Article Preview
TopIntroduction
The Centers for Medicare and Medicaid Services (CMS) estimated that the healthcare expenditure in the United States in 2016 alone was $3.4 trillion, an increase of about 4.8% from 2015. According to the CMS, this trend will continue, with an average growth rate of 5.6% per year until 2025 (“National Health Expenditure Data,” 2018), and this significant growth is partially the result of an aging population, increased lifespans, and growing costs associated with repeated hospital visits for patients with serious infectious diseases.
Sepsis is an inflammatory condition caused by infection and results in a relatively high mortality rate (Amland & Hahn-Cover, 2016; Singer et al., 2016). Sepsis and septic shock are common causes of morbidity and mortality (Raghavan & Marik, 2006). The condition of patients with sepsis can change from stable to near death in a very short period, from days to even just several hours (Taneja et al., 2017). Early diagnosis and prompt treatment have been associated with improved outcomes and alleviated risks (Nachimuthu & Haug, 2012). During the past decade, cases of sepsis have increased significantly (Raghavan & Marik, 2006), and sepsis is considered to be one of the main causes for admission to intensive care (Vincent et al., 1996). In the United States, for example, sepsis accounts for more cases of death than prostate cancer, breast cancer, and AIDS combined (“Sepsis Fact Sheet,” 2016). The Agency for Healthcare Research and Quality (AHRQ) indicated that sepsis is the most expensive condition treated in U.S. hospitals, costing more than $20 billion in 2011, with an average annual increase of 11.9%. Fleischmann et al. (2016) argued that reducing the burden of sepsis is a global challenge, and many countries suffer from a high level of mortality and morbidity from it. Given the current state of sepsis complications and the significant social and economic benefits of better treatment of this disease, it is paramount for health care providers to find more effective ways to deal with sepsis and improve the health outcomes of sepsis patients.
According to the Centers for Disease Control and Prevention, sepsis can be a challenge for medical providers and health care professionals (HCPs) because there is no standard diagnostic test for it (Epstein, 2016). Sepsis diagnosis relies on the judgment of HCPs. Furthermore, it has been estimated that if a country like the United States can achieve an earlier sepsis identification and evidence-based treatment, there will be 92 thousand fewer deaths and savings of more than 1.5 billion in medical cost annually (Shorr, Micek, Jackson, & Kollef, 2007). Predictive analytics in healthcare facilities has mostly been limited to the simple heuristics and scoring systems. HCPs can leverage predictive analytics and machine-learning (ML) techniques to harness the variables available through electronic health records (EHRs) to better predict patient outcomes (Bhattacharjee, Edelson, & Churpek, 2017).
In this study, we explored how predictive models can be used to identify patients with a high mortality risk while they are hospitalized. We empirically developed and examined our model using a large archival dataset from the National Inpatient Sample (NIS), which was created by the AHRQ through a federal-state-industry partnership. A total of 1,048,575 sepsis patients’ records were collected between the years 2008 and 2012. This sample was subdivided randomly into training, testing, and validation data partitions. Death during hospitalization (in-hospital mortality) was defined as a target variable. The independent predictors of mortality were identified using the Chi-square Automatic Interaction Detector (CHAID) model in testing. Additionally, the model was validated using receiver-operating characteristic (ROC) curves and accuracy metrics.