Semi-Blind and l1 Robust System Identification for Anemia Management

TL;DR

This paper develops and compares two robust system identification methods for personalized anemia management in CKD patients, demonstrating that semi-blind identification outperforms l1 in modeling drug-response dynamics.

Contribution

It introduces a semi-blind robust system identification approach for individualized drug response modeling in CKD anemia management, improving accuracy over existing methods.

Findings

Semi-blind identification yields lower MMSE than l1 method.

Patient-specific models improve drug response prediction.

The approach enhances personalized treatment strategies.

Abstract

Chronic diseases such as cancer, diabetes, heart diseases, chronic kidney disease (CKD) require a drug management system that ensures a stable and robust output of the patient's condition in response to drug dosage. In the case of CKD, the patients suffer from the deficiency of red blood cell count and external human recombinant erythropoietin (EPO) is required to maintain healthy levels of hemoglobin (Hb). Anemia is a common comorbidity in patients with CKD. For an efficient and robust anemia management system for CKD patients instead of traditional population-based approaches, individualized patient-specific approaches are needed. Hence, individualized system (patient) models for patient-specific drug-dose responses are required. In this research, system identification for CKD is performed for individual patients. For control-oriented system identification, two robust identification techniques are applied: (1) l1 robust identification considering zero initial conditions and (2) semi-blind robust system identification considering non-zero initial conditions. The EPO data of patients are used as the input and Hb data is used as the output of the system. For this study, individualized patient models are developed by using patient-specific data. The ARX one-step-ahead prediction technique is used for model validation at real patient data. The performance of these two techniques is compared by calculating minimum means square error (MMSE). By comparison, we show that the semi-blind robust identification technique gives better results as compared to l1 robust identification.