The Sepsis Six: Does Public Data Show Whether It Worked?

In 2005, an NHS intensive care consultant distilled a 58-recommendation international guideline into six actions any ward nurse could perform within one hour. The Sepsis Six is now used in 96% of British hospitals and 37 countries. Bootstrap CUSUM applied to every available public measure of sepsis mortality in England — 22 years of data, two series, six charts — finds one stage and no change point in either. The data cannot prove or disprove whether the intervention saved lives. This article explains precisely why — and why that matters for every clinical quality improvement programme in the NHS.

One hour. Six actions. A bundle that spread around the world.

In 2005, Dr Ron Daniels — an intensive care consultant at Good Hope Hospital in Sutton Coldfield — looked at the Surviving Sepsis Campaign’s international guidelines and saw a problem. The guidelines were 58 recommendations across 16 pages. Five years after publication, they were being achieved in fewer than one in seven patients. The gap between the evidence and the ward was not a knowledge problem. It was a translation problem.

Daniels and colleagues distilled the guidelines into six actions that any nurse or doctor could perform at the bedside, without specialist equipment, within one hour of recognising sepsis. They called it the Sepsis Six.

The bundle was pragmatic by design. Daniels recognised that a complex protocol requiring specialist knowledge would not reach the ward in time. Sepsis kills through a dysregulated inflammatory response that progresses from infection to organ failure to death within hours. The Sepsis Six addressed the hours immediately after recognition — the window when bedside action matters most.

In 2011, the UK Sepsis Trust published the first evidence that the Sepsis Six was associated with a 50% reduction in mortality compared to patients who did not receive the full bundle. The UK Sepsis Trust was formally founded in 2012. NHS England adopted the Sepsis Six as a CQUIN indicator — a financial incentive measure — in 2013. The NICE guideline NG51 on sepsis recognition and management was published in 2015. The Think Sepsis mandatory screening campaign launched in 2017.

By 2019, survival rates from sepsis in the UK had reportedly risen from 70% in 2012 to 80%. The Sepsis Six appeared to be a success story.

Bootstrap CUSUM asks a different question.

What is sepsis — and why measuring it is harder than it looks

Sepsis is not a disease. It is a syndrome — the body’s dysregulated response to infection, producing organ dysfunction that can progress to septic shock and death. The infecting organism can be bacterial, viral, or fungal. The source can be anywhere: a urinary tract infection, pneumonia, an infected wound, a bowel perforation. What makes sepsis sepsis is not the infection but the body’s catastrophic response to it.

This creates a fundamental measurement problem. When a person dies of pneumonia complicated by sepsis, what was the cause of death — the pneumonia or the sepsis? International death certification guidelines ask certifiers to record the underlying cause: the disease that started the chain of events. For most sepsis deaths, the underlying cause is the precipitating infection — pneumonia, urinary infection, bowel perforation — not the sepsis itself.

This is not a data error. It is a feature of how death certification works. But it has a critical implication for evaluating the Sepsis Six: the publicly available mortality series — deaths where A40 or A41 was the underlying cause — is both incomplete and unstable over time. If clinical campaigns train staff to recognise and code sepsis as the primary cause more frequently, the underlying cause series will rise even if mortality is actually falling.

📊 Data note: Analysis uses ONS “Deaths involving sepsis, England and Wales, 2001 to 2023” user-requested dataset (released 2024). Two series: Table 1 (sepsis as underlying cause, A40/A41) and Table 2 (sepsis mentioned anywhere on certificate). N=23 annual observations for both. Bootstrap CUSUM, Loops=5,000, Turn Length=3, Conf=90%. England and Wales population grew from approximately 52 million in 2001 to 60 million in 2023 — a 15% increase — and aged substantially over the same period. Both factors increase raw sepsis deaths independently of any change in clinical outcomes. The age-standardised rate would correct for this but is not available in this dataset. Data file: ⇣ sepsis-both-series.csv

The recognition campaigns that changed the coding

Four campaigns ran in sequence and each changed how sepsis was documented and coded:

• Surviving Sepsis Campaign (2002 onwards): International initiative raising awareness among intensivists and emergency physicians. Primarily an ICU-level intervention. Published major guidelines in 2004, 2008, 2012, and 2016.
• NHS England CQUIN for sepsis screening (2013): Financial incentive — trusts received additional payment for documenting sepsis screening. Directly incentivised staff to document and code sepsis. This is the year the Bootstrap CUSUM detects the ratio change point at 95.5% confidence.
• NICE guideline NG51 (2016, updated 2017 and 2019): First NICE guideline specifically addressing sepsis recognition in all clinical settings. Introduced “Red Flag Sepsis.” Extended the focus beyond ICU to ward and community settings.
• NHS England Think Sepsis mandatory screening (2017): Made sepsis screening mandatory in A&E and acute admissions. The most significant coding-level intervention — ensuring sepsis was formally documented whenever present.

The coding change point at 2013 aligns precisely with the CQUIN introduction. Financial incentives for documentation produce documentation improvements. That is not a criticism — accurate coding is genuinely valuable for resource allocation and research. But it must be separated from clinical outcome measurement. The Bootstrap CUSUM does exactly this: it detects a structural change in the coding ratio at 2013, while finding no structural change in either mortality series.

The Bootstrap CUSUM analysis — two series, six charts

The ONS dataset provides two distinct mortality series for England and Wales from 2001 to 2023. Running Bootstrap CUSUM on both — and comparing them — tells a more complete story than either series alone.

Series 1 — Deaths where sepsis was mentioned anywhere on the certificate

This is the more complete clinical measure. It captures deaths associated with sepsis regardless of which condition was coded as the primary cause. The values range from approximately 17,600 in 2001 to 26,200 in 2023, with a peak of approximately 27,400 in 2006 and a trough of approximately 19,300 in 2020.

Series 2 — Deaths where sepsis was the underlying (primary) cause

This is the headline measure most often cited in mortality statistics. It counts only deaths where A40 (streptococcal sepsis) or A41 (other sepsis) was recorded as the primary underlying cause. Values range from approximately 1,958 in 2001 to 4,276 in 2023.

The ratio — the article’s most important finding

Comparing the two series year by year reveals something the individual series cannot show. In 2001, for every death where sepsis was the primary underlying cause, there were approximately nine deaths where sepsis appeared somewhere on the certificate but not as the primary cause. By 2022 that ratio had fallen to approximately 6.8. By 2023 it was 6.1.

A falling ratio means an increasing proportion of sepsis deaths are being coded as primary sepsis rather than as the underlying infection. This is a coding improvement, not a clinical improvement. More deaths are being correctly identified as primarily attributable to the septic response.

Running Bootstrap CUSUM on the ratio itself — not on either mortality series, but on the ratio between them — produces the article’s most revealing result.

What the Surviving Sepsis Campaign claimed — and what the evidence actually shows

The UK Sepsis Trust published a 2011 study showing that full Sepsis Six compliance was associated with a 50–54% reduction in hospital mortality compared to non-compliance. This finding has been widely cited and has driven NHS adoption. It is genuine evidence — but it is observational evidence from a single hospital, and the comparison is between patients who received the full bundle and those who did not.

The question this analysis asks is different. Not “does the Sepsis Six work in patients who receive it?” but “did the national rollout of the Sepsis Six change the population-level trajectory of sepsis mortality in England?” Those are not the same question, and they require different evidence.

The observational evidence from 2011 is consistent with the Sepsis Six being effective in patients who receive all six components. But it tells you nothing about whether the NHS rollout achieved consistent enough compliance at sufficient scale to move the national mortality curve. The Bootstrap CUSUM answers that question — and finds no detectable change in either public mortality series across the 22 years of data.

A 2025 ecological study (Broad et al., Infection, Guy’s and St Thomas’ NHS Foundation Trust / King’s College London / UCL) used HES data to examine all sepsis-coded hospital admissions in England from April 1998 to March 2024, including mortality. It found sepsis-coded admissions increased 7.5-fold over that period — but concluded that this rise “may have been impacted by coding changes and improved disease recognition” rather than reflecting a true increase in sepsis incidence. That conclusion is precisely the coding change argument this article makes, confirmed by an independent research team using a different data source. The 2025 paper describes the epidemiological trends; it does not apply Bootstrap CUSUM to ask whether a structural change point in the mortality rate is detectable and whether its date is consistent with the Sepsis Six intervention timeline. That is what this article does. The two analyses are complementary, not competing — and together they make a stronger case than either alone.

The Deming/Joiner lens: what the null result actually means

The Bootstrap CUSUM finding of no change point is not the same as finding that the Sepsis Six did not work. It is a finding about the measurement system, not (only) about the intervention. Deming’s theory of knowledge is precise on this: you cannot learn from an outcome without a stable, consistent measure. When the measure changes during the intervention period, the study step of PDSA is compromised.

The comparison with the Never Events analysis on this site is instructive. The Never Events Bootstrap CUSUM finds one stage, no change point — and the reason is clear: the system has not been redesigned at the right level of intervention. The Sepsis Six Bootstrap CUSUM also finds one stage, no change point — but here the reason is different: the measure itself changed during the intervention period, making it impossible to detect a clinical signal even if one exists.

Same result. Different diagnosis. Same conclusion: the Study step of PDSA was never properly completed at national level.

The three measure types — applied to the Sepsis Six

The table below applies Langley et al’s three measure types directly to the Sepsis Six. Reading it makes clear why the NHS’s evaluation was incomplete and what a complete evaluation would require.

Bright Spots: where is the Sepsis Six working — and why?

The Bootstrap CUSUM null result — no detectable change point in either national mortality series — tells you the aggregate picture. It does not tell you whether some trusts, regions, or patient groups are doing substantially better than others. That question requires a different analytical approach: looking for bright spots.

The concept of positive deviance was coined by Marian Zeitlin, a nutrition scientist at Tufts University, in her 1990 book Positive Deviance in Child Nutrition (UN University Press). Zeitlin asked why some children in desperately poor communities were well-nourished while their neighbours were not — and found the answer in the specific behaviours of those families, using the same constrained resources as everyone else. Jerry and Monique Sternin operationalised the concept in rural Vietnam in the early 1990s, working with Save the Children to identify families who had avoided childhood malnutrition against all odds. Their pilot rehabilitated 93% of malnourished children and was scaled to 5 million families. Pascale, Sternin and Sternin brought the approach to healthcare and organisational improvement in The Power of Positive Deviance (Harvard Business Press, 2010). Chip and Dan Heath’s Switch (2010) popularised the same idea as “Bright Spots.”

The core question in all of these is the same: rather than asking “why is the system failing?”, ask “where is it working, and what are those places doing differently?” Applied to sepsis: if some NHS trusts are consistently achieving faster time-to-antibiotics, higher bundle compliance, and better survival outcomes than their peers — with comparable patient populations — the question is what they are doing that others are not. That knowledge is actionable in a way that a national average null result is not.

The Sepsis Six null result at national level is not evidence that the intervention has failed. It is evidence that the national average conceals substantial variation — some trusts performing much better, some much worse — and that the improvement opportunity lies in understanding and propagating the bright spots rather than continuing to push the aggregate number. Bootstrap CUSUM is the tool that makes that analysis rigorous: it distinguishes trusts where the outcome has structurally changed from those where variation remains within common cause noise, and it dates the change precisely enough to ask what happened there that year.

The patients who survived — and what happened to them

This analysis examines mortality. The 2025 ecological study by Broad et al. examines admissions and survival. Between the two papers, a picture emerges that neither alone provides: more patients are surviving sepsis — but we do not know what that survival means for their lives.

This connects directly to the second of Langley et al’s three questions: “How will we know that a change is an improvement?” The answer that is best for patients requires going further than either mortality data or survival data alone. A patient with sepsis wants three things — and public data currently measures only the first two:

• Not to die — mortality data addresses this partially, with the coding limitations this article identifies
• To leave hospital alive — HES admission and survival data addresses this, and the 2025 paper shows survival rates have improved
• To return to the life they had before — neither paper addresses this at all

That third question is where the evidence base is most conspicuously absent. Studies suggest that 40–60% of sepsis survivors develop significant long-term morbidity — cognitive impairment, fatigue, vulnerability to recurrent infections, and psychological trauma collectively described as post-sepsis syndrome. The UK Sepsis Trust now runs a dedicated post-sepsis support programme recognising this. NICE guideline CG191 (rehabilitation after critical illness) addresses it directly. A patient who survives sepsis after receiving the Sepsis Six but spends the following two years with moderate cognitive impairment, repeated hospital admissions, and an inability to return to work has a very different outcome from a patient who survives and returns to full function. Mortality data counts them identically — both survivors. So does the 2025 HES paper.

The Sepsis Six was designed to keep patients alive. The evidence suggests it may be doing so — more patients are surviving to discharge. But survival is not the same as recovery, and recovery is what patients actually want. The improvement cycle is not complete until the third question is answered.

📊 Post-sepsis resources: UK Sepsis Trust post-sepsis support: sepsistrust.org/patient-information — NICE guideline CG191: Rehabilitation after critical illness in adults, 2009. — Langley G, Moen R, Nolan K et al. The Improvement Guide. 2nd ed. Jossey-Bass, 2009 — source of the three measure types (outcome, process, balancing) and the Model for Improvement.

What would actually show whether the Sepsis Six worked?

The public mortality data cannot answer the question. Three other measures could — and one of them is publicly available and has been collected since 2018.

The NHS England sepsis audit is the most important of these for future evaluation. It collects time-to-antibiotics, bundle compliance, and outcomes from hospitals across England. With annual data from 2018, it will have sufficient observations for Bootstrap CUSUM analysis by approximately 2030. If a structural change point is detectable in that series — a sustained improvement in bundle compliance correlating with a sustained reduction in in-hospital mortality — that would be the first robust population-level evidence that the Sepsis Six is moving the mortality curve. The CUSUM is waiting. The data is accumulating.

The broader lesson for clinical quality improvement

The Sepsis Six is not unique in this problem. Many NHS quality improvement interventions have been adopted on the basis of observational studies, rolled out with enthusiasm, and evaluated against outcome measures that changed at the same time as the intervention. The result is that it is often impossible to say whether the intervention worked — not because it did not, but because the evidence infrastructure to answer the question was never built.

None of this is a criticism of the Sepsis Six itself, or of Ron Daniels and the UK Sepsis Trust who created it. The clinical rationale is sound. The 2011 observational evidence is real. The 96% NHS adoption is a genuine achievement of clinical advocacy and implementation science. The question this analysis raises is not whether the Sepsis Six should have been adopted — it should have been — but whether the NHS has the analytical infrastructure to know whether its adoption produced the population-level outcomes it was intended to achieve.

The Bootstrap CUSUM cannot answer that question from public data. But it can hold the question open, state clearly what evidence would answer it, and point to the data that is now accumulating that will provide the answer in the next decade.

📊 Data source: ONS, Deaths involving sepsis, England and Wales: 2001 to 2023. User-requested dataset, released 2024. Available at: ons.gov.uk. Table 1: underlying cause. Table 2: mentioned anywhere. Figures are registered deaths, England and Wales combined, all persons, all ages.