Modeling Phenotype Interactions in Sickle Cell Anemia.

Predicting a broad risk of selected serious vasoocclusive complications of sickle cell anemia is possible. For example, patients with higher hemoglobin concentrations are less likely to have a stroke. Yet, it has not been feasible to integrate the many clinical and laboratory abnormalities of sickle cell anemia into a predictive model that permits an understanding of the interactions among common clinical and laboratory abnormalities. From the database of the Cooperative Study of Sickle Cell Disease, we examined clinical and laboratory data from nearly 1500 individuals with sickle cell anemia, with or without coincident α thalassemia. We used these data to develop a Bayesian network that describes the interactions between clinical and laboratory data and their associations with the risk for complications of sickle cell anemia. Bayesian networks are multivariate models that represent the complex structure of interactions between many variables by a network of interrelated modules. The modules can be learned from data using statistical techniques and can be used to describe how changes in some variables affect other variables and ultimately the risk for phenotypes of interest. Our model shows that a complex network of interactions between clinical and laboratory variables underlies common complications of sickle cell anemia and ultimately death. Particularly important is the protective role that α thalassemia appears to play in common complications of sickle cell anemia. For example, α thalassemia, by decreasing erythrocyte density, reduces hemolysis and is associated with lower levels of bilirubin and an associated decreased risk for priapism. Bilirubin levels may reflect nitric oxide (NO) availability and NO may be invoved in the etiology of priapism. α thalassemia is also associated with smaller numbers of reticulocytes that are strongly associated with a decreased risk for acute chest syndrome and osteonecrosis; and it is associated with higher level of fetal hemoglobin and a reduction in leukocyte counts with a significant decreased risk for stroke and death. This model can be used to predict the occurrence of certain complications of sickle cell anemia and early death, given the presence of other disease complications and variations among common laboratory variables. For example, our model predicts a 10% risk for stroke at early age and an 11% risk for early death in patients with sickle cell anemia without α thalassemia compared with a 4% risk for stroke and 1.5% risk for early death for individuals with coincident α thalassemia. However, the predictive power of this model is limited, and we conjecture that this is a reflection of the omission of the genotypic changes that underlie the phenotypes. In related work using this same patient population, we analyzed genetic polymorphisms in candidate genes and showed than 25 SNPs and 4 clinical variables, including α thalassemia and fetal hemoglobin, were associated with increased risk of stroke and that this model predicted the occurrence of stroke in 114 individuals in a different population with 98% accuracy. The lack of the same predictive power of our current model suggests that genetic variants play a fundamental role in susceptibility to stroke and other complications of sickle cell anemia.