Ensemble learning uncovers novel metabolomic biomarkers for early osteoporosis prediction in Tibetan plateau populations.

BACKGROUND

Osteoporosis represents a prevalent metabolic bone disorder among middle-aged and elderly populations, with its prevention and early detection holding important implications for clinical practice and public health. While extensive research has characterized osteoporosis in low-altitude populations, plateau regions present unique challenges.

METHODS

This study enrolled 177 adult residents from Lhasa, Tibet. Data collection included demographic characteristics, health profiles, and female reproductive parameters. Osteoporosis is defined by T-score ≤ -2.5 (for males >50 years and postmenopausal women) or Z-score ≤ -2 (for males ≤50 years and premenopausal women). Serum metabolomic profiling identified 3381 metabolites via HPLC-MS/MS. Predictive models were constructed using Least Absolute Shrinkage and Selection Operator (LASSO) and Random Forest (RF) ensemble learning algorithm, with pathway enrichment analysis performed in MetaboAnalyst (P < 0.05).

RESULTS

Our cross-sectional analysis of 177 participants revealed 41 osteoporosis cases (23.2%), predominantly postmenopausal women (36/41). Baseline characteristics showed significant differences in age, gender, difficulty initiating sleep, and menopausal status across groups (all P < 0.05). The 9 metabolites screened by LASSO classification demonstrated exceptional predictive capacity for osteoporosis (AUC: 0.995), while 10 metabolites based on RF better predicted osteopenia progression (R2: 0.848). Z-score-based models outperformed T-score subgroups. Menopausal status and difficulty initiating sleep were identified as high-risk factors, and three metabolites-Dihydro-3-hydroxy-4,4-dimethyl-2(3H)-furanone, gamma-Glutamyltryptophan, and di-homo-gamma-linolenate-emerged as potential early indicators of bone loss. Pathway enrichment highlighted pyrimidine metabolism and related diseases as a potential therapeutic target for bone loss.

CONCLUSIONS

This research proposes a metabolomics-driven model for osteoporosis prediction, establishing an innovative biomarker framework for early clinical intervention in plateau areas.