Objective: Possible failures of either continuous glucose monitor (CGM) sensors or continuous subcutaneous insulin pumps expose type 1 diabetes patients to possibly severe risks, especially overnight. In a previous contribution, we proposed a failure-detection method (FDM) to detect possible overnight sensor–pump system failures in real time by exploiting CGM and insulin pump data streams. In this contribution, we assess a simplified version of FDM where an average, rather than an individualized, model of the glucose–insulin relationship is used. Method: The FDM consists of two main steps: first, a prediction of the future glucose level is obtained through a Kalman-filter estimator based on a model of the glucose–insulin relationship; second, glucose predictions are compared with CGM samples, and a failure alert is generated if the CGM is not consistent with the predictions. The FDM-a and FDM-p implementations of FDM employ, respectively, average and personalized versions of the glucose–insulin relationship model. Result: Both FDM-a and FDM-p were tested on simulated data created by using the University of Virginia/Padova type 1 diabetes simulator (US2008/067725), which include three types of failures: spike and transient loss of sensitivity of the CGM and failure of the pump in insulin delivery. FDM-a proved to perform satisfactorily even compared to FDM-p, generating a limited number of false negatives and a number of false positives (~20%) only slightly higher than FDM-p. Conclusion: FDM-a performed satisfactorily and similarly as FDM-p, with the significant practical advantage that it does not require going through the data-demanding model individualization phase and is more robust with respect to intraindividual variability. In any case, FDM-a is the approach of choice until enough data for model individualization in FDM-p are collected.
Improving Overnight Safety of Type 1 Diabetes Subjects: Failure Detection Method of Glucose Sensor–Insulin Pumps System Based on an Average Model
DEL FAVERO, SIMONE;SPARACINO, GIOVANNI;COBELLI, CLAUDIO
2012
Abstract
Objective: Possible failures of either continuous glucose monitor (CGM) sensors or continuous subcutaneous insulin pumps expose type 1 diabetes patients to possibly severe risks, especially overnight. In a previous contribution, we proposed a failure-detection method (FDM) to detect possible overnight sensor–pump system failures in real time by exploiting CGM and insulin pump data streams. In this contribution, we assess a simplified version of FDM where an average, rather than an individualized, model of the glucose–insulin relationship is used. Method: The FDM consists of two main steps: first, a prediction of the future glucose level is obtained through a Kalman-filter estimator based on a model of the glucose–insulin relationship; second, glucose predictions are compared with CGM samples, and a failure alert is generated if the CGM is not consistent with the predictions. The FDM-a and FDM-p implementations of FDM employ, respectively, average and personalized versions of the glucose–insulin relationship model. Result: Both FDM-a and FDM-p were tested on simulated data created by using the University of Virginia/Padova type 1 diabetes simulator (US2008/067725), which include three types of failures: spike and transient loss of sensitivity of the CGM and failure of the pump in insulin delivery. FDM-a proved to perform satisfactorily even compared to FDM-p, generating a limited number of false negatives and a number of false positives (~20%) only slightly higher than FDM-p. Conclusion: FDM-a performed satisfactorily and similarly as FDM-p, with the significant practical advantage that it does not require going through the data-demanding model individualization phase and is more robust with respect to intraindividual variability. In any case, FDM-a is the approach of choice until enough data for model individualization in FDM-p are collected.Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.