A new study found that two-thirds of epidemiologists in specialized training programs used artificial intelligence (AI) in their work, despite only one in five having received AI training and one-quarter expressing ethical concerns about accuracy and bias.
For the study, published in Eurosurveillance, researchers led by a team at the Public Health Agency of Canada surveyed 105 participants in Field Epidemiology Training Programs (FETPs), referred to as field epidemiologists, about their use of AI. The survey consisted of 20 open- and closed-ended questions written in a neutral tone to reduce response bias. The respondents hailed from Canada, Europe, and the United States.
FETPs are specialized programs that train participants to develop technical capabilities in applied epidemiology and improve their professional judgment and ability to communicate information.
While AI has the potential to improve efficiency for field epidemiologists, the researchers write, it also raises concerns about the development of critical-thinking skills and professional discernment in a field that “places a premium on judgment and interpretation, and the ability to communicate uncertainty.”
Troubleshooting coding errors
Among the 105 respondents, 66% said they use AI in their field epidemiology work. AI adoption was highest among fellows in the European program (32 of 36 [89%]), compared with 30 of 56 (54%) in the US program and seven of 13 (54%) in the Canadian program. Use difference was not statistically significant between first- and second-year fellows.
Among AI users, 42% said they use the technology weekly, 30% said they use it daily, and 26% reported using it occasionally. Most reported feeling either somewhat comfortable (49%) or comfortable (35%) using AI, while 12% reported feeling very comfortable, and 4% reported feeling uncomfortable.
ChatGPT was by far the most commonly used platform (87%), followed by institution-specific AI tools (25%).
The most common applications of AI were troubleshooting coding errors (91%), writing code (75%), and improving work efficiencies (42%). Qualitative responses suggested that AI generally improved coding efficiency.
“AI reduced the time spent troubleshooting coding errors, aided in learning new coding techniques, simplified existing code, facilitated code transfer across different coding software programmes, and was helpful for generating ideas for data analysis,” the researchers note.
Respondents also said AI helped improve writing quality and efficiency, as well as streamline routine administrative tasks, such as writing meeting minutes and organizing project timelines. Some said AI helped them better understand new topics, conduct background research, and summarize information.
Training lags behind adoption
At the same time, 41% of AI users reported barriers to adopting the new technology. Common concerns included technical limitations, limited access to AI tools, uncertainty about institutional rules governing AI use, and questions about the accuracy and reproducibility of AI-generated content. One-quarter of AI users also reported ethical concerns, including data privacy, potential bias, environmental impacts, and the possibility of overreliance on AI.
Respondents said they wanted practical instruction on using AI for coding, data analysis, scientific writing, data visualization, and outbreak detection.
The survey also found substantial gaps in institutional guidance and training. Only one in five fellows reported receiving any formal or informal AI training.
Respondents said they wanted practical instruction on using AI for coding, data analysis, scientific writing, data visualization, and outbreak detection, as well as training on ethical considerations and effective prompting strategies. Respondents also expressed interest in learning about the limitations of AI, such as increased awareness of common mistakes.
Curricula should promote AI literacy
The authors say FETP curricula should promote AI literacy and include education on the technology’s limitations, biases, and ethical implications, especially now that it has become a routine part of field epidemiology practice for many training program participants.
“Possible recommendations for training might include hands-on workshops, real-world case studies, and access to AI tools and platforms,” they write. “The FETPs need to ensure learners gain experience using AI in foundational epidemiological tasks like coding, study design, data analysis, scientific writing, and topics such as surveillance and outbreak detection.”
