ESPE Abstracts

Background: Childhood hematologic malignancies are no longer a death sentence. With survival rates significantly increasing, focus needs to shift towards diminishing long-term adverse effects, such as bone disorders. In childhood hematologic cancer survivors (CHCS), peak bone mass is not usually attained due to malginancy-related inflammation, treatments employed or subsequent endocrine complications. Thus, low BMD is frequent. However, in the absence of fragility fractures (FF), no clear indication of antiosteoportic agents is establised.

Study aim: As FF lead to pain, disabilities and physical deformities, altering quality of life (QoL), we aimed to develop a risk prediction model for FF in CHCS.

Methods: We registered 40 CHCS (17 with ALL, 9 AML, 11 HL, 2 CML, 1 JMML), aged between 4-19, in remission for at least one year, evaluated in our Pediatric Endocrinology Department between 2016 and 2023 (mean age 11.68 +/-4 years, time from diagnosis 4.71 years). Their assesment included: clinical and biochemical exams and DXA scan. TBLH BMD_HAZ Z-scores were used to account for the effect of stature on BMD. Patients symptomatic for FF were referred for imaging studies. Logistic regression was used to develop a prediction model. Employing backward selection, multiple independent variables (age at diagnosis, BMI, TBLH BMD_HAZ Z-scores, glucorticoids, methotrexate, HSCT, radiotherapy, hypogonadism, IGF1 z-scores for Tanner stages, vitamin D, PTH and TSH levels) were analysed and statistically non-significant predictors were removed, until a robust model was built. The model performance was evaluated using the area under the receiver operating characteristic curve (AUC).

Results: In the study population, 5(12.5%) children suffered FF (3 had vertebral fractures, 1 had radial and humeral fractures and 1 had femoral fracture). Among the prediction models analysed, the one based on TBLH BMD_HAZ Z-scores (β=−1.37, P =0.019) yielded the best results (Nagelkerke R Square =0.62) and correctly identified 89% of patients (AUC 0.89, SD 0.1, 95%CI 0.73-1.05). Using the ROC curve, we determined that a TBLH BMD_HAZ Z-score ≤ −2.09 predicts the risk of FF with 91% sensitivity and 75% specificity.

Conclusions: We developed a machine-learning model for determining the risk of FF in CHCS, based on TBLH BMD_HAZ Z-scores. Our results emphasise the need for close DXA monitoring in CHCS. Also, as Z-scores ≤ −2.09 predict the risk for FF with 91% sensitivity and 75% specificity, independently of other risk factors, the question of early antiresorptive therapies use rises. Our next aim is to prospectively test the model's validity.