Today's nursing executives and managers are inundated by a tsunami of data from a variety of sources, including inpatient and ambulatory electronic health records (EHRs), finance and accounting data, claims data, human resources reports, demographic data, national and internal benchmarking data, research data, quality improvement and process improvement project outcomes data, and satisfaction reports for patients, physicians, and employees. Large-scale electronic repositories are emerging that offer nursing management vast opportunities to discover patterns hidden in the data that could significantly improve patient care and management practices.1
The Lack of Usable Data in Nursing
Knowledge discovery and data mining (KDD) is an interdisciplinary area focusing upon methodologies for extracting useful knowledge from data.2 Knowledge discovery and data mining techniques can identify and categorize patterns while artificial intelligence can create computer algorithms that can predict events. The challenge of extracting knowledge from data draws upon research in statistics, databases, pattern recognition, machine learning, data visualization, optimization, and high-performance computing, to deliver advanced business intelligence and Web discovery solutions. These advanced computational methods have the capacity to identify patients at risk for such conditions as infections, stroke, heart failure, falls, and pressure ulcers, as well as potential readmission to the hospital. However, most nursing departments are just learning about the opportunities that big data and data science can contribute. Ironically, although health systems of today are often oversupplied with data, it is a lack of usable data that is the greatest barrier allowing nursing management to accelerate improvements in patient care and patient safety.
Although the potential for big data to support providers in their clinical decision making is evident, identifying and installing the crucial infrastructure to successfully demonstrate results are proving to be elusive and complex. Conducting studies on large-scale data sets requires the right combination of human capital (data scientists, data engineers, and domain experts), hardware considerations (processing speed, storage, number of servers), database platforms (relational, hierarchical, networked, object oriented), and software applications. Internet giants such as Google and Amazon have been at the forefront of developing the capability to analyze big data.
However, the shift from small data studies such as the traditional clinical trial to big data studies such as predictive analytics has created new challenges for data scientists. Knowledge discovery and data mining faces severe challenges because of the sheer volume, velocity, and variety of data being processed today, most of which require data scrubbing, normalization, and feature extraction because of incorrect, incomplete, improperly formatted, or duplicated entries. It is estimated that most of time required to complete a big data study is consumed by preprocessing. The phrase "garbage in, garbage out" is particularly applicable to KDD and projects.
Standardized Nursing Terminologies
Although the barriers to implementing a successful big data program are evident, nursing has some unique challenges.3 Foremost is nursing's ability to document and demonstrate its contribution to clinical and administrative outcomes through discoverable, consistent, and measurable data. Before the advent of EHRs, nursing clinical documentation was primarily represented by personalized and unit-specific methods. Consequently, a wide range of terminology arose to describe similar types of care.4 For example, consider documentation related to emesis. Nurses frequently document that the amount of emesis is small, medium, or large. But these terms are subjective and lack the precision and standardization needed to be "usable" data in studies.
Searchable Data Entry
The problem is further complicated by the fact that too much of nurses' clinical documentation is stored as unstructured text, requiring the additional step of natural language processing to extract key terms. Thus, the entire preprocessing effort is incrementally more challenging. This makes it difficult to compare nursing documentation to system clinical protocols and quality standards, and much less use in comparative studies or predictive analytics.
The Benefits of Data Models
As a result, the absence of standardized nursing terminology (SNT) in health systems data repositories adds to the existing burden of data preprocessing. Investigators are often tasked with having to map multiple terms describing the same care to a standardized nursing language (SNL) such as Systematized Nomenclature of Medicine-Clinical Terms or Logical Observation Identifiers Names and Codes. This is a labor-intensive task that contributes to delays in providing nursing leaders with the analytical benefits from big data studies. Rather than attempting to retrofit stored clinical documentation into an SNL after it has been entered, it is more efficient to build a "data model" that automatically maps terms to an SNL as they are entered in an EHR.
As previously described, a data model organizes data elements and standardizes how they relate to one another.1 Effective models provide an accurate distillation of clinical data in a consistent, safe, and meaningful way and have the capacity to adapt to changing information needs in an EHR. A data model is a blueprint of how data are structured around clinical concepts and describes the elements used in computer fields located within the EHR. These data elements include the use of SNT that can be used to label and organize computer fields for vital signs, assessments, interventions and nursing diagnosis.
The development of a data model follows the path described in Figure 1. The 1st step is to identify a clinical data model topic such as pain management, fall prevention, or pressure ulcers. The next step is to develop a list of concepts from research questions, clinical guidelines, and literature that describe care surrounding the topic.5 An SNL is an excellent resource for this step. For example, the topic of "Pain Management" may include subcategories that help to refine the understanding of the clinical observations, such as the type of pain, location, duration, a pain scale rating, and other relevant facts that can lead to improved diagnosis and treatment. Concepts are then mapped to electronic clinical documentation forms such as flowsheets and validated by an evaluation team. The Figure, Supplemental Digital Content 1, http://links.lww.com/JONA/A471, provides a description of a data model and how concepts can map to a standard flowsheet used for clinical documentation in an EHR. The completed data model is a great resource to supply terms for new dropdown menus in EHRs that will improve accuracy and reduce documentation time.
Conclusion
The emergence of large-scale electronic repositories populated with disparate data from EHRs and other sources has provided nursing management with innovative opportunities to reveal patterns of patient care hidden. Standardized definitions for concepts integral to big data exploration are also helpful (see Document, Supplemental Digital Content 2, http://links.lww.com/JONA/A472). Appropriately critiquing data will lead to improvements in patient care and safety. However, conducting big data studies requires a multidisciplinary team with new expertise and extensive resources to fully realize the benefits. One barrier that needs to be overcome is the need to standardize nomenclature. An additional barrier is the design and development of documentation workflow enhancements to the EHR that will simplify the nursing documentation process and make KDD available to nurse leaders.
References