Machine Learning In Healthcare

Application of Advanced Computational Techniques to Improve Healthcare

Introduction

About one trillion gigabytes of health care data is generated annually in the U.S. health care system.[i] This, of all statistics, sets healthcare apart from several other industries where advanced computational techniques can be applied. Artificial Intelligence (AI) modalities such as machine learning have been piloted and subsequently implemented. Machine learning, an important subset if AI, is the application of algorithms and statistical models to perform specific tasks and provide inferences without using explicit instructions or human input, relying on and learning from patterns within the data itself.[ii]

A key aspect of machine learning, especially relevant to healthcare is its ability to make predictions and drive decision making. While the application of machine learning has been widely prevalent in other industries, its application within the realm of healthcare is quickly expanding across the health care value chain (i.e., insurance companies, health care providers, pharmaceutical companies, etc.).[iii] Moreover, recent advances in machine learning have largely focused on specific aspects of a patients care journey, such as diagnostics and post-acute care optimization. That said, there also have been setbacks and opinions that machine learning has limited potential if / when healthcare is defined more broadly (e.g., population health management, global health, etc.)

Nevertheless, the overwhelming consensus seems to be that unlocking the full potential of machine learning will span a long-time horizon, but once realized, the potential opportunities could be substantial. This paper’s objective is to spotlight both areas within healthcare where machine learning has and/or is driving substantial impact and some of its limitations.

Findings

Machine learning has been driving substantial impact on several aspects of health care. This fact can be clearly seen / measured based on the number of new health care start-ups that have machine learning at their core. Four main areas where machine learning has made / is making meaningful contributions are hospital operations, medical imaging, drug discovery and development, and disease prediction and treatment.[iv],[v] Below, using the aforementioned categories, I have summarized how / where machine learning is playing a pivotal role in impacting health care.

• Hospital operations: Several upstarts, such as Quotient Health, are aiming to reduce the total cost of health care by reducing the cost of EMR systems – underpinning their value proposition is using machine learning to optimize and standardize system design.v Other areas where machine learning is being applied is in designing and managing clinical workflows to enable seamless information transfer within and across health systems.
• Medical imaging: A highly prevalent use case for machine learning has been medical imaging and diagnostics wherein with the help of advanced statistical models and particularly pattern recognition, companies are improving the speed and accuracy of medical diagnoses.[vi]
• Drug discovery and development: Another area where machine learning has had / is having substantial impact is pharmaceuticals, particularly drug development. Companies such as Pfizer have employed IBM Watson’s AI tool(s) to accelerate drug development across key stages in a drug’s development pipeline.v
• Disease prediction and treatment: Finally, in the realm of disease prediction, several start-ups are leveraging machine learning to tap into massive patient registries and extract key points of information (e.g., disease detection timeframe; gaps in care; therapy requirements, etc.) that could eventually have a substantial impact of defining clinical care pathways over time, and potentially enable precision medicine.[vii]

Limitations

Adoption of machine learning may not yet be widespread in all spheres of healthcare, but its application is inevitable. Therefore, it is important to consider its limitations and some barriers for its adoption.[viii] First, despite its potential, there aren’t many incentives for hospitals or physician groups to adopt this powerful method and in general there is skepticism within the medical community when a new technology is introduced. Second, the output of machine learning is only as good as its input. Currently, the fragmentation of the electronic medical record data, privacy concerns, and interoperability issues will likely hinder machine learning’s full potential. Lastly, overdependence on technology with minimal human input may potentially increase the risk of safety concerns. Algorithms may likely generate spurious interpretationsvii and it is important to validate them periodically. This needs to be considered when deciding what aspects of the health care delivery system this needs to be embedded into.

Conclusion

Machine learning, like several other things in healthcare that have been viewed with skepticism of late, such as value-based care, could be transformational for healthcare “at large”. Its transformative power is largely rooted in the fact that it has the potential to impact all health care stakeholders, most importantly to patients. Notwithstanding the prevailing skepticism surrounding machine learning, the fact that it has overwhelming support from key stakeholders such as insurance companies, drug manufacturers, researchers, etc. potentially means that it is here to stay.

[i] Travis May. The fragmentation of health data. Medium. July 31, 2018. Available at: https://medium.com/datavant/the-fragmentation-of-health-data-8fa708109e13

[ii] Machine Learning. What it is and why it matters. Available at: https://www.sas.com/en_us/insights/analytics/machine-learning.html#machine-learning-users

[iii] David Champagne, Sastry Chilukuri, Martha Imprialou, Saif Rathore, and Jordan VanLare. Machine learning and therapeutics 2.0: Avoiding hype, realizing potential. December, 2018. Mckinsey and Company. Pharmacuetical and Medical Products. December, 2018. Available at: https://www.mckinsey.com/industries/pharmaceuticals-and-medical-products/our-insights/machine-learning-and-therapeutics-2-0-avoiding-hype-realizing-potential

[iv] AI And Healthcare: A Giant Opportunity. Forbes. February 11, 2019. Available at: https://www.forbes.com/sites/insights-intelai/2019/02/11/ai-and-healthcare-a-giant-opportunity/#2c5cfb8e4c68

[v] Mike Thomas. Ultra-Modern Medicine: Example of Machine Learning in Healthcare. June 2, 2019. Available at: https://builtin.com/artificial-intelligence/machine-learning-healthcare

[vi] Erickson BJ, Korfiatis P, Akkus Z, Kline TL. Machine learning for medical imaging. Radiographics. 2017 Feb 17;37(2):505-15.

[vii] Obermeyer Z, Emanuel EJ. Predicting the future—big data, machine learning, and clinical medicine. The New England journal of medicine. 2016 Sep 29;375(13):1216.

[viii] Murdoch TB, Detsky AS. The inevitable application of big data to health care. Jama. 2013 Apr 3;309(13):1351-2.