At the American Heart Association meeting, Vishal Khetpal — a preventive cardiologist and assistant professor at the Northwestern University Feinberg School of Medicine — described his reaction to presentations of the STREAM trial. Investigators reported that older adults without prior heart attacks or strokes who stopped taking statins did not have significantly different outcomes from those who continued, meeting the trial’s statistical noninferiority threshold as presented. Yet when the Kaplan-Meier plots were shown, Khetpal noticed the survival curves did not run parallel; they separated. The group randomized to stop statins appeared to do worse over time. That visual divergence, he argued, left him puzzled about the way the finding was framed.
Short sentence: the curves diverged. Longer thought: in randomized noninferiority designs, a formal statistical boundary can allow an intervention to be judged similar even when point estimates favor the control arm and when event-time curves separate across follow-up, and the author used that technical distinction to question whether the headline interpretation matched the clinical signal.
Khetpal outlined the apparent paradox: the trial’s analysis, as presented in the meeting, satisfied the protocol’s noninferiority criterion. Yet the point estimates and the Kaplan-Meier visualization suggested inferiority in practice — that is, patients who stopped therapy experienced worse outcomes compared with those who continued. The article noted that this scenario can occur in noninferiority trials where prespecified margins and timing of events shape the statistical conclusion. The study team reported noninferiority; the cohort’s event-time behavior suggested a different clinical story.
The source article did not provide granular trial methodology in the text available: it did not list the exact primary endpoint definition, sample size, follow-up duration, or the noninferiority margin used. The STAT piece referenced the New England Journal of Medicine DOI for STREAM but did not reproduce trial-level numbers or analytic thresholds in the publicly accessible portion of the report. Therefore, readers relying solely on the STAT summary were not given the full numerical context that would allow independent assessment of the divergence versus the prespecified statistical claim.
The author reported that the trial’s data safety monitoring board recommended clinicians consider talking with patients who had stopped statins about resuming them. That recommendation implies the DSMB observed a safety or outcome signal concerning the off-treatment group. Khetpal emphasized that this real-world advisory came from the oversight body, rather than from a press release framing the result as a straightforward neutral finding.
Brief and direct: the DSMB urged discussion with clinicians. More detail: the publicly available story did not include verbatim DSMB language, committee minutes, or the board’s internal rationale; those materials were not reported in the source article.
Khetpal posed three questions after the STREAM presentation; the STAT article foregrounded his puzzlement. The piece made clear that the author wanted greater clarity about interpretation, but the subscriber-only portion of the column contains the full commentary. The source did not list the three questions in the free text, and it did not publish the detailed follow-up analyses or subgroup results in this summary.
Concrete limits from the source: the public article did not reproduce detailed trial statistics, did not provide subgroup or sensitivity analyses, and did not document dosing, adherence metrics, or how many participants crossed treatment arms. Those omissions matter when a visual divergence in Kaplan-Meier curves contrasts with a formal noninferiority declaration. Clinicians reading the summary were left with two competing impressions: a headline neutral finding and a data visualization suggesting harm when therapy was stopped.
The author closed his account by urging clinical caution. He recommended that patients who discontinued statins consider discussing resumption with their clinicians, reflecting the DSMB’s advice as reported. The article ends on specific, concrete open points — the need for access to full trial methods, the importance of reviewing subgroup data, and the unresolved interpretive gap between noninferiority statistics and clinical trajectories — rather than on broad speculation.
The STAT piece referenced the NEJM publication for STREAM and noted that the complete analysis and Khetpal’s extended commentary were available in the STAT+ subscriber version. The source did not provide further trial details within the public text, so readers seeking the full methodological and numerical context were directed to the original trial report and to the subscriber column for Khetpal’s full set of questions.