Saubhagya Joshi

Saubhagya Joshi

Auto-generated profile, based on publicly available information

Saubhagya Joshi

Rutgers, The State University of New Jersey, 4 Huntington Street, New Brunswick, NJ, 08901, United States, 1 848-932-7500

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
30 October 2024 in JMIR Formative Research
Verified
Journal Article
Open access
SJSaubhagya Joshi
Saubhagya Joshi
EHEunbin Ha
Eunbin Ha
AAAndee Amaya
Andee Amaya
MMMelissa Mendoza
Melissa Mendoza
YRYonaira Rivera
Yonaira Rivera
VSVivek Kumar Singh
Vivek Kumar Singh
Background: In the digital age, large language models (LLMs) like ChatGPT have emerged as important sources of health care information. Their interactive capabilities offer promise for enhancing health access, particularly for groups facing traditional barriers such as insurance and language constraints. Despite their growing public health use, with millions of medical queries processed weekly, the quality of LLM-provided information remains inconsistent. Previous studies have predominantly assessed ChatGPT’s English responses, overlooking the needs of non–English speakers in the United States. This study addresses this gap by evaluating the quality and linguistic parity of vaccination information from ChatGPT and the Centers for Disease Control and Prevention (CDC), emphasizing health equity. Objective: This study aims to assess the quality and language equity of vaccination information provided by ChatGPT and the CDC in English and Spanish. It highlights the critical need for cross-language evaluation to ensure equitable health information access for all linguistic groups. Methods: We conducted a comparative analysis of ChatGPT’s and CDC’s responses to frequently asked vaccination-related questions in both languages. The evaluation encompassed quantitative and qualitative assessments of accuracy, readability, and understandability. Accuracy was gauged by the perceived level of misinformation; readability, by the Flesch-Kincaid grade level and readability score; and understandability, by items from the National Institutes of Health’s Patient Education Materials Assessment Tool (PEMAT) instrument. Results: The study found that both ChatGPT and CDC provided mostly accurate and understandable (eg, scores over 95 out of 100) responses. However, Flesch-Kincaid grade levels often exceeded the American Medical Association’s recommended levels, particularly in English (eg, average grade level in English for ChatGPT=12.84, Spanish=7.93, recommended=6). CDC responses outperformed ChatGPT in readability across both languages. Notably, some Spanish responses appeared to be direct translations from English, leading to unnatural phrasing. The findings underscore the potential and challenges of using ChatGPT for health care access. Conclusions: ChatGPT holds potential as a health information resource but requires improvements in readability and linguistic equity to be truly effective for diverse populations. Crucially, the default user experience with ChatGPT, typically encountered by those without advanced language and prompting skills, can significantly shape health perceptions. This is vital from a public health standpoint, as the majority of users will interact with LLMs in their most accessible form. Ensuring that default responses are accurate, understandable, and equitable is imperative for fostering informed health decisions across diverse communities.

20 February 2026 in JMIR AI
Verified
Journal Article
Open access
SJSaubhagya Joshi
Saubhagya Joshi
MMMonjil Mehta
Monjil Mehta
SMSarjak Maniar
Sarjak Maniar
MWMengqian Wang
Mengqian Wang
VSVivek Kumar Singh
Vivek Kumar Singh
Background: Large language models (LLMs) are increasingly integrated into health care, where they contribute to patient care, administrative efficiency, and clinical decision-making. Despite their growing role, the ability of LLMs to handle imperfect inputs remains underexplored. These imperfections, which are common in clinical documentation and patient-generated data, may affect model reliability. Objective: This study investigates the impact of input perturbations on LLM performance across three dimensions: (1) overall effectiveness in different health-related applications, (2) comparative effects of different types and levels of perturbations, and (3) differential impact of perturbations on health-related terms versus non–health-related terms. Methods: We systematically evaluate 3 LLMs on 3 health-related tasks using a novel dataset containing 3 types of human-like variations (redaction, homophones, and typographical errors) at different perturbation levels. Results: Contrary to expectations, LLMs demonstrate notable robustness to common variations, and in more than half of the cases (151/270, 55.92%), the performance was stable or improved. In some cases (38/270, 14.07%), variations resulted in an increased performance, especially when dealing with lower perturbation levels. Redactions, often stemming from privacy concerns or cognitive lapses, are more detrimental than other variations. Conclusions: Our findings highlight the need for health care applications powered by LLMs to be designed with input variability in mind. Robustness to noisy or imperfect inputs is essential for maintaining reliability in real-world clinical settings, where data quality can vary widely. By identifying specific vulnerabilities and strengths, this study provides actionable insights for improving model resilience and guiding the development of safer, more effective artificial intelligence tools in health care. The accompanying dataset offers a valuable resource for further research into LLM performance under diverse conditions.

26 May 2026 in JMIR Formative Research
Verified
Journal Article
Open access
SJSaubhagya Joshi
Saubhagya Joshi
MMMonjil A Mehta
Monjil A Mehta
MMMelissa Mendoza
Melissa Mendoza
YRYonaira M Rivera
Yonaira M Rivera
VSVivek Kumar Singh
Vivek Kumar Singh
Background: Large language models (LLMs) such as ChatGPT are increasingly used to support health-related queries and decision-making. However, these models can be “jailbroken” through adversarial prompts that bypass safety filters and elicit harmful or medically inappropriate responses. In health care contexts, such vulnerabilities pose serious risks. Understanding how jailbreak susceptibility varies across languages is essential for developing robust safeguards and promoting equitable access to safe health information. This paper may contain examples that may be deemed harmful in terms of violence, self-harm, and drug abuse. Objective: This study aims to systematically compare and contrast the vulnerability of a health LLM for jailbreaking across 3 languages: English, Spanish, and Hindi (transliterated using the Latin alphabet), based on emoji and permutation cipher attacks. Methods: We analyzed 1000 input prompts per language, drawn from the BeaverTails dataset, across 3 harm categories: self-harm, violence, and drug abuse. Each prompt was modified using emoji and permutation cipher techniques, resulting in 6000 input-output pairs. Model responses were evaluated by human coders to determine the success rate of jailbreak attempts across languages and cipher types. Results: Hindi prompts showed the highest vulnerability, with 787 successful jailbreaks using emoji ciphers and 873 using permutation ciphers. Spanish and English followed, with lower success rates across both cipher types. Differences in jailbreak success across languages and cipher strategies were statistically significant. Additionally, attacks targeting violence-related prompts were more successful overall than those targeting drug-related or self-harm content, indicating variation in vulnerability by harm type. Conclusions: The findings of this formative study reveal that LLM safety performance varies substantially across languages and harm categories, raising concerns about equitable protection in multilingual health communication. Disparities in access to harmful content may contribute to downstream health risks. Strengthening multilingual content moderation and developing language-aware safety mechanisms are critical steps toward creating safer and more inclusive health AI systems.

5 September 2025 in JMIR Preprints
Verified
Preprint
Open access
SJSaubhagya Joshi
Saubhagya Joshi
MMMonjil Mehta
Monjil Mehta
SMSarjak Maniar
Sarjak Maniar
MWMengqian Wang
Mengqian Wang
VSVivek Kumar Singh
Vivek Kumar Singh
BACKGROUND Large Language Models (LLMs) are increasingly integrated into healthcare, where they contribute to patient care, administrative efficiency, and clinical decision-making. Despite their growing role, the ability of LLMs to handle imperfect inputs remains underexplored. These imperfections, which are common in clinical documentation and patient-generated data, may affect model reliability. OBJECTIVE This study investigates the impact of input perturbations on LLM performance across three dimensions: (1) overall effectiveness in different health-related applications, (2) comparative effects of different types and levels of perturbations, and (3) differential impact of perturbations on health-related terms versus non-health-related terms. METHODS We systematically evaluate three LLMs on three health-related tasks using a novel dataset containing three types of human-like variations (redaction, homophones, and typographical errors) at different perturbation levels. RESULTS Contrary to expectations, LLMs demonstrate notable robustness to common variations, with some cases showing improved performance at lower perturbation levels. Redactions, often stemming from privacy concerns or cognitive lapses, are more detrimental than other variations. CONCLUSIONS Our findings highlight the need for healthcare applications powered by LLMs to be designed with input variability in mind. Robustness to noisy or imperfect inputs is essential for maintaining reliability in real-world clinical settings, where data quality can vary widely. By identifying specific vulnerabilities and strengths, this work provides actionable insights for improving model resilience and guiding the development of safer, more effective AI tools in healthcare. The accompanying dataset offers a valuable resource for further research into LLM performance under diverse conditions. CLINICALTRIAL (N/A)

26 May 2024 in JMIR Preprints
Verified
Preprint
Open access
SJSaubhagya Joshi
Saubhagya Joshi
EHEunbin Ha
Eunbin Ha
AAAndee Amaya
Andee Amaya
MMMelissa Mendoza
Melissa Mendoza
YRYonaira Rivera
Yonaira Rivera
VSVivek Kumar Singh
Vivek Kumar Singh
BACKGROUND In the digital age, large language models (LLMs) like ChatGPT have emerged as important sources of health care information. Their interactive capabilities offer promise for enhancing health access, particularly for groups facing traditional barriers such as insurance and language constraints. Despite their growing public health use, with millions of medical queries processed weekly, the quality of LLM-provided information remains inconsistent. Previous studies have predominantly assessed ChatGPT’s English responses, overlooking the needs of non–English speakers in the United States. This study addresses this gap by evaluating the quality and linguistic parity of vaccination information from ChatGPT and the Centers for Disease Control and Prevention (CDC), emphasizing health equity. OBJECTIVE This study aims to assess the quality and language equity of vaccination information provided by ChatGPT and the CDC in English and Spanish. It highlights the critical need for cross-language evaluation to ensure equitable health information access for all linguistic groups. METHODS We conducted a comparative analysis of ChatGPT’s and CDC’s responses to frequently asked vaccination-related questions in both languages. The evaluation encompassed quantitative and qualitative assessments of accuracy, readability, and understandability. Accuracy was gauged by the perceived level of misinformation; readability, by the Flesch-Kincaid grade level and readability score; and understandability, by items from the National Institutes of Health’s Patient Education Materials Assessment Tool (PEMAT) instrument. RESULTS The study found that both ChatGPT and CDC provided mostly accurate and understandable (eg, scores over 95 out of 100) responses. However, Flesch-Kincaid grade levels often exceeded the American Medical Association’s recommended levels, particularly in English (eg, average grade level in English for ChatGPT=12.84, Spanish=7.93, recommended=6). CDC responses outperformed ChatGPT in readability across both languages. Notably, some Spanish responses appeared to be direct translations from English, leading to unnatural phrasing. The findings underscore the potential and challenges of using ChatGPT for health care access. CONCLUSIONS ChatGPT holds potential as a health information resource but requires improvements in readability and linguistic equity to be truly effective for diverse populations. Crucially, the default user experience with ChatGPT, typically encountered by those without advanced language and prompting skills, can significantly shape health perceptions. This is vital from a public health standpoint, as the majority of users will interact with LLMs in their most accessible form. Ensuring that default responses are accurate, understandable, and equitable is imperative for fostering informed health decisions across diverse communities.

9 November 2025 in JMIR Preprints
Verified
Preprint
Open access
SJSaubhagya Joshi
Saubhagya Joshi
MMMelissa Mendoza
Melissa Mendoza
YRYonaira Rivera
Yonaira Rivera
VSVivek K. Singh
Vivek K. Singh
BACKGROUND Large language models (LLMs) such as ChatGPT are increasingly used to support health-related queries and decision-making. However, these models can be “jailbroken” through adversarial prompts that bypass safety filters and elicit harmful or medically inappropriate responses. In healthcare contexts, such vulnerabilities pose serious risks. Understanding how jailbreak susceptibility varies across languages is essential for developing robust safeguards and promoting equitable access to safe health information. OBJECTIVE This study aims to systematically compare and contrast the vulnerability of a health LLM for jailbreaking across three languages: English, Spanish, and Hindi (transliterated using the Latin alphabet) based on emoji and permutation cipher attacks. METHODS We analyzed 1,000 input prompts per language, drawn from the BeaverTails dataset, across three harm categories: self-harm, violence, and drug abuse. Each prompt was modified using emoji and permutation cipher techniques, resulting in 6,000 input-output pairs. Model responses were evaluated by human coders to determine the success rate of jailbreak attempts across languages and cipher types. RESULTS Hindi prompts showed the highest vulnerability, with 787 successful jailbreaks using emoji ciphers and 873 using permutation ciphers. Spanish and English followed, with lower success rates across both cipher types. Differences in jailbreak success across languages and cipher strategies were statistically significant. Additionally, attacks targeting violence-related prompts were more successful overall than those targeting drug-related or self-harm content, indicating variation in vulnerability by harm type. CONCLUSIONS The findings of this formative study reveal that LLM safety performance varies substantially across languages and harm categories, raising concerns about equitable protection in multilingual health communication. Disparities in access to harmful content may contribute to downstream health risks. Strengthening multilingual content moderation and developing language-aware safety mechanisms are critical steps toward safer and more inclusive health AI systems.