Bioinformatics & Perioperative Analytics

The Division of Bioinformatics and Perioperative Analytics is committed to leveraging technology to improve delivery of patient care throughout the acute care period, conducting and supporting innovative research, and cultivating a community that is skilled in interpreting and utilizing healthcare data.

“Improving patient care bit by bit.”


We are involved in many short-term and long-term projects in medical informatics. Our group leads investigations in the following areas:

  1. Data extraction, validation, and reporting
  2. Predictive analytics and risk stratification
  3. Research in data-driven quality improvement

We are always interested in partnering with others in the use of “big data” to pursue scientific knowledge.  

Data extraction, validation, and reporting: The Perioperative Data Warehouse

The foundation of much of the work in the division is based on the creation and maintenance of the Perioperative Data Warehouse (PDW), a custom-designed data warehouse that enables extraction of data from EPIC’s “Clarity” database into a format that facilitates its use in research and quality improvement.  The PDW is currently in its third version, and the division is actively engaged in a multi-center project to demonstrate and better understand the application of this architecture at other institutions that use the EPIC EMR.

In addition to publishing work on the creation of the PDW, the Division continues to use the PDW as the source of data for retrospective and prospective trials based on large data sets. These have included multi-center collaborations of perioperative data, as well as single-center, hypothesis-driven examinations of perioperative questions.

Data Validity

Historically, administrative data (ICD codes or DRGs) have been used to classify patients for large retrospective reviews. Multiple previous studies have demonstrated that these data are highly specific but can lack sensitivity. We want to understand the validity of data extracted from various sources, especially the EMR, and use data from multiple sources to yield more accurate information.

Using a technique that we call “triangulation”, we have developed a series of algorithms to automate patient phenotyping and outcome calculation using structured data from the EMR. These techniques have been successfully used to populate over 600 discrete measures in the PDW. The group has been able to automate the extraction of the entire anesthesia component of the Society of Thoracic Surgeons (STS) cardiac surgery registry, and has further validated one component of the registry (postoperative ventilator requirement) to demonstrate increased accuracy of automated review compared to manual review.

Currently, our team is working on applying these techniques prospectively to improve the accuracy of administrative data in the medical record. After successfully developing algorithms to detect diseases, we are working with colleagues in our Preoperative Evaluation and Planning Clinic (PEPC) on a pilot project for risk-based screening of patients before elective surgery. Detecting preexisting diseases early in the admission process, we hope, will improve documentation by clinical teams, and facilitate both accurate coding and optimized care.

Predictive Analytics and Risk Stratification

We intend to use data from the EMR to develop improved risk models to identify those patients at greatest risk for postoperative complications and delayed return to function after surgery. We hope the data can also lead to earlier detection of patient decompensation after surgery. Toward this end, the group has developed several partnerships with colleagues in computer science to leverage deep learning and machine learning models. Our group includes engineers and computer scientists with advanced expertise in machine learning. We have used a “deep neural network” to predict in-hospital mortality based on intraoperative patient data. The abstract of this work was awarded “Best in Show” at the 2017 Society for Technology in Anesthesia conference.

In the future, the group hopes to create risk models that successfully predict event-free hospitalization after surgery or patient decompensation after surgery. We hope to utilize these tools in prospective studies to demonstrate that machine-learning technologies can help improve the efficiency of perioperative care.

Clinical Decision Support

We are especially interested in developing and testing clinical decision support tools, and in the process of integrating information into care workflows, to help clinicians deliver better care. This includes both real-time decision making and the reinforcing effect of physician feedback reports. Our division works directly with the EPIC build team, in Los Angeles and at the headquarters in Wisconsin. This work has been integral to developing perioperative care pathways, such as a protocol to reduce PONV. The next step is underway to create tools to enrich clinicians’ knowledge about patients in the preoperative phase.


In addition to these research projects, the division serves as a resource for all faculty requiring data for their research questions. We are participating in several ongoing prospective projects and data collection efforts with the UCLA Anderson School of School of Management and UCLA School of Engineering.