Health systems need to partner with universities to solve the inadequate supply of analytic talent

This week, I had the privilege of participating as a reviewer of a master’s thesis (“major project”) for a candidate for a master’s degree in economics at the University of Windsor.  The candidate’s project was a decision-analytic model that explored a patient’s decision to seek care and to pursue second opinions and the physician’s decisions regarding diagnostic testing, implemented as a simulation written in Python.  I got a chance to attend the candidate’s defense of the work, which led to his successful attainment of his degree.  I am always encouraged when young people show interest in healthcare analytics and always inspired by the rituals of higher learning.

The experience got me thinking about our general challenges in the identification, recruitment and development of healthcare analytic talent, and the proven and promising strategies to meet those challenges.  This topic is important because I believe that one of the root causes of our slow progress in transformational improvements to our healthcare system is a lack of access by health care provider and payer organizations to analytic talent and the analytic outputs they can produce. 

In this post, I will emphasize three proven tactics and three promising tactics.  The proven tactics include (1) developing what I call a “community of analysts,” (2) establishing analytic career pathways, and (3) collaborating with universities.  The three promising tactics include (1) the use of clinical decision-analysis projects to foster learning, (2) the provision of what I call “seed papers” to enable such projects to be on target and successful, and (3) the establishment of open-source modeling infrastructure to increase the value and effectiveness of such projects over time.

Sources of Analytic Talent

From the perspective of health care delivery and payer organizations, useful analytic outputs can come from external sources, such as academic institutions, government agencies and consultancies.  As a consultant, I benefit when organizations look outside for analytic talent. But the most valuable analytic outputs are the ones that come from internal staff that know the organization and can pursue lines of inquiry and tell actionable analytic stories over longer periods of time.  Internal sources of analytic talent include centralized, specialized analytics departments, such as the Center for Clinical Effectiveness that I led at Henry Ford Health System in the 1990s, the Biostatistics and Clinical Epidemiology Department that I established at Blue Cross Blue Shield of Michigan in the 2000s and the Medical Economics and Claims Analytics Department that I helped establish at Trinity Health in the 2010s.  Or they can come from decentralized analytic staff positioned within various clinical operations departments, regions, and facilities. As I’ll describe below, both have important roles to play.

Causes of deficiency of supply of analytic talent

There are a number of reasons why analytic talent is so scarce in healthcare organizations.  The most actionable cause of scarcity in the short term is defective knowledge of leaders of health care organizations.  In most organizations, leaders do understand the importance of analytics, and they are acutely aware of their lack of access to good analytic outputs.  I can’t remember the last time a leader said to me “I’m happy with our data and our analytics.”   But many leaders think that the problem is that they just do not have the right data reporting application.  I’ve seen many organizations go through cycles of investment in new reporting applications, each time leading to frustration because the reports do not make sense due to problems with the source data or because the reports are not designed to support the specific decisions they have to make.   Waves of new technologies are touted as the solution to these problems — including “data warehousing,” “business intelligence,” “big data,” and now the biggest of all, “AI.”   Hope springs eternal that the next shiny widget will auto-magically solve the problem.  I’ve lamented about this problem in many previous posts, such as this one back in 2011: We don’t need more data. We need better story-telling. – Reward Health. As a result, many health care leaders seem reluctant to recognize that useful analytic outputs require talented insightful human analysts.  As a result, some leaders fail to invest enough time and budget to the recruitment and development of analytic talent within their organizations. 

Note that I am talking about analytic talent.  I use that word to emphasize the idea that there are innate cognitive capabilities for analytic thinking.  Training and experience are necessary and important, but as with opera singers (and, to my dismay, swimmers), so is natural talent.  The last thing we need is to increase the supply of people lacking analytic talent that believe themselves to be analysts and that get hired or engaged into analytic roles, occupying spots where talented analysts could otherwise be.  Although there can be value in developing easy-to-use, intuitive analytic technology to enable more people to participate productively in analytic work, such tools sometimes have the disadvantage of empowering people lacking analytic talent to masquerade as talented analysts, clogging up the lanes in which valuable analytic outputs are intended to flow.

In this context, I have observed that the leaders of many otherwise sophisticated healthcare organizations lack of ability to recognize analytic talent and to distinguish real talent from masqueraders.  As an employee or consultant to many health care organizations, I have spent lots of time over many years assisting organizations in assessing and recruiting talented analysts. Such work has always seemed to pay dividends.

Developing analytic talent through a community of analysts

As noted above, analysts within healthcare organizations can be positioned in centralized or decentralized settings.  In my experience, the most productive and impactful analysts are those that reside in the decentralized settings.  By working in or near the operational front lines of an organization, analysts gain the context necessary to understand the challenges and questions that are the subject of analyses. And they gain access to facilitate the transfer of analytic outputs to the front-line leaders most directly able to put the insights to good use. 

However, such decentralized settings are not ideal for the development of analytic talent.  Decentralized settings do not have sufficient budget to hire a substantial group of analysts, nor do they have sufficient demand to keep such a group productively engaged.  Therefore, decentralized settings tend to have one or two analysts. They try their best to figure out how to do things on their own.  It can be lonely work.  Such analysts can feel frustrated and stuck. 

One important strategy to overcome this problem is to intentionally establish and nurture a “community of analysts,” allowing people to learn from one another.  The organization can bring decentralized and centralized analysts together for retreats and lunch-and-learn sessions. They can provide tools to foster collaboration and sharing of ideas and practices.   I’ve built such a community of analysts in a number of institutions and advocated for its development in many other institutions for which I have provided consulting support.  Such initiatives can promote professional development and sharing among analysts positioned in both centralized and decentralized organizational settings.  

Developing analytic talent through career pathways

To achieve deeper development of the analytic talent within a healthcare organization, it is necessary to create career development pathways. The idea is to initially hire new analysts into centralized analytic departments, where they can work directly with more experienced analysts. Then, as they mature, such analysts are intentionally moved to various decentralized areas with occasional stints back in the centralized analytic areas.  Such career development pathways make the analysts feel valued, improving retention and morale.  And, in my experience, such career pathways can provide a large benefit to an organization’s culture by reducing the silos separating departments and creating strong professional networks that cross the “whitespaces” in the organizational chart.

Opening the University pipeline for analytic talent

The strategies described above help to make the most of the analytic talent already existing within healthcare organizations.  But the deeper cause of analytic talent scarcity in healthcare organizations is a lack of supply.  There is just not a sufficient population of talented analysts available to be hired or engaged by health care organizations.  To address deficiencies in the supply of talented analysts, we need to go upstream to identify more people with analytic talents in graduate or undergraduate university settings and then attract them to the health care field.

Health care organizations can start this process by taking more proactive steps to build relationships with university departments that produce analysts.  Analytic talent most often begins development among students studying social science disciplines such as economics, epidemiology, biostatistics, and perhaps other areas such as business management, management engineering, organizational psychology, and the analytics and “data science” sections of information sciences departments.  When I was at Blue Cross Blue Shield of Michigan, we set up meetings with the leaders of the University of Michigan School of Public Health to raise awareness that we had interesting job openings in our Epidemiology and Biostatistics Department and asking for their help to hook us up with the best students.  We sponsored a reception offering a big charcuterie tray and refreshments to have a chance to show a few slides describing our opportunities and to meet with students.  Most importantly, we sponsored some summer internships to build deeper relationships with students and to be in a position to more thoroughly assess their capabilities.

Through this process, we discovered that the most talented analysts, particularly those that had had both talent and knowledge of advanced analytic methods, wanted to work in a place that offered an intellectually stimulating community and an opportunity to make a difference.  When we hired analysts with advanced degrees, we made candidates give a presentation of their work — typically their major project or thesis.  We invited a bunch of our existing analysts to attend and discuss the work.  This process took a lot of time, particularly because we were very picky, often interviewing ten candidates before we hired one.  But it allowed us to discern real talent as well as the professional characteristics that made for good analytic collaborators and effective communicators of analytic outputs.  It also allowed the candidate to see how serious we were and to allow them to meet people and confirm they would enjoy the stimulating environment.  It provided learning opportunities for the existing analysts.  And each time we hired an awesomely talented analyst, it made our team better and it made it easier and easier to attract more talent.  Over time, we developed a core analytic team that included 13 PhD-level analysts and 10 masters-level analysts, dramatically improving the organization’s analytic effectiveness and enabling us to pursue value-based contracting innovations and care management innovations that would otherwise have been impossible, or at least very risky.

Using analytic projects to improve analytic education

Of course, most health care organizations do not have the size and budget to build such large and sophisticated analytic core teams.  As a result, it is necessary to go beyond just improving the recruitment pipeline and the internal talent development process. In the longer term, it is valuable to go farther upstream to work directly with universities to improve the analytic educational experiences within undergraduate and graduate programs.   I believe that the most promising such initiative would be to establish and develop programs that give more students in analytic disciplines more opportunities to participate in interesting, applied healthcare analytic projects. 

In this regard, we can learn from the talent development approaches that work in the physical and biological sciences.  Science fairs motivate very young students to participate in science projects.  Most science projects at elementary and secondary schools are nowhere near the cutting edge.   For example “does soapy water inhibit plant growth” or “demonstrating a volcanic eruption with vinegar, baking soda and red paint” do not generate new discoveries.  But such projects expose young students to the tools, jargon and conventions of science — developing a hypothesis, collecting and analyzing data, preparing a poster, and making a presentation.  Then, in science graduate programs, science students have an opportunity to work on actual cutting-edge projects.  Where do these project ideas come from?  Occasionally, student can think up their own original ideas.  But, more typically, good candidate projects are identified by the leaders of the labs in which the student-scientists work.  The leader of a laboratory typically has some core projects that they personally lead, and the students and post-docs do the dirty work, getting experience with the equipment and methods of the field.  The laboratory leader identifies various side projects, typically related to the main projects, that are suitable for students and post-docs.  Through some matchmaking, the laboratory leader helps the students and post-docs to select projects of interest and guides them through the process.

We can adapt some of those concepts to apply them to healthcare analytic talent development.  In social science and data science programs, students may be required to do papers at the undergraduate level, a major project at the master’s level, and a thesis in a doctoral program.  In my experience, faculty members in social sciences tend toward theoretical projects, since they lack exposure to real-world questions and challenges.  Such theoretical projects are helpful to build understanding of concepts and methods, but they tend to be less helpful in developing in students a passion for applied analytics and, more specifically into a passion for analytics focused on supporting healthcare process improvement.  Therefore, I think it would be beneficial for health care organizations to partner with social science departments to proactively define applied analytic projects that are “real” in the sense of relating to actual decision-making and design issues within healthcare organizations, but still suitable to serve as educational projects for students.

The ideal types of analytic student projects

Health care organizations have care processes and administrative processes, and the care processes can be divided into clinical decision-making processes (those that produce changes to care plans) and care delivery processes (those that execute care plans).   Analytic projects can also be classified as (a) retrospective data analysis and reporting, and (b) prospective modeling and simulation.    Putting these two dimensions together, we can define six types of analytic projects relevant to healthcare organizations, as shown below.

	Admin & Finance	Clinical Decision-Making	Care Delivery
Retrospective Data Analysis and Reporting	(1) Admin Reporting. Financial Statement Analysis, Market Structure	(2) Clinical Practice Reporting. Disease incidence & prevalence, Utilization, Practice Pattern, Small Area Variation	(3) Clinical Operations Reporting. Cost Accounting
Prospective Modeling and Simulation	(4) Admin Modeling. Contract Modeling	(5) Clinical Decision Analysis. Cost-Effectiveness Analysis/CBA, Population Simulation	(6) Clinical Operations Modeling. Operational Simulation, Queuing Theory

The table above shows some types of analyses that could serve as projects for students learning healthcare analytics.  Among mature, fully developed healthcare analysts, the most effective are those that are familiar with all of these types, as I noted in a post back in 2011: We need to cultivate more analysts that can do both rigorous science and practical decision-making – Reward Health. But that still leaves the question about which type of project is the optimal starting point for student analysts.

When most people think about analytic projects, their minds tend to conjure the top row: retrospective data analysis and reporting.  In such projects, a student is provided access to a database. They do queries and regression analysis and apply other retrospective analytic methods to characterize something or draw conclusions about potential cause-effect relationships.  Such projects can indeed by valuable.  But they also have some disadvantages in terms of student education.  Educational institutions often do not have access to proprietary and confidential health care data, so retrospective analytic projects tend to be done on generic public databases, such as datasets produced by government agencies.  Also, papers describing things and exploring cause-effect relationships do not always feel compelling to students.  They feel theoretical.  It is not obvious to the students how such projects make a difference. 

Therefore, I have a non-traditional proposal: the ideal student analytic projects are actually prospective modeling and simulation projects. 

Such projects  can be done under the control of the student, without substantial proprietary database access dependencies.  They directly relate to decisions that need to be made by actual healthcare organizations.  Among the three types of modeling and simulation projects, I propose that the ideal type is modeling and simulation projects to support clinical decision-making.  Such projects involve decision analysis, cost-effectiveness analysis, cost-benefit analysis, and population simulation.  They compellingly relate to the front-line decisions at the heart of the mission of a healthcare organization. 

The importance of seed papers

Just as elementary schools provide students with lists of ideas for science projects and faculty directors of research laboratories maintain lists of ancillary research projects suitable for graduate students and post-docs, it is very helpful to students learning analytic modelling to have access to a list of good candidate projects.  But, where do we find such projects, specifically for clinical decision-making modeling and simulation?  I propose that the best way to find such good candidate is to search for what I call “seed” papers.  A seed paper is a recently published paper that does not involve modeling and simulation, but that serves as a great starting point for a student modeling and simulation project.  The best seed papers for clinical decision-analytic projects are randomized clinical trials or retrospective observational studies (outcomes studies).  The fact that they were recently published suggests that they address a currently important topic.  Such seed papers provide a background section that frames the topic and the associated questions and that summarizes the current state of the literature on that topic.  Such a seed paper includes a great starting list of references for the student.  And, most importantly, such a seed paper includes results that will provide some of the inputs to a model or simulation.   They typically do not provide data on all the relevant outcomes needed to support the modeling of alternative candidate protocols for a particular clinical decision-making process.  But that is a good thing, because it gives the student an opportunity to get some experience in finding or collecting data or eliciting expert opinions to fill in the missing input values. 

The identification and matching of such seed papers can happen in the following ways:

1. The student can be provided with a list of journals and other search criteria to be used to cast a net for such papers, and directed to skim through many papers to find candidates of interest to them.
2. The faculty of the educational institution can keep a list of good seed papers that were noticed by faculty members or found through a purposeful search by faculty members or by other students.
3. The partner healthcare organization can keep a list of good seed papers based on topics that are relevant to the organization, and/or through a purposeful search.

The importance of mentoring during the conceptualization phase

Regardless of the source of the candidate topic and the associated seed paper, the next step is for the student to have an opportunity to have conversations with people with sufficient experience to help the student develop the analytic questions and frame the project.  Good guidance during the concept phase of any project is of obvious importance to avoid wasting time and to prevent the student from later becoming confused, frustrated and turned off by analytics. 

The investment of time with students during this phase is also valuable to assess the talent of the student.  Not everyone has adequate analytic talent and some students can be gently guided to apply their other talents. 

Spending time with students during this early phase also helps to shape the students’ analytic habits and provides them with the cognitive “frames” that can assist in clear thinking across diverse analytic topics and methods.  In my experience, the most important of such habits is to start with the end in mind and work back from there. Students must develop the discipline to always start by carefully defining the question or the decision being supported and the associated decision alternatives under consideration. Then, a student with good analytic habits works back from the decision to the analytic outputs needed, identifies the outcomes that are thought to materially differ across the alternatives, thinks through the basic logic needed to estimate those outcomes, and then works back to the model inputs needed to support such logic and the information sources available to provide those inputs.  The frames can take the form of patterns, templates or exemplar models that illustrate how to solve common modeling challenges.

Target state vision

When thinking through improvements to our process for developing analytic talent for the healthcare field, it is helpful to envision the eventual target state.  I envision a future with the following target attributes:

1. Students in social science departments and medical educational departments of local institutions of higher learning view the program to develop healthcare analytic talent as potentially valuable to their own career trajectory and as a way to make a difference.
2. The clinical and administrative leaders of local healthcare organizations view the program as both a pipeline for good analytic talent and as a source for useful and actionable analytic outputs.
3. The leaders of the educational institutions view the program as a valuable part of their educational mission and an important contributor to the reputation of the institution and its standing in the larger community.
4. There are defined faculty members at the educational institution and defined clinical and administrative leaders at the healthcare organizations who actively participate and collaborate in the program, attending meetings, working with students and providing other types of support to the program. They feel a sense of reward for what they have contributed and pride for what has been accomplished.
5. The students participating in the program go through a process that balances between being tailored vs. standardized.  It needs to be sufficiently tailored to serve students with different interests and a different mix of specific talents, while being sufficiently standardized so as to constitute a controlled process that can be systematically improved over time.  The program has defined activities and is supported by defined resources.  It generates defined analytic outputs, along with the structured data needed to assess the performance of the program.  Just as the program is intended to produce people capable of supporting the systematic improvement of healthcare organizations’ processes, the program itself should be a process subject to systematic improvement.
6. In addition to providing educational experiences and defined project-level analytic outputs, the program should accumulate intellectual assets that increase the effectiveness of the program over time.  For example, for model-building projects, rather than each student building each model from scratch, the program can build up templates, patterns, and reusable components in a modeling and simulation platform that can enable future student projects to build upon them. For example, a student model might include a component that does geographical analysis using GIS data for the local region. That component can be retained, packaged and documented for use by other students who can use it in their own projects, possibly making their own improvements in the component.  In this way, the program can operate as a type of open-source project, teaching students how to collaborate to build big, complex things.
7. As alumni of the program enter the healthcare organization workforce, they continue to benefit from their affiliation with the program through opportunities for continual education, mentoring and access to the intellectual assets of the program.  And they continue to contribute to the program, helping to identify good projects and seed studies, providing access to datasets for student projects, and mentoring both students and other alumni of the program.
8. Over time, the program, initially focused only on healthcare analytics, inspires others to develop similar programs for other types of talent development and contributes to a larger reimagining of the relationship between educational institutions and other organizations within communities.