Mortality is an important metric in quality informatics for assessing the quality of critical care. In general we might conclude that intensive care units (ICUs) with low mortality rates provide better care than ICUs with higher mortality rates.
However, this conclusion may or may not be true because the risk of mortality varies from patient to patient. For example, in general an 80 year with septic shock and co-existing coronary artery disease will be at higher risk for death than an otherwise healthy 25 year old. Risk-adjustment of mortality rates is necessary to interpret them. The most robust methodology to accomplish this is prospective risk adjustment based on both co-existing diseases and acute physiology (for example, blood pressure, heart rate, temperature, critical laboratory values, etc). The best examples of this type of risk-adjustment are the APACHE (applied physiology and chronic health evaluations) scores that have evolved through several iterations.
These are used to predict the probability of mortality. These prospective scoring systems (there are several available besides APACHE) are invaluable for risk-adjusting mortality rates as a measure of ICU quality, as well as being used to have frank and open discussions with family members about prognosis. However, calculation of any of the prospective risk-prediction scores is a bit labor-intensive, sufficiently so to preclude manual calculation for each patient who comes to a busy ICU. Thus implementation of these prospective mortality prediction tools requires an informatics solution. The Cerner Corporation, which licenses the latest iteration of APACHE, has developed such a solution. However, not every institution has Cerner as a vendor.
At our institution we are working to implement a slightly simpler scoring system. This scoring system utilizes the following variables (present on admission): blood pressure, heart rate, Glasgow coma scale, age, as well as a series of binary variables, mechanical ventilation, acute renal failure, cardiac dysrhythmia , cerebrovascular accident , GI bleed , intracranial mass effect, history of chronic kidney disease, history of cirrhosis, history of metastatic neoplasm. The calculation is easily performed on an Excel spreadsheet if the above data is available. This data is available in most electronic medical records. The challenge is the database query, finding the data and populating the spreadsheet. However, implicit in this assumption is an even greater challenge. Most of the binary variables relate to specific diagnoses or conditions present on admission or as part of the patients medical history.
Thus this project cannot proceed without structured data entry by some clinician, either physician or nurse. In our particular institution, we are further developed with nursing documentation than physician documentation. We are currently evaluating the nursing database to determine if the needed data is documented, and if not, do we have the opportunity to revise the database. It is relatively easy using our current system to find the admission blood pressure or heart rate or patient’s age, but querying the database to determine if the patient has a history of chronic kidney disease, cirrhosis, or metastatic neoplasm is more challenging when the system was not designed for that from its inception.
A more remote possibility is to use natural language processing to extract key diagnoses from the physician’s narrative data. Given the current state of the art, this will probably not be feasible for a number of years.
- Higgins TL, Teres D, Copes WS, Nathanson BH, Stark M, Kramer AA. Assessing contemporary intensive care unit outcome: An updated Mortality Probability Admission Model (MPMo-III). Crit Care Med 2007; 35:827-835.