• English 
  • Spanish 

Potential factors and adjustment of health systems performance indicators. A novel approach with Bayesian networks

Journal: Artificial Intelligence in Medicine
Year: 2008  
Status: Waiting author review
In this status since: 12 Apr 2008
PDF file: 2008_cofino_AIIM.pdf
Authors:
, , Gogorcena, M.A., López, O.,

In this work we analyze the variability observed in hospital outcomes, as measured by some quantitative indicators related to hospital performance and quality of care (length of stay, mortality rate, etc.). In order to use these indicators for benchmarking purposes, the hospital indicator values must be adjusted according to the influence of external factors (e.g., different clinical and patient profiles) not attributable to the hospital performance.
To this aim, we use the Spanish National Health System (SNHS) discharges dataset corresponding to the year 2005 (~3.5 million records). This database collects information about the patient and hospital profiles (age, gender, type of discharge, etc.), the diagnostic and procedures associated with the disease (complexity, severity, diagnosis category, etc.) and also the corresponding indicator values for every discharge. In order to find relevant potential factors of influence among the indicators and the above set of variables we introduce a novel multivariate statistical
model for discrete data (a probabilistic Bayesian network). The resulting model allows explaining part of the indicators’ variability observed among different hospitals in terms of clinical and diagnostic profiles, thus, providing a procedure for the adjustment of the outcome indicators for an unbiased estimation of hospital care quality.