Is excess intensive care mortality in the United Kingdom concealed by ICU mortality prediction models?

The comparative performance of intensive care units (ICUs) is measured with casemix adjustment systems such as APACHE, SAPS and MPM [1]. These systems calculate predicted hospital mortality based upon reason for ICU admission, degree of physiological derangement, chronic health status, age and medical intervention. Predicted hospital mortality is calculated using data collected shortly before and after ICU admission [1]. There are considerable limitations to casemix adjustment systems [2, 3]. However, if the standardised mortality ratio (SMR; observed mortality/predicted hospital mortality) is taken as an indicator of the effectiveness of ICU treatment, outcome is not clearly worse in the United Kingdom compared with elsewhere [4–9]. If the average predicted hospital mortality of admissions, rather than the SMR, is used to compare ICUs then large differences between countries emerge. In the North Thames region of the United Kingdom a group of ICUs contribute information to a database [10]. The average predicted hospital mortality by APACHE II [11] for 12 762 patients from 15 ICUs in this database is 28.6%. Another British ICU database reports a predicted hospital mortality of 27.2% [4]. In other countries average predicted mortality is generally lower, for example 19.8% [5], 18.8% and 15.1% in the United States [12]. Out of 37 ICUs in the United States, four reported an average predicted hospital mortality of more than 25% [12] whereas only two of 15 ICUs in North Thames had a predicted mortality less than 25%. Data from over 13 000 ICU admissions in the United Kingdom, eight other European countries and North America showed that the British hospital mortality for these patients was highest at 32.4%, compared with a median of 21% for the other European countries and 19.7% for North America [13]. Preliminary data from the European Consortium for Intensive Care Data (ECICD) using SAPS II to assess severity of illness show that intensive care patients in the United Kingdom are sicker than any of the eight other participating countries (ECICD abstract, not published). A Canadian study reported a predicted hospital mortality of 24.7% [6] and one of 20% was calculated for a group of Brazilian ICUs [7]. Patients already in hospital account for a high proportion of high-risk ICU admissions. In our data, ICU admissions from the ward are 21.7% of total admissions. These have a 52.9% hospital mortality (1466 deaths), compared with 22.3% (1156 deaths) of those admitted from the operating theatre/recovery and 30.2% (1081 deaths) from the accident and emergency department. Of patients admitted to ICUs following external cardiac massage or defibrillation, 42.9% (677 patients) came from the ward. These patients had a 79.5% mortality. In our own hospital 34.8% of ICU admissions of patients who had been in hospital at least 24 h were following a respiratory or cardiac arrest on the ward [14]. In 1996 there were 142 cardiac arrest calls to the wards following which 33 patients (23%) were admitted to the ICU. Our research [14] and that of others [15] suggests that it is possible to identify early those ward patients likely to require ICU admission or suffer a cardiac arrest. Early recognition of these patients may allow management to prevent deterioration in physiological values or to prevent arrest. Such intervention is likely to improve outcome. The incidence of cardiac arrest on the ward may therefore be a useful indicator of the quality of care. Compared with the United Kingdom, in some other countries a higher percentage of resources is given to caring for critically ill patients [16, 17] and ICU admissions have a lower average predicted mortality suggesting that patients are likely to have access to appropriate care earlier. If patients are identified early and admitted to the ICU, in particular before a respiratory or cardiac arrest, their predicted mortality will be less but, given appropriate ICU treatment, so will the observed mortality. There will be no difference in the SMRs and no indication of the improved ICU outcome. The relative lack of critical care resources and the high predicted mortality of patients admitted to British ICUs point to the possibility of an excess mortality compared with better resourced medical systems. Early identification of critically ill patients may help improve care for these patients on the ward or facilitate early admission to an appropriate high-dependency area or ICU. This is likely to decrease the number of deaths without altering the SMR. The use of casemix adjustment systems to compare ICU performance will conceal rather than reveal this excess mortality. Funding to support the North Thames ICU audit system was provided by the North Thames (East) Regional Health Authority. We are grateful to the intensive care units who supplied data and appreciate the time and effort they contributed in order to collect the data.