A Preparation Framework for EHR Data to Construct CBR Case-Base

Introduction

Diabetes Mellitus (DM) is a serious disease. If it has not treated on time and properly, it can lead to serious complications including death. This makes diabetes one of the main priorities in medical science research, which in turn generates huge amounts of data. These data are transactional and distributed in the patient's EHR. An early diabetes diagnosis is the most critical step in diabetes management. The diagnosis of diabetes is an ill-formed problem and depends on the physician experience. Case Based Reasoning (CBR) is considered as the most suitable Clinical Decision Support System (CDSS) for dealing with these problems where physicians share their experience (Richter and Weber, 2013; Blanco, 2013). Therefore, case-base creation is a challenging step. On the other hand, CBR is appealing in medical domains because a case-base already exists as the stored symptoms, medical history, physical examinations, lab tests, diagnoses, treatments, and outcomes for each patient (Andritsos et al., 2014). However, because clinical data are usually incomplete, inconsistent, and noisy, these data need a set of preparation steps before converted into CDSS knowledge (Abidi & Manickam, 2002). The first step is the data preprocessing stage that is applied to enhance data quality. The application of a set of machine learning algorithms improves the accuracy of CBR case retrieval algorithms. The second step is the coding stage that is used to represent the pre-processed data with standard coding terminology such as SNOMED CT (SCT) (Lee et al., 2013). We have proposed a diabetes diagnosis reference set from SCT version 2013 and modeled it in an OWL 2 ontology (El-Sappagh et al., 2014). This ontology is used to encode the unstructured (i.e. textual) contents of the case base knowledge base. Lack of standard data affects the accuracy of CDSS implementation (Ahmadian et al., 2011). Data standardization is critical for CBR systems for many reasons. The encoded knowledge supports: (1) the creation of distributed CBR systems; (2) the integration and interoperability between CDSS and EHR environment (Ahmadian et al., 2011); and (3) the creation of knowledge-intensive CBR systems. As a result, CBR supports semantic retrieval algorithms, and its intelligence is increased (Melton et al., 2006). Finally, the third step is the data fuzzification stage that is used to handle vague knowledge. Physicians always describe patients using vague terms, such as the sugar level is high, the patient has obese, and so on. Moreover, the patients often describe their conditions using imprecise terms. As Zadeh (2003) argued much of the knowledge that humans acquire through experience be perception-based and thus subject to imprecision and inaccuracy. Such knowledge, when not treated in some suitable way that can consider and convey its inherent imprecision, usually leads to reduced effectiveness of the knowledge-based systems that use it. Vagueness can be handled using fuzzy logic (Zadeh 2003), which has been used in diabetes diagnosis rule-based systems (Lee and Wang, 2011). Moreover, fuzzy logic has been integrated with CBR in hybrid systems (Abdul et al., 2014) and used for calculating the fuzzy similarity between cases (Khanum et al., 2009). However, in diabetes diagnosis domain, there are no studies in fuzzy CBR systems.