10 Questions For @PrimeDimensions: Real-Time, Event-Driven, Complex Event Processing In Government and Healthcare

A few years ago I flew all the way to Cape Town, South Africa to present a paper at the triennial World Congress on Medical and Health Informatics, commonly know as MedInfo. My topic was a “big picture” framework of how electronic health records can be supplemented with process-aware Business Process Management ideas and tech. A key link in that framework was Complex Event Processing. I am not an expert on Event-Driven Architectures. So I am so delighted to convince Michael Joseph of Prime Dimensions, who is expert, to answer my questions! He’s experienced in both government and healthcare real-time information processing systems. By the way, I’m tweeting out chunks of this interview during the NextGov Prime2014 conference. Both Michael (@PrimeDimensions) and I (@wareFLO)will be responding to replies and retweets — in real-time!


Every hour or two I’ll update this blog post and tweet links to each added question for which there is a new answer. I’ll tweet on the #Prime2014 hashtag, so be sure to follow it during the Prime2014 conference, Monday, September 2014. Say hi to Michael for me if you bump into him at #Prime2014.

Thanks Michael! First of all. Enjoy the conference! Now, let’s dive into the real-time, event-driven weeds! Could you provide us a brief glossary of terms, so we won’t feel lost?

Sure, Chuck. The following is from one of my slide decks.

  • Real-time Data: Data streams processed and analyzed within seconds of data collection.
  • Business Process Management (BPM): Includes user-friendly business modeling and optimization tools, system integration tools, business activity monitoring tools, and rich workflow capabilities for end users.
  • Complex Event Processing (CEP): Advanced analytics on one or more streams of real-time data to quickly identify and respond to meaningful events.
  • Data in motion: Steaming data flowing within a system or from one system to another (ie: HL7, medical devices, real-time location services).
  • Data at rest: Data stored in a database or file.
  • Business Activity Monitoring: Real-time activity data converted to information and pushed to end-users through a visualization tool or dashboard.
  • Operational Business Intelligence (BI): Reporting of real-time data triggered by an end-user request.
  • Event-driven Architecture (EDA): A framework that orchestrates behavior around the production, detection, consumption of events as well as the responses they evoke.

Thanks, Michael, that was very helpful! If we count that as the first question in this interview, it get’s us to a nice round ten (plus one!). Here are the rest of the questions I have, base on past conversations with you. I know you’ve got a foot in both the government and healthcare spheres, so I’d appreciate it if you could address both areas, when practical.

  1. Could you provide us a brief glossary of terms before we dive into the Real-Time weeds?
  2. What are the advantages in adopting a real-time data capability?
  3. What are the advantages in adopting a real-time data capability in healthcare?
  4. Is there an industry-wide, accepted definition of “real-time?”
  5. What are typical sources of real-time data?
  6. What type of analytics can be generated with real-time data?
  7. How does this real-time capability influence or drive the future-state solution architecture?
  8. What are advantages of an Event-Driven Architecture?
  9. What technologies are required to deploy a real-time capability?
  10. What is the relationship between business process management (BPM) and Complex-Event Processing (CEP)?
  11. You’ve got a great marketecture of how an Event-Driven Architecture could fit into a hospital IT architecture. May I share it?

2. What are the advantages in adopting a real-time data capability?

Government agencies continuously generate and collect valuable real-time data associated with specific transactions, but they generally have limited ability to effectively analyze, process and present this data for actionable intelligence and real-time decision support.

Real-time data allows information to be disseminated in a timely manner, when and where it is needed. Most organizations have been focusing efforts on leveraging technology to become a data-driven enterprise; the next evolution is to also consider how to become an event-enabled enterprise. Real-time capability and related process automation assist in accessing data to build event-driven applications enriched with other relevant information and bringing them together in a scalable, vendor agnostic platform.

The expectation and pressure to deliver measurable results and meaningful change has never been more pronounced, as government executives face enormous challenges and risks as they navigate the complexity of our digital, data-driven world. These circumstances necessitate next-generation technologies designed to extract value from very large volumes of disparate, multi-structured data by enabling high-velocity capture, discovery, and analysis.

3. What are the advantages in adopting a real-time data capability in healthcare?

For the healthcare industry, a real-time capability is required to exceed future standards of care, provider experience and patient engagement expectations and to accommodate the massive transformation currently occurring in the healthcare industry — a transformation focused on opening up health data to facilitate exchange with providers, payers, and patients. For this reason, healthcare providers should be seeking to deploy an Enterprise-wide real-time processing capability that provides improved clinical insights, operational effectiveness and situational awareness associated with key indicators and events.

4. Is there an industry-wide, accepted definition of “real-time?”

Not exactly. The real-time platform should be designed based on requirements for providing minimally acceptable timeliness of information based on feasibility and clinical necessity. In collecting, processing and analyzing real-time data, there is inherent latency depending on data rates, volume, aggregation method, processing power, embedded analytics and throughput. In general, real-time data is defined as data streams that are processed and analyzed from milliseconds to approximately 30 seconds of collection. This is done either through a machine-to-machine or machine-to-human interface. Sensors and medical devices generate real-time data that are captured by other systems for continuous monitoring. Depending on the scenario, anomalies may be resolved by automated responses or alerts for human intervention.

5. What are typical sources of real-time data?

Sources of real-time data include data-in-motion, such as instant messages, flow sheets, device and sensor data, business process and workflow data, and real-time location services (RTLS). The goal of the real-time capability is not only to capture and integrate these sources, but also to transform and collect latent, transactional data as it becomes available in various source systems, such as financial, human resources, operations, and supply chain management. The challenge is efficiently integrating these disparate data sources across a fragmented information infrastructure, with multiple data silos, data marts and operational data stores. The real-time platform will provide specialized data services to extract data-at-rest that represents the most current records and maintains “a single version of the truth.”

6. What type of analytics can be generated with real-time data?

The real-time processing provides better visibility into all dimensions of healthcare delivery with real-time information. Disparate clinical and operational applications create the need to aggregate patient data in a new IT environment for real-time collection, processing and analysis. Analytic engines process vast amounts of data to identify and correlate the most important factors influencing patient care to develop an optimal treatment plan. In addition to real-time response and alerting, the platform enables the emergence of a new class of analytic applications, dashboards and visualizations for predictive modeling, clustering analysis, decision trees, root-cause analysis and optimization.

By supporting both on-demand and continuous analytics, the real-time platform extends and improves operational business intelligence and business activity monitoring through integration with Enterprise reporting and dashboard tools. On-demand real-time analytics are generated and delivered based on a user query; the data are pulled in real-time. Continuous real-time analytics notifies users with continuous updates in real-time; the data are pushed through on a regular basis. Algorithms include statistical analysis, predictive modeling, root-cause analysis, optimization, data mining, decision trees, clustering and natural language processing.

7. How does this real-time capability influence or drive the future-state solution architecture?

Event-driven architecture (EDA) integrates relational, non-relational and stream data structures to create a unified analytics environment. EDA describes an architecture where layers are partitioned and decoupled from each other to allow better or easier composition, asynchronous performance characteristics, loose coupling of APIs and dependencies and easier profiling and monitoring. Its goal is a highly reliable design that allows different parts of the system to fail independently and still allow eventual consistency.

EDA enables discovery, or exploratory, analytics, which rely on low-latency continuous batch processing techniques and high frequency querying on large, dynamic datasets. This type of architecture requires a different class of tools and system interfaces to promote a looser coupling of applications to streamline data access, integration, exploration, and analysis. It is also designed to deploy real-time Web applications using NoSQL databases, RESTful interfaces and advanced platforms that maximize throughput and efficiency by providing evented, asynchronous I/O and guaranteed, non-blocking libraries, thereby sharing code between the browser and server, effectively eliminating the Web server layer.

8. What are advantages of an Event-Driven Architecture?

  • Promotes operational effectiveness, process automation and analytic excellence
  • Enables advanced analytics and clinical informatics through an interoperable and scalable infrastructure
  • Streamlines technology insertion based on agile development methodology for rapid deployments
  • Controls IT operational costs by eliminating redundancies and aligning capabilities
  • Supports strategic planning, organizational alignment, and continuous process improvement
  • Provides a practical framework for defining and realizing the evolving future state
  • Integrates multi-structured and stream data using advanced technologies that provide high velocity data capture, discovery and analysis
  • Establishes a virtualized data environment and extensible service-oriented architecture that supports both Restful and SOAP APIs, allowing multiple data structures and formats (JSON, XML, etc.)
  • Provides an application development platform with domain-specific enclaves for evolving from “systems of record” to “systems of engagement”

9. What technologies are required to deploy a real-time capability?

A real-time platform ingests and processes data streams from clinical and operational systems, performs complex event processing (CEP), pattern matching and anomaly detection, applies on-demand and continuous analytics and triggers notifications and alerts based on an embedded rules engine. The platform also aggregates at-rest retrospective data from other source systems with real-time data streams for enhancing the context of information presented through operational business intelligence. With an in-memory cache, the platform has the ability to retain and persist data as long as it remains relevant to the real-time event. Detecting and reacting to events in real-time allows wide variety of business processes to be automated and optimized, improving a patient’s entire care team to improve communication and collaboration.

CEP provides an organization with the ability to detect, manage and predict events, situations, conditions, opportunities and threats in complex, heterogeneous networks. The in-memory cache provides the capability to run multiple, complex filters that compare events to other events in the same stream, or in another stream, or compare events to a computed metric. Moreover, in conjunction with the CEP rules engines, multiple algorithms can be deployed simultaneously and include the following rule types: (1) message syntax and semantic rules, (2) routing and decision rules, and (3) aggregation and transformation rules. Sophisticated rules engines can invoke in-memory analytics. To optimize performance, the platform can apply data compression and message parsing directly on the incoming streams, depending on the data rate, content and structure. Detection-oriented Complex Event Processing focuses on detecting combinations or patterns of events. Aggregation-oriented Complex Event Processing focuses on executing embedded algorithms as a response to data streams entering the system.

10. What is the relationship between Business Process Management (BPM) and Complex Event Processing (CEP)?

An ideal real-time platform integrates Business Process Management and Complex Event Processing. Together, these components create an agile, high performance, scalable platform that can deliver fast insights through real-time queries, notifications, alerts and advanced analytics. With BPM, the platform is able to detect discrete events and trigger a workflow that completes a specific process through a series of transactions. CEP extends this capability by correlating multiple events through a common interface that invokes an embedded rules engine. Event filtering evaluates a specified logical condition based on event attributes, and, if the condition is true, publishes the event to the destination stream as a notification or alert. Moreover, integrating BPM tools, and other line-of-business (LOB) applications, improves operational business intelligence and business activity monitoring (BAM), the use of technology to proactively define and analyze the most critical opportunities and risks in an enterprise. By deploying a BPM capability based on CEP technology, providers will be able to process high volumes of underlying technical events to derive higher level business and clinical decision support, extending the functionality of BPM tools and providing insights to a wider audience of users.

11. You’ve got a great marketecture of how an Event-Driven Architecture could fit into a hospital IT architecture. May I share it?

[Michael, you’ve got an incredibly detailed “marketecture” of how an Event-Driven Architecture integrates into a hospital venue. May I share it? Also, your detailed answers to my in-the-weeds question really started a bunch of wheels spinning in my head. Do you mind if I followup at a future date with ten more questions, even more healthcare focused?]

Please do! I look forward to it!



P.S. [Chuck writing] You can see that Michael is fount of detailed knowledge about real-time analytics. Michael comes from a business intelligence and analytics background. I, of course, come from a healthcare workflow tech background. It’s fascinating to see how event-driven architectures are so at the intersection between clinical business intelligence and clinical business process management. And I’d like to thank Michael for having a conversation about that intersection. By the way, as I mentioned at the outset, I’m tweeting out chunks of this interview during the NextGov #Prime2014 conference. And both Michael and I will be responding to replies and retweets — in real-time!

Day Two of Healthcare Business Intelligence Forum: Show Me More Workflow

I’m back for day two of the Healthcare Business Intelligence Forum in Washington DC! Here’s my report from Day One. I’m looking for even more workflow tech, and I find it.

Below is the venue, from the view of the stage, about a half hour before festivities begin.

Fantastic presentation by Glenn Steele MD of Geisinger Health.

Refer to the following tweeted link for an overview.

What I want to focus on is the idea that healthcare value and cost are not necessarily positively correlated. In fact, the opposite is true. The great thing about this is that it brings physicians on board relative to reducing cost as a path to increase value to patients.

So, how is it possible to reduce cost to increase value? Value Reengineering and then scaling and generalizing.

So, where’s the workflow tech? It’s under the hood. In fact, I happen to have a slide from a presentation given by the CIO (I believe) from Geisinger Health two years ago a National Library of Medicine. (It was sitting in a draft blog post I didn’t finish).

For me, the two best presentations were this one from Glenn Steele, MD and from Ray Hess yesterday. In both cases, the results of applying business intelligence to healthcare workflows were remarkable. And in both cases I happen to know they’re workflow automation, AKA BPM (Business Process Management) under the hood!

I enjoyed this presentation about the need to decrease healthcare data Semantic Incongruity — especially since it was an excuse to tweet a link to my From Syntactic & Semantic To Pragmatic Interoperability In Healthcare. My general point is basically that workflow tech can compensate for problems at the syntactic (transport) and semantic (meaning) layers. Workflow, or pragmatic, interoperability is the degree to which actual effect of a message matches its intended effect.

Issues regarding cost accounting systems have came up frequently. And here it is again. Activity-based costing is a form of micro-costing that can more accurately assign costs to specific healthcare workflows, such as an appendectomy. For a variety of technical reasons I won’t go in here (except to note I’m the only premed-accy major I’ve ever heard of) there’s a really great fit between ABC and workflow tech. In fact, I’d go as far as to say ABC won’t be practical without workflow tech. But that’s another blog post.

Now, finally, back to a bit of humor. Gus Venditto, emceeing the Healthcare BI Forum, mentioned me as a “Twitter Celebrity” so I just had to capture the moment using Google Glass. I grabbed 9 seconds and uploaded it on the spot.

This year’s Healthcare Business Intelligence Forum was a great conference!

I’ll be back!

Day One of Healthcare Business Intelligence Forum: Show Me The Workflow

I am delighted to be attending the Healthcare Business Intelligence Forum here in DC. As usual, Industrial Engineer/MD that I am, I looked for workflow tech.

That’s an old tweet, from the recent HIMSS14 conference. But it’s a fitting start of this blog post.

I’ve written a lot about the workflow technologies necessary to support care coordination, so my ears perked up at the above reference to secret sauce.

If care coordination is the secret sauce of population health management and if (as I argue) workflow automation is the secret sauce of care coordination, then isn’t workflow automation the secret sauce of the secret sauce of health population management? I certainly believe it is.

Above reminds me of a complaint about EHRs. They have the data but they don’t help users do what they need to do NEXT. Managing population health means taking that data and turning it into actions at the point of care. And that’s what workflow tech does.

More examples of workflow automation style task management: populating work lists and alerting users to actions needing to be taken.

Now what I find so interesting about this slide is the way in which “care coordination” clearly occurs outside of the EHR. This too bad, but unfortunately it’s the situation today. I read recently that many millions of investment money has gone into care coordination companies during the last couple years. Some are leveraging the kind of workflow management tech I’m looking for evidence of here at the conference.

Again this tweet echoes the previous one about NOW, NEXT, LATER. Workflow tech reminds you to do something you told it to remind you about a month ago. Actually, if you set up the workflow definitions right, it can figure out what it needs to remind you about when and where without you explicitly telling it to do so! Saves a lot of work that way.

BREAKING: workflow! 🙂

I walk around with Google Glass at health IT conferences and conduct what I call One-Minute Interviews. There’s just one question and I upload and tweet on the conference hashtag literally on the spot.

In the following video Christina remarks that one of the presentations emphasized that traditional business intelligence (dashboards, reports) is not enough. Business intelligence has to reflect, and in turn drive, actual, executable workflows. This approach is called Business Process Management, or BPM (though in a medical context, search engines sometimes think it means Beats Per Minute). I maintain an entire blog on Healthcare BPM.

It’s great to see a rep from a BPM vendor, especially Appian, at a health IT conference!

This tweet also echoes an earlier one about not knowing the true cost of healthcare. Which is another way of saying we don’t know the true cost of healthcare workflows. Well, healthcare workflow is made up of tasks. The most expensive resources to accomplish these tasks are people. And we don’t know how long which people spend on which tasks (see, I told you I’m an Industrial Engineer). And (I know, that’s the second “and”), I don’t think we’ll be able estimate the costs of these workflows without workflow technology to automatically measure these durations. Healthcare cost accounting systems are hobbled by lack of input from time stamped data about resources used to accomplish specific workflow tasks. See: Show me the workflow tech!

Rogers Diffusion of Innovation Curve was used to explain adoption of EHR tech. But I include it here because it also describes diffusion of workflow tech into healthcare.

By the way, the rest of the tweeted slides in this blog post are quite geeky. You have been warned.

Interesting: Core Applications and Workflow/Automation…

If you look up what this means (in the small print on the next slide), it “orchestrates execution of the activities of care coordination.” Wow! That sounds pretty important. In fact, this is the secret sauce of the secret sauce angle I riffed on earlier.

The above “marketecture” is sufficiently important that I’m not satisfied with its resolution (in other words, I can’t read the small print!). So here is the original photo. There are five main components.

  1. Connectivity, Security and Interoperability (dark blue line)
  2. Data Integration and Management in purple (including a Natural Language Processing inset in red)
  3. Data Analytics and Content in yellow
  4. Core Applications and Workflow/Automation in green
  5. Physician/Patient Engagement in red

My point here is to highlight the importance of workflow automation to the entire scheme. If you look up #4, here is what the small print says: Core Applications and Workflow/Automation: Orchestrates the execution of activities that constitute the care continuum, gathering contextual information from the transactional systems as well as the data warehouse.

Anyway, the first day of the Healthcare Business Intelligence Forum was just great. I’m seeing exactly what I hoped to see: workflow automation closing the loop from clinical and financial data back to executable workflows at the point of care. Exactly where is should be.

I’m looking forward to more of the same tomorrow, the second and closing day of the Healthcare Business Intelligence Forum!

Tweeting #HCBizIntel at the Healthcare Business Intelligence Forum

I’m attending, tweeting (and Google Glassing) the upcoming Healthcare Business Intelligence Forum here in DC on April 16-17.


This blog post is sort of a repeatedly updated running commentary before, during, and after #HCBizIntel. A place to embed tweets, Glass videos, and links to my own posts about clinical business intelligence.

In particular, as always, I’ll be looking for the workflow and workflow technology connection to clinical business intelligence…. as you know I would, must, and will. 🙂

And more to come!

P.S. I created a list of #HCBizIntel speakers, speaker organizations, and exhibitors. Here are their tweets!

Cutting Through The (Healthcare) Big Data Hype at #BigDataTechCon

The best antidote to hype about a technology (in this case, Big Data) is knowledge about the technology (especially the Hadoop ecosystem plus complementary and competing technologies).

So, off I went to Big Data Tech Con in Boston last week. I loved my deep dive into the software nuts and bolts under the hood of Big Data. I learned about (in some cases hands on) and live tweeted about:

  • Hadoop distributed processing of large datasets across clusters of computers using a simple programming model
  • Ambari Deployment, configuration and monitoring Hadoop clusters
  • Flume Collection and import of log and event data (of which is there is a lot!)
  • HBase Column-oriented NoSQL database scaling to billions of rows
  • HCatalog Schema and data type sharing over Pig, Hive and MapReduce
  • HDFS Distributed redundant file system for Hadoop
  • Hive Data warehouse with SQL-like access
  • Mahout Library of machine learning and data mining algorithms
  • MapReduce Parallel computation on server clusters
  • Pig High-level programming language for Hadoop computations
  • Oozie Orchestration and workflow management (BPM for Hadoop)
  • Sqoop Imports data from relational databases (extract/translate/load)
  • Zookeeper Configuration management and coordination of Hadoop nodes

Above is adapted from O’Reilly’s Big Data Now. It’s free. The first chapter is an excellent overview. Below are additional topics covered at Big Data Tech Con:

  • NoSQL non-relational databases (“Not Only SQL” has some SQL)
  • Cassandra highly available NoSQL database with no single point of failure
  • MongoDB scalable, high-performance, document-oriented NoSQL database
  • CouchDB easily replicated document-oriented NoSQL database
  • Google BigQuery web service for interactive analysis of massive datasets
  • Google Predictive API cloud-based machine learning tools for analyzing data
  • R plus Hadoop statistical analysis of massive datasets stored in Hadoop
  • Storm real-time complex event processor (Hadoop is batch)
  • Impala interactive SQL database using Hadoop’s HDFS file system

What do these technologies do, that makes Big Data possible? The shortest commonsense answer is that they can count higher than your laptop, or even the server down the hall or in your IS department.

Count? You say. Count? That doesn’t sound impressive or exotic at all! True. But it turns out that ability to count items in large sets enables remarkably intelligent software systems. One way to think about big data is that, in contrast to sophisticated statistical analysis using a single SQL database, big data is advanced applied counting, in parallel, across many databases.

Instead of complex algorithms applied to (relatively) small samples of data, Big Data is (relatively) simple algorithms (such as frequency counts) applied to (relatively) lots of data, sometimes so much, it is “all” the data generated by some software process. The more data, the better machine learning algorithms work. The more data monitored in real-time, the more individualized smartphone, web site, and desktop workflows.

What can you do with these counts? You can estimate probabilities. With these probabilities you can create automated systems that can predict

  • the next word (useful for natural language processing),
  • entities (people, places, diseases, etc.),
  • relationships (before/during/after, has-a/is-a, over/on/in/beside/below, etc.) among entities, and even
  • human behavior (when will you end your smartphone contract, will you stick to your meds, even will you reach for that next piece of pie).

I’ll cover more of the technologies listed above, perhaps in one of my occasional 5,000 word blog post opuses. Oozie is especially cool. It’s a workflow management system for Hadoop!

In my humble opinion, the best way for a health IT professional to cut through the healthcare Big Data hype is to learn about nuts and bolts of Big Data software. If you’ve ever taken an introduction to programming course (or willing to take one on Coursera) and know a bit of SQL, Bob’s your uncle!

Then, when someone makes an off-the-wall claim or wild-eyed guestimate about Healthcare Big Data you can at least try to imagine how it might be accomplished (a mental architectural sniff test). A little scientific and technical knowledge is exactly what more people need with respect to taking advantage of the benefits Big Data offers to healthcare.

Right now, if you’re in healthcare, and you’re interested in this stuff, check out the next Big Data Tech Con, in San Francisco. No, it’s not about healthcare Big Data. It’s about the nuts and bolts of the tools and techniques of Big Data (period), and that’s exactly what you need. It’ll be up to you to apply the tools to healthcare.

That said, healthcare did come up a couple times…

As an attendee I didn’t have access to the attendee list (apparently only exhibitors do, fair enough). But I did ask, and was told, a bunch of healthcare folks were there. In conversation, I heard of one well known EHR vendor investigating Hadoop for storing data from its customers, as well as a health insurance exchange doing similar.

The next tweet (not tweeted from #BigDataTechCon) links to a delightfully detailed (i.e. “non-hype”) description of one medical center’s use of Hadoop.

(I added the following two tweets on 4/15/2013.)

The software under Big Data’s hood does indeed have the potential to save billions of healthcare dollars. But, it has to be done right.

The biggest obstacle to doing it right? Workflow is the key. And health IT is not (yet) doing workflow right. (You knew I was going to eventually mention workflow, right? I always do.)

More later. Much more. But not too much later.

P.S. Just in case you don’t believe me, that you think I’m making this all up, here’s my certificate of completion! 🙂


P.S.S. Here are my tweets from #BigDataTechCon, in reverse order (so you don’t have to read from the bottom up).

Clinical & Business Intelligence, Meet Process Mining (Submitted to #HIMSS13 Blog Carnival)


[3/1/13 Update: This blog was one of three chosen by HIMSS to highlight in the mobile and clinical & business intelligence portion of the #HIMSS13 Blog Carnival. I am honored!]

EHRs increasingly mediate patient care effectiveness, resource efficiency, and user happiness. EHR process mining is a new medical “imaging” technique, one that allows process diagnosticians to view workflow blockages, errant processes, and unused resources. Process mining promises to do for healthcare workflow what Röntgen’s invention of X-rays and radiography in 1895 did for medicine proper.


Today, EHR process mining can discover, monitor and improve evidence-based processes (not assumed processes) by extracting knowledge from event logs available in (or “generatable” from) today’s EHRs. Process mining can answer three types of questions about a hospital or clinic: What is happening inside processes (Discovery)? It can compare what is happening with what should be happening (Conformance: especially relevant to medical error and patient safety). It can suggest ways to improve healthcare process effectiveness, efficiency, and user and patient satisfaction (Enhancement).

Of particular note to anyone interested in applying process mining techniques in healthcare is the Process Mining Manifesto. It is not specifically about healthcare, though it does mention “paper-based medical records” as examples of poor event logs. The Manifesto is authoritative (co-authored by more than 75 process mining experts), timely (recent and relevant to problems facing EHR adoption), and accessible to health IT and process improvement professionals.


The credited “godfather” of process mining is Professor Wil van der Aalst, a Dutch computer scientist and mathematician. Professor van der Aalst notes that healthcare is notorious for dismayingly complex “spaghetti” processes. Nonetheless, process mining research can learn a lot from tackling the healthcare domain. On one hand there is great opportunity to learn from intuitively creative medical experts. On the other hand spaghetti processes often are the greatest process improvement opportunities.

Process Mining and Clinical & Business Intelligence

The process mining of EHR event log data is a form of clinical & business intelligence. What van der Aalst notes generally about business intelligence also applies to clinical & business intelligence:

“Business intelligence tools tend to be data-centric while providing only reporting and dashboard functionality.”

This describes many clinical & business intelligence tools.

“They can be used to monitor and analyze basic performance indicators (flow time, costs, utilization).”

These are the KPIs, or Key Performance Indicators, in clinical & business intelligence reports and dashboards.

“However, they do not allow users to look into the end-to-end process.”

If you cannot look “into the end-to-end process” you cannot, in an evidence-based way, determine what is wrong—and therefore what is to be done—for ineffective, inefficient workflows and their unhappy users.

“Moreover, despite the “I” in BI, most of the mainstream BI tools do not provide any intelligent analysis functionality.”

Again, most current clinical & business intelligence tools are reports or dashboards. Without access to detailed evidence-based representations of end-to-end processes, clinical & business intelligence reporting and dashboard systems can flag process problems, but cannot diagnose and solve them.

EHR Event Logs

An EHR event log is a record of named activities (“Check Medications”, “Patient Examination”) created as a byproduct of EHR use. Encounter, or case, ids, tie together collections of events. Events occur in an order relative to each other, usually represented by time stamps. Intervals between time stamps can be years, in long-running chronic conditions; hours or minutes, in patient encounters; or seconds or less between user clicks on a single EHR screen.

The first three columns in the following EHR event log extract—CaseID, Activity, and TimeStamp—are required for process mining to create a process map, or model, from event data. The column of “…”s to the right represents additional data not shown: UserRole (a user such as Dr. Smith or Dr. Jones, or Physician vs. Nurse), EncounterType (such as Sick vs. Well Checkup vs. Vaccination), and Facility (such as Facility 5, 7 or 9, see upcoming illustrations).

CaseID Activity TimeStamp More columns→
7859, "Get Patient", 9/19/2011 15:44:27,
7859, "View Chart", 9/19/2011 15:53:58,
7859, "Current Meds", 9/19/2011 15:59:52,
7859, "Allergies", 9/19/2011 15:59:59
7859, "Labs", 9/19/2011 16:00:27,
7859, "New Note", 9/19/2011 16:05:46,
7859, "Examination", 9/19/2011 16:17:01,
More rows↓

Table 1: Portion of An EHR Event Log

Optional additional columns, over and above case id, activity name, and time stamp, depend on what you want to compare, explain, understand, or predict about your processes. Do you want to understand processes of a poorly performing clinic or hospital relative to a better performing clinic or hospital? You need a facility column. Do you want to do the same for users? You need a user column. Or do you want to understand workflows for sick visits compared to well checkups? Add a column for that. This additional information allows you to filter an event log and ask different questions about logged processes.

The bottom row of “…”s represents the many other rows, with different CaseIDs for separate process instances, usually required to generate useful process models. Healthcare processes generate a lot of time-stamped data that can result in large event logs. Process mining will be required to understand and leverage this “Big Data.”

Evidence-Based Process Maps

Below is a relatively unannotated (for example, no frequency or performance statistics) set of process models, or process maps, generated by ProM, a free and open source process mining tool. Even a simple example, with only five or six possible EHR activities, begins to looks like the aforementioned pile of spaghetti.


The process model can be simplified using event log filtering techniques and by asking specific questions to narrow investigations. The next illustration shows process-mined process maps comparing the most common workflow from three similar medical practices.


Suppose you know some Key Performance Indicators (KPIs) for these facilities, such as patient throughput and cycle time, cost per encounter or encounter type, or perhaps even measures of user or patient satisfaction. Process mining can generate process models that you can compare to explain differences between KPIs. Traditional clinical & business intelligence report and dashboard software may tell you what the KPIs are and help benchmark them. However, to understand the likely causes of flagged KPIs, you need evidence-based process models such as process mining provides.

Summary and Conclusion

Process mining of event log data from electronic health records promises new methods to systematically improve EHR-mediated patient care processes and workflow usability. Process mining is part of a larger front of process-aware business process management (BPM) technology diffusing into the healthcare information technology industry.


Process mining can discover evidence-based process models, or maps, from time-stamped user and patient behavior; detect deviations from intended process models relevant to minimizing medical error and maximizing patient safety; and suggest ways to enhance healthcare process effectiveness, efficiency, and user and patient satisfaction.

There's a great fit between traditional clinical & business intelligence KPIs, dashboards and process mining. Process mining provides an “X-ray” of workflows that can explain clinical & business intelligence KPIs. KPI dashboards alert users to systematic problems, while process mining shows subsystem tasks and workflows driving them. Combined, clinical & business process intelligence addresses central issues of healthcare reform: identification of best practices, coordination of care among clinical staff, consistency across patient care processes, and efficient use of healthcare resources.

For more information about process mining, the best place to start is the Process Mining Manifesto. It even mentions medical records.

P.S. This blog post was submitted to the #HIMSS13 Blog Carnival.

Clinical Intelligence, Complex Event Processing and Process Mining in Process-Aware EMR / EHR BPM Systems

Short Link: http://ehr.bz/8c


Last fall I presented a paper, co-written with Mark Copenhaver, at MedInfo2010 in Cape Town, South Africa.

Webster C. & Copenhaver, M. Process-aware EHR BPM Systems: Two Prototypes and a Conceptual Framework. In: Proceedings of the 13th World Congress on Medical Informatics, Studies in Health Technology and Informatics, Volume 160, 2010, pp 106-110. (indexed in MedLine)

You may have noted my photo travelogue at the time.

Process-aware health information system research and related industry undertakings have evolved since we built clinical intelligence and process mining prototypes in 2009 and thought how to bridge between healthcare IT and business process management. One of the goals of the blog is to draw readers, equally, from the realms of EMR/EHR/Health IT and BPM/Workflow/groupware. If you are from one of these industries but not the other, you’ll know some of the terminology, but not all. Hence a lengthy editorial preface plus a glossary of EMR / EHR workflow terminology as an addendum.

You can keep reading, skip to the abstract/slides/notes, or peruse the following outline and cut the (slide) deck wherever you like.

Short “Editorials” on EMR, EHR, BPM, BI, CI, CEP, Productivity and Usability

  • EMR, Electronic Medical Record
    • Perhaps the simplest definition of EMR is a “computerized ‘systematic documentation of a single patient’s long-term individual medical history and care'” where all the words between the single quotes are simply the definition of a “medical record” (wiki)
  • EHR, Electronic Health Record
    • Some use “EMR” and “EHR” synonymously. I often do. Others regard an EMR as being within a healthcare organization, such as medical office or hospital, while an EHR is a sum of capabilities to share and coordinate data and care across organizations. I’m OK with that too, but note use of the EHR acronym did not take off until the Federal government began using it instead of EMR. I sometimes hedge my semantic bets by using the phrase “EMR / EHR”. If a reader thinks they mean the same thing, then I appear to be noting synonymy. If they believe EMR and EHR mean different things, then I appear to refer to the totality of EMR plus EHR. Either works for me.
  • Workflow Management Systems, Business Process Management, BPM
    • As noted in Wil van der Aalst’s 2004 book Workflow Management: Models, Methods, and Systems, by analogy a workflow management system is to a workflow system much as a database management system is to a database system (for more on this distinction). In each case, the former creates and manages the latter. Workflow management systems are narrower in scope than business process management systems, sometimes designed to do little more than flexibly automate collections of tasks. Business process management systems, or suites, add capabilities from business activity monitoring and business intelligence to process mining and simulation to flexible user-customizable user interfaces.
  • Business Intelligence

“There is no clear definition for BI. On one hand it is a very broad term that includes anything that aims at providing actionable information that can be used to support decision making. On the other hand, vendors and consultants tend to conveniently skew the definition towards a particular tool or methodology. Clearly, process mining can be seen as a new collection of BI techniques. However, it is important to note that most BI tools are not really “intelligent” and do not provide any process mining capabilities. The focus is on querying and reporting combining simple visualization techniques showing dashboards and scorecards….Under the BI umbrella many fancy terms have been introduced to refer to rather simple reporting and dashboard tools.” (p. 21)

“Many vendors offer Business Intelligence (BI) software products. Unfortunately, most of these products are data-centric and focus on rather simplistic forms of analysis….process-centric, truly “intelligent” BI is possible due to advances in process mining.” (p. 261)

“BI products do not show the end-to-end process and cannot zoom into selected parts of this process….Another problem of mainstream BI products is that the focus is on fancy-looking dashboards and rather simple report, rather than a deeper analysis of the data collected. This is surprising as the “I” in BI refers to ‘intelligence’.” (p. 263, emphasis in original)

Also see How Process Mining is Related to BI where they write:

“The added value of process mining over traditional BI reporting tools lies in the depth of the analysis.

Traditional BI reporting tools focus on the display of Key Performance Indicators (KPIs) for executives in the organization. For example, the cycle times of a customer-facing process may be key in meeting certain service levels that have been agreed.

If the cycle times are out of the acceptable bounds, dashboards can highlight this problem. However, they cannot do much to uncover the root causes for this problem. Process mining can help to provide much deeper insight into the actual processes by uncovering the process flows and bottlenecks based on existing IT logs in a bottom-up manner.

Essentially, BI assumes that the underlying processes are known. Process mining takes the stand that even well-defined processes usually don’t go as planned and need to be brought into light objectively.” (my emphasis)

Out-of-bound KPI cycle times explained by process mined bottlenecks…this is exactly the capability demonstrated by one of the two EMR / EHR BPM modules presented below.

  • Clinical Intelligence
    • Definitions of clinical intelligence also vary according to vendor and consultant, tool and methodology. At this point, it is perhaps best defined as business intelligence tools and methods applied to patient care and health, and left at that. Later I’ll describe a specific clinical intelligence tool presented at MedInfo2010 (slides and notes also below).
  • Clinical Groupware
    • Clinical groupware is a combination of the “intentional care team processes and procedures pertaining to the observation and treatment of patients plus the tools designed to support and facilitate the care team’s work.” Note the emphasis on “team”. It’s unusual to see that same word, especially a noun, used more than once in a definition, so it must be important! I sometimes refer to clinical groupware as teamware (as opposed to “singleware”). Clinical groupware includes workflow systems, workflow management systems, business process management systems, and adaptive case management systems when applied to clinical coordination and collaboration.
  • Process-Aware Information Systems
    • PAISs, or Process-Aware Information Systems, include business process management systems (which in turn include workflow management systems). While database systems and email programs, for example, may execute steps in a process, they do not contain, consult, or are “aware” of, any explicit process models. Most current traditional EMR / EHR systems are not process aware. While EMR / EHRs are gradually incorporating more-and-more sophisticated task management features, most of these capabilities are relatively frozen, their workflow not amenable to editorial control by EMR / EHR users.
  • Process Mining
    • Since I own Process Mining by, arguably, the world’s expert on the topic, I might as well just quote Wil van der Aalst again:

“The goal of process mining is to use event data to extract process-related information, e.g., to automatically discover a process model by observing events recorded in some enterprise system.” (Process Mining: Discovery, Conformance and Enhancement of Business Processes)

“The healthcare industry includes hospitals and other care organizations. Most events are being recorded (blood tests, MRI scans, appointments, etc.) and correlation is easy because each event refers to a particular patient. The closer processes get to the medical profession the less structured they become. For instance, most diagnosis and treatment processes tend to be rather Spaghetti-like…. Medical guidelines typically have little to do with the actual processes. On the one hand, this suggests these processes can be improved by structuring them. On the other hand, the variability of medical processes is caused by the different characteristics of patients, their problems, and unanticipated complications. Patients are saved by doctors deviating from standard procedures. However, some deviations also cost lives. Clearly hospitals need to get a better understanding of care processes to be able to improve them. Process mining can help as event data is readily available. (emphasis in original)

I’ll illustrate the use of process mining to generate a process model for comparing nine busy pediatric practices in slides and speaker notes below.

If you are interested in process mining applied to healthcare, a good place to start are these three recent introductory posts.

Four Challenges for Process Mining in Healthcare

Process Mining in Healthcare – Case Study No. 1

Process Mining in Healthcare – Case Study No. 2

  • Usability, Human Factors
    • Usability is “The extent to which a product can be used by specified users to achieve specified goals with effectiveness, efficiency, and satisfaction in a specified context of use.” EMR / EHR usability, applying human factors principles and methods to EMR / EHRs, is a hot topic because physicians are “resisting” adoption of EMR / EHRs and there must be a good reason. The current reason célèbre is that EMRs are “clunky” and that usability engineers can fix this clunkiness. I’m a great fan of the cognitive science behind usability, but I do have a bone to pick (hmm, more like axe to grind). The relationship between EMR / EHR usability and workflow is both profound and misunderstood. When users complain about EMR usability they are often really complaining about EMR workflow that gets in their way, instead of paves their way. The biggest reason that current traditional-style EMRs are difficult to use is not so much because they weren’t created by an army of usability engineers, but because EMR / EHRs don’t rely on executable process models under editorial control (at both design- and run-time) of the users relying in them. It’s the users who should make their EMRs more usable, not EMR programmers or even usability experts. If non-programmer users could more easily push-pull-poke malleable EMR workflow into more usable workflow (for them, their goals, and their context), more usable EMRs and EHRs would result. I am *not* against usability testing. Usability testing *will* result in more usable EMRs. However, the most important form of EMR / EHR usability is a kind of meta-usability: usability of the tools users need to improve EMR usability, themselves. The most important meta-usability is the ability to improve the usability of EMR workflow. All the usability testing in the world won’t convert current document-based, data-centric, non-process-aware EMRs into workflow-based, process-aware EMRs. It’s too much of a paradigm shift and there is too much investment (and therefore design inertia) built into current legacy EMR / EHR product infrastructure.
  • Productivity
    • There’s also been a lot of press about current EMRs / EHRs reducing physician productivity (this letter to the New York Times is typical). Much of the current impetus to improve EMR usability comes from this press. Yes, usability is part of the problem. However, a traditional approach to EMR usability focuses on solitary users in front of solitary computers accomplishing (relatively) solitary tasks. The alternative is to study teams of users coordinating accomplishment of coordinated tasks “in the wild” (a reference to one of the first books about distributed cognition). The problem with the first approach is that it’s difficult to generalize from simulated laboratory experiments with individual users back to the real world. On the other hand, studying teams using groupware in the real world is fraught with its own problems. However, studying how better workflow leads to better usability and higher productivity will require it. We need hard data (such as, for example, process models generated from individually time-stamped user clicks) from teams of users actually using EMRs in the real world. Process mining can provide this. In fact, process mining is already used to study usability.
  • Complex Event Processing, Event-Driven Architectures
    • An event is a change in state, such as a patient who gains weight and moves from obese to morbidly obese state categories. A complex patient event is a pattern of detected events amidst a patient event stream (such as moving from obese to morbidly obese combined with being diabetic). Complex event processing, implemented in conjunction with a BPM system, provides means to react to events in real-, or almost real-, time. In the case of a process-aware information systems such as EMR / EHR workflow management (or business process management) systems, patient events can drive automated clinical workflows (such pushing action items to worklists) via workflow engines executing process definitions (more below).

Abstract, Slides, and Speaker Notes:

“Process-aware EHR BPM Systems:
Two Prototypes and a Conceptual Framework”


Systematic methods to improve the effectiveness and efficiency of electronic health record-mediated processes will be key to EHRs playing an important role in the positive transformation of healthcare. Business process management (BPM) systematically optimizes process effectiveness, efficiency, and flexibility. Therefore BPM offers relevant ideas and technologies. We provide a conceptual model based on EHR productivity and negative feedback control that links EHR and BPM domains, describe two EHR BPM prototype modules, and close with the argument that typical EHRs must become more process-aware if they are to take full advantage of BPM ideas and technology. A prediction: Future extensible clinical groupware will coordinate delivery of EHR functionality to teams of users by combining modular components with executable process models whose usability (effectiveness, efficiency, and user satisfaction) will be systematically improved using business process management techniques.

Keywords: EMR, Electronic Medical Record, EHR, Electronic Health Record, WfMS, Workflow Management Systems, Business Process Management, BPM, Business Intelligence, BI, Clinical Intelligence, Clinical Groupware, PAIS, Process-Aware Information Systems, Process Mining, Usability, Human Factors, Productivity, Complex Event Processing, CEP, Event-Driven, Clinical Quality Measures, Protocols, Guidelines, Compliance, Outcomes, Population Health Management, KPI, Key Performance Indicators, Closed-loop Patient Care


Thank you for attending this session on Process-Aware EHR Business Process Management Systems: Two Prototypes and a Conceptual Framework.


My presentation outline is as follows…


  • introduce EMR / EHR productivity, which can be systematically improved with business process management technology.
  • discuss a prototype clinical intelligence BPM module, called PROCARE, intended to systematically improve the state of health of patients in an EHR database.
  • describe a prototype process mining BPM module, called PROCESS, intended to systematically improve EHR workflow efficiency.
  • Finally I’ll list seven general advantages of process-aware EHR BPMS systems over EHRs that lack workflow engines, process definitions and the functionality these enable.

The red numbered bullet points (2, 2.a-d, 3, 3.a, 3.b) correspond to upcoming numbered slides.

EHR Productivity = Information Value / Information Cost


The concept of EMR / EHR productivity can bridge between EHRs and BPM technology realms. EHR productivity is the value of information contained in an EHR divided by the cost of obtaining that information. Information value and information cost can be systematically improved, maximized and minimized respectively, using BPM techniques such as business intelligence, process mining, and complex event processing.

Closed-Loop Optimization can Systematically Improve EHR Productivity


I am sure you are familiar with negative feedback loops such as implemented by thermostats operating to minimize the difference between observed and desired temperature. The difference between observed and desired state/output steers system state/output toward desired state/output. Many complex systems, from missiles to reactors, use sophisticated implementations of negative feedback loops to optimize system behavior.

[See Closed-Loop Strategies for Patient Care Systems for further overview of closed-loop control and its history, use, and future potential in healthcare.]

Healthcare information technology increasingly seeks to implement closed-loop systems, using estimated measures of clinical outcome and resource consumption to improve performance.

In order to systematically improve EHR productivity the information value numerator should be systematically increased while the information cost should be systematically decreased. In our formulation information cost is inversely proportional to efficiency level.

Closed-Loop Population Management and Closed-Loop Process Improvement


This slide is a graphical representation of an outline to this presentation. The numbered red boxes (2, 2.a-d, 3, 3.a, 3.b) correspond to upcoming slides.

You can think of improving EHR productivity in terms of negative feedback loops within a negative feedback loop. Two inner feedback loops implement systems for systematically increasing EHR information value and systematically decreasing EHR information cost. The outer feedback loop systematically increases the ratio of EHR information value to its cost to create.

Clinical Intelligence Plus CEP Drives Process Definition Execution


PROCARE, PROvision-based Clinically Active Reporting Environment, was a prototype BPM clinical intelligence module created to interact with an EMR / EHR workflow management system that relies on a workflow engine to execute process definitions.

A clinical intelligence reporting system without ability to trigger automated workflow is a passive reporting system (in which reports must be handed to staff for disposition, “Please put a note in each patient’s chart so that the next time they have an appointment…”). An active reporting system feeds directly back to a workflow engine executing clinical process definitions to automatically perform useful tasks–hence the “Active” in PROCARE’s Provision-based Clinically Active Reporting Environment.

“Provision” is borrowed from legal terminology. It means forward-looking restriction or qualification in a contract or agreement. For example, a patient can be in a predefined class of patients provided they meet that class’s predefined criteria (age between 0 and 18, BMI > 30, etc.). Many clinical intelligence reporting systems use predefined, or user defined, criteria to include or exclude patients from numerators and denominators of clinical performance measures.

To implement an EMR / EHR-based population health management you need a measure of health state, or a surrogate such as clinical performance (those clinical quality measures, with their numerators and denominators and exclusion categories and so forth). This direct or indirect measure (or combined measures) is compared to a goal value. The difference, or at least direction, is used to configure states, events, policies, and process definitions that use patient events to drive automated workflows improving health state/clinical performance. A human user, reacting to patient state reports and clinical dashboards, provides an important part of this negative feedback loop.

Clinical Dashboard Displays Patient On-Protocol/Compliant, Measured, Controlled Percentages


This slide shows a clinical dashboard relative to its function within this negative feedback formulation (“Health Monitor” in red box).



PROCARE’s clinical dashboard displays KPI’s (Key Performance Indicators) for each measure of clinical performance four numbers corresponding to the four levels of a “patient class event hierarchy” (which I’ll display several slides from now):

  • number of patients in the class for which the measure applies,
  • percentage of patients in each class that are compliant with a predefined protocol,
  • percentage of patients for whom appropriate and timely measurements are available, and
  • percentage of patients for whom observed measures are controlled (within target normal limits).

These colorful graphs on the right represent the same information in comparison to goal thresholds:

  • Green means performance measure above threshold
  • Yellow means at or near clinical performance threshold
  • Red means below clinical performance threshold.

Patient List Manager Enables Ad-Hoc and Policy-Based Intervention Planning



Selecting a measure of clinical performance in the summary display brings up a patient list management screen for intervention planning. Creation or refinement of automated workflow policies link patient class events to automated workflows. For example, process definition steps could include sending work items to roles or users, work items that appear when the patient next is physically present, instructions that appear automatically whenever a patient chart is opened, or messages to external systems that trigger email or phone calls.

While interventions can be triggered manually (from the patient list manager for individual patients or groups of patients meeting prior or user specified criteria) or automatically (via clinical CEP policies linking patient events to automated workflows) one optimization goal is to gradually replace manual interventions with automatic policy-based interventions, decreasing resource consumption while increasing predictability. Over time, slow, inconsistent, manual workflow “compiles” into fast, consistent, automated workflow.

Linking a clinical business intelligence system and a workflow engine automatically executing process definitions is what makes PROCARE an example of clinical complex event processing. A non-programmer, a clinical user, can create and edit patient state definitions, policies linking patient events (changes in patient state), and executable process definitions, turning an EMR / EMR, with a workflow engine, into an active clinical assistant, tirelessly working to achieve goals its human users program into it.

Patient Class Event Hierarchy Intermediates Patient Event Stream and Automated Workflow


Now we’ll drill down into the patient class event hierarchy (red box labeled Patient Class Events) used to trigger automated EHR workflows.

This decision tree is the critical intermediate representation mediating between low level patient events (state changes) and higher level concepts clinical concepts such as “on-protocol,” “compliant”, “measured”, and “controlled.”


Here you can see where the numbers in the PROCARE clinical performance dashboard come from:

  • #P+EHR, number of patients in EHR;
  • #P+MC, number of patients meeting clinical criteria;
  • #P-MC, number of patients not meeting clinical criteria;
  • #P+P/C, number of patients on-protocol/compliant,
  • #P-P/C, number of patients not on-protocol/compliant;
  • #P+MM, number of patients for whom target metrics have been measured within specified time interval;
  • #P-MM, number of patients for whom target metrics have not been measured within specified time interval;
  • #P+C, number of patients for whom target measures have been measured within specified time interval and are under control (within normal limits);
  • #P-C, number of patients for whom target measures have been measured within specified time interval and are not under control (not within normal limits).

Regarding #P+/-P/C (Patients On Protocol/Compliant, or not), if you have direct compliance-relevant data feeds from devices in the home, for example, this patient class event hierarchy likely should separate into On Protocol (#P+/-P) and Compliant (#P+/-C) levels.

To summarize…

Execution of appropriate automatic policy-based workflows (in effect, intervention plans),

  • for patients who aren’t on protocol but should be,
  • aren’t being measured but should be,
  • or whose clinical values are not-controlled,

moves patients from

  • off-protocol to on-protocol,
  • non-compliance to compliance,
  • unmeasured to measured, and from
  • uncontrolled to controlled state categories,

improving individual and collective patient health state, causing a shift from red to yellow to green graphical indicators on the clinical dashboard.

KPIs and Process Mining Flag and Identify EMR / EHR Workflow Bottlenecks


Now let’s take a look at the denominator of the EHR productivity formula. This module was called PROCESS, for PROcess Comparison for Efficient System Specification.

PROCESS is an example of process mining. Process mining generates process models from workflow, or event, logs. An event data point can be as little as a number identifying a patient encounter, the name of a task (“Record Allergies”), and a time stamp.


We needed a measure of global efficiency to optimize. We chose average throughput time, also called cycle time by industrial engineers. [Note red box, “Efficiency Monitor”.]


We process mined the workflow logs of nine pediatric practices to compare productivity measures and workflows, and highlight possible bottlenecks that could be alleviated by changing executed process definitions–hence the “Comparison”, “Efficient”, and “Specification” in PROCESS’s Process Comparison for Efficient System Specification. We benchmarked practice throughput volume and times against each other. Three practices stood out [see circled practices in previous slide]. We noticed that one of the three busiest pediatric practices (Practice 9, in blue) had a dramatically longer throughput time. Practices 5 (red) and 7 (green) took only 23 minutes and 44 minutes, respectively, to open and close a patient chart. In contrast, practice 9 took over eight hours to complete its charts. Obviously the patient was long gone by then.


Now we’ll drill down into a process model generated from the combined workflow/event logs of these nine pediatric practices. [Note red box, “Compare Processes”.]

Individual EMR / EHR Workflow Steps are Time-stamped and Logged


This EHR workflow management system has screens devoted to each possible data review and entry and order entry step. Table letters A through Y index the names of EMR screens: Allergies, Anticipatory Guidance, Chart Review and so on. Let me draw your attention to two pairs of task screen steps, Get Patient (H) and then Current Meds (E) versus Examination (F) and then New Note (G). In the former case (H to E) a nurse gets the patient and then asks about current medications. In the latter case (F to G) a patient examination is followed by creating a new note about the patient.

Process Mined Workflow/Event Logs Generate Detailed Process Models


This is the transcribed result of process mining the workflow logs for nine pediatric practices for the busy month of October 2008. The letters correspond to the individual screen tasks. Reviewing the process models revealed that practice 9 differed from practices 5 and 7 primarily in that many charts appeared to pile up between the Examination and a New Note steps and then stay there (red arrow from F to J). Practices 5 and 7 also showed some degree of congestion earlier in their workflow (red arrow from F to G), but this apparently did not have a dramatic impact on throughput time. The practice skills instructor took one look at this and said “Who is practice nine? They are doing something wrong and I need to fix their workplans!” (Process definitions are called “workplans” in this EHR workflow management system.)

[Recall that second quote about process mining and business intelligence? PROCESS is an example of what it described, explaining an out-of-bounds cycle time (the KPI, or key performance indicator) via a potential bottleneck in the process model generated by process mining EMR / EHR event data.]

Active Clinical Intelligence and Systematically Improvable Clinical Processes Require Process-Aware Foundations


Business process management systems, or suites, rely on workflow engines and process definitions but add additional value, such as user-friendly user interfaces or visual analytics to better understand processes. More recently, in the US, the phrase “clinical groupware” has also become popular. Workflow systems are classical examples of groupware. So, in keeping with recent trends in both the health information technology and business process management industries, I sometimes refer to these systems as “process-aware clinical groupware.” What all these technologies have in common is that they are “process-aware.”

Without a process-aware foundation, that is, without an executable process model, neither PROCARE nor PROCESS (or EMR / EHR BPM modules similar to them in functionality) would be possible and have practical effect. Actionable clinical intelligence, that is, active as opposed to passive clinical reporting requires some means to automatically detect salient patient events and then automatically trigger automated workflows, transparently and usably at physician behest. These automated workflows interleave with other workflows, manual and automated, to generate a deluge of time-stamped data, the basis for generating sophisticated operational clinical process intelligence to explain and improve clinical processes. Process mining is not just about improving efficiency. Any KPI–clinical outcomes; practice productivity and profitability; patient and user satisfaction–can be compared across medical practices and difference in KPI values (good or bad) explained by processes generating or influencing them. These fact- and process-based explanations can direct further investigation and intervention.

Seven Advantages of Process-Aware EMR / EHR BPM Over Process-Unaware Alternatives


Process-aware EMR / EHR workflow management systems/business process management systems have numerous advantages over their process-unaware cousins.

Non-process aware EHRs do not distinguish between unitary tasks at the same fine degree of granularity as process-aware EHRs. Traditional EHRs often have high resolution screens with a multitude of simultaneous data review and entry and order entry options. Multiple user events, spanning multiple tasks, are often committed together to the underlying database, conflating together logically separate workflow steps. In contrast, a process definition-driven EHR can present just the data review and entry and order entry options on each screen that are relevant to a single step in a task workflow sequence. For example, a nurse checking allergies and then current medications are two different tasks that at highly granular resolution should be distinct and acquire different time stamps.

Non-process aware EHRs do not capture all the potential meaningful timestamps for those events that they do log. They may log when data and orders are committed to a database but they do not typically log when tasks are first available to be accomplished, when they begin, when they complete, and other relevant timed-stamped events such as cancellation, postponement, or forwarding. Much of this missing temporal information is invaluable for understanding why bottlenecks occur, why certain tasks are subject to rework, and what slack resources are available elsewhere in the system.

Non-process aware EHRs, even if their event logs result in useful process models and actionable insights, lack means to actively influence changes to workflow. There are no process definitions or workflow engines to execute them; so there are no process definitions to change and thereby influence and improve effectiveness and efficiency. With respect to EHR effectiveness, a clinical intelligence reporting system without ability to trigger automated workflow is a passive reporting system (in which reports must be handed to staff for disposition, “Please put a note in each patient’s chart so that the next time they have an appointment…”). A more active clinical intelligence reporting system feeds directly back to a workflow engine and process definitions to automatically perform useful tasks. With respect to EHR efficiency, even if a process model has an obvious flaw, there is no way to consistently and automatically deflect behavior at critical process junctures in order to improve throughput and throughput time.

In summary, compared to process-aware EHR workflow management, or business process, management systems, traditional EHRs (1) do not track tasks at a sufficiently high degree of resolution, (2) do not distinguish among the large number of possibly useful time-stamped events, and (3) have no means for process model insights to drive improvement at the point-of-care through automated workflow.

The next four advantages of process-aware EHR BPM systems (or process-aware clinical groupware, if you will) are generally acknowledged advantages of BPM systems over non-BPM systems.

  • EHR BPM systems can be used to model and understand workflow,
  • coordinate patient care tasks handoffs,
  • monitor task execution in real time, and
  • systematically improve clinical workflow and outcomes.

The next step in the evolution of ambulatory EMRs is squarely at the intersection between two great software industries: electronic health record systems and workflow management/business process management systems. The hybrid EMR workflow systems that result will be more usable and more systematically optimizable than traditional EMRs with respect to user satisfaction, clinical performance, patient satisfaction, and practice profitability.

Thank you!

Epilogue: EMRs / EHRs Need to Perceive and Respond to Clinical Threats and Opportunities in Real-Time

Referring back to van der Aalst’s quote about business intelligence, he clearly considers process mining to be an example of sophisticated operational business intelligence. Just as clearly, therefore, both the numerator (PROCARE) and the denominator (PROCESS) in the EMR / EHR productivity ratio are examples of clinical business intelligence/clinical intelligence. The difference between this formulation of EMR / EHR-mediated business intelligence and most other formulations is the important role of an executable process model yoked to clinical complex event functionality. Without both capabilities–to both perceive and react to clinical threats and opportunities in real-time–transparently and under flexible human control, EMRs / EHRs will not become capable of automatable closed-loop patient care or its systematic improvement.

Addendum: Glossary of EMR / EHR Workflow Terminology

Phrase Definition Medical Example
Work Item Task to perform Vitals signs awaiting performance during a patient encounter
Workflow/ Process Definition Description of a process detailed enough to drive EMR / EHR behavior. van der Aalst refers to this as a formal process model, that is, one that can executed. Get the Patient, Take Vitals and a Chief Complaint, Review Allergies, Review Medications, Review of Systems, Examination Screen, Evaluation and Management, Billing Approval
Worklist List of tasks to perform A nurse’s To-Do list
Case Particular application of a EMR / EHR workflow management system / business process management suite A particular patient’s encounter managed by EMR / EHR workflow management system /business process management system
Process Order (though not necessarily sequence) of tasks to be performed and resource requirements A Well Child pediatric visit
Resource Something that accomplishes tasks (often a user) A physician, nurse, technician
Role Set of related skills accomplished by a resource The role of nurse or physician
Routing Types of routing include sequential, parallel, conditional, or iterative task execution Routing a recording to a transcriptionist and the report back to the physician
Task Unit of work carried out by a resource Obtain vital signs
Trigger An event that changes a work item into an activity Starting to accomplish the task of responding to a phone message by selecting a To-Do list item
Workflow A process and its cases, resources, and triggers The tasks and people involved in accomplishing a patient encounter
Workflow/ Process Definition Editor User application or interface for creating workflow/ process definitions An ordered picklist or flowchart diagram representing Get the Patient, Take Vitals and a Chief Complaint, Review Allergies, Review Medications, Review of Systems, Examination Screen, Evaluation and Management, Billing Approval
Activity Performance of a task Obtain vital signs within a patient encounter