Absence of Evidence Is Not Evidence of Absence in Head‐to‐Head Exercise Intervention Studies

1.

In head‐to‐head intervention studies comparing two treatments or exercise modalities, participants are often assessed before and after the intervention period. This corresponds to a two‐factor repeated‐measures design, in which the relevant statistical question is whether the change over time differs between groups. Statistically, this is captured by the group × time interaction term. However, misinterpretation of such designs remains frequent, particularly when statistically significant change within one group, but not the other, is taken to indicate superiority of one intervention over the other. A recent example is the study by Trinks et al., published in Cell Metabolism in January 2026 [1]. The authors evaluated blood‐flow restriction training (BFRT) versus conventional resistance exercise training (CREST) in individuals with type 2 diabetes (T2D), addressing an important and clinically relevant question. Supervised intervention trials in this population are inherently challenging [2], and the demonstration that BFRT can be safely implemented is clinically relevant. Resistance exercise performed with heavy loads is well known to elicit substantial gains in muscle mass and strength [3], yet recent years have seen increasing interest in lower load training strategies capable of inducing similar adaptations [4].

The central aim of the study was to determine whether musculoskeletal and metabolic adaptations differed between BFRT and CREST after 12 weeks of training. Of the 65 outcomes assessed from pre‐ to post‐intervention, 27 were discussed or presented in a way that, in our view, may lead readers to infer differential responses between training modalities, either through the presentation of group‐specific p‐values suggesting differing magnitudes of change, statistically significant change reported in one group but not the other, or through explicit statements in the text indicating a differential effect. Because group × time interactions were not consistently reported in the article, we examined the publicly available individual participant data to assess whether the reported differences between BFRT and CREST were supported by the corresponding interaction effects. We applied a linear mixed‐effects model with compound symmetric covariance in GraphPad Prism (v.10.6), including group, time (repeated), and their interaction as fixed effects and subject as a random effect (Table 1). A statistically significant group × time interaction was observed for only one outcome (citrate synthase activity; p = 0.0248), indicating a greater increase following BFRT relative to CREST. For the remaining outcomes described as differentially affected—including measures of body composition, oxidative capacity, and muscle signaling—corresponding interaction effects were not statistically supported.

TABLE 1.

Outcomes described by the authors (Trinks et al. [1]) as showing differential responses between groups, outcomes for which different within‐group p‐values were reported, and corresponding group × time interaction p‐values from our reanalysis.

Outcome	Figure/table	Reported by authors		Our reanalysis
		Described in terms suggesting differential response	Different within‐group p‐values shown	Interaction p (group × time)
[CI]p	Figure 2	x	x	0.1271
[CI + II]p	Figure 2	x	x	0.1949
CS activity	Figure.2	x	x	0.0248
Kmax	Figure 2	x	(x)	0.3348
Vmax	Figure 2	x	x	0.2051
AMPKa	Figure 3	x	x	0.569
HXK2	Figure 3	x	x	0.2895
Myostatin	Figure 3	x	x	0.1845
PGC‐1‐a	Figure 3	x	x	0.9514
TFAM	Figure 3	x	x	0.3833
Subcutaneous adipose tissue	Figure 5	x	x	0.5584
Visceral adipose tissue	Figure 5	x	x	0.0532
Whole‐body adipose tissue	Figure 5	x	x	0.8421
[CI + II]p (adipose tissue)	Figure S3	x	x	0.9886
mtDNA (adipose tissue)	Figure S3	x	x	0.0979
Vmax (adipose tissue)	Figure S3	x	x	0.3503
HXK1	Text	x	x	0.4647
Triglycerides	Table 1	x	x	0.0785
Waist circumference	Table 1	x	(x)	0.5582
Cross‐sectional area	Figure 1		x	0.4964
Quadriceps muscle volume	Figure 1		x	0.7689
mTOR	Figure S1		x	0.9485
Diastolic blood pressure	Table 1		x	0.4508
Heart rate	Table 1		x	0.9185
Systolic blood pressure	Table 1		x	0.9273
VO2max	Table 1		x	0.3352
Intramuscular fat fraction	Table 2		x	0.5749

Note: “x” indicates that the criterion was met. Significant interaction effects identified in our reanalyses are highlighted in green and bold.

Importantly, this is not a critique of the statistical method itself, but of the interpretation of the results. In the absence of a significant group × time interaction, a statistically significant within‐group change in BFRT but not CREST does not provide evidence that BFRT induced greater adaptation than CREST. Another issue relates to multiple testing. Given the large number of outcomes assessed, the nominally significant interaction for citrate synthase activity should be interpreted with caution in the absence of adjustment for multiplicity. When multiple related outcomes are examined, formal methods exist to account for multiplicity while reflecting the structure and relatedness of the tested hypotheses [5].

This does not diminish the potential clinical relevance of BFRT, which may still represent a valuable therapeutic option if it is more feasible or appropriate than high‐load resistance training in some individuals [6]. However, the results reported by Trinks et al. do not, in our view, provide consistent statistical support for the conclusion that BFRT is superior to conventional resistance exercise in T2D. More generally, this serves as a reminder that intervention studies should report main and interaction effects transparently and should not infer superiority from statistically significant change in one group but not the other.

Funding

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.