Access to Optimal Treatment of Acute Myeloid Leukemia Patients is Affected by Sociodemographic Factors: a French Population-Based Study.

Background During their care pathway, AML patients not admitted to Specialized Hematology Units (SHU) have less access to curative treatment. We aim to determine whether access to optimal curative treatment is affected by sociodemographic factors. Methods We included 1,033 incidents AML-cases diagnosed between 2012–2016 from three French “départements”. We considered patients managed in reference hospitals SHU within 5 days(n = 297) received “gold-standard” treatment. Treatment was "curative-treatment” if intensive chemotherapy and “non-curative” otherwise. Firstly, we trained a Gradian Boosting Machine (GBM) algorithm on 80%(n = 238) of "gold-standard" cases to learn how they were treated and validated the model on the remaining 20%(n = 59). Next, GBM predictions were contrasted with actual treatment. Using multivariable logistic regression, we examined how non-optimal treatment (discrepancy between predicted curative and observed non-curative treatment) was associated with sociodemographic factors. Patients with predicted non-curative treatment were excluded as uninformative on access to curative treatment (n = 471). Results The rate of “curative treatment” was 84.8% (252/297) for gold-standard patients vs. 33.5% (247/736) for others. The three most influential predictive factors in gold-standard patients were age (68.3%-influence), t-AML/MDS (15.8%), and the AML-others subtypes (5.4%). A total of n = 102(9.9%) patients were in non-optimal treatments. Living in Basse-Normandie (0.65-times;95%CI [0.5,0.8]) and over 30minutes from a reference hospital were strongly associated with a non-optimal treatment. Conclusion There are geographical disparities in access to optimal treatment, potentially linked to medical desert situations or medical system organization which must be addressed.