A high-dimensional incomplete-modality transfer learning (HD-IMTL) method for early prediction of Alzheimer’s disease
Dohyun Ku, Georgia Institute of Technology
Prediction of Alzheimer’s disease (AD) risk for individuals with mild cognitive impairment (MCI) allows early intervention. Neuroimaging has shown promise, but not every patient has all the modalities due to the cost and accessibility. We previously developed an incomplete-modality transfer learning (IMTL) model and extended the capacity of IMTL to handle high-dimensional feature sets, namely, HD-IMTL.
Our dataset included 1319 T1-MRI scans from MCI patients; 1002 had FDG-PET and 612 had amyloid-PET. 156 regional volumetric and thickness features were from MRI and 83 and 83 regional SUVR features from FDG-PET and amyloid-PET, respectively. The goal of HD-IMTL was to jointly train 4 ML models to predict MCI conversion to AD in 36 months, with each model based on a certain combination of available modalities, MRI, MRI+FDG, MRI+amyloid, and MRI+FDG+amyloid. We employed feature screening to remove uninformative features, performed modality-wise partial least squares (PLS) to condense remaining features, and used correlation tests to select PLS components. IMTL, a generative model using expectation-maximization for joint parameter estimation was used to facilitate transfer learning. Synthetic Minority Over-sampling Technique (SMOTE) was used to account for sample imbalance, and 20 training/test splits were repeated. For comparison, three existing ML models for incomplete-modality fusion were applied to the same dataset.
The AUCs by HD-IMTL were 0.802, 0.840, 0.868, and 0.880 for modality combinations. The AUCs by existing methods were lower, with ranges of 0.749-0.793, 0.769-0.826, 0.816-0.863, and 0.832-0.868. HD-IMTL demonstrated high accuracy in predicting MCI conversion to AD for patients with varying access/availability of imaging modalities.
Zhiyang Zheng, Ph.D. student, Georgia Institute of Technology
Lingchao Mao, Ph.D. student, Georgia Institute of Technology
Ruiqi Chen, Ph.D. student, Georgia Institute of Technology
Yi Su, Ph.D., Banner Alzheimer’s Institute
Kewei Chen, Ph.D., Banner Alzheimer’s Institute
David Weidman, MD, Banner Alzheimer’s Institute
Teresa Wu, Ph.D., Arizona State University
ShihChung Lo, Ph.D., MS Technologies Corp.
Fleming Lure, Ph.D., MS Technologies Corp.
Jing Li, Ph.D., Georgia Institute of Technology
A Preference-sensitive Approach to Optimal Screening Problems with Multiple Objectives: Application to Breast Cancer Screening
Sun Ju Lee, H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology
Sequential screening and treatment problems in medical decision-making often involve multiple objectives that must be balanced to find the optimal policy with a Markov Decision Process (MDP). We propose a lexicographic ordering method with tolerance to solve a multi-objective MDP and apply it to a disease screening problem. We show that our approach can facilitate a personalized and patient-centered decision-making process better aligned with the priorities of individual patients.
Gian-Gabriel Garcia, Assistant Professor, H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology
Adherence to Clinical Practice Guidelines and FDA Approved Psychotropic Medication to Treat Medicaid-insured Children Diagnosed with ADHD
Daniel Kim, H. Milton Stewart School of Industrial & Systems Engineering
Objective: This study examined adherence to clinical practice guidelines and the U.S. Food and Drug Administration (FDA) medication labels when prescribing psychotropic medications to Medicaid-insured children with attention deficit hyperactivity disorder (ADHD).
Methods: A retrospective analysis was conducted using 2015 – 2018 Medicaid claims data for children diagnosed with ADHD diagnosis and at least one psychotropic medication in nine southern U.S. states (total N=606,881). Medications recommended for pediatric use were identified from clinical practice guidelines by the American Academy of Child and Adolescent Psychiatry and the University of South Florida.
Results: In the nine states, 90.6% of children received recommended and FDA-approved medications for ADHD treatment and its comorbidities. Antidepressants, antihistamines, and atypical antipsychotics were prescribed without a documented diagnosis for 17.1% of children.
Conclusions: High compliance with clinical practice guidelines and FDA medication labels for ADHD prescribed medication was observed, suggesting the effectiveness of guidelines in driving evidence-based care.
Steven P. Cuffe, M.D., Department of Psychiatry, University of Florida College of Medicine
Michael W. Naylor, M.D., Department of Psychiatry, University of Illinois Chicago College of Medicine
Pinar Keskinocak, Ph.D., H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology
Nicoleta Serban, Ph.D., H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology
AI Frameworks for Data-centric Epidemic Forecasting
Arthi Rao, Georgia Tech (Atlanta, GA)
The recent COVID-19 pandemic has reinforced the importance of epidemic forecasting to equip decision makers in multiple domains, ranging from public health to economics. However, forecasting the epidemic progression remains a non-trivial task as the spread of diseases is subject to multiple confounding factors spanning human behavior, pathogen dynamics and environmental conditions, etc. Research interest has been fueled by the increased availability of rich data sources capturing previously unseen facets of the epidemic spread and initiatives from government public health and funding agencies like forecasting challenges and funding calls. This has resulted in recent works covering many aspects of epidemic forecasting. Data-centered solutions have specifically shown potential by leveraging non-traditional data sources as well as recent innovations in AI and machine learning. his poster exhibits various data-driven methodological and practical advancements. We discuss methods and modeling paradigms with a focus on the recent data-driven statistical and deep-learning based methods as well as novel class of hybrid models that combine domain knowledge of mechanistic models with the effectiveness and flexibility of statistical approaches. We also discuss experiences and challenges that arise in real-world deployment of these forecasting systems including decision-making informed by forecasts.
Alexander Rodríguez
Harsha Kamarthi
Jaiming Cui
B. Aditya Prakash
An integrated approach to creating healthy communities
Arthi Rao, Georgia Tech (Atlanta, GA)
The Center for Quality Growth and Regional Development (CQGRD) is a research center affiliated with the College of Design. CQGRD has expertise in planning healthy communities utilizing systems frameworks such as socioecological theory and health in all policies approaches. Dr. Arthi Rao is the current Interim Director with an interdisciplinary background in Urban Planning, Epidemiology and Spatial Analytics. This poster will highlight CQGRD’s research, teaching, and practice areas to promote the creation of healthy communities.
Automated Extraction of Social Determinants of Health (SDoH) from Electronic Health Records
Andy Stevens, Georgia Tech Research Institute
Social determinants of health (SDoH) data are important data points for both patient-level and population-level analyses. Unfortunately, this information is often unavailable due to limited presence in structured data, with most information residing in unstructured data, which require more advanced techniques to analyze. Most modern EHR systems expose their data as Fast Healthcare Interoperability Resources (FHIR) resources, with many EHRs also supporting the ability to write data back to the EHR using this same mechanism. Our work created a FHIR-based prototype for automated extraction of SDoH from EHRs, with support for writing this information back to the EHR as structured observations.To extract SDoH information from clinical notes, we developed custom SDoH modules for ClarityNLP, an open-source clinical NLP system developed by a team at Georgia Tech. These modules were leveraged to create a proof-of-concept application that can retrieve clinical notes from a FHIR-enabled server, run them through ClarityNLP, and translate the NLP findings into structured codes that conform to the United States Core Data for Interoperability (USCDI) guidelines. The application uses the SMARTonFHIR authentication and authorization protocol and supports a connection to any EHR system that exposes their data via FHIR using this framework. The developed application automatically retrieves clinical notes via FHIR, extracts SDoH information on various SDoH elements, and writes this information back to the EHR in structured form. Future work includes integrating with a real-world EHR to assess NLP performance, as well as extending the algorithms to other SDoH, including food insecurity, interpersonal violence, and other SDoH.
Causal Graph Discovery from Self and Mutually Exciting Time Series
Song Wei, Georgia Tech
We present a generalized linear structural causal model, coupled with a novel data-adaptive linear regularization, to recover causal directed acyclic graphs (DAGs) from time series. By leveraging a recently developed stochastic monotone Variational Inequality (VI) formulation, we cast the causal discovery problem as a general convex optimization. Furthermore, we develop a non-asymptotic recovery guarantee and quantifiable uncertainty by solving a linear program to establish confidence intervals for a wide range of non-linear monotone link functions. We validate our theoretical results and show the competitive performance of our method via extensive numerical experiments. Most importantly, we demonstrate the effectiveness of our approach in recovering highly interpretable causal DAGs over Sepsis Associated Derangements (SADs) while achieving comparable prediction performance to powerful “black-box” models such as XGBoost. Thus, the future adoption of our proposed method to conduct continuous surveillance of high-risk patients by clinicians is much more likely.
Yao Xie, Georgia Tech
Christopher S. Josef and Rishikesan Kamaleswaran, Emory university
Data-Driven Counterfactual Optimization for Personalized Medical Decision Making
Chey-Yi Liao, H. Milton Stewart School of Industrial & Systems Engineering
Machine learning-based health tools are becoming popular for informing treatment targets for high-risk patients with chronic diseases. However, using these tools alone, it is challenging to identify personalized treatment targets that lower the risks of adverse outcomes to a clinically acceptable range and are realistic, actionable, and robust to changes in these tools. To this end, we propose a data-driven approach called Distributionally Robust Selection of Clinical Role Models via Counterfactual Optimization (DISC²O). With a case study on 5-year cardiovascular disease, we show that our model can recommend treatment targets with much lower risks of adverse health outcomes and slightly higher costs, compared to clinical practice recorded in that dataset and other benchmarks including the clinical practice that does not consider uncertainty in these health tools.
Esmaeil Keyvanshokooh; Gian-Gabriel Garcia
Designing Accurate Concussion Assessments under Time Constraints
Himadri Pandey, H. Milton Stewart School of Industrial & Systems Engineering
Concussion is a common type of brain injury in sports and recreation, with 1.6-4 million cases yearly, half of which go unreported or undetected. Prompt and precise diagnosis is vital to manage the injury and reduce the risk of severe short-term and long-term effects, such as cognitive impairment, depression, and eurodegenerative disorders. However, identifying appropriate concussion assessment tools in time-constrained settings like athletic competitions is still challenging. We are proposing a novel approach to optimize the accuracy of a concussion assessment battery under time constraints. Specifically, we formulate the problem as a mixed-integer optimization problem with chance constraints. Our findings demonstrate that we can create precise assessment batteries even for short timeframes, resulting in better outcomes in time-constrained settings.
Gian-Gabriel P. Garcia, Landon Lempke
Enhancing Opioid Research with Feature Engineering Strategies
Guantao Zhao, Georgia Tech
Recent research indicates significant differences in the use and dosing of opioid medications in the management of pain in adults with traumatic injuries and pediatric patients following cardiac surgery. The complexity of these differences arises from multiple factors, including clinical data, prescription information, patient demographics, and social environment. Although studies on different types of opioid medications may involve diverse study populations, the data preprocessing procedures are typically similar. Therefore, our goal is to summarize our past experiences and findings related to opioid medications and establish a standardized data preprocessing process, aiming to provide more accurate and comprehensive data for downsteam analyses. This comprehensive data preprocessing approach will also contribute to a better understanding of the differences in opioid medication utilization. It will, in turn, offer more targeted guidance for future research and clinical practices, ultimately enhancing the quality of pain management and reducing potential biases and associated risks.
Manvitha Kalicheti, Georgia Institute of Technology
Nicoleta Serban, Georgia Institute of Technology
Exploring Educational Equity: A Machine Learning Approach to Unravel Achievement Disparities in Georgia
Yichen Ma, Georgia Tech ISYE
The COVID-19 pandemic has significantly exacerbated existing educational disparities in Georgia’s K-12 system, particularly in terms of racial and ethnic achievement gaps. Utilizing machine learning methods, the study conducts a comprehensive analysis of student achievement rates across different demographics, regions, and subjects. The findings highlight a significant decline in proficiency in English and Math during the pandemic, with a noticeable contraction in score distribution and a greater impact on economically disadvantaged and Black students. Socio-economic status, as represented by the Directly Certified Percentage – the percentage of students eligible for free lunch, emerges as the most crucial factor, with additional insights drawn from faculty resources such as teacher salaries and expenditure on instruction. The study also identifies disparities in achievement rates between urban and rural settings, as well as variations across counties, underscoring the influence of geographical and socio-economic factors. The data suggests that targeted interventions and resource allocation, particularly in schools with higher percentages of economically disadvantaged students, are essential for mitigating educational disparities.
Dr. Dima Nazzal, Principal Academic Professional, Georgia Tech ISYE
For RT-PCR Hospitalized Adults With COVID-19, Actionable Results Are Associated With Improved Outcomes
Qunzhi Xu, H. Milton Stewart School of Industrial & Systems Engineering
During the COVID-19 pandemic, test result delays occurred due to staff and supply shortages and off-site testing. The impact of delays is not documented; this study aimed to determine the impact of rapid test results. The results showed that rapid testing was associated with fewer adverse outcomes in COVID-19 IPs ≥ 50yrs.
D.M.Wolk, Geisinger, Danville, PA
D.J.Novak, Geisinger, Danville, PA
A.M.Tice, Geisinger, Danville, PA
A.M.Styer, Geisinger, Danville, PA
L.Olsho, Abt Associates, Cambridge, MA
A. Envers, Abt Associates, Cambridge, MA
J. Burns, Abt Associates, Rockville, MD
Y. Mei, Georgia Inst. of Technology, Atlanta, GA
Glucose Variability as a Predictor of Outcomes for Stroke Patients
Paul Horton, Georgia Institute of Technology
Glycemic control is a known predictor of patient outcome following acute ischemic stroke. However, the Stroke Hyperglycemia Insulin Network Effort (SHINE) randomized controlled clinical trial did not detect a difference in patient outcomes based on the intensity of glucose management during the acute post-stroke phase. It is unknown if other parameters of glycemic control, such as glucose variability, should be considered when a glycemic goal is set. Within, we perform a secondary analysis of the SHINE trial data.
We identified higher glucose variability over 72 hours initiated after stroke admission was associated with a reduced risk of poor outcome when analyzed across the entire SHINE cohort and by its subgroups (treatment group and HbA1c). However, further analysis of the first 8 hours suggested that patients with poor outcomes had a faster correction of glucose (-8.9 and -6.7mg/dL/hr for the unfavorable versus favorable one, respectively, p6), yet not in the standard of care nor in the normal HbA1c sub-groups.
Dr. Ofer Sadan, Dr. Yajun Mei
Interdisciplinary Collaborative Efforts to Advance Mental Health System
Pricilla Zhang, Health Analytics Lab
The modern mental health landscape faces unprecedented challenges, necessitating a dynamic and collaborative approach for meaningful progress. Our project introduces an interdisciplinary initiative aimed at advancing mental health systems through the synergistic efforts of data processing, analytical insights, and system modeling.
In an era of rapidly evolving healthcare paradigms, our objective is to foster collaboration among diverse fields, including mental health, data science, and engineering. By pooling resources and expertise, we aspire to develop innovative solutions that can be effectively distributed and continually improved to address the complex and multifaceted nature of mental health.
Our approach begins with the harnessing of comprehensive datasets, encompassing clinical records, patient-reported information, and socio-environmental factors. Leveraging cutting-edge data processing techniques, we aim to extract valuable insights from this wealth of information. These insights will inform and optimize the delivery of mental health services, enhancing patient outcomes and resource allocation.
Furthermore, system modeling will be employed to simulate and predict the impact of various interventions and policies. This modeling framework will provide a platform for testing and refining innovative strategies in mental health care.
Through this interdisciplinary collaboration, we seek to catalyze transformative change in mental health systems, improving the efficiency of the mental health system. By working together, we can create a more compassionate, data-driven, and responsive mental health ecosystem, ultimately improving the lives of those affected by mental health challenges.
Yujia Xie, Georgia Institute of Technology
Modeling and Mitigating Food Under-Served Areas: The Food Distributor and Policy-Maker Problems
Sofia Perez-Guzman, Georgia Institute of Technology
This study introduces innovative mathematical models to understand the influence of market-related factors on the existence of food under-served areas (FUAs), often known as food deserts, and aiding policymakers in their efforts to address this issue. The communities in these regions, predominantly low-income populations, have limited access to healthy food options. The initial model, the Food Distributor Problem (FDP), delves into FUAs’ formation by considering the economics of food distribution and demand, measured by order sizes. The FDP is an integer optimization program based on the Prize Collecting Traveling Salesman Problem with Profits, mimicking the decision-making process of food distributors when deciding whether to serve a particular food outlet. The study examines the impact of supply and demand factors through numerical experiments and a concept case. It reveals that the absence of financial incentives for food distributors to serve small food outlets significantly contributes to the persistence of FUAs, highlighting the importance of addressing supply and demand issues in combating the problem. The second model, the Policy-Maker Problem (PMP), is a work in progress. It aims to capture the decision-making behavior of policymakers striving to strike the right balance between community benefits and public-sector investments to alleviate FUAs. These investments support initiatives enhancing food distribution to small food outlets. Integrating the PMP with the FDP is under exploration to depict the interconnection between decisions in the food distribution process and the implementation of strategies to mitigate FUAs.
Jose Holguin-Veras, William H. Hart Professor, Rensselaer Polytechnic Institute
Modeling Relaxed Policies for Discontinuation of Methicillin Resistant Staphylococcus aureus Contact Precautions
Jiaming Cui, Georgia Institute of Technology
Objective: In this work, we assess the spread dynamics of MRSA and economic costs of reducing University of Virginia (UVA) hospital’s present “3-negative” policy (discontinuing the MRSA contact precautions for patients with three consecutive negative test results) to either two or one negative test policies.
Design: Cost-effective analysis.
Settings: Medical center
Patients: 41,216 patients in UVA hospital from 2015 to 2019.
Methods: We extend the existing 2-Mode-SIS model to simulate MRSA transmission in the UVA hospital, accounting for both environmental contamination and interactions between patients and providers. Additionally, we introduce a parameter adjustment method that utilizes calibrated parameters for the 3-negative policy to estimate the outcomes for 2-negative and 1-negative policies.
Results: Our findings indicate that the 1-negative policy has statistically significantly lower costs ($346,030 (95% CI: 285,560-409,909) annually, p<0.001) than the 2-negative ($357,776 (95% CI: 296,691-419,704)) and 3-negative ($358,923 (95% CI: 297,435-427,405)) policies.
Conclusions: Our study indicates that a single negative MRSA nares PCR test may provide sufficient evidence to discontinue MRSA contact precautions, and that the 1-negative policy may be the most cost-effective option.
Jiaming Cui (Georgia Institute of Technology)
Jack Heavey (University of Virginia)
Leo Lin (University of Virginia)
Eili Y. Klein (Johns Hopkins University)
Gregory R. Madden (University of Virginia)
Costi D. Sifri (University of Virginia)
Bryan Lewis (University of Virginia)
Anil K. Vullikanti (University of Virginia)
B. Aditya Prakash (Georgia Institute of Technology)
Modeling the Spread of Circulating Vaccine-Derived Poliovirus Type 2 Outbreaks and Interventions
Yuming Sun, Georgia Institute of Technology
Poliomyelitis (polio) is an infectious disease that paralyzed millions of people worldwide before polio vaccines were available. Despite the successes of the Global Polio Eradication Initiative, there are still circulating vaccine-derived poliovirus type 2 (cVDPV2) outbreaks which require an updated understanding of how factors such as outbreak responses and the existence of under-vaccinated areas affect the disease spread. We built a general differential-equation-based model to simulate polio outbreaks and interventions. In a case study of northern Nigeria, we fit the model using data on the reported cVDPV2 paralytic cases in 2018 ‒ 2021. By predicting cVDPV2 outbreaks in 2022 ‒ 2023 in northern Nigeria using the validated model, we tested the impact of different outbreak responses that varied in the number of vaccination rounds, the target geographical areas, and the start days. We compared the outcomes of outbreak responses in terms of the total number of cVDPV2 paralytic cases and the time when transmission stops in 2022 – 2023. Our findings highlighted that to stop the outbreaks, northern Nigeria would need at least 1 to 2 more vaccination rounds that covered all areas compared to the currently planned response. Stakeholders should achieve high coverage with improved access to under-vaccinated population areas to avoid persistent transmission which largely delayed successful outbreak interruption. Also, outbreak responses with shorter delays between detecting transmission and starting vaccinations should be implemented since they averted more paralytic cases and required fewer vaccination efforts to stop the outbreaks.
Pinar Keskinocak, Ph.D., Georgia Institute of Technology
Lauren N. Steimle, Ph.D., Georgia Institute of Technology
Stephanie Kovacs, Ph.D., Centers for Disease Control and Prevention
Steven Wassilak, M.D., Centers for Disease Control and Prevention
Modeling to Inform Strategies for Disease Eradication: The Case of Guinea Worm Disease
Hannah Smalley, Georgia Institute of Technology
In 1986, there were approximately 3.5 million cases of Guinea worm disease around the world. Since then, Guinea worm has been eliminated in 17 countries, and in 2022, there were only 13 human cases globally. However, in the same year, 521 dog infections were reported in Chad, the country with the highest incidence by far. Because the disease is spread through shared water sources, the human population is at risk. We built an agent-based simulation to model the transmission of Guinea Worm disease among dogs and shared water sources in Chad. We analyzed intervention strategies employed to reduce transmission, namely, tethering of dogs with emerging worms, and application of Abate® larvicide to reduce the number of infective copepods in the water sources. We also considered the potential benefit from a proposed diagnostic test capable of identifying infected dogs before signs of infection arise. Our results show that tethering and Abate levels need to increase above historical levels to achieve Guinea worm elimination in Chad. Targeting limited resources can lead to faster elimination. When used in conjunction with tethering and Abate, a diagnostic test could be helpful and support elimination efforts but the implementation strategy and level of education about tethering compliance following a positive test result are critical factors.
Pinar Keskinocak, Associate Chair for Faculty Development, William W. George Chair, Professor, Georgia Institute of Technology
Julie Swann, A. Doug Allison Distinguished Professor and Department Head, North Carolina State University
Chris Hanna, Global Project Partners, LLC
Adam Weiss, Director, Guinea Worm Eradication Program, The Carter Center
Tyler Perini, Georgia Institute of Technology
Yifan Wang, Georgia Institute of Technology
Multidimensional Hardships during the Covid-19 Pandemic in the United States
Shatakshee Dhongde, Georgia Institute of Technology
In this paper, we measure multidimensional hardships experienced by Americans during the Covid-19 pandemic. We use the Federal Reserve Board’s Survey of Household Economics and Decision-making (SHED) to compile data on self-reported economic hardships such as inability to pay bills, unemployment, and inability to afford health care. We find that during the pandemic, 28 percent or almost 1 in 3 adults experienced multiple hardships. We measure inter temporal movement in and out of hardships by building a panel within SHED with nearly 900 respondents between 2018 and 2021. Our panel regression estimates show that multidimensional hardships were particularly high women and among Blacks and Hispanics. Our results underscore the fact that the pandemic compounded hardships experienced by Americans and left a long-lasting impact on their well-being.
Roshani Bulkunde
Erdal Asker
Nonnegative Tensor Completion with Healthcare Applications
Caleb Bugg, Georgia Tech ISyE
Unlike matrix completion, tensor completion does not have an algorithm that is known to achieve the information-theoretic sample complexity rate. This paper develops a new algorithm for the special case of completion for nonnegative tensors. We prove that our algorithm converges in a linear (in numerical tolerance) number of oracle steps, while achieving the information-theoretic rate. Our approach is to define a new norm for nonnegative tensors using the gauge of a particular 0-1 polytope; integer linear programming can, in turn, be used to solve linear separation problems over this polytope. We combine this insight with a variant of the Frank-Wolfe algorithm to construct our numerical algorithm, and we demonstrate its effectiveness and scalability through computational experiments using a laptop on tensors with up to one-hundred million entries.
Chen Chen, Assistant Professor of Integrated Systems Engineering, Ohio State University
Anil Aswani, Associate Professor of Industrial Engineering and Operations Research, University of California-Berkeley
Predicted likelihood of successful vaginal birth after Cesarean using a race-unaware calculator in a safety net hospital
Amaya McNealey, Department of Industrial and Systems Engineering
Designed to mitigate concerns with the use of race and ethnicity in predictive algorithms “VBAC 2.0” was created in 2021 to replace its predecessor “VBAC 1.0” by replacing the use of race/ethnicity with an indication of chronic hypertension in its prediction. However, it is also unclear how these updated predictions affect the likelihood of recommending a trial of labor after cesarean (TOLAC) which is thought to be a more favorable outcome. To compare the effects of both calculators, we conducted a secondary analysis of pregnant patients who had a prior cesarean delivery at Grady Memorial Hospital. We used both the VBAC 1.0 and 2.0 calculators to separately estimate the predicted likelihood of a successful VBAC for each individual and examined if their estimated probability of successful VBAC from the 1.0 and 2.0 calculators would be favorable for recommending TOLAC. Finally, we computed the risk ratio for being recommended a cesarean under the VBAC 2.0 vs. 1.0 by race. 451 individuals met the inclusion criterion. The cohort was 84.7% Black, and 7.5% Hispanic, and 8.8% White. Under VBAC 2.0, 61% of Black patients would have been recommended a cesarean compared to 34% under 1.0 (RR: 1.78; 95% CI: 1.52, 2.10). The VBAC 2.0 calculator not only seems to limit the likelihood of a TOLAC recommendation in comparison with the original calculator but even more so for Black women. Our results emphasize the need for further research into when and how to consider race/ethnicity in clinical support tools and AI/ML models.
Meghan Meredith, H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology
Lauren N. Steimle, Assistant Professor, H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology
Gian-Gabriel P. Garcia, Assistant Professor, H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology
Kaitlyn K. Stanhope, Postdoctoral Fellow, Department of Gynecology and Obstetrics, Emory University
Marissa H. Platner, Assistant Professor, Department of Gynecology and Obstetrics, Emory University
Sheree L. Boulet, Associate Professor, Department of Gynecology and Obstetrics, Emory University
Racial/ethnic disparities in adult COVID-19 vaccination and outcomes in Georgia
Akane Fujimoto, Georgia Tech
The COVID-19 pandemic has exacerbated health disparities in racial and ethnic minority groups. It is important to identify these disparities to guide public health efforts. We analyzed COVID-19 vaccination, deaths, and hospitalizations in Georgia through February 2023. We computed vaccination, death, and hospitalization rates for Hispanic, non-Hispanic (NH) Asian, NH Black, and NH White adults by county of residence urban/rural classification and vaccination status. Rate ratios (RRs) were calculated to evaluate racial/ethnic disparities.
Racial/ethnic disparities in deaths, hospitalizations, and vaccination coverage varied by county urban/rural class. NH Black adults were at higher risk of being hospitalized with COVID-19 than NH White adults (RR 1.77), even when stratifying by vaccination status. NH Black adults were at higher risk of dying of COVID-19 than NH White adults (RR 1.37), except when they received a booster.
Stratifying vaccination and adverse outcomes using race/ethnicity, county urban/rural class, and vaccination status provided a better understanding of health disparities in Georgia to develop targeted interventions. Public health agencies should focus on both improving up-to-date vaccination coverage and removing barriers to access to care among communities that are underserved, particularly NH Black adults.
Pinar Keskinocak, Dima Nazzal
Resource Allocation for Different Types of Vaccines against COVID-19:Tradeoffs and Synergies between Efficacy and Reach
Daniel Kim, H. Milton Stewart School of Industrial & Systems Engineering
Objective: Vaccine shortage and supply-chain challenges have caused limited access by many resource-limited countries during the COVID-19 pandemic. One primary decision for a vaccine-ordering decision-maker is how to allocate the limited resources between different types of vaccines effectively. We studied the tradeoff between efficacy and reach of the two vaccine types that become available at different times.
Methods: We extended a Susceptible-Infected-Recovered-Deceased (SIR-D) model with vaccination, ran extensive simulations with different settings, and compared the level of infection attack rate (IAR) under different reach ratios between two vaccine types under different resource allocation decisions.
Results: When there were limited resources, allocating resources to a vaccine with high efficacy that became available earlier than a vaccine with lower efficacy did not always lead to a lower IAR, particularly if the former could vaccinate less than 42.5% of the population (with the selected study parameters) who could have received the latter. Sensitivity analyses showed that this result stayed robust under different study parameters.
Conclusions: Our results showed that a vaccine with lower resource requirements can significantly contribute to reducing IAR, even if it becomes available later in the pandemic, compared to a higher efficacy vaccine that becomes available earlier but requires more resources. Limited resource in vaccine distribution is significant challenge in many parts of the world that needs to be addressed to improve the global access to life-saving vaccines. Understanding the tradeoffs between efficacy and reach is critical for resource allocation decisions between different vaccine types for improving health outcomes.
Daniel Kim (Georgia Institute of Technology)
Pınar Keskinocak, Ph.D. (Georgia Institute of Technology)
Pelin Pekgün, Ph.D. (University of South Carolina)
Inci Yildirim, M.D., Ph.D. (Yale University)
The Shepherd Center-Georgia Tech Collaboration: Advancing Rehabilitative Patient Care and Research
Stephen Sprigle, Georgia Tech
Shepherd Center and Georgia Tech have announced a partnership that will unite researchers from both institutions and clinicians to improve care and create more success stories for people with spinal cord and brain injuries, pain, multiple sclerosis, and related neurological conditions. The strengths of GT and Shepherd creates a symbiotic inter-disciplinary environment that is unmatched in rehabilitation science.
Projects will focus on the constructs of usefulness and meaningfulness to insure impact on rehabilitation strategies and patient outcomes. Activities will span data science, artificial intelligence and machine learning, design of rehabilitation and assistive technologies, and development of revolutionary care models
Current activities include:
- Creating and utilizing Big Data System Architecture
- Data driven decision making for precision rehab & personal innovative solutions
- Development of Orphan Assistive Technology
The collaboration is currently seeking project ideas or areas of interest to link Georgia Tech faculty with Shepherd’s research and clinical staff.