Article Review

Breaking the paradigm: Scores are of no clinical relevance for outcome prediction in abdominal septic shock patients




Most clinicians can recognize septic shock, but if you ask them, you get a hundred definitions, although consensus conferences should have resolved this issue. In this paper we use strictly the term „septic shock“, the term „severe sepsis“ is intentionally not applied, since in a former study we could demonstrate that „severe sepsis“ comprises almost identical patients when considering exclusively abdominal septic shock.

Not only in order to document severity of illness, but also to estimate the prognosis of these critical ill patients, different scoring systems have been developed.

The best outcome predictor would be one that warns the physician on first day of ICU admission or when septic shock first appears (this is usually the second day of the patient's ICU stay according to our analysis). Our results demonstrate that none of the scoring systems achieves this goal. Only in the last three days of the ICU period, scores reach acceptable AUC values, whereby  the SOFA score, based  on ten variables, achieves the best AUC of all scores. Like the SOFA score, the data driven neural network approach performs similarly, using only three variables (bpt).

Although scores and neural network under investigation provide relevant outcome predicition information only in the last three days of the ICU stay of patients  (ie they are without clinical relevance), it is worth to look closer at the data.   

Looking at the CI values in Table II, we notice that scores are difficult to use for individual patients: a score value does not indicate death or survival with a high confidence resulting in long CIs. The neural network results on the non-score datasets are more reliable since CI length is usually shorter. The SOFA score has the lowest interval length (0.13) of all the scores. Therefore, it is the best score for abdominal septic shock patients from this point of view. For example, SOFA's CI length is 0.13, NN's CI length is only 0.09. Considering all datasets (e.g. lungs, heart, bpt, freq16), the results show the superiority of neural networks compared with scores when considering the confidence of a classification of individual patients. 

The resulting alarm system based on our analyses produces reliable alarms (in the last three days of the ICU stay there were ten times more alarms for deceased patients then for survivors).  The alarm system that was trained with data of the last three days represents the patient conditions that lead to death or survival with a high probability. Although the alarm system was trained with data of the last three days, it can be used as an online bedside alarm system. Right from the start of the patients' ICU stay physicians are warned when patients reach the same critical condition as deceased patients had within the last three days.

In April 2002  a prospective randomised multicenter study was initiated to check the clinical usefulness of the web-based alarm system (see study protocol at