Machine Learning: The Future of Healthcare

By: Puneet Gupta

The U.S. healthcare system is a mess. Both the system’s infrastructure, such as the role of insurance companies, and its clinical aspects, such as how care is provided, are lacking in multiple ways. Though improvements in the infrastructure are necessary, this article will primarily discuss and suggest changes to the clinical side of the healthcare system. A new movement to bring about change in private practices, hospitals, and other healthcare facilities revolves around one new innovative field of science and technology: machine learning (ML).

An Overview

Machine learning, in simple terms, focuses on developing algorithms and software based off of the machine’s past experiences. A program capable of machine learning is able to perform a certain task or improve how it performs a task through previous runs and without any additional changes in the software. In the fewest terms, machine learning is the extraction of knowledge from data.

Machine learning is split into three primary categories: supervised learning, unsupervised learning, and reinforcement learning. In supervised learning, a ML model is given data that has been labeled with a certain outcome, and then learns the relationship between both (data and outcome) to make predictions regarding the outcome for future data. In unsupervised learning, a ML model is given data that has not been labeled with an outcome, so it is able to sort and separate the data into groups of its choice, unlike supervised learning, which has certain outcomes or groups that the data must fit into. In reinforced learning, the model attempts to figure out the most effective way of achieving the highest ‘reward’ through choosing different sets of actions. In other words, the system is rewarded when it achieves a certain outcome, and it tries to determine the best way of achieving the highest reward (1). Overall, machine learning models attempt to adopt principles based on how humans innately learn and involves building systems that can ‘think’ and adapt themselves.

Machine Learning in Healthcare

In earlier decades, when walking into a healthcare setting, patients could see stacks of papers, piles of manila folders, and clutters of pens and pencils all over. Despite all the new advances in technology, at the turn of the millennium, offices and clinics are still filled with inefficient workspaces. In order to implement change, to transition into electronic health records, and to generally improve healthcare technology, the government issued the Health Information Technology for Economic and Clinical Health Act (HITECH) in 2009 (2). Though progress has been made in getting many healthcare systems to bring in new information technology (IT), there is still much room for innovation to be made to improve all aspects of patient care, including safety, patient experience, efficiency, and effectiveness. With the overall quality of care in the U.S. lacking in comparison to those of other countries, the demand for change has increased, with more people seeing machine learning as the solution. ML is currently being used in healthcare, but not to its full potential and capabilities, nor is it being applied to the extent that it is used in other industries, such as finance, where it has brought major positive changes and a variety of benefits.

ML’s primary use in the near future will involve data analysis. With each patient comes large bulks of data including X-ray results, vaccinations, blood samples, vital signs, DNA sequences, current medications, other past medical history, and much more. However, we still are not able to efficiently obtain, analyze, and reach conclusions well. One of the major challenges is integrating the data obtained for each patient into one system, as that will allow for efficient communication between providers, allow for rapid data analysis, and give providers all the information they need to accurately treat their patients. However, much of the data today is encrypted and has restricted access due to the constant efforts to protect patient privacy, making this transition difficult, alongside the fact that many medical devices are not interoperable (3). Once a single database can be established, the benefits of ML can be reaped.

One of the primary applications to healthcare for machine learning involves patient diagnosis and treatment. It is important not only in emergency medical situations, but also in general primary care and in specialized physicians as well. For example, ML can be used to predict mortality and length of life remaining using physiological patient vitals and other tools including blood test results, either in the immediate future, such as for a traumatic car accident, or in the long-run, such as for cancer (3). Most significantly, ML models can be used to help physicians diagnose patients, especially in cases involving relatively rare diseases or when outcomes are hard to predict. For example, in a recent clinical study, several machine learning models were used to analyze data from electronic health records to predict heart failures, and the outcomes indicated that these ML systems predicted outcomes well (4). Moreover, machine learning can be used to determine the most effective medication dosage, reducing healthcare costs for the patients and providers. ML can be used not only in determining dosage, but also in determining the best medication for the patient. Genetic variations among different races, ethnicities, and individual people in general impacts the effectiveness of certain drugs and people’s response to these drugs, such as HIV medications (3). Once more advanced ML algorithms and models are developed, they would be able to rapidly recognize these differences and reach accurate and reliable conclusions. Some technologies are being used currently for interpreting a variety of images, including those from magnetic resonance imaging (MRI), X-rays, and computed tomography (CT) scans.5 However, more advanced ML algorithms that can effectively identify potential regions of concern on these images and then develop possible hypotheses are needed. Even in surgery, new machine learning models need to be developed for robotic surgeries to increase the probability of successful surgical outcomes, which can potentially eradicate the need for human surgeons (6).

Many issues involving erroneous and imprecise data arise in data collection, as much data is simply wrong (3). This is especially true in waveform data, where environmental factors and patient movement can affect the recorded signals. New and advanced algorithms need to be established that can distinguish real data from artificial and poor data, thereby improving the reliability of the data gathered and allowing the physician to make an accurate diagnosis. Even in very common electrocardiogram readings, many physicians reach different conclusions in regards to the patient’s condition. Artificial data and data with poor signal quality play a major role in this analytical difference.3 Many times, physicians are overwhelmed by the plethora of data collected, but ML algorithms that can identify and streamline the most pertinent data without leaving behind other crucial information need to be developed. Moreover, ML algorithms that can allow the AI to explain the reasoning behind its proposed diagnosis or treatment plan is necessary.

Challenges and Controversies

Adapting artificial intelligence (AI) and machine learning into all healthcare systems is unfortunately not easy. Healthcare systems have been structured so that change is difficult. Much of the decision makers in healthcare systems and policies are elderly, who tend to have strong preferences for the typical ‘pen and paper’ and prefer simpler systems in which they have more control. ML systems are complex and need to be integrated into health care systems in the simplest yet most effective form. Moreover, many healthcare facilities are not motivated or incentivized enough to spend their budget in investing in adequate research, staff, and other support for developing these ML models. This adaptation of AI and ML is necessary not just in the United States health care system, but all across the world. However, as the U.S. is one of the leading places for innovation and development in this health information sector, the country needs to bring about a large-scale change in its system first, despite the difficulties in installing such a system, in order to start a ripple effect.

As with the rise of most new technologies, machine learning brings about a heated debate on ethics. When we train machines to ‘think for themselves,’ we have given up our control over them in that we don’t know what the system learned or what it is thinking, thereby putting our lives in danger. Some believe that our advancements in machine learning will reach a point at which we no longer need human physicians, which would significantly hurt the economy, workforce, and patient experience in clinics. Many are afraid that when they come into a doctor’s office, they will no longer have that physician-patient contact and connection, but instead must confront a machine. When building and training machine learning systems, access to large databases of patient information is needed, raising privacy concerns, for which there is still no accepted standard in regards to AI. Furthermore, advances in ML can lead to issues regarding insurance coverage. For example, some insurance companies may start demanding access to the AI that is tracking a patient’s health records to see how their overall health is and determine premiums based off that. Moreover, it is possible that when future research studies show the success of ML and AI, hospitals and clinics might increase the fees associated with these services, leading to inequality based off income. How will we react if the AI gives us wrong treatment or diagnoses? What if a physician’s diagnosis and an AI’s diagnosis are different? It is important to consider all these challenges as we further develop and improve our machine learning systems.

Future of Machine Learning

Today, many major companies and startups, including Enlitic, MedAware, and Google, have launched massive projects focused on improving AI and ML and bringing it to the healthcare system, such as Google’s DeepMind Health project and IBM’s Avicenna software (7). Moreover, IBM’s Watson Health is collaborating with the Cleveland Clinic and Atrius Health in using cognitive computing in their health system, from which experts are hoping to see reduced physician burnout (8). More recently, current ML algorithms being tested and developed include k-nearest neighbors, naive and semi-naive Bayes, lookahead feature construction, Backpropagation neural networks, and more (9).

Artificial intelligence and machine learning are undoubtedly the future, as refined automation of data collection and replacement of jobs in all industries by machine learning systems is inevitable. Scientists and researchers must focus on developing effective, efficient, and innovative algorithms while ensuring that their functions and models do not endanger the human job market. Both Elon Musk and Stephen Hawkings foresee AI and ML not only dangerous economically, but also physically (10). Nonetheless, it is imperative that we continue to work on transforming the quality of care and healthcare system as a whole through machine learning, a science and technology that is to revolutionize the world in all aspects of life for decades to come. The benefits of machine learning outweigh these theoretical nightmares.


Puneet Gupta ’18 is a junior in Dudley House concentrating in Biology.


Works cited

[1] Introduction to Reinforcement Learning. Salsa/rl.html (accessed Feb. 25, 2017).

[2] Rhodes, R. et al. Amer. J. of Hospice and Pall. Med. 2015, 33, 678-683.

[3] Johnson, A. E. W. et al. Proc. IEEE 2016, 104, 444-466.

[4] Wu, J. et al. Med. Care 2010, 48, S106-S113.

[5] Wang, S.; Summers, R. M. Medical Image Analysis. 2012, 16, 933–951.

[6] Nevett, J.; Waghorn, M. Robots ‘set to replace human surgeons entirely for complex operations’ potentially cutting risk of errors. Mirror, May 4, 2016. http:// (accessed March 3, 2017).

[7] Simonite, T. IBM’s Automated Radiologist Can Read Images and Medical Records. MIT Technology Review, Feb. 4, 2016. https://www.technologyreview. com/s/600706/ibms-automated-radiologist-can-read-images-and-medical-records/ (accessed Feb. 27, 2017).

[8] Slabodkin, G. Caradigm takes endto-end enterprise approach to population health. Health Data Management, March 2, 2017. (accessed March 3, 2017).

[9] Kononenko, I. Artif. Intell. Med. 2001, 23, 89-109. [10] Cellan-Jones, R. Stephen Hawking warns artificial intelligence could end mankind. BBC News, Dec. 2, 2014. http:// (accessed March 1, 2017).

Categories: Spring 2017

Tagged as: , ,

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s