Since the early 1990s, considerable efforts have been provided to improve treatment and global management of patients suffering from severe infection. In August 1991, the American College of Chest Physicians and the Society of Critical Care Medicine released a landmark Consensus Conference agreeing on definitions to be applied to patients with severe infections 1. Classifications were provided to assist clinicians and researchers including homogeneous and comparable populations in trials dealing with severe sepsis and septic shock. Following this conference, a number of major studies have been published that substantially modified the management and course of severe sepsis and septic shock patients. For the same period, the mortality of patients with severe infections did not decrease 2 or remained high, suggesting that physicians might be unaware of the scientific advances in the field of severe infection.

Severe sepsis is a major problem in intensive care, accounting for over 10 to 20% of stays of increased duration 3 and with a recent increase in hospitalization 4. The most severe presentation is septic shock with an individual mortality that has recently slightly decreased 5 but with an overall specific mortality that has increased 6 due to the higher number of patients, in particular older patients 7.

This observation prompted scientific societies to develop a partnership that intended to provide standards of care for management of severe sepsis and septic shock, based on the highest levels of evidence published scientific knowledge. Key recommendations overviewed management of septic patients, including specific treatments such as antimicrobial therapy, hemodynamic management and use of adjunctive therapies, and supportive care such as sedation, tight glucose control and mechanical ventilation 8.

Implementation of these guidelines, however, has been offset by a number of barriers. Heterogeneous levels of equipment and resuscitation skills and overburden in ICUs compromise the quality of care delivered to septic patients, especially those requiring a high level of technical support 910. Besides these organizational limitations, the way these guidelines are applicable or not to routine care patients is also controversial. One can suppose that differences between groups of homogeneous patients included in pivotal trials and real-life patients routinely admitted to the ICU may prevent guideline generalization. Even if non-inclusion criteria are known to explain these differences, their identification and frequency among routine ICU patients is lacking. As a consequence, it is actually difficult to assess the proportion of patients in whom the most current recommended treatments were in fact not tested. To specifically address whether international recommendations to treat septic patients fit routine ICU patients or not, we first checked for non-inclusion criteria in published randomized controlled trials (RCTs) dealing with severe sepsis and septic shock since 1992. In a second step, we investigated the frequency of these non-inclusion criteria in a large ICU population by using a multicenter ICU registry.

Exclusion criteria in RCTs dealing with severe sepsis and septic shock

A search using 'severe sepsis', 'septic shock' and 'treatment' as keywords in the PubMed database identified 7,012 publications from 1 January 1992 to 31 December 2008. Among these publications, 734 were 'randomized' trials. We selected 96 interventional trials including patients with severe sepsis and septic shock after careful assessment of each publication (Figure 1), corresponding to 26,875 included patients (see Additional file 1 for studies' characteristics and Additional file 2 for a complete publications' list). The main intervention referred to vasopressors in 38 trials; fluid loading/global hemodynamic-based strategy in eight trials; steroids in eight trials; modulation of immunity in 29 trials; coagulation-targeted therapy in 11 trials; and other adjunctive therapy in four trials. No study referred to antimicrobial agents. Results did not support the primary hypothesis in 63 studies. The number of patients screened and patients excluded were reported in only 23 trials.

Among the 96 selected studies, 24 (corresponding to 2,163 patients) did not mention any exclusion criteria. These trials were monocenter in 20 studies, and nine studies were controlled and eight studies were double blind. Results were positive for the primary endpoint in 33 studies. In these studies, the number of patients screened and excluded was not reported. Mortality was the primary endpoint in four studies.

The non-inclusion criteria used in the remaining 72 studies are reported in Table 1. Pregnant women were excluded from 46 studies, patients aged <18 years from 39 studies, and patients aged >75 years from four studies. The main underlying co-morbidities that led to patients' exclusion were congestive heart failure (34 studies) and cancer (30 studies). Gastrointestinal and liver disorders were also frequent non-inclusion criteria. In 33 studies, patients could not participate if they suffered from at least one cause of severe immune suppression (combining AIDS, neutropenia and transplant). Patients with coagulation disorders or treatment interfering with hemostasis could not be enrolled in 12 trials. Other miscellaneous non-inclusion criteria appeared in 60 trials.

In this article we demonstrate that RCTs dedicated to treatment of severe sepsis and septic shock do not fit a large part of the patients that are currently admitted to ICUs with severe infection. Patients with common co-morbidities, usual medications, cancer and various immune deficiencies were not included in clinical trials dealing with severe sepsis and septic shock although such patients are likely to develop severe infections 13. Whether the results of these trials are applicable to these frequently encountered ICU patients is questionable.

RCTs are believed to currently provide the best level of evidence that ensures efficacy of drugs interventions or strategies. Regarding severe infections, their results represent the basement of evidence-based clinical knowledge and practices as recommendations of the Surviving Sepsis Campaign 8. Feasibility of such trials requires homogeneous populations with selective characteristics to allow group comparisons and internal validation. Characteristics of the population are therefore tightly selected but lead to exclusion of potential participants. For instance, it has been reported that 23% trials included in meta-analyses excluded no patients 14, and the results were more significant in studies that excluded patients. In the present study, the majority (75%) of RCTs devoted to severe sepsis and septic shock had non-inclusion criteria. Selecting patients is also believed to decrease uncontrolled and unknown adverse events in patients with severe underlying disorders. As a consequence, more fragile patients cannot enter most RCTs. Patients enrolled in clinical trials therefore sometimes differ from the actual target population. This may result in a paradox where methodology oversights clinical relevance. Research conducted by more than 300 analysts observed that older people were often excluded from studies focused on Alzheimer's dementia, arthritis and incontinence 15. RCTs may therefore lack external validity. Consequently, efficacy and adverse outcome can be ignored when translating scientific evidence into daily practice.

Clinical practice addresses complex patients suffering from multiple health problems that require multiple medications. In a primary care population, patients eligible for five RCTs dealing with hypertension had 5 to 11 chronic conditions 16. Ideally, RCTs would include more complex and severe patients to allow translation of results in daily practice. A recent review examined the exclusion criteria in 283 RCTs published in high-quality medical journals 17. The subject of interest in these articles was infectious diseases in 55 (19%) studies and critical care in eight (2.8%). Exclusion related to medication and co-morbidities were frequent and poorly justified (for details, see Additional file 4). As an example, cardiac disorders are excluded from 24% of RCTs published in the selected set of journals 17. In western countries, congestive heart failure is the most common underlying disorder and the leading cause of death 12. Additionally congestive heart failure is a major co-morbidity associated with death in septic patients. In our study, patients with congestive heart failure were excluded in 34 (35%) articles whereas the disorder was encountered in 11% patients with severe infections from our database (Additional files 5 and 6).

Obviously disorders or treatments that deregulate innate and adaptative immunity increase the risk for severe infections. Neutropenia, steroids, cancer and especially hematological malignancies are basically responsible for a burden of infection. In a report of 283 trials including 35 RCTs for treatment of oncologic and hematological malignancies, cancer patients were excluded in 46 (16%) studies and those with blood disorders in 59 (21%) 17. Here we show that these were frequent exclusion criteria for patients with severe infection, whereas they were frequently encountered at the bedside. Besides these exclusion criteria that can be forecast, other frequent conditions might be underperceived as exclusion criteria whereas they frequently impair the eligibility of septic patients. This particularly applies to gastrointestinal and liver disorders.

Pharmacological properties of medications are modified in physiological changes. Anticipating these variations is sometimes challenging. Obesity, an increasing condition in western countries, is illustrative for this paradigm. In obese patients, pharmacokinetics of drugs may change as the volume distribution differs from nonobese patients. We observed that overweight was a criterion for exclusion in 7% trials while 18 to 32% of American citizens present a body mass index >30 18. We observed in our population that obesity was a condition that increased over time whereas it did not reach the incidence recorded in North America. This may be related to a different distribution of this condition in our area 19 or to recent perception of the problem or inadequate evaluation and underestimation of overweight. As obese patients may be at risk to develop severe infections 20, and because their prognosis is worsened 19, people with overweight should enter RCTs dealing with severe sepsis and septic shock.

We observed that a minority of trials dealing with severe infections excluded older patients. The burden of infection encountered in western countries is related to ageing. Additionally, age and mortality are closely related in infectious diseases 2122. A number of reports recently warned physicians on the exclusion of older patients from RCTs 2324. More than 38% of RCTs excluded patients over the age of 65 17. In trials devoted to treatment of cancer, only 25% of patients were older than 65 25. In our study, older patients were excluded from only 4% of studies but the most common cutoff was 75 years, which can be responsible for an underestimating of age impact on non-inclusion.

The weight of co-morbidities on patients' prognosis is a usual concept in several fields of medicine. Their impact has been extensively described in infectious disorders 132627. In this setting, underlying disorders are likely to impair prognosis by decreasing host defense against the microorganism 28. Alternatively, patients are unable to face increasing oxygen demand because of impaired physiological adaptation. In brief, co-morbidities expose the patient to more severe infections 28. Whereas excluding patients because of co-morbidities allows recruitment of a homogeneous population in RCTs, it may be detrimental for translation of evidence into daily practice.

A major issue is to determine barriers to participate in a clinical trial. Refusal occurs in one out of five patients approached. Factors that influence the decision to decline the invitation to enter a trial are poorly known 29. The most important reasons for failing entry into a study remain lack of an adequate trial and unmet inclusion criteria. This suggested a lack of pragmatic studies addressing frequent clinical questions in patients with common characteristics 30.

We identified several limitations in this study. First, there might be some differences between non-inclusion criteria extracted from RCTs and patients' characteristics. However, even if they exist, their relevance was limited and should have induced a negligible bias. Perhaps more important, we were unable to identify both coagulation disorders and use of steroids among the database population. Consequently, no firm conclusion can be drawn from our results regarding these two criteria. Also we did not restrict our analysis to positive studies but also to negative studies, since we considered that these negative studies could have influenced the sepsis guidelines in a similar manner. Finally, we tested only one French database population, and a replication in a different ICU population could have been useful. Whereas we provide data from a large multicenter series, patients we selected for severe sepsis and septic shock may differ from other cohorts, even in western countries. Indeed, assessment of these diagnoses may be subjective, and the incidence of underlying disorders may be not representative for other areas.

Here we demonstrate that RCTs dealing with severe sepsis and septic shock infrequently include patients with conditions that usually coexist and predispose to severe infection in routine ICU patients. Moreover, we also show that the frequency of these non-inclusion conditions increased over recent years. Excluding patients from analysis of intervention trials presumably results in biased estimates of treatment effects, because patients excluded may correspond to the actual target population 14. These findings advocate for extension of entry criteria for participants with multiple co-morbidities as they constitute the majority of patients with severe sepsis and septic shock.

• RCTs dealing with severe sepsis and septic shock support current guidelines; however, these studies use stringent exclusion criteria that lead to exclusion of numerous patients.

• In an important cohort of patients with severe sepsis and septic shock, 60% experienced at least one of the main exclusion criteria for RCTs dedicated to this disorder.

• A significant trend for increasing number of patients with these criteria is observed with time.

CuB-Rea: College des Utilisateurs de Bases de données en Réanimation; RCT: randomized controlled trial.

The authors declare that they have no competing interests.

Y-EC conceived the study, proceeded to selection of the studies used for the paper, and participated in data extraction and redaction of the manuscript. PA was responsible for extraction of data from the Cub-Rea database, statistical treatment of the data and redaction of the manuscript. HB proceeded to selection of the studies used for the paper and participated in data extraction and redaction of the manuscript. BG was responsible for the Cub-Rea network and database, and participated in conception of the study and critical review of the manuscript. AC participated in the study conception and design, and in redaction of the manuscript. All authors approved the final manuscript. Physicians from the Cub-Rea group participated in daily implementation of the database.