Section Snippets

Introduction

Exploratory Data Analysis (EDA) is a comparatively new area of statistics.(Bruce and Bruce, 2017) It is based on the principle that “It is important to understand what you CAN DO before you learn to measure how WELL you seem to have DONE it”(Tukey, 1977). EDA is complementary to confirmatory inferential statistics as it minimises violations of assumptions for model building. It also enables understanding of the data and guides appropriate questions, analysis, and models. Exploratory Spatial Data Analysis (ESDA) is an advancement to EDA for the detection of spatial patterns, hypotheses formulation based on spatial features, and assessing the appropriate spatial models. Similarly, for datasets including both space and time attributes, Exploratory Spatio-Temporal or Space-Time Data Analysis (ESTDA) has been recently introduced and is an active research domain in the field of Geographic Information Science (GIS). (De Smith et al., 2018) The current approach is preferred compared to other methods for time series classification, such as federated distillation learning system (EFDLS)(Xing et al., 2022a), robust temporal feature network (RTFN)(Xiao et al., 2021), and strategies for hybridisation of supervised learning, unsupervised learning, and Self distillation(Xing et al., 2022b). The mathematical foundations for the approach used are underpinned by the fact that dengue counts in routine data have Poison distribution and have been explored to subsequently develop spatiotemporal regression models and time series forecasting for a continuous outcome/ dependent variable. Further, underreporting of mild and missed cases in LMICs and the lack of adequate geocoding accuracy for spatial point pattern analysis pose a limitation for classification algorithms.

Material and methods

The present study included secondary data analysis of routinely collected datasets by healthcare system and multiple open-source datasets. The study design was an ecological study in healthcare epidemiology using the data science approach.

Dengue epidemiology in the state

The state dengue incidence rates in 2015–19 were 47.8, 33.6, 52.0, 49.7, and 33.4 per 100,000, respectively. The mean (SD) age of the reported dengue cases was 34.3 (16.8) years. Most patients were males (63.9%) and in the age group of 25–39 years (32.0%). The fourth quarter, October, and week 41 to week 46 were dengue’s peak periods. The Hurst coefficient of quarterly, monthly, and weekly time series was 0.5, 0.91, and 0.99, and the spectral entropy measure was 0.28, 0.53, and 0.72,

Discussion

Dengue occurrence in a population is influenced by ecological and socio-demographic factors (Farrar and Manson, 2014). Exploration of associations between disease occurrence and its eco-socio-demographic factors provides evidence for feature selection and development of early warning forecasting systems for strengthening disease surveillance, efficient resource allocation, and timely implementation of prevention and control measures. The present study is a first-of-kind study from the state

Conclusion

The present study provides evidence and a framework for the exploration of Spatio-temporal associations of dengue with ecological and socio-demographic variables in the local context. The study found a high dengue incidence in the state with seasonal patterns. At the sub-district level, changing epidemiology of dengue was observed, which strengthens the call for climate change policy implementation. A non-linear association of dengue with risk factors was seen at multiple lags and with

Availability of data and algorithms

All the analyses were carried out using open-domain data sources and reproducible codes, which can be shared with readers on reasonable requests. The data from NVBDCP, Punjab, and Sub-district level spatial files were obtained with restricted use, which can be shared only after additional permissions from the state directorate and Punjab Remote Sensing Authority.