The Workshop on Algorithmic Medical Decision Making: Leveraging Heterogeneous Data Sources for Quantifying Risk will spin off this WG to explore ways to make better use of human sensing and ubiquitous communications in public heath care planning and decision making. The goal is to discover and integrate algorithms that proactively sample humans to obtain information about their health and the health of those around them, then analyze these data to detect and reduce health risks by delivering care in a more targeted and cost-effective manner.
Recent methodological advances in the quantitative analysis of space-time data and ubiquitous communications enabled by widespread use of computer and wireless phone networks have created new opportunities to adaptively sample for locally-targeted data as well as for public health data over large-scale geographical areas. Recently, greater use has been made of wireless communications and social media, e.g., the use of ``twitter'' inputs by the CDC, to compute maps of the spread of the H1N1 virus. However, the potential of employing humans as sensors, and of integrating sampling decision making algorithms in more of an automated surveillance system, has not been fully explored.
This WG will address questions such as: (a) How should human ``sensors'' be selected for sampling? (b) What protocols and instruments are needed to acquire health-related information from human sensors, and how should these be adapted as more is learned? (c) What statistical algorithms are most effective for detecting health risks and exploiting spatial-temporal information? (d) What techniques are available for assessing reliability of analytical results and reducing false positive rates? (e) How might social media and peer-to-peer networking be used to reduce costs of healthcare data acquisition? (f) What privacy concerns arise in the use of human sensors and how can we exploit privacy-preserving data mining methods developed at DIMACS and elsewhere to enable the use of human sensors?
Over the last decade, public health researchers have produced a collection of methods under the rubric of syndromic surveillance to signal an outbreak. (An early DIMACS WG on ``Adverse Event Detection'' led to collaboration with the CDC in the creation of an annual Syndromic Surveillance conference.) Syndromic surveillance systems face inherent trade-offs that limit their effectiveness. Consequently, new decision making approaches are needed to improve the precision and timeliness of data collection, and to develop better metrics for the confidence in and reliability of results.
Typically, syndromic surveillance has involved collecting and analyzing statistical data on health trends, e.g., symptoms reported by people in emergency rooms. There are fewer examples of proactively sampling end users of computer and wireless communications networks about their health and the health of others in their social and physical environments. The WG will explore decision making algorithms to: implement adaptive sampling over space and time; gather space-time statistics, including spatial scan statistics and analysis of incomplete or censored data; use self-organizing systems and agent-based modeling to generate simulated data and assess efficacy of adaptive sampling procedures; manage mobile communications networks to ensure timely and efficient data acquisition and information sharing; and combine evidence in useful ways.
Next: Call for Participation
Workshop Index
DIMACS Homepage