The decision this page answers. This is a substitution question, not an add-vs-nothing one. The person is going to do aerobic exercise; the choice is the modality — high-intensity interval training (HIIT) versus moderate-intensity continuous training (MICT), the realistic alternative. So the effect that matters is HIIT minus MICT, not HIIT minus rest. (Whether to exercise at all is a different, larger decision settled elsewhere — see Physical Activity Dose and Mortality.)
The headline is narrower than the title. Liu 2019 is a gold meta-analysis of 13 RCTs, 345 participants (163, 47.2%, in HIIT arms) in people with T2D (mostly middle-aged/older adults; one trial, Lee 2015, in adolescents, mean age 15); HIIT protocols ran 11-16 weeks, 2-5x/week (median 3) (Liu et al., 2018). The paper is titled for glycemic control, but its only moderate-certainty finding is fitness, not glucose: «The principal finding of the current meta-analysis was that HIIT was more efficient than MICT in increasing VO2peak in T2D patients; they also found that reduction of BMI, body weight, and HbA1c (%) was less conclusive because of low quality of the evidence.» (Liu et al., 2018)
The effect table — HIIT vs MICT, with certainty
All estimates are the between-modality contrast (HIIT minus MICT). Negative favours HIIT for the composition/glucose/lipid rows; positive favours HIIT for VO2peak. GRADE certainty is Liu’s own rating of the vs-MICT comparison (Liu et al., 2018).
| Outcome (vs MICT) | Effect (95% CI) | Studies | I2 | GRADE certainty | Read |
|---|---|---|---|---|---|
| VO2peak, relative | +3.37 ml/kg/min [1.88, 4.87] | 7 | 48% | moderate | HIIT better |
| VO2peak, absolute | +0.37 L/min [0.28, 0.45] | 6 | 36% | moderate | HIIT better |
| HbA1c | -0.37% [-0.55, -0.19] | 9 | 0% | low | HIIT better (surrogate) |
| Body weight | -1.22 kg [-2.23, -0.18] | 8 | 0% | low | HIIT better |
| BMI | -0.40 kg/m2 [-0.78, -0.02] | 8 | 0% | low | HIIT better |
| LDL cholesterol | -0.25 mmol/L [-0.46, -0.04] | 6 | 0% | very low | HIIT better |
| Fasting glucose | +0.10 [-0.84, 0.65] | 8 | 0% | low | not significant (wide CI) |
| Fasting insulin | SMD -0.19 [-0.58, 0.20] | 4 | 0% | low | not significant (wide CI) |
| HOMA-IR | +0.13 [-0.10, 0.36] | 6 | 0% | very low | not significant (wide CI) |
| Total cholesterol | -0.18 [-0.44, 0.07] | 7 | 0% | low | no difference |
| HDL cholesterol | -0.04 [-0.10, 0.02] | 9 | 0% | very low | no difference |
| Body fat % | -0.50 [-1.18, 0.19] | 5 | 0% | very low | no difference |
| Waist circumference | -0.15 [-1.21, 0.91] | 6 | 0% | low | no difference |
What survives the certainty grading. Only cardiorespiratory fitness clears moderate certainty (rated up for a large effect, down for inconsistency and imprecision). Everything else is low or very low. So the defensible claim is: HIIT raises VO2peak more than MICT in T2D; on glucose, weight, and lipids the between-modality difference is at best a low-certainty point estimate. Liu’s own conclusion says as much — HIIT is «an effective strategy for improving cardiorespiratory fitness in patients with T2D, preferable to MICT», while «Results related to other parameters associated with the prognosis of T2D, such as HbA1c, body weight, and BMI, were not conclusive.» (Liu et al., 2018)
The glycemic finding is a low-certainty point estimate — and contested
The HbA1c advantage (-0.37% vs MICT) is the only glucose-family signal; fasting glucose, fasting insulin, and HOMA-IR show no significant HIIT-over-MICT difference (table above), with CIs too wide to exclude one — so the HbA1c signal is unsupported by the other glucose markers rather than contradicted by them, and the isolated point estimate should be read cautiously. Liu attributes the discordance partly to measurement and baseline: «The inconsistent results could partly be explained by the difference among methods used to measure insulin sensitivity, as well as the difference in the baseline of glycemic control.» (Liu et al., 2018)
Liu flags the contest directly: «HIIT showed a 0.37% greater reduction of HbA1c than MICT… This is inconsistent with the findings of a meta-analysis conducted by Jelleyman et al.», which found HIIT reduced HbA1c but with no significant difference versus continuous training (Liu et al., 2018). That prior MA is not held here — but the within-literature disagreement on the very finding the title emphasizes is itself the honest read: the HIIT-over-MICT glycemic edge is not settled, and an earlier (2015) MA in a broader diabetes + metabolic-syndrome population found no significant HIIT-vs-continuous HbA1c difference (unheld; quality and trial overlap unverified) (corrected 2026-10-06: one gold MA against it -> unheld, broader-population MA; self-critique).
HbA1c is a surrogate — do not read a hard-outcome benefit off this MA
This MA measures no patient-important outcome — no mortality, no microvascular or CV events, no adverse events. HbA1c is a surrogate (see Surrogate Outcomes). Liu cites the UKPDS-type transmission that a 1% HbA1c reduction lowers microvascular complications by «37% and that of death related to diabetes can be reduced by 21%» (Liu et al., 2018). Two cautions before anyone chains that to the modality choice:
- The HIIT-over-MICT HbA1c delta is 0.37%, well under the 1% the transmission is quoted for; a linear read would imply roughly a tenth of those figures, and only if the surrogate transmits and if the point estimate is real (it is low-certainty and contested, above).
- Surrogate transmission is itself an evidenced claim, not an assumption — and here it is borrowed from UKPDS 35, a prospective observational analysis of glycaemia vs complications (not an intervention contrast, and not an exercise trial) (corrected 2026-10-06: was a glucose-lowering-drug trial; self-critique). Whether a fitness-driven HbA1c change carries the same prognosis is untested by this evidence.
The better-anchored surrogate is fitness: the VO2peak difference is moderate-certainty, and CRF has a strong observational mortality gradient (see Cardiorespiratory Fitness and Mortality). Whether a training-induced VO2peak gain transmits to mortality is, like HbA1c, not tested by this MA — Liu treats both as predictors: «Both VO2peak and HbA1c are important predictors» of CV events (Liu et al., 2018). (corrected 2026-10-06: was the load-bearing benefit … at moderate certainty, fusing the between-arm GRADE with an ungraded transmission; self-critique)
Adherence, safety, and net effect — the substitution caveats
Adherence is part of the effect (see the schema’s Layer-3 rule): a harder protocol abandoned loses
to an easier one sustained. This MA cannot speak to it. Trials were short (11-16 weeks), and the MA
does not systematically report supervision (the only included trial it describes as supervised is Støa
2017, ref 15 — corrected 2026-10-06: was the Stoa background trial; self-critique); the only compliance proxy is per-trial dropouts (0-10 per arm, Table 1), and there is
no free-living or long-term adherence data. HIIT’s real-world edge or deficit versus MICT over years
is unmeasured.
(inferred from Liu et al., 2018)
(corrected
2026-10-06: supervised dropped as a property of the included trials, here and in question:; the only
hit is Støa 2017, an included trial) What is held is the affective precursor:
Affective Response to High-Intensity Interval Exercise (Niven 2020) finds higher-intensity work felt as less pleasant
in-task than moderate continuous work — an adherence hazard whose behavioural consequence is still
unproven, and which tracks intensity rather than the interval format.
Safety is a named gap. The MA reports no adverse-event or hypoglycemia comparison [searched: adverse/hypoglyc across chunks 01-02, 0 hits]. For prescribing high-intensity work to older or higher-risk T2D patients this matters, and it is exactly where a general fitness source can carry the load: Measuring and Raising Cardiorespiratory Fitness holds Poon 2024’s finding that safety was not a differentiator between HIIT and MICT across populations — a cross-link, not a claim this MA supports.
Time-efficiency is asserted as an aim, not cleanly tested. Liu frames HIIT as «an ideal time-efficient physical activity program» (Liu et al., 2018), but HIIT and MICT protocols were heterogeneous and not consistently volume-matched across the 13 trials, so this MA does not isolate whether HIIT delivers equal benefit in less time. The time-efficiency case is plausible and stated, not demonstrated here.
Where this sits in the T2D lever stack
For a person with T2D, the big rocks are weight loss, dietary pattern, and getting any activity at all (see Diets for Weight Management in Type 2 Diabetes, Carbohydrate Restriction and Type 2 Diabetes Remission, Lifestyle vs Metformin for Diabetes Prevention). Modality choice within aerobic exercise is a refinement below those — a Layer-1 small lever by construction once the person is already exercising. What Liu adds at the margin: if the person is choosing between interval and continuous work and cares about fitness (a moderate-certainty fitness gain — a surrogate with a strong observational mortality gradient), HIIT is the better bet; if the goal is glycemic control specifically, the modality difference is low-certainty and contested, so pick on preference, tolerability, and sustainability rather than on an expected HbA1c edge.
Independence note
The CRF finding here concurs with the general HIIT-over-MICT fitness result in Measuring and Raising Cardiorespiratory Fitness (Poon 2024 umbrella). This is not independent type-E backing: Poon is an umbrella of SR/MAs that plausibly overlaps this MA’s trials or includes it, so the agreement is shared-substrate, not a second route. Treat Liu as the T2D-stratum instance of a broadly-held direction, cross-linked, not as corroboration that raises confidence.