Discussion
The size of Oslo University Hospital and its patient base could provide data for calculating the rate of in-hospital mortality caused by healthcare-associated infections in Norway. We found that 11.4 % of all deaths can be related to adverse events, and that healthcare-associated infections accounted for approximately 84 % of these (9.6 % of all deaths). Determining whether a patient dies from or with an infection (or another adverse event), is particularly difficult for multi-morbid patients (14). The prognosis for most patients with a fatal healthcare-associated infection was considerably reduced due to advanced age, malignant disease or organ failure.
Some patients with late-stage disease are hospitalised in order to receive terminal care. This is particularly the case at Ullevål. Limited resources are put into pneumonia prevention for terminal patients and antibiotics are not always administered. It would be somewhat inappropriate to classify respiratory tract problems, clinically referred to as pneumonia, as a fatal healthcare-associated infection in these patients. However, we chose to include clinical pneumonia without radiologically proven infiltrate, despite the fact that this does not meet the criteria set by the Centers for Disease Control and Prevention. We would otherwise have ended up with lower pneumonia figures for the group of patients whose prognosis was less than four weeks, but only for this particular group. It is important to consider findings in relation to prognosis, as discussed by Flaatten et al. (12).
We have based our prognoses on the McCabe classification which involves an assessment based on age, diagnosis, stage, type of malignancy and severity of organ failure. This can be advantageous when judging prognoses, because the relevant information is most often found in the patient records. The Charlson system uses a numbered risk score for certain diagnoses and conditions (15). Despite a certain popularity, medical developments may suggest that the Charlson system is no longer a suitable tool for classifying prognoses, particularly in cases of lymphoma, leukaemia and HIV infection (16, 17).
Data published by the Norwegian Research Centre for Health Services and based on the Global Trigger Tool suggested that approximately 4723 patients died as a consequence of in-hospital trauma in 2010. This accounted for 32.7 % of all in-hospital deaths in Norway (2). The figures attracted significant public and political attention. They differ markedly from the international experience (4), and from our own findings for the same period. In our opinion, surrogate estimates based on the Global Trigger Tool method should not be used. In a case review study of all 372 patients above the age of 18 who died at Nordland Hospital in 2013, the researchers found, using a modified Global Trigger Tool method, a somewhat larger proportion of adverse events with a fatal outcome than we did (18). Approximately half were respiratory tract infections. Two other Norwegian studies, with considerably higher numbers of patients included, coincide with our findings (12, 19). These studies also involved a review of the medical records of patients who died in hospital, but they did not present figures specifically for infections.
Our choice of methodology is supported by a Dutch study which included 3 983 deaths at 21 hospitals of varying size and category (4). They found that 4.1 % of fatalities were caused by preventable adverse events, but did not specify infection rates. It is our opinion that a study of fatal adverse events, particularly infections, should be conducted by instigating a full case review of patient records, including access to all relevant clinical data as well as microbiology, histopathology, radiology and clinical chemistry findings, and in some cases autopsy records. This would be a time-consuming exercise, but has been achieved in some large-scale studies (6–9). An American study found that healthcare-associated infections accounted for one-third of unexpected deaths, but the threshold for 'unexpected' was set to six months (9). If we had set a six-month threshold, the proportion of infections in our material would have been higher (13/24).
Studying your own institution comes with a risk of bias. The figures from Oslo University Hospital are on a par with, or somewhat higher than, those of comparable studies (4, 12, 19, 20). Our infection data are based on accepted criteria (11). This carries less risk of bias than the recording of adverse events generally (5, 8, 21, 22). Our data reflect underlying diseases as well as treatment methods. Modern treatments include methods that expose patients to infection, such as immunosuppression, cytostatic drugs, organ transplants, extensive cancer surgery and demanding traumatology. We have presented some of the data from the Ullevål and Rikshospitalet hospitals separately. The differences primarily reflect these hospitals' dissimilarities in terms of patient base. Among patients with pneumonia we found that the majority had a prognosis of less than four weeks (62 %). Among patients with bloodstream infections, there was a wider range of estimated prognoses. The fact that there was a greater prevalence of bloodstream infections at Rikshospitalet/the Norwegian Radium Hospital, can be explained by differences in the patient base. Both pneumonia and sepsis prophylaxis are difficult conditions in high-risk patients (14). The range of microbiological findings in bloodstream infection cases may point to the endogenous flora being an extra challenge in cases of prophylaxis. Weakened barriers and bacteraemia secondary to processes in another body site are common. The patient records did not include sufficient information for us to assess the connection with central venous or peripheral catheters.
Our figures and assessments of comorbidity coincide with the data presented by others and show the justification for grouping patients according to their predicted prognosis (14). Researchers in Canada also accentuate pathological stage when assessing in-hospital morbidity (23). In our data, patients with a malignant disease make up a large group. As pointed out by Weinstein: 'Nosocomial infections typically affect patients who are immunocompromised because of age, underlying diseases, or medical or surgical treatments' (24).
We have not attempted to estimate how many infections could have been prevented. Advancing age and serious conditions, such as organ failure and malignant disease, as in our patients, make this a difficult exercise (25). As evidenced in the Netherlands, the potential for prevention is greatest if the prognosis is more than six months (4). Few of our patients were in this category. It is also possible to overestimate the potential for prevention (7).
Care must be taken when estimating how many deaths are preventable, as this is difficult to establish (26–28). The complexity of modern hospital treatment is a point to be considered, and considerable variation has been found in the assessments of different researchers (26, 27). A recent Canadian publication maintains that genuinely preventable deaths are rare (29).