Daniel Seung Kim

Objective: To estimate how often remission of type 2 diabetes is recorded in routine care, how long it lasts, and what ends it, separating a return of hyperglycemia from a restart of treatment. Research design and methods: We studied 136,781 adults aged 18 years or older with laboratory-confirmed type 2 diabetes in the Protégé linked electronic health record (EHR) and claims database. The 96,817 adults with glucose-lowering treatment recorded at or before cohort entry formed the primary (previously treated) cohort; 39,964 adults with no treatment recorded at entry were analyzed in parallel. Remission required two HbA1c values Why did we undertake this study? Most real-world studies count any end of remission as relapse, mixing rising HbA1c with treatment restart. What is the specific question we wanted to answer? How often is remission recorded in routine care, how long does it last, and what ends it? What did we find? About 1 in 23 previously treated adults had remission (4.33% ever recorded; 3.82% by 5 years); Remission episodes had a median duration of about 7.5 months (37.9% had ended by 5 months) and ended more often with a treatment restart than with rising HbA1c (47.83% vs 31.94% as first event by 5 years). Remission was less common with greater baseline treatment burden after exploratory adjustment (4.65% with 1 glucose-lowering class with a treatment episode recorded at or before entry, 3.62% with two, 2.69% with three, and 2.02% with four or more such drugs). What are the implications of our findings? Remission studies and quality measures should report glycemic relapse and treatment restart separately.

OBJECTIVE: To describe (1) US emergency department-treated pickleball injury burden and the injury-to-participation measure among total players, with burden among adults aged 60 to 79 years as a descriptive subgroup, and (2) determine whether general internet search interest for “pickleball” is also associated with total pickleball-related injuries. DESIGN: Ecological time series study, United States, 2014 to 2023. DATA SOURCES: National Electronic Injury Surveillance System (NEISS) emergency department microdata yielded 1,738 adjudicated pickleball cases, 921 of them orthopedic, weighted to an estimated 101,450 injuries nationally. Google Trends, English Wikipedia pageviews, and GDELT news volume served as general pickleball attention signals, with the Trends index for “sourdough” as a negative control. Sports and Fitness Industry Association (SFIA) participation estimates supplied a total player denominator. MAIN OUTCOME MEASURES: Annual national injury estimates with design-based 95 percent confidence intervals; injury-to-participation measures; and correlations between total pickleball-related injuries and general pickleball attention signals before and after removing trend and seasonality. RESULTS: Estimated emergency department-treated pickleball injuries rose 17.5-fold, from 1,391 in 2014 (highly unstable) to 24,294 in 2023. 70 percent of adjudicated cases involved adults aged 60 to 79. During 2017-2023, injury estimates rose 4.0-fold and participation rose about 4.4-fold. No temporal trend was detected in the injury-to-participation measure (P=0.59). Search interest looked like an excellent proxy: in annual levels it correlated with total pickleball-related injuries at R=0.96, and Pickleball Wikipedia pageviews at R=0.98. However, so did searches for "sourdough", at R=0.76. Removing trend and seasonality left little (monthly residual R=0.084; year-on-year R=0.28 over 96 months, 2015-2023), and the seasons ran opposite, with injuries peaking in February and searches in July, such that, on 12-month mean profiles, the sourdough negative control fitted the injury season (R=0.96) while pickleball ran against it (R=-0.40). General pickleball search interest and pickleball news coverage remained correlated in year-on-year changes in 2017-2023 (R=0.64), whereas their corresponding correlations with total pickleball-related injuries in 2017-2023 were 0.25 and 0.20. CONCLUSIONS: The absolute burden of emergency department-treated pickleball injury has grown substantially, but the available surveillance data cannot resolve whether the sport has become more hazardous per participant. The observed rise is consistent with growth in exposure alone. The strong correlation between search interest and injury counts was not specific, the association was substantially attenuated after detrending, and the seasonal patterns were opposed. These descriptive results do not establish that search interest can substitute for injury surveillance, and series that rise together should not be assumed to be related. WHAT THIS STUDY ADDS: Injury estimates rose 4.0-fold and participation rose about 4.4-fold during 2017-2023. No temporal trend was detected using the total-player participation denominator. However, this does not establish unchanged risk per pickleball player. Strong annual correlations between total pickleball-related injuries and general internet attention for “pickleball” (R=0.96 for search interest, 0.98 for Wikipedia pageviews) were also observed for a sourdough negative control (R=0.76). Injury-search correlations were substantially attenuated after accounting for trend and seasonality. General pickleball search interest remained associated with pickleball news coverage in year-on-year changes in 2017-2023 (R=0.64); however, their year-on-year correlations with total pickleball-related injuries in 2017-2023 were smaller (R=0.25 and 0.20). Taken together, these analyses caution against interpreting rising pickleball-related injury counts as rising individual risk or using search interest alone to extrapolate to injury burden.

PURPOSE: To demonstrate an end-to-end research workflow combining a large electronic health record (EHR) resource and a multi-agent artificial intelligence (AI) system governed by Human-Guided Orchestration (HGO), using sequential sacituzumab govitecan (SG) and trastuzumab deruxtecan (T-DXd) orders in breast cancer as the clinical use case. METHODS: A June 2026 EHR release contained 4.95 million patients, including 156,051 with breast cancer codes. Muse Spark supported cohort construction, executable statistical analysis, data-quality checks, literature retrieval, and reporting executed through Symbiotic Agent Teaming (SAT) under investigator direction. The main clinical measure was the second-drug record span, from first to last order. Analyses were descriptive and unadjusted; orders did not confirm administration or progression. RESULTS: The workflow identified 106 patients: 73 had T-DXd ordered first and 33 had SG ordered first. Median second-drug spans were 63 days (interquartile range [IQR], 29-116) for SG after T-DXd and 49 days (IQR, 21-126) for T-DXd after SG; the median difference was -14 days (95% bootstrap CI, -41 to 21; distributional P = 0.42). Reported outputs additionally included cohort definitions, aggregate analyses, evidence context, figures, and a preliminary draft manuscript. CONCLUSION: This study provides a concrete model for integrating large EHR resources, agentic AI, and human expertise across the research cycle. The ADC use case illustrates how computational execution and Expert-in-the-Loop (EITL) adjudication can produce interpretable research outputs and refine their scientific meaning, providing a foundation for extending and evaluating this approach across clinical questions.

BACKGROUND: Federal policy now requires sponsors to submit diversity action plans for clinical trials, yet the modifiable features of trial design associated with enrolled composition remain poorly characterized. English-language proficiency requirements are written into eligibility criteria and recorded as free text in a public registry, making them visible and, in principle, amenable to revision, but their relationship to enrolled composition has been examined only within single disease areas and in small samples of trials. We examined whether these requirements and documented language accommodation were associated with the racial and ethnic composition of enrolled participants. METHODS AND FINDINGS: We conducted a cross-sectional study of interventional trials registered on ClinicalTrials.gov with at least one US site and posted results, using a registry snapshot retrieved in October 2026. The primary cohort comprised the 44,487 trials with exclusively US sites, of which 21,924 (49.3%) reported baseline race and 17,434 (39.2%) reported baseline ethnicity with counts that reconciled with the posted denominator. Both exposures were classified from free-text eligibility criteria using regular expressions validated against independent physician review of 150 trials. For the English-language requirement classifier, sensitivity was 88.2% and specificity was 99.2%. The outcome was the trial-level proportion of enrolled participants in each category, analyzed using generalized linear models adjusted for intervention class, therapeutic area, sponsor, phase, enrollment, sites, start year, US Census region, and the population composition of the trial’s site states. An English-language proficiency requirement appeared in 5,736 trials (12.9%); its prevalence rose from 7.4% of trials starting in 2010 to 26.2% in 2023 (odds ratio [OR] per decade 2.91, PP =0.001) and Hispanic/Latinx participants (OR=0.85, 95% CI 0.79-0.91, P P =0.75). Documented accommodation was associated with a higher proportion of Hispanic/Latinx participants (OR=3.23, 95% CI 2.82-3.71, P <0.001). Geographic adjustment did not change the statistical significance of the requirement associations and attenuated the accommodation association. The main limitations were the cross-sectional design, which precludes causal inference; the restriction of enrollment analyses to trials reporting baseline demographics, which were more recent than those without such reporting; and possible misclassification of automatically classified exposures. CONCLUSIONS: English-language proficiency requirements became more than three times as common between 2010 and 2023 and were associated with lower enrolled proportions of Asian and Hispanic/Latinx participants. Less common documented language accommodation was associated with a higher enrolled proportion of Hispanic/Latinx participants. Because these criteria are investigator-written and publicly recorded, they may be a tractable target for improving enrollment equity, although whether revising them changes who enrolls requires prospective evaluation.

BACKGROUND: Hypertrophic cardiomyopathy (HCM) is a pan-ethnic disease, at least as common in Black as in White Americans, yet Black patients reach specialist care later, undergo septal reduction less often, and die sooner. Whether the defining trials enrolled representatively is unknown. We sought to measure Black participation in the landmark HCM trials against three expectations of increasing stringency, and to test whether trial site geography or quantitative eligibility thresholds account for the shortfall. METHODS: Of 197 registered interventional HCM trials, 32 met prespecified landmark tiers; demographics were extracted for 16 and 13 reported race. The Black share of Black plus White participants was pooled by random-effects meta-analysis and compared with the US adult population, with each trial’s own site counties, and with a disease-referenced echocardiographic expectation. Site geography and eligibility thresholds were tested using site addresses, census data, and NHANES. RESULTS: Across 13 trials, 70 of 1,780 Black or White participants were Black. The pooled Black share was 3.98% (95% CI 2.68-5.86), against a population expectation of 16.3%, giving a participation-to-prevalence ratio (PPR) of 0.24 (95% CI 0.16-0.36), the lowest of any racial and/or ethnic minority group. The clinical trial site network reached 63.7% of Black and 47.5% of White US adults, with geospatial analysis predicting 26.4% Black enrollment; however, every trial fell below this expectation. Quantitative eligibility thresholds closed only 2.0% of the enrollment gap. CONCLUSIONS: Black participation was about one quarter of the population expectation. Neither site geography nor eligibility thresholds explain it. CLINICAL PERSPECTIVE: What is New?: Across the 13 landmark hypertrophic cardiomyopathy trials that posted race, 70 of 1,780 participants recorded as Black or White were Black, a pooled share of 3.98% against a population expectation of 16.3%. The US counties hosting the trials’ own sites predicted 26.4% Black enrollment, and every trial enrolled below its own catchment expectation. Neither site geography nor quantitative eligibility thresholds explain the shortfall, because the site network reached 63.7% of Black adults and the extracted thresholds closed 2.0% of the gap in Black enrollment. What are the Clinical Implications?: Black patients with hypertrophic cardiomyopathy live within reach of the sites that enrolled them least, so clinicians should treat trial referral as a distinct clinical act and audit their own referral patterns. Future trials should report screening, referral, and offer by race prospectively, and should benchmark enrollment against the catchment population of their own sites. The loss occurs downstream of geography and eligibility, in site recruitment conduct and in the referral pathway, and that is where measurement and correction must begin.

Objective: Hispanic/Latinx populations in the U.S. experience higher rates of chronic disease linked to physical inactivity, yet digital health interventions remain largely inaccessible to more than 16 million Hispanic/Latinx adults with limited English proficiency. While large language models (LLMs) offer scalable personalization, their use in non-English behavioral coaching is unexplored. This study introduces MHC-Coach-ES, a Spanish-language LLM fine-tuned on the Transtheoretical Model (TTM) of behavior change. Materials and Methods: We fine-tuned Llama 3-70B-Instruct using a two-stage pipeline. First, the model was adapted to Spanish health and motivational language using a 2.21-million-token corpus. Second, it was instruction-tuned on 3,268 translated human written messages to align the model with the Transtheoretical Model (TTM) of Behavioral Change. We compared MHC-Coach-ES with Llama 3-70B-Instruct and translated human-expert messages using a forced-choice preference survey (N = 77) and blinded expert review (N = 2). Results: Spanish-speaking participants significantly preferred MHC-Coach-ES messages over translated human-expert messages (81% preference, P <0.001). Linguistic analysis showed that MHC-Coach-ES produced more temporally anchored messages than the base model (65% vs. 20%), while maintaining readability. In blinded evaluation, clinical experts rated MHC-Coach-ES higher for alignment with Transtheoretical Model stages than human-expert messages (4.83 vs. 4.38 out of 5). The base model also outperformed translated expert messages across preference and expert ratings. Conclusions: Generative AI can operationalize behavioral science frameworks in Spanish, offering a scalable approach to reducing health disparities. The strong performance of both MHC-Coach-ES and the base model highlights the promise of generative and personalized approaches over translation-based localization for theory-driven behavioral interventions.

BACKGROUND The 6-minute walk test (6MWT) is a cornerstone of functional capacity assessment, but its utility is constrained by the need for clinic-based, supervised administration. Advances in mobile and wearable technology offer a solution for scalable, remote assessment in real-world environments. However, existing 6MWT prediction equations were developed in controlled clinical settings and their applicability to unsupervised, smartphone-collected data is unknown. The My Heart Counts study collected >30,000 self-administered digital 6MWT across 8,922 participants, creating an opportunity to develop population-specific predictive equations for real-world, unsupervised environments. OBJECTIVE The primary objective of this study was to develop the first fully-digital prediction equations for smartphone-based 6MWT performance using data from the large-scale My Heart Counts cohort. A secondary objective was to quantify the transferability of established, clinic-based prediction equations to this real-world digital dataset. METHODS We analyzed 16,828 self-administered 6MWTs completed by 3,580 participants in the My Heart Counts study between March 2015 and April 2022. Of an initial cohort of 8,922, participants were included if they had at least one valid 6MWT and complete covariate data. Participants were stratified into Healthy (n=2,975) and Cardiovascular Disease (n=605) cohorts. Using linear mixed-effects models with random intercepts for participants, we developed separate prediction equations for men and women in each cohort. Predictor variables included age, age², height, weight, body mass index (BMI), self-reported physical activity, race, and ethnicity. Model performance was evaluated using the multiple correlation coefficients (R) between predicted and observed distances. RESULTS Correlations were R=0.24 for Healthy Men (n=2,152), R=0.36 for Healthy Women (n=823), R=0.49 for Men with Cardiovascular Disease (n=454), and R=0.54 for Women with Cardiovascular Disease (n=151). Critically, established clinical prediction equations performed poorly when applied to our digital cohort, yielding substantially lower correlations (R=0.12-0.15). This represents a substantial degradation from the performance reported in their original clinic-based validation studies (R=0.60-0.81). CONCLUSIONS The poor performance of traditional 6MWT prediction equations in a digital context highlights that they are not directly transferable to unsupervised, real-world, smartphone-based assessments. This study provides the first set of prediction equations derived entirely from and developed specifically for smartphone-based 6MWT data. While predictive strength is lower than in supervised settings, these fully-digital models represent a step toward enabling scalable and equitable remote monitoring of functional capacity. The equations are publicly available for clinical and research use.