(243f) An Experimental Evaluation of Pca Monitoring Strategies for Type 1 Diabetes Subjects | AIChE

(243f) An Experimental Evaluation of Pca Monitoring Strategies for Type 1 Diabetes Subjects


Finan, D. A. - Presenter, University of California, Santa Barbara
Palerm, C. C. - Presenter, University of California Santa Barbara
Seborg, D. E. - Presenter, University of California Santa Barbara
Bevier, W. C. - Presenter, Sansum Diabetes Research Institute
Zisser, H. - Presenter, Sansum Diabetes Research Institute
Jovanovic, L. - Presenter, Sansum Diabetes Research Institute

Diabetes mellitus is a disease characterized by elevated concentrations of blood glucose, or hyperglycemia. As of 2005, approximately 20.8 million Americans suffer from diabetes, or roughly 7.0% of the population. People with type 1 diabetes compose 5-10% of this total [1]. The hyperglycemia that results from diabetes is known to cause a litany of long-term complications: heart disease and stroke, hypertension, retinopathy, nephropathy, and neuropathy, to name a few. The achievement of normal glucose levels, or normoglycemia, realized through intensive insulin therapy can greatly reduce the risk of developing such complications. In one study, such therapy reduced the risk of developing retinopathy by 76% over a period of 6.5 years [2].

Subjects with type 1 diabetes produce no insulin from their pancreatic β-cells, and thus rely on exogenous insulin for survival. They control their blood glucose levels by injecting insulin; a basal insulin dose covers their requirements in fasting conditions, and is complemented by bolus injections to compensate for glucose resulting from carbohydrate meals. As many factors influence the insulin dosing requirements, subjects depend on periodic measurements of their glucose levels (e.g., finger sticks) to constantly adjust therapy. Therefore, an ?artificial pancreas? consisting of a continuous glucose sensor, an insulin infusion pump, and a feedback control algorithm to automatically determine the insulin delivery rate has the potential to significantly improve glycemic control in diabetic subjects [3].

However, many factors affect insulin pharmacokinetics and pharmacodynamics. Stress, for instance, lowers a subject's insulin sensitivity, thereby increasing the insulin required to achieve normoglycemia. This stress can be both psychological (e.g., grief) or physical (e.g., illnesses or hormonal cycles) in nature, and can occur over a wide range of time scales. Exercise, on the other hand, increases a subject's insulin sensitivity.

From a process control perspective, these unanticipated changes in insulin sensitivity can be viewed as process changes, or ?faults.? In order for an artificial pancreas to function properly in the presence of such faults, a practical and robust monitoring strategy is an essential component of any glucose regulation control strategy.

This paper is the first reported experimental application of a PCA monitoring strategy for subjects with type 1 diabetes. The data for each subject consisted of 5-minute blood glucose measurements, insulin infusion data (i.e., basal rates and boluses), and subject estimates of the carbohydrate content of each meal. From a process control perspective, the data were considered to consist of one output variable (blood glucose concentration) and three input variables (basal insulin infusion rate, bolus insulin infusion rate, and rate of glucose appearance in the blood from a carbohydrate meal). The basal and bolus insulin inputs were considered separately for the following reason: the basal input consists of infrequent step changes to compensate for the subject's diurnal insulin sensitivity variations, while a bolus is an impulse input.

Each subject wore a portable Continuous Glucose Monitoring System (CGMS System Gold, Medtronic MiniMed, Northridge, CA) blood glucose sensor and one of two types of insulin infusion pumps. Each subject was monitored for a period of 6-30 days, including three days in which they were administered a drug to pharmacologically induce stress. The objective of this experimental study was to determine if suitable monitoring strategies could be developed to detect this induced stress state while signaling a minimal number of false alarms.

The subject monitoring was based on two types of applications of principal component analysis (PCA), a well-known multivariate statistical analysis tool. Due to the inherent variability among individuals, the PCA evaluation was performed on a subject-to-subject basis. In the first approach, a PCA model was constructed from several days of normal (i.e., non-stress) calibration data for each subject. Data for new days were then projected onto the PCA model. These new data included both stress days and additional normal days. The effects of alternative data scaling techniques and different numbers of principal components were investigated, and standard metrics such as the Q and T2 statistics were calculated. In general, the stress days violated the confidence limits for these metrics while the normal validation days did not.

In the second PCA method, a PCA model was developed from each day of experimental data. Each model was then compared to the other models using similarity factors. In general, normal days were similar to other normal days, and not similar to stress days. By contrast, the stress days were similar to each other. Thus, in both PCA applications, the monitoring strategies could distinguish between normal days and the days of induced stress.

This joint UCSB-SDRI research project is sponsored by the National Institutes of Health, grant R21-DK069833.


[1] Centers for Disease Control and Prevention. National diabetes fact sheet: general information and national estimates on diabetes in the United States, 2005. Atlanta, GA: U.S. Department of Health and Human Services, Centers for Disease Control and Prevention (2005).

[2] DCCT ? The Diabetes Control and Complications Trial Research Group. The effect of intensive treatment of diabetes on the development and progression of long-term complications in insulin-dependent diabetes mellitus. N. Engl. J. Med., 329, 977 (1993).

[3] Parker, R.S., F.J. Doyle, and N.A. Peppas. A model-based algorithm for blood glucose control in type 1 diabetic patients. IEEE Trans. Biomed. Eng., 46, 148 (1999).