LETTERS TO THE EDITOR

Clarifying Denominators in Surveys of Ethical Concerns About AI

Tolga Akben, MD

Fam Med.

Published: 11/18/2026 | DOI: 10.22454/FamMed.2026.0586

TO THE EDITOR:

Cabrol and colleagues provide useful evidence on artificial intelligence (AI) use and training needs among family medicine residents and supervisors.1 Clarification of the denominators underlying their comparisons of ethical concerns about AI use in written academic work would help readers judge how these findings should inform training.

Table 2 reports cognitive off-loading in 82 residents (75%) and 240 supervisors (81%), and confidentiality concerns in 65 residents (59%) and 124 supervisors (42%). These percentages use 110 residents and 296 supervisors reporting ethical concerns as denominators. Using these denominators in 2×2 Pearson χ2 tests, without continuity correction, yields P=.148 for cognitive off-loading and P=.002 for confidentiality. The corresponding published values are .004 and .14. Those published values are reproduced, to the reported precision, if all 175 residents and 402 supervisors are used as denominators.

The intended comparison therefore remains unclear. Restricting an analysis to respondents who acknowledged ethical concerns estimates the pattern of specific concerns within that subgroup. An analysis of all respondents estimates overall endorsement only if the coding of skipped items justifies that interpretation. Table 2 lists 65 residents and 106 supervisors as “not displayed” for concern types. Because the preceding screening item also includes “not applicable” responses, respondents who were never shown an item should not automatically be treated as having answered no. The distinction matters for the estimated group differences, as well as the P values. Aggregate counts alone cannot establish how the original responses were coded.

Could the authors provide the 2×2 counts used for these comparisons, explain the questionnaire branching rules, and specify how “not displayed,” “not applicable,” and missing responses entered the analyses? The Checklist for Reporting of Survey Studies recommends explicit reporting of analytic methods and missing-data handling.2 Recent guidance in family medicine likewise emphasizes aligning survey questions, variables, and analyses.3 Reporting percentages and tests for the same clearly defined population, or labeling separate populations explicitly, would help educators distinguish overall endorsement of concerns about AI use in academic writing from priorities among respondents already concerned about AI.

ARTIFICIAL INTELLIGENCE USE DISCLOSURE STATEMENT

ChatGPT (OpenAI) was used solely to improve readability and language accuracy. The author is responsible for the accuracy, interpretation, and final content of this letter, including the accuracy of references and calculations.

References

  1. , , , , , . Artificial intelligence in family medicine: a comparative study of use, attitudes, ethics, and training needs between residents and supervisors. Fam Med. 2026;58(6):427–433. doi:10.22454/FamMed.2026.939793
  2. , , , et al. A consensus-based checklist for reporting of survey studies (CROSS). J Gen Intern Med. 2021;36(10):3179–3187. doi:10.1007/s11606-021-06737-1
  3. , , , , , . Making the most of CERA studies: how to develop excellent survey research in academic family medicine. Fam Med. 2026;58(2):118–122. doi:10.22454/FamMed.2026.478409

Lead Author

Tolga Akben, MD

Affiliations: Çankaya District Health Directorate, Ankara, Türkiye

Corresponding Author

Tolga Akben, MD

Correspondence: Çankaya District Health Directorate, Ankara, Türkiye

Email: tolga_akben@hotmail.com

Fetching other articles...

Loading the comment form...

Submitting your comment...

There are no comments for this article.

Downloads & Info

Share

Related Content

Tags

Searching for articles...