Page 22 - AIH-2-4

Artificial Intelligence in Health AI editorial policy ethics

“outside scope” – feedback suggesting that editorial boards • Who determines what constitutes valid evidence?
often defer scientific vetting to original authors, a process • Who is accountable when predictive models reinforce
colloquially referred to as pre-clearance. While this practice structural bias or contribute to diagnostic error?
may be intended to streamline correspondence handling, it In the absence of systemic safeguards, the pre-mature
effectively allows original authors to veto external critique, adoption of under-evaluated AI tools threatens not just the
compromising the neutrality and independence of peer integrity of the scientific record but the safety and equity of
review. 32-38 patient care. 30,31,39,40

This gatekeeping is further exacerbated by a systemic
lack of technical and ethical expertise among clinical 6. Comparative analysis across cases
journal editors to assess AI-related submissions. As ML The rejection of substantive methodological critiques in
models become more complex and deeply integrated both the Haghish and Ding et al. case studies reveals
12
11
into healthcare, editorial boards must be equipped to consistent patterns of editorial gatekeeping, technical
evaluate not only clinical relevance but also algorithmic exclusion, and ethical under-evaluation. While the studies
validity, interpretability, and fairness. 30,31 Without such addressed different domains (i.e., text-based versus speech-
expertise, editorial decisions may inadvertently privilege based suicide prediction), the nature of the overlooked
esthetic novelty or positive results over scientific rigor and issues and the editorial rationale for rejection were
replicability. strikingly similar. These cases demonstrate that systemic
In many journals, the peer review process itself editorial deficiencies can transcend methodological
remains opaque and insufficiently diverse, further domain, modality, and even discipline.
contributing to biased publication outcomes. Studies Table 1 summarizes the critical methodological concerns
show that increasing gender and international diversity raised in each case, mapping them to potential clinical
among reviewers correlates with fairer evaluations and consequences and corresponding editorial responses. This
higher-quality editorial outcomes. 33-35 Yet, even in journals side-by-side view makes visible the shared vulnerabilities
that acknowledge these disparities, few have adopted in AI health research publication and underscores the
concrete reforms, such as blind review, reviewer training urgency for reform in peer review protocols.
in AI ethics, or structured checklists for evaluating ML Figure 1 shows a conceptual model depicting the
studies. 19,30,39-45 multi-layered nature of editorial gatekeeping and its
As generative AI continues to scale across clinical consequences. Critique pre-clearance, limited AI literacy,
domains, scholars have increasingly called for the and narrow definitions of clinical relevance combine to
integration of embedded ethics into the development, create significant obstacles. Together, these factors build
evaluation, and dissemination of medical AI research. 39-45 barriers that obstruct scientific accountability.
This approach demands that ethical concerns – such as Together, these cases illustrate a systemic breakdown
algorithmic bias, safety, transparency, and explainability – in editorial accountability. When valid methodological
be addressed from the outset, not appended post hoc. In this critiques are filtered out by opaque editorial practices
model, ethics is not a checkpoint at the end of the pipeline or vetoed by original authors, the epistemic integrity
but a structural element of rigorous scientific inquiry. of the scientific record is compromised. Moreover, the

Despite these calls, the editorial handling of the critiques publication of inadequately vetted AI models has serious
toward the works of Haghish and Ding et al. suggests that clinical and ethical implications.
11
12
present publishing norms fall short of AI-driven studies
and studies using AI methods. The absence of substantive 7. Limitations
engagement with these challenges implies that many In discussing the constraints within editorial decision-
journals remain ill-equipped – or unwilling – to enforce making, I recognize several key factors that shape how
ethical scrutiny as part of peer review. Without meaningful scholarly work and professional discourse are disseminated.
reform in areas, such as editorial independence, reviewer Journals operate within specific editorial frameworks that
training, and conflict-of-interest transparency, flawed AI dictate what content is selected for publication. These
models may continue to bypass critical evaluation and policies may prioritize particular research methodologies or
enter the clinical literature unchallenged. thematic focuses, inadvertently shaping which perspectives
This failure is not merely procedural. It raises enter the broader academic conversation. 20,21,32-38
foundational questions about epistemic authority in The peer-review process, while intended to ensure rigor
clinical AI: and credibility, is subject to variability in reviewer expertise,

Volume 2 Issue 4 (2025) 16 doi: 10.36922/AIH025210049