Akshay Swaminathan

ORCID: 0000-0003-3426-9289
Publications
Citations
Views
---
Saved
---
About
Contact & Profiles
Research Areas
  • Child Nutrition and Water Access
  • Machine Learning in Healthcare
  • Topic Modeling
  • Artificial Intelligence in Healthcare and Education
  • Cancer Genomics and Diagnostics
  • Natural Language Processing Techniques
  • Biomedical Text Mining and Ontologies
  • Global Maternal and Child Health
  • COVID-19 epidemiological studies
  • Poverty, Education, and Child Welfare
  • Food Security and Health in Diverse Populations
  • Health Systems, Economic Evaluations, Quality of Life
  • Birth, Development, and Health
  • Digital Mental Health Interventions
  • Gestational Diabetes Research and Management
  • Artificial Intelligence in Healthcare
  • Big Data and Business Intelligence
  • Crop Yield and Soil Fertility
  • Genetic factors in colorectal cancer
  • AI in cancer detection
  • Statistical Methods in Clinical Trials
  • Speech and dialogue systems
  • Agriculture Sustainability and Environmental Impact
  • Agronomic Practices and Intercropping Systems
  • Radiomics and Machine Learning in Medical Imaging

Stanford University
2019-2025

Stanford Medicine
2025

Center for Innovation
2024

Keck Hospital of USC
2024

Harvard University
2019-2024

Santa Clara Valley Medical Center
2024

Menlo School
2024

Harvard University Press
2024

University of Otago
2024

Palo Alto University
2024

Abstract Large language models (LLMs) with retrieval-augmented generation (RAG) have improved information extraction over previous methods, yet their reliance on embeddings often leads to inefficient retrieval. We introduce CLinical Entity Augmented Retrieval (CLEAR), a RAG pipeline that retrieves using entities. compared CLEAR embedding and full-note approaches for extracting 18 variables six LLMs across 20,000 clinical notes. Average F1 scores were 0.90, 0.86, 0.79; inference times 4.95,...

10.1038/s41746-024-01377-1 article EN cc-by npj Digital Medicine 2025-01-19

Abstract Because meat is more resource intensive than vegetal protein sources, replacing it with efficient plant alternatives potentially desirable, provided these prove nutritionally sound. We show that conserving to rigorously satisfy key nutritional constraints while minimizing cropland, nitrogen fertilizer (Nr) and water use greenhouse gas (GHG) emissions exist, could improve public health. develop a new methodology for identifying whose satisfaction by eaters challenging,...

10.1038/s41598-019-46590-1 article EN cc-by Scientific Reports 2019-08-08

Gestational diabetes is common in pregnancy and associated with adverse fetal outcomes. Currently, population-based data on the prevalence of gestational are limited India.To provide a comprehensive national assessment India its socioeconomic, demographic, geographic associations, using elevated random blood glucose as proxy for diagnosis.This cross-sectional study analyzed fourth National Family Health Survey, conducted between January 2015 December 2016. This nationally representative...

10.1001/jamanetworkopen.2020.25074 article EN cc-by-nc-nd JAMA Network Open 2020-11-09

Despite growing interest in using large language models (LLMs) healthcare, current explorations do not assess the real-world utility and safety of LLMs clinical settings. Our objective was to determine whether two can serve information needs submitted by physicians as questions an informatics consultation service a safe concordant manner. Sixty six from consult were GPT-3.5 GPT-4 via simple prompts. 12 assessed LLM responses' possibility patient harm concordance with existing reports...

10.48550/arxiv.2304.13714 preprint EN cc-by arXiv (Cornell University) 2023-01-01

Patients experiencing mental health crises often seek help through messaging-based platforms, but may face long wait times due to limited message triage capacity. Here we build and deploy a machine-learning-enabled system improve response crisis messages in large, national telehealth provider network. We train two-stage natural language processing (NLP) with key word filtering followed by logistic regression on 721 electronic medical record chat messages, of which 32% are potential...

10.1038/s41746-023-00951-3 article EN cc-by npj Digital Medicine 2023-11-21

Agriculture is the backbone of a country in terms economy and survival people.To maintain high efficiency crop production we look to avoid plant diseases.The proposed algorithm optimize information from resources available us for betterment result without any complexity.The neural network used classification Dense Convolution Neural Network (DCNN).In this project, pre-trained model (densenet-121) which imported keras library has been training.A convolution may be simple application filter an...

10.34218/ijeet.12.5.2021.005 article EN INTERNATIONAL JOURNAL OF ELECTRICAL ENGINEERING & TECHNOLOGY 2021-05-24

0. Abstract Background The integration of large language models (LLMs) in healthcare offers immense opportunity to streamline tasks, but also carries risks such as response accuracy and bias perpetration. To address this, we conducted a red-teaming exercise assess LLMs developed dataset clinically relevant scenarios for future teams use. Methods We convened 80 multi-disciplinary experts evaluate the performance popular across multiple medical scenarios. Teams composed clinicians, engineering...

10.1101/2024.04.05.24305411 preprint EN cc-by-nc-nd medRxiv (Cold Spring Harbor Laboratory) 2024-04-07

BackgroundAbout 3 million stillbirths occur each year, 98% of which are in low-income and middle-income countries (LMICs). Interpregnancy interval is a key risk factor interest, because it modifiable. We aimed to investigate whether there causal relationship between the length interpregnancy subsequent stillbirth.MethodsWe used Demographic Health Surveys (2002–18) from 58 LMICs study reproductive histories women identify livebirths preceding 5 years. Countries were selected on basis...

10.1016/s2214-109x(19)30458-9 article EN cc-by The Lancet Global Health 2019-12-12

Methods to ensure factual accuracy of text generated by large language models (LLM) in clinical medicine are lacking. VeriFact is an artificial intelligence system that combines retrieval-augmented generation and LLM-as-a-Judge verify whether LLM-generated factually supported a patient's medical history based on their electronic health record (EHR). To evaluate this system, we introduce VeriFact-BHC, new dataset decomposes Brief Hospital Course narratives from discharge summaries into set...

10.48550/arxiv.2501.16672 preprint EN arXiv (Cornell University) 2025-01-27

Red teaming, the practice of adversarially exposing unexpected or undesired model behaviors, is critical towards improving equity and accuracy large language models, but non-model creator-affiliated red teaming scant in healthcare. We convened teams clinicians, medical engineering students, technical professionals (80 participants total) to stress-test models with real-world clinical cases categorize inappropriate responses along axes safety, privacy, hallucinations/accuracy, bias. Six...

10.1038/s41746-025-01542-0 article EN cc-by npj Digital Medicine 2025-03-07

Background: Schools play a central role in addressing the mental health needs of young people. Here, we analyze data from school-based teletherapy intervention delivered two socioeconomically vulnerable school districts to characterize sociodemographics referred students, identify variables associated with session completion, and assess associations attendance, grade point average (GPA), disciplinary actions. Methods: Guidance counselors healthcare professionals eligible students for...

10.1101/2025.03.11.25323793 preprint EN cc-by-nc-nd medRxiv (Cold Spring Harbor Laboratory) 2025-03-13

Since the onset of COVID-19 pandemic, countless disease prediction models have emerged, shaping focus news media, policymakers, and broader society. We reviewed accuracy forecasts made during prior twenty-first century epidemics, namely SARS, H1N1, Ebola. found that while were relatively nascent as a research SARS for Ebola, numerous such published. deaths Ebola often far from eventual reality, with strong tendency to over predict. Given societal prominence these models, it is crucial their...

10.1007/s10654-020-00661-0 article EN other-oa European Journal of Epidemiology 2020-07-17

There are emerging opportunities to assess health indicators at truly small areas with increasing availability of data geocoded micro geographic units and advanced modeling techniques. The utility such fine-grained can be fully leveraged if linked local governance that accountable for implementation programs interventions. We used from the 2011 Indian Census village-level demographic amenities features 2016 Demographic Health Survey in a bias-corrected semisupervised regression framework...

10.1073/pnas.2025865118 article EN cc-by Proceedings of the National Academy of Sciences 2021-04-26

Over 250 million children in developing countries are at risk of not achieving their developmental potential, and unlikely to receive timely interventions because existing assessments that help identify who faltering prohibitive for use low resource contexts. To bridge this "detection gap," we developed a tablet-based, gamified cognitive assessment tool named DEvelopmental on an E-Platform (DEEP), which is feasible delivery by non-specialists rural Indian households acceptable all end-users....

10.3389/fpsyg.2020.01202 article EN cc-by Frontiers in Psychology 2020-06-10

In India, data on key developmental indicators used to formulate policies and interventions are routinely available for the administrative unit of districts but not political parliamentary constituencies (PC). Recently, Swaminathan et al. proposed two methodologies generate PC estimates using randomly displaced GPS locations sampling clusters ('direct') by building a crosswalk between PCs boundary shapefiles ('indirect'). We advance these precision-weighted estimations based hierarchical...

10.1016/j.ssmph.2019.100375 article EN cc-by-nc-nd SSM - Population Health 2019-02-10

7048 Background: Castleman disease (CD) has three subtypes: Unicentric (UCD), Human herpesvirus-8 associated multicentric (HHV-8 MCD) and idiopathic (iMCD). Outcomes for patients with iMCD are poor treatment options limited, only one FDA-approved therapy (siltuximab in April 2014). Further, the lack of CD-specific ICD codes until 2017 limited real-world evaluation. We identified an electronic health record (EHR)-derived dataset described their clinical characteristics, patterns, overall...

10.1200/jco.2021.39.15_suppl.7048 article EN Journal of Clinical Oncology 2021-05-20

Aim: To assess concordance between HER2 status measured by traditional methods and ERBB2 amplification next-generation sequencing its association with first-line trastuzumab clinical benefit in patients advanced esophagogastric cancer. Methods: Retrospective analysis of HER2/ERBB2 using a deidentified USA-based clinicogenomic database. Clinical outcomes were assessed for HER2+ cancer who received trastuzumab. Results: Overall was 87.5%. Among trastuzumab, concordant associated longer time to...

10.2217/fon-2021-0203 article EN cc-by-nc-nd Future Oncology 2021-08-31

Background Education is considered one of the most robust determinants health. However, it unclear whether maternal education and paternal have differential impacts on perinatal health outcomes. We assess differences their association with adverse birth outcomes in a large cohort from Ontario, Canada.Methods The OaK Birth Cohort recruited patients Canada, between October 2002 April 2009. mothers were 12 20 weeks' gestation collected both mother infant data. final sample size was 8,085...

10.1080/14767058.2022.2049751 article EN The Journal of Maternal-Fetal & Neonatal Medicine 2022-03-14

Abstract Objective While there are currently approaches to handle unstructured clinical data, such as manual abstraction and structured proxy variables, these methods may be time-consuming, not scalable, imprecise. This article aims determine whether selective prediction, which gives a model the option abstain from generating can improve accuracy efficiency of data abstraction. Materials Methods We trained classifiers (logistic regression, random forest, support vector machine) extract 5...

10.1093/jamia/ocad182 article EN cc-by-nc Journal of the American Medical Informatics Association 2023-08-24
Coming Soon ...