Dartmouth Institute for Health Policy and Clinical Practice
Dartmouth Digital Commons (Dartmouth College)Not a member yet
8214 research outputs found
Sort by
A longitudinal observational study with ecological momentary assessment and deep learning to predict non-prescribed opioid use, treatment retention, and medication nonadherence among persons receiving medication treatment for opioid use disorder
Full text linkBackground Despite effective treatments for opioid use disorder (OUD), relapse and treatment drop-out diminish their efficacy, increasing the risks of adverse outcomes, including death. Predicting important outcomes, including non-prescribed opioid use (NPOU) and treatment discontinuation among persons receiving medications for OUD (MOUD) can provide a proactive approach to these challenges. Our study uses ecological momentary assessment (EMA) and deep learning to predict momentary NPOU, medication nonadherence, and treatment retention in MOUD patients. Methods Study participants included adults receiving MOUD at a large outpatient treatment program. We predicted NPOU (EMA-based), medication nonadherence (Electronic Health Record [EHR]- and EMA-based), and treatment retention (EHR-based) using context-sensitive EMAs (e.g., stress, pain, social setting). We used recurrent deep learning models with 7-day sliding windows to predict the next-day outcomes, using Area Under the ROC Curve (AUC) for assessment. We employed SHapley additive ExPlanations (SHAP) to understand feature latency and importance. Results Participants comprised 62 adults with 14,322 observations. Model performance varied across EMA subtypes and outcomes with AUCs spanning 0.58–0.97. Recent substance use was the best performing predictor for EMA-based NPOU (AUC = 0.97). Life-contextual factors were best performers for EMA-based medication nonadherence (AUC = 0.68) and retention (AUC = 0.89), and substance use risk factors (e.g., nicotine and alcohol use) and self-reported MOUD adherence performed best for predicting EHR-based medication nonadherence (AUC = 0.79). SHAP revealed varying latencies between predictors and outcomes. Conclusions Findings support the effectiveness of EMA and deep learning for forecasting actionable outcomes in persons receiving MOUD. These insights will enable the development of personalized dynamic risk profiles and just-in-time adaptive interventions (JITAIs) to mitigate high-risk OUD outcomes
The Fangs of Philosophy: Religious Experience as a Persuasive Device
Full text linkThe philosophical systems of Plato and Lucretius use the religious experience to compel readers into their distinct philosophical versions of reality. This compulsion is necessary for philosophy. This study’s three philosophical systems reappropriate this experience from the existing structures of religion to violently produce conviction in readers. The religious experience is not window dressing over the arguments of the philosophical systems, but a necessary component that reorients the audience into a new discursive reality. Ultimately, the philosophers we examine will show how the knowledge structure of philosophy becomes able to grip the audience, consuming them into its reality – how philosophy became fanged
Data-driven Dynamic Decision-making Using Discrete Optimization and Supervised Machine Learning
Full text linkIn recent years, the operations research community has developed data-driven optimization techniques to solve complex combinatorial problems with the aid of machine learning. This thesis contributes to these efforts by combining machine learning with optimization to expedite online decision-making, with applications in transportation and healthcare.
In the domain of airline operations recovery, the focus is on the aircraft recovery process—repairing disrupted schedules by minimizing overall disruption costs. Traditional exact methods are too time-consuming, while heuristic approaches often yield poor solution quality and lack generalizability across varying formulations. To address these challenges, this research employs supervised machine learning to identify near-optimal solution components by leveraging historical data. By integrating binary classification methods into a decision-aware framework, our approach prunes the decision space effectively, yielding high-quality solutions in significantly shorter runtimes than both exact and heuristic methods.
In the healthcare domain, for the diagnosis of occult hemorrhage, we develop a data-driven framework that takes advantage of multisensor data and vital signs to detect internal bleeding early, particularly under capacity constraints. We conduct extensive experiments with animal and human data to engineer high-fidelity features and maximize predictive accuracy. Building on these predictions, we propose a decision-aware machine learning approach that dynamically updates patient risk scores over time and optimizes admissions to resource-intensive care units. Our method balances the trade-off between acting early (with less accurate information) versus waiting for more precise data, thus reducing both false positives and missed diagnoses. Through experiments based on data sets from our preclinical study, we show that our dynamic optimization framework surpasses traditional risk-based heuristics, leading to significantly improved patient outcomes.
Through extensive computational experiments on realistic data sets based on our preclinical study, this thesis demonstrates that our integrated, decision-aware framework not only accelerates online decision-making but also consistently produces near-optimal solutions, offering significant improvements in both airline operations recovery and hemorrhage diagnosis
Pitch probability learning in auditory selective attention
No full textExtensive evidence suggests that previous experience guides visuospatial attention. For instance, studies of location probability learning demonstrate that repeatedly finding a search target in a particular region of the visual field produces attentional biases for that region (Jiang et al., 2013; Addleman, 2019). Similarly, Addleman (2019) demonstrated a comparable auditory effect whereby repeated location of an auditory target induced attentional preference for the specific region. However, there have been fewer explorations of whether the repeated selection of auditory non-spatial features, such as pitch, influences auditory selective attention. Can participants incidentally learn the probability structure of simple auditory features such as pitch, and use this experience to more effectively select more probable target sounds?
To investigate this question, we developed an auditory search paradigm where participants heard two sounds presented to both ears simultaneously and were tasked to report the location (left vs. right) of a target sound, which contains a brief gap. Unbeknownst to the participants, targets were disproportionately more likely to appear in one pitch (either low, medium, or high, counterbalanced across participants) than the other two. In a subsequent extinction phase, targets were equally likely to occur in all three pitches. We tested whether participants would be faster for frequent versus infrequent target sounds across both experimental phases. Our results showed that participants were faster at detecting that target sound that was associated with the frequent pitch, and this learning persisted, at least to some extent, into the extinction phase. However, we also observed that the magnitude of this probability learning was stimulus-dependent, such that higher pitched sounds exhibited the strongest learning benefit. This work suggests a critical role for stimulus salience in long-term probability learning, as stimulus salience may modulate the strength of probability learning, at least in auditory selective attention. Broadly, this study generalizes effects of probability learning in the visual domain to feature-based attention in the auditory domain, providing a first stepping stone to understanding how experience shapes selective attention across sensory domains.https://digitalcommons.dartmouth.edu/wetterhahn_2025/1000/thumbnail.jp
