Poster
COMPOSE: A model for composable computer interpretable guidelines using data from smart wearable systems
Motivating Scenario
The metabolic syndrome (MS) is a cluster of health conditions that occur together and increase the risk of heart disease, stroke and diabetes. As the availability of wearable sensors is becoming more popular, the collection of frequent physiological data from individuals has become easier than ever. This raises a need for new models that interpret continuous physiological values and provide meaningful interpretation for patients and caregivers. One way of interpreting these data is by automating existing evidence-based guidelines. The assumption is that, by combining different clinical guidelines relating to the metabolic syndrome with the physiological data of the patient, we can predict deterioration states that may require medical attention. Such solution can assist caregivers in identifying high-risk patients and provide patient-tailored interventions.
Challenges
Languages for representing computer interpretable guidelines (CIGs) can be partitioned in: models based on creating meta-data over documents and using these to create recommendations; decision-tree models; and task-network models. The task network models are of particular interest because they execute the represented clinical knowledge against patient’s data. These declarative models have a big advantage in that they do not require all the possible scenarios that can occur during the execution of a rule to be predicted in advance. However, the current solutions have two main drawbacks: (i) they do not combine different, competing, declarative models, and (ii) they do not focus on both temporal and frequency patterns to support the reasoning. This is very important in our motivating scenario where there are many guidelines referring to different conditions, and the evidence can conflict if the patient has more than one medical condition.
Ongoing work
In COMPOSE1 we have extended the event calculus approach [1] with temporal patterns that facilitate the expression of the CIG and the reasoning with the frequent temporal data. In addition to the temporal patterns, we defined a model for representing and executing different CIGs and the algorithms for combining the outcomes from reasoning with different guidelines [2] into consistent advice based on the patient’s profile. Several clinical guidelines relating to diabetes, hypertension and chest pain (as specified by the English National Institute for Health and Care Excellence [NICE]2) are automated using the extended temporal calculus. Several forms of physiological data (including heart rate, activity, glucose levels and breathing) coming from wearable sensors are used to generate the alerts and to score the health and fitness level of the patient. To evaluate the results, tests are executed to compare the scores with the GOLD standard.
Conclusions
We address the problem of using multiple clinical guidelines to reason with temporal frequent data provided by the patient. Incorporating data coming from wearable sensors as health data can help with patient-tailored early interventions, provide personalised care and improve adherence of the medical professionals to clinical guidelines; in turn, this may significantly improve patient care.
1 http://www.hevs.ch/fr/mini-sites/projets-produits/aislab/projets/compose-1866
2 www.nice.org.uk
Correspondence
Correspondence:
Dr. Visara Urovi
University of Applied Sciences
Techno-Pôle 3
CH-3960 Sierre
visara.urovi[at]hevs.ch
References
1 Shanahan M. The event calculus explained. In: Wooldridge MJ, Veloso M, editors. Artificial intelligence today. Berlin Heidelberg, Germany: Springer-Verlag; 1999. p. 409–30.
2 Urovi V, Bromuri S, Stathis, K Artikis Al. Initial steps towards run-time support for norm-governed systems. In: De Vos M, Fornara N, Pitt JV, Vouros G, editors. Coordination, Organizations, Institutions, and Norms in Agent Systems VI. Berlin Heidelberg, Germany: Springer-Verlag; 2011. p. 268–84.
Copyright
Published under the copyright license
“Attribution – Non-Commercial – NoDerivatives 4.0”.
No commercial reuse without permission.
See: emh.ch/en/emh/rights-and-licences/