Effectiveness of different exercise types for improving bone mineral density in adults: A pairwise, network, and dose-response meta-analysis of 162 randomized controlled trials.

BACKGROUND: Comparative effectiveness of exercise modalities across skeletal sites and the dose-response pattern for bone mineral density (BMD) remain incompletely characterized. This review aimed to compare exercise modalities across six skeletal sites and to explore whether MET-min/week could describe a dose-response relationship between exercise exposure and BMD change. METHODS: We conducted a systematic review and meta-analysis of RCTs (PROSPERO: CRD420261345811), searching four databases to March 2026. Three complementary frameworks were applied: three-level pairwise meta-analysis with robust variance estimation, Bayesian network meta-analysis (NMA) with prespecified vague priors and SUCRA rankings, and exploratory dose-response model-based NMA using restricted cubic spline modelling of MET-min/week. Methodological quality was assessed using ROBUST-RCT, GRADE, and an NMA-specific certainty assessment considering intransitivity and incoherence. RESULTS: We included 162 RCTs (10,475 participants; age range, 18-81.6 years). Pairwise GRADE certainty ranged from high (femoral neck) to very low (total hip). Significant pooled effects were observed at five of six sites: lumbar spine (Hedges' g = 0.22), femoral neck (g = 0.28), Ward's triangle (g = 0.23), trochanter (g = 0.15), and total body (g = 0.26). NMA suggested site-specific ranking patterns, but credible intervals were often wide and NMA certainty was commonly limited by imprecision, heterogeneity, and sparse direct evidence. Dose-response analyses suggested an overall peak near 1100 MET-min/week for lumbar spine BMD, but agent-specific optima, especially AT+RT at 1800 MET-min/week, were boundary-sensitive and exploratory. CONCLUSION: Exercise is associated with small-to-moderate, site-specific improvements in BMD across adulthood. Modality rankings and dose-response estimates should be interpreted as hypothesis-generating rather than definitive prescriptions, particularly where direct evidence was sparse or certainty was low.