Regression modeling to inform cell incorporation into therapies for craniosynostosis

Designing an appropriate tissue engineering solution for craniosynostosis (CS) necessitates determination of whether CS-derived cells differ from normal (wild-type, WT) cells and what assays are appropriate to test for differences. Traditional methodologies to statistically compare cellular behavior may not accurately reflect biologically relevant differences because they poorly address variation. Here, logistic regression was used to determine which assays could identify a biological difference between WT and CS progenitor cells. Quantitative alkaline phosphatase and MTS proliferation assays were performed on adipose, muscle, and bone marrow-derived cells from WT and CS rabbits. Data were stratified by assay, cell type, and days in culture. Coefficients of variation were calculated and assay results coded as predictive variables. Phenotype (WT or CS) was coded as the dependent variable. Sensitivity-specificity curves, classification tables, and receiver operating characteristic curves were plotted for discriminating models. Two data sets were utilized for subsequent analyses; one was used to develop the logistic regression models for prediction, and the other independent data set was used to determine the ability to predict group membership based on the predictive equation. The resulting coefficients of variation were high for all differentiation measures. Upon model implementation, bone marrow assays were observed to result in 72%-100% predictability for phenotype. We found predictive differences in our muscle-derived and bone marrow-derived cells suggesting biologically relevant differences. This data analysis methodology could help identify homogenous cells that do not differ between pathologic and normal individuals or cells that differ in their osteogenic potential, depending on the type of cell-based therapy being developed.