Posts classified under: Machine Learning

Enhancing Lung Cancer Screening

Lung cancer is the leading cause of cancer-related mortality in the United States and globally. A key issue has been that lung cancer diagnosis often occurs quite late and in its advanced stages, resulting in poor outcomes. Within the past decade, landmark randomized clinical trials including the National Lung Screening Trial (NLST) and more recent NELSON study, have demonstrated the efficacy of low-dose computed tomography (LDCT) for detecting lung cancer sufficiently early. Consequently, new lung cancer screening programs have launched, providing ongoing surveillance of individuals deemed at high risk for developing this disease. However, these programs face numerous pragmatic challenges, ranging from low enrollment and adherence to ongoing surveillance given a lack of patient and provider understanding about lung cancer and screening, through to costs and limited resources that need to be optimally utilized.

Alongside UCLA’s lung cancer screening experts, the Center for SMART Health is helping set the stage to improve screening by exploring several different strategies, including outreach through online patient portals and working with community partners to develop strategies that encourage enrollment and continuance with recommended imaging follow-up; developing novel ways to conduct risk stratification of at-risk individuals to optimize screening frequency; and seeking new ways to provide better patient engagement for cancer screening.

Predicting Chronic Kidney Disease

In partnership with UCLA’s Faculty Practice Group (FPG), we are looking to better identify individuals within our healthcare enterprise who are at risk for worsening kidney function. Chronic kidney disease (CKD) affects a significant number of Americans, and if left untreated results in the need for dialysis and/or kidney transplant. Proper clinical management of individuals early in the course of the disease can sometimes recover impaired kidney function, if not slow or stabilize disease progression. Unfortunately, a number of patients suffer from rapid decline of kidney function, and it is presently unclear who these individuals will be. As part of the UCLA CTSI’s Clinical Pathways program, we are using machine learning (ML) methods on EHR-derived data to better stratify UCLA patients by identifying early patterns indicative of CKD patients who may be at risk of rapid decline. By finding these patterns in a timely manner, more targeted preventive interventions can be put in place to help the patient and prevent disease progression.

Predicting Future Hospitalizations

Some patient visits, either to the emergency room (ER) or in-patient admissions, could be prevented through timely outpatient interventions by healthcare providers. By identifying such situations early, we can not only improve a patient’s end outcomes and quality of life, but significantly reduce healthcare resource utilization and the high costs associated with such visits. Working with UCLA’s Faculty Practice Group (FPG), we have developed new predictive models from the electronic health record (EHR) to identify those patients who are at high risk for admission within the next year. Based on this risk prediction, UCLA healthcare providers and staff reach out periodically to check on the individual and provide guidance and monitoring to prevent subsequent healthcare issues that will result in the need for admission. Being deployed across UCLA’s primary care clinics, this new risk assessment tool is being consider for broader adoption across the entire UC Health system.

Improving Colon Cancer Screening

Colorectal cancer (CRC) is the second most common cause of cancer-related mortality, killing over 50,000 Americans each year – a number that is unwarranted as it many of these cases are preventable via screening and surveillance of high-risk abnormalities. For instance, many people with high-risk polyps that are detected during initial colonoscopies fail to follow-up in a timely manner.  Often such findings are not fully documented and/or issues exist with inconsistencies in reporting, making it challenging to definitively know which patients need to come back and when. Working with investigators in the Division of Digestive Diseases, the Center for SMART Health is helping to identify these individuals at UCLA through automated natural language processing (NLP) tools, looking to merge colonoscopy and pathology report observations in the electronic health record (EHR) into a comprehensive picture that can then be used to guide follow-up recommendations.

Remote Monitoring of COVID-19 Symptoms

With increased COVID-19 testing and the detection of more positive cases (COVID+), many “at-risk” individuals who appear asymptomatic or only with mild symptoms are being told to self-quarantine at home. Unfortunately, there are situations where a sudden worsening of the disease occurs and symptoms progress quickly, usually within 24-48 hours. This rapid health decline often goes unreported until too late. More proactive surveillance of these COVID+ individuals could provide an alert to healthcare providers to initiate earlier intervention and save lives.

Our project adapts our current mHealth framework to the immediate needs of the COVID-19 crisis. Unlike other mHealth efforts that are using apps for contact tracing, self-reporting of symptoms, or new sensors for detecting disease onset, we focus on using mHealth to enable (safe) remote monitoring of known COVID+ patients. We originally created the Sensing for At-Risk Patients (SARP) platform to automatically collect information on older, frail patients with different conditions. SARP uses a smartwatch and different sensors to capture this information, automatically and securely sending the data to UCLA Health for review by healthcare staff. We are extending SARP by adding several FDA-approved devices (thermometer, pulse oximeter, respiratory monitor) to augment its present capabilities. SARP will be used to gather objective observations for at-home COVID+ cases, with the data reviewed by healthcare providers to assess patient health.

Capturing this data will further elucidate the myriad presentation of symptoms in COVID+ patients in a natural environment, helping establish a better understanding of this disease and the evolving pandemic. As more insight is gained about these patients and the presentation of the disease, we will explore the application of machine learning methods on this data to detect early patterns of worsening symptoms, leading to potential clinical decision support tools for physicians.

Detecting Cognitive Changes

Can we automatically detect changes in speech to detect cognitive changes? The Center for SMART Health is helping to develop and test digital technologies for cognitive testing in the context of benzo/z-drug tapering/insomnia treatment. Specifically, we are exploring the use of digital vocal biomarkers for assessing changes in cognition. This study is recruiting participants who will answer web-based tests with verbal responses. The voice data from their responses are collected and analyzed using machine learning algorithms, identifying potential patterns reflecting cognitive changes. We are also partnering with industry leaders to perform neurocognitive testing to provide real-time results in different areas, including: instant and delayed verbal memory, impulse control, attention, focus, emotion, identification, information processing speed, flexible thinking, working memory, executive function, visual learning, and spatial memory.

Understanding Depression & Anxiety

Depression affects an astounding 300 million people worldwide and one of society’s biggest problems – yet it is poorly understood. In 2018, UCLA launched the Depression Grand Challenge (DGC) with the ambitious goal of cutting in half the worldwide burden of depression by 2050. New diagnostic and treatment methods are being explored, including mHealth. The DGC, in a partnership with Apple Inc., launched a major study to provide greater insights into how we diagnose and treat anxiety and depression. Using devices connected to a smartphone/smartwatch, we are obtaining objective measures around sleep, physical activity, heartrate and daily routines to illuminate the relationship between these factors and symptoms of depression and anxiety. This joint effort has the very real potential to transform behavioral health research and clinical care by enabling healthcare providers to note warning signs before severe mental health crises occur. The study is also an important step toward greater understanding of the different types of depression, and which treatments work best for each. As part of this effort, we are helping with subject recruitment, as well as coordinating secure sensor data collection and integrated analysis.