Methodological Considerations in Assessing Intra‐Aortic Balloon Pump Use in Non–Cardiogenic Shock Acute Coronary Syndrome

We read with great interest the study by Fan et al. examining intra-aortic balloon pump (IABP) use in acute coronary syndrome patients without cardiogenic shock [1]. Although this large multicenter study shows real-world value, some methodological concerns limit the value of the results and require special attention. First, the lack of IABP timing data severely impairs causality inference. The authors state IABP was placed at the physician's discretion in those with “hemodynamic instability or procedural challenges,” but do not say whether IABP was used prophylactically or as rescue therapy for intra-procedural deterioration. This timing matters is important. If IABP was mostly used as rescue therapy for patients with decompensating disease, then the mortality association you report is one of the concomitant disease severity and not of device harm. All prior randomized studies showed neutral IABP effects and used prophylactic IABP placement, a situation really different than the one in which IABP is used as a rescue therapy [2]. This issue is best studied in the future by longitudinal registry studies with timed documentation and other standardization. Second, the meta-analysis findings contradict the main findings results. The pooled analysis shows no association with mortality (RR = 1.06, 95% CI: 0.79−1.42), but in their propensity-matched analysis, there was substantial harm (OR = 2.56, 95% CI: 1.30−5.04). Rather than describing the effect as “moderate heterogeneity,” the authors need to explain why their observational study diverges from the rest of the evidence, which includes randomized trials [3], as this lack of agreement strongly suggests that unmeasured confounding dominates their findings. Third, there are apprehensions concerning over-adjustment bias related to the statistical adjustments of Model 4. Authors adjusted for procedural variables such as thrombus aspiration and stent placement, which are likely to be mediators, not confounders, in the causal pathway connecting the use of IABP and mortality. Adjusting for mediators can lead to collider stratification bias and can, paradoxically, strengthen an association. This is worrying, especially given that the odds ratio went from 2.05 to 2.56. This seems to suggest that there is an artifact of the method rather than a true effect. Fourth, I think the overall outcome definition, “death from any cause including cardiac and non-cardiac causes,” is a bit limiting from a mechanistic perspective due to the lack of adjudication. If IABP causes harm, there are clear patterns that should emerge, namely vascular complications, death from bleeding, and infection. Without complication and cause-specific mortality data, the attributable death to IABP is conjectural. I think this revolves around a question of design credibility, and I would argue that the outcome should center on adjudicated endpoints and comprehensive safety outcomes [4]. Finally, having a 93% sample reduction because of propensity matching considerably constrains the statistical power of the analysis. Out of the 11 interaction subgroups, only one reached a statistically significant result, which could be a type I error given the lack of a multiple comparisons adjustment [5]. In summary, Fan et al. provide important preliminary evidence regarding IABP use in non-shock ACS patients. Moving forward, the research will need to include standard IABP protocols, careful documentation of timelines, adjudicated cause-specific outcome measures, systematic complication measures, and the potential associations of the findings to causality, patient safety, and clinically relevant outcomes. The authors declare no conflicts of interest.