This in-depth evaluation of sepsis and septic shock in France from 2010 to 2015 using information from the French national hospital administrative database that records all hospital stays showed an increased incidence of sepsis and septic shock and a decrease in the associated risk of death limited to patients with septic shock. These results should be interpreted in light of our subgroup analyses and the current literature and deserve a few comments.
First, we found an increase in the incidence of sepsis and septic shock in the whole cohort and most of our subgroups. Several studies focusing on sepsis (Additional file 1: Table S1) [7, 9, 12,13,14,15,16,17,18,19,20,21] also reported an increased incidence. For instance, Fleischmann-Struzek et al.  reported an incidence of sepsis that increased from 108 per 100,000 population in 2010 to 158 in 2015. As neither the identification codes nor the reimbursement strategies were modified from 2010 to 2015 in France, the increased incidence of sepsis or septic shock could have been related to an actual increase in the disease incidence due to greater risk factors for sepsis such as the ageing of the population, the increasing burden of comorbidities or the increasing rate of resistances to antibiotics. However, we believe that much of the increase could also be secondary to better identification and coding . It is unfortunately impossible to directly determine how many patients were captured due to better identification. The solution might be to work on detailed clinical data extracted from the electronic health record systems of hospitals . However, these data are not yet available for all hospital stays in France. Nevertheless, it is important to underline that the incidence of sepsis or septic shock due to septicemia has remained fairly stable over time, which indirectly supports a global increase in recognition of other sepsis. The recognition of other sepsis than septicemia could be related to the improvement in the diagnosis of sepsis thanks to successive surviving sepsis campaigns [3, 23], to an improved identification of the pathogens involved or to the rise in electronic medical records and hospital coding systems, which has led to the inclusion of more patients with perhaps less severe septic shock [18, 24, 25].
Septic shock, with a death rate of 45%, is associated with a high risk of mortality. We observed a decrease in the adjusted risk of death by − 0.33% (IQR: − 0.4; − 0.27) per year. This decrease was also found in patients with most severe form of sepsis, i.e., those admitted to the ICU or with bacteremia. Those varying results prompt several remarks. First, to date most studies have reported a decrease in the risk of death over time, from 0.6% per year to 17% per year [7, 9, 12,13,14,15,16,17,18,19,20,21]. For instance, Kadri et al.  reported similar results for septic shock patients, with a decrease in the death rate of around 1.22% per year, from 48.3% in 2005 to 39.3% in 2014. Similar results were also observed among patients with sepsis . Our results, however, were the first to be obtained after adjustment, and compared to those from other countries the decrease although quite smaller was probably more reflective of reality. Our findings could be explained by an improvement in care over time thanks to earlier detection and identification of pathogens, and earlier specific treatment included in care bundles [25, 27]. In contrast, mortality increased in patients not admitted to the ICU regardless of their disease severity. Such results could be explained by pre-mortem misdiagnosis, suboptimal care or early limitation of care. The implicit definition of sepsis using ICD-10 codes included heterogeneous septic patients, almost half of whom were not admitted to the ICU and this could explain why the risk of death was unchanged when all septic patients were considered together.
Finally, we calculated an average cost per hospital stay for septic shock in France of €16,449 compared to the sum of €17,261 in 2010 and the current cost of €16,365. This decrease in costs could have been due to a shorter hospital LOS, mainly because of a shorter ICU LOS, which in turn could be related to the inclusion of older patients with more comorbidities and earlier limitation of care or to earlier discharge of patients to rehabilitation. Another contributing factor could have been an overall improvement in care, with reduction in the time spent on organ support such as ventilation and subsequent reduction in sedation. It is unfortunately difficult to interpret this reduction because of all the confounding factors to be taken into account.
Our work has several strengths. Use of the PMSI allowed all patients with sepsis and septic shock admitted to French hospitals to be included in the study, which with one of the highest number of sepsis and septic shock cases ever recorded gives an accurate picture of the burden they represent for a country with a high standard of care, high hospital bed availability, an integrated healthcare organization structure, and similar admission policies for all patients thanks to its public insurance system. Analysis of a period of time during which diagnostic codes and insurance policies were unchanged ensured that the rise of coding was kept to a minimum. In addition, all procedure codes, diagnostic codes, and the SAPS II scores in the subgroup of ICU patients were taken into account to provide a comprehensive description of septic shock that includes adjustment for mortality risk to a degree not previously attained. We also analyzed hospital stays from admission to death or discharge, taking into account all potential transfers, and adjusting for readmissions, which has not been performed before in similar studies.
Our study has several limitations, mostly related to the nature of the database: the specificity of the coding, the rise in coding due to better recognition and coding, missing data due to billing reasons, and the lack of detailed clinical, paraclinical and drug exposure data. Thus, it was not possible to compare our inclusion criteria with those based on clinical data using the sepsis 3.0 definitions as done elsewhere [28, 29] to determine whether our increase was related to coding and to assess the number of misdiagnosed patients. Nor was it possible to differentiate primary infections from nosocomial infections, to know whether limitations of care were achieved or to take into account discharges to a hospice. Finally, our study did not allow causal inferences between sepsis and/or septic shock and death.