Big Data Analytics and Machine Learning
in EHRs
Nandini Bansal1
1Department of Engineering and Technology, Apeejay Stya University
Sohna – Palwal Road, Sohna – 122103
Gurgaon, Haryana, India
1nandini.bansal@asu.apeejay.edu, 1agilist.nandini@gmail.com
Abstract— Big Data Analytics is expanding rapidly to all the spheres of human life, including healthcare where it has immense potential when combined with Machine Learning. Whether this potential can be exploited depends on the methods and technologies available to process and use Big Data. In the recent years, it has provided tools to accumulate, manage, analyze, and assimilate large volumes of structured, and unstructured data produced by healthcare systems.
Big data analytics has been recently applied towards aiding the process of care delivery and disease exploration. In this paper, we discuss the applications of Big Data Analytics with Machine Learning on EHR data for the greater good of the society.
A common platform where medical history of patients can be stored and shared securely within various medical platforms for research purpose along with helping doctors in typical cases when diagnosis of disease becomes difficult due to varied symptoms is also discussed. Potential challenges in achieving all above are discussed too.
Keywords— Big Data, Machine Learning, EHR, healthcare, medical platforms, technologies
I. INTRODUCTION
The concept of “big data” is not new; however the way it is defined is constantly changing. Various attempts at defining big data essentially characterize it as a collection of data elements whose size, speed, type, and/or complexity require one to seek, adopt, and invent new hardware and software mechanisms in order to successfully store, analyse, and visualize the data [1–3]. Healthcare is a prime example of how the three Vs of data, velocity (speed of generation of data), variety, and volume [4], are an innate aspect of the data it produces. This data is spread among multiple healthcare systems, health insurers, researchers, government entities, and so forth. Furthermore, each of these data repositories is siloed and inherently incapable of providing a platform for global data transparency. To add to the three Vs, the veracity of healthcare data is also critical for its meaningful use towards developing translational research.
Despite the inherent complexities of healthcare data, there is potential and benefit in developing and implementing big data solutions within this realm. A report by McKinsey Global Institute suggests that if US healthcare were to use big data creatively and effectively, the sector could create more than $300 billion in value every year [10]. After decades of technological laggard, the field of medicine has begun to acclimatize to today’s digital data age. New technologies make it possible to capture vast amounts of information about each individual patient over a large timescale. However, despite the advent of medical electronics, the data captured and gathered from these patients has remained vastly underutilized and thus wasted.
Important physiological and pathophysiological phenomena are concurrently manifest as changes across multiple clinical streams. This results from strong coupling among different systems within the body (e.g., interactions between heart rate, respiration, and blood pressure) thereby producing potential markers for clinical assessment.
Recently there has been an increasing adoption of electronic health records (EHRs) in different countries. Thanks to these systems, multiple health bodies can now store, manage and process their data effectively. However, the existence of such powerful and meticulous entities raise new challenges and issues for health practitioners. In fact, while the main objective of EHRs is to gain actionable big data insights from the health workflow, very few physicians exploit widely analytic tools, this is mainly due to the fact of having to deal with multiple systems and steps, which completely discourage them from engaging more and more [5].
We find very few research papers that evaluate the integration of analytic solutions within EHR systems [6].
In this paper, we will be addressing two things primarily. First, how structured and unstructured data from all the medical platforms can be stored at a common platform and various challenges in Centralization of healthcare data. Second, how Big Data analytics and Machine Learning can prove to be a boon after patient data has been centralized.
II. ELECTRONIC HEALTH RECORDS
EHR or electric health record are digital records of health information. They contain all the information you’d find in a paper chart – and a lot more. EHRs include past medical history, vital signs, progress notes, diagnoses, medications, immunization dates, allergies, lab data and imaging reports. They can also contain other relevant information, such as insurance information, demographic data, and even data imported from personal wellness devices.
They are considered as the modern and the digital version of the health information system, which provides information on diseases, previous consultations and exam results, the EHR allows patients and healthcare professionals to store, process and share electronically medical data for the coordination of care. Through EHR systems, patient information is more easily accessible to the different departments of health care facilities for various basic health care systems. From preliminary interviews to exams, diagnostics, eventual follow- up examinations and treatment, healthcare providers can quickly have the right information in case of emergency. The blood type, allergies, diseases, possible medications or any other vital signs measurements, everything is centralized and searchable at a glance [5].
Electronic health record systems can bring instant benefit to medical organizations by reducing administrative activities, ensuring data availability, minimizing waste, enabling faster time to treatment, reducing costs and overall improving the quality of care within a health entity. The main purpose behind setting up an electronic health records is to be able to analyse voluminous, varied, unstructured health data and acquire meaningful insights through analytical and decision-making tools [5].
EMR has to reach at every lowest level of Healthcare Services Providers and we will term that as PoS [Point of Service].
III. CENTRALIZATION OF HEALTHCARE DATA
The data has to be collected from each of the PoS and maintained at Central Level. In this we would have issues like data structure, data security/privacy, data storage and transfer of the same to Central Data Repository (CDR). The process to sync the data to CDR should be made easy and quick so that it can be integrated into EHRs without unnecessary hassles. Again, internet availability at each PoS for syncing data is must.
Every PoS should be registered with Central Agency and allotted a Unique Identification Number which shall be used to maintain patient records in EHR with unique ID, PoS_ID+ “-” +Patient_ID.
Every record of the patient visits to the PoS and diagnosis with Vital Stats, Medical Diagnosis Reports, and medicines prescribed and for how long they’re to be consumed should be recorded in EHR without a fail.
Once EMR data is synced into the CDR, a unique ID should be generated which can be accomplished easily and very quickly. Identifications like Adhaar Card Number, or PAN Number or any other such identification should not be included as that may lead to breach of security personal data of patient. The data in CDR should be handled independently from previously existing identifications.
The unique ID thus generated can be shared with the patient through SMS on registered mobile number. So next time, any patient visits PoS, his/her medical history can be extracted from the system with the use of this unique ID. Hence, there will be no need for the patient to carry his/her previous diagnosis reports or doctor’s prescriptions to every other doctor he consults with in different PoS.
Another very important aspect of this EHR system would be to come at a standard terminology in terms of Diseases and Procedures/Services. In USA industry, we noticed usage of International Classes of Diseases (ICD) and ICD 9 had been in use for a long time but now they’ve come up with new versions [8]. The newer versions of ICDs are found as ICD 10 and ICD 11. These ICD specifications came from World Health Organization or WHO. It has defined a standard way of Disease Diagnosis and that can be a way to be used in EHR so that during analysis of data, the variations/ differences in terminology being used by each PoS can be avoided. Just like ICD we have CPT or Current Procedural Terminology that can be adopted to reduce terminological differences and come at par in terms of Disease Identification and Procedural Services.
Before syncing data into CDR, a validation check must be done so as to make sure data is in proper format as prescribed by the Central Agency.
IV. CHALLENGES TO DATA CENTRALIZATION AND PROPOSED SOLUTIONS
A. Data Privacy and Security
The Privacy Standards and the Security Standards are necessarily linked. Any health record system requires safeguards to ensure that the data is available when needed and the information is not used, disclosed, accessed, altered or deleted inappropriately while being stored or retrieved or transmitted [7].
It can be achieved by adoption of following measures:
1. Authentication:
A person or entity seeking access to electronic health information is indeed the one as claimed and is also authorized to access such information must be verifiable [7].
2. Automatic log-off:
An electronic session after a predetermined time of inactivity must be forcibly terminated. To log in back, the user will have to initiate a new log in session.
However, it is recommended that the unsaved state of the system at the time of automatic log-off be saved and presented back to the user for further action [7].
3. Access Privileges:
Ideally only clinical care providers should have access rights to a person’s clinical records. However, different institutional care providers have widely varying access privileges specified that are institution-specific.
In cases of emergency where access controls need to be suspended in order to save a life, authorized users (who are authorized for emergency situations) will be permitted to have unfettered access electronic health information [7].
4. Audit Log:
All actions related to electronic health information must be recorded with the date, time, record identification, and user identification whenever any electronic health information is created, modified (non-clinical data only), deleted (stale and non-clinical data only), or printed; and an indication of which action(s) took place must also be recorded [7].
B. CONVINCING HEALTHCARE ORGANIZATIONS TO BE A PART
Next big challenge we can face is WHY the doctors and Healthcare Organizations would be ready to be a part of this system. They can be made aware of the benefits of such data collection and big data analysis but that would not look very lucrative to the health care provider’s community.
The whole process needs few additional things to be done at PoS level These would need to have extra manpower at PoS level to execute these things, may be doctor has to put few extra minutes on each patient, record those things and get that entered in the system with someone in the office.
With those things, thought was coming to my mind to compensate PoS specially individuals to be ready to execute these steps and be a part of the system. That can be in terms of Income Tax advantages or some other way to compensate and benefit them. Other than Individuals, there are many organizations with organized structures [Hospitals and Large Clinics] already in place, the Government may have to compensate them to make these changes in their systems so that those would be ready to be a quick part of this big system. These organizations would play a vital role specially starting with this endeavor.
In the United States, financial incentives offered for the “meaningful use” of health information technology has spurred growth in the adoption of the EHR and other enabling health-related technology. That can be done with along with a Central Data Agency, Government Ruling to provide benefits to adoption of these Standard EMRs, Provision of Government Ruling to provide incentives, there would be a need of an online System to evaluate EMRs certifying them ready for Data Collection, so that we get a validated population of EMRs compatible to the integrated data. Moreover, a system would need to be added at the level of Central
Data Agency to allow the PoS to sync data with use of APIs and other ways. That would be an altogether time consuming and lengthy process required as well as we would need a system for Data Analysis.
V. BIG DATA IN HEALTHCARE
Every second, dozens of terabytes of data are generated and accumulated from various sources, e.g. internet browsing, social networks, mobile transactions, online shopping and many others. Indeed, the big data paradigm has taken an expended shape, and the abundance of such structured and unstructured data has made it possible to be open to new perspectives. These new sources of data increase the chances of understanding one's behaviour and motivations, identifying instant signals and triggers for someone’s interest in a specific offer or product. Getting meaningful insights from voluminous and varied amounts of data helps to understand and extract hidden information, which can be used and exploited for the proper improvement of the users’ experiences [5].
Indeed, the analysis of health data can assist the improvement of the quality of care for a whole population, predict new epidemics and ensure equal access to care for everyone. While health analytics is represented as one of the most important technologies in e- health, their proper deployment and integration to EHRs is not as simple as it seems.
Currently, the process of implementing an EHR system is as much a challenge for health practitioners as it is the case for improving analytics and gaining meaningful insights [5].
The interpretation of big health data is not limited to a single approach or model. Both descriptive and predictive analytics need to be included for either or both can be convenient in different cases.
V. THE 4 “VS” OF BIG DATA ANALYTICS IN HEALTHCARE
Analytics associate big data with three primary characteristics: volume, velocity, and variety.
The healthcare data is produced in daunting volumes, created and accumulated continuously, resulting in an incredible volume of data that includes personal medical records, radiology images, etc.
With advancement of technologies, better platforms are being developed for storing and manipulating large amount of data effectively. Data is accumulated in real-time and at a rapid pace or velocity.
Future applications of real-time data, such as detecting infections as early as possible, identifying them swiftly and applying the right treatments (not just broad-spectrum antibiotics) could reduce patient morbidity and mortality and even prevent hospital outbreaks. Already, real-time streaming data monitors neonates in the ICU, catching life-threatening infections sooner [12]. The ability to perform real-time analytics against such high-volume data in motion and across all specialties would revolutionize healthcare [11]. Therein lies variety.
Some practitioners and researchers have introduced a fourth characteristic, veracity, or ‘data assurance’. That is, the big data, analytics and outcomes are error-free and credible. Of course, veracity is the goal, not (yet) the reality. Data quality issues are of acute concern in healthcare for two reasons: life or death decisions depend on having the accurate information, and the quality of healthcare data, especially unstructured data, is highly variable and all too often incorrect.
VI. ADVANTAGES TO HEALTHCARE
Potential benefits include detecting diseases at earlier stages when they can be treated more easily and effectively; managing specific individual and population health and detecting health care fraud more quickly and efficiently.
Numerous questions can be addressed with big data analytics. Certain developments or outcomes may be predicted and/or estimated based on vast amounts of historical data, such as length of stay (LOS); patients who will choose elective surgery; patients who likely will not benefit from surgery; complications; patients at risk for medical complications; patients at risk for sepsis, MRSA, C. difficile, or other hospital-acquired illness; illness/disease progression; patients at risk for advancement in disease states; causal factors of illness/disease progression; and possible co-morbid conditions (EMC Consulting) [9].
McKinsey believes big data can help reduce waste and inefficiency in the following areas [10]:
1. Clinical Operations:
To determine more effective and less expensive ways of disease diagnosis.
2. Research & Development:
2.1. Faster development of more accurate and effective vaccines through analysing data.
2.2. Analysing clinical trials and records of patients to identify patterns and discover adverse effects before products are released for public
2.3. Speeding new treatments to market and reducing trial and errors for analysing better treatments.
3. Public Health:
3.1. Analysing disease patterns and tracking disease outbreaks
3.2. Large amounts of data can be converted into useful information to identify needs of public, provide service and prevent crisis.
VII. MACHINE LEARNING
Just like big data, machine learning too has immense potential when it comes to healthcare. A lot has been achieved and a lot can be done to help practitioners to provide better services and patient care.
Having easy access to the blood pressure and other vital signs when a physician sees his patient is routine and expected. Imagine how much more useful it would be if he was also shown his patient’s risk for stoke, coronary artery disease, and kidney failure based on the last 50 blood pressure readings, lab test results, race, gender, family history, socioeconomic status, and latest clinical trial data.
We need to advance more information to clinicians so they can make better decisions about patient diagnoses and treatment options, while understanding the possible outcomes and cost for each one. The value of machine learning in healthcare is its ability to process huge datasets beyond the scope of human capability, and then reliably convert analysis of that data into clinical insights that aid physicians in planning and providing care, ultimately leading to better outcomes, lower costs of care, and increased patient satisfaction.
VIII. APPLIED MACHINE LEARNING IN HEALTHCARE
Machine learning in medicine has recently made headlines.
• Google has developed a machine learning algorithm [13] to help identify cancerous tumours on mammograms.
• Stanford is using a deep learning algorithm [14] to identify skin cancer.
• A recent JAMA article [15] reported the results of a deep machine-learning algorithm that was able to diagnose diabetic retinopathy in retinal images.
IX. REAL-TIME CLINICAL DECISION TO HELP PHYSICIAN
It’s clear that machine learning puts another arrow in the quiver of clinical decision making. Algorithms can provide immediate benefit to disciplines with processes that are reproducible or standardized.
At the same time a physician sees a patient and enters symptoms, data, and test results into the EMR, there’s machine learning behind the scenes looking at everything about that patient, and prompting the doctor with useful information for making a diagnosis, ordering a test, or suggesting a preventive screening. This will prove to be immensely helpful in cases where doctors struggle to put one and one together. Early and accurate diagnosis of disease is another big advantage.
Machine learning can be trained to look at images, identify abnormalities, and point to areas that need attention, thus improving the accuracy of all these processes.
Long term, the capabilities will reach into all aspects of medicine as we get more useable, better integrated data. We’ll be able to incorporate bigger sets of data that can be analysed and compared in real time to provide all kinds of information to the provider and patient.
X. ETHICS OF USING MACHINE LEARNING IN HEALTHCARE
Could there be a tendency for physicians to view machine learning as an unwanted second opinion?
There may be physicians who fear that machine learning is the beginning of a process that could render them obsolete. But it’s the art of medicine that can never be replaced. Patients will always need the human touch, and the caring and compassionate relationship with the people who deliver care. Neither machine learning, nor any other future technologies in medicine, will eliminate this, but will become tools that clinicians use to improve ongoing care.
The focus should be on how to use machine learning to augment patient care. For example, if a doctor is testing a patient for cancer, then he wants the highest-quality biopsy results he can possibly get. A machine learning algorithm that can review the pathology slides and assist the pathologist with a diagnosis, is valuable. If he can get the results in a fraction of the time with an identical degree of accuracy, then, ultimately, this is going to improve patient care and satisfaction.
And considering rare diseases with low data volumes, it should be possible to merge regional data into national sets to scale the volume needed for machine learning.
We must understand that data drives machine learning and with creating a centralized platform of EHR, we can have enough data to generate effective algorithms for predictive analysis.
XI. CONCLUSION
Healthcare needs to move from thinking of machine learning and big data analytics as a futuristic concept to seeing it as a real-world tool that can be deployed today. If they are to have a role in healthcare, then we must take an incremental approach. We must find specific use cases in which their capabilities provides value from a specific technological application (e.g., Google and Stanford). This will be a step-by-step pathway to incorporating more analytics, machine learning, and predictive algorithms into everyday clinical practice.
Medicine has a method for proving that treatments are safe and work effectively. It’s a long process of trial and error. We need these same processes in place as we look at machine learning to ensure its safety and efficacy. We need to understand the ethics involved in handing over part of what we do to a machine.
References
[1] A. McAfee, E. Brynjolfsson, T. H. Davenport, D. J. Patil, and D. Barton, “Big data: the management revolution,” Harvard Business Review, vol. 90, no. 10, pp. 60–68, 2012.
[2] C. Lynch, “Big data: how do your data grow?” Nature, vol. 455, no. 7209, pp. 28–29, 2008.
[3] A. Jacobs, “The pathologies of big data,” Communications of the ACM, vol. 52, no. 8, pp. 36–44, 2009.
[4] P. Zikopoulos, C. Eaton, D. deRoos, T. Deutsch, and G. Lapis, Understanding Big Data: Analytics for Enterprise Class Hadoop and Streaming Data, McGraw-Hill Osborne Media, 2011.
[5] Fadoua Khennou Youness Idrissi Khamlichi Nour El HoudaChaoui, Improving the Use of Big Data Analytics within Electronic Health Records: A Case Study based OpenEHR, 2018
.
[6] Myers, L. and J. Stevens, (2016) Using EHR to Conduct Outcome and Health Services Research, in Secondary Analysis of Electronic Health Records. Springer International Publishing: Cham. 61-70.
[7] e-Health Division, Department of Health & Family Welfare, Ministry of Health & Family Welfare, Government of India, Electronic Health Records Standards for India, 2016
[8] World Health Organization. The International classification of Diseases (ICD) [Internet]. Geneva, Switzerland: World Health Organization; c2018 [cited at 2018 June 18]. Available from: http://www.who.int/whosis/icd11.
[9] Wullianallur Raghupati and Viju Raghupati, Big data analytics in healthcare: promise and potential, 2014
[10] Manyika J, Chui M, Brown B, Buhin J, Dobbs R, Roxburgh C, Byers AH: Big Data: The Next Frontier for Innovation, Competition, and Productivity. USA: McKinsey Global Institute; 2011
[11] Feldman B, Martin EM, Skotnes T: “Big Data in Healthcare Hype and Hope.” October 2012. Dr. Bonnie 360; 2012. http://www.west-info.eu/files/big-data-inhealthcare.pdf
[12] IHTT: Transforming Health Care through Big Data Strategies for leveraging big data in the health care industry; 2013. http://ihealthtran.com/ wordpress/2013/03/iht%C2%B2-releases-big-data-research-reportdownload-today/.
[13] https://www.mercurynews.com/2017/03/03/google-computers-trained-to-detect-cancer/ [INTERNET]
[14] https://news.stanford.edu/2017/01/25/artificial-intelligence-used-identify-skin-cancer/ [INTERNET]
[15] Varun Gulshan, PhD; Lily Peng, MD, PhD; Marc Coram, PhD; et al, Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs
https://jamanetwork.com/journals/jama/article-abstract/2588763