cell anemia (see Fig. 6) offers an example of the power of these models. The network captures the interaction between 31 SNPs in 12 genes that, together with fetal hemoglobin, modulate the risk for stroke. We showed the prognostic accuracy of this model by predicting with 98.2% accuracy the occurrence of stroke in 114 subjects not included in the primary analysis and showed that this approach outperformed statistical models based on logistic regression [122]. We used the same approach to develop a model of severity of sickle cell disease defined by survival. Phenotypic heterogeneity is a well known characteristic of this disease. Patients have different rates of complications, such as pulmonary hypertension, painful episodes, acute chest syndrome, and osteonecrosis, as well as variations in levels of laboratory variables. To integrate individual disease variables into a global measure of severity, we developed a network that describes the complex associations of 25 clinical and laboratory variables, deriving a score that we used to define disease severity (0, least severe to 1, most severe) as the risk of death within 5 years [129]. This initial network was validated in 140 patients whose disease severity was assessed by expert clinicians and 210 adults where
