Thoracolumbar segment-specific volumetric BMD thresholds as candidate alternatives to L1-L2 in opportunistic QCT...

To assess whether alternative thoracolumbar segments reproduce L1-L2-based QCT classification and to derive segment-specific candidate vBMD thresholds corresponding to the L1-L2 cutoffs when L1-L2 cannot be reliably assessed.

In this retrospective cross-sectional study, vBMD from T1 to L5 was measured using a previously validated AI-based opportunistic QCT platform. Mean L1-L2 vBMD classified by ACR cutoffs was the reference. Fifteen alternative two-vertebra segments were assessed by correlation and linear regression, with secondary Youden optimization. Regression-derived thresholds underwent 1,000-resample bootstrap OOB internal validation, with additional patient-level misclassification and age- and sex-stratified analyses.

In 1,244 patients, 15 segments correlated with L1-L2 vBMD (r = 0.873-0.993, all p < 0.001). Regression-derived thresholds for the 80- and 120-mg/cm³ cutoffs were 71.2-122.5 and 112.3-166.2 mg/cm³, respectively. In bootstrap OOB internal validation, AUCs were 0.943-0.996, with sensitivity 66.8 %-93.5 %, specificity 95.0 %-98.7 %, and weighted κ 0.775-0.953. Exact agreement was 73.8 %-94.0 %, with major misclassification ≤ 0.6 %. Weighted κ was 0.783-0.954 with Youden versus 0.773-0.953 with regression-derived thresholds. Correlations were significantly lower in patients ≥ 70 versus < 50 years across five segments but remained > 0.97 at T12-L1 and L2-L3. Age-related threshold variation was greatest at upper thoracic levels and smallest at T12-L1 and L2-L3, with smaller sex-related variation.

Segment-specific thresholds may serve as candidate fallback values for reproducing L1-L2-based QCT classification when L1-L2 cannot be reliably assessed, with L1-L2 preferred when evaluable. Exploratory age-stratified analyses suggested greater stability for segments closest to L1-L2 in patients aged ≥ 70 years.