The importance of interdisciplinary collaboration in AI projects
Machine learning/AI capability is increasingly transforming how we engage with technology. Just look at how mainstream digital voice assistants such as Alexa and Siri, customer-service chat bots and bank-fraud detection tools have become.
I see promise in healthcare using machine learning/AI technology. One example is using AI to detect breast cancer by analyzing mammograms. Leveraging the expertise of subject matter experts in healthcare is crucial to successfully applying AI to solving healthcare challenges.
U.S. employers seeking healthcare data-scientist talent recognize the importance of working in interdisciplinary teams. As I work with Allscripts data scientists, it’s clear they seek and value the experience and expertise that clinicians bring to solving healthcare problems using AI.
A deep dive into the world of data scientists
I have the privilege of being able to contribute to a new textbook, Introduction to Biomedical Data Science, which will be published later this year. In it I highlight the importance of interdisciplinary collaboration in data science. Below is an excerpt from the book.
Challenges in healthcare are, of course, complex and require a multidisciplinary effort that often includes experts with clinical, workflow, patient safety, process improvement and IT knowledge. Also necessary are those who understand where healthcare IT fits into driving more efficient, safer care and better patient outcomes.
Problems that involve healthcare data are no exception
The data scientist can benefit from close collaboration with non-data scientists to help understand the question(s) being asked and the problem(s) that need to be solved. For example, non-data scientists can help define cohorts of interest; clarify diagnostic and therapeutic terms as well as provide context; and define relevant workflows (e.g. CPOE, clinical documentation) and data points important to capture within those workflows.
Non-data scientists can also provide feedback about the analysis from a clinician and informatics perspective and help determine whether that insight is useful clinically. Translating the output from a predictive or prescriptive model that is easy to understand by clinicians and patients is also crucial.
While a binary classification model may be helpful in some instances, physicians often use probability in clinical decision making and during discussion with patients. Non-data scientists can also help to identify emerging forms of data outside of the healthcare system that might be useful for analysis, such as the internet of things, social media, personal devices, publicly available data sets or social determinants.
Operationalizing predictive and prescriptive analytics models are essential to improving healthcare outcomes for patients.
Health informatics professionals — uniquely positioned at the intersection of the healthcare system, clinical care and health information technology — are particularly important in this translational step. Operationalizing may require review by various stakeholders, such as a Clinical Decision Support Committee, Institutional Review Board and others.
Clinicians involved need to understand the model’s potential benefits and limitations and, to the extent possible, how the model works (e.g. what model features most and least strongly contribute to the model’s output). While delivering the model at the point of care in easily understandable prose is important, the use of data visualization may also be helpful.
Finally, ongoing monitoring after model implementation is important to determine if it has made a favorable impact on patient health.
To learn more about population health analytics, go here.