By Sangita Paul, Research Officer (Data Analyst) UNDP Bangladesh Research Facility Policy and Strategic Advisory Unit, UNDP

The second thematic area of United Nations World Data Forum 2023 focuses on the use of reliable data and statistics to generate insightful knowledge for decision and policymakers to improve people's lives. This blog outlines how analysis of crowdsourced data can enable early prediction of disease spread, allowing for necessary actions to combat a pandemic.

On 8 March 2020, the first case of Covid-19 was diagnosed in Bangladesh and the country went into lockdown for two months from 26 March to 30 May 2020. During this period there was a significant movement of people across the country, leading to extensive disease spread.

Like in other countries, challenges such as limited testing capabilities, insufficient healthcare capacity and a lack of urgent and accessible data hindered efforts to monitor the epidemic’s spread in real time. Hospital-based surveillance data was slow, particularly from rural and remote areas. To combat these challenges, the a2i programme of UNDP Bangladesh together with government of Bangladesh introduced a unique participatory syndromic surveillance system at the initial stage of the COVID-19 pandemic to detect disease outbreak in advance.

Managing Crowd- Sourced Data

Three main sources of crowd-sourced data are utilized, including: 1) an interactive voice response system and national hotline numbers, and 2) an unstructured supplementary service data-based messaging system, and 3) various web and mobile applications. People are asked to report symptoms such cough, fever, and shortness of breath, along any contact with symptomatic people and/or with someone who tested positive for COVID-19, as well as their age and gender. Each response in the system is geolocated based on the nearest cell phone tower and mapped to a sub-district. After a preliminary diagnosis based on responses to questions in each data stream, individuals are classified as low and high-risk using algorithms set by each mobile phone operator. Callers are contacted and reconfirmed through the automated outbound calls. Potential high-risk patients are then verified by a doctor’s pool and referred to the Director of Health for further processing, including sample collection.

Validation of Participatory Syndromic Data

There are two syndromic indicators: number of people reporting multiple symptoms per 100,000 and number of people classified as high-risk per 100,000. While validating the indicators, statistically, we found a positive association between them and confirmed cases data with a one- or two-weeks lag. This indicates that people who report multiple symptoms at one time point will be potentially diagnosed with Covid-19 one or two weeks later. Similarly, people, who are classified as high risk at one time may become confirmed covid-19 case one or two weeks later (Mahmud, A. S., et al. (2021)). The days between the first high-risk classification and the first confirmed case are analyzed for each sub-district. The median lead time was found to be 10 days for most of the sub-districts. Further details of the analysis and results can be found here.

Practical implications

In May 2020, hot zone areas were identified in Narayanganj, Gazipur, Jessore, Madaripur and other districts (Table 1). The system identified Narayanganj district the as a hot zone on 29 April 2020, and the first confirmed case was found on April 7, 2020, in those areas, i.e., indicating a prediction lead time of 9 days. Similarly for other hot zones, the system predicted outbreaks 5 to 12 days in advance.

TABLE 1: IDENTIFICATION OF HOT ZONES: HOW EARLY PREDICTION WAS MADE THROUGH SELF-REPORTING SYSTEM

This surveillance system has evolved into the National COVID-19 Data Intelligence system, which includes a national dashboard for policy makers at the local level. The dashboard integrates COVID-19 case data, test data, test positivity, death data, and hospitalization data providing visual analytics to support decision-making. Policymakers can use this information to prioritize interventions in health services for the most affected areas (e.g., provide mass testing, oxygen supply, etc.). The system also enables timely resource allocation to manage patient overload, by providing medical personnel, hospital equipment, ICU beds, and other necessary resources.

Way forward

This participatory syndromic surveillance system can be further enhanced with innovative solutions for strategic preparedness such as mortality surveillance, contact tracing, geofencing, and epidemiological modelling to identify hot spots, track suspected cases, and forecast disease progression accurately before testing. To minimize the noise and volatility of self-reported syndromic surveillance data, we can conduct awareness campaigns at the community level to increase citizens’ awareness of diseases and their symptoms. Additionally, we need to include more women, elderly and marginalized population who are often underreported in the CDR data due to their mobile SIM cards (Subscriber Identity Module) not being registered under their names. To build in-house analytical capacity, we can recruit public servants with technical expertise and offer capacity building trainings related to statistics, data analysis and analytics. Finally, we should prioritize the use of updated data for making timely decisions while ensuring data privacy and security.

Concluding remarks

By combining additional syndromic data from clinics and hospitals, hospitalization, and mortality rate, this participatory surveillance system allows early detection of the outbreaks, disease control, and implementation of preventative measures by the decision-makers and policymakers. It can also help schedule hospital beds and other necessary medical supplies in advance to handle patient overloads. Additionally, the system enables the community support team to promote non-pharmaceutical interventions and raise awareness nationwide. This approach can be particularly useful during epidemics with limited testing availability and non-specific symptoms. This blog post on crowdsourcing syndromic surveillance system can bring some new insights to the United Nations World Data Forum 2023 session TA2.34 titled "Toward better collection and analysis of data in healthcare applications" to help identify future pandemics early and prevent deaths.