Imputation of clinical covariates in time series

Diabetes mellitus and blood glucose variability increases the 30-day readmission rate after kidney transplantation

INTRODUCTION

Inpatient hyperglycemia is an established independent risk factor among several patient cohorts for hospital readmission. This has not been studied after kidney transplantation. Nearly one-third of patients who have undergone a kidney transplant reportedly experience 30-day readmission.

METHODS

Data on first-time solitary kidney transplantations were retrieved between September 2015 and December 2018. Information was linked to the electronic health records to determine diagnosis of diabetes mellitus and extract glucometric and insulin therapy data. Univariate logistic regression analysis and the XGBoost algorithm were used to predict 30-day readmission. We report the average performance of the models on the testing set on bootstrapped partitions of the data to ensure statistical significance.

RESULTS

The cohort included 1036 patients who received kidney transplantation; 224 (22%) experienced 30-day readmission. The machine learning algorithm was able to predict 30-day readmission with an average area under the receiver operator curve (AUC) of 78% with (76.1%, 79.9%) 95% confidence interval (CI). We observed statistically significant differences in the presence of pretransplant diabetes, inpatient-hyperglycemia, inpatient-hypoglycemia, minimum and maximum glucose values among those with higher 30-day readmission rates. The XGBoost model identified the index admission length of stay, presence of hyper- and hypoglycemia, the recipient and donor body mass index (BMI) values, presence of delayed graft function, and African American race as the most predictive risk factors of 30-day readmission. Additionally, significant variations in the therapeutic management of blood glucose by providers were observed.

CONCLUSIONS

Suboptimal glucose metrics during hospitalization after kidney transplantation are associated with an increased risk for 30-day hospital readmission. Optimizing hospital blood glucose management, a modifiable factor, after kidney transplantation may reduce the risk of 30-day readmission.

GDP Forecasting: Machine Learning, Linear or Autoregression?

This paper compares the predictive power of different models to forecast the real U.S. GDP. Using quarterly data from 1976 to 2020, we find that the machine learning K-Nearest Neighbour (KNN) model captures the self-predictive ability of the U.S. GDP and performs better than traditional time series analysis. We explore the inclusion of predictors such as the yield curve, its latent factors, and a set of macroeconomic variables in order to increase the level of forecasting accuracy. The predictions result to be improved only when considering long forecast horizons. The use of machine learning algorithm provides additional guidance for data-driven decision making.

Prediction of Neutropenic Events in Chemotherapy Patients: A Machine Learning Approach

PURPOSE

Severe and febrile neutropenia present serious hazards to patients with cancer undergoing chemotherapy. We seek to develop a machine learning-based neutropenia prediction model that can be used to assess risk at the initiation of a chemotherapy cycle.

MATERIALS AND METHODS

We leverage rich electronic medical records (EMRs) data from a large health care system and apply machine learning methods to predict severe and febrile neutropenic events. We outline the data curation process and challenges posed by EMRs data. We explore a range of algorithms with an emphasis on model interpretability and ease of use in a clinical setting.

RESULTS

Our final proposed model demonstrates an out-of-sample area under the receiver operating characteristic curve of 0.865 (95% CI, 0.830 to 0.891) in the prediction of neutropenic events on the basis of only 20 clinical features. The model validates known risk factors and offers insight into potential novel clinical indicators and treatment characteristics that elevate risk. It relies on factors that are directly extractable from EMRs, provided a tool can be easily integrated into existing workflows. A cost-based analysis provides insight into optimal risk thresholds and offers a framework for tailoring algorithms to individual hospital needs.

CONCLUSION

A better understanding of neutropenic risk on an individual level enables a more informed approach to patient monitoring and treatment decisions.