The study was conducted in two acute National Health Servie (NHS) Trust Teaching hospitals in England. The protocol was considered by the Local Research and Ethics Committee, and the need for informed consent was waived in view of the observational and anonymous nature of the study.

From 1 November 2004 to 31 March 2005, four authors (TM, DFD, SG and SF) ran daily screening on new admissions (aged 18 or over) into medical and surgical wards and accident and emergency areas for patients with severe sepsis or septic shock as defined by the International Sepsis Definitions Conference 9. We then followed them up and used proximate look-back data extraction to record the time '0' when signs and symptoms of infection, documented source of infection and ≥1 organ disfunction had all been fulfilled.

All the patients were eligible for 6-hour basic ward care: such as oxygen, iv access and Modified Early Warning Scores (MEWS) 10 as well as the 6-hour sepsis bundle.

The elements of basic ward care and the 6-hour sepsis bundle adapted from the SSC standard sepsis resuscitation bundle are listed in Additional file 1. Our bundle differed from the sepsis resuscitation bundle as we used a haemoglobin target of 7 to 9 g/dl instead of haematocrit ≥ 30%, and used remaining hypotension after fluid resuscitation for threshold of inotropes instead of central venous oxygen saturation (ScVO2). A 'yes' score was obtained if the element had been executed, as documented on charts or notes, within the first six hours after time '0' (diagnosis of severe sepsis); a 'no' score was obtained otherwise.

If the process of severe sepsis was progressing and organ function support was required (for example vasopressors, mechanical ventilation), patients were reassessed for the appropriateness of critical care admission and of the 24-hour bundle; 69/71 patients were eligible to receive the 24-hour bundle as part of critical care management.

The elements of the 24-hour sepsis bundle adapted from the SSC sepsis management bundle is listed in Additional file 1. Again, a 'yes' score was obtained if the element had been executed, as documented on charts or notes, within the first 24 hours after time '0' (diagnosis of severe sepsis); a 'no' or 'not applicable' score was obtained otherwise.

We assessed compliance using 'all or none' as a pass-fail basis for the whole bundle of elements. We used hospital death rate as the outcome measure.

We applied Chi squared test, relative risks (RR) and their 95% confidence intervals (95%CI), and the number needed to treat (NNT), as appropriate, to compare hospital mortality between compliant and non-compliant groups.

We identified 101 consecutive patients who met inclusion criteria for severe sepsis or septic shock on the wards (n = 90), or in accident and emergency areas (n = 11). Of the 101, 71 (70%) were admitted into critical care units (high dependency unit, n = 20; ICU, n = 51) with a mortality rate of 39.4% (n = 28). The in-hospital mortality rate was 35.6% (n = 36). Figure 1 shows the patients' flow chart. General information about the patients is given in Table 1. The median age of the patients was 69 years (interquartile range 51 to 78), 53% were male and 56 (55%) were medical and 45 (45%) were surgical. The major sources of infection were chest (50%) and intra-abdomen (22%).

Within the first 6 hours following the diagnosis of severe sepsis, when patients had already developed one organ failure, we found that of the 101 patients, 8% had no oxygen administered, 14% had no iv access established, and 14% had no essential monitoring, including blood pressure, heart rate, respiratory rate, oxygen saturation, temperature, urine output and level of consciousness described as MEWS. One-third of the patients received no outreach service within the first 24 hours following the diagnosis of severe sepsis.

Of the 101 patients, within the first 6 hours after diagnosis of severe sepsis: 74% had a presumptive diagnosis made, including blood culture; 74% had antibiotics administered; 52% had serum lactate measured; in the event of hypotension, 84% had immediate 0.5 litre fluid administered; and in 70%, when MAP < 65 mmHg despite fluid resuscitation, a vasopressor was used and/or blood transfusion given to a haemoglobin target of 7 to 9 g/dl. All the elements of the first 6-hour sepsis bundle were received by 52% of patients and, by definition, the rate of compliance was 52%, with the lowest compliant element being the measurement of serum lactate.

Dividing the 101 patients into compliant (n = 52) and non-compliant (n = 49) 6-hour sepsis bundle groups, we found that the two groups were comparable in age, gender, sources of infection and type of specialties (Table 2). The two groups were not only comparable in their severity of sepsis at the points for interventions of the 6-hour sepsis bundle, assessed using median MEWS, but were also comparable for the appropriateness of further interventions, assessed using their requirement for the 24-hour sepsis bundle in critical care settings. Compared with the compliant group (Figure 1), however, we found that the non-compliant group had a more than twofold increase in hospital mortality (49% versus 23%, RR 2.12 (95% CI 1.20 to 3.76), P = 0.01). The number needed to treat to save one life was approximately 4.

Of the 101 patients, 71 (70%) were admitted into critical care and 59% (42/71) of these patients achieved all goals in the 6-hour sepsis bundle. The acute physiology and chronic health evaluation (APACHE) II score and the predicted hospital mortality were similar between the compliant and the non-compliant groups, although the hospital mortality was significantly higher in the non-compliant group (55% versus 29%, RR 1.93 (95% CI 1.08 to 3.45), P = 0.045) (Table 3). The number needed to treat remained approximately 4.

Of the 71 critical care patients, 2 (2%) required central venous pressure monitoring only prior to emergency laparotomy and wound debridement. Postoperatively, they were discharged to wards and required no further special care. Of 71 patients requiring organ support, 69 (98%) were qualified for the 24-hour sepsis bundle for clinical care. Of these 69 patients, within the first 24 hours following the diagnosis of severe sepsis: 64% received glucose control < 8.3 mmol/l; 43% had low-dose steroids given when requiring continued use of vasopressors; activated protein C was considered in only 30% of patients; and plateau pressures were maintained < 30 cm H₂O in 85% of ventilated patients. The entire 24-hour sepsis bundle was achieved in only 30% of eligible candidates (21/69) and the rate of compliance, by definition, was 30%. Again, the compliant and the non-compliant groups were comparable in their characteristics and severity of sepsis, but hospital mortality was increased in the non-compliant group from 29% to 50% with a 76% increase in risk for death, although the difference did not reach statistical significance (RR 1.76 (95% CI 0.84 to 3.64), P = 0.16).

The study has some notable strengths. To date, this is the first study to demonstrate the impact of compliance with an adaptation of the SSC 6-hour and 24-hour sepsis bundles on hospital mortality in patients with severe sepsis. Our findings add to the limited literature supporting the association between the use of a group of evidence-based interventions, executed together, and improved outcomes. In addition, the results suggest that if the association between use of process, indicated by compliance with evidence-based treatments, and improved mortality holds true, using process measures rather than the more resource-intensive outcome measures may be the better way for the NHS healthcare system to monitor performance and for the NHS hospitals to compare performance. Process measures based on the results of randomised controlled trials are able to detect relevant differences between hospitals that would not be identified by comparing hospital specific mortality, which is an insensitive indicator of the quality of care 19. Finally, if the NNT is confirmed by future studies, sepsis care bundles will become the most powerful interventions in clinical care.

We recognise that the study also has some limitations. First, the nature of this observational study may have led to some unknown bias that, rather than interventions, may actually be the cause of both the differences in compliance with interventions and the differences in mortality observed. Second, we may not have measured the full clinical impact of these interventions. For example, we did not measure other risk factors for death, such as the severity of late stage of severe sepsis using a sequential organ failure assessment (SOFA) score, or patients' co-morbidity, which may have had an impact on decisions of withholding or withdrawal. Thirdly, we did not assess the patients (n = 30) who did not require critical care admission for inotropes, mechanical ventilation or drotrecogin alfa for glucose control. This approach also deviated from the SSC method. Finally, small sample size has resulted in the failure to demonstrate an association between compliance with the 24-hour sepsis bundle and hospital mortality.