Ran Wei

Rutgers University - New Jersey Medical Shcool

Graduate Student

NLST (Learn more about this study)

NLST-222

May 26, 2016

A Novel Approach to Facilitate Interoperability between Clinical Decision Support Systems and Electronic Health Records

Clinical practice calls for reliable diagnosis and optimized treatment; however, human errors in health care remain as a severe issue even industrialized countries. The application of clinical decision support systems (CDSS) casts light to this problem. By generating patient-specific assessment and clinical suggestions with better adherence to standards, CDSS may reduce medication errors and increase efficacy of care services. However, given the great improvement in CDSS contribution to quality of care in past several years, the problems pointed out by previous studies still persist, to various degrees. Challenges to adoption of CDSS include (1) decision making of CDSS is complicated by complexity of human physiology and pathology, which may render the whole process time-consuming by loading huge patient-related data; (2) incompatibility to current clinical workflow makes CDSS a lonely session, i.e. additional input work of patient information may be required, further increasing the burden of clinicians; (3) CDSS diagnosis model has to be reconstructed dynamically due to fluctuation of patient health data, which may greatly slow down system performance given huge amount of data sets. One popular strategy is integration of CDSS in health information systems (HIS) to directly read electronic health records (EHR) for analysis. However, misinterpretation of health information by this heterogeneous HIS may occur, as the document standards of EHR are not unified and on the other hand, HIS may use different default clinical terminologies to define input health data. Several proposals have been published so far to guarantee correct CDSS-EHR communication via redefinition of terminologies according to standards used by recipient of each data flow, but they mostly aim at specific versions of CDSS guideline.

We are trying to view this problem in different way. Compared to conventional approaches, we suggest more fundamental change – uniform clinical terminology should be used by EHR and HIS (with its integrated CDSS). Facilitated data exchange will increase overall data loading efficacy for analysis. Furthermore, this updated CDSS should be machine-learning based, which dynamically update knowledge according to constantly incoming training dataset.

As the last section of this research, a case study will be done to examine feasibility of our concepts in EHR and HIS design. First we will test whether data can be transferred between two sides unambiguously. The second task is to compare diagnosis and therapy recommendation generated by our suggested algorithm to conclusions made by clinicians. In order to better simulate clinical practice, we hereby request NLST datasets for this study. Usage of datasets will be cited in our work and will be restricted to current proposed research plan. This study is proposed for paper call of Journal of Medical Systems.

Aim 1. Unify the data format of CDSS and EHR. In order to facilitate interoperability between CDSS and EHR, one solution is to define the same data format standard and protocol for these two systems. In this study, new data format protocol will be raised to prevent the misinterpretation due to different definitions taken by different systems for the same clinical concepts.

Aim 2. Design a machine-learning based CDSS algorithm. Currently the majority of CDSS are knowledge-based, i.e. CDSS make decisions following pre-set rules defined by committee of experts. These rules should be updated frequents to keep up with the novel findings in clinical practice. Here we suggest a machine-learning based CDSS, which can update the clinical knowledge dynamically to save the labor of maintenance and update. Primary structure of algorithm for this novel system will be given.

Aim 3. Check the function of this new system in a case study. In this aim this newly built system will be tested using data obtained from NLST. Certain diagnostic and therapeutic data sets will be used as training data, and the system will try to diagnose and give treatment suggestions using other data, and decisions made by system will be compared with those from clinicians to determine the reliability of algorithm.