Key points

“Big data” is a broad term for collection of data sets so large and complex that they are difficult to process using on-hand database management tools or traditional data processing applications. Predictive analytics encompasses a variety of statistical techniques from modeling, machine learning, and data mining that analyze current and historical facts to make predictions about the future, or otherwise unknown events.

Changes in the health care delivery model over the past decade in the United States have led to renewed interest and newer mechanisms in data-driven methods for managing quality of care. Health systems and provider groups are now evaluated using quality metrics for mortality, length of stay, and for hospital readmissions within 30 days. Rankings are increasingly made public, and are being used by insurers to institute pay-for-performance programs. Hospitals need to continually monitor and understand their performance, and determine whether interventions to improve have been successful.

Definition of analytics and its impact on health care

Analytics is the systematic use of data combined with quantitative as well as qualitative analysis to make decisions. Health care analytics efficiently applies clinical and administrative data in electronic health records (EHRs), and knowledge of clinical practice standards and guidelines, by way of order sets and protocols, to manage metric-driven quality improvement. The goals of such an initiative are to identify inefficiencies in care delivery and variations from standard practice, explore opportunities for expanding services that may enhance quality of care, efficient targeting of limited resources, and peer benchmarking. The analytical techniques needed to achieve these goals require mature EHR implementations.

EHRs with the required breadth of data increasingly are available at children’s hospitals. Institutions may make these data available through a clinical data warehouse, a relational database fed by an EHR’s transactional system that supports efficient population queries. National repositories of administrative and sometimes clinical data provide institutions with access to comparative data; for example, the Children’s Hospital Association (CHA) Pediatric Health Information System (PHIS). The CHA, through a wide variety of data programs, helps pediatric hospitals drive improved clinical, operational, and financial performance. With shared data sets, potentially better practices across organizations can be identified, ultimately improving children’s health care. The PHIS, a comparative pediatric database, includes clinical and resource use data for inpatient, ambulatory surgery, emergency department, and observation unit patient encounters for 45 children’s hospitals. PHIS supports a wide range of improvement activities, including clinical effectiveness, resource use, care guideline development, readmission analysis, antimicrobial stewardship, and physician profiling (ongoing professional practice evaluation). Such data sets represent a rich potential source of data for analytics, but they also present substantial challenges in their use. Diagnoses, comorbidities, and procedures typically are represented indirectly as billing codes. Clinical data such as laboratory test results and medication histories may be recorded as local codes or in text, and they may require substantial clinical context to interpret. Efforts are underway to create publicly available libraries of phenotypes defined in terms of patterns in codes and discrete data. Tools for translating these phenotypes into queries of local data warehouses and national repositories could make data-driven quality improvement more broadly practical.

Definition of analytics and its impact on health care

Analytics is the systematic use of data combined with quantitative as well as qualitative analysis to make decisions. Health care analytics efficiently applies clinical and administrative data in electronic health records (EHRs), and knowledge of clinical practice standards and guidelines, by way of order sets and protocols, to manage metric-driven quality improvement. The goals of such an initiative are to identify inefficiencies in care delivery and variations from standard practice, explore opportunities for expanding services that may enhance quality of care, efficient targeting of limited resources, and peer benchmarking. The analytical techniques needed to achieve these goals require mature EHR implementations.

EHRs with the required breadth of data increasingly are available at children’s hospitals. Institutions may make these data available through a clinical data warehouse, a relational database fed by an EHR’s transactional system that supports efficient population queries. National repositories of administrative and sometimes clinical data provide institutions with access to comparative data; for example, the Children’s Hospital Association (CHA) Pediatric Health Information System (PHIS). The CHA, through a wide variety of data programs, helps pediatric hospitals drive improved clinical, operational, and financial performance. With shared data sets, potentially better practices across organizations can be identified, ultimately improving children’s health care. The PHIS, a comparative pediatric database, includes clinical and resource use data for inpatient, ambulatory surgery, emergency department, and observation unit patient encounters for 45 children’s hospitals. PHIS supports a wide range of improvement activities, including clinical effectiveness, resource use, care guideline development, readmission analysis, antimicrobial stewardship, and physician profiling (ongoing professional practice evaluation). Such data sets represent a rich potential source of data for analytics, but they also present substantial challenges in their use. Diagnoses, comorbidities, and procedures typically are represented indirectly as billing codes. Clinical data such as laboratory test results and medication histories may be recorded as local codes or in text, and they may require substantial clinical context to interpret. Efforts are underway to create publicly available libraries of phenotypes defined in terms of patterns in codes and discrete data. Tools for translating these phenotypes into queries of local data warehouses and national repositories could make data-driven quality improvement more broadly practical.

Analytics in the form of decision support and monitoring systems

Analytics play a key role in predictive risk assessment, clinical decision support, home health monitoring, finance, and resource allocation. They may also be used to enhance less sophisticated, rules-based systems that are already in use. One of the benefits of EHRs has been the integration of clinical decision support (CDS) systems. CDS systems have been shown to reduce errors and improve clinical outcomes in certain settings, such as pediatric intensive care units, and can result in performance improvement on perioperative quality and process measures. Some CDS systems that are designed to prevent medication errors are based largely on commercially available software packages that rely on simple rules. Often, these products do not provide ideal rule sensitivity, and institutions must perform manual reclassification of drug-drug interactions to improve the efficacy of their CDS systems. Analytics may offer a solution to this challenge, because clinicians can use analytics techniques to query and mine the EHR for meaningful connections, and then synergistically combine the knowledge-based rules with analytics applied to EHR data.

Analytics are used in health care applications outside of the traditional inpatient and outpatient care settings, such as wearable monitors that patients use at home. Wearable health monitoring systems consist of a variety of sensors, actuators, and multimedia devices, and enable low-cost, noninvasive options for continuous monitoring of health, activity, mobility, and mental status, both indoors and outdoors. Thus, wearable monitoring systems provide continuous physiologic data that may reflect the general health of the monitored individuals, and the use of wearable sensors in health monitoring systems is an emerging field that necessitates data mining and analytics of physiologic measurements in a nonclinical setting. Such health monitoring systems may reduce health care costs by disease prevention and enhance the quality of life with disease management.

Architectural framework

The conceptual framework for a big data analytics project in health care is similar to that of a traditional health informatics or analytics project. The key difference lies in how processing is executed. In a regular health analytics project, the analysis can be performed with a business intelligence tool installed on a stand-alone system, such as a desktop or laptop. Because big data are by definition large, processing is broken down and executed across multiple nodes. The concept of distributed processing has existed for decades. What is new is its use in analyzing very large data sets as health care providers start to tap into their large data repositories to gain insight for making better-informed health-related decisions. In addition, open-source platforms such as Hadoop/MapReduce, available on the cloud, have encouraged the application of big data analytics in health care. By discovering associations and understanding patterns and trends within the data, big data analytics has the potential to improve care, save lives, and reduce costs.

The volume and variety of health care–related data continue to grow, spurred on by the increasing adoption and use of EHRs, the expansion of omics data, and the proliferation of actigraphy data from wearable self-tracking devices and mobile applications. The use of these data for health care analytics such as data visualization, patient stratification, predictive modeling, personalized medicine, and drug discovery will increasingly depend on the ability to appropriately collect, curate, and integrate disparate data from many different sources. The linking of many different types of data will allow the analysis and modeling of complex multidimensional interactions that can enable deeper insights. A robust data collection, curation, and integration infrastructure should comprise the following stages: data collection, data understanding, data validation, data cleaning, data integration, and data enrichment.

Beyond early detection of disease, predictive modeling can also facilitate greater individualization of care. Predictive models can enable clinicians to distinguish individuals who will benefit from a specific therapeutic intervention from those that will not. For example, it is known that obesity, with or without the metabolic syndrome, increases children’s risk for developing diabetes, but most of these children do not develop diabetes. The ability to refine the predictive ability for developing diabetes would allow therapies to be better targeted to the individuals most likely to benefit. As the understanding of genomic information increases, the combination of genomic, environmental, and clinical variables in predictive modeling is likely to be critical for the individualization of care in a broad array of disease processes.

Data scientists

The primary value of analytics to end users in health care is the access to actionable information, which needs to be synthesized from various data sources. A personnel resource that has come to the forefront lately is the data scientist, who is an expert analyst, and can serve as a 1-stop shop for data management, analysis, and interpretation of electronic data. In health care, this is particularly important for translating electronic bits into meaningful data. These data scientists need to be proficient in a broad spectrum of analytical methodologies that encompass traditional (biostatistics, epidemiology, discrete-event simulation, and causal modeling) as well as emerging methods (data mining, bayesian statistics, optimization modeling, social network analysis, and agent-based simulation).