Laboratory Notes: Model-Based Patient Monitoring

George Verghese and Thomas Heldt
Laboratory for Electromagnetic and Electronic Systems

Patients in intensive care units (ICUs) are closely monitored through sensors that continuously track electrical activity of the heart, blood-pressure waveforms, and oxygenation levels. A variety of intermittent recordings also capture such items as medication levels, ventilator settings, body temperature, fluid input and output, and much more. In addition, various invasive measurements requiring skilled operators are undertaken from time to time. Even outside the ICU, increasingly large amounts of data are routinely collected, facilitated by rapid advances in physiological sensor technologies and wireless communication.

Advanced cardiovascular monitoring revealing total peripheral resistance (TPR) for an ICU patient. The five red dots constitute the intermittent measurements available using the highly invasive methods that current practice is limited to (two such measurements suffice to calibrate our method). The lower trace shows the level of a medication used to increase TPR, and it is evident that our TPR estimate follows the time course of the medication quite closely, a fact that is not nearly as evident from the intermittent invasive measurements.

Advanced cardiovascular monitoring revealing total peripheral resistance (TPR) for an ICU patient. The five red dots constitute the intermittent measurements available using the highly invasive methods that current practice is limited to (two such measurements suffice to calibrate our method). The lower trace shows the level of a medication used to increase TPR, and it is evident that our TPR estimate follows the time course of the medication quite closely, a fact that is not nearly as evident from the intermittent invasive measurements.

The interpretation of this massive influx of raw data is typically left to clinical staff, who can of necessity only deal with a small fraction of the data, and only in relatively limited ways. The premise of work being done in our group—in collaboration with Professors Roger Mark (HST, EECS) and Peter Szolovits (CSAIL, EECS), and under funding from the NIH, NASA and Philips Medical Systems—is that medical care can be significantly enhanced by the use of computational tools incorporating well-established static and dynamic models from physiology. The resulting advanced monitoring system would integrate, analyze, and interpret clinical data in real time, relating measured variables to each other, extracting underlying parameters, generating more sophisticated alarms and alerts, and presenting distilled information in useful forms to the clinical staff.

The premise of work being done in our group—in collaboration with Professors Roger Mark (HST, EECS) and Peter Szolovits (CSAIL, EECS), and under funding from the NIH, NASA and Philips Medical Systems— is that medical care can be significantly enhanced by the use of computational tools incorporating well-established static and dynamic models from physiology.

The premise of work being done in our group—in collaboration with Professors Roger Mark (HST, EECS) and Peter Szolovits (CSAIL, EECS), and under funding from the NIH, NASA and Philips Medical Systems— is that medical care can be significantly enhanced by the use of computational tools incorporating well-established static and dynamic models from physiology.

Our focus so far has been on cardiovascular monitoring. With Tushar Parlikar, who recently completed his PhD in the group, we have established model-based algorithms to monitor non-invasively and continuously several hemodynamic variables, such as cardiac output and total peripheral resistance (TPR), that are of particular importance for diagnosis and tracking of conditions such as heart failure and shock. As an example, the figure (above) shows, in the upper continuous trace, our estimate of TPR for an ICU patient. The five red dots constitute the intermittent measurements available using the highly invasive methods that current practice is limited to (two such measurements suffice to calibrate our method). The lower trace shows the level of a medication used to increase TPR, and it is evident that our TPR estimate follows the time course of the medication quite closely, a fact that is not nearly as evident from the intermittent invasive measurements.

Leave a Reply

Replies which add to the content of this article in the EECS Newsletter are welcome. The Department reserves the right to moderate all comments. If you would like to provide any updated information please send an email to newsletter@eecs.mit.edu.