Three Decades of Sepsis in the United Kingdom: Tracing the Tides of Time

Our understanding of sepsis has evolved significantly over the past decades (1). Whether these advances improved outcomes is less clear. Several studies suggested improvement (2, 3), but they are challenging to interpret because of heterogeneity of patient populations, evolving diagnostic criteria, and changes in background care and population health. In this issue of the Journal, Prescott and colleagues (pp. 507–516) present arguably the most comprehensive evaluation to date, examining sepsis epidemiology and outcomes over a 30-year span from the United Kingdom’s Acute Physiology and Chronic Health Evaluation II (UK APACHE II) and Intensive Care National Audit and Research Centre Case Mix Program databases (4). The study encompasses more than 426,000 ICU admissions with sepsis across 261 ICUs from 1988 to 2019, applying standardized data collection that includes patient demographics, five advanced chronic comorbidities, ICU admission diagnoses, APACHE II scores, and ICU and hospital mortality outcomes.

To ensure consistency in measuring sepsis over the entire 30-year study period, the authors uniformly applied Sepsis-3 criteria retrospectively across all epochs (5). They also used a hierarchical classification method from the UK APACHE II study for admission diagnoses (6). They conducted three separate analyses. The primary analysis examined changes in patient case mix and outcomes across nine 3-year epochs, adjusting for variables such as age, sex, infection site, chronic conditions, and APACHE II scores within 24 hours of ICU admission. Next, they conducted a matched cohort analysis to understand shifts in acute severity and outcomes among patients with similar preadmission characteristics. Here, they matched patients from the initial epoch (1988–1990) with those from later epochs based on age, chronic conditions, and infection sites. Finally, because the number of ICUs contributing data grew 10-fold (from 26 to 261) over time, they conducted a subset analysis restricted to the 10 ICUs that contributed data throughout. In addition to adjusting for known confounders, they also calculated E-values to estimate the impact an unmeasured variable would have to exert to explain the association between epoch and hospital mortality rates (7).

There were several interesting findings. First, sepsis accounted for a greater proportion of all ICU cases over time, rising 2.6-fold, from 7.4% in the first epoch (1988–1990) to 19.6% in the last epoch (2017–2019). One reason was that the proportion of ICU admissions transferred directly from emergency departments also more than doubled from 18.1% to 41.3%, with sepsis often being the reason for admission. The source of infection also changed, with an increase in urogenital and a decrease in gastrointestinal infections. Patients tended to be older (median age climbed from 63 to 66 yr), have less chronic illness (except for immunosuppression), and have less acute organ dysfunction at presentation. Together, these changes resulted in a lower presenting APACHE II score over time (from 20 to 14). Not surprisingly, unadjusted mortality declined considerably: ICU mortality fell from 46.6% to 24.5%, and hospital mortality from 54.6% to 32.4%. Importantly, the adjusted mortality also decreased. For example, the adjusted hospital mortality fell 8.8%, reflecting a 40% relative reduction over 30 years. Findings were consistent across all three analyses. The authors also found that any unmeasured confounder would have to exert a fairly large effect.

Broadly speaking, this study can be interpreted as yielding highly positive results. Campaigns such as the Surviving Sepsis Campaign and many regional quality improvement efforts have focused on raising awareness of sepsis and implementing strategies to ensure prompt diagnosis and intervention and a greater commitment of ICU resources (8). These data would appear to support these efforts. Sepsis appears to be a more dominant reason for admission; patients seem to be admitted sooner, as evidenced by less organ dysfunction at presentation; and outcomes are better than would be expected after adjusting for several known confounders. Improved outcomes could well be due both to earlier diagnosis and improvements in ICU care.

There are, however, some residual methodologic challenges, some of which are inherent to ICU database studies. In particular, the problem with an analysis of patients admitted to the ICU is the inability to account for patients who were not admitted to the ICU—a form of selection bias. This limitation is particularly significant if the goal is to interpret changes over time, and the likelihood of being admitted also changed over time. Over the 30-year period, there was considerable growth in ICU resources in the United Kingdom. Indeed, the number of admissions per ICU grew in this study from ∼700 to 2,400. Consider a scenario involving two patients with sepsis with identical APACHE scores seeking emergency care during a period of limited ICU availability. The clinician, faced with just one available ICU bed, must make a choice. This decision is unlikely to be random and may favor the patient perceived to benefit most. For instance, if one patient shows a slower response to a fluid bolus, this might influence the clinician’s decision. In contrast, in later years, with more ICU beds available, both patients might be admitted. Thus, subtler aspects of patient condition, not captured by the APACHE II score, could lead to the systematic exclusion of less critically ill patients in earlier periods when ICU resources were scarce. This substantial change in admission likelihood over time could introduce significant selection bias and potentially confound mortality findings. The study also contends with the risk of “diagnostic drift.” Over time, the methods for recording laboratory values and physiological data may have changed. Because unmeasured variables are presumed normal, infrequent measurements could lead to underestimating illness severity. However, the direction of this drift is uncertain, as the extent of data missingness could have either increased or decreased.

These methodologic concerns notwithstanding, the study by Prescott and colleagues offers unique insights into the epidemiology of sepsis over 3 decades. Given the worldwide effort to fight sepsis, understanding whether we are making progress or not is crucial. In that regard, this study is extremely helpful. At least in the United Kingdom, sepsis appears to be recognized more frequently and earlier, with greater use of ICU services. And, although there may be some residual confounding, it seems that early recognition, prompt intervention, and higher-quality ICU care are likely leading to considerable improvements in short-term mortality.