Machine Learning in Healthcare

-
 Ghassemi (2020) presents in a TED.com video how machine learning (ML) algorithms and statistics have been used in healthcare processes, but the goal of predicting disease and employing artificial intelligence (AI) in medical decision-making, has not yet been reached.
Ghassemi explains the data used now to train ML algorithms comes from practice and knowledge. Practice data refers to clinical records in hospitals and clinics, treatments offered, including how patients interact with their doctor. Training data derived from knowledge is the kind that is generated in medical trials, articles, and textbooks. The main issue with this algorithm training data is that is generated by doctors who often burn out after providing patient care for long hours each day, resulting in the introduction of bias in the records they generate. 

Ghassemi states 35% of doctors report burn out since they not only provide patient care, but several other different tasks. From those reporting burn out, 56% feel they do not have time to be empathetic towards their patients, and even though ethics training is a must in the medical field, biases seep through and are reflected in the type of health care they provide, and may be associated with racial and ethical disparities, bias against women, weight bias against patients with obesity, etc.

Ghassemi (2020) explains randomized controlled trials (RCTs) are by far the largest generators of medical knowledge. These are events where a group of people are given a treatment and medical teams observe and record how well they do with that treatment. Unfortunately, RTCs are expensive and too often the data generated from them is just wrong. Ghassemi states more than 10% of over 3,000 top journal studies are "medical reversals." This means the treatment or guidance previously deemed appropriate, is no longer.

The current state of affairs presents a provider searching through similar cases when a patient needs medical care, but even if records exist for similar patients, the same treatment may not work on someone else because as humans, each one of us is unique, and so is the treatment (Ghassemi, 2020). Ghassemi explains the current practice data and knowledge available to train ML algorithms is highly biased and cannot be used, but suggests the application of the same technology to improve such data
by generating complex models that feed from hard coded information derived from mobile data, social networking, medical test results, genomic and environmental data, internet usage, etc., to train algorithms without the effects of human bias. Predictions of type of care are ethical and social in nature as they will depend on the patient's gender, race, etc., and in using properly trained ML algorithms, the focus will shift towards actionable insights in patient health and keeping them out of the hospital all together (Ghassemi, 2020).

Ghassemi's (2020) TED.com video in ML algorithms to enhance healthcare faces social, economical, cultural, and other concerns. The Commonwealth Fund Survey from 2002, identified healthcare inequalities between White and minority populations in the U.S. that persist to this day. Issues like patient-physician communication, language barriers, feelings of disrespect, health status and more chronic conditions, low rates of insurance coverage, and less access to healthcare, stem from the different racial and ethnic groups. This aligns with Ghassemi's (2020) statements of bias being present in ML algorithm training data as minority Americans are reported to earn lower incomes, for example, and may not afford to procure insurance coverage. Social and cultural concerns may include culturally appropriate healthcare policy to increase the quality level of care for Americans speaking a language other than English.



References

The Commonwealth Fund. (2002, March 6). Minority Americans lag behind Whites on nearly every measure of health care quality. https://www.commonwealthfund.org/press-release/2002/minority-americans-lag-behind-whites-nearly-every-measure-health-care-quality

Ghassemi, M. (2020). How machine learning enhances healthcare [Video]. TED.com. https://www.ted.com/talks/marzyeh_ghassemi_how_machine_learning_enhances_healthcare

Comments

Post a Comment

Popular posts from this blog

Think Tanks

-
Image
  Lewis (2019) describes think tanks, also known as think factories, idea factories, reflection pools, or policy institutes, as organizations where scholars representing different disciplines assemble to think without any restrictions, conduct research associated with specific issues, ideas, and policies (Aly, 2019; Whittenhauer, n.d.).   Think tanks conduct research, generate ideas, debate issues, and monitor public policy by means of implementing various strategies, which are associated with the business model utilized by the think tank, such as consulting, advocacy, research, and contract efforts (Lewis). There are four types of think tanks as follows: (1) Ideological or advocacy tanks are organizations where problem solving is rooted on an ideological philosophy, (2) Specialist tanks are institutes with a specific focus or theme, such as the environment or foreign policy, (3) Subnational tanks are government-related entities that typically work on specific state policies, ...
Read more

The Rise of Data Scientists

-
Image
  Data science has been used in physics and astronomy, but it is being applied to businesses and finance, manufacturing, and other fields. The demand for data science professionals exploded about two decades ago when businesses began collecting massive amounts of data,  and a shortage of talented individuals, or data scientists, with a specific set of skill sets able to extract the most benefit from the available data (World Data Science, 2022). Data science is a multi-disciplinary field that converges programming, mathematics, statistics, computer science domain expertise, and communication skills, yet seldom organizations are able to find a data scientist that excels at all of them (Bansal, 2021; World Data Science, 2022).  The World Data Science Initiative (2022) reports data science jobs were expected to grow by 364,000 in 2020, but the real number came in at 2,720,000. Data scientists became part of the solution to address the challenges of growing organizational da...
Read more

Big Data Analytics in the Banking Industry

-
Image
  The implementation of big data analytics (BDA) in organizations is a costly and complex undertaking, and given there is no recipe that guarantees a successful implementation and execution, a careful and honest assessment of different aspects of the business must be conducted, from the technology, data collection, the human resources involved in the process, and managerial support (Hajiheydari et al., 2021). Hajiheydari et al. state that approximately 60% to 85% of BDA projects fail not because the technology is bad, but because organizations do not have a clear understanding of how to utilize data analytics to construct achievable business goals,  or how to successfully integrate its practice with day-to-day business operations (Sun et al., 2020). BDA offers financial institutions countless benefits, and provides valuable insights to personalize customer service via the analysis of transactional records and other data, including social media information (Flynn, 2020; Sun et ...
Read more