A brief Survey of Recent Clinical Dashboards Sheler Maktoobi1 , Michele Melchiori1 University of Brescia, Dip. di Ingegneria per l’Informazione, via Branze, 38 25123 Brescia - Italy {sheler.maktoobi@yahoo.com, michele.melchiori@unibs.it} Abstract. Application of data analytics and visual analytics to the medical domain, and specifically to support clinical activities is more and more both a need and an opportunity. For example, as reported by the US Institute of Medicine, positive transformation of current health systems to make them cost-sustainable and providing higher care qual- ity, requires to capture more clinical data and generate knowledge for better healthcare management and enhancement of medical research. In this brief survey we review recent papers (since 2013 on) describing inno- vative clinical dashboards which promise better support of clinical and medical research activities. We also explain how in some of these tools analytics features are implemented to cope with clinical dataset and get insights from them. We try therefore to identify for each considered tool a set of features concerning data sources, tasks/workflows supported, indicators and their graphical representation, analytics support, target users and data processing approaches. Keywords: Clinical Decision Support Systems, Clinical Dashboards, Data an- alytics, Big data for healthcare 1 Introduction Business dashboards are mostly used to gather summary data and provide neces- sary information to take critical decisions for the management of single business processes and overall organizations, that are decision support systems. Clini- cal dashboards specialize this concept to healthcare management but also to everyday clinical and research activities. Some recent clinical dashboards provide also, often as visual and interactive data visualization features, data analytics and inference capabilities in order to extend the usual capabilities of visualizing key performance indicators (KPI) and summarized data. Actually, application of data and visual analytics to the medical domain and specifically to support clinical activities is currently con- sidered both a need and an opportunity. In fact, as described in a report [8] of the U.S. Institute of Medicine (IOM), transformation of current health systems is needed in order to contain the costs of the systems becoming too complex to manage and to provide high care quality especially for patients with chronic 1 Discovering of Filtering on: papers related Filtering on: the tool has to to: - paper has to support: The selected - clinical describe a - physicians, papers are dashboards specific tool or analyzed against - analytics - publishing - clinical nine points tools for year >= 2013 researchers healthcare Fig. 1. Selection method for the reviewed tools diseases that require better coordination in patient data sharing and manage- ment among healthcare institutions. One consequent recommendation given in the report, is to gather more clinical data and generate knowledge for better coordination and management, for improving care practices and enhancement of medical knowledge. As well as, analytics is also an opportunity. The cited IOM report envisions the idea of US healthcare as continuous learning system where data produced by healthcare processes is used to improve coordination of these processes and focusing on activities that improve patient health. And to accelerate integration of the best clinical knowledge into care decisions. In this perspective, we notice how the recent evolution of the healthcare interoperabilty standard HL7, that is FHIR, 1 can support the spread of analytics tools by providing models and methods for decoupling data sources and analytics applications operating on this data thus permitting reuse of advanced applications on various data sources. Other surveys on clinical dashboards have been previously published, for ex- ample [4] and the more recent one [3]. In our brief survey we use a more specific perspective by describing innovative clinical dashboards that promise to better support clinical and medical research activities. Moreover, we explain how some of these tools provide analytics features that are implemented to deal with large clinical datasets and in order to get insights from them. For each considered tool we describe a set of features concerning: data sources used by the tool, types of tasks/workflows supported, indicators and their graphical representation, ana- lytics support, target users and adopted data processing approach. 2 Method of selection and analysis of the literature Reviewed papers have been selected according the criteria reported in the first three steps in Fig. 1 and analyzed according to nine points reported in the following section. Concerning the papers selection process we used as sources: (i) Google Scholar; (ii) Google; (iii) The Proc. of the 2014 Workshop VAHC 2014 2 . For example, in Google Scholar we made searches based on combinations of groups of terms tool clinical dashboard and medical healthcare analytics. Even restricting these searches to the range of years 2013-2016 we got thousands of results. Thus, for each search we set in Google Scholar the option of ordering the result by relevance and inspected title and abstract of the first 50 results. 1 http://www.hl7.org/fhir/ 2 Proceedings of the 2014 Workshop on Visual Analytics in Healthcare (VAHC 2014), available at http://www.visualanalyticshealthcare.org/proceedings.html. 2 Table 1. Non analitycs-based clinical dashboards Short name Tool class Main indicators Main function- Activities sup- Dimensions Visualiz. alities ported by the mean tool Real-time en- Web-based Daily enroll- Manage multi- Conduction of Hospital Table; rollment dash- dashboard. ment for nine site clinical trial clinical trials. and time Bar- board [7] hospitals partic- enrollment. chart ipating in the study Measuring Maternal- Six clinical Visualization Support hospi- Hospital/ Table Quality in newborn performance and comparison tals and care unit and Maternal- dashboard indicators of of indicators. providers for time. Newborn quality care. quality improve- Care [9] ment. By looking for paper describing software tools and not general discussions or methodologies we identified 12 papers (from all the considered sources). Some of these papers were further removed because they do describe tools aimed at hospital management and not at physicians/researchers. The nine points we considered for analyzing the chosen tools and the re- lated literature were devoted to determine for each tool: 1) purpose and applica- tion, 2) target users, 3) tool class, 4) data sources used by the tool, 5) types of tasks/workflows supported by the tool, 6) indicators the tool is able to display, 7) dimensions associated with the indicators, 8) data processing methodology, 9) Type of information inferable by the analytics functionalities of tool (e.g., the tool discovers and shows correlations among patient features in order to help physicians to identify relevant cohorts of similar patients). For each tool we complete a form with the answers to these questions. These answers were sum- marized and used to fill in the Tables 1 and 2. These tables focus on six aspects we considered relevant for our survey in order to balance general features and domain-specific ones. 3 Comparison and features of selected clinical dashboards The first table describes approaches that do not clearly provide data and visual analytics functionalities. In the second table, approaches that have analytics features are listed. The Main indicators column describes the clinical dashboard indicators (KPIs) for each approach. Moreover, we tried to understand the use of each tool in clinical or research contexts. Thus, the Activities supported by the tool provides information about the tasks or workflows supported. In a perspective of multidimensional data, typical of dashboard tools, the Dimensions column explains, when applicable, which dimensions are applied to indicators. Finally, Visualization means lists diagram types or visualization approaches used in each tool. Non Analytics-based tools. The two tools summarized in Table 1 are clinical dashboards without specific analytic or inferential capabilities. The first one [7] offers support for conduction of clinical trials concerning studying pneumonia and helps management of multi-site enrollment. Physicians can use the enroll- ment dashboard showing various daily enrollment numbers, also represented as 3 Table 2. Analitycs-based clinical dashboards Short name Tool class Main indicators Main function- Activities sup- Dimensions Visualiz. alities ported by the mean tool Patient-Like- Web-based Summary in- Decision sup- Improve Clin- Time Graph Mine A Real Visual dicators from port for: formu- ical Pathway Time [6] Analytics electronic med- lation of specific compliance, rec- Tool for ical records care plan, find- ognize patterns. Clinical (EMR). ing patients’ Decision cohorts. Support. An Analytics Analytics Proportions of Analytics ap- Application of Number of Barchart; Approach tool, Sim- patients suc- proach for analytics to patient per Linechart to Manag- ulation cessfully treated improving clin- EHR; simula- cluster. ing Provider tool. and cost per ical outcomes tion of therapies Treatment patient. by identify- Variety [5] ing successful treatments. A Visual Clinical Objective data Creation and Visualization of Flexible: Barplot; Analytics dashboard. on the patients comparison indicators for a one vari- Scatter- Approach expressed by of cohorts. single cohort of able can plot for Care KPIs like acuity, Comparison of patients and two be shown Processes [1] health status, care processes cohorts. in scatter- total charges. with pathway plots as a guidelines. function of another one. Cancer Cohort Clinical Objective (e.g., Visualization of Creation and Typical Barcart; Visual Analy- dashboard PSA values over multiple patient evaluation of dimen- linechart sis [2] based on the time) and histories and cohorts as inte- sions: time, static and subjective (e.g., identification of grated process. therapy dynamic well being eval- similar patient type. indicators. uation over a histories. scale) data on the patients. barcharts, for the nine hospitals participating in the study. The tool in [9] is a dashboard for measuring and monitoring quality care for maternal-newborns. The purpose is increasing awareness about KPIs identified as relevant (e.g., Proportion of newborn screening samples that are unsatisfactory for testing). Moreover, other targets are reporting on a selection of performance indicators (feedback), comparing performance to established ideal levels (benchmarking) and providing alerts when performance is sub-optimal to trigger action (warn- ings). Indicators are represented in tabular form. Analytics-based tools. The Web-based visual analytics tool in [6] provides clinical decision support. It appears to be the only non domain/pathology specific tool we reviewed. In particular, it creates a transparent, interactive environment that enables a physician to formulate more specific plan for a given patient using real world data from a high number of EHRs (i.e., electronic health records). More- over, it provides support to find patients’ cohorts in order to improve clinical pathway compliance. By using a flexible UI, the physician or team can also ex- plore what-if scenarios that would have previously required statistical/database skills and effort to develop. For example, use of graphics allows the provider to quickly determine if a patients clinical parameter is within given bounds, based on visual comparison with a cohort of similar patients. 4 Also the work [5] applies analytics to EHRs in order to improve clinical out- comes. In particular, it utilizes EHR data to identify different types of successful treatments that physicians deliver for different types of patients with Type 2 dia- betes. Then a simulated environment is used to assigns patients to physicians for better treatment of type-2 diabetes in a clinic setting. Simulations of a diabetes clinic relies on different physician models for treating patients. The simulated clinic comprises models of patients with type 2 diabetes (T2DM) and models of physician decision making processes. The tool clusters patients into groups based on their initial states then assigns a physician model for each cluster of patients. It also permits to evaluate the physician-patient assignments by visu- alizing indicators of treatment success and cost per patient per year, calculated on the outcomes of simulations. We focus now on two interesting tools, the ones in the last two row of Table 2. The first one, in [1], provides visualization of indicators mainly oriented to study either a single cohort of patients, that is a group of patients with similar features, or to study and compare two cohorts (i.e., comparing populations of patients). The application domain is care of asthma in children. The main purpose of this system is to permit analysis of care processes in the reference domain. Basically the system provides functions to: – create/modify patient cohorts based on demographic (age, gender), process (length of stay, number activities), clinical (e.g., acuity, initial clinical respi- ratory score (CRS)) and others; – analyze the care processes of single cohort of patients by visualizing indica- tors values and distributions (e.g., histograms); – comparing a single cohort of patients by visualizing indicators values and distributions (e.g., histograms); In general, the system provides multiple and coordinated visualizations. The tool has not statistical inference capabilities. So identification of correlations in co- horts has to be performed visually. However, it is able to perform process mining concerning care processes and presents in a graphical way frequent subprocesses in care therapies by using Sankey diagrams. Target users are clinical physicians who specialize in the research and treatment of children asthma. The last tool we consider is [2] which permits to visually analyze multi at- tribute datasets including time dependent data. The dataset concerns patient records of patients suffering from prostate cancer. Visualization of large sets of patient histories as well as aggregate visualization of specific aspect of sets of patient histories are offered. That helps to identify patient cohorts based on a visual and inferential support and permits to make shorter the clinical task of identifying such as cohorts. Statistical inference functionalities are implemented in order to identify correlations between features that help physicians identify patient cohorts. In particular, the system offers a guided analysis of correlations between the current patient cohort and all static attributes provided in the data. The correlations are calculated automatically using statistical dependency mea- sures. 5 4 Conclusions In this survey we reviewed clinical dashboards that provide access to informa- tion for clinicians and researcher supporting improved care process and quality. Generally, these tools allow to visualize and compare specific profiles of different patients at the same time and help recognizing patients’ histories. It is interest- ing to notice that some recent dashboards either provide analytics features or are based on analytics approaches, like clustering. This may help physicians to make inferences and/or to discover patterns in large EHRs datasets. Concern- ing visualization approaches, the dashboards here considered adopt simple, and probably familiar to physicians, metaphors like: barcharts, linecharts and tables. Since the growing interest and expectation for applying big data techniques to medical research and practice, we suppose more and more analytic-based medical applications will be soon presented. References 1. R. C. Basole, H. Park, M. Gupta, M. L. Braunstein, D. H. Chau, and M. Thomp- son. A visual analytics approach to understanding care process variation and con- formance. In Proceedings of the 2015 Workshop on Visual Analytics in Healthcare, VAHC ’15, pages 6:1–6:8, New York, NY, USA, 2015. ACM. 2. J. Bernard, D. Sessler, T. May, T. Schlomm, D. Pehrke, and J. Kohlhammer. A visual-interactive system for prostate cancer cohort analysis. Computer Graphics and Applications (CG&A), IEEE, 35(3):44–55, 2015. 3. Dawn Dowding, Rebecca Randell, Peter Gardner, Geraldine Fitzpatrick, Patricia Dykes, Jesus Favela, Susan Hamer, Zac Whitewood-Moores, Nicholas Hardiker, Eliz- abeth Borycki, and Leanne Currie. Dashboards for improving patient care: Review of the literature. International Journal of Medical Informatics, 84(2):87 – 100, 2015. 4. M. Egan. Clinical dashboards: impact on workflow, care quality, and patient safety. Critical care nursing quarterly, 29(4):354–361, 2006. 5. Ramsey G., Gupta A., and Kwon Y. An analytics approach to managing provider treatment variety to improve patient outcomes for a type-2 diabetes clinic. In Proceedings of the Ninth Midwest Association for Information Systems Conference, MWAIS 2014, Ames, Iowa, 2014. 6. P. Li, S. N. Yates, J. K. Lovely, and D. W. Larson. Patient-like-mine: A real time, visual analytics tool for clinical decision support. In Big Data (Big Data), 2015 IEEE International Conference on, pages 2865–2867, Oct 2015. 7. William A. Mattingly, Robert R. Kelley, Timothy L. Wiemken, Julia H. Chariker, Paula Peyrani, Brian E. Guinn, Laura E. Binford, Kimberley Buckner, and Julio Ramirez. Real-time enrollment dashboard for multisite clinical trials. Contemporary Clinical Trials Communications, 1:17 – 21, 2015. 8. M. Smith, R. Saunders, L. Stuckhardt, and J.M. McGinnis. Best Care At Lower Cost: The Path To Continuously Learning Health Care In America. Institute of Medicine, National Academies Press, Washington, USA, 2012. 9. Ann E. Sprague, Sandra I. Dunn, Deshayne B. Fell, JoAnn Harrold, Mark C. Walker, Sherrie Kelly, and Graeme N. Smith. Measuring quality in maternal-newborn care: Developing a clinical dashboard. Journal of Obstetrics and Gynaecology Canada, 35(1):29 – 38, 2013. 6