ObjectivesThe objective of this study was to compare the performance of two popularly used early sepsis diagnostic criteria, systemic inflammatory response syndrome (SIRS) and quick Sepsis-related Organ Failure Assessment (qSOFA), using statistical and machine learning approaches.MethodsThis retrospective study examined patient visits in Emergency Department (ED) with sepsis related diagnosis. The outcome was 28-day in-hospital mortality. Using odds ratio (OR) and modeling methods (decision tree [DT], multivariate logistic regression [LR], and naïve Bayes [NB]), the relationships between diagnostic criteria and mortality were examined.ResultsOf 132,704 eligible patient visits, 14% died within 28 days of ED admission. The association of qSOFA ≥2 with mortality (OR = 3.06; 95% confidence interval [CI], 2.96–3.17) greater than the association of SIRS ≥2 with mortality (OR = 1.22; 95% CI, 1.18–1.26). The area under the ROC curve for qSOFA (AUROC = 0.70) was significantly greater than for SIRS (AUROC = 0.63). For qSOFA, the sensitivity and specificity were DT = 0.39, LR = 0.64, NB = 0.62 and DT = 0.89, LR = 0.63, NB = 0.66, respectively. For SIRS, the sensitivity and specificity were DT = 0.46, LR = 0.62, NB = 0.62 and DT = 0.70, LR = 0.59, NB = 0.58, respectively.ConclusionsThe evidences suggest that qSOFA is a better diagnostic criteria than SIRS. The low sensitivity of qSOFA can be improved by carefully selecting the threshold to translate the predicted probabilities into labels. These findings can guide healthcare providers in selecting risk-stratification measures for patients presenting to an ED with sepsis.

Diabetic retinopathy (DR) is a leading cause for blindness among working-aged adults. The growing prevalence of diabetes urges for cost-effective tools to improve the compliance of eye examinations for early detection of DR. The objective of this research is to identify essential predictors and develop predictive technologies for DR using electronic health records. We conducted a retrospective analysis on a derivation cohort with 3749 DR and 94,127 non-DR diabetic patients. In the analysis, an ensemble predictor selection method was employed to find essential predictors among 26 variables in demographics, duration of diabetes, complications and laboratory results. A predictive model and a risk index were built based on the selected, essential predictors, and then validated using another independent validation cohort with 869 DR and 6448 non-DR diabetic patients. Out of the 26 variables, 10 were identified to be essential for predicting DR. The predictive model achieved a 0.85 AUC on the derivation cohort and a 0.77 AUC on the validation cohort. For the risk index, the AUCs were 0.81 and 0.73 on the derivation and validation cohorts, respectively. The predictive technologies can provide an early warning sign that motivates patients to comply with eye examinations for early screening and potential treatments.

No abstract

Summary Objectives: Our goal was to develop predictive models for sepsis and in-hospital mortality using electronic health records (EHRs). We showcased the efficiency of these algorithms in patients diagnosed with pneumonia, a group that is highly susceptible to sepsis. Methods: We retrospectively analyzed the Health Facts® (HF) dataset to develop models to predict mortality and sepsis using the data from the first few hours after admission. In addition, we developed models to predict sepsis using the data collected in the last few hours leading to sepsis onset. We used the random forest classifier to develop the models. Results: The data collected in the EHR system is generally sporadic, making feature extraction and selection difficult, affecting the accuracies of the models. Despite this fact, the developed models can predict sepsis and in-hospital mortality with accuracies of up to 65.26±0.33% and 68.64±0.48%, and sensitivities of up to 67.24±0.36% and 74.00±1.22%, respectively, using only the data from the first 12 hours after admission. The accuracies generally remain consistent for similar models developed using the data from the first 24 and 48 hours after admission. Lastly, the developed models can accurately predict sepsis patients (with up to 98.63±0.17% accuracy and 99.74%±0.13% sensitivity) using the data collected within the last 12 hours before sepsis onset. The results suggest that if such algorithms continuously monitor patients, they can identify sepsis patients in a manner comparable to current screening tools, such as the rulebased Systemic Inflammatory Response Syndrome (SIRS) criteria, while often allowing for early detection of sepsis shortly after admission. Conclusions: The developed models showed promise in early prediction of sepsis, providing an opportunity for directing early intervention efforts to prevent/treat sepsis.

Healthcare industry generates streams of data in different problem domains. Analysis of such data requires stream analytics tools and techniques to generate useful insights. Stream analytics involve analysis of time variant events. The specific patterns in the events can indicate some imminent outcomes such as state of a heart, etc. Therefore, novel ways to find specific patterns in the events generated by multiple sources are required. A key requirement for applying any such method is data preparation and organization to enable such analysis. In this paper, we extend the CRISP-DM process to include data preparation approaches for sequence mining. We present progression analysis, an approach for converting multidimensional time variant streams of health records in a form to be able to detect useful sequential signals. To illustrate the process, we use patient health history stored in an Electronic Medical Record system (EMR) and present a healthcare application to compare progression of diseases over time between patients diagnosed with Tobacco Use Disorder (TUD) and non-tobacco users. Interestingly, many diseases follow the same path for TUD and non-TUD patients. Finally, the generalizability of the progression analysis is discussed.