Big Data in cardiovascular disease

“Big Data is a buzzword, or catch-phrase, used to describe a massive volume of both structured and unstructured data that is so large it is difficult to process using traditional database and software techniques”. ¹ In other words, Big Data refers to a collection of data sets, typically several billions or trillions of records, which are so large and complex that traditional systems of database management are unable to capture, store, process, and analyse the volume of information.Reference Haar ² The complex and varying amount of data in paediatric cardiovascular disease include data from surgery, anaesthesia, history, physical examination, cardiac intensive care, laboratory, electrocardiogram, Holter monitoring, event recording, pacemaker, echocardiography, CT, MRI, cardiac catheterisation, and angiography.

Big Data are unstructured, unlike traditional data systems, and are devoid of traditional relational database design techniques such as data normalisation, standardisation of data interfaces, data extraction, and data analytics.Reference Adamson ³ Big Data does not have the traditional table – the row and column structure that relational databases have. In a relational database, a schema for the data is required; every piece of data exists in a well-defined place, and tables are linked or joined by a shared identifier such as a unique patient identifier. Primary and foreign key constraints are required to enforce integrity of data in the tables within relational databases. In contrast, Big Data is often raw data with limited or no structure at all. Little or no data cleaning is performed, and business rules do not exist. Omics data sets are examples of Big Data in cardiovascular medicine. Profiles of biomedical data in the form of genomics, proteomics, lipidomics, and metabolomics are acquired using unbiased methodology such as mass spectrometry.Reference Adamson ^{3 , 4} These data sets can be re-used and re-purposed to answer clinical questions. These biomedical data can help create “personalised medicine”, whereby the Big Data can facilitate choice of therapy to match the unique biomedical profile of a given patient.

Many paediatric institutions are aggressively promoting evidence-based medicine by integrating patient-specific data into complex algorithms to help physicians streamline care. In order to achieve this, Big Data combines electronic health records, patient-generated data sources, bio-specimens, etc., from several healthcare facilities and from millions of patients. Some of these data are structured – for example, International Classification of Diseases 10th revision or ICD-10 – but a significant majority are unstructured – such as memos, clinical notes, radiology reports, cardiac angiography reports, etc.Reference Denaxas and Morley ⁵ The application of Big Data to medicine is exemplified by an analysis, where the review of clinical data obtained from an intensive care unit hours before a cardiac arrest event in a patient revealed several subtle events that could have alerted providers to intervene many hours before the event.Reference Busenbank ⁶

What makes Big Data unique is the ability to utilise multiple data sets to identify patterns of information that together can lead to an accurate diagnosis and guide therapy; one of the major advantages of Big Data is the ability to access existing patient records for an automated analysis of outcomes.Reference Haar ^{2 ,} Reference Denaxas and Morley ⁵ Unlike structured data, Big Data can be searched by concept rather than by key words.

Variability, inconsistency, and uncertainty in medical practice have been shown to compromise quality of care and cost efficiency. A total of 25% of heart attack victims do not receive aspirin early as recommended by evidence-based medicine.Reference Chen, Podchiyska and Altman ⁷

Medical practice is often driven by individual expert opinion and anecdotal experience. The advent of meaningful use of electronic medical records is creating a new opportunity for data-driven clinical decision support that utilises the expertise of many practitioners in a new learning health system.Reference Busenbank ^{6 –} Reference Richardson, Berwick, Bisgard and Bristow ⁸ Clinical decision support tools such as order sets are already helping to reinforce consistency and compliance with best practices.

The five “Vs” of Big Data are:

• ∙ data Volume – the massive amounts of new and historical data points;

• ∙ data Variety – structured versus unstructured, text, multimedia, and data graphs;

• ∙ data Velocity – how fast data comes in;

• ∙ data Value – the importance or significance of the data; and

• ∙ data Veracity – how accurate the data are.

In a specialised field such as paediatric cardiac care, special hardware and software are needed to transform high-end processing into objective treatment algorithms. The operationalisation of this objective is facilitated by the concept of semantic technology, which gives computers a deeper insight into the meanings and relationships of different medical terminology. It is becoming increasingly evident that the field is moving away from purely anecdotal information and textbooks to search engines and data-driven practice.Reference Frankovich, Longhurst and Sutherland ⁹

Healthcare organisations are driving quality improvement using descriptive and comparative analytics. Although descriptive analytics describe a group of patients, comparative analytics examine the differences between groups of patients by risk stratification or treatment effectiveness. Descriptive analytics are being used to develop evidence-based care guidelines and algorithms. The use of comparative analytics to compare different performance measures has proven critical in improving outcomes. In order to drive tangible, long-term changes, hospitals and physicians must move beyond descriptive and comparative analytics into predictive analytics. This technique has the ability to use known information in the medical literature, including best practices, latest scientific research, and existing databases, to develop a predictive model that allows clinicians to improve continuously the effectiveness of their treatments.Reference Frankovich, Longhurst and Sutherland ⁹ Predictive analytics allow clinicians to develop algorithms or tools to predict outcomes such as 30-day mortality after complete repair of transposition of the great arteries in the neonatal period. This approach is the basis for evidence-based medicine.

An example of effective utilisation in paediatric cardiovascular disease is the development of critical pathways for the management of infants with the spectrum of cardiac malformations related to tetralogy of Fallot by the cardiac programme at Lucile Packard Children’s Hospital at Stanford University.Reference Denaxas and Morley ⁵ The Stanford program prototyped an organisation-specific approach to treat patients with tetralogy of Fallot. The rationale for using this multi-disciplinary team approach to creating and deploying this clinical pathway was to reduce unnecessary and wasteful variations in treatment, ultimately resulting in improved outcomes as well as reduction in cost, and therefore improving value. Tetralogy of Fallot is a complex diagnosis with several distinct subsets requiring different treatment approaches. There are several institution-specific approaches to treating this disease entity. The Stanford program used a form of Big Data analytics to track compliance and outcomes. They reviewed the resource utilisation report from Pediatric Health Information Service, which is a collaborative of the Children’s Hospital Association and the Pediatric Research in Inpatient network. The resource utilisation report compared resource utilisation among 38 Pediatric Health Information Service hospitals with regard to the most costly inpatient conditions.Reference Denaxas and Morley ⁵

Unlike adult cardiac disease, paediatric heart disorders are relatively uncommon. A single institution may not treat enough children with these rare and complex diseases to generate a large enough sample size. By combining data from several institutions, one is able to generate large volumes of data to help make meaningful and statistically significant inferences, in order to expand our knowledge of paediatric diseases. The use of data from registries and collaboratives is important to achieve this goal.

Examples of such national and international registries and collaboratives that are specific to pediatric cardiovascular medicine and surgery are:

• ∙ Society of Thoracic Surgeons Congenital Heart Surgery Database,

• ∙ IMPACT Registry of the National Cardiovascular Data Registry of the American College of Cardiology,

• ∙ Joint Congenital Cardiac Anesthesia Society–Society of Thoracic Surgeons Congenital Cardiac Anesthesia Database,

• ∙ Paediatric Cardiac Critical Care Consortium,

• ∙ Virtual Pediatric Intensive Care Unit Performance System (VPS)

• ∙ Pediatric Health Network,

• ∙ United Network of Organ Sharing,

• ∙ Pediatric Heart and Lung Transplant Study Consortium,

• ∙ Intermacs/PediMacs,

• ∙ Congenital Heart Surgery Society database,

• ∙ Pediatric Cardiomyopathy Registry, and

• ∙ Pediatric Pulmonary Hypertension Registry.

Several of these databases, including Pediatric Health Network, attempt to utilise Big Data on a national and even international stage. The Pediatric Health Network is a collaboration of children’s hospitals and a data coordinating centre that conducts research studies in children with congenital or acquired heart diseases. It was created and is funded by the National Heart, Lung, and Blood Institute to improve outcomes and quality of life in children with cardiac disease. Congenital cardiovascular malformations affect ~40,000 infants in the United States of America each year, and are a leading cause of infant death. The incidence of CHD is three times that of childhood cancers, and is substantially greater than paediatric AIDS; yet, limited funded research has been conducted in the domain of paediatric cardiac disease. In the past 25 years, <40 randomised clinical trials have been carried out in patients with congenital cardiac disease or acquired paediatric cardiac disease, and most of these trials are on neonatal patent ductus arteriosus. The major barriers to clinical studies in paediatric cardiac disease include small numbers of individuals with a particular congenital cardiac defect at any one centre, lack of standardisation in treatment approaches, and lack of resources to coordinate collaborative research at the national or international level. Furthermore, parents are usually unaware of ongoing research studies. The collaborative approach of Pediatric Health Network brings together a large pool of patients with uncommon diseases through a common infrastructure for recruiting and monitoring, as well as standardising treatment. All the centres carefully follow a study protocol, collect identical data, and treat patients in similar ways, so that study results are accurate. ¹⁰

The Scientific Registry of Transplant Recipients is another example. The Scientific Registry of Transplant Recipients is contracted with the United States Department of Health and Human Services to collect, analyse, and report outcomes of all solid organ transplant programmes in the United States of America. The data collected include reliable transplant information for patients, families, and medical professionals, a complete list of United States hospitals with transplant centres, waiting time and organ availability, and survival statistics.Reference Salkowski, Snyder and Zaun ¹¹ Information about organ transplant centres and recipients in the United States of America is provided by the Scientific Registry of Transplant Recipients in collaboration with the Organ Procurement and Transplantation Network under contract with the Health Resources and Services Administration of the federal government of the United States of America. The Scientific Registry of Transplant Recipients utilises a Bayesian approach to assess outcomes of transplant programmes. The programme-specific reports are released publically twice a year. At present, this method applies only to monitoring of post-transplant outcomes, such as survival of patients and grafts. The Bayesian methodology allows the Scientific Registry of Transplant Recipients to provide a probability distribution indicating where the performance of a transplant programme is most likely to be in relation to what would be expected if the programme was performing at the level predicted based on national experience. This is a huge data repository as the database includes all solid organ transplant programmes in the United States of America.

In the current healthcare environment of accountability and cost-cutting, healthcare organisations will not survive without a vision and a plan. Even with rapidly declining reimbursement that makes resource allocation tight, hospital and practice executives must address what it will cost to sustain a commitment to quality, and then move forward in building capacity into the organisation to provide that quality.

Perceptions of standardisation and personalisation vary widely from institution to institution. Advocates of standardised care say that it improves clinical efficiency, outcomes, and patient safety. Advocates of personalisation, however, do not want to see a one-size-fits-all approach to delivering care become the norm; they want to see a healthcare system in which physicians treat patients like unique individuals. Utilisation of data enables clinicians to make informed decisions and plan customised treatments.

The keys to successful continuous quality improvement in the hospital setting are as follows:

• ∙ Engaging physicians and other leaders to identify and bring together the right people for championing effective collaboration.

• ∙ Leveraging the data warehouses of the hospital for extracting, organising, and analysing data, as well as identifying opportunities for improvement.

• ∙ Distributing the resulting analyses in real time, so that physicians and frontline caregivers can deliver customised care to each patient.

• ∙ Implementing change rapidly.

• ∙ Spreading expertise across the organisations – ensuring that individuals take the knowledge and skills they have acquired and apply them to other problem areas throughout the organisation.

• ∙ Understanding cohort trends and analysing those trends.

• ∙ Operationalizing an evidence-based, data-driven plan for optimising outcomes.

• ∙ Using the trends and the data-rich environment to predict variations, clinical severity, and endpoints of disease.Reference Oshiro ¹²

Big Data in healthcare today is mostly limited to research because of the highly specialised skill set required to run it. Hospital information technology experts are more familiar with using traditional relational databases. In order to utilise the power of managing or adopting principles of Big Data, hospitals need data scientists. These are PhD-level mathematicians, statisticians, and computer programmers commonly found in banking, finance, software companies, and industry. Some high-level research institutions also have data scientists on their staff. The goal is to have Big Data embrace SQL as the querying language, as this will then help bridge the gap between relational database Information Technology (IT) specialists and Big Data scientists.

Although research is expensive and complex, it is a vital component of delivery of care. All attempts must be made by institutions to promote research, as it can easily be hindered by a variety of organisational and technical roadblocks. Hospitals and health systems can overcome these common roadblocks to research, and turn analytics-powered research into improvements in the delivery of healthcare.

Healthcare data are now aggregated into single, centralised systems such as Enterprise Data warehouses. This aggregation of data makes the data accessible and actionable. Eventually, however, healthcare will want to look forward and consider adopting the processing power of advanced storage and analytic applications such as Hadoop. This application is an open-source platform that can handle essentially unlimited volumes of data. Hadoop was developed by Yahoo, and is now used by companies such as Facebook, Google, and Yahoo. Hadoop is touted as the most significant data processing platform for Big Data analytics in healthcare.Reference Crapo ¹³

Transparency

Public reporting of medical and surgical outcomes has gained widespread support. One rationale that supports public reporting is the theory that public access to data about provider performance will incentivise providers to improve their performance through a “peer pressure” effect.Reference Crapo ¹³ This theory offers that this publicly available information about outcomes and performance will give consumers the opportunity to be able to select providers on the basis of performance, shifting market share to those who provide the highest quality care. This concept relates closely to the “pay for performance” model in the healthcare reform act developed by the Center for Medicare and Medicaid services.Reference Joynt ¹⁴ Pay for performance is a payment model that is designed to shift from “paying for volume” to “paying for value”.

The Scientific Registry for Transplant Recipients bi-annual public reporting of programme and centre outcomes is one example of transparency in reporting of outcomes. These programme-specific reports are used by the Centers for Medicare and Medicaid Services, private insurance providers, and the general public to assess outcomes at transplant programmes.Reference Salkowski, Snyder and Zaun ¹¹

The PEDSnet learning health system is another example of transparency in reporting of outcomes. PEDSnet is a national network that supports the efficient conduct of clinical trials, observational research, and quality improvement across diseases, specialties, and institutions. PEDSnet combines Big Data with family engagement, and involves parents as partners in research design in order to address concerns that are important to families. PEDSnet is currently composed of eight paediatric academic health centres, and has data on 4.5 million children dating back to 2009. It has enabled centres to share data by standardising the definitions and descriptions of clinical observations for paediatric care. This strategy overcomes one of the biggest hurdles to implementation and use of Big Data.Reference Busenbank ⁶ All medications, laboratory data, and diagnoses are coded in the same way, allowing for rapid analysis. These data are entered once by clinicians as part of routine clinical care, but because these data are transformed into common standards a resource is created that can easily be used for research, quality improvements, population health, and/or benchmarking. PEDSnet data also can be pulled to identify patients for clinical trials, often in a matter of minutes.Reference Busenbank ⁶

As children cannot give consent, their parents or guardians make decisions and sign consent for what physicians and hospitals do with the health data of their children. Parents and guardians also require information and education in order to effectively participate in making informed decisions about the healthcare of their children. Traditional sources of information for families include internet search engines, social media, national hospital ranking systems, anecdotal peer consultation, and trusted but potentially misinformed practitioners. These sources may be biased and misleading. Public reporting of outcomes attempts to address this issue. In January, 2015, the Society of Thoracic Surgeons began to publicly report outcomes of paediatric and congenital cardiac surgery on the Society of Thoracic Surgeons website on a voluntary basis.Reference Jacobs and Jacobs ¹⁵ Although over 95% of paediatric cardiac surgical programmes in the United States of America participate in the Society of Thoracic Surgeons Congenital Heart Surgery Database, as of March, 2016, only 60% of these programmes voluntarily publicly report their outcomes – a notable improvement from the 23% of these programmes that voluntarily publicly reported when the initiative began in 2015. The Society of Thoracic Surgeons continues to strive to increase participation in its voluntary public reporting initiatives.

Recently, the National Pediatric Cardiology Quality Improvement Collaborative partnered with Sisters by Heart, a national group supporting parents and families of patients with hypoplastic left heart syndrome, to form a Transparency Workgroup, which includes an equal number of clinicians and parents.Reference Lihn, Kugler, Peterson, Lannon, Pickles and Beekman ¹⁶ The initial overall aim of the National Pediatric Cardiology Quality Improvement Collaborative was to decrease interstage mortality in patients with hypoplastic left heart syndrome, which varied across centres. Algorithms were developed by the National Pediatric Cardiology Quality Improvement Collaborative to guide nutritional optimisation of these fragile babies, and the spread of optimal nutritional practices led to decreased variation in interstage growth. A substantial reduction in interstage mortality across the collaborative was reported in 2015. Parents voiced their frustration about variations in practices among centres, and requested that member centres share their outcomes data to enable all to learn and improve. The following four levels of transparency were proposed:

• ∙ Individual Transparency – proactively sharing data about expected outcomes prior to the delivery of healthcare, within and between members of the care team including a surgeon, a cardiologist, and parents/patients. These data can help parents choose a care team.

• ∙ Organisational Transparency – ensuring that all clinicians and team members within a centre are aware of the outcomes and performance of their centre. These data can help drive the performance improvement of a programme.

• ∙ Collaborative Transparency – sharing centre-specific and aggregate information about processes and data about outcomes among members of a multi-centre collaborative. These data can help identify best practices that can subsequently spread to other centres.

• ∙ System Level Transparency – an individual centre or a multi-centre collaborative sharing data with the public, such as on a website or through the press. This transparent public reporting can help drive system-wide improvement.Reference Lihn, Kugler, Peterson, Lannon, Pickles and Beekman ¹⁶

Summary

Although data about healthcare are currently being managed by existing relational databases and enterprise data warehouse systems, it is becoming increasingly evident that several institutions have begun to direct their attention towards the more sophisticated systems of Big Data, that are capable of acquiring, storing, analysing, and reporting large amounts of data. Big Data is an evolving domain of healthcare with far-reaching implications. In fact, it is estimated that the integration of Big Data analytics into healthcare would save the healthcare industry at least $300–450 billion a year;Reference Groves, Kayyali, Knott and Van Kuiken ¹⁷ however, major challenges to using Big Data in healthcare are the technical expertise required to use it and the lack of robust, integrated security to manage it. Compliance with regulations to protect the privacy of patients, such as the Health Insurance Portability and Accountability Act of 1996 of the United States of America (HIPAA), must always be aggressively enforced to protect the privacy and security of data about patients.

Transparency on quality, price, and safety can help reduce the costs of healthcare and improve outcomes by empowering consumers to make informed choices. The proper use of Big Data can facilitate transparency. Furnishing patients, their families, and insurance companies with data about comparative performance, so that they can make informed healthcare choices, is one of the few strategies for creating consequences for low quality of care or high quality of care.Reference Hibbard, Peters, Slovic, Finucane and Tusler ¹⁸ It is not enough to simply dump raw data on consumers; it is best to deliver meaningful, accurate, understandable, and usable information, based on comparative risk analytics and risk adjusted, audited, and verifiable data, in order to empower parents, promote patient-engaged and family-engaged healthcare, and potentially improve outcomes in the domains of both the health of our patients and the experience of our families.Reference Basken ^{19 ,} Reference Jacobs ²⁰