No Significant Difference is not Equivalence: Statistical Considerations on Phenylephrine Requirement in Left-Tilt versus Supine Cesarean Delivery [Letter]

Dear editor

Yu et al¹ addressed a practical question in obstetric anesthesia: whether a 15° left-tilt position reduces prophylactic phenylephrine requirements during elective cesarean delivery under combined spinal-epidural anesthesia. The study contributes useful dose-finding evidence for prevention of spinal anesthesia-induced hypotension. The similar ED50 estimates, 0.33 μg/kg/min in the left-tilt group and 0.30 μg/kg/min in the supine group, add to the reassessment of routine left uterine displacement. Several points deserve clarification.

The main interpretive issue is that absence of statistical significance is not equivalence. The reported relative median potency for phenylephrine in the left-tilt versus supine group was 1.06, but the 95% CI ranged from 0.86 to 1.45. This interval is wide. It allows the possibility that 15° left tilt could reduce, have no effect on, or increase phenylephrine requirement by a clinically relevant amount. Figure 2 shows scatter and overlap in individual effective and ineffective responses; Figure 3 shows nearly superimposed probit curves, but visual overlap is not proof of interchangeability. Cuzick and Sasieni emphasized that an inconclusive superiority comparison only means that no statistically significant difference was detected; noninferiority requires a prespecified margin and a confidence interval excluding unacceptable loss.² The CONSORT extension for noninferiority and equivalence trials likewise calls for explicit margins and design-specific interpretation.³ Therefore, the statement that routine tilt solely to reduce vasopressor need “may not be necessary” is acceptable as a cautious clinical hypothesis, but not as formal evidence that supine positioning is equivalent or noninferior to 15° left tilt.

A smaller reporting inconsistency concerns categorical analyses. The methods state that categorical variables were evaluated using Fisher exact test, yet the footnote to Table 2 identifies the chi-square test. This matters because several endpoints had sparse events: hypertension 0 of 40 versus 1 of 40, bradycardia 3 of 40 versus 2 of 40, and nausea or vomiting 2 of 40 versus 3 of 40. For such 2×2 tables, Fisher exact testing is preferable when expected counts are low.⁴ A revised table specifying the exact test used for each endpoint, with risk differences or odds ratios and 95% CIs, would make negative findings more transparent. In addition, Table 1 and Table 2 label several non-normal variables as median (range), but values such as total phenylephrine 345 (290 to 443) versus 310 (243 to 390) appear more consistent with interquartile ranges.

The authors deserve appreciation for challenging a long-standing positional habit.