Low and Very Low Bone Mineral Density Among Adult Survivors of Childhood Cancer Can be Predicted

Background: Although low bone mineral density (BMD) is prevalent among adult survivors of childhood cancer, evidence-based guidance for surveillance of low BMD among survivors by dual X-ray absorptiometry (DXA) is limited. Therefore, we aimed to develop and validate prediction models for low and very low BMD based on patient and treatment characteristics that identify adult survivors who require screening by DXA.

Methods: Caucasian survivors (N=2412, median attained age, 30.9 [range: 18.1-65.9] years, 60% survivors of hematologic malignancies) enrolled in the St. Jude Lifetime Cohort (SJLIFE, development) and survivors previously treated at the Erasmus Medical Center (validation), the Netherlands (N=410, median age, 24.5 [range: 18.0-49.3] years, 73% survivors of hematologic malignancies) were evaluated with DXA to determine lumbar spine (BMD_LS) and total body (BMD_TB) BMD, with low and very low BMD defined as BMD_LS and/or BMD_TB Z-score ≤-1 or ≤-2, respectively. Backward multivariable logistic regression was used to build prediction models; performance was assessed using receiver operating characteristic curves. Diagnostic values were calculated at different probabilities. Diagnostic values were calculated at different probabilities. The formula P_lowBMD=e^(β₀+β₁*x₁+...+β_n*x_n)/(1+e^(β₀+β₁*x₁+...+β_n*x_n)) was used to predict individual probability of low BMD (β₀=intercept; β_n=regression coefficient (β) of each predictor x_n).

Results: Low BMD was prevalent among 52% and 44% of SJLIFE and Dutch participants, and very low BMD among 20% and 10%, respectively. The model for low BMD included male sex (β=0.50), height (β=-0.05), weight (β=-0.03), age at cancer diagnosis (β=-0.02), current smoking (β=0.19) and cranial irradiation (β=0.31) as significant predictors. Areas under the curves (AUC's) were 0.72 (95%-CI 0.70-0.74) in SJLIFE and 0.70 (95%-CI 0.65-0.75) in the Dutch cohort. Sensitivity (69.6%), specificity (63.9%) and accuracy (66.9%) were optimized at a probability of 0.50. The model for very low BMD included male sex (β=0.55), height (β=-0.05), weight (β=-0.04), cranial (β=0.31) and abdominal (β=0.22) irradiation, yielding AUC's of 0.75 (SJLIFE cohort) and 0.72 (Dutch cohort).

Conclusion: Validated models for low BMD including sex, age at cancer diagnosis, height, weight, smoking status, cranial irradiation and abdominal irradiation correctly identified BMD status in most Caucasian survivors through age 50 years. Clinical applicability can be warranted using our online calculator.