TABLE 4

Propensity Score Matching to Estimate the Effect of NICU Levels 2 Through 4 or PICU Compared With Newborn Level 1 or Ward

LOS, dCost, Adjusted 2010 US $Probability of 30-d Readmission Estimated difference from propensity score matching0.11 (−0.02 to 0.26)a506 (325 to 688)a,b.00 (−0.02 to 0.02)a Multivariable regression coefficients1.02 (0.91 to 1.15)a1.51 (1.47 to 1.56)a,b.74 (0.50 to 1.09)a Estimated difference from multivariable regression models0.03602b−.01 • From propensity score matching, one assumes an increase of 0.11 d in LOS for NICU. Instead, researchers used a model to predict a 0.03-d greater LOS in the NICU. In cost, one assumes an increase of$506 on the basis of propensity scores, whereas in the model, we instead expect \$602. The difference in probability of revisit is almost identical between the 2 methods. With the possible exception of cost, there does not seem to be a practical difference between findings from the 2 different statistical methods, and there seem to be no statistically significant differences.

• a Values in parentheses in data cells are 95% confidence intervals of point estimates.

• b P < .01.