Dataset Information

Multi-perspective predictive modeling for acute kidney injury in general hospital populations using electronic medical records.

ABSTRACT:

Objectives

Acute kidney injury (AKI) in hospitalized patients puts them at much higher risk for developing future health problems such as chronic kidney disease, stroke, and heart disease. Accurate AKI prediction would allow timely prevention and intervention. However, current AKI prediction researches pay less attention to model building strategies that meet complex clinical application scenario. This study aims to build and evaluate AKI prediction models from multiple perspectives that reflect different clinical applications.

Materials and methods

A retrospective cohort of 76 957 encounters and relevant clinical variables were extracted from a tertiary care, academic hospital electronic medical record (EMR) system between November 2007 and December 2016. Five machine learning methods were used to build prediction models. Prediction tasks from 4 clinical perspectives with different modeling and evaluation strategies were designed to build and evaluate the models.

Results

Experimental analysis of the AKI prediction models built from 4 different clinical perspectives suggest a realistic prediction performance in cross-validated area under the curve ranging from 0.720 to 0.764.

Discussion

Results show that models built at admission is effective for predicting AKI events in the next day; models built using data with a fixed lead time to AKI onset is still effective in the dynamic clinical application scenario in which each patient's lead time to AKI onset is different.

Conclusion

To our best knowledge, this is the first systematic study to explore multiple clinical perspectives in building predictive models for AKI in the general inpatient population to reflect real performance in clinical application.

SUBMITTER: He J

PROVIDER: S-EPMC6447093 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:BackgroundAs a result of demographic changes, physicians are required to deliver needed services with limited resources. Research suggests that tablet PCs with access to patient data may streamline clinical workflow. A recent study found tablets with mobile electronic medical records (EMRs) can facilitate data retrieval and produce time savings across the clinical routine within hospital settings. However, the reasons for these time savings, including details on how tablets were being used, remain unclear. The same applies to physicians' perceptions of this tool within an inpatient setting.ObjectiveThis study examined physicians' perception of tablets with EMRs in an inpatient setting. The rationale was to identify both subjective and objective factors that impacted the successful implementation and use of tablets running an EMR.MethodsWe developed a 57-item survey questionnaire designed to examine users' perception of and attitude toward tablets, which was administered to 14 participating physicians following 7 weeks of tablet use. Five participants volunteered to participate in a second study that investigated physicians' patterns of tablet use within the EMR environment by digitally tracking and storing usage behavior. Statistical analyses of questionnaire results included mean values with their bootstrapped 95% confidence intervals and multivariate analysis of variance to identify predictors of tablet use.ResultsPhysicians reported high degrees of satisfaction with the tablets. There was a general consensus among physicians that tablet use streamlined clinical workflow through optimized data retrieval (rated 0.69, 0.23-1.15 points better than control) and improved communication with patients and other physicians (rated 0.85, 0.54-1.15 and 0.77, 0.38-1.15 points better than control, respectively). Age (F3,11=3.54, P=.04), occupational group (F1,11=7.17, P=.04), and attitude toward novel technologies (F1,11=10.54, P=.02) predicted physicians' satisfaction with the devices and their motivation regarding their further use. Tracking data yielded that only a few of the available functions were used frequently.ConclusionsAlthough tablet PCs were consistently perceived as beneficial, several factors contributed to the fact that their full potential was not fully exploited. Training in functionality and providing a reliable infrastructure might foster successful tablet implementation.

Project description:ObjectiveAttention networks learn an intelligent weighted averaging mechanism over a series of entities, providing increases to both performance and interpretability. In this article, we propose a novel time-aware transformer-based network and compare it to another leading model with similar characteristics. We also decompose model performance along several critical axes and examine which features contribute most to our model's performance.Materials and methodsUsing data sets representing patient records obtained between 2017 and 2019 by the Kaiser Permanente Mid-Atlantic States medical system, we construct four attentional models with varying levels of complexity on two targets (patient mortality and hospitalization). We examine how incorporating transfer learning and demographic features contribute to model success. We also test the performance of a model proposed in recent medical modeling literature. We compare these models with out-of-sample data using the area under the receiver-operator characteristic (AUROC) curve and average precision as measures of performance. We also analyze the attentional weights assigned by these models to patient diagnoses.ResultsWe found that our model significantly outperformed the alternative on a mortality prediction task (91.96% AUROC against 73.82% AUROC). Our model also outperformed on the hospitalization task, although the models were significantly more competitive in that space (82.41% AUROC against 80.33% AUROC). Furthermore, we found that demographic features and transfer learning features which are frequently omitted from new models proposed in the EMR modeling space contributed significantly to the success of our model.DiscussionWe proposed an original construction of deep learning electronic medical record models which achieved very strong performance. We found that our unique model construction outperformed on several tasks in comparison to a leading literature alternative, even when input data was held constant between them. We obtained further improvements by incorporating several methods that are frequently overlooked in new model proposals, suggesting that it will be useful to explore these options further in the future.

Project description:BackgroundModels developed to predict hospital-acquired acute kidney injury (HA-AKI) in non-critically ill patients have a low sensitivity, do not include dynamic changes of risk factors and do not allow the establishment of a time relationship between exposure to risk factors and AKI. We developed and externally validated a predictive model of HA-AKI integrating electronic health databases and recording the exposure to risk factors prior to the detection of AKI.MethodsThe study set was 36 852 non-critically ill hospitalized patients admitted from January to December 2017. Using stepwise logistic analyses, including demography, chronic comorbidities and exposure to risk factors prior to AKI detection, we developed a multivariate model to predict HA-AKI. This model was then externally validated in 21 545 non-critical patients admitted to the validation centre in the period from June 2017 to December 2018.ResultsThe incidence of AKI in the study set was 3.9%. Among chronic comorbidities, the highest odds ratios (ORs) were conferred by chronic kidney disease, urologic disease and liver disease. Among acute complications, the highest ORs were associated with acute respiratory failure, anaemia, systemic inflammatory response syndrome, circulatory shock and major surgery. The model showed an area under the curve (AUC) of 0.907 [95% confidence interval (CI) 0.902-0.908), a sensitivity of 82.7 (95% CI 80.7-84.6) and a specificity of 84.2 (95% CI 83.9-84.6) to predict HA-AKI, with an adequate goodness-of-fit for all risk categories (χ2 = 6.02, P = 0.64). In the validation set, the prevalence of AKI was 3.2%. The model showed an AUC of 0.905 (95% CI 0.904-0.910), a sensitivity of 81.2 (95% CI 79.2-83.1) and a specificity of 82.5 (95% CI 82.2-83) to predict HA-AKI and had an adequate goodness-of-fit for all risk categories (χ2 = 4.2, P = 0.83). An online tool (predaki.amalfianalytics.com) is available to calculate the risk of AKI in other hospital environments.ConclusionsBy using electronic health data records, our study provides a model that can be used in clinical practice to obtain an accurate dynamic and updated assessment of the individual risk of HA-AKI during the hospital admission period in non-critically ill patients.

Project description:Reduction of preventable hospital readmissions that result from chronic or acute conditions like stroke, heart failure, myocardial infarction and pneumonia remains a significant challenge for improving the outcomes and decreasing the cost of healthcare delivery in the United States. Patient readmission rates are relatively high for conditions like heart failure (HF) despite the implementation of high-quality healthcare delivery operation guidelines created by regulatory authorities. Multiple predictive models are currently available to evaluate potential 30-day readmission rates of patients. Most of these models are hypothesis driven and repetitively assess the predictive abilities of the same set of biomarkers as predictive features. In this manuscript, we discuss our attempt to develop a data-driven, electronic-medical record-wide (EMR-wide) feature selection approach and subsequent machine learning to predict readmission probabilities. We have assessed a large repertoire of variables from electronic medical records of heart failure patients in a single center. The cohort included 1,068 patients with 178 patients were readmitted within a 30-day interval (16.66% readmission rate). A total of 4,205 variables were extracted from EMR including diagnosis codes (n=1,763), medications (n=1,028), laboratory measurements (n=846), surgical procedures (n=564) and vital signs (n=4). We designed a multistep modeling strategy using the Naïve Bayes algorithm. In the first step, we created individual models to classify the cases (readmitted) and controls (non-readmitted). In the second step, features contributing to predictive risk from independent models were combined into a composite model using a correlation-based feature selection (CFS) method. All models were trained and tested using a 5-fold cross-validation method, with 70% of the cohort used for training and the remaining 30% for testing. Compared to existing predictive models for HF readmission rates (AUCs in the range of 0.6-0.7), results from our EMR-wide predictive model (AUC=0.78; Accuracy=83.19%) and phenome-wide feature selection strategies are encouraging and reveal the utility of such datadriven machine learning. Fine tuning of the model, replication using multi-center cohorts and prospective clinical trial to evaluate the clinical utility would help the adoption of the model as a clinical decision system for evaluating readmission status.