Akansh Khurana, CEO,THB (Technology, Healthcare & Bigdata Analytics) states how dengue outbreaks can be predicted in advance with machine learning and artificial-intelligence-based surveillance systems
Dengue Burden in India
Dengue is amongst the most rapidly spreading vector-borne viral disease globally. WHO estimates that close to 87 per cent of the total population in the South East Asia region may be at risk of dengue. As per recent reports, India contributes to nearly 35 per cent of the total global dengue infections and that actual number of cases may be much more than reported as most surveillance systems cannot capture the subclinical symptoms.
Prevalence of dengue in India differs significantly between regions due to variability in size, geographical differences, urbanisation and weather patterns. Rain and temperature are important climatic factors that influence mosquito population and dengue transmission dynamics. Warm temperature and high humidity increase the lifespan of the adult mosquitoes,shorten the viral incubation time and thereby increase virus replication due to increased frequency of blood-feeding.
Changes in weather patterns, due to global warming and climate change, have led to variation in precipitation (prolong flood/excess rainfall) and warm temperatures. Floods can potentially increase the transmission rates of dengue. Standing waters caused by the overflow of rivers can act as a breeding ground for mosquitoes, thereby enhancing the potential for dengue exposure to the flood-affected populations.
Month-Wise Trends in Dengue Prevalence
In the absence of any specific drugs or vaccines against dengue, accurate and early forecast of dengue outbreaks are essential to reduce preventable deaths. THB India, an independent clinical intelligence company, recently analysed de-identified data of around three lakh Indian patients that tested for dengue. These records were analysed from January 2017 till September 2019 and dengue positivity rates were assessed in five zones: North, South, East, West and Central India. Month-wise trends in dengue prevalence were then mapped by plotting positive test results in each zone.
- For North Zone (Chandigarh, Delhi, Haryana, Punjab, Uttar Pradesh, Uttarakhand), the highest increase in the number of positive cases were seen in the months of September, October and November (34 per cent, 40 per cent, and 34 per cent, respectively).
- For South Zone (Telangana, Andhra Pradesh, Tamil Nadu, Karnataka, Kerala), the number of positive cases increased from May till July (~26 per cent), dropped in August, September (~20 per cent) and then rose again in October and November (~23 per cent).
- For East Zone (West Bengal, Odisha, Jharkhand) the months with the highest dengue prevalence were September, October and November (19 per cent, 17 per cent, and 11 per cent, respectively).
- For West Zone (Gujarat, Maharashtra, Rajasthan, Goa), the number of positive cases increased in August, September, October, and November (23 per cent, 27 per cent, 31 per cent, and 27 per cent, respectively).
- For Central Zone (Madhya Pradesh, Chhattisgarh), October (36 per cent) and November (36 per cent) showed the most increase.
This variability in dengue prevalence can be attributed to several environmental and non-environmental factors. Temperature, precipitation, and humidity are significant factors that influence the incidences of dengue fever. This, in addition to unplanned urbanisation, host-pathogen interactions, population immunological factors and inadequate vector control measures can also create habitable conditions for dengue mosquitoes.
According to Dr Manmohan Singh, Medical Director, THB India, “Aedes aegypti, the primary transmission vector of dengue viruses, is more adapted to an urban environment and is common in north and east cities of India. Whereas, Aedes albopictus, the secondary vector is more common forested areas and is more prevalent in southern India. A. albopictus can also survive longer – 65 days as compared to Aedes aegypti, which has a survival time of 22 days.”
Eggs of dengue mosquito can survive without water for one year. In temperatures below 15° C, the survival of mosquitoes and their eggs becomes difficult. Hence, there is a decline in dengue cases towards the end of November.
Dengue Forecast Model with AI and Machine Learning
Dengue is a significant public health problem in tropical countries and preventing it with vector control remains the primary option. Machine learning and artificial-intelligence-based surveillance systems can help to predict dengue outbreaks up to three to four months in advance. These early-warning systems are currently being utilised in several countries across Asia and Latin America: Singapore, China, Malaysia, Thailand, Rio de Janeiro, Manila.
These predictive models have been built to mine and analyse vast amounts of meteorological and epidemiological data, in addition to other parameters like built environment, demographics, to create risk maps and determine dengue-endemic regions. These forecast models can also predict where the next outbreak will be and advise public health agencies to prepare for dengue outbreaks – supply diagnostic kits, increase hospital beds, and make sure that mosquito foggers are ready.
Machine learning is a component of AI in which machines learns from creating self-learning algorithms – takes data and learns from it. With more data and better algorithms,it fine-tunes its learning model to predict dengue outbreaks from local conditions. These tools can provide real-time predictions with up to 80-90 per cent accuracy and have the potential to forecast large dengue outbreaks accurately.
“We are creating a reservoir of data for dengue incidences in India and trying to integrate it with local and regional climate patterns. This will help us to gain a better understanding of dengue trends and incidences and may help us design an early warning system in India,” says Rohit Kumar, Chief of Analytics and Co-Founder, THB India.
“This area is full of opportunities and big rewards; however, big data is yet to reach its full potential in India. Data scarcity is the biggest challenge. Hospitals, institutes, and research centres have different systems of heterogeneous datasets. We are trying to overcome these inefficiencies which are associated with managing huge volumes of data and those which slow down the overall process,” he adds.
Kumar, who works with a team of data scientists to build algorithms for predictive modelling, says, “For best results, it is essential to have access to live data streams with timely, detailed, and accurate values of predictor variables. An ideal system should have the potential to report and predict the next dengue outbreaks in real-time and even suggest vector control management.”