Interest in machine-learning applications within medicine has been growing, but few studies have progressed to deployment in patient care. We present a framework, context and ultimately guidelines for accelerating the translation of machine-learning-based interventions in health care. To be successful, translation will require a team of engaged stakeholders and a systematic process from beginning (problem formulation) to end (widespread deployment).
Many complexities exist in developing and deploying effective ML systems across domains as diverse as health care, self-driving cars and space exploration. We propose a robust overarching framework, including people and processes, within which many issues stemming from the complexity of adopting ML in practice, especially in health care, can be successfully avoided. We encourage all researchers working on deploying ML systems to consider the broader context early in the process.
There are many unanswered questions remaining in the field of ML: how much accuracy is sufficient for deployment? What level of model transparency is required? Do we understand when the model outputs are likely to be unreliable and therefore should not be trusted?
Although there is still a long way to go, impactful innovation with ML in health care is now clearly a team activity.
Ultimately, meaningful progress will require both system-wide changes driven by policy-makers and health-system leaders, and individual researchers working on bottom-up solutions. From understanding the data to deploying the model, to have the greatest impact, developers must work as well-informed, interdisciplinary teams including health-system leaders, frontline providers and patients.
As the field hurtles forward, it is worth pausing to remember the Hippocratic oath: “first, do no harm.” It is imperative that all stakeholders work together to understand and appropriately address the nuances and potential biases lurking in health data, before deploying solutions. Doing so will not only decrease the potential for unintended consequences but also reduce rather than amplify existing social inequalities and ultimately lead to better care.