The impact of statistical choices on neonatal intensive care unit quality ratings based on nosocomial infection rates.