In this population-based cohort in China, we found that CDC Adult Sepsis Event eSOFA criteria identified a smaller group of sepsis patients with higher mortality compared to Sepsis-3 criteria. These findings were consistent with previous study in US hospitals that reported lower prevalence (4.4% vs. 6.1%) and higher mortality (17.1% vs. 14.4%) of sepsis patients defined by eSOFA versus Sepsis-3 criteria [6].
In our study, eSOFA criteria had high PPV (82.9%). Only 98 patients (false positive rate 12.5%) were misdiagnosed as sepsis by eSOFA criteria, with hyperlactatemia without evidence of concurrent organ dysfunctions being the major cause. Moreover, the mortality rate of these patients was significantly higher than those fulfilling neither Sepsis-3 nor eSOFA criteria (16.3% vs. 5.6%) [15,16,17]. This indicates that SOFA score alone is unable to detect all infected patients with high risk of mortality [18], and further studies are needed to assess whether additional screening parameters such as lactate could be helpful. In comparison, eSOFA criteria identified more severely ill patients, possibly by including invasive procedures (such as vasoactive agents or invasive mechanical ventilation) in the criteria, although they might not be the same patients diagnosed as septic at the bedside. Meanwhile, our study presented a severely ill cohort of infected patients with older age, more pneumonia (60.3%), higher rates of mechanical ventilation (13.6%) and vasopressor use (7.9%), which might explain the high PPV.
The sensitivity of eSOFA criteria (50.8%) was considerably reduced by patients with mild hypoxia that did not reach the eSOFA point of mechanical ventilation, those with mild thrombocytopenia that did not reach a decrease of platelets by 50% or more from baseline, as well as those with elevated creatinine that was lower than the doubling of baseline. As a result, eSOFA criteria might miss the diagnosis of septic patients who are less severely ill but with significant risk of in-hospital mortality (adjusted OR 2.09, 95%CI 1.35–3.25), questioning the potential use of eSOFA alone as a surveillance tool for sepsis [6].
Our study has some strengths. First, no study has evaluated the external validity of eSOFA criteria or investigated the characteristics and outcomes of infected patients missed and misdiagnosed by eSOFA criteria. Moreover, we manually reviewed the medical records of all patients to obtain relevant data and made diagnosis of infection and sepsis using standardized definitions, whereas previous studies used EHR-based proxies as culture orders and antibiotic administrations for infection [3, 6].
Our study has several limitations. First, this study is a secondary analysis of a database that was not originally designed for the study purpose. Among the 3449 patients with infection in the original cohort, we only collected data from 1716 patients who met Sepsis-1 criteria. This might introduce selection bias to the current study. However, Luo et al. reported that, among 38 infected patients who did not fulfill Sepsis-1 criteria, only 5 were diagnosed as Sepsis-3 [18]. Therefore, it was unlikely that addition of these patients in our analysis would change the major results. Second, the mortality rates of septic patients were significantly higher than those in Rhee et al.’s study. Older age (median age 82 for eSOFA+ patients, and 81 for Sepsis-3+ patients in our study) and higher rate of pneumonia might cause the difference since they were proved as independent risk factors for mortality of sepsis patients [4, 5, 19, 20]. Meanwhile, mortality rates of septic patients vary in previous studies by factors including age, sex, comorbidities, and acuity of illness [2, 19,20,21,22]. Third, our cohort represented a patient population significantly different from the original patient population from which eSOFA criteria were developed, as suggested by older age, more pneumonia, higher rate of mechanical ventilation, and higher mortality. Although our results might not be generalized to other patient populations, it clearly demonstrated the same advantages and disadvantages of eSOFA criteria, i.e., external validity, in different settings and patients [6]. Fourth, some data necessary for the calculation of SOFA and/or eSOFA score were missing, particularly with high proportion of missing data from lactate values. However, the frequency of missing data in our cohort was comparable to that in previous studies [3, 6]. As a matter of fact, this might be regarded as limitations of the scoring systems, rather than limitations of the current and previous studies, because both eSOFA and SOFA scoring systems contained variables such as lactate and, possibly, bilirubin that were not routinely monitored in general wards. Last, due to the lack of gold standard for diagnosis of sepsis [4, 5, 23, 24], concordance between eSOFA and Sepsis-3 criteria, rather than sensitivity and specificity, should be reported. However, in order to highlight the limitations of applying eSOFA criteria as a surrogate of SOFA score, as well as to compare our results with those in previous study [6], we still calculated sensitivity, specificity, and PPV of eSOFA criteria compared to Sepsis-3.