Scholarly Works, Statistics

Permanent URI for this collection

https://hdl.handle.net/10919/24334

Research articles, presentations, and other scholarship

Browse

Now showing 1 - 20 of 278

The impact of imidacloprid on the subterranean survivorship of Laricobius (Coleoptera: Derodontidae), a biological control agent of Adelges tsugae (Hemiptera: Adelgidae)
Hillen, Ashleigh P.; Foley, Jeremiah R.; Gross, Aaron D.; Mayfield III, Albert E.; Williams, Jacob; Xia, Kang; Salom, Scott M. (Wiley, 2025-05-01)
The invasive hemlock woolly adelgid (HWA), Adelges tsugae (Annand) (Hemiptera: Adelgidae), has spread throughout most of the range of eastern hemlock, Tsuga canandensis (L.), and the entire range of Carolina hemlock, Tsuga caroliniana (Engelman), in the United States. Integrated pest management (IPM) of HWA combines chemical applications with the release of biological control agents on untreated trees within the same stand. Laricobius spp., Rosenhauer (Coleoptera: Derodontidae), have been used as biological control agents of HWA since 2003 and have subterranean and arboreal life phases that are synchronous with HWA's lifecycle. When utilizing IPM tactics, there is potential for Laricobius spp. to settle below an insecticide-treated tree for its subterranean phase. Field investigations assessed the impact of historical (five years post treatment in 2017) and recent imidacloprid soil treatments (via soil injection, soil drench, and tablet in November 2020) on the subterranean survivorship of Laricobius spp. by quantifying concentrations of imidacloprid and its metabolites to determine its potential impact on percent adult emergence from the soil. We observed a significant treatment effect on mean soil concentration among application methods at the recent treatment site, but not the historical treatment site. Additionally, at the more recently treated site, significantly lower mean percent emergence was observed from soil drench and tablet imidacloprid applications after one year (2021), but by the following year (2022), that effect was no longer present. This study supports recent recommendations to delay releases of Laricobius spp. for one-year post-treatment with imidacloprid. Furthermore, these data suggest when applying imidacloprid to a stand established with Laricobius spp., soil injection techniques pose the least risk to their subterranean survivorship.
The current landscape of academic statistical and data science collaboration units with examples
Sharp, Julia; Griffith, Emily H.; Craig, Bruce A.; Hanlon, Alexandra; Peskoe, Sarah; Van Mullekom, Jennifer (Wiley, 2024-09-01)
The delivery of academic statistical collaboration resources can vary among types of institutions and across time. In particular, this variation might occur in the management of infrastructure and the business model, the staffing model and opportunities for staff development. In this manuscript, we present examples of these three themes in modern academic statistical collaboration units and describe key advantages and challenges.
Treatment and recovery from opioid use disorder: The role of pain severity in individuals with moderate to severe pain
Tegge, Allison N.; Ferreira, Marco A. R.; Garafola, Peter M.; Xu, Shuangshuang; Farrell, Michael; Marsden, John; Lee, Ken; Le Moigne, Anne; Gray, Frank; Bickel, Warren K. (Elsevier, 2025-09-25)
Background: Pain is a frequent comorbidity among individuals with opioid use disorder (OUD), yet its impact on treatment outcomes is unclear. This study examined associations between pain severity and OUD treatment outcomes, including abstinence, craving, retention, and psychological functioning, in participants receiving longacting buprenorphine (BUP-XR). Methods: This secondary data analysis investigates participants from a BUP-XR phase 3 program: randomized clinical trial (NCT02357901; N = 192), open-label study (NCT02510014; N = 410); and a longitudinal observational follow-up (NCT03604861; N = 350). Pain was measured using the Brief Pain Inventory (BPI) at each treatment visit. Additional measures included demographics, opioid use, participant retention, opioid withdrawal, craving, depression, and quality of life. Analyses were performed on the full sample and the subgroup of individuals with moderate-to-severe pain (BPI≥4). Results: Participants averaged 40 years old, predominantly male (67%) and White (66%). Pain decreased after starting BUP-XR, and the reduction in pain continued throughout treatment (p-values<.001). For individuals with moderate-to-severe pain, greater concurrent pain severity was associated with lower abstinence rates (odds ratios: [0.801,0.852]; p-values<.001) in two datasets. Pain was not associated with participant retention. Lastly, greater pain severity was associated with worse physical quality of life (p-values<.001) and opioid withdrawal (pvalues<. 001), and greater depression (p-values<.001) and opioid craving (p-values<.001). Collectively, these findings are well replicated across three studies. Conclusions: Pain severity is a clinically relevant predictor of opioid use and psychosocial outcomes, but not treatment retention, in patients receiving BUP-XR. Routine pain severity monitoring may provide valuable insight into patient trajectories and support more tailored treatment approaches in OUD.
Poison Ivy (Toxicodendron radicans) Leaf Shape Variability: Why Plant Avoidance-By-Identification Recommendations Likely Do Not Substantially Reduce Poison Ivy Rash Incidence
Jelesko, John G.; Thompson, Kyla; Magerkorth, Noah; Verteramo, Elizabeth; Becker, Hannah; Flowers, Joy G.; Sachs, Jonathan; Datta, Jyotishka; Metzgar, Jordan (Wiley, 2023-09-28)
Avoidance of poison ivy plants is currently the primary approach to prevent the estimated 30–50 million annual poison ivy skin rash cases. The “leaves of three let it be” mnemonic device lacks specificity to differentiate poison ivy from other three-leaflet native plants. This report demonstrated that poison ivy leaves show marked total leaf shape variability that likely confounds accurate poison ivy plant identification, and significantly undermines a poison ivy avoidance strategy for mitigating poison ivy rash cases. Therefore, there is an ongoing need to develop prophylactic medical procedures to prevent poison ivy rash that do not depend on human poison ivy plant identification. Summary: Urushiol is the natural product produced by poison ivy (Toxicodendron radicans) that is responsible for millions of cases of delayed contact allergenic dermatitis in North America each year. Avoidance of poison ivy plant material is the clinically recommended strategy for preventing urushiol-induced dermatitis symptoms. However, poison ivy leaf shape is anecdotally notoriously variable, thereby confounding accurate poison ivy identification. This study focused on quantitative analyses of poison ivy and a common poison ivy look-alike (American hog peanut) leaf shape variability in North America to empirically validate the high degree of poison ivy leaf shape plasticity. Poison ivy and American hog peanut iNaturalist.org records were scored for seven attributes of compound leaf shape that were combined to produce a total leaf complexity score. Both the mean and the distribution of poison ivy total leaf complexity scores were significantly greater than that of American hog peanut. Non-metric multidimensional scaling analyses corroborated a high degree of poison ivy leaf shape variability relative to American hog peanut. A poison ivy accession producing frequent palmate penta-leaflet compound leaves was evaluated using linear regression modeling. Poison ivy total leaf complexity was exceedingly variable across a given latitude or longitude. With that said, there was a small but significant trend of poison ivy total leaf complexity increasing from east to west. Palmate penta-leaflet formation was significantly correlated with a stochastic leaflet deep-lobing developmental process in one unusual poison ivy accession. The empirically-validated poison ivy leaf shape hypervariability described in this report likely confounds accurate poison ivy identification, thereby likely resulting in many accidental urushiol-induced clinical allergenic dermatitis cases each year.
Are Vision LLMs Road-Ready? A Comprehensive Benchmark for Safety-Critical Driving Video Understanding
Zeng, Tong; Wu, Longfeng; Shi, Liang; Zhou, Dawei; Guo, Feng (ACM, 2025-08-03)
Vision Large Language Models (VLLMs) have demonstrated impressive capabilities in general visual tasks such as image captioning and visual question answering. However, their effectiveness in specialized, safety-critical domains like autonomous driving remains largely unexplored. Autonomous driving systems require sophisticated scene understanding in complex environments, yet existing multimodal benchmarks primarily focus on normal driving conditions, failing to adequately assess VLLMs’ performance in safety-critical scenarios. To address this, we introduce DVBench—a pioneering benchmark designed to evaluate the performance of VLLMs in understanding safety-critical driving videos. Built around a hierarchical ability taxonomy that aligns with widely adopted frameworks for describing driving scenarios used in assessing highly automated driving systems, DVBench features 10,000 multiple-choice questions with human-annotated ground-truth answers , enabling a comprehensive evaluation of VLLMs’ capabilities in perception and reasoning. Experiments on 14 state-of-the-art VLLMs, ranging from 0.5B to 72B parameters, reveal significant performance gaps, with no model achieving over 40% accuracy, highlighting critical limitations in understanding complex driving scenarios. To probe adaptability, we fine-tuned selected models using domain-specific data from DVBench, achieving accuracy gains ranging from 5.24 to 10.94 percentage points, with relative improvements of up to 43.59%. This improvement underscores the necessity of targeted adaptation to bridge the gap between generalpurpose vision-language models and mission-critical driving applications. DVBench establishes an essential evaluation framework and research roadmap for developing VLLMs that meet the safety and robustness requirements for real-world autonomous systems. We released the benchmark toolbox and the fine-tuned model at: https://github.com/tong-zeng/DVBench.git.
Learning Data Heterogeneity with Dirichlet Diffusion Trees
Huo, Shuning; Zhu, Hongxiao (MDPI, 2025-08-11)
Characterizing complex heterogeneous structures in high-dimensional data remains a significant challenge. Traditional approaches often rely on summary statistics such as histograms, skewness, or kurtosis, which—despite their simplicity—are insufficient for capturing nuanced patterns of heterogeneity. Motivated by a brain tumor study, we consider data in the form of point clouds, where each observation consists of a variable number of points. Our goal is to detect differences in the heterogeneity structures across distinct groups of observations. To this end, we employ the Dirichlet Diffusion Tree (DDT) to characterize the latent heterogeneity structure of each observation. We further extend the DDT framework by introducing a regression component that links covariates to the hyperparameters of the latent trees. We develop a Markov chain Monte Carlo algorithm for posterior inference, which alternatively updates the latent tree structures and the regression coefficients. The effectiveness of our proposed method is evaluated by a simulation study and a real-world application in brain tumor imaging.
Testing the Temperature-Mortality Nonparametric Function Change with an Application to Chicago Mortality
Mahmoud, Hamdy F. F. (MDPI, 2025-08-03)
The relationship between temperature and mortality is well-documented, yet most existing studies assume this relationship remains static over time. This study investigates whether the temperature-mortality association in Chicago from 1987 to 2000 has changed in shape or location of key features, such as change points. We apply nonparametric regression techniques to estimate the temperature-mortality functions for each year using daily mortality and temperature data from the National Morbidity, Mortality, and Air Pollution Study (NMMAPS) database. A permutation-based test is used to assess whether the shapes of these functions differ across time, while a bootstrap procedure evaluates the consistency of change points. Intensive simulation studies are conducted to evaluate the permutation-based test and bootstrap procedure based on Type I error and power. The proposed tests are compared with F tests in terms of Type I error and power. For the real data set, the analysis finds significant variation in the functional shapes across years, indicating evolving mortality responses to temperature. However, the estimated change points—temperatures associated with peak mortality—remain statistically consistent. These findings suggest that while the population’s overall vulnerability pattern may shift, the temperature threshold linked to maximum mortality has remained stable. This study contributes to understanding the temporal dynamics of climate-sensitive health outcomes and highlights the importance of flexible modeling in public health and climate adaptation planning.
Few-shot Learning over Graphs Using Topological Prompts
Goel, Jaidev; Chen, Yuzhou; Gel, Yulia (ACM, 2025-05-08)
Prompt-based fine-tuning of pre-trained models has recently emerged as a promising trend for few-shot learning over graphs. Despite its significant potential, high variability and sensitivity to noise and perturbations remain the major challenges on the way of a wider adoption of prompt-based fine-tuning. We propose a new solution to these open problems by introducing the machinery of persistent homology to graph prompts. In particular, to better guide the fine-tuning process on downstream tasks, we extract intrinsic topological descriptors of the activation graphs of the pre-trained models in a form of Fréchet Means and incorporate this inherent topological information into the prompt-tuning process. Additionally, we implement bootstrapping over the topological summaries to mitigate the high variability, typically observed in prompt-based methods. Our extensive validation shows that the new Topo-Prompt tool results not only in relative gains in node classification accuracy up to 11% but also in up to 4 times reduction of variability with respect to the state-of-the-art prompt tuning methods. Furthermore, Topo-Prompt delivers superior robustness to perturbations, outperforming its competitors up to 25% under noisy conditions.
Metrics of glycemic control but not body weight influence flavor nutrient conditioning in humans
Baugh, Mary Elizabeth; Ahrens, Monica L.; Burns, Amber K.; Sullivan, Rhianna M.; Valle, Abigail N.; Hanlon, Alexandra L.; DiFeliceantonio, Alexandra G. (Elsevier, 2025-07)
The modern food landscape, marked by a rising prevalence of highly refined, ultra-processed, and highly palatable foods, combined with genetic and environmental susceptibilities, is widely considered a key factor driving obesity at the population level. Gaining insight into the physiological and behavioral mechanisms that shape food preferences and choices is crucial for understanding obesity's development and informing prevention strategies. One factor influencing habitual eating patterns, which may impact body weight, is flavor-nutrient learning. Research suggests that post-oral signaling is diminished in both animals and humans with obesity, potentially affecting flavor-nutrient learning. By analyzing pooled data from two similar preliminary studies, we found that markers of glycemic control-specifically fasting glucose and HbA1C-rather than BMI, were negatively correlated with changes in flavor liking in our flavor-nutrient learning task. These findings contribute to the expanding body of research on flavor-nutrient learning and underscore the variability in individual responses to these paradigms. Obesity is increasingly recognized as a complex and heterogeneous condition with diverse underlying mechanisms. Together, our findings and existing evidence emphasize the importance of further investigating how phenotypic factors interact to shape food preferences and eating behaviors.
A Pilot Study Exploring Obesity-Related Differences in Fall Rate and Kinematic Response Resulting From a Laboratory-Induced Trip
Garman, Christina R.; Nussbaum, Maury A.; Franck, Christopher T.; Madigan, Michael L. (Taylor & Francis, 2016-08-15)
Background: Obese adults are reported to fall at a higher rate than non-obese adults. Purpose: To help determine the reason for this higher fall rate, we quantified fall rates, kinematics at trip onset, and kinematics during the response to a laboratory-induced trip among two groups of young adults with higher and lower body mass indexes (BMI) that approximated obese and healthy-weight ranges. Our focus was on young adults given that they comprise a substantial portion of the workforce. Methods: Twenty-one young adult subjects, including 10 with a lower BMI (19.4–25.7 kg/m2) and 11 with a higher BMI (29.8–42.9 kg/m2), walked along a 10 m walkway at a purposeful speed. During a randomly selected walking trial, an obstacle was raised to elicit a trip. Results: Among the 19 subjects who unambiguously fell or recovered, 30% of subjects with higher BMI fell and 0% of lower BMI subjects fell, but this difference did not reach statistical significance. Among the 15 subjects who used an elevating strategy, all recovered balance, and the only kinematic response variable that differed between BMI groups was that recovery step time was longer among the higher BMI group. Among the four subjects who used a lowering strategy, no statistical analysis was possible due to a small number of subjects, but several measures were consistent with a less favorable kinematic response among the three higher BMI fallers compared to the one lower BMI subject who recovered. Conclusions: This study provides preliminary evidence that obesity may adversely influence fall rate and recovery kinematics after tripping among young adults. Additional larger scale studies are needed to better understand contributing and modifiable factors that can be targeted via intervention.
A bootstrapping method to assess the influence of age, obesity, gender, and gait speed on probability of tripping as a function of obstacle height
Garman, Christina R.; Franck, Christopher T.; Nussbaum, Maury A.; Madigan, Michael L. (Elsevier, 2015-02-03)
Tripping is a common mechanism for inducing falls. The purpose of this study was to present a method that determines the probability of tripping over an unseen obstacle while avoiding the ambiguous situation wherein median minimum foot clearance (MFC) and MFC interquartile range concurrently increase or decrease, and determines how the probability of tripping varies with potential obstacle height. The method was used to investigate the effects of age, obesity, gender, and gait speed on the probability of tripping. MFC was measured while 80 participants walked along a 10-m walkway at self-selected and hurried gait speeds. The method was able to characterize the probability of tripping as a function of obstacle height, and identify effects of age, obesity, gender, and gait speed. More specifically, the probability of tripping was lower among older adults, higher among obese adults, higher among females, and higher at the slower self-selected speed. Many of these results were not found, or clear, from the more common approach on characterizing likelihood of tripping based on MFC measures of central tendency and variability.
Machine Learning–Based Prediction and Optimization of Balanced Mixture Design Performance Indices
Tong, Bilin; Huang, Wenjiang; Habbouche, Jhony; Boz, Ilker; Guo, Qing; Diefenderfer, Stacey D.; Flintsch, Gerardo W. (SAGE Publications, 2025-04-26)
The balanced mix design (BMD) concept is an emerging methodology that facilitates the design of engineered asphalt mixtures. This approach is particularly beneficial for mixtures containing conventional and high reclaimed asphalt pavement, for which the traditional volumetric design methods may fail to effectively address the performance characteristics. However, given production variability, these engineered mixtures can still fail to meet the required thresholds. Additionally, identifying the cause of this imbalance is challenging. To maximize the benefits of BMD implementation, this study introduces machine learning (ML) algorithms including linear regression (LR), random forest (RF), extreme gradient boosting (XGB), and support vector regression (SVR) as strategic tools to predict mixtures’ BMD performance indices. 648 specimens fabricated for quality acceptance as part of the 2020 Virginia Accelerated Pavement Testing Program is used for the modeling and analysis. The durability, cracking, and rutting susceptibility of the specimens were evaluated using the Cantabro test, the indirect tensile cracking test (IDT-CT), and the asphalt pavement analyzer (APA) rut test. Key outcomes include: a) ML models, including RF, XGB, and SVR, demonstrated superior performance compared with LR; b) feature importance analysis from ML models identified dominant factors for each BMD test, also highlighting the reheating process; and c) a pseudo in situ deployment was simulated to optimize BMD implementation. The dimensionality reduction analysis—uniform manifold approximation and projection—highlighted the practical challenges associated with concurrently improving multiple performance metrics. Ultimately, the pivotal role of ML in advancing both the design and production phases was emphasized.
Age Differences in the Required Coefficient of Friction During Level Walking Do Not Exist When Experimentally-Controlling Speed and Step Length
Anderson, Dennis E.; Franck, Christopher T.; Madigan, Michael L. (Human Kinetics, 2014-06-30)
The effects of gait speed and step length on the required coefficient of friction (COF) confound the investigation of age-related differences in required COF. The goals of this study were to investigate whether age differences in required COF during self-selected gait persist when experimentally-controlling speed and step length, and to determine the independent effects of speed and step length on required COF. Ten young and 10 older healthy adults performed gait trials under five gait conditions: self-selected, slow and fast speeds without controlling step length, and slow and fast speeds while controlling step length. During self-selected gait, older adults walked with shorter step lengths and exhibited a lower required COF. Older adults also exhibited a lower required COF when walking at a controlled speed without controlling step length. When both age groups walked with the same speed and step length, no age difference in required COF was found. Thus, speed and step length can have a large influence on studies investigating age-related differences in required COF. It was also found that speed and step length have independent and opposite effects on required COF, with step length having a strong positive effect on required COF, and speed having a weaker negative effect.
Comparison of Treadmill Trip-Like Training Versus Tai Chi to Improve Reactive Balance Among Independent Older Adult Residents of Senior Housing: A Pilot Controlled Trial
Aviles, Jessica; Allin, Leigh J.; Alexander, Neil B.; Van Mullekom, Jennifer; Nussbaum, Maury A.; Madigan, Michael L. (Oxford University Press, 2018-12-21)
Background: There is growing interest in using perturbation-based balance training to improve the reactive response to common perturbations (eg, tripping and slipping). The goal of this study was to compare the efficacy of treadmill-based reactive balance training versus Tai Chi performed at, and among independent residents of, older adult senior housing. Methods: Thirty-five residents from five senior housing facilities were allocated to either treadmill-based reactive balance training or Tai Chi training. Both interventions were performed three times per week for 4 weeks, with each session lasting approximately 30 minutes. A battery of balance tests was performed at baseline, and again 1 week, 1 month, 3 months, and 6 months post-training. The battery included six standard clinical tests of balance and mobility, and a test of reactive balance performance. Results: At baseline, no significant between-group differences were found for any balance tests. After training, reactive balance training participants had better reactive balance than Tai Chi participants. Maximum trunk angle was 13.5 smaller among reactive balance training participants 1 week after training (p =. 01), and a reactive balance rating was 24%-31% higher among reactive balance training participants 1 week to 6 months after training (p <. 03). Clinical tests showed minimal differences between groups at any time point after training. Conclusion: Trip-like reactive balance training performed at senior housing facilities resulted in better rapid balance responses compared with Tai Chi training.
Comparing Trunk Kinematics Computed by Optical Marker-Based Motion Capture System and Inertial Measurement Units During Overground Trips
Lee, Youngjae; Alexander, Neil B.; Franck, Christopher T.; Madigan, Michael L. (SAGE Publications, 2023-10-25)
Falls are the most common cause of non-fatal injuries, and trips are responsible for high percentages of those falls in the United States. Traditional method for estimating trunk kinematics during overground trips uses optical marker-based motion capture systems. However, their cost and space requirements can often be barriers in this research field. Inexpensive and portable inertial measurement units may be an appropriate alternative. This study compared trunk flexion angle and angular velocity at touchdown of the initial recovery step after laboratory-induced trips while walking captured by the optical markerbased motion capture system versus IMUs. Our results provide evidence that a sternum-worn IMU can provide trunk kinematic measurements of clinical relevance and may be used to provide meaningful data to understand kinematic responses to trips or trip-induced falls that occur in real life.
Approximate Bayesian Techniques for Statistical Model Selection and Quantifying Model Uncertainty-Application to a Gait Study
Franck, Christopher T.; Arena, Sara L.; Madigan, Michael L. (Springer, 2022-08-20)
Frequently, biomedical researchers need to choose between multiple candidate statistical models. Several techniques exist to facilitate statistical model selection including adjusted R2, hypothesis testing and p-values, and information criteria among others. One particularly useful approach that has been slow to permeate the biomedical literature is the notion of posterior model probabilities. A major advantage of posterior model probabilities is that they quantify uncertainty in model selection by providing a direct, probabilistic comparison among competing models as to which is the “true” model that generated the observed data. Additionally, posterior model probabilities can be used to compute posterior inclusion probabilities which quantify the probability that individual predictors in a model are associated with the outcome in the context of all models considered given the observed data. Posterior model probabilities are typically derived from Bayesian statistical approaches which require specialized training to implement, but in this paper we describe an easy-to-compute version of posterior model probabilities and inclusion probabilities that rely on the readily-available Bayesian information criterion. We illustrate the utility of posterior model probabilities and inclusion probabilities by re-analyzing data from a published gait study investigating factors that predict required coefficient of friction between the shoe sole and floor while walking.
Inertial measurement units worn on the dorsum of the foot and proximal to the ankle can provide valid slip recovery measures
Morris, Michelle A.; Franck, Christopher T.; Madigan, Michael L. (Elsevier, 2024-11-03)
Background: Slips are a leading cause of injury among older adults. Slip recovery measures are often captured using optoelectronic motion capture (OMC) systems that can be costly and typically require a laboratory setting. Inertial measurement unit (IMU) systems show promise as a lower cost, portable, and wearable form of motion capture. Question: Can IMUs worn on the dorsum of the feet and proximal to the ankles be used to capture valid slip recovery measures? Methods: Thirty older adults (ages 65–80; 18 females) were exposed to a laboratory slip while wearing OMC markers, IMUs on the dorsum of the feet, and IMUs proximal to the ankles. To evaluate the concurrent validity of IMU-based slip recovery measures using the OMC-based measures as our standard, we determined whether the IMU-based slip recovery measures differed between falls and recoveries, and evaluated the strength of correlation between IMU-based measures and OMC. We also defined the difference between foot IMU-based and OMC-based slip recovery measures to be the system offset, and compared the system offset variance between participant-placed IMUs and researcher-placed IMUs. Results: All IMU-based and OMC-based slip recovery measures differed between falls and recoveries (p ≤ 0.008), and all IMU-based measures exhibited strong correlation (r ≥ 0.94) with OMC-based measures. The system offset variance was larger when foot IMUs were participant-placed than when researcher-placed for anterior-posterior slip distance (p = 0.032), but not other slip recovery measures (p ≥ 0.054). Significance: IMUs worn on the dorsum of the feet and proximal to the ankle can provide valid slip recovery measures in a laboratory setting. This includes IMUs placed by participants on the dorsum of the feet that might be needed for the long-term monitoring of these measures by participants outside the laboratory setting.
Low responsiveness of machine learning models to critical or deteriorating health conditions
Pias, Tanmoy Sarkar; Afrose, Sharmin; Tuli, Moon Das; Trisha, Ipsita Hamid; Deng, Xinwei; Nemeroff, Charles B.; Yao, Danfeng Daphne (Springer Nature, 2025-03-11)
Background: Machine learning (ML) based mortality prediction models can be immensely useful in intensive care units. Such a model should generate warnings to alert physicians when a patient’s condition rapidly deteriorates, or their vitals are in highly abnormal ranges. Before clinical deployment, it is important to comprehensively assess a model’s ability to recognize critical patient conditions. Methods: We develop multiple medical ML testing approaches, including a gradient ascent method and neural activation map. We systematically assess these machine learning models’ ability to respond to serious medical conditions using additional test cases, some of which are time series. Guided by medical doctors, our evaluation involves multiple machine learning models, resampling techniques, and four datasets for two clinical prediction tasks. Results: We identify serious deficiencies in the models’ responsiveness, with the models being unable to recognize severely impaired medical conditions or rapidly deteriorating health. For in-hospital mortality prediction, the models tested using our synthesized cases fail to recognize 66% of the injuries. In some instances, the models fail to generate adequate mortality risk scores for all test cases. Our study identifies similar kinds of deficiencies in the responsiveness of 5-year breast and lung cancer prediction models. Conclusions: Using generated test cases, we find that statistical machine-learning models trained solely from patient data are grossly insufficient and have many dangerous blind spots. Most of the ML models tested fail to respond adequately to critically ill patients. How to incorporate medical knowledge into clinical machine learning models is an important future research direction.
A framework for developing a real-time lake phytoplankton forecasting system to support water quality management in the face of global change
Carey, Cayelan C.; Calder, Ryan S. D.; Figueiredo, Renato J.; Gramacy, Robert B.; Lofton, Mary E.; Schreiber, Madeline E.; Thomas, R. Quinn (Springer, 2024-09-20)
Phytoplankton blooms create harmful toxins, scums, and taste and odor compounds and thus pose a major risk to drinking water safety. Climate and land use change are increasing the frequency and severity of blooms, motivating the development of new approaches for preemptive, rather than reactive, water management. While several real-time phytoplankton forecasts have been developed to date, none are both automated and quantify uncertainty in their predictions, which is critical for manager use. In response to this need, we outline a framework for developing the first automated, real-time lake phytoplankton forecasting system that quantifies uncertainty, thereby enabling managers to adapt operations and mitigate blooms. Implementation of this system calls for new, integrated ecosystem and statistical models; automated cyberinfrastructure; effective decision support tools; and training for forecasters and decision makers. We provide a research agenda for the creation of this system, as well as recommendations for developing real-time phytoplankton forecasts to support management.
Augmenting a Simulation Campaign for Hybrid Computer Model and Field Data Experiments
Koermer, Scott; Loda, Justin; Noble, Aaron; Gramacy, Robert B. (Taylor & Francis, 2024-05-24)
The Kennedy and O’Hagan (KOH) calibration framework uses coupled Gaussian processes (GPs) to meta-model an expensive simulator (first GP), tune its “knobs” (calibration inputs) to best match observations from a real physical/field experiment and correct for any modeling bias (second GP) when predicting under new field conditions (design inputs). There are well-established methods for placement of design inputs for data-efficient planning of a simulation campaign in isolation, that is, without field data: space-filling, or via criterion like minimum integrated mean-squared prediction error (IMSPE). Analogues within the coupled GP KOH framework are mostly absent from the literature. Here we derive a closed form IMSPE criterion for sequentially acquiring new simulator data for KOH. We illustrate how acquisitions space-fill in design space, but concentrate in calibration space. Closed form IMSPE precipitates a closed-form gradient for efficient numerical optimization. We demonstrate that our KOH-IMSPE strategy leads to a more efficient simulation campaign on benchmark problems, and conclude with a showcase on an application to equilibrium concentrations of rare earth elements for a liquid–liquid extraction reaction.

Browse

Recent Submissions