Sepsis, most succinctly defined as organ dysfunction due to infection, is estimated to account for more than 5 million deaths around the world each year and to cause or contribute to approximately half of all deaths occurring in hospitals in the United States.1,2 A 2016 report from the Healthcare Cost and Utilization Project estimated the cost of treating sepsis in US hospitals in 2013 at $24 billion, making it the most expensive condition treated in US hospitals among all payers.3 In recent years, the substantial burden of sepsis at the individual, health system, and societal levels has become increasingly recognized by the medical community and the public, culminating in the World Health Organization (WHO) adopting the Improving the Prevention, Diagnosis, and Management of Sepsis resolution at the World Health Assembly in May 2017.4
Despite the increased visibility of sepsis and increasing awareness of the global health challenge it poses, accurate quantification of sepsis incidence, mortality, and morbidity remains elusive. In the absence of a definitive diagnostic test for sepsis, the gold standard remains the clinical diagnosis of infection and identifying new organ dysfunction caused by the host response to that infection. This basic concept remains unchanged by the third consensus definition of sepsis promulgated in 2016, which, unlike the first 2 consensus definitions, proposed quantifying the degree of organ dysfunction needed to diagnose sepsis by using the Sequential Organ Failure Assessment (SOFA) score.5
Estimates for the incidence of sepsis in the United States vary between 850 000 and 3 000 000 cases annually, the greater than 3-fold difference explained by the variation in case definitions applied to hospital administrative databases.6 The only available global estimate, 19.4 million cases per year, is based on extrapolation of estimates of hospital-treated sepsis in high-income countries. This is likely an underestimate, given the near total lack of data from low- or middle-income countries, where the burden of infectious diseases is higher.1,7
Practical methods available to quantify the burden of sepsis include clinical diagnosis through prospective cohort studies or retrospective chart reviews, the use of administrative data (International Classification of Diseases [ICD] codes associated with hospital discharges, emergency department visits, or reimbursement claims), and, where they are available, interrogation of electronic health records (EHRs). To date, most large-scale population-based studies have used administrative data and identified cases of sepsis via an “explicit” diagnosis if the patient was allocated an ICD code for severe sepsis (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] code 995.92) or septic shock (785.52), or an “implicit” diagnosis when allocated a code indicating infection together with a code indicating acute organ dysfunction (eg, ICD-9-CM code 569.83, perforation of intestine, combined with 584, acute renal failure). The majority of such studies have used a combination of codes taken or adapted from the approach used by Angus et al.8 One important concern with this methodology is the inability to determine a causal relationship between infection and acute organ dysfunction, which might result in overcounting sepsis cases. A more recent concern with this methodology is so-called upcoding, in which the diagnosis of sepsis may be sensitive to changes in awareness of the diagnosis and may be influenced by financial incentives if coding for sepsis results in higher reimbursement.
Collectively, approaches using administrative data have demonstrated a substantial increase in sepsis incidence in the United States over the past 2 decades, accompanied by a significant decrease in sepsis-specific mortality.6 A 2017 report found that between 2005 and 2014, inpatient stays for “septicemia” in the United States almost tripled from 518 000 to 1 514 100 cases.9 Possible explanations for these observations are an actual increase in the incidence of sepsis with improved outcomes, improved coding of patients with sepsis but including patients with milder disease, a combination of the two, or so-called upcoding.
In this issue of JAMA, Rhee and colleagues10 report the findings of a carefully conducted and analyzed study that seeks to determine whether the reported trends in sepsis incidence and mortality in the United States are real. Additionally and importantly, the authors provide estimates for the accuracy of 2 administrative data–based approaches in estimating the total number of sepsis cases. The study included more than 7 million adult hospitalizations over a 5-year period at 409 hospitals representative of all US hospitals in terms of geographic region, size, and teaching status. The authors generated estimates for sepsis incidence and mortality from 2009-2014 using 3 methods. The first method was based on an analysis of EHR clinical data, which the authors termed a “clinical surveillance” definition. The second method involved an administrative data–based method that replicated the methodology of Angus et al8 to identify sepsis through explicit or implicit ICD-9-CM coding. The authors refer to these as “claims-based” definitions, a somewhat US-centric view, as the ICD evolved from the International List of Causes of Death adopted by the International Statistical Institute in 1893. Entrusted to the WHO on its founding in 1948, its primary use globally is for epidemiology, health planning, and quality purposes rather than for financial claims. The third method used by Rhee et al was retrospective clinician chart review of 510 randomly selected hospitalizations, stratified by EHR definition (yes or no), drawn from 3 academic centers and 2 community hospitals. The authors reported the incidence and mortality rates derived from the EHR and administrative data approaches and assessed the sensitivity, specificity, and positive and negative predictive values of these approaches using clinician chart review as the gold standard.
For the EHR clinical surveillance–based approach, sepsis was defined as presumed infection leading to serious organ dysfunction. “Presumed infection” required the combination of a blood culture draw and at least 4 consecutive days of antibiotic administration. In keeping with the Sepsis-3 criteria, “organ dysfunction” was identified by an increase in the SOFA score of at least 2 points, with a score of at least 2 points in a single organ system. Clinical characteristics such as hypoxemia without mechanical ventilation, thrombocytopenia with platelet count 100 to 150 cells/μL, hypotension not requiring vasopressors, and altered mental status, although components of the SOFA score, were not included in the surveillance definition.
The authors report that in 2014, a total of 173 690 cases of sepsis were identified using EHR clinical criteria among 2 901 019 adults admitted to study hospitals (overall incidence, 6%). Of the patients identified with sepsis, 26 061 (15.0%) died in the hospital and 10 731 (6.2%) were discharged to hospice. The authors highlight that from 2009-2014, the administrative data (explicit sepsis/septic shock codes) support an increasing incidence (+10.3% per year) and declining in-hospital mortality (−7.0% per year) from sepsis, whereas EHR data (clinical) suggest a stable incidence with a significant but lesser decrease in mortality (incidence, +0.6% per year; mortality, −3.3% per year). As the authors suggest, by using a consistent methodology to analyze data that are less subject to ascertainment error and less subject to bias or gaming than coding, the EHR clinical data are more likely than the coding data to produce accurate data for trends over time. Based on the chart review of the 510 randomly selected hospitalizations, neither method accurately identified all cases of sepsis; both the EHR clinical method and counting both explicit and implicit cases by coding missed approximately 20% of sepsis cases, whereas counting only explicit cases identified by coding missed approximately 40%. Conversely, the EHR clinical method had a 20% false-positive rate, thus arriving at the same total number of cases as chart review; the explicit/implicit coding method had a much higher false-positive rate and would result in a sepsis count twice that identified by chart review.
If the EHR-derived clinical estimate is considered the most accurate, the data reported by Rhee et al support the contention that sepsis is a major public health problem in the United States. The estimated national weighted incidence (after adjusting for hospital region, size, and teaching status) of 5.9% among hospitalized patients, and the 15.6% in-hospital mortality rate, suggest that in 2014 there would have been approximately 1.7 million hospitalizations for sepsis and 270 000 deaths related to sepsis. This represents more than 10% of the 2.63 million deaths reported in that year.11
The data on sepsis incidence and mortality reported by Rhee et al are informative regarding the disease burden in the United States; however, sepsis is also a major public health issue at the global level, and the global burden of sepsis is poorly documented. To date, sepsis has not been featured in the Global Burden of Disease statistics, with deaths due to sepsis being attributed to other primary causes, such as underlying infection. Recognizing the global importance of sepsis, new collaborations are needed to develop separate estimates for the global epidemiology of sepsis. Although EHRs are common in the United States and increasingly common in other high-income countries, they are rare in many low- or middle-income countries, where approximately 87% of the world’s population lives. In most countries, analysis of ICD coding remains the only practical method available to quantify the burden of sepsis. Notably, the WHO resolution urges member states “to apply and improve the use of the ICD system to establish the prevalence and profile of sepsis.”4 The study by Rhee et al not only provides a clearer understanding of trends in the incidence and mortality of sepsis in the United States but also a better understanding of the challenges in improving ICD coding to accurately document the global burden of sepsis.
Corresponding Author: Simon Finfer, MD, FCICM, Level 5, 1 King Street, Newtown, NSW 2042, Australia (email@example.com).
Published Online: September 13, 2017. doi:10.1001/jama.2017.13697
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none were reported.
Funding/Support: Dr Rudd is supported by grant T32-HL007287-39 from the National Heart, Lung, and Blood Institute. Dr Finfer is supported by National Health and Medical Research Council Practitioner Fellowship 1117230.
Role of the Funders/Sponsors: The funders had no role in the preparation, review, or approval of the manuscript or the decision to submit the manuscript for publication.
Rudd KE, Delaney A, Finfer S. Counting Sepsis, an Imprecise but Improving Science. JAMA. Published online September 13, 2017. doi:10.1001/jama.2017.13697