Nimit Desai

All Publications

• An Artificial Intelligence-Enabled Cardiopulmonary Resuscitation Instructor. JAMA internal medicine Desai, N., Majhail, N., Dredze, M., Hogarth, M., Smith, D., Callaway, C., Dezfulian, C., Padman, R., Kochanek, P. M., Ayers, J. W., Horvat, C. M. 2026; 186 (7): 829-836

  Abstract

  High-quality, timely cardiopulmonary resuscitation (CPR) after out-of-hospital cardiac arrest (OHCA) is vital, but bystanders need help to act. Telecommunicator-CPR by 911 dispatchers promotes bystander CPR but may be limited by inherent variability and delays in human-led systems.To evaluate the performance of widely available artificial intelligence (AI) models in delivering guideline-concordant CPR instruction and to develop and assess ChatCPR, a purpose-built AI CPR instructor agent in an early proof-of-concept study design.In this cross-sectional study, simulated emergency scenarios were used to evaluate baseline CPR instructing capabilities of widely available AI models (ChatGPT, Claude, Gemini, Grok, Llama, and Mistral). Models were evaluated for their adherence to minimally viable criteria, which included instructions like performing appropriate-depth chest compressions, and maximally effective criteria, which included more nuanced instructions like ensuring that compressions achieved full recoil, all derived from major association CPR guidelines. These insights informed development of the purpose-built AI CPR instructor agent that was subsequently evaluated using standard test-retest methods across the same simulated scenarios and 911 dispatcher-assisted calls where CPR was indicated. Data were analyzed from April through December 2025.In simulated OHCA scenarios among 6 widely available AI models, the models achieved 89.7% (95% CI, 84.8%-93.2%) of minimally viable criteria, ranging from 79.4% (95% CI, 63.2%-89.7%) for Gemini to 97.1% (95% CI, 85.1%-99.5%) for Grok and Claude. Models achieved a mean of 69.8% (95% CI, 65.5%-73.7%) of maximally effective criteria, ranging from 61.3% (95% CI, 50.3%-71.2%) for Llama to 75.0% (95% CI, 64.5%-83.2%) for GPT-4o. The instructor agent achieved 100% (95% CI, 89.8%-100%) and 100% (95% CI, 95.4%-100%) adherence to minimally viable and maximally effective criteria, respectively, for the same scenarios. In a retest using 911 calls, the agent achieved 100% (95% CI, 88.6%-100%) adherence to minimally viable and 98.9% (95% CI, 94.0%-99.8%) adherence to maximally effective criteria, representing absolute improvements of 15.5 percentage points (95% CI, 2.1-27.4 percentage points; P = .02) and 36.1 percentage points (95% CI, 27.2-44.3 percentage points; P < .001) over dispatchers (84.5%; 95% CI, 73.1%-91.6% and 62.8%; 95% CI, 55.0%-70.0%, respectively).These findings suggest that AI-enabled CPR instruction shows promise for supporting bystanders. Further validation in diverse, general population settings is warranted to define the role of AI-based CPR instruction as a scalable public health intervention for OHCAs.

  View details for DOI 10.1001/jamainternmed.2026.1552

  View details for PubMedID 42149572

  View details for PubMedCentralID PMC13184806

• Growing Health Concern Regarding Gambling Addiction in the Age of Sportsbooks. JAMA internal medicine Yeola, A., Allen, M. R., Desai, N., Poliak, A., Yang, K. H., Smith, D. M., Ayers, J. W. 2025; 185 (4): 382-389

  Abstract

  The US Supreme Court decision Murphy v National Collegiate Athletic Association allowed states beyond Nevada to legalize sports betting, including online wagers. How sports betting has evolved and its association with gambling harms has not been studied.To describe how US sports betting evolved after Murphy v National Collegiate Athletic Association and offer insights into the potential health effects of sportsbooks, which are platforms for wagering on sporting events.Enactment of (1) Murphy v National Collegiate Athletic Association nationally and (2) the opening of retail or online sportsbooks in states.In this longitudinal study, aggregate US internet search trends for gambling addiction and wagers on sports were described before and after the emergence of legalized sportsbooks.Internet searches per 10 million queries that mentioned gambling and addiction, addict, anonymous, or hotline (such as gambling addiction hotline) made to Google from January 1, 2016, through June 30, 2024.The number of states with operational sportsbooks increased from 1 during 2017 to 38 during 2024. Total sports wagers increased from $4.9 billion during 2017 to $121.1 billion during 2023, with 94% of wagers during 2023 being placed online. There were 23% (95% CI, 15%-30%) more searches nationally for gambling addiction help-seeking after Murphy v National Collegiate Athletic Association. Massachusetts (47%; 95% CI, 21%-79%), New Jersey (34%; 95% CI, 21%-45%), New York (37%; 95% CI, 26%-50%), and Pennsylvania (50%; 95% CI, 35%-66%) each had more searches than expected after the opening of any sportsbooks in their state. Additional analyses suggest the opening of online, vs retail, sportsbooks corresponded with a larger increase in searches.The results of this time series study suggest that access to sportsbooks, sports wagers, and potential help-seeking for gambling addiction increased substantially and highlight the need to address the health implications of sportsbooks, including recognition and treatment of gambling problems and their broader societal implications.

  View details for DOI 10.1001/jamainternmed.2024.8193

  View details for PubMedID 39960737

  View details for PubMedCentralID PMC11833657

• Using Large Language Models to Support Content Analysis: A Case Study of ChatGPT for Adverse Event Detection. Journal of medical Internet research Leas, E. C., Ayers, J. W., Desai, N., Dredze, M., Hogarth, M., Smith, D. M. 2024; 26: e52499

  Abstract

  This study explores the potential of using large language models to assist content analysis by conducting a case study to identify adverse events (AEs) in social media posts. The case study compares ChatGPT's performance with human annotators' in detecting AEs associated with delta-8-tetrahydrocannabinol, a cannabis-derived product. Using the identical instructions given to human annotators, ChatGPT closely approximated human results, with a high degree of agreement noted: 94.4% (9436/10,000) for any AE detection (Fleiss kappa=0.95) and 99.3% (9931/10,000) for serious AEs (kappa=0.96). These findings suggest that ChatGPT has the potential to replicate human annotation accurately and efficiently. The study recognizes possible limitations, including concerns about the generalizability due to ChatGPT's training data, and prompts further research with different models, data sources, and content analysis tasks. The study highlights the promise of large language models for enhancing the efficiency of biomedical research.

  View details for DOI 10.2196/52499

  View details for PubMedID 38696245

• Characteristics of X (Formerly Twitter) Community Notes Addressing COVID-19 Vaccine Misinformation. JAMA Allen, M. R., Desai, N., Namazi, A., Leas, E., Dredze, M., Smith, D. M., Ayers, J. W. 2024

  Abstract

  This study evaluated the topics, accuracy, and credibility of X (formerly Twitter) Community Notes addressing COVID-19 vaccination.

  View details for DOI 10.1001/jama.2024.4800

  View details for PubMedID 38656757

• AI-Generated Draft Replies Integrated Into Health Records and Physicians' Electronic Communication. JAMA network open Tai-Seale, M., Baxter, S. L., Vaida, F., Walker, A., Sitapati, A. M., Osborne, C., Diaz, J., Desai, N., Webb, S., Polston, G., Helsten, T., Gross, E., Thackaberry, J., Mandvi, A., Lillie, D., Li, S., Gin, G., Achar, S., Hofflich, H., Sharp, C., Millen, M., Longhurst, C. A. 2024; 7 (4): e246565

  Abstract

  Importance: Timely tests are warranted to assess the association between generative artificial intelligence (GenAI) use and physicians' work efforts.Objective: To investigate the association between GenAI-drafted replies for patient messages and physician time spent on answering messages and the length of replies.Design, Setting, and Participants: Randomized waiting list quality improvement (QI) study from June to August 2023 in an academic health system. Primary care physicians were randomized to an immediate activation group and a delayed activation group. Data were analyzed from August to November 2023.Exposure: Access to GenAI-drafted replies for patient messages.Main Outcomes and Measures: Time spent (1) reading messages, (2) replying to messages, (3) length of replies, and (4) physician likelihood to recommend GenAI drafts. The a priori hypothesis was that GenAI drafts would be associated with less physician time spent reading and replying to messages. A mixed-effects model was used.Results: Fifty-two physicians participated in this QI study, with 25 randomized to the immediate activation group and 27 randomized to the delayed activation group. A contemporary control group included 70 physicians. There were 18 female participants (72.0%) in the immediate group and 17 female participants (63.0%) in the delayed group; the median age range was 35-44 years in the immediate group and 45-54 years in the delayed group. The median (IQR) time spent reading messages in the immediate group was 26 (11-69) seconds at baseline, 31 (15-70) seconds 3 weeks after entry to the intervention, and 31 (14-70) seconds 6 weeks after entry. The delayed group's median (IQR) read time was 25 (10-67) seconds at baseline, 29 (11-77) seconds during the 3-week waiting period, and 32 (15-72) seconds 3 weeks after entry to the intervention. The contemporary control group's median (IQR) read times were 21 (9-54), 22 (9-63), and 23 (9-60) seconds in corresponding periods. The estimated association of GenAI was a 21.8% increase in read time (95% CI, 5.2% to 41.0%; P=.008), a -5.9% change in reply time (95% CI, -16.6% to 6.2%; P=.33), and a 17.9% increase in reply length (95% CI, 10.1% to 26.2%; P<.001). Participants recognized GenAI's value and suggested areas for improvement.Conclusions and Relevance: In this QI study, GenAI-drafted replies were associated with significantly increased read time, no change in reply time, significantly increased reply length, and some perceived benefits. Rigorous empirical tests are necessary to further examine GenAI's performance. Future studies should examine patient experience and compare multiple GenAIs, including those with medical training.

  View details for DOI 10.1001/jamanetworkopen.2024.6565

  View details for PubMedID 38619840

• Regulate Artificial Intelligence in Health Care by Prioritizing Patient Outcomes. JAMA Ayers, J. W., Desai, N., Smith, D. M. 2024; 331 (8): 639-640

  View details for DOI 10.1001/jama.2024.0549

  View details for PubMedID 38285467

• C-X-C Motif Chemokine Receptor 3 Splice Variants Differentially Activate Beta-Arrestins to Regulate Downstream Signaling Pathways. Molecular pharmacology Smith, J. S., Alagesan, P., Desai, N. K., Pack, T. F., Wu, J. H., Inoue, A., Freedman, N. J., Rajagopal, S. 2017; 92 (2): 136-150

  Abstract

  Biased agonism, the ability of different ligands for the same receptor to selectively activate some signaling pathways while blocking others, is now an established paradigm for G protein-coupled receptor signaling. One group of receptors in which endogenous bias is critical is the chemokine system, consisting of over 50 ligands and 20 receptors that bind one another with significant promiscuity. We have previously demonstrated that ligands for the same receptor can cause biased signaling responses. The goal of this study was to identify mechanisms that could underlie biased signaling between different receptor splice variants. The C-X-C motif chemokine receptor 3 (CXCR3) has two splice variants, CXCR3A and CXCR3B, which differ by 51 amino acids at its N-terminus. Consistent with an earlier study, we found that C-X-C motif chemokine ligands 4, 9, 10, and 11 all activated G αi at CXCR3A, while at CXCR3B these ligands demonstrated no measurable G αi or G αs activity. β-arrestin (βarr) was recruited at a reduced level to CXCR3B relative to CXCR3A, which was also associated with differences in βarr2 conformation. βarr2 recruitment to CXCR3A was attenuated by both G protein receptor kinase (GRK) 2/3 and GRK5/6 knockdown, while only GRK2/3 knockdown blunted recruitment to CXCR3B. Extracellular regulated kinase 1/2 phosphorylation downstream from CXCR3A and CXCR3B was increased and decreased, respectively, by βarr1/2 knockout. The splice variants also differentially activated transcriptional reporters. These findings demonstrate that differential splicing of CXCR3 results in biased responses associated with distinct patterns of βarr conformation and recruitment. Differential splicing may serve as a common mechanism for generating biased signaling and provides insights into how chemokine receptor signaling can be modulated post-transcriptionally.

  View details for DOI 10.1124/mol.117.108522

  View details for PubMedID 28559424

  View details for PubMedCentralID PMC5508197