Development and validation of an interpretable machine learning scoring tool for estimating time to emergency readmissions

A novel interpretable machine learning system to generate clinical risk scores: An application for predicting early mortality or unplanned readmission in a retrospective cohort study

“…The inclusion of age in Model 2 is not surprising, as old age generally associates with alleviated risk of adverse clinical outcomes, but the clinical basis of the use of sodium level may warrant further investigation. The focus on metastatic cancer and renal disease among the 17 comorbidities is consistent with a recent study of time to emergency readmission using a similar cohort to that in our study [ 22 ], which developed a 6-variable risk score using number of ED visits in previous year, age, cancer history, renal disease history and inpatient measures of creatinine and album. Three variables in Model 1B (i.e., number of ED visits in past 6 months, inpatient LOS and metastatic cancer) were also included in the LACE index.…”

A universal AutoScore framework to develop interpretable scoring systems for predicting common types of clinical outcomes

“…Using AutoScore, users can easily generate data-driven clinical scores while concomitantly incorporating clinical expertise and practical considerations. 22 , 23 , 24 , 25 , 26 Besides binary outcomes, 1 AutoScore has been methodologically extended to survival outcomes, 2 unbalanced binary data 27 and ordinal outcomes. 3 The modularized structure allows AutoScore to be integrated with more advanced interpretable machine learning methods (e.g., the Shapley variable importance cloud 28 ) for improved robustness, interpretability and transparency in the risk score development.…”

“…Moreover, it enables users to build transparent and interpretable clinical scores quickly in a straightforward manner. It has been extensively used in different clinical applications, e.g., for general risk assessments in the emergency department, 22 , 23 , 35 , 36 and for prediction of disease-specific outcomes in specific patient cohorts. 24 , 25 , 26 , 37 , 38 , 39 , 40 , 41 …”

Artificial Intelligence: its Future and Impact on Acute Medicine

“…The applications of AI in this domain hold immense possibilities, with notable focus areas including triage, providing suggestions for diagnostic workup, supporting clinical decisionmaking concerning medications and interventions, and offering accurate prognoses of disease progression or outcomes 6 . Over the last few years, we have seen the first examples of such tools being developed and tested within research environments, such as a risk prediction model to support ambulance transport decisions 7 , an AI triage tool in the emergency department (ED) 8 , a screening tool for early sepsis detection 9 , a blood culture stewardship tool 10 , a machine learning tool for predicting ciprofloxacin resistance 11 , an AI audit system to minimize prescription errors 12 , a tool for estimating time to emergency readmissions 13 , and a model to predict admission to the neuro intensive care unit directly from the ED 14 . Furthermore, there is potential in harnessing AI tools before patients reach the ED or an acute care setting, leveraging data from devices and smartwatches for remote patient monitoring, alerting in a timely manner, and triage 15 .…”