Publication Details



Clinical decision support systems (CDSSs) have been shown to assist clinical decision making in health- care while also enabling timely and appropriate integrated care services. The clinical pathway (CP), in particular, delivers and outlines an optimal logical path and plan of care from assessment to treatment at the primary and secondary health care level. Clinical pathways are increasingly used in routine pa- tient care to maintain care process standardization, improve patient outcomes, reduce costs, and empower local healthcare practitioners. However, these clinical decision support and/or clinical pathway systems have remained out of reach for low-resource settings (LRSs). In LRSs, following the paper-based clinical guideline is a traditional practice and the only choice utmost. The service is suboptimal and challenged to deliver accurate and adequate evidence for decision making. Furthermore, low clinical competence, limited diagnostic capabilities, high turnover and low health professional motivation are some of the challenges that most public health facilities in developing countries are facing on a day-to-day basis. This dissertation demonstrates and develops computer-aided point-of-care decision support instru- ments for identifying referral and locally treatable cases with the goals of: (i) empowering local and (inexperienced) care providers to decide about the need for referral or not and make better clinical referral decisions towards primary and first-level hospital care, (ii) assisting the management of care processes by generating a CP or plan of care to reduce delays, costs, and improve patient outcomes, (iii) improving diagnosis and follow up (by learning about the trends from evidence and historical records), (iv) assisting in the diagnostic process and helping caregivers by utilizing a knowledge-based comput- erized clinical workflow and processes in conjunction with data-driven approaches such as proba- bilistic Bayesian learning, (v) providing multi-criteria decision analysis (often expressed in concordance tables) to adjust or re- adjust decision priority as well as issues like reliability of the prediction, patient referral and classi- fication analysis by estimating the level of risks and providing best practices for improvement. To develop CDSSs for LRSs, five key steps were completed. The overall need for the development of clinical decision support systems was initially assessed, and it was found that: (i) existing paper-based point-of-care instruments have the disadvantage of being non-interactive and difficult to use for extracting relevant clinical information, summarizing the patient history, constructing a patient flow diagram, diagnosing all potential underlying diseases, and ultimately delivering optimal CPs, ii) utilizing existing care information to deliver adaptive evidence-based health services will require the development of a suitable algorithm that works with limited clinical input and incrementally updates the generated CP each time new information becomes available, (iii) accessing, monitoring, and tracking the history of referred patients and referral feedback is chal- lenging with the present paper-based referral coordination and communication system. Then, a state-of-the-art review was conducted to investigate design approaches for executable CPs at the point-of-care, and the results show that exploring a trade-off mechanism between knowledge-based and data-driven techniques is critical for promoting data-driven decision-making approaches. Addition- ally, LRS challenges such as limited infrastructure, resource constraints, deficient data readiness, data inconsistency, data incompleteness, and adaptability (understanding the context) must be addressed to close the evidence-practice gap. - An algorithm for the automated and dynamic generation of CPs was developed using knowledge-based and data-driven approaches. The key principle of our proposed algorithm is that it operates with mini- mal clinical input and may be updated as new information becomes available, and it dynamically maps and validates the initial knowledge-based CP based on the local context and historical evidence in order to provide a multi-criteria decision analysis. Our proposed algorithm was evaluated using 719 ‘preg- nancy, childbearing, and family planning’ dataset records from Jimma Health Center (Ethiopia), and promising results were found. - The proposed solution was then deployed on an edge device, the Raspberry Pi 4 Model B, to provide a point-of-care clinical reference, data processing, and workflow generator, as well as an interactive data visualization and clinical guidance wizard for LRS. The Raspberry Pi 4 is intended to work with a power bank when there is no electricity in remote areas. Furthermore, the CDSS is accessed via a smart phone’s mobile data or wireless network. - Finally, user acceptance of the CDSS at the point-of-care in LRSs was evaluated using 22 parameters organized into six major categories, namely ease of use, system quality, information quality, deci- sion changes, process changes, and user acceptance. The care givers at the health center were asked to express their level of agreement using a think-aloud approach. The evaluation received a favor- able agreement score in all six categories by obtaining primarily strongly agree and agree responses. A follow-up interview, on the other hand, indicated a variety of reasons for disagreement based on the neutral, disagree, and strongly disagree responses. Furthermore, the overall acceptability was simulated using partial least squares structural equation modeling, and a variety of factors impacting the accep- tance of the CDSS in LRSs were examined. In all, the key features of the CDSSs are able to provide low-cost, automated, adaptable, interactive, and applicable CPs for LRSs. A wider scale evaluation and longitudinal measurements, including CDSS usage frequency, speed of operation and impact on intervention time have not been included in the thesis work, because they require a larger deployment in daily practice.