Machine Learning Models Can Predict Urgent Care Eeds For Patients With Non–small Cell Lung Cancer

A caller study published successful JCO Clinical Cancer Informatics demonstrates that instrumentality learning models incorporating patient-reported outcomes and wearable sensor information tin foretell which patients pinch non–small compartment lung crab are astir astatine consequence of needing urgent attraction during treatment. The study was led by researchers and clinicians astatine Moffitt Cancer Center.

Patients undergoing systemic therapy for non-small compartment lung crab often acquisition treatment-related toxicities that tin consequence successful unplanned urgent attraction visits. In this study, Moffitt researchers tested whether integrating aggregate sources of patient-generated wellness information including self-reported quality-of-life surveys and wearable instrumentality metrics specified arsenic slumber and bosom complaint could amended predictions beyond modular objective and demographic information. 

The squad utilized explainable instrumentality learning approaches called Bayesian Networks to build predictive models among 58 patients monitored pinch Fitbit devices and surveyed done a questionnaire. Machine learning models that included patient-reported outcomes and wearable sensor information importantly outperformed models based connected objective information unsocial connected expertise to separate betwixt high-risk and low-risk patients. 

By combining accusation patients supply astir their symptoms pinch continuous monitoring from wearable devices, we tin amended place who is astir astatine consequence for curen complications. Our extremity is to springiness clinicians devices to intervene earlier, amended diligent experiences and perchance forestall hospitalizations." 

Brian D. Gonzalez, Ph.D., lead writer and interrogator successful Moffitt's Department of Health Outcomes and Behavior

The findings propose that integrating multidimensional information into instrumentality learning models whitethorn heighten personalized crab attraction and let providers to proactively reside toxicities earlier they escalate. While nan study was constricted to a azygous halfway and a humble sample size, researchers opportunity nan attack holds committedness for broader application. 

"What makes this attack powerful is not only nan accuracy of nan predictions, but besides nan expertise to understand why nan exemplary reaches those predictions," said Yi Luo, Ph.D., co-author and interrogator successful Moffitt's Department of Machine Learning. "By utilizing explainable instrumentality learning methods, we tin spot really factors for illustration denotation reports, slumber value and laboratory results interact to power risk. This transparency is captious for building spot pinch clinicians and ensuring that nan models tin beryllium utilized to guideline existent world decisions successful crab care." 

Future investigation will grow nan models to see further objective and molecular data, arsenic good arsenic validate results successful larger, multi-institutional cohorts. 

This study was supported by nan National Institutes of Health (P30 CA076292).