Big Data Health Science Conference Gathers Global Cohort to Bridge Discovery and Delivery

The University of South Carolina’s Big Data Health Science Center hosted the 7th National Big Data Health Science Conference on February 12 and 13, 2026, in Columbia, South Carolina, under the theme “Unlocking the Power of Big Data in Health: Bridging the Gap Between Discovery and Delivery.” Supported in part by the National Institutes of Health, the conference brought together researchers from academia, government, and industry to exchange theoretical, methodological, and substantive knowledge about big data in biomedical science and public health. The proceedings, published in BMC Proceedings, capture the breadth of work presented, from artificial intelligence methods for clinical diagnosis to population-scale analyses of Medicaid claims and national survey data.

The 2026 meeting marked a record year for the annual event. Organizers received more than 100 abstract submissions, and over 200 participants attended from 17 US states and 7 countries, including South Africa, Nigeria, Pakistan, Rwanda, Lebanon, and the United Kingdom. The scientific program featured 6 plenary speakers, 55 poster presentations, and 7 breakout sessions with more than 40 oral presentations. Session themes ranged from big data analytics and emerging methodologies to public and community health, biomedical research, clinical practice and patient care, health research perspectives from the Global South, and data-driven solutions for implementation challenges. An NIH trainee session highlighted research training experiences from four NIH- and NIAID-funded programs at the university, three R25 awards and one T35, focused on big data and infectious diseases.

The conference also served its educational mission through a pre-conference SAS Viya training workshop provided by SAS, Inc., a training session with the South Carolina Department of Public Health Data Repository, and a formal networking reception. A virtual student case competition preceded the main meeting, running February 6 through 8, 2026, with 32 teams totaling roughly 100 undergraduate and graduate students from 17 universities across the United States. Teams received healthcare problems and datasets and were judged by panels of industry and academic experts on their use of the full analytics process, from framing the problem through methodology selection, data use, model building, and innovation. This year’s challenge asked teams to develop data-driven strategies supporting shared decision-making for patients newly diagnosed with musculoskeletal conditions, comparing surgical intervention against conservative non-surgical management while modeling clinical variation and patient preference heterogeneity.

CareLogic, a team from Rutgers Business School at Rutgers University, took first place and a $5,000 prize. Dynamic Dog from the University of South Carolina earned second place and $3,000, while HAC from Missouri State University placed third with $2,000. Honorable mention awards of $500 went to teams from the University of Iowa, Dartmouth Geisel School of Medicine, Carnegie Mellon University, and Florida State University. Among the presented research, a retrospective quasi-experimental study from South Africa examined whether an eLearning management system could improve quality assurance of rapid HIV testing. The research team, led by Amanda Mohlala of the South African Medical Research Council and University of Johannesburg, allocated 2,526 healthcare workers in Gauteng primary health facilities who had received HIV quality assurance training into two groups: 813 workers exposed to the eLearning platform and 1,744 exposed to conventional didactic training.

Using the Stepwise Process for Improving the Quality of Rapid Testing checklist version 3, evaluated across 1,609 primary health care facilities in Gauteng Province, the researchers analyzed program data collected annually by the South African National Department of Health from 2019 to 2023. Healthcare workers exposed to the eLearning platform achieved significantly higher certification rates than unexposed workers in every year measured: 32.26 percent versus 10.97 percent in 2021, 42.96 percent versus 23.98 percent in 2022, and 68.21 percent versus 48.95 percent in 2023, all with p values below 0.001. Facility-level quality assurance scores rose from a mean of 0.79 to 0.94 over the study period, with a Cohen’s d of approximately 2.0, and both Welch’s t-test and the Mann-Whitney U test confirmed statistically significant year-on-year gains. The authors noted that digital training effectiveness can depend on contextual factors such as reliable technology access and internet connectivity, and that in-person training may yield higher initial knowledge gains in some settings.

Machine learning also featured prominently in the presented work. Kesheng Wang and colleagues at the University of South Carolina analyzed 6,793 adults from the 2024 Health Information National Trends Survey, covering 93 variables across sociodemographic, behavioral, health-belief, and media-use domains, to classify awareness of the link between alcohol consumption and cancer risk. Among respondents, 36.6 percent believed alcohol increases cancer risk, 9.2 percent believed it has no effect or decreases risk, and 54.2 percent did not know. The team applied seven algorithms, including support vector machines, logistic regression, random forest, gradient boosting, k-nearest neighbor, naive Bayes, and XGBoost, with feature selection via Boruta and LASSO. Random forest achieved the highest discrimination for the “Yes” category with an F1 score of 0.792, while XGBoost performed best for “Don’t know” responses with an F1 score of 0.664. Key predictors included beliefs about air pollution, palliative care knowledge, social media use, low income, marijuana use, hookah smoking, education level, and cancer information-seeking behavior.

Population-scale claims data anchored another line of investigation. Chao Cai of the University of South Carolina College of Pharmacy conducted a retrospective cohort study of 2,870 South Carolina Medicaid beneficiaries with schizophrenia or schizoaffective disorder who filled antipsychotic prescriptions between 2012 and 2018. Using latent class mixture models on quarterly proportion-of-days-covered measures over 24 months, the analysis identified four adherence trajectories: a best-adherent class comprising 56 percent of the population, an increasing-adherence class at 8 percent, an early drop-off class at 12 percent, and a worst-adherent class at 24 percent. Long-acting injectable antipsychotics reduced the odds of belonging to the worst-adherent class compared with oral therapy, with an odds ratio of 0.47 and a 95 percent confidence interval of 0.38 to 0.60. During months 13 through 24, the early drop-off and increasing-adherence classes showed the highest emergency department visits and hospitalizations, suggesting that fluctuating adherence itself carries risk beyond consistently low adherence.

Environmental determinants of substance use were examined through the National Institutes of Health All of Us dataset. Gazi Sakir Mohammad Pritom and colleagues analyzed 138,515 participants with complete social determinants of health survey items and electronic health record data, excluding those with baseline stimulant use disorder, which had a prevalence of 0.27 percent in the cohort. Structural equation modeling showed that higher neighborhood vulnerability significantly increased perceived stress, with a standardized coefficient of 0.35, and that both stress and neighborhood vulnerability independently predicted stimulant use disorder, with coefficients of 0.12 and 0.13 respectively. The indirect effect of neighborhood vulnerability through perceived stress was significant, indicating partial mediation. The authors emphasized that the very large sample enabled precise estimation despite the low prevalence, and called for longitudinal designs and formal measurement validation in future work.

Healthcare costs and basic biology rounded out the program. A retrospective cohort study of 53,159 Medicaid-enrolled birthing people in South Carolina, using linked hospital, claims, and birth certificate data from 2019 to 2021, found that emergency department initiation of prenatal care was associated with 12 percent higher total pregnancy-related expenditures than outpatient initiation, while telehealth initiation was cost-neutral. Notably, among rural outpatient initiators, telehealth follow-up without emergency department care was associated with higher expenditures, with an adjusted incidence rate ratio of 1.21, suggesting telehealth may function differently in rural care environments. On the laboratory side, researchers from Claflin University and Clemson University presented a convolutional neural network approach for segmenting aligned collagen fibers in second-harmonic-generation imaging volumes, trained on 2,000 synthetic images generated with Stable Diffusion 3.5 Large, achieving precision and recall above 0.90 despite observed overfitting. Cell biology work from the University of South Carolina demonstrated that pharmacological inhibition of the DDX3 helicase reshapes RNA pools within stress granules, reducing their abundance and size in U2OS cells under oxidative stress. The 8th National Big Data Health Science Conference is scheduled for February 11 and 12, 2027, in Columbia, South Carolina.

Subject of Research: Big data applications in health science, from HIV testing quality and cancer risk awareness to antipsychotic adherence and stress granule biology

Article Title: Proceedings of the 7th National Big Data Health Science Conference

Article References: Proceedings of the 7th National Big Data Health Science Conference. (2026). BMC Proceedings, 20(36), 40. https://doi.org/10.1186/s12919-026-00400-5 (Author names are not included in the publisher’s citation export.)

Image Credits: AI Generated

DOI: 10.1186/s12919-026-00400-5

Keywords: big data health science, University of South Carolina, machine learning, HIV testing, healthcare worker training, cancer risk awareness, antipsychotic adherence, stimulant use disorder, prenatal care costs, stress granules, collagen biofabrication, student case competition