Metrics, even when well intentioned, can distort. Particularly when you are dealing with something as amorphous as health treatment outcomes. A better hospital may actually have a higher mortality rate because its reputation brings a lot of more critically ill patients there... and the general disease categorization would not necessarily reflect this.

In the US (and probably elsewhere as well) there are complex standardized codes used for insurance, medicare and other issues. Doctors and hospitals MUST encode patients and this too gets to be a distorting force. Outright fraud notwithstanding, there is also a natural inclination to code as many codes as appropriate for each patient to optimise reimbursement. There are in every organization either in house or outside specialists who do the coding (not generally the doctor) because, besides lost revenue, miscoding can result in fines.

The bad thing about statistics like this is that you get what feels like solid information, clear crisp values, to multiple decimal places. But it's standing on a foundation of quicksand.