The efficacy question for exercise on the bone surrogate in postmenopausal women — and the surrogate discipline that gates what the answer is worth. The safety page (Exercise Safety in Osteoporosis) asks whether exercise breaks a fragile skeleton; the falls page (Exercise for Preventing Falls in Older Adults) asks whether it prevents the fall that fractures one. This page asks the third, distinct decision: does loading the skeleton raise bone mineral density (BMD), and — the load-bearing question — does a BMD gain actually mean fewer fractures, the outcome a person cares about. First source held: Mohebbi 2023, a gold updated SR-MA (80 studies, 94 training and 80 control groups, 5581 participants pooled) of controlled exercise trials >= 6 months, applying the inverse-heterogeneity (IVhet) model. (Mohebbi et al., 2023)
The effect on the surrogate: real, positive, small-to-moderate
Exercise raises BMD at all three sites, but the effect is modest and the magnitude framing is the source’s own:
- Lumbar spine (LS): SMD 0.29 (95% CI 0.16-0.42); heterogeneity substantial (I2 = 68%), and publication-bias/small-study evidence HIGH at this site.
- Femoral neck (FN): SMD 0.27 (95% CI 0.16-0.39).
- Total hip (TH): SMD 0.41 (95% CI 0.30-0.52); heterogeneity low (I2 = 20%), publication-bias evidence negligible.
The authors classify the size plainly: «the average SMD for BMD effects on LS, FN, and TH can be classified as moderate at best (i.e., 0.2 to 0.5)» (Mohebbi et al., 2023). So the honest headline is a small positive standardized effect on a surrogate, best-evidenced and cleanest at the total hip, and weakest (noisiest, most bias-suspect) at the lumbar spine.
The moderators are all NULL — the for whom question does not separate
The update’s stated purpose was to find effect modification — who benefits more. It found none. Across three pre-specified group comparisons the exercise effect on BMD did not differ significantly (all p > .09):
- osteopenia/osteoporosis vs normal BMD (bone status),
- early vs late postmenopausal (menopausal status),
- predominantly supervised vs predominantly non-supervised programs.
This is a route-(b) null: the search for a subgroup with a larger relative BMD effect came up empty on the three axes most expected to modify it. It does not deny route-(a) stratification — absolute fracture benefit still scales with baseline fracture risk even when the relative BMD effect is flat (Baseline Risk and the Relative-Absolute Split) — but it removes the positive interaction evidence that would license personalizing the exercise prescription by bone or menopausal status. The high residual heterogeneity (I2 up to 68%) says trials differ a lot in effect, but the measured moderators do not explain it.
The load-bearing limit: BMD is a SURROGATE, and transmission to fracture is not shown here
(inferred from Mohebbi et al., 2023) — Mohebbi measures BMD, not fracture. BMD is a surrogate for the patient-important outcome (fracture, and downstream function/independence), and the whole surrogate discipline applies (Surrogate Outcomes). Two cautions run in opposite directions and both matter:
- BMD may UNDER-capture exercise’s true fracture benefit. Exercise prevents fractures through at least two channels, and BMD sees only one. The bone-loading channel (mechanostat: load raises BMD) is what this page measures; the falls channel (balance, strength, gait -> fewer falls -> fewer fractures) is largely BMD-independent and is where the larger fracture signal for older adults sits (Exercise for Preventing Falls in Older Adults: exercise cuts falls, RaR ~0.77). A BMD meta-analysis is structurally blind to that channel — the streetlight effect on the outcome side.
- A small statistical BMD gain may NOT transmit to meaningful fracture reduction. The corpus already holds a worked inversion of exactly the BMD->fracture link: GRADE’s handbook names «bone mineral density for fractures» as a proximate surrogate (rate down only one level), yet in the TRAVERSE testosterone RCT the surrogate rose while clinical fractures rose too (HR 1.43) -> Surrogate Outcomes. Proximity plus mechanism plus a consistently-moving marker were still not enough. That case is a different exposure (a drug, not loading) and cannot be transported to exercise, but it is the standing proof that a BMD gain is not self-evidently a fracture reduction. The corpus also holds a cleaner, non-drug within-exposure inversion — Fluoride and Bone Health, where fluoride raises BMD while fractures rise above ~1.5 mg/L, because it adds density but degrades bone quality. Exercise, fluoride and testosterone are three exposures raising the same surrogate with the outcome moving different directions, and the transmission tracks how the density was gained (loading builds structurally normal bone) -> Surrogate Outcomes (the three-exposure cluster).
(inferred from Mohebbi et al., 2023) Mohebbi measures BMD, not fracture, and points at the fracture SR from its own group — «the favorable effect of supervised exercise protocols on fracture» (Mohebbi et al., 2023). That fracture SR is now held (Hoffmann 2022, next section) and closes the surrogate->outcome step for the loading channel — but from the same group, so it refines rather than independently corroborates.
The fracture endpoint is now held: exercise reduces major osteoporotic fractures (type-F)
The patient-important outcome the BMD surrogate above only stands in for is directly measured by Hoffmann 2022, a gold SR-MA (11 controlled exercise studies; 9715 exercise vs 9592 control participant-years; mixed-effects conditional Poisson) — the fracture-incidence MA this page had been holding open. Exercise cuts major osteoporotic fracture (MOF): pooled «RR: 0.75, 95% CI: 0.54–0.94, p = .006» (Hoffmann et al., 2022) (the Results section reports the same 0.75 with a slightly tighter lower bound, 95% CI 0.59-0.94 (Hoffmann et al., 2022) — a within-paper discrepancy; both exclude 1.0). Heterogeneity was negligible (I2 <1%) with no funnel-plot asymmetry. MOF is defined per FRAX (hip, vertebral, forearm/wrist, humerus), a low-trauma / fragility endpoint. (Hoffmann et al., 2022)
- Absolute frame (inferred from Hoffmann et al., 2022): from the raw counts (151 MOF in 9715 EG participant-years vs 196 in 9592 CG) the crude rate falls ~2.0 -> ~1.6 per 100 participant-years — roughly one MOF averted per ~200 participant-years of exercise. Crude and unadjusted (the pooled RR 0.75 is the adjusted estimate), shown only to size the rock.
- This is a type-F refinement, not independent (type-E) corroboration. Hoffmann is the SAME Erlangen/Kemmler group as Mohebbi (Kemmler, von Stengel, Jakob, Kerschan-Schindl, Lange, Peters, Thomasius, Uder shared) drawing on an overlapping data pool and shared methods. It supplies the outcome leg the BMD MA lacked; it does not independently confirm the BMD effect. The wiki still holds no independent group’s exercise-fracture MA — the robustness gap persists. (inferred from Hoffmann et al., 2022)
- The protocol question stays open — which is where the BMD-moderator nulls and the fracture data agree. Hoffmann’s two pre-specified moderators are both NULL: intensity progression (p=.133) and study duration (p=.883, though only the >12-month arm reached significance on its own, IR 0.77, 95% CI 0.61-0.97, vs the wide-CI short arm IR 0.70, 95% CI 0.23-2.15) (Hoffmann et al., 2022). The authors decline to derive a prescription: «diverging study and exercise characteristics along with the close interaction of exercise parameters prevented the derivation of reliable recommendations for exercise protocols for fracture reductions» (Hoffmann et al., 2022). So on both the surrogate (Mohebbi’s bone-status / menopause / supervision nulls) and the patient-important outcome (Hoffmann’s progression / duration nulls), the effect is real but the modifiers do not separate — a consistent route-(b) null across both legs of the same group’s work.
The supervision discordance — surrogate silent, real outcome not
A sharp instance of why the surrogate can mislead: supervision made no significant difference to BMD in this meta-analysis (the null above), yet Mohebbi’s group argues supervision matters more for bone than for falls, and their separate fracture SR credits supervised protocols with the fracture benefit (Mohebbi et al., 2023). So a program feature that is null on the BMD surrogate may still bear on the patient-important outcome (adherence, dose fidelity, safety) — a caution against reading no BMD difference as no decision difference. (inferred from Mohebbi et al., 2023)
Confidence, and the non-independence caveat
confidence: medium-high, and the legs are deliberately split:
- BMD-surrogate efficacy exists and is positive: well-evidenced (gold updated SR-MA, consistent across three sites, robust IVhet model). This leg alone would be high.
- Moderators: NULL / low — no effect modification shown on bone status, menopausal status, or supervision (Mohebbi), nor on intensity progression or duration (Hoffmann); do not personalize the prescription on them.
- Fracture transmission: now DIRECTLY measured, moderate — the upgrade from medium. Hoffmann 2022 (gold SR-MA) shows exercise cuts major osteoporotic fracture (RR 0.75, CI excludes 1.0), so the surrogate->outcome step is no longer a gap for the loading channel. What still caps it below high: the fracture MA is from the same group (non-independent, below), the effect is moderate with a CI near 1.0, and no protocol modifier separates.
Non-independence (binding). Mohebbi is the Erlangen/Kemmler group (Kemmler, von Stengel, Shojaa, Jakob among the authors) and is an explicit update of Shojaa et al. — «further evidence» in the source’s own words. Hoffmann 2022, the fracture endpoint, is the SAME group (8+ shared authors) — so the surrogate (BMD) and outcome (fracture) legs are not independent of each other: convergence across these reviews reflects a shared team, data pool, and method, and must not inflate confidence. The wiki holds no independent research group’s exercise-BMD or exercise-fracture meta-analysis yet — a genuine robustness gap for both the surrogate effect and its outcome confirmation.
The author’s own verdict on meta-analytic precision here
Notably, the source itself doubts that pooling delivers a prescription: «their practical application for deriving dedicated and reliable exercise recommendations is rather limited» (Mohebbi et al., 2023). The high between-trial heterogeneity means the mean SMD hides wide variation in what individual protocols achieve — the pooled number says exercise-in-general nudges BMD, not which program a person should do. The load/impact/dose specifics live in the programming frontier (Resistance Training Prescription - Load Sets and Frequency), kept peripheral by Layer-1.
The creatine adjunct does not add to loading’s BMD effect (bounds the supplement space)
A natural follow-on question — can a supplement amplify resistance training’s bone effect? — has a clean null for the most-touted candidate. Forbes 2018, a gold MA (5 RCTs, n = 193, adults >50 / postmenopausal), pooled creatine + resistance training vs resistance training + placebo and found no added BMD at any site (whole body, hip, femoral neck, lumbar spine — all pooled MD ~0, CIs within about ±0.01–0.03 g/cm²), despite every trial showing creatine enhanced muscle mass or strength: «these adaptations did not translate to greater BMD» (Forbes et al., 2018). So the loading effect above is not augmentable by creatine on the BMD surrogate at studied doses — a muscle-yes / bone-no dissociation that bounds the adjunct space (inferred from Forbes et al., 2018). Two escape hatches stay insufficient rather than closed: bone geometry / strength (one 12-month trial increased femoral shaft subperiosteal width, which BMD cannot see) and longer / higher-frequency / relative-dose protocols showed positive signals; and no creatine trial measured fracture (Forbes et al., 2018). Full appraisal: Creatine Supplementation (Bone leg).
Decision relevance
- The exposure is worth pulling, and the fracture case is now direct. For a postmenopausal woman, exercise produces a small positive BMD effect on a surrogate AND a measured reduction in major osteoporotic fractures (Hoffmann RR 0.75) — plus the largely BMD-independent falls channel (Exercise for Preventing Falls in Older Adults). Frame the recommendation on the composite (bone + falls + strength + function), not on the BMD number alone; the fracture endpoint firms the bone leg but the modifiers (which protocol) stay unresolved.
- Non-substitutable by the obvious drug — for the pleiotropy, not the BMD. An antiresorptive (bisphosphonate) raises BMD and cuts fracture and would, for the bone-density outcome alone, shrink the exercise rock. But exercise is pleiotropic (falls, muscle, gait, cardiometabolic, mortality -> Muscle-Strengthening Activity and Mortality) and a single-channel drug does not substitute for those other-channel benefits — so the lever’s rank falls only for the BMD outcome the drug actually covers. The drug’s efficacy/limitations are not held in the fabric — a genuine G-gap (bisphosphonate/antiresorptive comparator); do not assert its effect as fact here, and the choice between lever and drug stays the person’s (Layer 3).
- Do not personalize the exercise prescription on bone or menopausal status — the moderator evidence is null; the stratifying variable that does work is baseline fracture risk (route (a), absolute benefit), which BMD/FRAX supplies as a prognostic input, not a treatment target.
- Menopause context: accelerated post-menopausal bone loss is one of the shifting levers (Menopause and the Shifting Levers); exercise is a modifiable counter-lever on the bone axis, weighted alongside the others there.
- The loading->BMD mechanism is the SAME one that inverts stress-fracture risk across strata. The mechanostat here (load raises BMD) is why a lean athlete can be bone-fragile despite exercising — the low-energy-availability runner on Stress Fracture Risk Factors in Runners has low BMD and high bone-stress risk — while a high-mass obese novice loads the skeleton heavily and plausibly has higher BMD, shifting their binding injury constraint to soft tissue/joint rather than bone. Same mechanism, opposite stratum conclusions.