Daniel Smolyak

Background: Latina pregnant women and mothers of infants in the United States experience elevated risk for perinatal anxiety, stress, and depression alongside persistent barriers to culturally and linguistically responsive care. AI-powered digital health tools offer a potentially scalable approach to address these gaps, yet evidence on their acceptability and effectiveness for this population remains limited. Objective: We aimed to evaluate the feasibility, acceptability, and preliminary outcome changes of Rosie, a Spanish-language AI-powered chatbot designed with and for Latina mothers, on maternal mental health and healthcareuse outcomes. Methods: We conducted a mixed methods pilot randomized controlled trial with 30 Spanish-speaking Latina participants who were pregnant or parenting an infant younger than 6 months. Participants were randomized 1:1 to the Rosie intervention or a comparison book club condition that provided monthly children’s board books. Surveys administered at baseline and 3-month follow-up assessed depression (Patient Health Questionnaire-9), anxiety (Generalized Anxiety Disorder-7), perceived stress (Perceived Stress Scale), and health care use. Quantitative analyses examined within- and between-group changes, and qualitative feedback assessed acceptability and user experience. Results: Baseline mental health scores were low in both groups; however, several baseline demographic differences were observed between study arms, including age, pregnancy vs parenting status, marital status, education, and country of origin. Over 3 months, participants in the Rosie group demonstrated statistically significant within-group reductions in anxiety (mean baseline 3.27, SD 2.89; mean change −2.00, SD 2.94; P =.02) and perceived stress (mean baseline 30.00, SD 5.53; mean change −12.20, SD 12.99; P =.003). Depression scores declined modestly in both groups, with no significant between-group differences. No significant differences were observed in maternal emergency department use or adherence to well-baby visits. Engagement with Rosie was high, and qualitative feedback indicated that participants found the chatbot useful and supportive. Conclusions: This pilot study suggests that a culturally and linguistically tailored AI-powered chatbot is feasible and acceptable among Latina mothers during the perinatal period. Although anxiety and perceived stress decreased within the Rosie group, no significant between-group differences were observed. Within-group reductions in anxiety and perceived stress among Rosie participants are preliminary and should be interpreted cautiously because the trial was not powered to detect between-group effects and no statistically significant between-group differences were observed. Larger, adequately powered trials are needed to evaluate effectiveness.

Objectives: Social determinants of health (SDOH) are incompletely captured in structured electronic health records (EHRs) but are frequently documented in unstructured clinical notes. We evaluated large language models (LLMs) for extracting SDOH from clinical text. Materials and Methods: We constructed an adult sepsis cohort from the Medical Information Mart for Intensive Care-IV (Sequential Organ Failure Assessment ≥2) and analyzed clinical notes from 1 year prior to 30 days following suspected infection. Three instruction‑tuned, decoder‑only LLMs (Mistral‑Instruct‑7B-v0.2, DeepSeek‑R1‑Distill‑Qwen‑14B, and GPT‑oss‑20B) were evaluated using structured prompts with predefined label schemas and few‑shot examples. Performance was benchmarked against a clinically validated annotated dataset and compared with a fine‑tuned encoder‑decoder baseline. Macro‑F1 scores were reported. A gold‑standard Intensive Care Unit (ICU) sepsis subset was independently annotated by 3 reviewers to assess domain‑level performance and ensemble strategies. Results: Decoder‑only models outperformed the fine‑tuned encoder‑decoder baseline across SDOH domains. GPT‑oss achieved the highest macro‑F1 score (0.79) compared with Flan‑T5‑XXL (0.57). Prompt refinement substantially improved extraction accuracy. Ensemble majority voting increased robustness across domains, while unanimous agreement yielded high precision but limited coverage. In a subsequent mortality analysis, extracted SDOH did not independently predict 30-day mortality, which was instead associated with established clinical and demographic risk factors. Discussion: Instruction‑tuned decoder‑only LLMs can reliably extract multiclass SDOH from unstructured clinical notes without task‑specific fine‑tuning. Ensemble and agreement‑based strategies provide practical operating points for high‑precision clinical deployment. Conclusion: These findings support the feasibility of leveraging LLMs to enrich EHRs with structured SDOH data, providing a scalable approach for incorporating social context into downstream risk stratification and health outcome prediction.

BACKGROUND Latina pregnant women and mothers of infants in the United States experience elevated risk for perinatal anxiety, stress, and depression alongside persistent barriers to culturally and linguistically responsive care. Artificial intelligence (AI)-powered digital health tools offer a potentially scalable approach to address these gaps, yet evidence on their acceptability and effectiveness for this population remains limited. OBJECTIVE To evaluate the feasibility, acceptability, and preliminary efficacy of Rosie, a Spanish-language AI-powered chatbot designed with and for Latina mothers, on maternal mental health and healthcare utilization outcomes. METHODS We conducted a mixed-methods pilot randomized controlled trial with 30 Spanish-speaking Latina participants who were pregnant or parenting an infant younger than six months. Participants were randomized 1:1 to the Rosie intervention or a comparison Book Club condition that provided monthly children’s board books. Surveys administered at baseline and three-month follow-up assessed depression (PHQ-9), anxiety (GAD-7), perceived stress (PSS), and healthcare utilization. Quantitative analyses examined within- and between-group changes, and qualitative feedback assessed acceptability and user experience. RESULTS Baseline demographic characteristics and mental health scores were generally comparable between groups. Over three months, participants in the Rosie group demonstrated statistically significant within-group reductions in anxiety (mean change -2.00, p = 0.02) and perceived stress (mean change -12.20, p = 0.003). Depression scores declined modestly in both groups, with no significant between-group differences. No significant differences were observed in maternal emergency room utilization or adherence to well-baby visits. Engagement with Rosie was high, and qualitative feedback indicated that participants found the chatbot useful and supportive. CONCLUSIONS This pilot study suggests that a culturally and linguistically tailored AI-powered chatbot is feasible, acceptable, and shows preliminary promise for reducing anxiety and stress among Latina mothers during the perinatal period. Larger, adequately powered trials are needed to determine effectiveness and inform integration of AI-enabled tools into maternal and child health care to advance health equity.

Smartphone applications routinely collect precise location data from users by offering free services and this information is often monetized for advertising and marketing purposes. While many companies limit sales to aggregate behaviors, data aggregators and data brokers (DA&DB) provide access to individual-level location data. Some DA&DB have implemented privacy-preserving rules and the Federal Trade Commission (FTC) has also intensified its regulations on location-data practices. This paper presents an in-depth exploration of U.S. privacy perceptions regarding specific location features that can be derived from data made available by DA&DB. These findings can provide policy-relevant insights that could assist organizations such as the FTC in defining clearer access rules. Using a factorial vignette survey, we collected responses from 1,405 participants to assess comfort levels with sharing various types of location features, including individual trajectories and visits to points of interest, which are currently available for purchase from DA&DB worldwide. Our results show that trajectory-related features elicit higher privacy concerns, that certain data broker based obfuscation practices increase comfort, and that race, ethnicity and education have an effect on data sharing privacy perceptions.

Background: COVID-19 forecasting models have been used to inform decision-making around resource allocation and intervention decisions, such as hospital beds or stay-at-home orders. State-of-the-art forecasting models often use multimodal data, including mobility or sociodemographic data, to enhance COVID-19 case prediction models. Nevertheless, related work has revealed under-reporting bias in COVID-19 cases as well as sampling bias in mobility data for certain minority racial and ethnic groups, which affects the fairness of COVID-19 predictions across racial and ethnic groups. Objective: This study aims to introduce a fairness correction method that works for forecasting COVID-19 cases at an aggregate geographic level. Methods: We use hard and soft error parity analyses on existing fairness frameworks and demonstrate that our proposed method, Demographic Optimization (DemOpts), performs better in both scenarios. Results: We first demonstrate that state-of-the-art COVID-19 deep learning models produce mean prediction errors that are significantly different across racial and ethnic groups at larger geographic scales. We then propose a novel debiasing method, DemOpts, to increase the fairness of deep learning–based forecasting models trained on potentially biased datasets. Our results show that DemOpts can achieve better error parity than other state-of-the-art debiasing approaches, thus effectively reducing the differences in the mean error distributions across racial and ethnic groups. Conclusions: We introduce DemOpts, which reduces error parity differences compared with other approaches and generates fairer forecasting models compared with other approaches in the literature.

Objective. Demographic groups are often represented at different rates in medical datasets. These differences can create bias in machine learning algorithms, with higher levels of performance for better-represented groups. One promising solution to this problem is to generate synthetic data to mitigate potential adverse effects of non-representative data sets. Methods. We build on recent advances in LLM-based synthetic data generation to create a pipeline where the synthetic data is generated separately for each demographic group. We conduct our study using MIMIC-IV and Framingham "Offspring and OMNI-1 Cohorts" datasets. We prompt GPT4-Turbo to create group-specific data, providing training examples and the dataset context. An exploratory analysis is conducted to ascertain the quality of the generated data. We then evaluate the utility of the synthetic data for augmentation of a training dataset in a downstream machine learning task, focusing specifically on model performance metrics across groups. Results. The performance of GPT4-Turbo augmentation is generally superior but not always. In the majority of experiments our method outperforms standard modeling baselines, however, prompting GPT-4-Turbo to produce data specific to a group provides little to no additional benefit over a prompt that does not specify the group. Conclusion. We developed a method for using LLMs out-of-the-box to synthesize group-specific data to address imbalances in demographic representation in medical datasets. As another "tool in the toolbox", this method can improve model fairness and thus health equity. More research is needed to understand the conditions under which LLM generated synthetic data is useful for non-representative medical data sets.

Objectives: Given substantial obstacles surrounding health data acquisition, high-quality synthetic health data are needed to meet a growing demand for the application of advanced analytics for clinical discovery, prediction, and operational excellence. We highlight how recent advances in large language models (LLMs) present new opportunities for progress, as well as new risks, in synthetic health data generation (SHDG). Materials and Methods: We synthesized systematic scoping reviews in the SHDG domain, recent LLM methods for SHDG, and papers investigating the capabilities and limits of LLMs. Results: We summarize the current landscape of generative machine learning models (eg, Generative Adversarial Networks) for SHDG, describe remaining challenges and limitations, and identify how recent LLM approaches can potentially help mitigate them. Discussion: Six research directions are outlined for further investigation of LLMs for SHDG: evaluation metrics, LLM adoption, data efficiency, generalization, health equity, and regulatory challenges. Conclusion: LLMs have already demonstrated both high potential and risks in the health domain, and it is important to study their advantages and disadvantages for SHDG.

One of the central difficulties of addressing the COVID-19 pandemic has been accurately measuring and predicting the spread of infections. In particular, official COVID-19 case counts in the United States are under counts of actual caseloads due to the absence of universal testing policies. Researchers have proposed a variety of methods for recovering true caseloads, often through the estimation of statistical models on more reliable measures, such as death and hospitalization counts, positivity rates, and demographics. However, given the disproportionate impact of COVID-19 on marginalized racial, ethnic, and socioeconomic groups, it is important to consider potential unintended effects of case correction methods on these groups. Thus, we investigate two of these correction methods for their impact on a downstream COVID-19 case prediction task. For that purpose, we tailor an auditing approach and evaluation protocol to analyze the fairness of the COVID-19 prediction task by measuring the difference in model performance between majority-White counties and majority-minority counties. We find that one of the correction methods improves fairness, decreasing differences in performance between majority-White and majority-minority counties, while the other method increases differences, introducing bias. While these results are mixed, it is evident that correction methods have the potential to exacerbate existing biases in COVID-19 case data and in downstream prediction tasks. Researchers planning to develop or use case correction methods must be careful to consider negative effects on marginalized groups.