Is the Assumption of Homogeneity Violated?

Box's M is significant, while Levene's test is not. This can happen for several reasons:

The covariance between Length of stay and Log-cost is not constant across cells, and thus the model results are suspect.
The covariances are unequal, though not not by much, but the large size of the data file causes Box's M to be overly sensitive to this departure from homogeneity.
The covariances are equal, but the test procedure for computing Box's M, a multivariate test, simply comes up with a different result than the univariate test.
The distribution of Length of stay and Log-cost is different enough from a multivariate normal distribution to cause Box's M to be significant.

In order to help decide whether you should be concerned about the significance of Box's M, some exploratory data analysis is in order. You can use the Explore procedure to check the assumption of normality. With the data file split by the cells, you can use the Bivariate Correlations procedure to see whether the correlations are constant across cells.