The decision here is not should an older adult resistance-train (that is settled — see Muscle-Strengthening Activity and Mortality, Exercise for Preventing Falls in Older Adults) and not the load/sets/frequency dial of Resistance Training Prescription - Load Sets and Frequency. It is a distinct axis: once training, should the emphasis be on moving fast (power / velocity) rather than lifting heavy (strength)? The answer, on a gold head-to-head, is yes for physical-function-test performance — but the endpoint stops at lab/clinic function tests, not falls or daily-life activity.
Evidence tier (confidence: moderate). One gold SR+MA, but with an active comparator (strength
training, itself effective) and RCT-grade primaries: el Hadouchi 2022, 15 RCTs / 583 participants,
healthy older adults (mean age >65), Cochrane RoB + GRADEpro. The activity-test outcomes are GRADE HIGH;
what caps confidence is (a) surrogate distance — the trials measured function tests, and none
measured daily-life physical activity — and (b) unblindable primaries (RoB rated serious, blinding of
outcome assessment high-risk by default). A second gold MA (Jimenez-Lupion 2023) targets fall-risk-predictor
tests but its pooled effects fall below the tests’ minimal detectable change and it still does not measure
fall incidence, so it neither upgrades certainty nor closes the falls gap (see the type-F section below).
(inferred from el Hadouchi et al., 2022; Jiménez-Lupión et al., 2023)
Why the mode matters — power declines faster than strength, and it is the limiting factor for daily tasks
Two mechanistic facts, both with human corroboration, make velocity rather than force the better training target for this stratum:
- Power falls faster than strength with age. «Several studies have revealed that the annual decline in muscle power is larger than the annual decline in muscle strength in older adults», the direction «consistent» across studies though the magnitude varies by age and sex. The substrate is fast-twitch fibre loss: «The degeneration of fast twitch fibers, which are responsible for explosive power, occurs more rapidly than the decrease of muscle strength.» (el Hadouchi et al., 2022)
- Power, not force, is usually the binding constraint on function. «in daily activities, such as getting up from a chair, the ability to move with a sufficiently large speed (emphasizing muscle power) is more often the limitating factor than the ability to produce a sufficiently large moment.» So the capacity that degrades fastest is also the one the task actually gates — a Layer-1 argument for targeting it. (el Hadouchi et al., 2022)
The head-to-head effects — power beats strength on power itself and on function tests
Standardized mean differences (random-effects), power training vs an age-matched strength-training comparator. Anchored on the Discussion Summary / Results body / GRADE table (which agree); a labelling caveat on the abstract follows the table.
| Outcome | SMD (95% CI) | k | GRADE certainty | Importance |
|---|---|---|---|---|
| Muscle power (leg / chest press, UE+LE pooled) | 0.99 (0.54 to 1.44) | 16 | moderate | important |
| Generic activity tests (400 m / 6 min walk, chair rise reps, sit-to-stand, balance) | 0.43 (0.23 to 0.62) | 11 | HIGH | critical |
| Speed-emphasis activity tests (chair rise s, SPPB, gait speed, TUG, stair climb, CMJ, floor rise) | 0.36 (0.06 to 0.68) | 9 | HIGH | important |
- Read the sizes right. The large SMD (~0.99) is on muscle power itself — the trained capacity, nearest the intervention. On the function tests that matter to a person, the advantage over strength training is small-to-moderate (0.36-0.43). That gap is the expected shape: the benefit attenuates as you move from the trained quantity toward function, and the comparator is already effective. (inferred from el Hadouchi et al., 2022)
- Certainty is best where it counts. The generic-function outcome is both GRADE HIGH and rated critical — the cleanest decision anchor. The muscle-power estimate carries the heterogeneity (I2 75-86%) and an asymmetric funnel plot flagging possible publication bias, so it is graded moderate in the GRADE table (the abstract prose calls LE power “high” — an internal inconsistency; the table is the authoritative artifact). (el Hadouchi et al., 2022)
- The abstract transposes two labels. The abstract reports «generic activity-based tests SMD: 0.37, 95% CI 0.06 to 0.68 … activity-based tests emphasizing movement speed SMD: 0.43, 95% CI 0.23 to 0.62», i.e. the two CIs are swapped relative to the Results body, GRADE table, and Discussion (generic 0.43 [0.23-0.62]; speed 0.36 [0.06-0.68]). The decision is unchanged either way — power wins on both, both CIs exclude zero — but the per-category CI comes from the body/table, not the abstract. (el Hadouchi et al., 2022)
Against doing nothing, the effect is larger — the active comparator is deflating the head-to-head
A secondary meta-analysis pooled the studies with a non-training control. Power training vs not training: muscle power SMD 1.12, generic tests 0.73, speed tests 0.74 — «even larger» than versus strength training. This is the substitution frame made explicit: judged against the realistic alternative (an older adult already doing conventional strength work), power emphasis adds a small-to-moderate function increment; judged against inactivity, both the power and the strength arm carry most of their value from training at all. The mode choice is a second-order refinement on top of the first-order train vs not decision — the same Layer-1 ordering that governs Resistance Training Prescription - Load Sets and Frequency. (inferred from el Hadouchi et al., 2022)
The load-bearing gap — function TESTS are not daily-life activity, and falls were never measured
This is where the page stops honest, and it is exactly the Surrogate Outcomes boundary:
- No study measured the participation domain. «None of the included studies used the level of physical activity in daily life as an outcome», and the authors are explicit that the chain does not close itself: «Neither an increase in muscle strength nor a better performance on activity tests guarantees an effect on a physical activity in daily life (participation domain).» So the HIGH-certainty result is on lab/clinic capacity tests, one step short of what an older adult actually does at home. (el Hadouchi et al., 2022)
- Falls were not an endpoint here — and the companion MA does not close the gap. The function tests el Hadouchi pools — gait speed, timed-up-and-go, chair-rise time, SPPB, stair climb — are the predictors of fall risk, but this MA did not measure falls or fractures. The companion power-training MA that does target fall-risk-predictor tests (Jimenez-Lupion 2023) still measures only surrogates, not fall incidence, and its two pooled effects fall below the tests’ minimal detectable change — so whether power training’s function-test edge converts to fewer falls remains an open gap, folded in the type-F section below. (inferred from el Hadouchi et al., 2022; Jiménez-Lupión et al., 2023)
- Heterogeneity in the intervention itself may understate the effect. The review’s power-training definition «does not specify the load», so studies mixed a high-velocity/low-force approach with a low-velocity/high-force one; «Literature suggests an ideal load of 20-30% of the 1-repetition maximum (1RM) when training muscle power in older adults», and «muscle power is best trained through the high velocity and low force approach» — so the pooled speed-test SMD (0.36) is plausibly a floor, diluted by studies that trained power the less-effective way. (el Hadouchi et al., 2022)
The second gold MA attenuates rather than confirms — type-F, not independent corroboration
A second gold SR+MA (Jimenez-Lupion 2023, 12 studies / 478 subjects, PEDro) pooled power training on the two most-used fall-risk-predictor tests. It looks like independent corroboration of the function-test story, but it is not — and reading it carefully lowers rather than raises what can be claimed.
Independence check (why this is type-F, not type-E). The two MAs share trials, so their agreement is not independent backing. Bottaro 2007, Henwood, and Marsh appear in both trial lists (searched across both source directories); author lists are disjoint but that does not restore independence — the same primary studies drive both pools. And the two are not even estimating the same contrast:
| Parameter | el Hadouchi 2022 | Jimenez-Lupion 2023 | Same quantity? |
|---|---|---|---|
| Comparator | strength training, active (+ secondary vs no-training) | «another type of training program or control group» (mixed) | NO — power-vs-strength vs power-vs-mixed |
| Effect metric | SMD (standardized) | MD (raw: seconds, reps) | NO — not directly poolable |
| Certainty tool | Cochrane RoB + GRADE (function tests HIGH) | PEDro scale, no GRADE | NO |
| Outcome | function-test batteries (generic + speed) | two named tests: TUG (s), 30s-STS (reps) | PARTIAL — TUG/STS sit inside el Hadouchi’s speed battery |
| Endpoint class | surrogate (function tests) | surrogate (fall-risk predictors); NOT fall incidence | both surrogate; neither measures falls |
| Trials | 15 RCTs | 12 studies, 6 pooled | OVERLAP -> not independent |
(inferred from el Hadouchi et al., 2022; Jiménez-Lupión et al., 2023)
The pooled effects are below the minimal detectable change — and the authors say so. Power training beat the comparator on both tests, but only just: TUG MD -0.31 s (95% CI -0.63, 0.00; P=.05) and 30s-STS MD 1.71 reps (95% CI -0.26, 3.67; P=.09), each with a CI that touches or crosses zero. Both sit below the reported minimal detectable change for the test (TUG 3.2 s; 30s-STS 3.3 repetitions), so the difference is smaller than the test’s own measurement noise. The authors’ own limitations section calls this not sufficient «to accurately state that power training provides a relevant improvement over other training modalities and their relation to fall risk» — a walk-back that contradicts the paper’s own abstract conclusion, «power training increases functional capacity related to fall risk further than other types of exercise». The honest reading is the limitations section, not the headline. (A second internal inconsistency: the abstract and the Data-Synthesis body transpose the two subgroup counts — which test had 6 studies vs 4 — so subgroup k is not reported reliably; the MD/CI agree across both and are what carry the claim.) (Jiménez-Lupión et al., 2023)
What it does add: a fall-timing mechanism. The one genuine forward step is mechanistic, and it sharpens why velocity is the right target for falls specifically. A fall from standing lasts «between 0.27 and 0.63 s» — too short a window to recruit maximal strength — so the decisive quantity is «the force that can be generated in the initial 200 ms of muscle contraction» (rate of force development, RFD), which the source calls «decisive in preventing a fall», and RFD is what power (velocity) training raises. This is a directional, human-corroborated mechanism (marked as mechanism, not an outcome finding): it explains the plausibility of the function-test edge reaching falls, without providing the falls-incidence evidence that would confirm it. (Jiménez-Lupión et al., 2023)
Net effect on the fabric (type-F attenuate/bound). The composite of the two MAs beats either alone by bounding the claim: the function-test advantage of velocity over strength is real and GRADE-high (el Hadouchi), but on the specific tests that predict falls the between-modality difference is below clinical detectability (Jimenez-Lupion), and neither MA measures fall incidence. The falls-incidence gap therefore persists as a named G-gap — it would need a power-training RCT with prospective fall-count follow-up, which the authors explicitly call for («a direct outcome measure of fall risk (post-intervention fall history)»). (inferred from el Hadouchi et al., 2022; Jiménez-Lupión et al., 2023)
Intersections — how this refines the held exercise pages
- It refines resistance training alone does not reduce falls. Exercise for Preventing Falls in Older Adults holds that RT alone carries no fall benefit (Sherrington RaR 1.14 [0.67-1.97], very low), and that the active ingredient is balance. El Hadouchi does not overturn this — it measured function tests, not falls — but it says the mode of resistance training is not neutral for the functional-performance tasks (chair rise, TUG, gait speed) that sit upstream of falls: velocity emphasis beats force emphasis on exactly those. The companion MA that targeted fall-risk-predictor tests (Jimenez-Lupion 2023) does not carry the edge to the falls endpoint either — it measures the same surrogate class with below-MDC effects (type-F section below) — so treat power training as a function-test lever, not a proven fall lever. (inferred from el Hadouchi et al., 2022; Jiménez-Lupión et al., 2023; Sherrington et al., 2019)
- It sharpens the function is the stronger card reading. Muscle-Strengthening Activity and Mortality argues strength training’s strong card is function/falls, not its very-low-GRADE mortality signal. El Hadouchi adds a GRADE-high functional result to that card — and specifies the kind of resistance training (power/velocity) that maximizes it for older adults. (inferred from el Hadouchi et al., 2022)
- It is a different axis from the prescription dial. Resistance Training Prescription - Load Sets and Frequency optimizes load/sets/frequency for strength vs hypertrophy surrogates in general adults; this page optimizes training mode for function in older adults. Complementary decisions, not the same one — which is why this is its own nucleus rather than a facet there. (inferred from el Hadouchi et al., 2022)
- Mechanistic complement to the protein lever. The older-adult stratum’s blunted anabolic response (Anabolic Resistance) and higher per-meal protein target (Protein Intake for Older Adults) are the nutritional side of defending function; power training is the non-nutritional side, and both act on the same fast-twitch-fibre / neuromuscular substrate.
Decision relevance
- For the older adult already resistance-training, add a velocity/power emphasis. On GRADE-high evidence it buys a small-to-moderate increment in physical-function-test performance over conventional heavy-load strength work — on the chair-rise, gait-speed and timed-up-and-go tasks that gate independence.
- But the first-order lever is training at all. Versus not training, both power and strength arms carry most of the benefit; the mode choice is a refinement, not the rock. Someone not training should not wait to find the “right” mode.
- How to do it (directional, not a located optimum): a high-velocity, low-force approach — literature suggests ~20-30% 1RM moved fast — rather than slow heavy lifting, since velocity is what degrades and what the tasks demand. Hold the load number loosely (the review could not standardize it).
- Do not oversell the endpoint. This is function-test capacity, GRADE-high but one step short of daily-life activity (which no study measured) and of falls (never an endpoint here). Frame it as a function lever pending the falls evidence.
(inferred from el Hadouchi et al., 2022)
Limits
- Active comparator; no independent corroboration. The GRADE-high head-to-head is one gold SR+MA (el Hadouchi), and its active comparator deflates the effect (strength training is itself effective). The second gold MA (Jimenez-Lupion) shares trials, so it is not independent backing (type-F, not type-E), and its own fall-predictor effects are below the minimal detectable change.
- Surrogate distance. Outcomes are lab/clinic function tests; no daily-life physical activity, no falls, no fractures, no mortality. Post-intervention only — no follow-up, so durability is unknown.
- Unblindable primaries. RoB serious (blinding of outcome assessment high-risk by default; lack of allocation concealment); muscle-power funnel plot asymmetric (possible publication bias); heterogeneity I2 75-86% on the muscle-power and several speed subgroup pools.
- Intervention under-specified. No standardized power-training load/velocity protocol across studies; baseline degree of power loss not accounted for, though power training «should ideally be targeted in persons with an established decline in muscle power.»
- Internal numeric inconsistencies in the source (abstract vs body label transposition; a GRADE prose-vs-table grade mismatch for LE muscle power) — handled above by anchoring on the body/GRADE table.