Dezhi Wu

Background: AI-powered clinical decision support systems (CDSS) have shown promise in improving prediction, monitoring, and treatment optimization across clinical domains, including HIV care. However, translating AI outputs derived from electronic health records into clinically meaningful, trustworthy, and actionable decision support remains challenging, underscoring the need for more human-centered and socioecologically grounded CDSS design. Objective: This study aimed to explore how we can effectively translate the outputs of machine learning models based on HIV electronic health records into a real AI-powered CDSS for HIV care. Using the human-in-the-loop method, we engaged a set of stakeholders, including HIV physicians, nurse practitioners, infectious disease pharmacists, social workers, and case managers. Stakeholders interacted with an AI-powered CDSS prototype to identify barriers and challenges to adoption, as well as to inform a more holistic and context-aware AI-powered CDSS design. Methods: We conducted a field study at Prisma Health in South Carolina that included pre- and postsurveys, interactive usability testing sessions, think-alouds, and in-depth interviews with 16 clinicians providing HIV care between March and September 2025. We analyzed survey responses using descriptive statistics, and then transcribed and analyzed think-aloud and interview data using an etic and emic approach. Results: Clinicians identified multiple challenges and design considerations for AI-powered HIV CDSS, demonstrating that clinician-AI interaction is inherently sociotechnical and embedded across multiple socioecological levels. While clinicians relied on familiar clinical indicators as cognitive anchors for interpreting AI predictions, they emphasized that social determinants of health were central to their own risk assessment and clinical decision-making. Additionally, clinicians' trust in AI is conditional and develops over time, with explainability and actionability emerging as critical factors for translating predictions into meaningful clinical interventions. Conclusions: Findings highlight the need to move beyond technically accurate predictions toward AI-powered CDSS designs that align with clinicians’ cognitive practices and socioecological realities of HIV care. By extending a sociocognitive framework through empirical grounding in HIV clinical practice, this study offers design insights for developing AI-powered CDSS that are trustworthy, context-aware, and capable of supporting actionable decision-making in HIV care settings and beyond.

Objectives: We examine how hospital characteristics relate to clinical and operational artificial intelligence (AI) adoption and implementation stages and characterize AI deserts and spatial clustering patterns to highlight place-based AI access gaps among United States (US) hospitals. Materials and Methods: We used the 2024 American Hospital Association Annual Survey data from 2720 hospitals with at least one AI response. We applied logistic regression models to examine the associations between hospital characteristics and AI adoption, local indicators of spatial association to identify local clusters, and distance analyses to locate AI desert hospitals (ie, geographically isolated nonadopters located >50 miles from the nearest AI adopters). Results: We found that system membership, larger bed size, and nurse staffing intensity are positively associated with clinical and operational AI adoption and implementation stages, whereas rural location, for-profit ownership, and physician intensity are negatively associated with them. Teaching status is more strongly associated with clinical AI, while system membership is more positively associated with operational AI. Although 68.0% of hospitals adopted ≥1 clinical AI functionalities and 60.7% adopted ≥1 operational AI functionalities, 12.1% of clinical and 13.2% of operational non-adopters are classified as AI desert hospitals. Discussion: AI deserts reveal regional implementation gaps; sustained diffusion requires building shared regional capacity rather than relying only on hospital-level incentives. Policy implications to emerging divides may include tracking implementation stage, providing domain-specific support, and funding regional partnerships. Conclusion: US hospital AI diffusion is uneven and spatially structured. AI deserts mark regional gaps in access to AI-adopting hospitals.

Games and puzzles play important pedagogical roles in STEM learning. New AI algorithms that can solve complex problems offer opportunities for scaffolded instruction in puzzle solving. This paper presents the ALLURE system, which uses an AI algorithm (DeepCubeA) to guide students in solving a common first step of the Rubik's Cube (i.e., the white cross). Using data from a pilot study we present preliminary findings about students' behaviors in the system, how these behaviors are associated with STEM skills - including spatial reasoning, critical thinking and algorithmic thinking. We discuss how data from ALLURE can be used in future educational data mining to understand how students benefit from AI assistance and collaboration when solving complex problems.

Background: Black Americans experience a higher incidence of colorectal cancer than Whites despite undergoing prevention screenings similar to those of Whites since 2010. We compared the colorectal adenoma status of Black and White patients at screening colonoscopy, a measure of colorectal cancer risk. Methods: Using cross-sectional, observational data, we studied colorectal adenomas at first-time screening colonoscopy in average-risk patients aged 40 to 89 years, screened between September 2001 and July 2016 in South Carolina. We analyzed the age-adjusted odds of Black men and women (vs. White) having adenomas, advanced adenomas, ≥3 nonadvanced adenomas, and right hyperplastic polyps, and compared their total polyp burden (the sum of the diameters of all adenomas and right polyps detected). Results: Among 28,100 patients (58.4% Black and 53.8% women), we found that Black patients had lower age- and gender-adjusted odds than White patients of having adenoma (OR = 0.88, P < 0.01) and right hyperplastic polyp (OR = 0.74, P < 0.01), with a similar pattern within gender groups. Black and White patients were similar about advanced adenoma and 3+ nonadvanced adenoma. Among patients with lesions, mean polyp burden ranged from 8.5 mm (±7.2) for Black women aged 40 to 49 years to 12.3 mm (±7.4) for Black men aged more than 70 years. Age-adjusted polyp burden was 0.4 mm higher for Black men than for White men and 0.3 mm lower for Black women than for White women patient groups (P < 0.01). Conclusions: In a large, racially balanced patient sample, Black and White patients showed similar polyp profiles. Impact: Given similar adenoma status, other evidence-supported clinical factors associated with suboptimal polyp detection should be explored to understand the continuing colorectal cancer disparities affecting Black Americans.

Healthcare staffing shortages and high 30-day readmission rates are pressing challenges in today's healthcare systems. While extensive studies have explored the positive impact of Health Information Technology (HIT) on health outcomes, the effect of hospital staffing on HIT effectiveness, particularly Clinical Decision Support Systems (CDSS), is underexplored. Therefore, we examine whether physician and resident staffing levels moderate the impact of CDSS on 30-day readmission rates. After analyzing data from 1,605 hospitals in the United States from 2017 and 2018, we found that the effect of CDSS implementation in reducing 30-day readmission rates is more pronounced in hospitals with lower bed-to-physician and bed-to-resident ratios, suggesting a higher number of physicians and residents per bed enhances the effectiveness of CDSS. The study also highlights different effects based on hospital size and complexity levels and emphasizes the importance of considering staffing levels when implementing CDSS and other HIT to maximize their benefits.

Background: Geographic, demographic, and socioeconomic differences in health outcomes persist despite the global focus on these issues by health organizations. Barriers to accessing care contribute significantly to these health disparities. Among these barriers, those related to travel time—the time required for patients to travel from their residences to health facilities—remain understudied compared with others. Objective: This study aimed to explore the impact of telehealth in addressing health disparities associated with travel time to hospitals for patients with recurrent hospital admissions. It specifically examined the role of telehealth in reducing in-hospital length of stay (LOS) for patients living farther from the hospital. Methods: We sourced the data from 4 datasets, and our final effective sample consisted of 1,600,699 admissions from 536,182 patients from 63 hospitals in New York and Florida in the United States from 2012 to 2015. We applied fixed-effect models to examine the direct effects and the interaction between telehealth and patients’ travel time to hospitals on LOS. We further conducted a series of robustness checks to validate our main models and performed post hoc analyses to explore the different effects of telehealth across various patient groups. Results: Our summary statistics show that, on average, 22.08% (353,396/1,600,699) of patients were admitted to a hospital with telehealth adopted, with an average LOS of 5.57 (SD 5.06) days and an average travel time of about 16.89 (SD 13.32) minutes. We found that telehealth adoption is associated with a reduced LOS (P<.001) and this effect is especially pronounced as the patients’ drive time to the hospital increases. Specifically, the coefficient for drive time is –0.0079 (P<.001), indicating that for every additional minute of driving time, there is a decrease of 0.0079 days (approximately 11 minutes) in the expected LOS. We also found that telehealth adoption has a larger impact on patients frequently needing health services, patients living in high internet coverage areas, and patients who have high virtualization potential diseases. Conclusions: Our findings suggest that telehealth adoption can mitigate certain health disparities for patients living farther from hospitals. This study provides key insights for health care practitioners and policy makers on telehealth’s role in addressing distance-related disparities and planning health care resources. It also has practical implications for hospitals in resource-limited countries that are in the early stages of implementing telehealth.

Background: In 2021, the United States experienced a 14% rise in fatal drug overdoses totaling 106,699 deaths, driven by harmful opioid use, particularly among individuals in the perinatal period who face increased risks associated with opioid use disorders (OUDs). Increased concerns about the impacts of escalating harmful opioid use among pregnant and postpartum persons are rising. Most of the current limited perinatal OUD studies were conducted using traditional methods, such as interviews and randomized controlled trials to understand OUD treatment, risk factors, and associated adverse effects. However, little is known about how social media data, such as X, formerly known as Twitter, can be leveraged to explore and identify broad perinatal OUD trends, disclosure and communication patterns, and public health surveillance about OUD in the perinatal period. Objective: The objective is 3-fold: first, we aim to identify key themes and trends in perinatal OUD discussions on platform X. Second, we explore user engagement patterns, including replying and retweeting behaviors. Third, we investigate computational methods that could potentially streamline and scale the labor-intensive manual annotation effort. Methods: We extracted 6 million raw perinatal-themed tweets posted by global X users during the opioid epidemic from May 2019 to October 2021. After data cleaning and sampling, we used 500 tweets related to OUD in the perinatal period by US X users for a thematic analysis using NVivo (Lumivero) software. Results: Seven major themes emerged from our thematic analysis: (1) political views related to harmful opioid and other substance use, (2) perceptions of others’ substance use, (3) lived experiences of opioid and other substance use, (4) news reports or papers related to opioid and other substance use, (5) health care initiatives, (6) adverse effects on children’s health due to parental substance use, and (7) topics related to nonopioid substance use. Among these 7 themes, our user engagement analysis revealed that themes 4 and 5 received the highest average retweet counts, and theme 3 received the highest average tweet reply count. We further found that different computational methods excel in analyzing different themes. Conclusions: Social media platforms such as X can serve as a valuable tool for analyzing real-time discourse and exploring public perceptions, opinions, and behaviors related to maternal substance use, particularly, harmful opioid use in the perinatal period. More health promotion strategies can be carried out on social media platforms to provide educational support for the OUD perinatal population.

Background: Smart tracking technology (STT) that was applied for clinical use has the potential to reduce 30-day all-cause readmission risk through streamlining clinical workflows with improved accuracy, mobility, and efficiency. However, previously published literature has inadequately addressed the joint effects of STT for clinical use and its complementary health ITs (HITs) in this context. Furthermore, while previous studies have discussed the symbiotic and pooled complementarity effects among different HITs, there is a lack of evidence-based research specifically examining the complementarity effects between STT for clinical use and other relevant HITs. Objective: Through a complementarity theory lens, this study aims to examine the joint effects of STT for clinical use and 3 relevant HITs on 30-day all-cause readmission risk. These HITs are STT for supply chain management, mobile IT, and health information exchange (HIE). Specifically, this study examines whether the pooled complementarity effect exists between STT for clinical use and STT for supply chain management, and whether symbiotic complementarity effects exist between STT for clinical use and mobile IT and between STT for clinical use and HIE. Methods: This study uses a longitudinal in-patient dataset, including 879,122 in-patient hospital admissions for 347,949 patients in 61 hospitals located in Florida and New York in the United States, from 2014 to 2015. Logistic regression was applied to assess the effect of HITs on readmission risks. Time and hospital fixed effects were controlled in the regression model. Robust standard errors (SEs) were used to account for potential heteroskedasticity. These errors were further clustered at the patient level to consider possible correlations within the patient groups. Results: The interaction between STT for clinical use and STT for supply chain management, mobile IT, and HIE was negatively associated with 30-day readmission risk, with coefficients of –0.0352 (P=.003), –0.0520 (P<.001), and –0.0216 (P=.04), respectively. These results indicate that the pooled complementarity effect exists between STT for clinical use and STT for supply chain management, and symbiotic complementarity effects exist between STT for clinical use and mobile IT and between STT for clinical use and HIE. Furthermore, the joint effects of these HITs varied depending on the hospital affiliation and patients’ disease types. Conclusions: Our results reveal that while individual HIT implementations have varying impacts on 30-day readmission risk, their joint effects are often associated with a reduction in 30-day readmission risk. This study substantially contributes to HIT value literature by quantifying the complementarity effects among 4 different types of HITs: STT for clinical use, STT for supply chain management, mobile IT, and HIE. It further offers practical implications for hospitals to maximize the benefits of their complementary HITs in reducing the 30-day readmission risk in their respective care scenarios.

Background: According to the Morbidity and Mortality Weekly Report, polysubstance use among pregnant women is prevalent, with 38.2% of those who consume alcohol also engaging in the use of one or more additional substances. However, the underlying mechanisms, contexts, and experiences of polysubstance use are unclear. Organic information is abundant on social media such as X (formerly Twitter). Traditional quantitative and qualitative methods, as well as natural language processing techniques, can be jointly used to derive insights into public opinions, sentiments, and clinical and public health policy implications. Objective: Based on perinatal polysubstance use (PPU) data that we extracted on X from May 1, 2019, to October 31, 2021, we proposed two primary research questions: (1) What is the overall trend and sentiment of PPU discussions on X? (2) Are there any distinct patterns in the discussion trends of PPU-related tweets? If so, what are the implications for perinatal care and associated public health policies? Methods: We used X’s application programming interface to extract >6 million raw tweets worldwide containing ≥2 prenatal health- and substance-related keywords provided by our clinical team. After removing all non–English-language tweets, non-US tweets, and US tweets without disclosed geolocations, we obtained 4848 PPU-related US tweets. We then evaluated them using a mixed methods approach. The quantitative analysis applied frequency, trend analysis, and several natural language processing techniques such as sentiment analysis to derive statistics to preview the corpus. To further understand semantics and clinical insights among these tweets, we conducted an in-depth thematic content analysis with a random sample of 500 PPU-related tweets with a satisfying κ score of 0.7748 for intercoder reliability. Results: Our quantitative analysis indicates the overall trends, bigram and trigram patterns, and negative sentiments were more dominant in PPU tweets (2490/4848, 51.36%) than in the non-PPU sample (1323/4848, 27.29%). Paired polysubstance use (4134/4848, 85.27%) was the most common, with the combination alcohol and drugs identified as the most mentioned. From the qualitative analysis, we identified 3 main themes: nonsubstance, single substance, and polysubstance, and 4 subthemes to contextualize the rationale of underlying PPU behaviors: lifestyle, perceptions of others’ drug use, legal implications, and public health. Conclusions: This study identified underexplored, emerging, and important topics related to perinatal PPU, with significant stigmas and legal ramifications discussed on X. Overall, public sentiments on PPU were mixed, encompassing negative (2490/4848, 51.36%), positive (1884/4848, 38.86%), and neutral (474/4848, 9.78%) sentiments. The leading substances in PPU were alcohol and drugs, and the normalization of PPU discussed on X is becoming more prevalent. Thus, this study provides valuable insights to further understand the complexity of PPU and its implications for public health practitioners and policy makers to provide proper access and support to individuals with PPU.