, which requires future research. There are many other confounding factors and building the understanding of interactions among natural, ecological, and anthropogenic changes to the urban setting could be beneficial. Although the embedded ecological bias may initiate concern due to coarse-grained data that inadequately captures micro-environmental heterogeneity and urban-specific transmission determinants, particularly in densely populated megacities like Dhaka, Bangladesh, we have to be careful about misclassifying areas, overestimating suitability in low-risk regions and underestimating it in high-risk ones.

The identification of dengue hotspots will allow the plan to implement biocontrol including Wolbachia-infected dengue vectors or traditional interventions including those suggested by the United States Environmental Protection Agency [35]. Underlying all these efforts, strengthening epidemiological and entomological surveillance systems remains a foundational requirement for improving the efficacy of spatiotemporal modeling.

This study represents a significant advancement over previously developed SCI by examining two highly dengue-endemic countries in the Southeast Asia region. Approximately 1.3 billion people in this region are exposed to the risk of dengue infection, and dengue incidence rose by 46% between 2015 and 2019 [36]. The findings of this study indicate the need to investigate additional environmental factors related to dengue incidence in this region and to provide recommendations for improving the surveillance system to enhance accuracy. We supplemented our investigation by evaluating the predictive accuracy of the SCI through major diagnostic measures such as the CAR and the measures of sensitivity and specificity. Adding this key hydrological variable improved the ecological realism of the vector habitat suitability assessment. The district-wise dengue surveillance data for the period 2019–2024 for Bangladesh were used to enhance the temporal coverage and the epidemiological utility of the analysis. The actual incidence of dengue was utilized for model validation enabling the robustness of the method and providing intriguing insights on gaps observed at the spatial level and informing further refinement of the model.

However, there are a few limitations worth considering. First, a primary constraint related to the temporal scope of the dengue surveillance data. While the dataset from Bangladesh encompasses the most extensive period available since the initiation of district-level surveillance in 2019, the Sri Lankan data, published weekly by the national Epidemiology Unit as part of the notifiable disease system and reported by Medical Officers of Health, extend from 2007 to the present [37]. Despite its range, inconsistencies in granularity and reporting mechanisms may limit cross-country comparability, and future investigations should seek to harmonize temporal coverage for improved inference. Second, our analytical framework assumed the independence of dengue incidence rates across districts. Nonetheless, dengue transmission is inherently spatially structured by ecological, socio-demographic, and mobility-related factors [30]. Omission of the analysis or modeling for spatial autocorrelation or the absence of regression modeling may ignore geographically dependent patterns of transmission, presenting a potential future research area for development. Furthermore, the Sri Lankan dataset’s BLR models showed non-significant associations potentially explainable by reduced statistical power stemming from few data points or suboptimal spatial resolution. Either increased spatial granularity for the epidemiological data or inclusion of additional covariates such as vector density, movement of population, or urbanization, may increase model sensitivity and explanatory capacity. Finally, our model cross-validation process, using classification performance at a 0.5 threshold in probability, showed increased predictive capacity of the WWDA-SCI than the precipitation-weighted version. However, the absence of empirical evidence to define a biologically meaningful precipitation threshold limited our capacity to establish a validated cutoff. Consequently, our analysis relied on temperature suitability parameters derived from previously published experimental entomological studies. Any future shifts in these fixed thresholds, potentially due to vector adaptation or changes in viral–vector–host interactions should be rigorously monitored to detect evolutionary or ecological shifts in vector competence or environmental suitability. Finally, several significant yet presently absent predictors including humidity, built-up area index, urbanization, and dengue vector breeding site data may be integrated into future research.

Conclusions

This study found that the temperature and waterbody refined SCI had superior predictive accuracy and could capture the spatial heterogeneity in the observed dengue incidence. Therefore, the importance of applying this method as a critical tool in early warning systems as well as future prediction is strongly supported. We showed how two tropical countries located almost within the same region may require different types of refinement, as the variation was observed in total amount of precipitation, elevation, and the presence of waterbody, still known as the top-most dengue-endemic countries in the world. We found that the whole Sri Lanka country is susceptible to the distribution of Ae. aegypti so as the most parts of Bangladesh. An important insight we observed was in the classification of precipitation, where low to medium precipitation had a strongly positive association with dengue incidence, though when used as an aggregated amount may have a negative association. It implies creating a scientific threshold or using this parameter cautiously when refining the index. Most importantly, the hotspots we discovered in this study require robust vector control measures. Overall, we recommend using the strength of this ecological research method to extend this to more parameters and countries. Strengthening the dengue data, robust methodologies with appropriate model validation will lead us to reliable future prediction and thus help policymakers to act accordingly.

Abbreviations

SCI

Suitable conditions index

NDWI

Normalized Difference Water Index

BDA-SCI

Baseline daily average suitable conditions index

PWDA-SCI

Precipitation-weighted daily average suitable conditions index

WWDA-SCI

Waterbody-weighted daily average suitable conditions index

BLR

Binomial logistic regression

AIC

Akaike Information Criterion

IR

Incidence rate

AUC

Area under the curve

CAR

Crude accuracy rate

SD

Standard deviation

CI

Confidence interval

RR

Relative risk

GAUL

Global Administrative Unit Layers

FAO

Food and Agriculture Organization

NDCU

National Dengue Control Unit

OR

Odds ratio

TP

True positive

TN

True negative

FN

False negative

FP

False positive

Author contributions

J.I. was involved in data acquisition, undertook analysis, interpreted the results, and drafted the manuscript; W.H. is the supervisor of the project and served as the senior author. He conceived and designed the study and managed the overall research activities. All coauthors including W.H., F.D.F., C.K.D., X.Q., and H.B. contributed to the interpretation of the results, critical review, and revision of the manuscript.

Funding

None.

Data availability

The datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This work was based on secondary epidemiological data. The research is approved by the University Human Research Ethics Committee of the Queensland University of Technology (LR 2023-8014-16778), Brisbane, Australia.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.