High incidence of immune‐mediated inflammatory diseases in sepsis survivors: A nationwide exposed‐nonexposed epidemiological study

Sepsis is characterized by an excessive release of inflammatory cytokines. Cytokine dysregulation is pivotal to the pathophysiology of immune‐mediated inflammatory diseases (IMIDs). We aimed to analyze the incidence of IMIDs in patients who survived sepsis.


Introduction
Sepsis affects 50 million individuals worldwide and is associated with more than 10 million deaths annually [1].Most sepsis survivors experience Several elements support a link between severe infections and immune-mediated inflammatory diseases (IMIDs).First, overproduction of proinflammatory cytokines is central to the pathogenesis of both IMIDs and severe infections [3][4][5][6][7][8][9].Second, neutralizing anti-cytokine autoantibodies, reported in the setting of both IMIDs (such as systemic lupus erythematosus [SLE]) and severe infections (such as severe COVID- 19), play a Janus effect on IMIDs activity and infection risk [10][11][12].Third, patients with sepsis exhibit a high prevalence of autoantibodies against self-antigens that, which, because of their role in autoimmune diseases, may pave the way for the development of IMIDs in sepsis survivors [13][14][15].Fourth, the association between some specific pathogens and some IMIDs has been well demonstrated, as recently with Epstein-Barr virus and multiple sclerosis (MS) [16].
Previous studies on the temporal immune dynamics in sepsis have shown that immune dysregulation associated with severe infection diminishes over time with recovery but persists after discharge in two thirds of surviving patients and is associated with worse long-term outcomes [17,18].Hypothesizing that persistent immune dysregulation associated with sepsis may be involved in the development of IMIDs; we sought for this association using a nationwide database to analyse the incidence of IMIDs after sepsis.

Data source
Comprehensive data on all exposed and nonexposed patients admitted to all French hospitals from January 2011 to November 2020 were collected from the national medical-administrative database, the PMSI (Programme de Médicalisation des Systèmes d'Information, Information system medicalization program).The PMSI database provides a summary of diagnoses, procedures, and individual medical conditions at discharge from all French healthcare facilities [19].Each facility produces its own anonymous standardized data, which are then aggregated at the national level.Routinely collected medical data include principal and secondary diagnoses, coded according to the International Classification of Diseases, tenth revision (ICD-10), and medical procedures (e.g., radiological exams, technical care, and surgical procedures) coded according to the Classification Commune des Actes Médicaux (CCAM, French common classification of medical procedures).Administrative data include age, sex, year, length of hospital stay, and hospital site.Inhospital deaths are also reported.Since 2004, the budget of each hospital depends on the medical activity described in this specific program.The social insurance authority carries out regular checks to ensure that the data are correctly attributed.For patients admitted to intensive care units (ICUs), the severity illness at admission was assessed by using the Simplified Acute and Physiology Score II.The reliability and validity of PMSI data have been assessed elsewhere [20][21][22].

Definitions and study population
All diagnosis codes are listed in the electronic supplementary material (ESM).Sepsis (exposed) was defined by the combination of (i) a diagnosis code for infection (ICD-10 code A00-B99 + others) and (ii) a diagnosis and/or procedure code consistent with organ failure, as previously reported [21][22][23].The comparator condition for sepsis must (i) be common, (ii) be severe enough to require admission and prolonged follow-up after discharge, (iii) be captured by a clear and specific ICD-10 code, and (iv) not require prescription of immune system-modifying treatment.Patients who experienced an acute myocardial infarction (AMI) meet all of these criteria and were considered the unexposed control group.AMI was defined by ICD-10 code I21.IMIDs (outcomes) included immune thrombocytopenia (ITP), autoimmune hemolytic anemia (AHAI), SLE, systemic sclerosis (SSs), Sjögren's syndrome (SS), antineutrophil cytoplasmic antibody (ANCA)-associated vasculitis, giant cell arteritis (GCA), polymyalgia rheumatica (PMR), rheumatoid arthritis, spondyloarthritis (SpA), MS, inflammatory bowel disease (IBD), and Behçet's disease.Only patients with (i) a first diagnosis of sepsis (exposed) or AMI (nonexposed) in 2020 and (ii) no history of IMIDs reported in PMSI between January 1, 2010 and the index stay were included.Demographics, underlying comorbidities, medical history (including diagnoses and procedures), and the updated Charlson comorbidity index (CCI) [24,25] associated with the index stay were assessed.

Matching procedure
Exact 1/1 random matching without replacement of exposed (i.e., sepsis) and unexposed (i.e., AMI) patients was performed on the basis of on age ±2 years, sex, active cancer, active malignant hemopathy, HIV infection, and organ transplant history (Table S1).Matching variables were putative confounding factors chosen a priori in all cases.Matching accuracy and efficacy were assessed by calculating the standardized differences for the matching variables between the two populations.

Nine-month IMIDs-free survival analysis
The first day of index stay for sepsis (exposed) or AMI (unexposed) in 2020 was defined as Day 0 (D 0 ).Censoring was performed at 9-month follow up, at last hospital stay or at death, whichever occurred first.The incidence of IMIDs was first assessed in the exposed and unexposed populations and compared with the incidence of IMIDs reported in the general population.Hazard ratios (HRs) of IMIDsfree survival between exposed (sepsis) and matched nonexposed (AMI) patients were then estimated.As the proportionality of the hazard-assessed graphically (Fig. S1)-did not hold for the main outcome (IMDs), survival analyses were run starting (i) from D 0 (whole period analysis) or (ii) from D 16 after admission, to the end of follow-up.Further sensitivity analyses were performed including (i) a 6-month censoring (to account for follow-up gaps between the study groups) (Table S2), (ii) the exclusion of patients with IMIDs diagnosed during the index stay (to account for death gaps between the study groups) (Table S3), and (iii) the exclusion of patients with no evidence of prior hospital stay (to optimize the exclusion of patients with unreported IMIDs prior to the index stay) (Table S4).As requested during the reviewing process, we did a post hoc analysis using competing risk models for clustered data taking into account death during the follow up as a competing event (more details in ESM, Table S5) [26].We were also asked to run additional sensitivity analyses by doing two others matching using propensity scores: One including comorbidities associated with IL-1 production and one including the CCI.Details about the matching procedures are reported in ESM.

Statistical analyses
Categorical variables are presented as counts (percentages).Quantitative variables are presented as mean (standard deviation) or median (first quartile-third quartile) for time variables.Differences between matched groups were assessed using standardized mean differences.The number of patient-years-defined as the sum of the duration of follow-up (in years) for each patientand the incidence rate-defined as the number of events divided by the number of patient-years-were calculated for each group.Wald 95% confidence intervals (95% CIs) were calculated for the proportions.HRs and their respective 95% CIs were estimated using a marginal Cox proportional hazards model [27], according to populations matching.To account for remaining differences between groups despite matching, the estimated HRs were adjusted for ICU admission during the index stay.The Kaplan-Meier method was used to present 9month IMID-free survival.All analyses were performed with SAS V.9.4 software.Kaplan-Meier curves were generated using R V.4.2.0 software with survival package.

Ethical considerations
In accordance with the French regulatory system for personal and medical data and with the approval of the French data protection authority (Commission nationale de l'informatique et des libertés, CNIL), our institution was granted access to the PMSI database according to MR-005.

Characteristics of patients experiencing sepsis
From January to November 2020, a diagnosis of sepsis and AMI was reported in 460,707 and 62,258 patients, respectively.According to the inclusion criteria, no patient had prior evidence of IMIDs.The overall characteristics of patients with sepsis and AMI were different, highlighting the need for a matching process for further comparison (Table S1).The matching process between sepsis (i.e., exposed) and AMI (i.e., unexposed) patients was successful in all but one case (n=62,257 sepsis patients matched with n=62,257 AMI patients among 62,258 patients with AMI in 2020).Notably, inhospital mortality was higher in sepsis patients at both the index hospitalization (12.1% vs. 4.5% in nonexposed patients) and 9-month followup (16.9% vs. 6.0%).Rates of 9-month censoring differed between exposed (n = 49,405; 79.4%) and unexposed (n = 57,061; 91.6%) patients.Remaining differences between exposed and unexposed matched populations are shown in Table 1.

Nine-month IMIDs-free survival analysis
Nine-month cumulative incidence curves comparing IMIDs onset between exposed and nonexposed patients showed that the proportional hazards assumption did not hold for the main analysis (Fig. 2, Fig. S1).Interestingly, the survival analysis showed, after adjustment for ICU admission and using IMIDs onset after AMI as reference, an increased risk for IMIDs after sepsis of 2.80 (HR; 95% CI [2.22-3.54])starting from day 16 after admission.The adjusted HRs for the different IMIDs onset following sepsis are shown in Fig. 3. Sensitivity analyses performed after (i) a 6-month censoring, (ii) exclusion of patients with IMIDs diagnosed during the index stay, and (iii) exclusion of patients without evidence of prior hospitalization were consistent with the main analysis (Tables S2-S4).The post hoc analysis using competing risk models for clustered data taking death during follow up as a competing event showed a sub-distribution hazard ratio of 2.17 [1.43-3.27]for the period starting after day 16 for the risk of IMIDs onset after sepsis (Table S5).The propensity scores matching analyses including comorbidities in the calculation of the propensity scores produced also similar results to the main analysis (Tables S6 and S7).
Of note, the risk of IMID onset was different according to the type of the autoimmune disease.Indeed, in sepsis survivors, it was higher for ITP (5. ), using AMI patients as reference (Fig. 3).
Notably we found that the association between sepsis and IMIDs onset appeared well balanced across pathogen categories (Fig. 4).

Discussion
All patients with sepsis admitted to French hospiover a 1-year period were enrolled in a study cohort to capture the incidence of IMIDs following sepsis.Our primary analysis of 60,167 patients with sepsis and 60,167 matched controls showed a dramatically high incidence rate of IMIDs after severe infection.To our knowledge, no large-scale comparative study of IMIDs after sepsis has been published so far.
The onset of IMIDs onset following sepsis was delayed over time and became apparent only after day 16 after admission.Such a time lag may reflect a transient immunosuppressive phase at the onset of sepsis, as previously reported [41].Although bias in the survival analysis cannot be completely ruled out, sensitivity analyses confirmed the strong association between sepsis and IMIDs onset.
Even though our study is the first to examine the association between sepsis and incident IMIDs at a nationwide level, previous reports have already described an increased risk of IMID incidence after infection.Nielsen et al. [42] found an association between hospital admission for an infection and 29 autoimmune diseases in the Danish register data.These associations were time and « dose » dependent with adjusted incidence rate ratios ranging from 1.24 to 2.58.The spectrum of the IMIDs that are more frequent after infection is so broad that the authors argue that infections should be considered an environmental risk factor for IMIDs onset.Among patients already diagnosed with IMIDs, the impact of infections has also been studied during epidemiological studies.Buljevac et al. [43] observed that patients with MS had a risk ratio of 2.1 (95% CI 1.4-3.0) of experiencing flares-up of the disease during a period of 2 weeks before and up to 5 weeks after the onset of a clinical infection, compared to the other periods.
The choice of the control group was key in ensuring the reliability of our study.We figured out that AMI was a good option.Although up to 30% of the patients admitted for an AMI are readmitted in the 30 days following initial discharge [44] medical monitoring and follow-up might also have been different between the two study groups.Our sensitivity analysis reducing the follow-up period to 6 months was consistent with the main analysis and thus limits the risk of a surveillance bias.On the other hand, the majority of patients in both groups were censored at 5 months follow-up, meaning that for most of them, information on the occurrence Fig. 3 Risk for immune-mediated inflammatory diseases (IMIDs) onset after sepsis.Hazard ratios (HRs) are given for the risk of IMIDs after sepsis compared to acute myocardial infarction (AMI).HRs are calculated using a marginal Cox model, on intensive care units (ICU) stay during the first index stay.A log10-scale is used for the X axis.HRs are reported using the risk in nonexposed patients (AMI) as reference.Dark red is used for the risk during the whole period, whereas light red is used for the period after day 16.AIHA, autoimmune hemolytic anemia; AAV, ANCA-associated vasculitis; ANCA, anti-neutrophil cytoplasmic antibody; GCA, giant cell arteritis; IBD, inflammatory bowel disease; IMIDs, immune-mediated inflammatory diseases; ITP, immune thrombocytopenia; MS, multiple sclerosis; PMR, polymyalgia rheumatic; RA, rheumatoid arthritis; SLE, systemic lupus erythematosus; SS, Sjögren's syndrome; SSs, systemic sclerosis; UC, ulcerative colitis.Fig. 4 Immune-mediated inflammatory diseases (IMIDs)-free survival analysis according to the pathogen identified during sepsis.Hazard ratios (HRs) associated with bacteria (n = 1382 patients-years), virus (n = 1065 patients-years, including SARS-CoV-2 in most (88%) cases), other pathogens than virus or bacteria (n = 2816 patients-years) and no identified pathogen (n = 3585 patients-years) have been calculated considering the overall follow-up time.HRs are reported using the risk in nonexposed patients acute myocardial infarction (AMI) as reference.
of IMIDs, whether positive or negative, were not available.However, and regardless of the AMI or sepsis group, IMIDs like ANCA-associated vasculitis or AHAI-the most frequently IMIDs diagnosed after a sepsis-are severe condition that led directly to hospitalization in most cases.If some incident IMIDs have been missed in the follow-up, there is no reason for this hypothetical bias to be more pronounced in one group.
The validity of the IMIDs diagnoses need to be questioned.The coding list for the sepsis and the comorbidities is extensive but, PMSI data being strictly anonymous, we have no access to medical charts to gain granularity.This may be particularly true when there is no specific marker for the diagnosis of IMIDs such as PMR.However, we tested the accuracy of the PMR coding diagnosis (M353) by reviewing the medical records of 227 consecutive patients admitted to our institution between 2017 and 2021 and confirmed the accuracy of the coding procedure in almost all cases (n = 212, 93.4%).
The majority of IMIDs occurred in the short-term postinfection period.The strength and temporal pattern of the association between sepsis and increased risk of IMIDs suggest a causal relationship.Some IMIDs have been associated with specific pathogens, such as SLE and MS with Epstein-Barr virus, or polyarteritis nodosa with hepatitis B [16,[45][46][47].The association between sepsis and IMIDs onset appeared well balanced across pathogen categories and viruses, bacteria, and parasites all induce multiple inflammatory cascades leading the activation of innate and adaptive immune cells [48].Pathogens can induce autoimmune disease through mechanisms involved in the breakdown of self-tolerance, such as bystander activation, pathogen-induced necroptosis, epitope spreading, superantigen cross-linking, and molecular mimicry [48,49].Alternatively, patients who suffered from severe infection and IMIDs may share a common genetic susceptibility background.Our main analysis is based on admissions for sepsis in 2020 and thus used data collected during the first wave of COVID-19.Although this may have resulted in different follow-up of patients from usual, particularly those with long COVID, and less than usual follow-up for patients admitted with AMIs, our sensitivity analyses performed in comparison to patients with a sepsis stay in 2019 (i.e., prior the COVID-19 pandemics) remain unchanged.
IMIDs differ in their clinical phenotype, tissue distribution, and response to treatment [50].Interestingly, the risk for IMIDs seems to be higher for ANCA-associated vasculitis, ITP, and AHAI.Although thrombocytopenia is common in patients with sepsis [51], the identification of antiplatelet autoantibodies already reported in this setting supports an immune-related process [52].ANCAs are an important biomarker for ANCAassociated vasculitis [53].ANCA targets different components of human neutrophils.Pathogenic ANCA may be secondary to the release of neutrophilic enzymes triggered by infection.Production of other autoantibodies, such as rheumatoid factor, antinuclear, and antiphospholipid antibodies, has also been described after severe infection [54][55][56].
This study has several limitations.First, although the PMSI database links all hospital stays at the individual level, only inhospital diagnoses and procedures are captured by the database.Second, IMIDs may have been diagnosed in outpatients before the index stay and a classification bias cannot be excluded.However, our sensitivity analysis focusing on patients admitted before the index stay confirmed the main analysis.Third, we did not have access to medical records and cannot rule out misclassification bias.Fourth, we did not have access to immune-modifying treatments such as steroids that patients may have received during sepsis care.Fifth, our cohort consists only of admitted patients and an information bias cannot be excluded.This may indeed contribute to the higher incidence rate of IMIDs observed in our study as compared with the general population.The studies used as reference from the general population had different methods and populations: not only inhospital like ours and coming, for some of them, from countries other than France.Finally, the exact delay between the diagnosis of sepsis and the diagnosis of IMIDs diagnosis was difficult to determine using the PMSI database.
In conclusion, our study shows an intriguing and extremely high incidence of IMIDs among survivors of severe infections.Further studies are needed to investigate the relationship between severe infections and IMIDs.Coupling sound epidemiologic evidence with a comprehensive immunemonitoring approach should help to characterize the patterns of immune activation at sepsis onset that may pave the way to autoimmunity.

Table 1 .
Characteristics of the matched populations.