Fig. 5

Integrated analysis of EHR data using AI and introduction of its application to diagnosis and treatment. A Figure showing a series of processes to predict the risk of preventable ACU after chemotherapy initiation using the machine learning algorithm trained on comprehensive EHR data, modified from Ref. [151]. Nine machine learning models were developed, validated, and compared to predict ACU at 3, 6, and 12 months after chemotherapy initiation in patients presenting to an oncology clinic affiliated with a large academic cancer center. Patient-reported outcomes were also incorporated to assess the impact of these data in predicting the risk of preventable ACU. B Schematic of a lung cancer prognostic study using a clinical cohort constructed from EHR data, modified from Ref. [158]. Initially, data are obtained from the EHR and lung cancer diagnosis codes are used as filters. After creating a data mart containing structured data and narrative notes, the structured data are queried and the narrative notes are processed using NLP tools. A phenotyping algorithm has been developed using a combination of structured data and narrative notes to extract variables of interest. The performance of the phenotyping algorithm is compared to a random sample of patients selected for EHR review. The performance of the phenotyping algorithm is compared to a random sample of patients selected for EHR review. The accuracy of the extracted variables is compared to the EHR reviewed sample and the Boston Lung Cancer Study cohort data