Development of Electronic Health Record–Based Prediction Models for 30-Day Readmission Risk Among Patients Hospitalized for Acute Myocardial Infarction

Abstract

Key Points Question Can machine learning deployed in electronic health records be used to improve readmission risk estimation for patients following acute myocardial infarction? Findings In this cohort study examining externally validated machine learning risk models for 30-day readmission of 10 187 patients following hospitalization for acute myocardial infarction, good discrimination performance was noted at the development site, but the best discrimination did not result in the best calibration. External validation yielded significant declines in discrimination and calibration. Meaning The findings of this study highlight that robust calibration assessments are a necessary complement to discrimination when machine learning models are used to predict post–acute myocardial infarction readmission; challenges with data availability across sites, even in the presence of a common data model, limit external validation performance.

Information Extraction From Electronic Health Records to Predict Readmission Following Acute Myocardial Infarction: Does Natural Language Processing Using Clinical Notes Improve Prediction of Readmission?

“…More important, metrics presented in Table 4 can vary based on the threshold cutoff value, which offers flexibility to optimize the desired metric. 6 …”

“…Even though OMOP was developed at the external site, DHMC, based on the primary site, VUMC, using standardized variable definitions and code sharing, there were differences in EHR mappings that limited the availability of data at DHMC. 6 …”

“…Cohort derivation and inclusion criteria are described elsewhere. 6 Briefly, patients hospitalized at VUMC with a primary diagnosis of AMI and discharged alive between January 1, 2007, and December 31, 2016, were included in the cohort (n=6165). A separate cohort of hospitalized patients discharged between March 2, 2011, and December 31, 2016, from DHMC was included as the validation cohort.…”

“… 4 , 5 Prior prediction models experienced limited discrimination, with area under the curve values ranging from 0.63 to 0.75, did not leverage the wealth of patient data generated by an AMI hospitalization, and lack portability because of interhospital data collection variability. 6 , 7 , 8 , 9 In addition, social risk factors are rarely included in readmission prediction models, despite their established role in health outcomes and hospital readmissions. 5 , 10 , 11 , 12 , 13 …”

Development of a complete blood count with differential—based prediction model for in-hospital mortality among patients with acute myocardial infarction in the coronary care unit

“…Since machine learning algorithmic models can include more variables and produce more flexible relationships between predictors and outcomes, they have shown significant value in risk model development. For patients with AMI, state-of-the-art machine learning models have steadily improved the discrimination performance of risk stratification ( 3 , 4 , 37 ).…”