Clinicians Call Out 'Research Pollution' and Demand Better Appraisal Training

Key Takeaways

• Clinician frustration is aimed less at information volume than at claims that sound more causal, precise, or practice-ready than the data can support.
• Journal-club and literature-review formats need to build appraisal habits, not just summarize new studies.
• AI education belongs in the same family of work: teaching verification, doubt, and last-mile responsibility rather than tool familiarity alone.

Clinicians on X are calling out p-value misuse, overstated causal claims from observational data, and ignored confounding as widespread problems that undermine evidence-based practice. The signal remains narrow but points to a concrete provider need: appraisal skills must become recurring infrastructure in CME rather than occasional add-ons.

Research summaries are not enough

The sharpest clinician language this week centered on p-values, observational data, confounding, and the marketing-like use of terms such as “emulated randomized” or “pseudo randomized.” One oncology-oriented thread argued that “Statistical significance term is misleading and total badness; Causal inference continues to be the scam of the century,” while later posts pressed for clearer limitation statements and less causal overreach in observational-study claims (source). A second thread described a high tolerance for weak biomedical research and called much of it “research pollution” in service of career incentives (source).

The oncology presence is prominent, but the learning problem is broader: clinicians are asking for help distinguishing signal from persuasive packaging. This extends an earlier brief on clinicians saying medical training never taught them to read the literature they now must apply, but the tone this week was more confrontational and more statistical.

For CME providers, the issue is not whether to add another lecture on study design. It is whether existing journal clubs, rapid updates, and literature reviews actually make learners practice skepticism: identifying the claim, naming the design constraint, locating the unmeasured-confounding risk, and rewriting the conclusion in language the evidence can bear. The test for a program is simple: after the activity, can the learner spot when a study summary has made the evidence sound stronger than it is?

AI training has the same appraisal problem

The week’s AI signal came from a CPD-focused academic podcast, so it should be read as provider-adjacent rather than independent clinician demand. Still, it connects directly to the appraisal theme. The episode argued that CPD should not stop at teaching clinicians how to use specific tools; it should prepare them to evaluate AI outputs, understand tool families, maintain human oversight, and avoid hype-driven use in high-stakes care (source).

That matters because tool training decays quickly. A module built around one model or interface can become stale, while the underlying clinician task remains: verify the output, ask what evidence supports it, recognize missing uncertainty, and decide where human judgment has to override fluent text.

For CME teams, AI education should be built less like software onboarding and more like applied critical appraisal. Learners need repeated cases where the AI answer is plausible but incomplete, overconfident, poorly referenced, or unsafe without clinical context. The outcome to measure is not whether the learner can prompt a tool; it is whether the learner knows when not to trust the answer.

What CME Providers Should Do Now

• Audit journal-club and literature-review formats for whether learners actively critique causal language, confounding, and limitations—or only receive expert summaries.
• Add exercises that require learners to rewrite overstated conclusions into transparent, evidence-matched statements.
• In AI education, assess verification behavior: source checking, uncertainty detection, bias recognition, and final clinical accountability.

What to reconsider

The common thread this week is not statistics versus AI. It is whether CME is teaching clinicians to slow down at the exact point where information becomes persuasive. If a program leaves learners with more facts but no stronger habit of challenge, it may be helping them keep up while leaving the hardest work untouched.