Doctor of Philosophy, Case Western Reserve University, 2022, Clinical Translational Science

The translation of medical therapies from basic and preclinical research to efficacious human interventions is challenging. The majority of candidate therapies fail in early-stage human trials, after showing promise in preclinical work. The primary aim of the research presented herein is to explore the potential role that poor statistical practice in preclinical animal trials might play in contributing to translational failure. First, a comprehensive appraisal of current statistical practice in one area of preclinical neuroscience research was carried out. A close review of the current related literature is presented, and the appraisal includes a tutorial to explain how certain statistical mistakes might result in overly optimistic results, as well as practical recommendations for improvement. A majority of articles included in this appraisal failed to account for sources of non-independence in the data (74-93%) and/or did not analytically account for mid-treatment animal attrition (78%). Ordinal variables were often treated as continuous (37%), outliers were predominantly not mentioned (83%), and plots often concealed the distribution of the data (51%). Next, a sample including both successful and failed human trials for neurologic targets was identified, and rates of statistical mistakes in the associated preceding rodent trials were compared. Failed human trials were found to have higher rates of select sources of potential statistical bias in preceding rodent trials, compared to successful trials. This research provides evidence that a contributing factor to translational failure is statistical misapplication in preclinical animal research in the neurosciences. It provides the groundwork for future research that will provide practical solutions to translational researchers and funders, facilitating preclinical experimental validity to increase the translational success rate.

Committee: Mary Dolansky PhD RN FAAN (Committee Chair); Kenneth Baker PhD (Committee Member); Nancy Obuchowski PhD (Committee Member); Jill Barnholtz-Sloan PhD (Advisor) Subjects: Animal Sciences; Biostatistics; Neurosciences; Statistics

Master of Science, The Ohio State University, 2019, Industrial and Systems Engineering

Despite growing pressure for organizations to implement more science-based solutions into practice, efforts to successfully achieve this task have been known to fail due to the tensions that exist between science and application. While there has been a great push in the implementation science, translational science, evidence-based practice, and human factors literature, a void still remains regarding a framework that details the necessary requirements for bridging this known gap. In order to fill this void, I propose The IMPActS Framework, which is founded on the existing literature but acts as a new frame of reference for those trying to translate science into implementations. IMPActS proposes a new standard of what it means to make organizational “impact”, which is now defined as science-based solutions that maintain the maximum appropriate levels of scientific integrity while also being implementable and sustainable in real world practice. IMPActS also acts as an acronym for the five necessary factors each necessary but only jointly sufficient in making this successful definition of impact. These factors are Ideas, Model alignment, Pragmatics, Actors, and Sustainment, and can be thought of as the barriers to making impact that need to be overcome. In this paper, I will describe the IMPActS Framework in more detail and through the lens of three clinical cases, all of which deal with implementing clinical alarm interventions over the last 30 years. The purpose of introducing this framework and comparing it against real-world case studies is to highlight the barriers to making successful impact in hopes that the pathways to successful impact will become more salient, navigable, and tangible for all of those involved. Solution designers should use IMPActS as a means of assessing where to invest their future resources and efforts in order to overcome these barriers in practice.

Committee: Michael Rayo (Advisor); David Woods (Committee Member) Subjects: Industrial Engineering

Doctor of Philosophy, Case Western Reserve University, 2024, Systems Biology and Bioinformatics

The rapid advancement of personal wearable devices has allowed for the inception of novel applications of deep phenotyping for characterization of disease. The need to advance deep phenotyping and analysis methods for personalized wearable devices is crucial to the advancement of personalized remote patient monitoring. We developed an end-to-end digital health infrastructure designed for fast, secure, and effective patient recruitment, data collection, and analysis reporting. We analyzed the efficacy of patient recruitment through our end-to-end patient interface and found that recruitment methods from traditional means such as through clinical sources and university sources resulted in more consents ([0.015, 0.030]; p << 0.001) and more active patients initially (2 = 23.65; p < 0.005). Additionally, we noted that online recruitment through Facebook advertising and Google advertising produced a more ethnically diverse population compared to regional clinical recruitment (2 = 231.47; p < 0.001). We investigated the use of the previously reported NightSignal algorithm, originally developed for SARS-CoV-2 detection, on the detection of abnormal resting heart rate observations for cardiothoracic surgical patients collected through our infrastructure. We found The NightSignal algorithm had a sensitivity of 81%, a specificity of 75%, a negative predictive value of 97%, and a positive predictive value of 28% for the detection of postoperative events. When compared to patients who did not experience a postoperative event, patients who did experience a postoperative event had a significantly higher proportion of red alerts issued by the NightSignal algorithm during the first 30 days after surgery (0.325 vs. 0.063; p<0.05)]. Finally, we then investigated the potential for latent subgroup identification using physiological parameters generated from personal wearable devices. We found latent subgroups at 30-days, 60-days, and 90-days post-operatively. Each latent group was we (open full item for complete abstract)

Committee: Mark Cameron (Committee Chair); Jing Li (Committee Member); Wai Hong Wilson Tang (Committee Member); Xiao Li (Advisor) Subjects: Bioinformatics; Biomedical Research

Doctor of Philosophy, Case Western Reserve University, 2024, Clinical Translational Science

Artificial intelligence (AI) is a transformative force in healthcare, holding the potential to revolutionize patient care, diagnostics, treatment plans, and administration. The applications of AI in healthcare range from wearable devices with AI-powered algorithms monitoring vital signs to sophisticated clinical decision support tools used by healthcare professionals. For AI to have successful implementation in healthcare, there must be a deeper understanding of both the technical aspects of AI and the human dimension, considering the experiences, expectations, and needs of key stakeholders, including patients, physicians, and data scientists developing healthcare applications. The purpose of this research was to (1) characterize patient, physician, and data scientist perspectives and acceptance of the integration of artificial intelligence (AI) into healthcare (Section 2) and (2) use this qualitative data collected to understand acceptance and use of AI by key stakeholders using a system dynamics approach (Section 3). In the first part of this work, semi-structured individual interviews and focus groups were held with patients (n=23), primary care physicians (n=26), and data scientists (n=14) at the Cleveland Clinic. While recognizing the value of AI as a diagnostic aid, patients envisioned a collaborative approach where physicians retain the role of final decision-makers. Data scientists described systems they use to develop, implement, and maintain AI tools used in healthcare and agreed that a system for monitoring the performance of an AI tool post-implementation must be in place. Physicians placed specific emphasis on improving efficiencies and reducing their burden of work in providing care. All stakeholders regarded AI as a tool, not a replacement for the human aspect in the provision of care. As the second step of this work, data from the focus groups was then viewed with a system dynamics perspective to an existing frameworks of understanding technolog (open full item for complete abstract)

Committee: Mary Dolansky PhD, RN, FAAN (Committee Chair); Susannah Rose PhD (Advisor); Peter Hovmand PhD, MSW (Committee Member); Ronald Hickman PhD, RN, ACNP-BC, FAAN (Committee Member); Colin Drummond PhD, MBA (Committee Member) Subjects: Health Care; Medicine