Developing a NITROglycerin Dose Titration Decision Support System (NITRO DSS)

Background. Angina is one of the most common reasons people visit Canadian emergency departments, with approximately 800,000 annual visits. Nitroglycerin is a first-line medication for the acute relief of angina. There is no standardized dose of intravenous nitroglycerin because its impact on blood pressure varies significantly among patients. Trained critical and coronary care nurses manually increase or decrease the dose to achieve optimal titration, defined as obtaining the intended therapeutic effect (relief of angina) while avoiding side effects associated with nitroglycerin. Nurses use their clinical judgment and past experiences to inform their clinical decision-making about nitroglycerin dose titration. Many nurses have trouble selecting the right dose and anticipating the patient’s response to the dose adjustment, leading to suboptimal titration of nitroglycerin. Clinical decision support systems have been shown to support nurses’ titration-based decision-making for other medications in critical care. However, a system for nitroglycerin titration currently does not exist. A decision support system designed for the titration of nitroglycerin infusions would facilitate the prediction of blood pressure responses consequent to potential adjustments in dosage. This system, when integrated with the clinical judgment and experiential knowledge of nursing professionals, would enhance the precision and effectiveness of nitroglycerin titration. Objectives. To develop a Nitroglycerin Dose Titration Decision Support System (nitro DSS) that provides blood pressure predictions following a nitroglycerin dose change. Design. A multi-method design with quantitative and qualitative methods was used to develop the clinical decision support system for nitroglycerin dose titration. A systematic review and meta-analysis was conducted to compare blood pressure measured using a continuous non-invasive and invasive arterial pressure device in adult patients admitted to a critical care setting (Study One). This was followed by a retrospective observational design study to predict the subsequent systolic blood pressure following dose titration (defined as any change in the dose of nitroglycerin) within a 30-minute window (Study Two). The accuracy of a linear model, least absolute shrinkage and selection operator, ridge regression, and a stacked ensemble model trained using the Auto Gluon-Tabular framework were investigated. A persistence model, where the future value in a time series is predicted as equal to its preceding value, was used as the baseline comparison for model accuracy. The nitro DSS interface (visual display) was designed using a user-centred approach consisting of two phases (Study Three). The first phase was a qualitative study with semi-structured interviews to identify design specifications for the visual display of nitro DSS. The second phase was three iterative rounds of usability testing to test and refine the prototype. In each round of testing, participants completed two questionnaires: the System Usability Scale (to measure usability) and the Ottawa Acceptability of Decision Rules Instrument (to measure acceptability). Results. The systematic review and meta-analysis revealed substantial differences between blood pressure measurements obtained from continuous non-invasive and invasive monitoring devices. Given the critical differences, continuous non-invasive arterial pressure monitoring is not a reliable replacement for invasive monitoring in adult patients requiring critical care. As a result, continuous non-invasive arterial pressure monitoring is unsuitable for developing a clinical decision support system that aims to incorporate predictions of blood pressure. Therefore, available electronic health record data was used to train machine learning models to predict blood pressure responses to nitroglycerin titrations. The results of study two identified the stacked ensemble model developed using the AutoGluon-Tabular framework to have the lowest root-mean-square error of all models, producing a 22% improvement against the baseline (a persistence model). The results of phase one of the user-centered design revealed four themes for the interface design: (1) Clear and Consistent, (2) Vigilant, (3) Interoperable, and (4) Reliable. Nurses suggested the initial and subsequent prototype versions incorporate features reflecting the four identified themes. The findings from study one informed the development of an initial prototype, which underwent three iterative rounds of usability testing in phase two. Nurses tested the prototype and provided feedback to improve its usability and acceptability. All study participants' ratings on usability and acceptability exceeded the minimum threshold. Conclusion. This thesis successfully applied a multi-method design to develop nitro DSS, a clinical decision support system that predicts blood pressure responses to a potential nitroglycerin dose change. Upon completion of three rounds of usability testing, a refined nitro DSS prototype was identified which demonstrated excellent usability and acceptability as defined in the System Usability Scale and the Ottawa Acceptability of Decision Rules Instrument, respectively. Subsequent research includes developing a high-fidelity prototype, testing nitro DSS in silent trials, and conducting a randomized control trial.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.