Sickle severity selectors strike out

In a report in this issue of Blood, Quinn and colleagues apply a published model for early prediction of sickle cell disease severity to an independent cohort of sickle cell patients from Dallas, identified by newborn screening; in this analysis, the published predictors turn out to have negligible positive and negative predictive value.

The tested model, from a landmark 2000 report of the Cooperative Study of Sickle Cell Disease (CSSCD), defined “adverse outcomes” as proxies for disease severity, to include stroke, death, recurrent acute chest syndrome, and frequent pain crises.¹  CSSCD investigators used multivariate analyses to determine risk factors for subsequent events in a cohort enrolled between 1978 and 1988, followed for a mean of 10 years. In this test population, and in a validation cohort, 3 independent predictors were identified: dactylitis before age 1, hemoglobin less than 7 g/dL at steady state during the second year of life, and leukocytosis.

In the Dallas cohort, with initial assessments between 1990 and 2005 and with a 7.1-year mean follow-up, the presence or absence of these risk factors was unrelated to subsequent adverse events.

Aside from 2 general limitations (retrospective analysis and a relatively small sample size, criticisms equally applicable to the CSSCD initial and validation datasets), one potential weakness of the current work is that several study aspects differ from the CSSCD cohort, thus limiting direct comparison between studies. Quinn et al do describe these differences, as well as possible explanations for discrepant findings. There were fewer adverse outcomes in the Dallas group. The Dallas children were followed clinically, with fewer early visits for research-grade assessments and laboratory testing during toddlerhood. Pain episodes were captured in the Dallas cohort only if patients were hospitalized, whereas CSSCD included pain at home, in clinic, and at emergency department visits. Improved adherence to penicillin prophylaxis and improved pneumococcal vaccination occurred after the CSSCD was launched. Hydroxyurea use in children²  and transfusions for stroke prevention based on the Stroke Prevention Trial in Sickle Cell Anemia (STOP) study³  came into widespread use during the Dallas enrollment period. Even if these recent treatments did not affect incidence of severe outcome in the Dallas cohort, since subjects were censored after starting disease-modifying therapy, they can be expected to reduce adverse outcomes in the future.

These methodological differences between the studies do not alter Quinn et al's key finding and its implications. The CSSCD model, applied in an independent cohort, simply did not work to identify a high-risk group. This is not to criticize the prior work. Indeed, in their original paper, Miller et al¹  were careful to call for a validation data set and to suggest conservative application of the model. Rather, Quinn et al conclude that the predictive failure of the model means that, despite the enthusiasm with which the study was greeted initially, these 3 risk factors cannot be used as the selector for referral of patients for high-risk procedures—including potentially curative, but potentially dangerous, hematopoietic stem cell transplantation. Faced with this important new information, hematologists will need to reevaluate prognostication and therapeutic approaches in young children with sickle cell disease.