The objective of this K01 application is to give Dr. Hofer the necessary training and research experience to establish himself as an independent investigator focused on using machine learning (ML) on a variety of healthcare data to predict outcomes during the perioperative period. The career development activities consist of escalating coursework on machine learning beginning with an online course of ML fundamentals and ending with a UCLA course on ML applications in healthcare. Augmenting these courses are tutorials on the application of these techniques to healthcare data and a research program is designed to use ML on healthcare data to predict perioperative cardio-respiratory instability (CRI) ? specifically hypotension and arrhythmia. To achieve these goals, Dr. Hofer has established an outstanding team of leaders in machine learning, perioperative medicine, and clinical informatics. Dr. Maxime Cannesson, his primary mentor, is an expert in perioperative medicine and the use of ML on physiologic signals. Dr. Eran Halperin, the co-mentor for this pro- posal, is an expert in ML and its application to genomic and other healthcare data. Dr. Hofer has ongoing collab- orations with Drs. Cannesson and Halperin on joint projects. Both Drs. Cannesson and Halperin have a strong track record of mentoring individuals who have progressed to independent and productive academic careers. Dr. Hofer will be aided by an advisory committee consisting of Dr. Douglas Bell (who will provide guidance on integrating data from multiple sources), Dr. Mohammed Mahbouba (providing support regarding data security and creating enterprise level analytic solutions) and Dr. Jeanine Wiener-Kronish (providing guidance on the most relevant questions in perioperative outcome prediction). Challenges managing CRI have been implicated in the more than 15 million annual postoperative com- plications, costing more than $165 billion, however no scores exist to predict CRI. This study will leverage unique infrastructure at UCLA where whole EMR data has been combined with physiologic waveforms and genomic data on more than 30,000 patients. This proposal will use a variety of ML techniques on these data to create predictive models for CRI. In summary, this proposal will provide Dr. Hofer with both technical training in ML and hands on experi- ence in using ML to predict perioperative outcomes. This study has the potential to create models that will help clinicians predict, and thus avoid, perioperative instability, thereby improving patient outcomes. Additionally, this program will provide Dr. Hofer with the tools he needs to successfully compete for a R01 focusing on using ML models on a variety of healthcare data to predict the downstream effects of CRI ? perioperative complications.
During the perioperative period hemodynamic instability is the norm rather than the exception; the ability of clinicians to manage this instability has been associated with postoperative complications affecting more than 15 million Americans and costing more than $165 billion each year. There are no current risk scores that predict instability in real time, however the recent widespread adoption of electronic medical records and creation of genomic biobanks creates unique opportunities to develop scores that are both accurate and precise. We will use machine learning techniques to create risk scores for perioperative cardiorespiratory instability ? specifically hypotension and arrhythmia ? using combined electronic medical record, genomic and physiologic waveform data.
