OUTCOME PREDICTION FOR THE INDIVIDUAL PATIENT IN THE ICU

A very difficult clinical problem facing surgeons is knowing when furt her treatment is futile and no longer appropriate in a patient who has developed severe complications after surgery and is being treated in an intensive care unit. It is now possible to prolong the process of d ying among such patients. This results in unnecessary pain and loss of dignity for the patient, anguish and distress for the patients's rela tives and is dehumanising for the clinical and nursing staff. It has a lso tremendous implications in the use of limited health care resource s. A computer model designed to aid this process has to have the follo wing properties: it must reflect the dynamic pathophysiological proces s and be able to predict death with extreme accuracy and early in the clinical course. The Riyadh algorithm uses computerised dynamic, trend analysis of daily organ failure scores (APACHE II score corrected for the number and duration of organ failures), noting the rate of change in score relative to that of the previous day and an absolute thresho ld to predict death has been developed for this purpose. The algorithm was developed by tracking the daily scores of 200 IUC patients until their death or discharge from the intensive care unit. It was subseque ntly validated perspectively on 831 patients. During the validation pr ocess, the clinicians were blinded to the predictions. There were 290 deaths and the program predicted 109 deaths (38% of all deaths) with n o false-positive predictions. Forty per cent of the predictions were m ade within 48 h in the ICU and 74% within a week. In another study inv olving 227 patients to compare the accuracy of early outcome predictio ns by doctors and nurses with those by the program, it was found that doctors and nurses had false-positive diagnosis rates of between 11% a nd 17%, while program had none. The implications on resource allocatio n was studied at Guy's Hospital, London, where the cost of the treatme nt of each patient was also recorded. There were 3498 consecutive admi ssions of whom 555 died. The program predicted 136 patients to die and 130 of these patients died. Two patients died subsequently more than 3 months after discharge from the ICU but were omitted in the analysis . The false-positive diagnosis rate in the series is 4.41%, the sensit ivity 23.4% and the specificity 99.8%. The cost to treat the 136 patie nts who were predicted to die was 1,635,000 Pounds, which was 16.7% of the entire ICU expenditure over the study period and 46.6% of the cos t of treating the 555 patients who died. If the assumption is made tha t treatment was withdrawn on the day of prediction and the patients di ed on the following day, then the cost of care after prediction for th e 136 patients was 745,213 Pounds. This sum of money could have been u sed to treat 500 uncomplicated post cardiac surgery patients.