REF-1442 · Peric et al., 2026 · PeerJ · Paper · PubMed · Free full text

Does work below threshold move it as much as work above?

Pooled pre-post gains looked graded (below +2.6%, above +12.2%), but "below" is one mostly easy running arm. Across studies, not head-to-head; tempo not tested.

Main caveat: Within-group changes compared across studies; no pooled randomised contrast of intensity, and the authors state causality is not established.
moderate evidenceMeasured on non-athletes to elite athletes

Peric and colleagues asked which features of training most move the aerobic threshold (AeT, first threshold) and anaerobic threshold (AnT, second threshold). They included studies in healthy adults where intensity was normalised to an individual threshold, took each group's pre-to-post change, and sorted groups as below, at or above the threshold being measured.

Thirteen studies gave 24 groups: 9 for AeT and 15 for AnT. Overall AeT rose 21.0% (95% CI 11.8-31.0) and AnT 7.7% (4.9-10.7). Intensity category was a significant moderator for both (AeT p = 0.017, AnT p = 0.008); participant status and intervention length were not.

The percentages in the abstract (AeT above +38.7%, at +14.7%, below +18.4%; AnT above +12.2%, at about +7%, below +2.6%) are unweighted descriptive means from Table 2. The authors say the evidence "does not establish an independent causal effect of intensity" and that the results are "strong observational trends".

Our reading: with the older REF-1005 (A-), whose intensities did not differ significantly within a fitness level, one of two syntheses in the base sorted by intensity relative to a threshold, useful for the direction "work above a threshold moved it more". It does not grade the tempo band.

The one group below AnT is Enoksen's high-volume arm, 87% of its running at 65-82% of HRmax, while the same trial's "at" arm still ran 67% in that band (REF-1448); Inglis's lower-heavy arm (110% of the first threshold, REF-036) is filed as "above" in the AeT analysis and entered on AeT only, so the second-threshold analysis holds no arm clearly below the ceiling inside the heavy domain: the Inglis arm it codes "at" is the one REF-036 labels upper heavy, at the ceiling itself.

Protocol

Inclusion: healthy adults 18-60, training intensity normalised to an individual threshold (lactate or ventilatory), pre- and post-intervention threshold values; RCTs, non-randomised trials and single-group pre-post studies. Work matching not required.
Sample: n = 338 (239 men, 99 women) in 13 studies; 1,496 records screened, 62 full texts assessed, 25 groups reduced to 24 (one hypoxic group dropped).
Coding: "sub-threshold" below 100% and "supra-threshold" above 100% of the anchor; if a study gave no percentage, it "was calculated from the available data".
Analysis: log response ratio per group, three-level random-effects models stratified by threshold, cluster-robust variance with small-sample correction, RoB 2 or ROBINS-I, leave-one-out and bootstrap sensitivity, Egger's test.
Funding and interests: "The authors received no funding for this work."; competing interests as printed: "Carlo Ferri Marini is an Academic Editor for PeerJ." (PMC XML; the saved text dropped this footnote; PeerJ is the publishing journal).

Key Results

ThresholdOverall change (95% CI)95% prediction intervalI²AboveAtBelowTests
AeT (k = 9)+21.0% (11.8-31.0)−1.8 to +49.2~89%+38.7% (k = 3)+14.7% (k = 5)+18.4% (k = 1)moderator p = 0.017; at < above p = 0.046
AnT (k = 15)+7.7% (4.9-10.7)−0.3 to +16.4~53%+12.2% (k = 7)8 in Table 2, about 7 in the abstract (k = 7)+2.6% (k = 1)moderator p = 0.008; below < above p = 0.010; at vs above p = 0.054

Category means are unweighted.

CellGroup (Table 1)Relative intensity (%)Change
AnT belowEnoksen 2011, high-volume arm (REF-1448)84+2.6%
AnT atEnoksen 2011, around-LT arm (REF-1448)100+4.1%
AnT atInglis 2024, 30 min at the steady-state ceiling (REF-036)100+14.7%
AeT belowInglis 2024, 50 min moderate (REF-036)90+18.4%
AeT aboveInglis 2024, 41 min lower heavy (REF-036)110+28.0%
Other moderators: participant status (AeT p = 0.802, AnT p = 0.858) and intervention length not significant.
Method bias: ventilatory thresholds gave smaller AeT effects than lactate (β = −0.18, p = 0.018); for AnT, speed outcomes smaller than %VO2max (β = −0.07, p = 0.014). Publication bias indicated for AnT, so effects "may be overestimated".
Residual heterogeneity after intensity: significant for AeT (QE = 17.58, p = 0.007), not for AnT (QE = 20.49, p = 0.058).
A code we cannot trace: the 84 assigned to Enoksen's high-volume arm appears nowhere in that paper as a training intensity; its Table 1 gives 84.0 as that arm's pre-test %VO2max at vLT, and its training was set as 65-82% and 82-92% of HRmax (REF-1448). The review does not show its calculation.

Mechanisms

Measured: larger pre-post rises in threshold values in groups trained above that threshold; no mechanism was measured.
Authors' interpretation: greater homeostatic disturbance above a threshold (phosphocreatine depletion, AMP/ATP, PGC-1α signalling; lactate transporters for AnT); sub-threshold work "could elicit comparable shifts if sufficient volume is accumulated", which was not quantified.
Inference: intensity is confounded with study, sport, population, test and volume here, because no group was compared with another at random within the pooled contrast; a 2.6% gain in well-trained young runners and a 23.6% gain in an Inglis arm drawn from sedentary to recreationally active adults sit in the same second-threshold model.

Practical Application

Does work below threshold move it as much as work above? Pooled pre-post gains looked graded (below +2.6%, above +12.2%), but "below" is one mostly easy running arm. Across studies, not head-to-head; tempo not tested.
What it is good for: support for the direction "work above a threshold went with larger rises in it", not a dose. It holds no data on athletes whose threshold has stalled; using it as a reason to try such work after a stall is the base's inference. The authors say the data cannot define an optimal supra-threshold intensity. Compare REF-989 (A-) and REF-1005 (A-).
What it does not grade: training between the thresholds against the second threshold. For that the base holds REF-137 (B+) and REF-421 (B), one-week blocks in well-trained cyclists, and the arms of REF-1444 (B+) and REF-150 (B+). Not from this paper; the base's own practice: use sub-threshold work inside the quality budget (SYNTH-27, SYNTH-26).
What it does not show: that "above" beats "at" on the second threshold (p = 0.054; on the first, p = 0.046 with three "above" groups), any benefit in elite athletes, or anything about a band defined from FTP. It gives no prescription for sweet spot, a term and format it never mentions.

Limitations

Within-group changes compared across studies; no pooled randomised contrast of intensity, and the authors state causality is not established.
k = 1 in both "below" cells; the AnT one is mostly easy-paced running, and a single-group cell has no spread of its own to estimate.
Unweighted cell means quoted as headline numbers; no CI per cell is printed in the text.
Heterogeneous outcomes (power, speed, VO2 at threshold, %VO2max) and threshold methods, with LT and VT pooled as one anchor.
Search limited to PubMed and Google Scholar; four groups from one included study (Hebisz & Hebisz 2024) have a mean age of 17.9, below the stated 18-60.
Population labels are the primary authors': "amateur" covers both sedentary-to-recreational adults (REF-036) and runners with VO2max about 70 (REF-1448).
Few elite and female data by the authors' own account; supplementary file not read.
Funding and interests: no funding; one author is an Academic Editor at PeerJ, the publishing journal, as printed.
How this entry was checked and changed
Why this level

the rubric's A- line is an SR/MA with serious heterogeneity or mostly small trials; held at B+ because every effect is a within-group change compared across groups and studies rather than a randomised contrast, the sub-threshold estimate rests on k = 1 for each threshold, I² is about 89% and 53%, publication bias was indicated for the second threshold, and almost no study blinded its assessors

Full text (checked 2026-10-30)
Three-level meta-analytic models with cluster-robust variance estimation quantified Log Response Ratio effect sizes across 13 studies (24 intervention groups, 338 participants) (full text)
For AnT (k = 15) a significant categorical effect of training intensity was observed (p = 0.008), with post-hoc comparisons revealing greater adaptations for supra-threshold versus sub-threshold training (p = 0.010), below: +2.6%, at: ∼+7%, above: +12.2% (full text)
Training below the AnT resulted in significantly smaller adaptations than training above the AnT (p = 0.010), while the difference between training at and above the AnT showed only a non-significant trend (p = 0.054) (full text)
Table 2, AnT: Above k 7 mean 12.2; At k 7 mean 8; Below k 1 mean 2.6. AeT: Above k 3 mean 38.7; At k 5 mean 14.7; Below k 1 mean 18.4 (full text)
Enoksen, Shalfawi & Tønnessen (2011): 10 M, Running, AnT, BELOW, 84, 15.3 to 15.7 km/h, 2.6; 9 M, AnT, AT, 100, 14.6 to 15.2 km/h, 4.1 (full text)
These results should be interpreted as strong observational trends rather than absolute evidence of causality (full text)
The authors received no funding for this work (full text)