Trajectories of Sepsis-Related Case Fatality in Patients With Cancer in the United States.

Abstract

The recent study by Liu and colleagues1 provides timely data on the temporal trends of case fatality among patients with and without cancer who were hospitalized for sepsis, using the National Inpatient Sample (NIS) data set. However, several areas of the methodological approach of this study confound the interpretation of their findings.The authors have examined the temporal trends of case fatality in sepsis hospitalizations using generalized linear models, but they do not provide details on model construction. Specifically, what were the covariates included in these models to ensure that the resultant trends are risk-adjusted over time?In addition, in the propensity score–adjusted models comparing case fatality in patients with and without cancer who were hospitalized for sepsis, the authors report constructing the propensity score only on the basis of the patient’s age, sex, and Charlson comorbidity score. What were the other covariates included in these models?More broadly, the trend analyses of case fatality and the estimates of the prognostic impact of cancer in patients hospitalized for sepsis have not been adjusted for measures of severity of illness, which most likely differed between patients with and with- out cancer and may have changed over time. Although the NIS data set does not include clinical information that is commonly used in calculation of severity-of-illness scores, data on acute organ dysfunction (which was obtained by the authors) are readily obtainable and have been used as measures of severity of illness in prior epidemiological studies of claims-based data.2,3Finally, the authors report 30-day mortality data for their cohort. However, the NIS does not provide time-based mortality data, and mortality information is confined to in-hospital deaths.4 Because the NIS data are deidentified, with no patient-specific identifier codes, the data set cannot be linked to other data repositories to derive time-based mortality on the patients in the study cohort. How did the investigators obtain the vital status at 30 days for each hospitalized patient in their cohort?As importantly, if the case fatality data in this study were based on in-hospital mortality, the short- term mortality among patients with and without cancer who were hospitalized for sepsis may have been underestimated because of increasing rates of discharge to hospice, as was previously documented for patients hospitalized for sepsis in the United States.5 Indeed, use of the combination of in-hospital mortality or discharge to hospice in patients with sepsis was shown to eliminate findings of decreased in-hospital death alone as an end point of short-term mortality.5 Discharge to hospice was reported in the NIS during the first 6 years of the present study (through 2011).4 What were the trends of discharge to hospice and those of combined rates of in-hospital death and discharge to hospice among patients with cancer hospitalized for sepsis during those years?A better understanding of the epidemiology of sepsis in cancer patients can inform efforts to improve patient outcomes.