AWAbdi D. Wakene

Abdi D. Wakene

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Abdi D. Wakene has a total of 9 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Top Cited Papers
    Open Access

    Assessing Racial and Ethnic Bias in Text Generation by Large Language Models for Health Care–Related Tasks: Cross-Sectional Study

    by John J Hanna, Abdi D Wakene, Andrew O Johnson, Christoph U Lehmann, Richard J Medford

    Background: Racial and ethnic bias in large language models (LLMs) used for health care tasks is a growing concern, as it may contribute to health disparities. In response, LLM operators implemented safeguards against prompts that are overtly seeking certain biases. Objective: This study aims to investigate a potential racial and ethnic bias among 4 popular LLMs: GPT-3.5-turbo (OpenAI), GPT-4 (OpenAI), Gemini-1.0-pro (Google), and Llama3-70b (Meta) in generating health care consumer–directed text in the absence of overtly biased queries. Methods: In this cross-sectional study, the 4 LLMs were prompted to generate discharge instructions for patients with HIV. Each patient’s encounter deidentified metadata including race/ethnicity as a variable was passed over in a table format through a prompt 4 times, altering only the race/ethnicity information (African American, Asian, Hispanic White, and non-Hispanic White) each time, while keeping all other information constant. The prompt requested the model to write discharge instructions for each encounter without explicitly mentioning race or ethnicity. The LLM-generated instructions were analyzed for sentiment, subjectivity, reading ease, and word frequency by race/ethnicity. Results: The only observed statistically significant difference between race/ethnicity groups was found in entity count (GPT-4, df=42, P=.047). However, post hoc chi-square analysis for GPT-4’s entity counts showed no significant pairwise differences among race/ethnicity categories after Bonferroni correction. Conclusions: A total of 4 LLMs were relatively invariant to race/ethnicity in terms of linguistic and readability measures. While our study used proxy linguistic and readability measures to investigate racial and ethnic bias among 4 LLM responses in a health care–related task, there is an urgent need to establish universally accepted standards for measuring bias in LLM-generated responses. Further studies are needed to validate these results and assess their implications.


  • Open Access

    A pandemic of COVID-19 mis- and disinformation: manual and automatic topic analysis of the literature

    by Abdi D. Wakene, Lauren N. Cooper, John J. Hanna, Trish M. Perl, Christoph U. Lehmann, Richard J. Medford

    Objective: Social media’s arrival eased the sharing of mis- and disinformation. False information proved challenging throughout the coronavirus disease 2019 (COVID-19) pandemic with many clinicians and researchers analyzing the “infodemic.” We systemically reviewed and synthesized COVID-19 mis- and disinformation literature, identifying the prevalence and content of false information and exploring mitigation and prevention strategies. Design: We identified and analyzed publications on COVID-19-related mis- and disinformation published from March 1, 2020, to December 31, 2022, in PubMed. We performed a manual topic review of the abstracts along with automated topic modeling to organize and compare the different themes. We also conducted sentiment (ranked −3 to +3) and emotion analysis (rated as predominately happy, sad, angry, surprised, or fearful) of the abstracts. Results: We reviewed 868 peer-reviewed scientific publications of which 639 (74%) had abstracts available for automatic topic modeling and sentiment analysis. More than a third of publications described mitigation and prevention-related issues. The mean sentiment score for the publications was 0.685, and 56% of studies had a negative sentiment (fear and sadness as the most common emotions). Conclusions: Our comprehensive analysis reveals a significant proliferation of dis- and misinformation research during the COVID-19 pandemic. Our study illustrates the pivotal role of social media in amplifying false information. Research into the infodemic was characterized by negative sentiments. Combining manual and automated topic modeling provided a nuanced understanding of the complexities of COVID-19-related misinformation, highlighting themes such as the source and effect of misinformation, and strategies for mitigation and prevention.


  • Open Access

    Mortality in hospitalized SARS-CoV-2 patients with contemporaneous bacterial and fungal infections

    by John J. Hanna, Zachary M. Most, Lauren N. Cooper, Abdi D. Wakene, Alexander P. Radunsky, Christoph U. Lehmann, Trish M. Perl, Richard J. Medford

    Background: The interplay between SARS-CoV-2 and contemporaneous bacterial or fungal culture growth may have crucial implications for clinical outcomes of hospitalized patients. This study aimed to quantify the effect of microbiological culture positivity on mortality among hospitalized patients with SARS-CoV-2.Methods: In this retrospective cohort study, we included adult hospitalized patients from OPTUM COVID-19 specific data set, who tested positive for SARS-CoV-2 within 14 days of hospitalization between 01/20/2020 and 01/20/2022. We examined outcomes of individuals with organisms growing on cultures from the bloodstream infections (BSIs), urinary tract, and respiratory tract, and a composite of the three sites. We used propensity score matching on covariates included demographics, comorbidities, and hospitalization clinical parameters. In a sensitivity analysis, we included same covariates but excluded critical care variables such as length of stay, intensive care unit stays, mechanical ventilation, and extracorporeal membrane oxygenation.Results: The cohort included 104,560 SARS-CoV-2 positive adult hospitalized patients across the United States. The unadjusted mortality odds increased significantly with BSIs (98.7%) and with growth on respiratory cultures (RC) (176.6%), but not with growth on urinary cultures (UC). Adjusted analyses showed that BSIs and positive RC independently contribute to mortality, even after accounting for critical care variables.Conclusions: In SARS-CoV-2-positive hospitalized patients, positive bacterial and fungal microbiological cultures, especially BSIs and RC, are associated with an increased risk of mortality even after accounting for critical care variables associated with disease severity. These findings underscore the importance of stringent infection control and the effective management of secondary infections to improve patient outcomes.


  • Open Access

    Assessing Racial and Ethnic Bias in Text Generation by Large Language Models for Health Care–Related Tasks: Cross-Sectional Study (Preprint)

    by John J Hanna, Abdi D Wakene, Andrew O Johnson, Christoph U Lehmann, Richard J Medford

    BACKGROUND Racial and ethnic bias in large language models (LLMs) used for health care tasks is a growing concern, as it may contribute to health disparities. In response, LLM operators implemented safeguards against prompts that are overtly seeking certain biases. OBJECTIVE This study aims to investigate a potential racial and ethnic bias among 4 popular LLMs: GPT-3.5-turbo (OpenAI), GPT-4 (OpenAI), Gemini-1.0-pro (Google), and Llama3-70b (Meta) in generating health care consumer–directed text in the absence of overtly biased queries. METHODS In this cross-sectional study, the 4 LLMs were prompted to generate discharge instructions for patients with HIV. Each patient’s encounter deidentified metadata including race/ethnicity as a variable was passed over in a table format through a prompt 4 times, altering only the race/ethnicity information (African American, Asian, Hispanic White, and non-Hispanic White) each time, while keeping all other information constant. The prompt requested the model to write discharge instructions for each encounter without explicitly mentioning race or ethnicity. The LLM-generated instructions were analyzed for sentiment, subjectivity, reading ease, and word frequency by race/ethnicity. RESULTS The only observed statistically significant difference between race/ethnicity groups was found in entity count (GPT-4, df=42, P=.047). However, post hoc chi-square analysis for GPT-4’s entity counts showed no significant pairwise differences among race/ethnicity categories after Bonferroni correction. CONCLUSIONS A total of 4 LLMs were relatively invariant to race/ethnicity in terms of linguistic and readability measures. While our study used proxy linguistic and readability measures to investigate racial and ethnic bias among 4 LLM responses in a health care–related task, there is an urgent need to establish universally accepted standards for measuring bias in LLM-generated responses. Further studies are needed to validate these results and assess their implications.