Background:

Vision loss from diabetic retinopathy remains the leading cause of preventable blindness in working-aged adults in the United States (US).¹ Advanced diabetic retinopathy is referred to as proliferative diabetic retinopathy (PDR).² In many patients, blindness associated with PDR can be prevented with appropriate and timely diagnosis and treatment.³ In order to prevent PDR and improve visual outcomes, we need to better risk-stratify and follow patients who are at higher risk.

Machine Learning Based Analysis of Allele-Specific Expression Genes in Localized Prostate Cancer Patients 

Hanbing Song, Margaret Tsui, Sarah Hsu, Franklin W. Huang

The goal of our project is to enhance identification of the earliest safe extubation time in the PICU with a clinical decision support tool based on a machine learning model to predict when to extubate patients. We hypothesize that employing artificial intelligence could safely shorten extubation times by identifying subsets of patients for earlier extubation. Embedding a state-of-the-art predictive model into the EHR could augment real-time decision-making and better standardize care, resulting in more ventilator-free days, fewer ventilator-associated adverse events, and shorter length of stay.

The clinical problem: functional, cognitive, and mental health outcomes are common in ICU patients and after critical illness.

Background 

Sepsis, a multiorgan syndrome induced by infection, is a major public health concern, accounting for more than $20 billion (5.2%) of total US hospital costs¹. From 2005 to 2018, 6.7% of decedents had a diagnosis of sepsis². Over the past decade, sepsis-related mortality rates remained stable in both males (57 deaths per 100,000) and females (45.1 deaths per 100,000). 

The Clinical Problem: Heart failure (HF) is among the most common reasons for hospitalization and is a major cause of morbidity and mortality. Yet its most common symptoms, dyspnea and edema, are non-specific, which means incident cases rarely present directly to cardiologists. Thus, the initial diagnosis is challenging. Though echocardiography, which requires highly trained providers, can identify hallmarks of HF, it cannot make what is ultimately a clinical diagnosis.

In collaboration with individuals from the Bakar Institute, Center for Digital Health Innovation, and the Department of Medicine Informatics Core, the UCSF Center for Colitis and Crohn’s Disease has developed a suite of tools to better characterize the patients receiving care at our center. These include: