OSLER inHealth Advisory Board Minutes
PCORI OSLER inHealth Advisory Board Meeting
Tuesday, August 16, 2016
Vincent Carey, Yates Coley, Mary Cooke, Francesca Dominici, Patrick Heagerty, William Lewis, Martin Morgan, Desi Stringer, Zhenke Wu, Peter Zandi, Scott Zeger, Risha Zuckerman
Dr. Scott Zeger welcomed everyone to the meeting.
Dr. Zeger gave an overview of the project and the Individualized Health Initiative, including recent accomplishments and items in progress.
Prostate Cancer Active Surveillance Tools
Dr. Yates Coley shared the goals of the active surveillance project before sharing the current work, which is to design and deploy a decision support tool for clinical practice, evaluate the impact of the support tool, and continue to refine the software package. She then shared anticipated work for the multi-site implementation of the model. Dr. William Lewis, a patient stakeholder, asked if/when a larger context will be brought into the decision support tool. Dr. Zeger shared that the team has taken the feedback to understand and include the full context and pathways inherent in decision-making, and have mapped out all major decisions. Now the team needs to fill out this map with data and expert opinions.
Diagnosing Childhood Pneumonia
Dr. Zhenke Wu shared the goals of the childhood pneumonia project, the current data, and the methods on diagnosis of childhood pneumonia. Dr. Wu introduced the Bayesian Analysis Kit for Etiology Research (baker) user interface and detailed the statistical methods used. He also shared how the model is structured to use data from all other cases the doctor has seen when diagnosing one child. Since the last meeting, an adjustment has been made to the model to include the dependence of health state distribution upon covariates (season, rain, age, etc).
Tracking Depression and Schizophrenia
Dr. Peter Zandi detailed the analysis that is being done on schizophrenia data since the team just received the depression data from the NNDC. Dr. Zandi shared how because there are no biomarkers or measures in depression, we just have clinical symptoms with the goal of inferring from those symptoms if there are subgroups that are more similar than others. He shared information on the statistical model and gave examples of predictions the team aims to do, including predicting the trajectory of symptoms and disease state. The overall goal is to feed this data into a clinical support tool for the clinician and patient to use together.
Dr. Vince Carey shared how the team is thinking through how to take individual software packages and turn them into a repository. They want to attract statisticians and clinicians to use the models. The team is going to think about the quality of the data going in and how to cope with incomplete data. Dr. Carey shared the team’s short range targets, which include expressive formalism for dynamic models of latent health state, data schema for dynamically treated/observed cohorts, and to connect the schema provisions to the model requirements. He questioned how to bridge the short and the long range goals now that the prostate cancer surveillance methodology will be published. This will involve exploring the ability of non-programmers to use baker’s shiny interface and the deployment of the prostate cancer modeling tools in the Epic EMR at Hopkins.
Feedback was provided throughout the meeting but more formal feedback was given from each member of the Board in a roundtable fashion facilitated by Dr. Francesca Dominici.
Dr. Heagerty encouraged the team to think about how to do a better job benchmarking the individual we’re predicting for relative to the data we’re using as it is an important part of disclosure to individual. Benchmarks should be developed with respect to input measures in the model. Uncertainty can be represented through intrinsic outcome modeling. Dr. Coley shared that the team is building benchmarking into a dashboard for clinicians.
Dr. Dominici expressed that the work was very ambitious and impressive in terms of impact, research, and software. She encouraged the team to think about how the information will go back to the patient. What type of treatment strategy has been more effective for that patient? To what degree should the team push the research to provide information to estimate a personalized treatment strategy?
Dr. Morgan urged the team to develop explicit software best practices as each case study develops and to have a formal discussion about how good the software is and how to make it robust for clinical trials. Dr. Zeger expressed that developing a method for curating the software is important, and there are talks about putting in another PCORI grant to do that.