5.1 An introduction to data management and visualization

Researchers and clinicians have been collecting many types of data for years—so why is data getting so much attention now? A number of technological drivers have led to an explosion in the amount of data. According to a 2017 report, we create 25 quintillion bytes (or 2.5 million terabytes) of data every single day (IBM Marketing Cloud, 2017). Although some of those data comprise cat videos or the latest meme, but the growth of data is not limited to this type of popular data—clinical and research data are also being generated at a rapidly increasing speed. For example, the size of the world’s genomic data doubles every 7 months; depending on how new technologies develop, we could sequence up to 2 billion human genomes by 2025, which would account for more data than YouTube or Twitter (Stephens et al., 2015).

The huge growth in data is due in part to the fact that we now have the technology to generate a large amount of data more quickly and cheaply than ever before. The process of sequencing the first human genome cost at least $50 million and took about 13 years; today, a human genome can be sequenced in about 24 hours for less than $1000 (National Human Genome Research Institute, 2016). Moreover, the cost of storing that data has dropped precipitously, meaning that not only can we cheaply generate data, we can also continue to store them at a comparatively low cost (Komorowski, 2014). In addition, with the increase in internet transfer speeds, researchers often do not need to store data themselves. Instead, data can be stored using cloud services, often for less than it would have cost to maintain a local copy, and transferred or accessed as needed.

The move to electronic health records (EHRs) has also helped to create a wealth of digital data. Analog patient records lacked widespread utility beyond the clinical care of the individual patient, but EHRs make it possible for data to be used not only to improve the care of the individual patient, but also to the benefit of the broader population. The Centers for Medicare and Medicaid Services (CMS) created criteria to incentive “meaningful use” of EHR systems, among them a number of criteria that address public and population health (Centers for Medicare and Medicaid Services, 2016). While patient privacy must be protected, hospitals and providers can use EHRs to submit data to public health registries, clinical data registries, and other centralized sources that allow for the collection of data for epidemiological and clinical research, among other uses. Utilizing EHR data allows researchers to include more patients than they could ever hope to recruit for a trial and saves time by making use of existing data rather than collecting their own new data. Such approaches have been used in a variety of contexts in the last several years (Dewdney & Lachance, 2017; Ehrenstein, Nielsen, Pedersen, Johnsen, & Pedersen, 2017; Schofield, 2017).

In addition to technological drivers, new policies about data management and sharing have also been instrumental in shaping the current research data landscape. In 2011, the National Science Foundation (NSF) became one of the first US federal funders to require that all grant proposals include a data management plan (National Science Foundation, 2010). This policy helped address a problem that has become increasingly recognized as significant in many scientific communities: how to ensure preservation and accessibility of federally funded research data. Two years later, the US Office of Science and Technology Policy (OSTP) issued a memorandum directing federal agencies that spend more than $1 million annually on research funding to develop policies regarding access to the results of federally funded scientific research (Holdren, 2013). Specifically, the OSTP memo instructed agencies to consider how to provide access to not only the peer-reviewed articles resulting from their funded research, but also the data underlying these results. As a result, federal agencies funding a variety of scientific research areas have since developed policies regarding article and data sharing (SPARC, 2016).

Because the technological landscape of scientific culture around data is changing so rapidly, funder policies regarding data are likely to evolve significantly in the next decade. Informationists who support research data services should therefore make an effort to remain apprised of the policies that are likely to apply to their users. As of this writing, NIH's data-related guidance comes primarily from two broadly applicable policies, along with a handful of policies more specific to certain types of data and subject areas (National Institutes of Health Office of Extramural Research, 2016b). The NIH Data Sharing policy applies to grants since 2003, but only those receiving more than $500,000 per year in direct costs (National Institutes of Health Office of Extramural Research, 2016a). The policy specifies that researchers must make the research data underlying their results available no later than acceptance for publication of the article reporting on their findings, though it does not prescribe any specific repository or mechanism for doing so. The other major policy, the NIH Genomic Data Sharing (GDS) policy, applies to all NIH-funded research, regardless of funding mechanism or amount, that generates any type of human or nonhuman genomic data (National Institutes of Health Office of Science Policy, 2017).

Although NIH has addressed data sharing through such policies, at present, the NIH does not require researchers to submit a data management plan; however, plans to require a “data management and sharing plan” (DMSP) are included in the NIH’s 2015 response to the previously discussed 2013 OSTP memo (National Institutes of Health, 2015). The proposed DMSP policy will require all researchers, regardless of funding type or amount, to submit a detailed plan outlining how they will manage their data throughout the research cycle, as well as how they will make the data available upon publication. Although no start date has been formally announced as of this writing, researchers and others who receive NIH funding of any type should expect that they will be required to include DMSPs with submissions in the near future.

In addition to policies from funders, researchers increasingly face other data sharing mandates that may require them to make their data publicly available. Many journals and publishers, such as PLOS, the Nature Publishing Group, and Science journals, all require that authors make their data available upon publication. These data may be crucial for readers to fully understand and assess the findings of an article, so making data available is a logical step for journals aiming to increase transparency and reproducibility of the articles they publish. Although data sharing policies do not specifically address data management, the expectation that data will be eventually shared and necessitates proper management to ensure that data will be ready to share when the need arises.

All of the technological, policy, and cultural changes discussed here have resulted in a dramatically shifting academic and clinical landscape around how data are handled. Despite the increasing importance of being able to competently handle large amounts of data, many researchers and clinicians are not introduced to these skills in their training programs, and those individuals already working in these settings often have difficulty in finding time to learn new skills or even confidently identifying the types of skills that they would need to learn to be more effective in their work. Informationists can play an important role in filling these knowledge gaps by providing targeted training, offering individual or group consultations, and developing services aimed at addressing problems in these areas. While many examples of informationists providing support across the entire research data life cycle exist, this chapter focuses on two specific areas that are of particular interest and need for researchers and clinicians: research data management and data visualization.

5.1.2 Making sense of data through visualization

While data visualization may seem simply like a way to add some visual interest to an article or scientific poster, visualization can in fact play an essential role in analyzing and understanding data throughout the research process and within the context of clinical care. Presenting data in visual form allows us to utilize some of our most powerful perceptual abilities to detect meaning in data. Effective visualizations allow users to interact with and manipulate data, view relationships among multiple variables, track changes in data over time, and generally gain insights that would not be possible simply by looking at the raw data.

To understand the importance of visualization in exploratory data analysis and detection of patterns and meaning, consider Table 5.1, a set of four datasets known as Anscombe’s quartet (Anscombe, 1973). Each of the four sets, I–IV, consists of 11 pairs of X and Y values. Examining the raw X and Y values likely provides little in the way of useful information, even to an astute reader. Conducting statistical calculations on each set of values might lead the reader to believe that the four datasets are similar; after all, the values for five different descriptive statistical measures are nearly identical across all the four datasets.

Table 5.1

Anscombe’s quartet, a classic demonstration of the power of data visualization

I		II		III		IV
X	Y	X	Y	X	Y	X	Y
10	8.04	10	9.14	10	7.46	8	6.58
8	6.95	8	8.14	8	6.77	8	5.76
13	7.58	13	8.74	13	12.74	8	7.71
9	8.81	9	8.77	9	7.11	8	8.84
11	8.33	11	9.26	11	7.81	8	8.47
14	9.96	14	8.1	14	8.84	8	7.04
6	7.24	6	6.13	6	6.08	8	5.25
4	4.26	4	3.1	4	5.39	19	12.5
12	10.84	12	9.13	12	8.15	8	5.56
7	4.82	7	7.26	7	6.42	8	7.91
5	5.68	5	4.74	5	5.73	8	6.89
Mean(x) = 9 SD(x) = 3.32 Mean(y) = 7.5 SD(y) = 2.03 Corr coef(x, y) = 0.816		Mean(x) = 9 SD(x) = 3.32 Mean(y) = 7.5 SD(y) = 2.03 Corr coef (x, y) = 0.816		Mean(x) = 9 SD(x) = 3.32 Mean(y) = 7.5 SD(y) = 2.03 Corr coef (x, y) = 0.816		Mean(x) = 9 SD(x) = 3.32 Mean(y) = 7.5 SD(y) = 2.03 Corr coef (x, y) = 0.817

The pairs of X and Y values in these four datasets are identical (or nearly identical) in several statistical measures, including mean and standard deviation of x and y and correlation coefficient.

Upon viewing the data visualized as a scatterplot in Fig. 5.1, it becomes obvious that the four datasets are actually quite different. The red regression line, which demonstrates an estimate of the relationship between the X and Y values, is identical across all four datasets, yet the position of the points in each chart differs dramatically. Charts III and IV, for instance, demonstrate how a single outlier can significantly skew the statistical analysis of a dataset. As Anscombe’s quartet demonstrates, visualization can provide a clear understanding of data distributions and patterns in ways that statistical analysis alone cannot.

Visualization can be a highly effective means for understanding the types of large and complex datasets that characterize both scientific research and clinical practice. For example, EHRs contain large amounts of data about many patients, including both narrative and structured data, but the structure and organization of these data make it nearly impossible to detect patterns about what is happening at the care setting on the whole, locate similarities between patients, or even detect some patterns of disease within an individual patient. Software that presents the same data visually can make such patterns, similarities, and anomalies easily detectable (Clark, Grove, Shneiderman, & Chute, 2012; Monroe, Lan, Lee, Plaisant, & Shneiderman, 2013). Within biomedical research, interactive visualization in particular can be especially helpful in detecting patterns and discovering knowledge within large datasets, such as those frequently used in genomics and related disciplines (Turkay, Jeanquartier, Holzinger, & Hauser, 2014).

5.2 Roles for informationists in data management and visualization

Given the quickly changing landscape of data management and visualization, many researchers and clinicians face significant challenges in keeping up with policies and mandates as well as finding the best solutions to the data problems they face in their day-to-day work with various types of data. Informationists have a variety of opportunities to work with their users to provide helpful solutions to many of these data management and visualization challenges. Many libraries have started to offer data services, and a growing number of informationists and librarians work as dedicated data specialists.

Although informationists may need some additional training to successfully provide a high level of data services, many informationists will find that their skills and experience will enable them to effectively and meaningfully engage with their users on data-related issues. Many of the information management skills that informationists are trained to utilize are as applicable to data as they are to other types of information. Topics such as information-seeking behavior, metadata and information standards, and information literacy all have corollaries to data-related work. Informationists who have already worked in positions where they are embedded with users are also at an advantage in that they have likely obtained some subject matter knowledge, either through formal training or simply through working closely with users in clinical or research settings. In addition, such informationists are likely familiar with the typical practices and workflows of their users, and are therefore aware of the types of data their users work with and the challenges they face in doing so.

Whether an informationist has a long history of working with their users or is relatively new to the field, it is important to recognize that gaining at least some background information about the setting in which their users work is essential to successful provision of data services. Informationists will be able to best serve their users by assessing users’ needs and knowledge gaps and focusing on developing the types of services that will best address those challenges. A one-size-fits-all approach is unlikely to be effective in developing data services, since users within different fields and at different institutions likely have different needs; therefore, this section aims to provide some ideas about general services and approaches that informationists can adapt to their particular institution’s or groups’ needs.

5.3 Case study: Data services at the NIH library

The NIH Library began to see a need among their users for support for data management, around the same time that many libraries started to hire positions devoted full time to data support. I was hired for the Research Data Informationist position at the Library in 2013 and began development of the Data Services program. Knowing that researchers at the NIH likely had unique needs, we conducted a needs assessment to determine how best to serve our users. When asked which data-related topics were most relevant to their work, users reported that most of the tasks were highly relevant, but they generally rated their expertise in these tasks as lower, suggesting a need for additional training and support (Federer et al., 2016).

The first classes that I offered at the Library primarily focused around general data management. While these classes were well-attended, I realized from interacting with users and feedback obtained on class evaluations that users needed more in-depth support in data science-related topics, particularly data visualization and programming with R and Python. While users were generally highly interested in such topics, few resources existed at NIH for them to learn these skills. With the support of my library, I began a course of study to build my own expertise so that I could better provide the support that our users needed. I completed a variety of both free courses, including Coursera’s Data Science Specialization and Indiana University’s Information Visualization Massive Online Open Course (IVMOOC), and paid certifications, including New York University’s Data Visualization graduate certificate and Georgetown University’s Data Science graduate certificate.

Although building expertise in R programming, data visualization, and advanced data management was a time-consuming process, the investment of time and effort paid off when I was able to begin offering specialized courses and consultations. After 3 years of offering courses in R and data visualization on an approximately monthly basis, classes are still wait-listed almost every month. These classes also offer a means for gaining rapport with users and an entrée into working with them to provide more significant data-related support, sometimes meeting needs that they did not even know they had. The following case study provides an example of how an informationist might provide support across the research data life cycle, starting with just one contact.

5.4 Becoming a successful data informationist

As this chapter has emphasized, much of the knowledge and expertise needed to provide data services align with the skills many informationists already have. Informationists who have experience working with clinical or research groups would likely have gained knowledge of these groups’ work and needs that provide a basis for understanding the types of data services that could be useful to these groups. In addition to the subject matter expertise needed to work with such groups generally, the successful data informationist should also have an awareness of data policies and mandates to which their groups may be subject, as well as a familiarity with some of the tools that can be used in working with, managing, and visualizing data.

Fortunately, many resources exist for gaining the skills and knowledge necessary to provide data services. The New England Collaborative Data Management Curriculum (NECDMC) provides a useful starting point for informationists who would like to start teaching data management at their institution (Martin, Creamer, & Kafel, 2013). The NIH’s Big Data to Knowledge (BD2K) initiative has also funded the development of open educational resources, some geared specifically toward librarians and informationists, that will provide new learning opportunities for those interested in developing skills to support data science and data management. A number of continuing education courses related to data services and data management are offered through the Medical Library Association, through webinar as well as in-person at the organization’s annual meetings. Online courses such as those offered through Coursera can be helpful in learning some of the technological skills relevant to working with data.

In addition to online learning opportunities, the quickly growing community of librarians and informationists who provide data services has begun to create group and resources that allow them to network, provide each other support, and share ideas. Many library and information science journals have begun to feature articles that discuss data services within specific domain contexts; the Journal of eScience Librarianship focuses specifically on “services related to data-driven research in science, technology, engineering, math, social sciences, medicine, and public health” (University of Massachusetts Medical School and Lamar Soutter Library, 2017). Although no large-scale professional organization dedicated specifically to data-related library work yet exists, a number of special interest groups exist within larger professional organizations, such as the Data Special Interest Group within the Medical Library Association. Informal groups, such as the email list Datacure, also allow for networking among like-minded professionals. Even more casually, many data librarians and informationists communicate on Twitter using the hashtag #datalibs.

Any informationist interested in starting a new data service should consider how to do so using evidence-based practices to ensure success. As this chapter has discussed, different user groups at different institutions may have widely diverging needs. While keeping up with the literature and knowing what types of services other informationists have developed can be helpful, the success of a service at one institution does not necessarily mean it will be well received in other contexts. The first step to ensuring a successful data service is identifying the users’ most pressing needs and considering how the informationist can best address them. Ongoing evaluation and assessment is also essential to ensure that the service is meeting users’ needs.

Informationists embarking on data services will likely find that outreach and marketing are crucially important to the success of such new services. Although librarians and informationists at hundreds of libraries around the world have been providing data services for years, many users still do not think of libraries as a place to look for assistance with their data problems. In fact, in many institutions, the library is the only place such services are offered, and users are happy to discover that solutions exist for many of their data problems. Although developing and implementing data services can be time consuming and challenging, informationists will likely find that the work is rewarding and can be of great value to their users.