Where have all the people gone?
This article was co-authored with David Lindenbaum of CosmiQ Works.
At B.Next, we are keenly interested in improving the use of data for the management of infectious disease outbreaks. Because a better understanding of how people move will allow more rapid responses during outbreak events, we are exploring, in collaboration with CosmiQ Works, ways to improve inference of human population movements in low-resource settings.
How does seasonal movement impact disease dynamics?
Population size and movement are important factors that shape the timing, magnitude, and spread of many infectious diseases. Caroline Buckee, Harvard, and her colleagues provide an excellent review [1] of how movement impacts outbreak dynamics. Movement can impact several important factors, such as how people interact with infectious encounters (i.e., infectious people, insects, or environments); the population level immunity (e.g., typically lower immunity in displaced groups of people because of increased stress, among other reasons); and access to healthcare. For example, increases in population aggregation, such as mass gatherings (e.g., Olympics, Hajj, public schools in session), will increase the probability of contact among infectious and susceptible individuals and the number of expected cases. This pattern of outbreaks driven by population aggregation is evident in a range of pathogens that flare when groups of people come together (e.g., influenza, measles, rubella, pertussis, adenovirus).
Surprisingly, estimating the size of national populations and their dynamics (e.g., movement) remains poorly understood in low-resource settings [1],[2]. Population size estimates have significant implications for deriving key population health metrics, such as disease incidence.
Disease incidence is defined as the number of new cases (numerator) divided by the size of the population of interest (denominator). Improving the ability to quantify the number of new cases of an outbreak has been a concerted focus of public health agencies, civil society, and international policies (e.g., International Health Regulations, Global Health Security Agenda). Refining our collective means of aggregating disease case data through improved local community engagement, diagnostics, data systems, and public health surveillance remain ongoing critical efforts that require sustained attention and consistent resources.
However, the denominator (the size of the population of interest) has received less attention. Historically, estimating average population sizes has been a challenging area of research especially in low-resource settings, and estimating sub-populations and dynamic population sizes has been even more challenging [3],[4]. Traditionally, incidence estimates have assumed a static denominator. This assumption is pragmatic, given the available data in many locations, yet provides incorrect estimates because of, in part, human movement.
How is it done now, and what are the key data sources?
Andrew Tatem, at the University of Southampton, reviewed the range of data and analytical approaches that have been used to infer human movement patterns at scales from daily local commuting to permanent intercontinental migrations [5]. The figure from Tatem’s review (see below) provides a quick overview of the types of data that have been used and are being used at different spatiotemporal scales.
The figure highlights relatively recent data sources that have been demonstrated to provide useful inference of population size and movement at regional and seasonal scales. This work and others indicate that data (in order of effectiveness) from call data records (CDRs), satellite imagery, and social media have been the most effective sources of data for addressing human movement at seasonal and regional scales in low-resource settings.
Our focus: Satellite Imagery
Satellite imagery as a data source is especially intriguing to us because of the work being done within CosmiQ Works. Because of this expertise, we will pursue efforts in the near-term and the long-term. In the near-term, we will work with Nita Bharti, at Penn State University, to review the workflow currently used to analyze time-series of nighttime satellite imagery, which is highly manual, labor intensive, and time consuming. We aim to demonstrate alternative approaches to improve the speed and scalability of the workflow.
In the long-term, we want to double down on improving capabilities for identifying structures from satellite imagery. An emerging ‘gold standard’ technique of measuring populations is called ‘bottom-up’ population estimation. In brief, this approach uses micro-censuses (counts of population in small areas) and correlates these counts with geospatial covariates. Inferences can be made for unsampled areas based on the established correlations with geospatial covariates to provide an estimate, and uncertainty, of the population across a broader geographic region. A main predictor in this work has been number and types of structures. This highlights the importance of recent SpaceNet competitions — satellite imagery object detection competitions hosted by CosmiQ Works, DigitalGlobe, and NVIDIA — that focused on advancing machine learning approaches to classify structures from satellite imagery. So, in the long-term, we aim to scope a future project that will assess how algorithms derived from SpaceNet can support ‘bottom-up’ population estimations.
References:
[4] Bharti et al 2015. Remotely measuring populations during a crisis by overlaying two data sources. Int Health 2015; 7: 90–98 doi:10.1093/inthealth/ihv003.
[5] Tatem AJ. Mapping population and pathogen movements. IntHealth 2014; 6:5–11.
B.Next is designing a biodefense technology strategy, demonstrating the potential that innovative tools and techniques can provide, and supporting the investment strategies of these innovations.
Check out our work at www.bnext.org
