Robust Causal Estimation from Observational Studies Using Penalized Spline of Propensity Score for Treatment Comparison