REF-1229 · Li, Yang & Wang, 2026 · Journal of Strength and Conditioning Research · Paper · PubMed

Is polarized training the best intensity distribution?

No model clearly beat it, or lost to it: in this network meta-analysis of RCTs all comparisons with polarized were inconclusive for VO2max and time trial.

A-fairly strong evidenceMeasured on trained to highly trained athletes

Li, Yang and Wang pooled randomized controlled trials of training-intensity distribution (TID) models in endurance athletes in a Bayesian network meta-analysis, with VO2max and time-trial (TT) performance as outcomes. Against polarized training, no other model showed a definite advantage for either outcome; all 95% credible intervals crossed zero.

The posterior ranking placed lactate-threshold training (THR) as most likely best for VO2max (rank-1 probability 65.4%, SUCRA 84.8%) and high-intensity interval training (HIT) as most likely best for TT (53.9%, 81.5%).

Age, training status, the method used to quantify intensity, intervention duration and sex were identified as potential moderators, and a cluster analysis found shared and sport-specific TID patterns. The authors advise tailoring TID to the athlete and the sport, naming THR and HIT as "the most promising options" for VO2max and TT.

Our reading: the decisive finding is the null one; it is close to the base's two pooled analyses, no POL/PYR difference (REF-991) and only a small VO2peak edge for polarized, under 12 weeks and in highly trained athletes, with no performance edge (REF-1213), and the rankings that the conclusion leans on are not differences.

Abstract only: the number of trials and athletes, which models were in the network, the effect sizes and the certainty of evidence could not be checked.

Protocol

Design: systematic review and Bayesian network meta-analysis; searches in PubMed, Web of Science, EBSCO, the Cochrane Library and Scopus; risk of bias with the Cochrane tool; randomized controlled trials only.
Population: endurance athletes; the number of trials and participants, sports, sexes and training levels are not given in the abstract.
Models compared: "common TID models"; polarized (reference), lactate-threshold training and high-intensity interval training are named in the abstract; whether pyramidal or high-volume low-intensity models formed separate nodes is not given in the abstract.
Outcomes: VO2max and TT performance.
Analysis: Bayesian network meta-analysis with 95% credible intervals, posterior rank probabilities and SUCRA; moderator analysis (age, training status, intensity-quantification method, intervention duration, sex); cluster analysis of TID patterns. Heterogeneity, inconsistency checks and certainty grading are not given in the abstract.
Funding: National Social Science Fund of China (21BTY119) (record metadata); conflicts not given in the abstract.

Key Results

OutcomeAny model vs polarizedMost likely best (rank-1 probability; SUCRA)
VO2maxno definite advantage (all 95% credible intervals cross zero)lactate-threshold training (65.4%; 84.8%)
Time-trial performanceno definite advantage (all 95% credible intervals cross zero)high-intensity interval training (53.9%; 81.5%)
Moderators: age, training status, intensity-quantification method, intervention duration and sex ("potential moderators"; directions and sizes not given in the abstract).
Cluster analysis: both shared and sport-specific TID patterns (details not given in the abstract).
Abstract tension: the Results say no model had a definite advantage over polarized; the Conclusion still names THR and HIT as "the most promising options". A rank-1 probability of 54-65% with intervals crossing zero leaves a 35-46% chance that another model is best (our arithmetic) and says nothing about how large any difference is.

Mechanisms

Measured (from the abstract): no conclusive difference between polarized and any other distribution for VO2max or TT.
Authors' interpretation: TID strategies should be tailored to athlete characteristics and sport demands; THR for VO2max and HIT for TT are the most promising.
Inference: rankings in a sparse network of small trials are unstable; a model compared in few trials with a lucky result can top the ranking without a real advantage. The intensity-quantification moderator echoes the counting problem in the base: the same athletes read 75/8/17 by session goal and 91/6/3 by time in zone (REF-467, C+ commentary), so "polarized" and "threshold" groups in different trials may not have trained as differently as their labels suggest.

Practical Application

Is polarized training the best intensity distribution? No model clearly beat it, or lost to it: in this network meta-analysis of RCTs all comparisons with polarized were inconclusive for VO2max and time trial.
Do not sell any distribution as proven superior: this agrees with the base's pooled analyses, no POL/PYR difference (REF-991, A-) and a small VO2peak edge for polarized only under 12 weeks and in highly trained athletes (REF-1213, A-).
Do not switch to threshold-heavy training on the ranking: "most likely best" here is a probability with intervals crossing zero, not a measured gain (A- from the abstract).
Choose the distribution by the athlete's hours, recovery and goal rather than by a label (SYNTH-26); training status and how intensity is counted moved the results here, which is consistent with stating the counting method (time in zone vs session goal) whenever a split is prescribed.

Limitations

Abstract only: number of trials, participants, sports, training levels, network nodes, effect sizes, heterogeneity, inconsistency and certainty of evidence are not given; nothing beyond the abstract's statements could be checked.
Rankings overread: the conclusion recommends THR and HIT on rank probabilities and SUCRA while every interval against polarized crossed zero.
Definitions: TID models are defined differently across trials (by time in zone or by session goal), and the abstract names the quantification method as a moderator; pooled labels may hide different training.
Bearing on [SYNTH-26]: nothing here tests low weekly volume; the guide's hour bands remain house practice.
Funding: a national social-science grant (record metadata); conflicts not given in the abstract.
Written from the abstract and record metadata (full text not read).
How this entry was checked and changed
Why this level

(from the abstract) — a network meta-analysis of randomized trials, which the rubric would place at A, held at A- because the number of trials, their sizes, heterogeneity, inconsistency and certainty are not given in the abstract and trials of intensity distribution are typically small (base inference, unverified here; REF-1213 is A- for mostly small trials, REF-991 for 13 studies, n = 348); the null result itself is not a reason for the level. Closed access; no open copy via Europe PMC or OpenAlex