The initial search for systematic reviews and meta-analyses did not identify sufficient evidence to answer the review questions. We screened the full texts of 36 reviews and no eligible reviews were found (Fig. 1). The main reasons for exclusion were reviews including younger participants (n = 34), participants with osteoporosis at baseline (management instead of prevention, n = 12), and not investigating whole body physical activity (e.g., whole body vibration, n = 7).

Flow chart of studies investigating physical activity and osteoporosis prevention in older people included in the WHO report (left size), in the expanded search for individual studies (middle) and expanded search for systematic reviews (right side)

Amongst the 36 reviews which had their full text screened, 25 reviews included potentially eligible studies and their full texts were identified and assessed by two reviewers. We used the same eligibility criteria, but no restriction was applied for publication year of individual studies. We found 36 studies (trials and observational studies) investigating the association between physical activity and prevention of osteoporosis (34 identified from the reviews and 2 from hand searching) [41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76].

The expanded search yielded 772 records and the full texts of 92 records were screened (Fig. 1). A total of 24 studies met the eligibility criteria, 23 identified via PubMed search and one via hand searching. Out of the 24 studies identified, five had already been included in the WHO report [42, 47, 66, 68, 71]. Therefore, the expanded search found 19 additional studies [77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94,95].

The expanded search for systematic reviews identified a total of 366 reviews. We screened the full texts of 58 reviews and no eligible reviews were found (Fig. 1). Amongst the 58 reviews which had their full text screened, 30 reviews included potentially eligible individual studies and after assessing their full text we included 4 additional studies [96,97,98,99], resulting in a total of 59 studies included in this manuscript. The included studies were published between 1980 and 2020. There were 39 randomised controlled trials, 8 quasi-randomised trials and 12 observational studies (8 prospective and 4 retrospective studies). There were three cases where results from the same study were reported across multiple articles [53, 77, 81, 94, 96], all of which were included in this review as they reported results for different follow-up timepoints.

Within the included trials, 40 compared physical activity with a control intervention (Table 1); 11 compared two physical activity programs (Table 2); six trials (Table 3) and eight observational studies (Table 4) investigated different doses of physical activity. A total of 12 observational studies were included, seven investigated total physical activity, one leisure-time physical activity (exercise, transportation and sport), and five planned physical activity (four exercise and one sport-tennis) (Table 4).

The included trials comprised a wide range of physical activity and exercise modalities. Following the ProFaNE taxonomy, most studies (n = 19) investigated more than one category of exercise (classified as multiple); 11 studies investigated balance and functional exercises, 12 resistance; five endurance; nine investigated a combination of balance and functional exercise or resistance with bone loading; and one 3D exercises (Tai Chi).

Most included studies recruited from the general older population. Studies in which all participants had already been diagnosed with osteoporosis were excluded. Four studies excluded participants with osteoporosis at baseline [53, 93, 95, 96]. Three studies included participants on the basis of having some level of frailty [46, 73, 74]; five articles reporting results from two studies included only participants with osteopenia [77, 81, 91, 92, 94]; two studies included only obese participants [87, 88]; two studies investigated prostate cancer survivors without osteoporosis [93, 95]; one study included participants who had had surgical repair of a hip fracture no more than 16 weeks prior to study entry [43]; and one study included participants with increased risk for falls and fracture [90]. One study investigated lifelong tennis athletes. Twenty-eight studies included only women whereas six investigated only men. Five studies (reported in 8 articles) included participants who were younger than 65 years at study entry, but met the age criteria at follow-up [66, 77, 81, 86, 91, 92, 94, 98].

The included studies reported results for a range of different outcomes (n = 32), and the most common ones were measures of BMD and BMC. We performed an overall assessment of the evidence according to the study’s main outcome. If the study did not specify a main outcome, we selected the outcome we considered to be most relevant to the intervention (e.g., whole body for exercises involving the whole body). We selected lumbar spine in preference to hip when both were presented, and the exercise was primarily undertaken in a standing position. Where exercises were mostly performed in non-standing positions (e.g., seated, supine) and targeted the lower limb, hip measures were preferred. For studies that reported multiple hip measures, preference was given to total hip measures, if available. Preference was given to BMD when compared to other measures, such as BMC. We undertook two additional assessments according to the two most commonly reported outcomes across the included studies, which were measures of femoral neck BMD and lumbar spine BMD.

The overall quality of included trials was moderate (median 5, range 1 to 7). The PEDro total scores are reported for all relevant studies in Tables 1, 2, and 3 and the scores for each item are reported in Additional file 4, Table 1. The overall risk of bias of longitudinal studies using the modified QUIPS tool is reported in Table 4. Six longitudinal studies had low risk of bias (Additional file 4, Table 2). The most common sources of bias were related to exposure measurement, study attrition and study confounding.

A total of 40 articles reporting on 37 studies (30 randomised and 7 quasi-randomised trials) investigated physical activity interventions compared with a control group (Table 1). Overall the sample size for the trials was small (median: 50, range: 16 to 283) and the median follow-up length was 12 months (range 3 to 144). Meta-analysis revealed a significant but relatively small overall effect of exercise when the results of the main outcome from each study were pooled (standardised effect size 0.15, 95% CI 0.05 to 0.25, 20 trials, Fig. 2). The quality of evidence was moderate as per GRADE system, downgraded for study limitations, meaning that the true effect is likely to be close to the estimated results (Table 5 and Additional file 5, Supplementary Table A). The overall results suggest that physical activity interventions probably improve bone health and prevent osteoporosis in older adults.

Effect size (95% confidence interval) of physical activity interventions on the main outcome by pooling data from 20 studies comparing physical activity versus control using random-effects meta-analysis (n = 1560)

We also summarised the evidence for the two most commonly reported outcome measures across the included studies. Meta-analysis found a non-significant and small overall effect of physical activity on femoral neck BMD (standardised effect size 0.09, 95% CI − 0.03 to 0.21, 14 trials; Fig. 3). The quality of the evidence was low, downgraded for study limitations and publication bias, suggesting limited confidence in the results (Table 5 and Additional file 5, Supplementary Table B). Overall, these results suggest that physical activity interventions may improve BMD of the femoral neck in older adults.

Lumbar spine BMD was the second mostly commonly reported outcome measures. Meta-analysis found a significant but relatively small overall effect of physical activity on lumbar spine BMD (standardised effect size 0.17, 95% CI 0.04 to 0.30, 11 trials; Fig. 4). The quality of the evidence was moderate, downgraded for study limitations, suggesting that the true effect is likely to be close to the estimated results (Table 5 and Additional file 5, Supplementary Table C). The overall results suggest that physical activity interventions probably improve BMD of the lumbar spine in older adults.

We included 12 observational studies. Since the studies varied in terms of design, statistical approach and measures of physical activity, we did not perform meta-analysis and apply the GRADE approach. Overall, studies showed a positive effect of physical activity on bone health (Table 4).

As shown in Table 1, programs which had significant impacts were generally of a higher dose. Typical program for which significant intervention impacts were detected in randomised controlled trials were undertaken for 60+ mins, 2–3 times/week for 7+ months [45, 52, 59, 63, 71]. The randomised controlled trials (n = 6) investigating different doses of physical activity on bone health did not suggest a clear dose-response relationship (Table 3) but were probably too small (i.e., lacked statistical power) to detect differences between different doses of physical activity. All eight longitudinal studies investigating different doses of total or planned physical activity on bone health found that higher levels of physical activity were associated with better bone health (Table 4).

Meta-regression revealed a non-significant trend for studies with a higher overall intervention dose (i.e., 7800+ total mins) to have greater effects on femoral neck BMD (p = 0.144), where high dose interventions (7800+ mins) had a moderate effect with a standardised effect size of 0.26, 95% CI − 0.01 to 0.52 and lower dose interventions (