10/10: Big Data In Medicine

[smiklin]

Comment below with questions or thoughts about the reading's for this week's workshop. See here for guidelines on what kind of content to post. Please make your comments by Wednesday 11:59 PM, and upvote at least five of your peers' comments on Thursday.

This week, there are two readings, one of them is a rough draft of considerable length—don't be discouraged and give it a go!

[zeyuxu1997]

Thanks for presentation. It seems that one of the applications of big data to medical area is to assist physicians to make a diagnosis now. I wonder if we will have a reliable system for self-diagnosis that is understandable for most people in the predictable future, and what's the major difficulties to realize that except for those mentioned in the article.

[ShuyanHuang]

Thanks for presenting. I am wondering how the development of wearable devices can contribute to medical big data. For example, Apple Watch can monitor people’s EKG, which may be sent to doctors and provides abundant information about people’s long-term heart activity pattern.

[liu431]

Thanks for the discussion. Digital technologies for medicine could also eliminate the gap of high-quality medicine resources between developed and developing regions. How should we quantify this technology adoption factor to study economic changes? With the evolution of Big Data, how should today's medicine schools adapt to train future doctors?

[Leahjl]

Thanks for presentation.New technologies have been increasingly implemented to improve on the capabilities of previously established systems and surgical ergonomics. Is there any evidence show that a robotic surgery has better medical outcome than a traditional one? What's the future of robotic surgical systems?

[wanitchayap]

Thank you for your presentation in advance! Greene & Lea (2019) mentioned "entrusting life-and-death matters to unseen algorithm". I have 2 specific questions about this quote:

1. How do you envision the field of medical legal situation to change base on the change in digital-human landscape in medical practice. Are there already any changes currently undergoing in the law?
2. Berkeley J. Dietvorst, Prof. at Booth, demonstrates that people have algorithm aversion because they prefer uncertainty from human decision more than the predictability/stability of algorithm. Essentially, people prefer to have hope for an extreme favorable outcome, even with the probability for extreme unfavorable outcome from human decision, rather than a predictable moderate outcome. However, his works primarily explore the domain of economic. How do you think this concept apply to the same algorithm aversion in medicine. Could we use his model in economic realm to facilitate the communication between medicinal personnel and general public?

[KenChenCompEcon]

Thanks for presenting! I am particularly excited about the progress that we've made to enhance computerized diagnostics, such as the digital computer IBM devised... In the meantime, I also want to know your view on how these algorithmic technologies can be encompassed into doctors' diagnosing in the long term? It is tending to be more of a displacing factor or complementary to their current work? Thanks

[ruixili]

Thanks for presenting! In the past, due to the large imbalance in education level, there is a huge asymmetry in information between doctors and patients, and doctors may use this asymmetry to manipulate the information they delivered to patients thus make patients pay more than they need. As mentioned in Greene & Lea (2019), the usage of "electronic diagnosis" can easily solve this problem. However, correctly diagnose disease and give treatment is of great importance especially for severe patients. How should we handle the precision of the big data approach and reduce misdiagnosis in practice?

[YuxinNg]

Thanks for the presentation. I have a few questions:

1. It seems that big data is or will be really helpful to assist physicians to make a diagnosis. I am wondering how this new trend will impact the training process of physicians. That is, will there be more data related or computer science related courses in Medical School? Will the future medical students learn less about professional medical knowledge (because machine can do that for them!)

2. Though digital technology may be more advanced or more accurate. I think there are still many people out there who would believe human diagnosis more than digital diagnosis. Do you think this will be an obstacle for the further development of big data in medicine.

[HaowenShang]

Thanks for your presentation!

In the paper ‘The Long Arc of Big Data in Medicine’, you introduce the history of using big data for implementing digital medicine. That’s a really interesting topic. Since medical data has many sensitive information for patients, such as identification information, how do you handle these sensitive informations? Especially for handling the big data, could you share your experience with how to deal with the ethical problems?

[YanjieZhou]

Thanks for the presentation. Technologies of machine learning have made great progress now compared with the 20th century. So I am wondering how well machine learning can be applied to digital diagnoses nowadays. It seems that the combination of machine learning and human diagnoses can serve better by dividing the work into seperate parts: basic and easier parts for computers, and more complex problems for doctors. Does it work in the current medical systems or is it a possible way of improving diagnostic efficiency in the near future?

[romanticmonkey]

Thank you for the presentation! I am impressed by your efforts to put together the struggle medical digitalization had gone through. Your writings provide great insights into developing new methods in a traditional field, which helps (and warns) whoever ambitious with their unprecedented design to be careful with the potential obstacles in their way.

In addition, I find telling these stories to the public audience will raise more awareness in accepting and bolstering creativity, especially accepting "computer service" into our daily life. Speaking of such, I hope to know what you think of developing digitalized services in areas that are still predominantly occupied by manual labor nowadays. For example, psychiatry and psychotherapy. Although survey-type screenings for depression, etc. exist in the current medical system. Do you think there will be more space for digitalized diagnosis in psychotherapy? As digitalized physical health diagnosis becomes much more accepted, do you think the same progress can be applied to psychological health diagnosis? Furthermore, do you think we should have psychological health screenings added to government-funded annual health visits?

[yutianlai]

Thanks for the presentation! Now I get an idea of digital technology application in medical area and I want to further explore the medical-SPECIFIC problems you might encounter comparing with other fields, which might include ethical problems such as how to store and manage sensitive information of patients, and the change of treatment plans accordingly since medical field is not only about technology, but also about how physicians interact with and care for the patients.

[jsgenan]

Thank you for the great history lookback! You talked about automation efforts in diagnostic algorithms, electronic medical records, and medical informatics. Would you please talk more about information systems in pharmaceuticals? That's where I believe the business application of computational systems might be most useful in the healthcare industry.

[sunying2018]

Thanks for your presentation. It is an amazing development progress of automatic diagnosis you mentioned in the articles. At the meantime, I have two specific questions:

1. There are a lot of doctors make their diagnosis based on their experience and past cases. I am wondering if the diagnosis made by doctors conflicts with the automatic diagnosis, how should we deal with these conflicts? How doctors choose between their experience and machine?
2. Automatic diagnosis needs the access to patients' personal and private information. How can we obtain permits of patients to use these data?

[ydeng117]

Thanks for the detailed presentation of the history of the digital health industry. There are two concerns I have towards the future development of automatic health testing:

1. As digital health testing data become more accessible to individuals, will people tend to make unprofessional self-diagnose themselves rather than turning to doctors for more accurate and professional information? If it is the case, should we limited this technology only to the hospitals and not open to the public?
2. When I read about the enthusiasm of using the automated health clinic, I can help thinking about the fraud of Theranos' Edison blood testing. How should we protect the integrity of this technology from opportunists who utilize our optimism and passion for digital medical testing to hoax us?

[nswxin]

Thank you for the presentation! I am curious about whether there exist some “intrinsic” features of medical records collected digitally which differ from their counterparts. I mean, there are certain types of medical records that are very difficult to collect, or even impossible to obtain. For example, there are medical records from the less socioeconomically- developed area where technology hasn’t reached to keep them, as well as medical records from a different diagnostic system, which are based on experience rather than scientific reasoning and data evaluation (traditional Chinese medication). Thus, at least to me, the available digital database for medicine may not be entirely representative of the entire domain of human knowledge and experience in medicine. Thus, is it possible that with the development of technology accompanied by increasing emphasis on big data analysis, only approaches whose records can be kept digitally would be valued and studied in the future, while the rest being gradually forgotten? Then, we, human beings, on the one hand, are moving away from our dream of equal, and, one the other hand, are blundering away potentials of medical developments.

[vinsonyz]

Thank you for your presentation! It is good for us to know the development of the modern medical technologies.While the prediction for a whole population is feasible, predicting for a single subject is extremely hard. Since big data render us the opportunities for automatic diagnosis, however, the type 2 error is unavoidable when we run a model. My questions are: Knowing that the type 2 error is tremendously severe in the medical diagnosis, how can we strive to eliminate it in diagnosis? In the reality, how accurate the diagnosis could be? And which is more accurate, the doctors or the computers? How accurate it can be with a combination of these two?

[di-Tong]

Thank you for presenting! The papers elegantly delineate the ways big data and automation facilitate the development in health care. Aside from learning about how the human-machine complementarity plays out in the field of medicine, I'd like to also hear a bit about skill-substitutability in this field under the context of computational revolution. What is the impact of digitalization and automation on the workforce of the medical industry? What kind of changes in demand and payoff to different skills and jobs have taken place in the digitalization and automation transformation of medical care?

[SixueLiu96]

Thanks for your presentation in advance! It's always thrilled to know that the development of technology and science could actually help improving people's lives. Computer revolution happens in medical area. With the advanced progress in computation skills, the physicians' ability could be used in a more producible way in medical field. My question is: in terms of advanced development in computer decisions, what do you think the future of physical physicians? In the future, the development of which areas will be more suitable for human beings, and which areas will be replaced by artificial intelligence?

[chiayunc]

Thank you in advance for your presentation. I have 2 questions pertaining to the impact of big data application in medicine to the marginalized group in society and does such application help patient empowerment.

1. Research on the application of big data in medicine focuses almost entirely on high-end technology development such as wearable devices and AI diagnosis, or chronic diseases resulting from refined lifestyles. The involvement of big data in medicine does not necessarily makes health care more accessible or guarantee a better health care system for “everyone.” How do you see lower-income and marginalized individuals be impacted by a data analysis driven health care system?

2. One of the biggest selling points of data application in medicine is the possibility of self-care and patient-empowerment. With the growing complexity of data analysis, do you think it is somewhat an oxymoron in that the patient is not really empowered? Rather, it is only a power shift from doctors to data scientists?

[timqzhang]

Thank you for the presentation. It is quite a valuable paper since nowadays more and more industries are introducing big data and relevant methodologies, while it also raises much more concerns. I think this paper is quite objective to present the evolution of big data implementation in the medical industry and how people change their views from time to time. Here are my two questions:

1. How do you envision the ideal relationship between human (especially physicians, doctors, etc.) and digital system in medicine? Will the digital system serve more as a partner of human or still as a supplementary role in the diagnose considering so many worries on it? The answer will decide how advanced the technology we should develop in the future.

2. Suppose there is a consensus from the society on the implementation of such digital system, and as a normal person, I understand it may cost a lot at the beginning, but then I will focus more on when it could cover the majority of people to make almost everyone afford such service?

[bakerwho]

I would like to explore how the model of data analysis for distributive medical care would likely reach the people.

Would this be a state-owned technology, or otherwise? Developments on medical technology seem to currently be both publicly and privately funded. Which sphere is more rapidly expanding into AI for healthcare?

[goldengua]

Thanks for presenting. I feel it could sometimes be hard to get access to data as a social science student, while you are dealing with health records. I was wondering how people in your area find the balance between accessibility of data and privacy of participants? A seperate question would be, with the development of electronic, portable medical device, how could medical practice be changed in the future?

[nwrim]

Thank you for the presentation! My question is related to the computational (or electrical) diagnosis and the factor of gender and race. Many researches in medical anthropology and sociology suggest that diagnosis of disease is heavily biased by the patient's gender and race (e.g. Good, Good & Baker, 2003). Some people claim that by making computers assist in the diagnosis process, this problem can be partially resolved. Others claim that algorithms used in such diagnosis will be also biased - through the algorithm itself or by the "big data" inputted to the algorithm - and computational analysis will not resolve this problem. (As an indirect example, Professor M'Charek's RaceFaceID project suggest that biometric identification systems "rely on and reiterate racial ways of understanding differences") What side do you generally agree on in this matter? How can we assure that computational analysis will not replicate the bias?

[skanthan95]

Thanks for the discussion! I was concerned about the ethical implications of cases that are misclassified by algorithms trained to ease the workload on medical professionals. How can false negative cases be arbitrated and how will the industry deal with legal suits of this kind?

[Yilun0221]

Thanks for the presentation and I really enjoy reading the papers! I do believe that big data can play an important role in medicine in the future. However, I wonder what should we do in protecting the privacy of the patients with the growing power of data?

[tonofshell]

This was a very comprehensive read on the history of computing in healthcare. What I found most interesting was how early computational methods were being leveraged in the field. However, it seems that methods as basic as all-digital records have been filtering into the mainstream within just the past decade. What do you think have been some of the hurdles to the adoption of simple technologies (like all-digital records) into the majority of doctor's offices, and will these be the same for the adoption of future computational healthcare technologies?

[dongchengecon]

Thanks a lot for the presentation! The history of the interactions between diagnostic algorithms, electronic medical records, and medical informatics would absolutely be an interesting topic. Apart from the application in clinical practice, big data has also had great influence on health policy issues. Could you also give us an overview about the interactions of health data and policy? There are many ongoing research in this field. For example, the effect of subsidizing health insurance for low income adults, healthcare spending in public and private medicine.

[chun-hu]

Thanks for sharing these wonderful articles! As we experience more developments in the digitizing society, ethical issues are never absent from the context. Ethical boundaries are especially present in the medical field because of its nature to include human interactions. Robotic surgery, which allows the doctor to perform more precise procedures, already helps to complete complicated tasks and operations in many areas. However, how should we define the ethical responsibility and accountability of these technologies and algorithms? The same issue also applies to the use of digital records and automated diagnosis. How can we define and resolve this issue considering the increasingly immense social power of technology?

[boyangqu]

Thanks so much for presenting. The application of big data in the medical area is definitely a promising field, and it can be in many areas as diagnostic and treatments. I wonder that how do researchers handle privacy issue in using patients' data as training dataset (consent?), and also what is the current situation in these applications in pratice?

[anqi-hu]

Thank you for the presentation in advance. In the presence of rapidly evolving technologies, related ethical challenges arise. What are some measures, if any, taken to ensure stakeholders' responsibilities in protecting individuals using such technologies from potential harms? Several nationwide initiatives (e.g. Connected and Open Research Ethics, etc.) are in progress in an effort to address ethical, legal and social implications within this "gray area" of digital healthcare. What is your view on the essentiality of putting forth regulations in this regard?

[ChivLiu]

Thanks for the presentation! The development of electronic medicine was innovative with new algorithms at every stage. I found the data from medical records is big since it contains the information of a large population, and using washed data from a large scale could help to find some patterns and create regression models. It would be easier to train learning models when having data from various backgrounds, but I wonder if adding some social data of the patients, such as work locations and networking groups, would help to identify the patients' behavior rules. As well, Could we create clusters such as patients from similar family backgrounds, patients from close neighborhoods, patients having similar sleep qualities, to further monitor the infection rate of certain diseases or forecast the health status of individuals?

[fulinguo]

Thanks very much for your presentation. I was wondering what is the relationship between the tradition medical theories and the knowledge provided by the machine. Does the access to big data allow us to test medical theories, or do the big data help us build new medical knowledge? How we deal with the situations when the outcomes provided by the data appear in contradiction to our knowledge of the theories?

[lulululugagaga]

Thank you for presenting, this is a new and meaningful field that I have never explored before. I'm interested in how the role of doctors may change due to the impact brought by diagnostic algorithms, electronic medical records, and medical informatics, or how might the medical system develop in terms of the refreshed human-to-machine relations?

[MegicLF]

Thank you for your work and presentation. I believe that data science will be applied in more and more clinical practices and medical researches. And I think the introduction of the automated clinic in both materials is quite interesting. Although it failed the last century in the U.S., this concept achieved great success in Japan. Do you think the automated clinics will somehow rejuvenate in the United States? If so, in what forms? Moreover, consider the case of electronic records, what ethical concerns do people (e.g. patients and specialists) have towards the application of big data in medical practice? And how can we deal with those issues?

[wu-yt]

Thanks for your presentation! As there have been concerns about how algorithms or big data can be biased and negatively influence the results, how will these issues influence the medical information? Will rare diseases become harder or even falsely detected?

[yongfeilu]

Thanks for the presentation! After reading and getting familiarized with the general development of the electronic medical system, is it possible that we use the big data to isolate the effects of the electronic medical system on the time medical students spend on their education? Does the advent of electronic medical system increase the education load of the medical students by releasing them from easily-tractable diseases but meanwhile forcing them to concentrate on difficult diseases? If so, how do patients benefit from such a process in terms of their medical treatment cost? Or maybe is it true that the promising technology induces people to spend more on medical care? Is there a way we can use the big data to analyze such effects?

[bjcliang-uchi]

Thanks for the presentation. You mentioned the massive changes due to the digitalization of medical records. But who defines the standard (format, content, structure, etc) for record digitalization? If there is no such a universal standard, how do physicians and engineers work on disorganized data? If there is, Is there any tension (or politics) envolved in setting and adopting such standards locally, federally, and internationally?

[rkcatipon]

I have two questions:

1. What would you say are the implications of the privatization of hospitals and healthcare in both accessing healthcare data and in implementing standard healthcare models across the industry?

2. In the Long Arc of Big Data in Medicine you mentioned that some healthcare professionals felt that the focus on data-driven decision making did not improve patient care. What can model creators do to privilege the right metrics/variables so as to aid hospital functions in practice? Can this process be generalized broadly for those interested in applying large-scale data models outside of healthcare?

[policyglot]

I'd like to build on @skanthan95's critical question. In her book, ‘Weapons of Math Destruction', Harvard mathematician Cathy O'Neil writes about the dangers of opaque algorithms to human lives. According to your draft, the aim of Big Data in medicine is to diagnose and prevent diseases. But the underlying algorithms may become opaque too. How would the average doctor (without training in computer science) be expected to interpret the following for herself and her patients: a) The principal components derived from high dimensional Big Data b) The inner workings of neural network models Not everyone can be Morris Collen. Does the progress of technology in medicine so far imply that the years of medical training will now require possibly more years to understand quantitative analysis?

PS- Thank you for this comprehensive and highly readable history.

[WMhYang]

Thank you very much for your presentation in advance. I have two questions.

1. The paper talks about how big data in medicine could be used by in clinics, hospitals, and universities. I wonder whether it is possible that part of big data could be used by the general public to prevent diseases or to self-diagnose the diseases in order to save medical costs and hospital resources. If so, how to limit the degree of usage by the public given that some medical data can be sensitive?

2. Some physicians seem to rely more heavily on different and complicated examinations to make their final diagnoses, but the examinations can be costly for patients. Could we make use of big data to reduce certain examinations and reach the same or similar diagnoses?

[AllisonXiong]

Thanks for presenting! The development of digital medicine is inspiring for all areas in which data science has getting involved, especially the fact that despite technical improvement, some fundamental problems remains unsolved. What interests me most is the section of electronic diagnosis and I have two related questions:

1. As mentioned in the article, one major obstacle for effective electronic diagnosis system is the medical heterogeneity. That is, the lack of “ground truths” for certain problems even among medical experts would lead to the lack of standards for machines. However, what if we take an opposite perspective, that instead of presetting standards of diagnosis, we let machines with super-human computing capability and access to big database set the standards for unsettled medical problems? Like determinant symptoms and other independent variables and their statistical weight in diagnosis.

2. One basic requirement for digital diagnosis assistant is to distinguish between the monotonous mechanizable process and the complex, human interaction part. However, human relies heavily on heuristics to make decisions. Heuristics, in some cases, are proven to be more reliable than statistical models. These “instincts” or heuristics may be based on details/facts during the human-interaction process that is hard to be quantified or mechanized, and thus hard to be captured by machines. Do you think heuristics is important for correct diagnosis and how would digital medicine solve such problem?

[SoyBison]

Thank you for coming to present to our workshop! I'd like to pose a question about the possible consequences of techniques like the "big data" systems you describe being employed in medicine. It's clear that the diagnoses which such a system would be working with are highly individualized, as health ultimately is, and such a predictive system is always working with statistical constructs. That being said, when you change perspectives from the "bird's eye" to the position of the patient's point of view you can get into situations where the patient can be worse off due a system not considering the possibility that the literature could be wrong or in active debate. Take for instance the case of post-lyme disease syndrome, in which about 10% of lyme disease sufferers report musculo-skeletal pain chronically for years after being ostensibly cured of the disease. How do you think these sorts of dangers can be mitigated?

[anuraag94]

Thank you for the presentation in advance.

The Long Arc of Big Data in medicine concludes by opening the door on an interesting question: "what happens when human intelligence cannot comprehend the pathways of computer decisions?" This is a question of AI safety - when machine objectives don't fully capture our own real objectives, the risk is that incentives may come to catastrophically conflict with what we want, which will pose problems even if an intelligent machine can explain its decision-making inputs and process.

What does AI incentive alignment look like for medical diagnoses? What kinds of sociological diagnostic biases do we want to address? Which ones (if any) are more advantageous to retain in the long run? By treating diagnostic algorithms as black boxes, can human agents classify outputs (biased diagnoses) to understand the landscape of computer-mediated medical bias?

When it comes to the role of the physician in interacting with diagnostic algorithms, what will be the relationship between interpreting diagnoses and iteratively tuning AI incentive alignments? Handling the tradeoff between more incentive-aligned yet more inscrutable AI diagnoses?

[luyingjiang]

Thanks for your presentation! I am very interested in the diagnostic algorithms part and curious about the underlying mechanism of processing of big data.

Here are some questions: What factors are being considered when deducing the diagnoses? Are things like social background, mental state, living habit taken into consideration? Furthermore, as mentioned, datas are often biased and there are lot of disagreement between the medical experts and the computer. How is the standard of accuracy defined in this case? Is there a way to minimize the disparity?

[weijiexu-charlie]

Thank you for presenting. My question is about the application of big data on rare or even new diseases? As we know, the biological world is by no means static: viruses and bacteria may evolve very rapidly, leading to new diseases we've never faced before. Would it be risky if our physicians heavily rely on the automatic diagnosis driven by big data in such cases? Or is it possible that big data could help us to predict the emergence of new diseases?

[Yiqing-Zh]

Thank you for your presentation. I was so impressed that people started exploring digital medicine so many years ago. I have two related questions. 1. So far computer diagnoses are based on previous knowledge or records and are programmed by human. Does it mean the increasing accuracy of computer diagnoses must be based on the progress of medical science? Or can computer technology itself facilitate medical science by finding something new based on existing knowledge? 2. Physicians' roles would encounter some changes and become ambiguous as computer technology and algorithms become more and more advanced. Actually, despite the role of an expert treating patients, there also exist emotional links between patients and physicians. Will the absence of such links bring any negative impacts on treatment?

[sanittawan]

I found your 2019 article in the NEJM to be extremely illuminating. When I hear talks about big data and medicine, the conversations are usually very forward-looking. As you have shown in the article, it is equally important to look back and see where we've come from. Thank you very much for that. Based on this article and your research project, The Electronic Patient, what do you think about how easily accessible "big data," both scholarly research papers and mass websites, has changed the relationship between doctors and patients? In what ways are the changes positive/negative? Do you think technologically enhanced methods of communication with physicians like Teladoc have made care more accessible and affordable?

[lyl010]

Thanks for your presentation! It is amazing to see this meandering history of machine and medicine in 20th century: people could be so optimistic about the medical diagnosis, machine recording and stuff like this though we had comparatively less information about machines at that time. Fancy as these dreams were, some of them have come true nowadays. Compared to the past light mood towards technology, people are seemingly much more conservative nowadays---we all know machines can not replace physicians. But I think dreamers are important for the development of technology and knowledge. So is there still seemingly 'impossible' dream about machine and medicine that researchers are still trying to make it come true currently? And what are the most impressive communication and computation technology in the medical field for the moment? Thank you again!

[SiyuanPengMike]

Thanks a lot for your informative paper. Your paper, again, raises my curiosity about the importance of interpretable results by machine learning. As we all know, one of the biggest advantages of ML is that it could find some latent relationship in the data which is quite hard for humans to find. Even though it might be hard for us to explain, it still has some potential values. Considering that medical data is a great source to check the posterior probability， I'm wondering that is there such research to check the accuracy of those unsupervised ML learning models. Or to compare the accuracy between the doctor's judge and the machine. If it turns out that those 'non-sense' ML model's accuracy is even higher than experts, could we change our focus to find a reasonable explanation for those ML models? Or could we just let it go and simply use the ML model without explanation?