Gradient Boosted Trees - Confusion matrix

I am running a Gradient Based Trees model. I am getting the attached confusion table which seems 'incorrect' or inverted in terms of positioning of the scores. The ROC curve also is unusual in that it is slopping downwards left to right.

The model is looking at clinical tests by patient admitted to ICU - the original test data had each test labelled 1 - Died or 2 - survived. I have converted this to binomial hence the True-True, True-False categories on the confusion matrix.

Given that the focus should be to highlight those patients who died rather than survived I am concerned that I have omitted some parameter  or block that is that is  distorting the table.

Copy of model and confusion table below.

Thank you.