Abstract

The purpose of this report is to explore the future of Information Technology (IT) and its effect on Healthcare Information Systems (HIS), by looking at the present, and future trends in healthcare. Looking at the implementation of HISs, the opportunities and challenges are discussed, and the necessary skill for developing such a system are analysed. Finally, the digital transformation of South London and Maudsley NHS Foundation Trust is reviewed, what was presented at City, University of London by Stephen Docherty Chief Information Office (CIO).

Introduction

Purpose of the report

This aim of this report is to explore the future of the IT industry and Artificial Intelligence (AI)and their effect on the healthcare industry. First, the terms AI and data mining will be introduced and defined, then the actual trends in health care will be discussed in details.
In the literature review, the scientific views of the current opportunities of applying good computer science practices will be dissected, as well as the applications of data mining in healthcare. The review will further investigate what the main challenges are of this digital transformation. In the discussion, the British trends will be discussed in health informatics, and how the legal and ethical issues relate to computer science in healthcare. Finally, the required technical and non-technical skills will be analysed, that are required for a computer science professional in healthcare.

Background of AI

Artificial Intelligence (AI) has been around since the 1950s, when Alan Turing has written an article about artificial intelligence, titled as Computer Machinery and Intelligence (Turing, 1950).
In this article, Turing started by asking the question: “Can machines think?” Since it was difficult to define “machine” or “think”, Turing proposed another question: “Can a machine win a game, called ‘Imitation Game’?” The Imitation Game (according to Turing) is played by three players: Player A as a man, player B as a woman, and the interrogator C. The interrogator can ask questions from the players however, the messages are passed by notes, therefore player C cannot see which player is a man or woman. During the game player B’s goal is to convince the interrogator that she is a woman, and the goal of A is to trick C. Turing was curious, what would happen if player A’s role would be taken by a computer.
Later, the original test was modified, and now several versions exist. For example, in one version, both player A and B are trying to convince the interrogator that they are the human. Player C has to decide which player is the computer. To pass the test, a computer must trick a human more than 30% of the time during a series of five minutes keyboard conversations. First time in the history, this was achieved in 2014 on an event organised by University of Reading (University of Reading, 2017).
Part of artificial intelligence is data mining, which can be defined as the process of finding previously unknown patterns in large datasets, and then by using these patterns, a predictive model then can be built (Kincade, 1998). Data mining is especially powerful on medical data, because usually medical datasets have hundreds of thousands of data points.
As it was prior discussed, AI itself has been around for a while, and many of the applied statistical and mathematical models were discovered decades ago, however, the field did not have many practical applications, simply because of the lack of computer power, and data. However, the processing power and the memory of the computers have dramatically improved in the last couple of decades (Moravec, 1998).
Besides the computational power, the available digital data has also been increasing in the past ten years. In 2014, the International Data Corporation (IDC) has conducted a study for Dell EMC, and predicted that the amount of digital data will double in every two years, and by 2020, the data we create annually will reach 44 trillion Gb (Zwolenski and Weatherill, 2014).

Trends in health care

To investigate the opportunities for computer science and AI in healthcare, first, one has to consider the main trends in health care that support digital transformation.
Moving from a paper-based system to more computer-based systems is common today over the major industries. In the medical industry in the past, most of the data was recorded and stored mainly on paper however, the migration to paperless approach has already been started (Haux et al., 2002). Increasing the computer system’s boundaries, has also been a trend, especially with the appearance of the Internet. While in the past, an information system in health care was limited to a department, or to a particular hospital, now new systems are targeting to record whole cities, regions or even countries (Linberg, 1968). This expansion of the system boundaries also has an effect on the type of users of a system. While earlier, a specific system was designed for the physicians and the administrative workers, later it was passed on to be used by nurses (Ball et al., 1994). Today, even patients might have some limited access to check their test results, or to consult with doctors. Finally, the way as medical data is used has changed from using it only to patient’s care to use it for research or educational purposes (Leiner et al., 2002; Kuhn and Guise, 2001).
As the consequence of these trends, today, the available digital data is more than ever, and as it was mentioned before, this is considered to be one of the main requirements for building an effective data mining tool.

Literature review

Healthcare Information Systems (HIS) refers to the complete information processing and information storing subsystem of a hospital (Haux et al., 2004). The medical data of a patient is stored in an Electronic Health Record (EHR), and mainly it is either clinical data such as medical test results and notes from consultations, or social data, like notes and reports from home visits. With advanced HIS, the data can also come from the patient, recorded by smart sensors.
The variety of the data sources have led to an increase in the complexity of the actual data recorded for each patient.
EHR might have high resolution images from dermatoscopes or other sensors, or low resolution data, such as manual clinical observations. This increase in both quantity and quality of the recorded data presents a challenge for the clinical practice as well as for medical research to exploit the useful information from such a data set (Mernik and Menzies, 2012).

Challenges

Since HIS consist of a wide range of technologies, such as databases, websites, applications, biomedical, administrative and financial technologies, technical challenges are arising during the implementation of such a system.
One of the biggest challenges are the lack of standardization of the digital data and the lack of a well-developed healthcare information exchange. During the development and design, the implementers have to spend a significant amount of time with ‘cleaning’ and formatting the medical data in order to be able to record it into a central, common HIS.
Often another technical challenge can arise, which is data sharing over networks. Sharing such data can lead to security and data privacy issues. Other obstacles often arise on the management / business level (Blumenthal, 2009).
Managers, and business owners are typically concerned about the cost and the return of their investment. This remains a major issue, as HIS projects typically requires big initial investment, because of the software development cost. Also, HIS implementation would typically require huge organizational changes, including buying new devices and training the users (Blumenthal, 2009). Finally, as Lapointe and Rivard (2005) pointed out, user resistance to IT is a novel and rising issue, and IT resistance researchers need to figure out how and why it occurs, especially in HIS environment.

Opportunities

Given the high frequency of medication errors, their prevention is a worldwide priority for health systems. In the UK, out of 1000 consecutive claims reported to the Medical Protection Society from 1996, 193 were related to drug prescriptions. Healthcare information systems, with components such as computerized physician order entry, bedside bar-coded medication administration, and electronic medication reconciliation are key to prevent medication errors (Agrawal, 2009).
Clinical decision making is a complex process, with several steps that requires the physician to potentially memorise and recall a huge amount of data related to a particular patient and drug. Since most of the errors occur at the prescription step, a good HIS with a computerized physician order entry (CPOE) is a powerful tool to improve patients’ safety. A CPOE would make sure that the order is complete and legible, including all necessary information such as; prescribed dosage, route of administration and unit doses, patient’s allergies, drug recall warnings and drug-drug interactions. It also helps the physician to calculate the correct dosage based on clinical features such as the weight of the patient. Another component of a HIS is the bar-coded medication administration (BCMA) system. This would help to avoid wrong dosage of medication, or mismatch with the patient’s identity. Previous studies have shown 54-87% reduction in medication errors where BCMA systems were used (Agrawal, 2009).
HIS can be a great tool to reduces the cost of keeping health records, furthermore it also can improve workflows, practice management and pricing. It is expected to permit data sharing between healthcare providers, which will reduce office and hospital visits and hospital admissions, and even reduce malpractice lawsuits (Goldschmidt, 2005).
Reducing medical errors and the cost of healthcare are important to increase the overall quality of healthcare. However, by harvesting the capabilities of data mining (provided by the huge amount of structured data) disease research, prevention, and clinical decision making can be significantly improved (Goldschmidt, 2005).

Data mining opportunities in healthcare

Opportunities of data mining in healthcare can generally grouped as management of healthcare, customer relationship management, detection of fraud and evaluation of treatment effectiveness (Kincade, 1998).
In healthcare management, data mining aims to better identify, track and aid chronic disease states and patients, design interventions, and overall reduce the number of hospital admissions. For example data mining can be used to group patients by their demographic characteristics and medical conditions to determine which group uses the most resources, and develop programs to educate people to prevent or manage their conditions (Kincade, 1998). In Carnegie Mellon University, several machine learning approaches were used to detect heart failure in patients. The university was working on EHR data, collected by the Geisinger Clinic between 2001 and 2006. By using logistic regression and boosting techniques, researchers predicted heart failure more than 6 months before the diagnosis (Wu, Roy and Stewart, 2010). Data mining also can aid successful infection control or as an automated early-warning system in the event of epidemics (Kreuze, 2001). A syndromic system, based on patterns of symptoms, is likely to be more efficient and effective than a traditional system that is based on diagnosis. An early warning of the global spread of the SARS virus is an example of the usefulness of a syndromic system based on data mining (Brewin, 2003).
The next data mining opportunity option is through the Generally Customer Relationship Management (CRM). CRM is an approach at commercial organisations, however the relationship between the patients and the providers is also important. As in the case of any commercial organisation, data mining applications can be developed in the healthcare industry to determine the preferences, usage patterns, and current and future needs of individuals to improve their level of satisfaction (Biafore, 1999). Through the use of data mining, Customer Potential Management Corporation has developed a Consumer Healthcare Utilization Index that provides an indication of an individual’s propensity to use a specific health care service. The index is generally defined by twenty-five major diagnostic categories, selected diagnostic related groups or specific medical service areas. This index (based on millions of healthcare transactions of several million patients) can identify patients who can benefit the most from specific healthcare services. It can encourage patients who most need a specific care to access it, and continually refine the channels and messages used to reach appropriate audiences for improved health and long-term patient relationships and loyalty. The index has already been used by OSF Saint Joseph Medical Centre to get the right messages and services to the most appropriate patients at strategic times. The result is more effective and efficient communications as well as increased revenue (Paddison, 2000). Pharmaceutical companies can also benefit from data mining by tracking the physicians and the drugs prescribed for patients. It can be checked and evaluated which is the least expensive or most effective treatment plan for an ailment, as well as to help identify physicians whose practices are suited to specific clinical trials (for example, physicians who treat a large number of a specific group of patients) (Brannigan, 1999).
Data mining applications that attempt to detect fraud and abuse, first establish norms then identify abnormal patterns in claims of physicians, laboratories or clinics. A successful example of using data mining to detect fraud and abuse was when ReliaStar Financial Corporation has started to use data mining and has reported a 20 percent increase in annual savings (Christy, 1997).
Data mining applications can be developed to evaluate the effectiveness of certain medical treatments. By comparing and contrasting causes, symptoms, and courses of treatments, data mining can deliver an analysis of which courses of action prove effective (Milley, 2000). For example, the outcomes of patient groups treated with different drug regimens for the same disease or condition can be compared to determine which treatments work best and are most cost effective (Kincade, 1998).
The final category is a more specialized application of data mining in healthcare, such as predictive or personalised medicine, analysis of the genome or images produced by neuroscience. There are plenty of examples for researchers using machine learning algorithms to find patterns in genetic data to find biomarkers for certain diseases. These biomarkers then later can be used to identify disease in potential patients. Boutorh and Guessoum (2016) used a hybrid machine learning technique, based on Association Rule Mining, Neural Networks, and Evolutionary Algorithms on Single Nucleotide Polymorphism (SNP) data, to classify patients with Autism, Mental retardation, Colon cancer or Breast cancer. At the end of the classification, the selected features were identified as possible biomarkers for the particular disease. Their Focused Time Delay Neural Network achieved 100% classification result for breast -and colon cancer, and more than 90% for the mental disorders (Boutorh and Guessoum, 2016).
Another example is Yang et al., 2010, where another hybrid machine learning method was used, on a combined data of SNPs, and functional magnetic resonance imaging (fMRI). The selected combined model has achieved 87.2% accuracy in predicting schizophrenia in patients (Yang et al., 2010).

Discussion

Digital innovation - Trends in the industry

As it was earlier discussed earlier, there are several global trends in healthcare, such as moving from paper based to paperless systems, or increasing the boundaries of the healthcare system, which allows data sharing across the healthcare providers and the patients.
In United Kingdom, National Health Service (NHS) was founded in 1948 with the ideal, that good healthcare should be available to everybody regardless of their wealth. NHS always aims to take advantage of the latest opportunities that science and technology has to offer, however these changes mean they need to take a longer view called a Five-Year Forward View (FYFV) to consider possible opportunities (NHS England, 2017).
Several issues were identified which are in-line with the global trends such as the little amount of time that patients, even with long term conditions, actually spend with healthcare professionals. This is mainly because of the divide between family doctors and hospitals, and physicians and mental health. This separation makes patients to visit multiple facilities with complex conditions, or to bring test results from one physician to another.
NHS as a world leader in primary care computing recognizes the importance of HIS, and the challenges and opportunities come with it. However, in the past decades the progress of the development was slower than it should have been, mainly because of the two opposite approaches for digital transformation. First, they tried a highly centralised, national implementation, however it has failed, because of lack of local engagement, and sensitivity of local circumstances. The other approach was to let several independent systems to be developed, however they ended up with several systems that don’t share their data.
In the future, the focus will be on systems, that provide a glue between the different parts of the healthcare systems, provided they meet the nationally specified interoperability and data standards. To advance the implementation of the FYFV, a National Information Board (NIB) has been established, which will publish a set of “road maps” to support the transformation. Some of the key elements will include: comprehensive transparency of the data, expanding the NHS accredited health apps that patients will be able to use, fully interoperable electronic health records, family doctor appointments and electronic prescriptions availability online, and bringing together medical data from hospitals and GPs to support quality improvement and research.
Research is vital for NHS, therefore they will continue supporting the National Institute of Health Research (NIHR) and the network of specialist clinical research facilities in NHS. Further steps will be taken to speed up discoveries in medicine and diagnostics such as reducing the cost of Randomised Controlled Trials (RCTs), expand the Early Access to Medicine programme, rolling out new devices and equipment and decommission old ones, and to accelerate the adoption of new innovations. To support this acceleration, NHS will develop a small number of “test bed” sites, that will serve as a real-world site to test innovations and to help their integration to the existing technology.

Legal, professional and ethical issues

Ethical issues mainly raise from three sources, the autonomy of users and privacy. Autonomous patients expect access to their data, and many patients will desire some level of control over their records. Such an expectation conflicts with the medical and legal utility of a health record and HIS. To tackle this problem, patients should be restricted to view, update, but not to delete any information from their health records. The ownership of the records is another problem, because given the fact that a health record holds a patient’s private, intimate data, the patients claim, that they are the rightful owners. On the other side, companies who develop the HIS, maintain the data storage and servers, but even the healthcare providers claim the ownership of the same data. These conflicting interests, must be rectified both ethically and legally before a HIS is implemented (Mercuri, 2015).
The privacy of data is important in any information system, however in healthcare it is even more crucial. Since patient’s record contains intimate data, once it is revealed, patients can lose trust in the healthcare, which will damage the unique doctor-patient relationship. Strict privacy policies, however conflict with data sharing, therefore a good public policy has to be defined that protects privacy but in the same time provides a level of openness that will benefit the public health (Mercuri, 2015).
For example, a good policy that protects the patient’s privacy and autonomy, would give the patient’s the right not just to decide if they wish to share their data, but they could also decide what level of access they want to guarantee for which providers, or scientists (Mercuri, 2015).
A large interconnected system, with shared data is a very important part of a good HIS, because of the great potentials of the massive amount of raw data, as it was discussed earlier. As part of the system, patients should be able to provide access to their data for researchers and scientist, but also, approved scientist should be able to access data that was made public (Mercuri, 2015).
In such a system, data security issues arise, especially when portable devices are involved too, therefore robust authentication and access control is required, however keeping the balance between making a system secure or easy to use is one of the greatest challenge. This is even more difficult, when some of the users might be physically or mentally impaired (Cushman et al., 2010).

Necessary technical and non-technical skills, and personal development

The necessary non-technical skills will depend on the specific role. For software developers, these skills are the same regardless if it’s a healthcare company or a bank. Cognitive abilities such as critical and analytical thinking or problem solving is crucial for a programmer. Since most of the companies are working in agile now, communication skill became more important. In agile teams, there are regular discussions and “show and tells”, where every member of the team (including developers), must stand in front of other people and present their progress, or discuss the problems.
Some level of business awareness is necessary even for developers, because they should understand the background of a project to be able to deliver a good product. With higher positions, such as team leaders and managers, business awareness and service delivery management becomes more important. People in these positions are responsible for a team, therefore skills like leadership, planning, and time management are very important.
There are several different types of developers in healthcare, and the required technical skills depends on the type of the role. In general, employers are looking for multi-talented programmers, with as many skill as possible, with proven commercial experience. There are mainly two types of roles: web developers, and application developers. For web developers, javascript seems mandatory nowadays, regardless if it is JQuery, NodeJS, Angular, or React. Html, CSS is basic for any web developer, C# with .Net, python, or ruby on rails are used at many companies, knowing at least one of these technologies can make a big difference.
For application developers, the necessary skills and languages depend on the project that they must work on. Every language has the advantages and disadvantages, for example Java is popular because it is platform independent, and C++ is usually used when the focus is on the performance. Python became more popular recently, mainly because it is easy to learn, and it has many libraries that are used in machine learning. One of these software libraries is Tensorflow developed by Google’s Brain Team. Tensorflow is a very powerful, open source library, that can be used in research and production without rewriting any code. The flexible architecture allows the developers to deploy the computation to one or more CPUs or GPUs in any environment (server, desktop, mobile) with a single API. Knowing python and / or Tensorflow is a great advantage for any developer who wish to work in data science.

Expert analysis

Stephen Docherty is a Chief Information Officer (CIO) for South London and Maudsley (SLaM) NHS Foundation Trust. SLaM provides mental health services and substance misuse services in the community of Lambeth, Southwark, Lewisham, and Croydon. It is the only NHS Trust with a biomedical research centre, hosted in King’s College. Besides the health services, SLaM also provides a range of education, training and learning opportunities (Slam.nhs.uk, 2017).
Stephen, during his expert talk introduced us an innovative IT strategy, that was signed off at SLaM in March 2015.
Some of the key points of the new strategy were: “cloud first” approach, service management, frameworks, and culture change.
Cloud based services played a key point in the transformation. During the past two years, SLaM migrated all staff to Microsoft Office 365, which is the cloud based version of Microsoft Office. This helps the support of the users, since the application is cloud based, therefore the maintenance is Microsoft’s responsibility. As a business intelligence solution, Microsoft PowerBI has been introduced, which is another Microsoft cloud based application providing modern data visualization and analytical tools. More than 2000 old devices were also replaced, and the electronic health record system called electronic Patient Journey System (ePJS) has been tested on mobile devices. In 2013, SLaM released myHealthLocker with the purpose of engage patients online in their care, however the take up was poor. Now, a new version of the application (HealthLocker V2) is being developed and tested through a user-led design process, focused on delivering a clinically useful tool. Through the application, patients will be able to send confidential, secure messages to the staff, access their care plan, or even track their sleep. The development team is using an agile development process, with involving service users, carers, clinicians, and researchers to make sure that the application will meet their needs. HealthLocker is going to be launched by the end of July, but the development won’t stop, it is planned to add new features in the coming months (Healthlocker.uk, 2017).
As discussed earlier, the successful adoption of digital and technical innovations, requires active support of service users. SLaM created a forum called SL@M Connect, which is a weekly meeting between clinicians, digital services and communications team to discuss the issues, inform people about digital developments, and to discuss new ideas.

Conclusion

This analysis has looked at future of information technology, and how it relates to healthcare. British trends in healthcare are following the global trends, which was proven by looking at the Five Forward Year View of NHS. The opportunities with healthcare information systems are certainly outweigh the challenges, however to make a successful digital transformation that satisfies all stakeholders and users of the system, the designers and project leaders must achieve a cultural change, and they must involve every user in the design process. Such systems also raise several security, data privacy, and piracy issues, that has to be taken very seriously, because of the nature of the data.
Despite of these challenges, digital innovation is already happening in healthcare, one great example is CIO Stephen Docherty at South London and Maudsley NHS Foundation Trust. Stephen and his team, restructured the IT department and rolled out several products that will help data sharing, business intelligence, and remote patient care, and the foundation also has a biomedical research centre where data mining algorithms and tools are used on the collected data for research purposes.

References

Agrawal, A. (2009). Medication errors: prevention using information technology systems. British Journal of Clinical Pharmacology, 67(6).
Bellman, R. E. (1978). An introduction to artificial intelligence: can computers think? San Francisco: Boyd & Fraser Pub. Co.
Biafore, S. (1999). Predictive solutions bring more power to decision makers. Health Management Technology, 20(10).
Blumenthal, D. (2009). Stimulating the adoption of health information technology. The New England Journal of Medicine, 360(15).
Brannigan, M. (1999). Quintiles seeks mother lode in health “data mining.” Wall Street Journal, 2 (1).
Brewin, B. (2003). New health data net may help in fight against SARS. Computerworld, 37(17).
Boutorh, A. and Guessoum, A. (2016). Complex diseases SNP selection and classification by hybrid Association Rule Mining and Artificial Neural Network—based Evolutionary Algorithms. Engineering Applications of Artificial Intelligence, 51.
Christy, T. (1997). Analytical tools help health firms fight fraud. Insurance & Technology, 22(3).
Cushman, R., Froomkin, A., Cava, A., Abril, P. and Goodman, K. (2010). Ethical, legal and social issues for personal health records and applications. Journal of Biomedical Informatics, 43(5).
Emc.com. (2017). Executive Summary: Data Growth, Business Opportunities, and the IT Imperatives | The Digital Universe of Opportunities: Rich Data and the Increasing Value of the Internet of Things. [online] Available at: https://www.emc.com/leadership/digital-universe/2014iview/executive-summary.htm [Accessed 9 Jul. 2017].
Goldschmidt, P. G. (2005). HIT and MIS: implications of health information technology and medical information systems. Communications of the ACM, 48(10).
Haux, R., Winter, A., Ammenwerth, E. and Brigl, B. (2004). Strategic Information Management in Hospitals. New York, NY: Springer New York.
Healthlocker.uk. (2017). Healthlocker. [online] Available at: https://www.healthlocker.uk/ [Accessed 22 Jul. 2017].
Kincade, K. (1998). Data Mining: Digging for Healthcare Gold. Insurance & Technology, 23.
Khosrowpour, M. (2015). Encyclopedia of information science and technology. 3rd ed. Hershey: Information Science Reference.
Kreuze, D. (2001). Debugging hospitals. Technology Review, 104(2).
Kurzweil, R. (1990). The age of intelligent machines. London: MIT Press.
Mercuri, J. (2015). The Ethics of Electronic Health Records | Clinical Correlations. [online] Clinicalcorrelations.org. Available at: http://www.clinicalcorrelations.org/?p=2211 [Accessed 18 Jul. 2017].
Mernik, M. and Menzies, T. (2012). 2012 First International Workshop on Realizing AI Synergies in Software Engineering. 1st ed. NJ, USA: Press Piscataway.
Milley, A. (2000). Healthcare and data mining. Health Management Technology, 21(8).
Moravec, H. (1998). When will computer hardware match the human brain? Journal of Evolution and Technology, Vol. 1(1541-0099).
NHS England, N. (2017). NHS England » NHS Five Year Forward View. [online] England.nhs.uk. Available at: https://www.england.nhs.uk/five-year-forward-view/ [Accessed 12 Jul. 2017].
Paddison, N. (2000). Index predicts individual service use. Health Management Technology, 21(2), 14-17.
Rich, E. (1991). Artificial intelligence. New York: McGraw.
Slam.nhs.uk. (2017). Home - South London and Maudsley NHS Foundation Trust.. [online] Available at: http://www.slam.nhs.uk/ [Accessed 17 Jul. 2017].
Turing, A. (1950). Computing Machinery And Intelligence. Mind, LIX(236).
University of Reading. (2017). University of Reading. [online] Available at: http://www.reading.ac.uk/news-and-events/releases/PR583836.aspx [Accessed 9 Jul. 2017].
Wu, J., Roy, J. and Stewart, W. (2010). Prediction Modeling Using EHR Data. Medical Care, 48.
Yang, H., Liu, J., Sui, J., Pearlson, G. and Calhoun, V. (2010). A Hybrid Machine Learning Method for Fusing fMRI and Genetic Data: Combining both Improves Classification of Schizophrenia. Frontiers in Human Neuroscience, 4.
Zwolenski, M. and Weatherill, L. (2014). The Digital Universe Rich Data and the Increasing Value of the Internet of Things. Australian Journal of Telecommunications and the Digital