Most epidemiological data on sepsis refers to the first decade of twenty-first century and almost exclusively refer to the USA. This is a large observational study of patients discharged from all national health system acute-care hospitals conducted in a European setting. We estimated the mean sepsis incidence to be 212.7 cases per 100,000 inhabitants/year and in-hospital mortality to be 21.6%. Incidence and mortality varied over time, with a yearly increase in incidence of 7.3%, a yearly relative reduction of 3.3% in length of stay and a yearly reduction in in-hospital mortality of 3.4%. After adjustments for relevant clinical and epidemiological variables, the reduction in mortality remained statistically significant.
The estimated incidence of sepsis in our study was lower than reported in the USA and slightly higher than reported in smaller European studies [2, 6, 7]. Previous studies conducted in Spain reported incidences of between 110 and 230 cases per 100,000 inhabitants/year [15, 16], versus the 212.7 cases observed in our study. Differences in calculated incidences may be related to structural or functional organization [17] or may be due to discharge diagnosis coding. Nonetheless, we would like to emphasize the importance of using local data to monitor trends in activity and results over time. Moreover, the number of sepsis cases in our study increased yearly, a finding which is consistent with findings reported in other epidemiological studies [10–12].
Estimates of sepsis incidence and trends are also essential to estimate the resources needed to care for these patients. Sepsis incidence is increasing compared to incidence for other leading causes of mortality such as acute myocardial infarction or ischaemic stroke. The CatSalut data on hospital admissions/year for severe sepsis (five-year mean, 16,460 cases) are close to the combined numbers for acute coronary syndrome and ischaemic stroke admissions together, at 11,000 and 8000, respectively [18, 19]. However, incidence rates for acute coronary syndrome and ischaemic stroke, unlike for sepsis, are stable [19–21]. In the USA, the percentage of septic patients with a fatal outcome increased from 14% in 2000 to 16% in 2010; in contrast, mortality for respiratory failure decreased from 25 to 17%, for heart attack from 10 to 8%, for cancer from 8 to 4% and for stroke from 6 to 5% [22].
Prospective versus retrospective analysis observed differences in incidence and source of sepsis. Prospective monitoring is laborious, costly and complex and can also be affected by issues such as inclusion criteria or data sources [23, 24]. Although retrospective analyses from hospital discharges—as in our study—can also be affected by definitions, codes and analytical methods, they serve an important function in analysing local trends and outcomes. Gaieski [24] observed a 3.5-fold difference in estimates of absolute incidence using different database abstraction methods. Nonetheless, trends were similar irrespective of the methodology. Stevenson et al. [12] recently found that severe sepsis mortality was 10% higher for patients included in the control group of clinical trials compared to administrative data (collected according to Angus’ criteria) [5]; nonetheless, mortality trends were similar, irrespective of the data source—and were also similar to the 3% yearly reduction found in our study. Stevenson et al. consequently concluded that administrative data are useful in monitoring mortality trends in patients with severe sepsis.
Length of stay and mortality both decrease yearly during the study period—to a statistically significant degree according to the multivariate analysis adjusted for demographic data, comorbidities, infection source and number of organ failures. The external validity of our findings is supported by the fact that mortality in 2008 was the same as that reports by the PROWESS-SHOCK study placebo group [25]. Other recent large randomized clinical trials (RCTs) have reported mortality rates of 18–30% [26, 27], further confirming the likely validity of 2008 as our baseline year. Kaukonen et al. [11], who recently reported similar results for Australia and New Zealand, observed an annual absolute decrease of 1.3% in risk, from 24% in 2008 to 19% in 2012. In our study, multivariate analysis revealed a robust association between mortality and year of detection, as adjusted for confounding factors including sex, age group, comorbidities, ICU admission, emergency admission, organ dysfunction, number of organ failures, septic focus and the presence of bacteraemia. Although our study was not designed to address this issue, a potential improvement in the management of sepsis could be suggested, explained in part as a consequence of training and increased clinical awareness [3, 28,29,30].
Our results suggest also that sepsis outcomes should be interpreted according to the year of data collection and the presence of comorbidities. Moreover, underpowered RCTs would be avoided if these effects were taken into account in estimating statistical power and sample size. Yearly reduction in crude mortality rates should be expected, bearing in mind that overestimated mortality rates may lead to underpowered studies which might, in turn, lead to potentially useful treatments being downgraded due to lack of evidence. Furthermore, excluding elderly patients and patients with comorbidities from RCTs represents a form of selection bias; Kaukonen et al. [11], for instance, reported a 4.6% mortality rate for comorbidity-free patients and young patients (versus our rate of 21.6%). Another issue is that sepsis outcomes are too often viewed as binary: The patient dies (failure) or survives (success). Studies also tend to focus on in-hospital mortality and length of stay as an outcome measure for ICU patients, overlooking the fact that many patients admitted for sepsis die after discharge. There is an unmet need to improve knowledge regarding long-term effects in patients with sepsis [1–8], so other outcome indicators such as long-term morbidity and quality of life are likely to be included in future trials.
Just under a quarter (24.7%) of our patients presented with bacteraemia, associated with higher mortality. Patients with bacteraemia could represent a suitable population to monitor prospectively in clinical practice, as bacteraemia, unlike sepsis, is easily identified retrospectively, is easily distinguished from other non-infectious diseases that cause organ dysfunction and is also easily stratified using the sepsis-related organ failure assessment (SOFA) instrument [31].
The main strengths of our study are the large cohort of patients included in a European setting, the fact of including 100% of admissions to both public and private hospitals of Catalonia Health System, the long period of data collection and the use of a previously validated strategy. Our study has several limitations. The fact that cases of sepsis were identified indirectly using ICD-9-CM codes implies less accuracy in identifying cases and a poor clinical analysis compared to prospective methods. Urinary infections appear as the main focus of sepsis in our study. The relevance of each focus can be affected by population characteristics or methodology. It also can affect incidence or severity results. Case recruitment may also have been affected by coding, as reflected in our results for cardiovascular dysfunction. Hypotension was poorly documented on discharge. Incidence, as reported in our study, probably does not reflect clinical incidence. Cardiovascular dysfunction incidence rates of 7.2–42% have been reported for epidemiological or retrospective studies, in contrast to rates of up 90% for prospective studies [4, 5, 9, 13, 15, 32, 33]. Our study cannot account for reasons for reduced mortality and shorter stays. Inclusion of a specific severity scores, such as SOFA, in the multivariate analysis would allow insights into whether mortality reduction is related to the inclusion of less severe patients. Unfortunately, our study design does not admit this conclusion. Given that the CMBD-HA does not specifically collect data about ICU admission, the category ‘ICU stay’ was deduced from procedures typically used in ICUs. We think that since this definition is highly specific but not sensitive, we cannot rule out the possibility that some septic patients with less severity were excluded from ICU admission.