Session P7D.5

A Temporal Search Engine for a Massive Multi-Parameter Clinical Information Database

LH Lehman*, TH Kyaw, GD Clifford, RG Mark

Massachusetts Institute of Technology
Cambridge, MA, USA

Efficient searching for clinically significant physiological events in a large-scale, multi-parameter, time-series medical database is a challenging problem. In particular, one main challenge is in the detection of clinical events that may involve complex dynamics of multiple physiological parameters over multiple time scales. Traditional threshold-based searching algorithms are incapable of detecting the dynamics of physiological changes over time. We describe a novel search engine that is capable of rapid execution of queries concerning changes in the gradients and absolute (and relative) values of multiple irregularly sampled and asynchronous physiological parameters over many time scales. The parameter list includes 128 different important physiological indicators such as heart rate, blood pressure, temperature, cardiac output, laboratory values, IV drug levels, microbiology results, Glasgow Coma Scale scores, etc. Our approach has three major novel aspects. First, the search engine provides a temporal query syntax and framework that are capable of defining search criteria for multiple physiological parameters using gradient bounds, rates of change, and threshold breeches over various time scales. Multiple signals can be searched and combined in a Boolean manner to form complex queries. Search criteria can be defined to detect changes in physiological conditions that have occurred over a specified range of time scales. Secondly, the search engine implements an efficient time series search algorithm that uses pre-computed ranges and multi-scale gradients to significantly reduce the search time for locating temporal events. Finally, to illustrate the use of our search approach, a set of numerical search criteria were developed by clinicians using our temporal query syntax that allows the location of evidence for important pathophysiological conditions such as paroxysmal tachyarrhythmia, hemorrhagic shock, acute renal failure, metabolic acidosis and multi-organ failure. We have implemented the search engine in MATLAB and tested the algorithm on a massive multi-parameter intensive care unit database (MIMIC II). Specific illustrative examples of its flexibility and power will be presented, such as identifying the temporal location of episodes of metabolic acidosis in patients with acute myocardial infarction.

(Abstract Control Number: 266)