The delivery-format decision for the physical-activity lever: when exercise is the prescription, must it be supervised (a trainer/clinician present, usually facility-based) or can it be done unsupervised at home? The question matters because unsupervised home delivery removes the two largest real-world barriers to the lever — cost and travel/scheduling — so if it retains most of the benefit it is the realistic default for a self-directed person, with supervision reserved for where it is load-bearing.
Bottom line
For physical-function and well-being outcomes in older adults (>=60 y), supervised delivery (SUP) produces at most a small advantage over unsupervised (UNSUP), and that advantage mostly does not survive sensitivity analysis. Of the outcomes where SUP beat UNSUP, only knee extension strength remained significant when fragile studies were removed (Gómez-Redondo et al., 2024). On the substitution framing (judge a strategy against its realistic alternative), the alternative to unsupervised home exercise is often no program at all — so UNSUP is a sufficient default for most outcomes, and the source’s own verdict is that it is a safe, cost-effective alternative where SUP is not feasible (Gómez-Redondo et al., 2024).
The evidence
Gomez-Redondo (SR+MA of RCTs, Sports Med 2024): 34 studies, n=2830, older adults >=60 y, direct SUP-vs-UNSUP head-to-head (Gómez-Redondo et al., 2024); mean study quality TESTEX 10 (range 4-14) (Gómez-Redondo et al., 2024).
SUP was significantly superior to UNSUP on six outcomes, but only one was robust to sensitivity analysis (Gómez-Redondo et al., 2024):
| Outcome | k (n) | SUP vs UNSUP effect (95% CI) | p | Robust to sensitivity? |
|---|---|---|---|---|
| Knee extension strength | 10 (1,126) | SMD 0.18 (0.07, 0.30) | 0.002 | Yes (leave-one-out; I^2 0%) — but Begg’s p=0.037, mean TESTEX 9 |
| Usual gait speed | 8 (834) | SMD 0.29 (0.04, 0.55) | 0.026 | No |
| Sit-to-stand | — | SMD ~0.25 | 0.050 | No |
| Timed-up-and-go | — | SMD ~0.21 | 0.035 | No |
| Lean mass | 5 (118) | MD 1.05 kg (0.76, 1.34) | <0.001 | No (lost on removing one study) |
| HRQoL | 9 (953) | SMD 0.21 (0.02, 0.40) | 0.035 | No (lost on removing almost any study) |
| Handgrip, functional reach, balance, max gait speed, 6MWT, VO2peak, BMI, body mass, body fat | — | all non-significant | ns | — |
The surviving effect (knee extension SMD 0.18) is small, and sits near the lower SMD band; the others cluster at SMD ~0.2-0.3 but are not stable (Gómez-Redondo et al., 2024).
Attendance does not explain the SUP edge (adherence largely unmeasured)
The intuitive story is that supervision helps by keeping people turning up. Attendance does not support it: pooled attendance was similar in both arms. «The incidence of adverse events and falls as well as the attendance to the pro- gram (81%) were similar in the SUP and UNSUP groups.» (Gómez-Redondo et al., 2024)
But the two 81% figures pool different study sets (21 SUP, 20 UNSUP; UNSUP range 38-100% vs SUP 60-100%), and the authors flag them as possibly biased: UNSUP attendance was mostly diary-reported, and 21 of the 34 UNSUP arms involved some supervision. Adherence proper (doing the prescribed intensity / duration / exercises) was measured in only two studies, with mixed results: in Sandberg, full adherence was 24% (SUP) vs 26% (UNSUP), partial adherence 71% vs 48%, per the review’s text; its summary table gives both pairs the other way round — full adherence «SUP: 26% UNSUP: 24%», partial (labelled attendance there) «SUP: 48% UNSUP: 71%» — so neither the 2-point nor the 23-point gap has a reliable direction (Gómez-Redondo et al., 2024) (Gómez-Redondo et al., 2024).
So the SUP edge is not explained by attendance, which was similar on pooled, partly diary-reported figures within the trial window; whether it is adherence-mediated is largely unmeasured — within-session compliance (doing the prescribed intensity) stays a live mediator alongside what the supervised sessions contained. (corrected 2026-10-06: cannot be an adherence effect … not in the dose completed -> attendance-not-explanatory, adherence largely unmeasured; self-critique.)
Caveat on horizon: these are trial-window attendance rates. Whether unsupervised programs sustain adherence over the longer term (after the trial’s contact and measurement fall away) is not resolved by this evidence — a named gap, not a null.
The dose confound — a likely contributor to the non-robust edges
SUP and UNSUP arms were generally not dose-matched: supervised sessions tended to use harder exercise selection, higher intensity, and more volume. «However, it must be noted that the number of adverse events and falls reported in this study may be underestimated given that exercise dose variables are generally not equated between SUP and UNSUP groups (i.e., more difficult exer- cise selection, higher intensity, and volume for SUP).» (Gómez-Redondo et al., 2024)
So the comparison is confounded: SUP is partly a proxy for more/harder exercise, not for supervision as such. This reframes the decision — the question is less “supervised vs unsupervised” than “can an unsupervised person reach the same dose (intensity/volume) a trainer would push them to?” Where they can, the format gap should shrink, to the extent dose is the channel. The source’s own similar-programme sub-analysis partly bears this out and partly does not: «Sub-analyses revealed superior benefits of SUP on knee extension strength when only considering those studies that applied a similar intervention in both SUP and UNSUP groups. However, no significant benefits were found for the remaining outcomes. Beneficial effects of SUP over UNSUP were also observed for knee extension strength, STS, functional reach test, TUG, usual gait speed, lean mass, and HRQoL when separately analyzing those studies in which participants performed ≥ 66% of the sessions in the assigned condition.» (Gómez-Redondo et al., 2024) The ≥66% subset is a supervision-fidelity filter, not an adherence one — «the SUP group performed at least 66% of the sessions under supervision» — and the similar-intervention subset means «targeting the same muscle groups and with similar characteristics» (Gómez-Redondo et al., 2024). So the dose reading fits the matched subset (only knee extension survives, below; TUG and usual gait speed show «A trend towards statistical significance» there that did not reach it — not a shown zero (Gómez-Redondo et al., 2024); corrected 2026-10-06: 5-study null -> trends; self-critique), while the broad edge in the genuinely-supervised subset does not separate the readings, because those programmes were not required to match; hold the dose explanation as a likely contributor, not a shown one (refined 2026-10-06, Gomez-Redondo chunks 01-03). The source’s only dose statement is the safety-section caveat above, not a measured efficacy comparison; its discussion names other candidate mechanisms beside it — «obtaining direct feedback from a professional, the social component of being with other participants, or having greater material resources» (Gómez-Redondo et al., 2024).
One finding resists this deflation: knee extension strength stayed significant even in the sub-analysis restricted to studies that applied a similar intervention in both arms, and in those where >=66% of SUP sessions were actually supervised (Gómez-Redondo et al., 2024). So knee extension is the one place a genuine supervision effect is hardest to explain away as a pure dose artifact. Two cautions: SMD 0.18 is small, and the row carries a nominally significant Begg’s test (p=0.037; mean TESTEX 9, range 4-13) — a possible small-study signal the source does not discuss (Begg’s appears only in its methods and Table 4; searched: Begg / publication bias / small-study across chunks 01-04) (Gómez-Redondo et al., 2024). Why supervision would help here is a hypothesis: better loading of a lower-limb strength exercise that is easy to under-load alone is the wiki’s guess, untested in the source; the authors instead propose training specificity (lower-body exercises «with similar movement patterns» to the test) and a «lower starting level» on these outcomes (Gómez-Redondo et al., 2024). Knee extension strength is also a prognostically weighty target — it tracks ADL/function and mortality -> Grip Strength and Mortality, Low Muscle Mass and Mortality.
Substitution framing and who this changes
- For a self-directed, reasonably capable person: unsupervised home exercise is a sufficient realistic default for most function and well-being outcomes; the measured cost of going it alone is small and mostly not robust. The exception worth considering paying supervision for is dedicated lower-limb strength loading (the knee-extension signal — a small, possibly bias-inflated one), plausibly because unsupervised loading is lower (untested here; the authors instead propose training specificity and low baseline strength).
- For someone who would otherwise not exercise at all: the realistic alternative to UNSUP is nothing, so UNSUP dominates — the source’s explicit position: «However, for those unable to perform SUP, UNSUP may represent a safe and cost-effective alternative to ensure physical exercise.» (Gómez-Redondo et al., 2024)
- Safety is not a reason to insist on supervision here: adverse events and falls were similar in both arms (Gómez-Redondo et al., 2024) — though note dose was not equated, so a dose-matched unsupervised program might carry a different (unmeasured) injury profile for a deconditioned novice -> Resistance Training Prescription - Load Sets and Frequency.
Transportability — older adults, not middle-aged
The entire evidence base is >=60 y. The barriers supervision addresses (technique, loading confidence, fall risk during exercise) are plausibly larger in older adults than in a middle-aged person, so the small SUP advantage found here is, if anything, an upper bound for a younger, more capable stratum — the delivery-format gap should be no wider, and likely narrower, for the middle-aged. But this is a directional transport argument, not measured: no head-to-head SUP-vs-UNSUP evidence in a middle-aged population is held. Named gap.
AWAITS
Synthesis
The page is configurative, single-source (a provisional domain-opener for the delivery/adherence cell). The decision-relevant move it makes, present in no outcome row on its own: the SUP-vs-UNSUP contrast is partly a dose contrast wearing a format label — unequated dose is a confounder the authors flag (SUP arms trained harder), pooled attendance was similar though adherence was largely unmeasured, and the matched subset leaves knee extension standing; feedback, social contact and resources are the authors’ other named candidates — so attainable dose is a decision variable alongside presence of a supervisor, with lower-limb strength loading (a small, possibly bias-inflated signal) the one residual place supervision most plausibly earns its cost. (corrected 2026-10-06: largely a dose contrast … because attendance was equal -> partly; self-critique.) Confidence is low: single source, small and mostly non-robust effects, trial-window horizon, and an older-adult population that may not transport.