Algorithm could predict acute kidney injury
Findings showed that Previse could predict the onset of AKI up to 48 hours in advance of onset, sooner than the standard hospital systems like XGBoost AKI prediction model and the Sequential Organ Failure Assessment (SOFA).
Dascena Inc., a machine learning diagnostic algorithm company that is targeting early disease intervention to improve patient care outcomes, recently published results from a study evaluating the company’s machine learning algorithm, Previse, for the earlier prediction of acute kidney injury (AKI) in Kidney International Reports.
Findings showed that Previse could predict the onset of AKI up to 48 hours in advance of onset, sooner than the standard hospital systems like XGBoost AKI prediction model and the Sequential Organ Failure Assessment (SOFA).
Previse has previously received Breakthrough Device designation from the FDA.
“AKI is a severe and complex condition that presents in many hospitalized patients, yet it is often diagnosed too late, resulting in significant kidney injury with no effective treatments to reverse damage and restore kidney function,” said David Ledbetter, chief clinical officer of Dascena. “If we are able to predict AKI onset earlier, physicians may be able to intervene sooner, reducing the damaging effects.”
“These findings with Previse are exciting and further demonstrate the role we believe machine learning algorithms can play in disease prediction. Further, with Breakthrough Device designation from the FDA, we hope to continue to efficiently advance Previse through clinical studies so that we may be able to positively impact as many patients as possible through earlier detection,” Ledbetter added.
The study was conducted to evaluate the ability of Previse to predict for Stage 2 or 3 AKI, as defined by KDIGO guidelines, compared to XGBoost and SOFA. Using convolutional neural networks and patient electronic health record data, 12,347 patient encounters were analyzed, and measurements included “area under the receiver operating characteristic” (AUROC) curve, positive predictive value (PPV), and a battery of additional performance metrics for advanced prediction of AKI onset.
Findings from the study demonstrated that on a holdout test set, the algorithm attained an AUROC of 0.86, compared to 0.65 and 0.70 for XGBoost and SOFA, respectively. As for the PPV, it scored 0.24, relative to a cohort AKI prevalence of 7.62 percent, for long-horizon AKI prediction at a 48-hour window prior to onset.
Both TXR-1208 and TXR-1210 also showed significant decreases in myofibroblast activation, significant decrease in infiltrating T cells and excellent tolerability by body weight.
