Fostering a Data-Driven Hospital Environment

Data is transforming the world in which we live, from smart refrigerators that notify us of when we’re running low on milk to the emergence of self-driving cars. While the healthcare sector may be considered ‘data rich’, the actual use of data and advanced analytical approaches such as machine learning in practice is quite limited despite its enormous potential to improve clinical care and efficiency. Ironically, healthcare is among the most ‘evidence-based’ sectors despite our limited ability to truly leverage our existing data, relative to other sectors, to drive our business. Numerous challenges to creating a ‘data driven hospital environment’ are often raised, as highlighted below. While we are only starting our journey in fostering such an environment, we’d like to highlight 4 (among many) key facets, namely infrastructure, people, process, and culture.

Infrastructure: Hospital adoption of electronic medical records and digital data greatly facilitate data acquisition and analytics. The Electronic Medical Record Adoption Model (EMRAM) assesses an organization’s degree of adoption of electronic medical records and analytics, and is composed of 7 stages ranging from Stage 0 where practice is almost entirely paper chart based to Stage 7 which is a nearly paperless environment that enables analytics to improve care¹. While the majority of hospitals in the United States were estimated to be at Stage 5 or higher in electronic medical record (EMR) adoption, less than 6% of Canadian hospitals achieved this status in late 2017¹. Clearly, hospitals, especially in Canada, need to do a better job at digitizing data. However, once data is available electronically, it must stored and organized in a manner that facilitates rapid access and advanced analytics. Typically, data is siloed in different ‘data buckets’ within hospitals that make it difficult to access and analyze efficiently across these silos. Ideally, relevant data should be held in a common area such as a ‘data lake’². However, data lakes often contain data in their ‘natural’ state, which isn’t ideal for analytics. The process for linking and cleaning the data can be arduous, and busy clinicians and management leadership may not have the time or patience to wait if this process needs to be repeated for every data-related problem. In contrast, data warehouses often contain linked, ‘cleaned’ data that are readily accessible and easier to analyze, although they tend to be quite expensive and perhaps more cumbersome to modify. At our institution, we were very concerned about the timeliness of getting good, reliable data and analytics in a timely manner to our clinicians and management leadership. We also recognized the need for some flexibility to accommodate evolving data sources. Consequently, we opted for both – we have established a ‘raw staging area’, which serves as a data lake where data from key ‘data buckets’ (source systems) are housed, that regularly feeds into a comprehensive data warehouse housing over 2,200 variables that are ‘analytics ready’. For our data scientists, we’ve also brought in tools such as R, Python and PyTorch, and TensorFlow as well as software for natural language processing, simulation modeling, and optimization. Further, we work collaboratively with our Hospital Decision Support team who actively uses Tableau as our core data visualization tool to provide dashboards and reporting to clinical and management leadership

People: Having a fulsome data infrastructure isn’t very helpful unless there are talented individuals who know how to leverage this data for insights to meet the needs of end-users and convey analyses in an easy-to-understand, actionable manner. We have established a data science team at our institution – the Li Ka Shing Centre for Healthcare Analytics Research and Training (LKS-CHART) – whose goal is to work with end-users (clinical and management leadership) on issues where data and analytics can drive change. We have three different ‘streams’ of data scientists, namely computer science with a focus on machine learning, operations engineering with a focus on simulation modeling and optimization, and traditional biostatistics. While many people claim to be ‘data scientists’, finding highly qualified data scientists who are truly knowledgeable is a difficult process. Our recruitment process for data scientists is highlighted here. Further, our colleagues in Decision Support have established a similar process to recruit individuals with expertise in data visualization and use of business intelligence tools.

Process: A fulsome data infrastructure and highly skilled data scientists is wonderful for academics, but a process to bridge these with end-users is needed to enact change. A major issue we struggle with is data governance and access, which continues to evolve. As with most industries, privacy and security are extremely important considerations in determining who gets access to which data and when. Two main groups of ‘data handlers’ have access to our data infrastructure, namely Decision Support and the LKS-CHART. Two ‘main’ groups of end-users at our hospital are clinician leaders and management leaders. Both end-user groups have access to highly developed data dashboards through Decision Support that report on key metrics and processes, although not everything they could want is represented in these dashboards. For specific quality improvement initiatives, a process for data access is highlighted here. For specific research projects, a submission to the Research Ethics Board (REB) at our institution is required and data would be accessed through our Decision Support team in these cases. For specific projects that are more analytics intensive, end-users are able to approach the LKS-CHART team once a project proposal is signed off by their Program Director, Medical Director, and Division / Department Head. This ensures that proposed LKS-CHART projects have buy-in from key decision-makers who are committed to addressing the issue at hand. The LKS-CHART focuses on initiatives that have the potential to improve patient outcomes and/or hospital efficiency. Further, most projects are reviewed at a regularly scheduled advisory group meeting composed of senior hospital leadership (Vice-President level, privacy and risk, quality improvement, and decision support) for institutional support. Through this process, end-users engaged in ‘day-to-day’ activity can raise important questions that can be escalated to senior leadership. End-users are expected to be heavily engaged in the analytics initiative – our typical structure is biweekly meetings with end-users and data scientists for 3-6 months – to ensure their needs are being addressed. For example, our data scientists have worked closely with our clinical and management team in our emergency department (ED) to forecast expected ED patient volumes days and weeks in advance with > 90% accuracy. The metrics and visuals were all driven by the end users, which maximizes uptake of the end product since the output of complex models are represented in easy-to-understand, actionable ways.

Culture: Perhaps the most challenging issue is creating a ‘data culture’. With the establishment of appropriate data infrastructure, buy-in from senior leadership, access to highly skilled data scientists who are oriented to meet end-user needs, and a highly motivated and engaged end-user base, there certainly is a ‘buzz’ at our institution. That being said, we still have a long ways to go. Numerous data-driven initiatives are currently underway (see projects here) and we are actively working on implementation with hopes to demonstrate improvements in patient outcomes and hospital efficiency in the future. The ability to show ‘value’ to end-users is expected to generate greater interest in data and advanced analytics to promote a data culture. This interest will hopefully drive greater data literacy and awareness – we currently have institutional videos addressing data access, basic statistics, and dashboards and have established an educational plan – and excite each and every member of our hospital community, including the patients we serve. We look to role models such as Intermountain Health³ to continue to learn how to drive a data culture at our institution so that using data becomes pervasive⁴ in everything that we do.

References:

1. http://www.himssanalytics.org/emram
2. https://en.wikipedia.org/wiki/Data_lake
3. https://sloanreview.mit.edu/case-study/when-healthcare-gets-a-healthy-dose-of-data/
4. https://www.gartner.com/smarterwithgartner/the-key-to-establishing-a-data-driven-culture/