Introduction

As early as 1912, post mortem examination (PME) was being promoted as a useful medical procedure for correct diagnosis in difficult deaths and for the audit of clinical practice contributing to quality assessment [1]. In recent years, the PME rate has fallen worldwide: In the United States, PME rates decreased from 41% in the 1960s to 22% in the 1970s [2]; in Ireland, rates fell from 30.4% in 1990 to 18.4% in 1999 [3] and in Sweden rates fell from 80 to 39% in just one decade (1977/78 to 1987/88) [4]. Baker et al. [5] reported a mean autopsy rate of 12.7% in 418 US and Canadian institutions in 1989, with nearly two thirds of centers having an autopsy rate between 0 and 10%. A number of reasons can explain this decline, including improved modern diagnostic techniques, the fear of diagnostic errors being identified by autopsy findings, difficulty in obtaining consent from the family, increased work load for the pathologist and doubts regarding the cost/value ratio of the procedure [3, 6].

Published discrepancy rates between ante and post mortem diagnoses vary between 10 and 50% depending on the population studied, the criteria for PME, the completeness of the PME and the techniques used to evaluate the differences [4, 613]. Principal pathological entities uncovered by PME have included mainly undiagnosed nosocomial (particularly fungal) infections [1012], pulmonary embolus [6, 8, 10, 14], myocardial infarction [13] and malignancy [14].

In our multidisciplinary Department of Intensive Care, PMEs are facilitated by the Belgian law of presumed consent, whereby post mortem examination is allowed provided that no formal opposition has been registered before death. Nevertheless, post mortem examinations are not performed routinely if the cause of death is considered obvious. The objective of the present study was to compare pre and post mortem diagnostic findings and investigate the relationship between unexpected findings and the length of ICU stay.

Materials and methods

We reviewed all autopsy findings from patients who died in the 31-bed, medico-surgical Department of Intensive Care of the Erasme University Hospital and who had an autopsy performed in 1999 (Jan 1–Dec 31). All PME were performed on the entire body, with the brain being examined when considered necessary by the pathologist, for example, in trauma patients, comatose patients, patients who had had severe infections with neurological manifestations where the responsible microorganism was not detected, cases of intoxication and when specifically requested by the physician responsible.

The medical records (including demographic data, admission diagnosis, length of stay, clinical, laboratory, radiological and other diagnostic findings) were reviewed by the first investigator (GD) to establish the principal diagnoses and the presumed cause of death for each patient. The autopsy reports were reviewed by the second investigator (MP) and were noted independently. Special histopathological and microbiological techniques for the final post mortem diagnostic protocol were also noted. Unexpected findings, so-called ‘discrepancies’, were classified using a modification of Goldman’s criteria (Table 1) [15]. Patients were classified into two groups according to the duration of stay in the ICU: fewer than 10 days and more than 10 days. The results were also analyzed according to patient age.

Table 1 Modified Goldman’s criteria

Statistical analysis included Χ2-test or Fisher’s exact test to compare categorical variables and t-test to compare mean values. A p less than 0.05 was considered as statistically significant.

Results

From 2,984 ICU admissions in 1999, there were 489 deaths (mortality rate 16.3%). A total of 222 PMEs were performed (autopsy rate 45.4%). The brain was examined in 34 cases (15.3%). The clinical data of the patients are presented in Table 2. The mean ICU stay was 17±13 days (range 1–86 days). PME revealed unexpected findings in 50 (22.5%) patients, which were Class I in 12 (5.4%) patients, Class II in 7 (3.1%) and Class III in 31 (14%) patients.

Table 2 Clinical data

The discrepancies detected by PME are presented in Table 3. All the detected discrepancies were confirmed by histopathological and microbiological techniques during the PME. Major (Class I/II) findings included fungal infections, other nosocomial infections, endocarditis, pulmonary embolism, mesenteric infarction and myocardial infarction. Six of seven patients with fungal infections had not been immunocompromised and all these patients had been treated empirically with the appropriate antifungal agents despite negative fungal cultures. In the patient with endocarditis, the diagnosis had been suspected, but two transesophageal echocardiographic techniques had failed to confirm the diagnosis and blood cultures had remained negative. Minor findings included non-metastatic malignancies, Hashimoto’s disease, pulmonary embolism, Barrett’s esophagus and Candida esophagitis.

Table 3 Major discrepancies (Class I/II) in relation to the fundamental disease

When divided according to length of stay, there were more major findings in patients staying fewer than 10 days than in those staying more than 10 days (10.1% vs 6%). The reverse was seen with minor findings, which were more prevalent in the longer stay patients (26.2% vs 6.5%). All together, there were fewer discrepancies in patients staying fewer than 10 days than in those staying more than 10 days (16.7% vs 32.1%, p<0.001) (Table 4). The mean lengths of stay preceding major and minor unexpected findings were 8.3±6.9 and 16.8±8.4 days, respectively (p<.001).

Table 4. Pre versus post mortem diagnoses in relation to the length of stay

Of the 222 autopsies, 204 (91.9%) were performed in patients older than 50 years. Overall, discrepancies were detected more often in older (21.2%) than in younger patients (1.3%), but this was not statistically significant (p>0.05). However, minor discrepancies were significantly more common in older than in younger patients (31 [14%] vs 0, p<0.05). In older patients, the commonest discrepancies detected were primary malignancies, while in younger patients unexpected findings included disseminated aspergillosis, nosocomial infection, viral pneumonia in one double-lung transplant patient and pulmonary embolism in a trauma patient.

Discussion

The present study reviewed all PMEs performed during a 1-year period in a large, mixed ICU. The total discrepancy rate in our study was 22.5%, and 8.5% of the unexpected findings were related to the cause of death, with 5.4% being Class I errors. This Class I discrepancy rate is lower than that reported in many other studies (T.S1 in the electronic supplementary material). Blosser et al. [6] reported a Class I discrepancy rate of 27% in 41 autopsies from 132 deaths and Roosen et al. [12] a rate of 16% in medical ICU patients. Mort and Yeston [11] found a Class I discrepancy rate of 23% in 149 surgical ICU patients (including transplant patients), with 85% being undiagnosed infections (mainly fungal and viral pneumonia). More recently, in a prospective study, Combes et al. [16] reported a Class I discrepancy rate of 10.2% in 167 ICU autopsies. The most frequently missed diagnoses were infarction, thrombosis and pulmonary emboli. However, in a recent study in a medical ICU population, Twigg and colleagues [13] reported similar rates to ours with an overall discrepancy rate of 23.7%, 4.1% being major discrepancies. In this study, hemorrhage was the commonest major discrepancy detected. Nadrous et al. [17] also noted, in their study of 455 autopsies, a major discrepancy rate of 4%, with a 17% rate of minor discrepancies; cardiac tamponade was the most common major error. These differences in discrepancy rates among studies may be explained not only by different ICU populations, but also by different criteria for autopsy. Studies from hospitals in which autopsies are largely performed only in complicated cases may be expected to show higher discrepancy rates [18].

Interestingly, several groups have studied changes over time in the rates of diagnostic discrepancies detected by autopsy. Two studies reported no changes in discrepancy rates for the decades 1960s, 1970s and 1980s with 10% of studies in all three decades reporting a major discrepancy [15, 19]. However, more recently, Sonderedder-Iseli et al. noted that the frequency of major discrepancies in a university hospital in Switzerland had decreased significantly, from 30% in 1972 to 18% in 1982 and 14% in 1992, p=0.007 [20]. The authors suggested that this decrease was due to improved diagnostic procedures. Using a systematic literature search, Shojania and colleagues [21] identified 53 autopsy series between 1966 and 2000 and reported an average major error rate over the 40 years of 23.5% (ranging from 1.1 to 49.8%). Major errors significantly decreased with time with a relative reduction of 19.4% per decade (95% CI, 1.8–33.8%). Nevertheless, extrapolating their results to a contemporary US hospital, the authors suggest that autopsies would still reveal a major error rate of 8–24%.

Several authors have stressed the value of the PME and the need to encourage its continued use [18, 22, 23], although the PME rate has decreased worldwide (T.S1 in the electronic supplementary material) [15] with recently quoted autopsy rates for ICU populations generally in the region of 20–30% [6, 11, 24], although another Belgian hospital recently reported PME rates as high as 93% [12] and, in Finland, Silfvast et al. reported a rate of 89% over the years 1996–2000 [25]. A study in the UK from 1998–2001 reported an autopsy rate of just 7.7%, with 50% of these being ordered by the coroner [18]. Some have suggested that the fall in autopsy rates may have been in part related to the fact that in 1970 the Joint Commission on Accreditation of Healthcare Organizations dropped the accreditation requirement that hospitals perform a set percentage of autopsies [26]. Anderson et al. [27] reported that autopsy results should be an integral part of the quality assurance process, and Mort and Yeston [11] go so far as to say that the PME is “the ultimate quality yardstick”. However, despite the percentage of “missed diagnoses”, there is no agreement regarding the use of autopsy as a quality assurance tool and there are no data on the cost-effectiveness of systematic autopsy.

Major (Class I/II) discrepancies in our study were associated with diseases characterized by being difficult to diagnose, while minor discrepancies (Class III) included mainly unknown diseases present before ICU admission. Two decades ago, Goldman et al. reported similar unexpected findings in a non-ICU population, including pulmonary embolus, acute myocardial infarction, tumors and infections [15]. The so-called ‘missed diagnosis’ does not, therefore, necessarily imply medical negligence but rather an increased incidence of complicated diseases in the critically ill patient despite advanced medical technology, improved diagnostic tests and techniques, and increased clinical awareness. The development of spiral computed tomography has facilitated the diagnosis of pulmonary embolism [28], yet in seven of our patients autopsy showed a diagnosis of pulmonary embolism that had not been made clinically. The diagnosis of endocarditis may also be particularly difficult in ICU patients [29]. In one patient, autopsy revealed endocarditis even though the patient had had two negative transesophageal echocardiograph studies and repeated negative blood cultures.

Several previous autopsy studies have reported that infections, particularly fungal infections, are the most common discrepant finding [1012, 25]. The diagnostic methods available for fungal infections do not reliably differentiate colonization from systemic infection, and blood cultures (indicating invasive infection) are negative in 50% of patients with invasive candidosis [30]; our results showed that PME identified nine unexpected fungal infections. Silfvast et al. [25] recently reported that five of the eight major (Class I) discrepancies in their analysis of 346 autopsy reports on ICU patients were infections which occurred in patients already treated for another infection, highlighting the difficulties in diagnosing infectious complications.

We identified an important relationship between the length of ICU stay and the type of discrepancy detected. In patients with an ICU stay of fewer than 10 days, the detected discrepancies were mainly major (Class I/II) and included fungal infections, pulmonary embolism, mesenteric infarction and endocarditis, while in patients who died following a prolonged (>10 days) ICU stay, unexpected findings usually related to coexisting diseases (mainly non-metastatic malignancies) that did not directly affect the cause of death. Overall, patients with a longer length of stay were more likely to have a discrepancy between the pre and post mortem diagnoses. Mort and Yeston [11] reported that patients who stayed longer than 48 h were more likely to have a major discrepancy than those who died within 48 h of ICU admission. Other studies have reported no relation between the length of stay and post mortem findings [1618, 31, 32].

Several studies have reported the relationship between patient age and the number of missed diagnoses, with increased discrepancies among older patients [10, 31, 33]. Our results did not show a statistically significant relationship between the number of discrepancies and patient age, although, perhaps not surprisingly, there was a greater number of minor unexpected findings in older patients (>50 years old).

We acknowledge the limitations of this study. Indeed, all autopsy studies have some limitations in that there is no standard procedure for selection of patients for autopsy or for autopsy technique. Post mortem exams were conducted on fewer than half of the patients who died on the ICU during the 1-year study period, representing another potential limitation, although this rate is considerably higher than in many other recent studies. A further limitation is that the study was conducted in a single ICU and our findings may, therefore, not be applicable to other units.

In conclusion, following a short ICU stay, PME can detect findings whose diagnosis is difficult even though they may be suspected by the intensivist. Following a longer ICU stay, PME is more likely to detect coexisting disease processes unrelated to death. The PME should not be seen as a means of providing evidence of clinical malpractice, but rather as a positive educational tool to improve patient care.