Brisk walking led at all three bone sites. At the hip, on three trials.

Six kinds of exercise went into the comparison, and the same name came out at the top of all three columns. A hierarchical bayesian network meta-analysis published in The BMJ on 9 September 2026 pooled 124 randomised trials of structured exercise in 18,429 adults aged 40 and over, sorted every programme into one of six categories, and compared each against non-exercise controls for bone mineral density, the amount of mineral packed into a given area of bone, read off a dual energy x ray scan in grams per square centimetre. Brisk walking or jogging came first at the lumbar spine, first at the femoral neck, and at the total hip it was the only one of the six to move the measurement at all.

At the lumbar spine it added 0.013 g/cm2 over controls, with a 95 percent credible interval, the range the data leave plausible for the estimate, of 0.006 to 0.021. Combined aerobic and resistance training matched that figure exactly, 0.013 g/cm2, on a tighter interval of 0.010 to 0.016. Whole body vibration followed at 0.009 (0.004 to 0.013) and resistance training at 0.005 (0.002 to 0.008). Mind-body exercise and mixed aerobic exercise shifted lumbar spine density by no amount the analysis could separate from zero.

The order changes with the site. At the femoral neck, brisk walking or jogging sat at 0.009 g/cm2 (0.001 to 0.018) and mind-body exercise, meaning tai chi and yoga, at 0.007 (0.002 to 0.013), with resistance training and combined training level behind on 0.005 each. At the total hip the list runs to one item: brisk walking or jogging, 0.021 g/cm2 (0.005 to 0.038). Nothing else there separated from control.

Then comes the outcome that actually breaks. Twenty six of the 124 trials, covering 11,132 participants, counted fractures, and the ranking rearranges itself. Mixed aerobic exercise showed the largest reduction, an odds ratio of 0.29 (0.11 to 0.78), meaning the odds of a fracture in that group ran at 29 percent of the odds in the control group. Mind-body exercise followed at 0.58 (0.36 to 0.93). Brisk walking or jogging came in at 0.68, on an interval of 0.44 to 1.06 that crosses 1, the value meaning no difference. Combined aerobic-resistance training sat at 0.94 and resistance training at 0.91, neither of them separable from control. No kind of exercise came out with more fractures than its control group.

How we know

A network meta-analysis is a way of ranking several treatments when few trials have compared them head to head. It uses the comparisons that exist to infer the ones that do not, so a trial of walking against nothing and a trial of resistance training against nothing can be made to speak to walking against resistance training. The authors searched PubMed, Embase, the Cochrane Central Register of Controlled Trials and Web of Science from inception to January 2025, updated the search to January 2026, applied no language restriction, and registered the protocol in advance with PROSPERO. They took randomised trials enrolling adults aged 40 and over that compared structured exercise with a non-exercise control or with another exercise programme, and that reported bone density at the lumbar spine, femoral neck or total hip by dual energy x ray absorptiometry, or counted fractures. Programmes shorter than four weeks were excluded, as were trials in which participants started or changed osteoporosis drugs at randomisation. Control groups were usual care, no structured exercise, or low intensity stretching or relaxation not expected to influence bone density. The work was funded by the National Natural Science Foundation of China, and the authors declare no competing interests beyond that support.

What went in was 124 trials, 480 treatment arms and 18,429 participants, on programmes running from 12 to 208 weeks. The headline count is not the count behind each result. Lumbar spine density came from 94 of those records (222 arms, 6,553 participants), femoral neck from 86 (201 arms, 6,462), total hip from 49 (115 arms, 3,670) and fractures from 26 (62 arms, 11,132). Ninety of the 124 trials, 73 percent, enrolled women only; 24 enrolled both sexes and 10 enrolled men only. The United States contributed 22 trials, China 14, Australia 13 and England 9.

To put a yoga class and a jogging programme on one scale, every intervention was converted into metabolic equivalents of task in minutes per week, written METs-min per week: a MET is a multiple of the energy the body uses at rest, and the measure multiplies that intensity by the minutes spent at it each week. The authors then had to decide how much change in density would count as worth having. They set it against the clock. Adults past 40 lose roughly 1 to 2 percent of bone density a year, so they prespecified 1 percent and 2 percent of the pooled baseline density as the lower and upper bounds of a minimal clinically important difference, the smallest change a patient would notice the benefit of. On their own reading, reaching the lower bound means a programme has held off about a year of ordinary decline, and they note that individual kinds of exercise rarely reached the upper bound.

The dose, and where it stops rising

Pooling every modality, the relation between weekly dose and density was not a straight line. It rose, curved and flattened, an inverted U at nearly every site and for nearly every kind of exercise, mind-body exercise being the exception. At the lumbar spine the estimated effect peaked at 914 METs-min a week (0.023 g/cm2, 0.016 to 0.031) and had already reached 0.019 (0.012 to 0.027) at 600. The minimal dose predicted to reach clinical importance was 391 METs-min a week at the lumbar spine, 600 at the femoral neck and 641 at the total hip. Brisk walking or jogging taken on its own needed more than the pooled figure: 427 at the lumbar spine, 753 at the femoral neck, 789 at the total hip.

The authors' summary number is 600 METs-min a week, which they put at roughly 2 to 3 hours of brisk walking or jogging, with gains clustering near their maximum around 900 to 1,000 and not climbing beyond. They are careful about the falling right-hand side of that curve. It may be real, or it may be that people in the most demanding trial arms did less of what was asked of them, or that so few trials tested such volumes that the dip is noise. The subgroup patterns get the same treatment: men reached the threshold more readily than women, middle aged adults more readily than those over 60, and people of normal weight more readily than those with overweight or obesity, but all of that is rated very low certainty and none of it was tested by formal interaction analysis.

What holds the ranking up

The numbers behind first place are small. Brisk walking or jogging entered the lumbar spine network on 10 trials and 245 participants, the femoral neck on 8 trials and 206, and the total hip on 3 trials and 68. Combined aerobic-resistance training, level with it at the spine, came from 39 trials and 1,456 participants. The authors flag the consequence themselves. Funnel plots, which plot each trial's effect against its size and should fan out symmetrically, came out lopsided at the lumbar spine and the total hip, the pattern left behind when small studies reporting large effects reach print and small studies reporting little do not. Several small walking and jogging trials reported unusually large effects, and the authors say those estimates may be inflated and should be read judiciously.

Nobody can be blinded to whether they are doing tai chi, so 91 of the 124 trials carried some concerns or a high risk of the bias that follows from participants and staff knowing who is in which group. Overall on the Cochrane risk of bias tool, 28 trials (22.6 percent) came out at low risk, 59 (47.6 percent) at some concerns and 37 (29.8 percent) at high risk. Graded through CINeMA, the framework that rates how much to trust each comparison in a network, confidence ranged from moderate to very low, and for the top-ranked modalities it was low to moderate. One further limit belongs to the unit itself: a MET counts energy spent, not the peak force or the loading rate that actually deforms bone, so these dose figures are public health approximations rather than mechanical thresholds.

Why it matters

Osteoporosis and the fractures that follow it affect more than 200 million people, and lifetime fracture risk runs at 30 to 50 percent in women and 15 to 30 percent in men. Guidelines have long agreed that physical activity helps, without agreeing on which kind or how much: some professional bodies suggest 30 minutes of daily walking, the World Health Organization sets 150 to 300 minutes of moderate activity a week for general health, and clinicians, worried about falls, hesitate to hand older patients high impact work. What this review adds is a ranking by modality with a number attached, derived rather than asserted, and the authors propose 600 METs-min a week as a baseline for guidelines to work from. Their physiological argument for walking is that it lands a moderate, repeated load of about 1.5 to 2 times bodyweight, which is enough to reach the threshold at which bone responds, while the heavy loading that works best in a laboratory is often prescribed too cautiously in older adults to reach it at all.

What the review does not add is a promise about fractures. The modality that led on density did not reduce them here, and the two that did are supported by 119 and 524 participants. Density and fracture are related outcomes, not the same one. The conclusion the authors reach is cautious rather than triumphant: confidence low to moderate, subgroup patterns untested, and a literature in which 73 percent of trials enrolled only women. What would settle it is a trial powered for fractures rather than for scans, long enough to see whether early density gains hold, and with enough men in it to say so.