Karen Smilowitz gave a fascinating talk in the INFORMS Healthcare Conference about planning for the Chicago Marathon. Her talk addressed
long-term planning, such as course planning and locating water and aid stations
short-term planning for emergency medical services using demographic information provided when runners register months before the race
online decision-making during the race to manage health emergencies
Several people from the medical field chimed in with comments for how to possibly plan for cardiac arrests during the race, such as monitoring runners using their RFID chip (all runners are tracked during races to record their times) or slightly more invasive methods that could track the runners’ temperatures or pulses.
Karen talked in detail about route planning, where criteria included routes close to area hospitals and that visit many neighborhoods. No loops are allowed in the course to avoid massive race chaos if the course passes over itself. The course start and finish need to be somewhat near each other, and there has to be flexibility in the finish line location to allow for possible reroutes near the end of the race (many large races have been rerouted prior to the 2013 Boston marathon). There are constraints to avoid U-turns and frequent turns. Feasibility is easy, optimality is hard. Their next steps are to integrate the route planning with aid station location and coordination with surveillance cameras.
As a marathon runner, I appreciated the use of OR to make for a more enjoyable and safer run. It’s painful to run around a curve after almost 26.2 miles, and while I’ve never had to visit an aid station, it’s reassuring when they are optimally located.
I enjoyed today’s networking luncheon (minus the part where the vegetarian person at the table didn’t get her entree until we were eating our desserts, and coffee never got served to our table or the tables around us). The best part of it was that an acquaintance from graduate school whom I had lost touch with ended up sitting at my table, and it was great to catch up with her. Her name is Sommer Gentry of the Naval Academy and she has done a lot of work on transplant kidney exchanges, along with her surgeon, a transplant surgeon at Johns Hopkins. She started working on operations research in healthcare long before it was fashionable, and I enjoyed hearing her share her insights. Among many other things, she made this great point that when she introduces herself as a mathematician to a member of the medical community (she is on the faculty in the math department at the Naval Academy, which does a lot of operations research, and it seems a lot easier than explaining what operations researcher is), the person will always interpret “being a mathematician” as “being a statistician”. And statistics are of course hugely important in determining which treatment is better, but I don’t need to tell anyone reading this blog that there is a lot more to math than statistics.
Sommer has published extensively on operations research models for kidney and now liver transplants, and has been mentioned in top media outlets such as the Baltimore Sun and the New York Times. She has designed optimization methods used for nationwide kidney paired donation registries in both the United States and Canada. Imagine situations where patient A needs a kidney and person 1 is willing to give him a kidney but isn’t a match, and patient B needs a kidney and person 2 is willing to give her a kidney but isn’t a match, but person 1 is a match for B and person 2 is a match of A: finding the best matching pairs is where the optimization will come in. The static version of the problem is already challenging due to its size, and the dynamic one (where people sign up over time for the registry and others leave it) is extremely difficult. But this is exactly the sort of problems where operations research expertise gets to make a real difference.
Kidney exchanges were also investigated in one of the papers finalists in the student paper competition, which ultimately received second prize: Iterative Branch-and-Price for Hierarchical Multi-Criteria Kidney Exchange by Kristiaan Glorie of Erasmus University Rotterdam. Congratulations to him!
At my table at the luncheon were also people from UT Dallas, NIST, Intermountain Healthcare and more. This made for many great conversations on the theory and practice of operations research and healthcare.
Many of the healthcare talks I have attended rely on optimize to allocate scarce resources or to match patients to healthcare resources. It’s worth pointing out that data-driven healthcare analytics doesn’t always imply statistics or simulation. Optimization is a great tool in the analytics toolbox. The optimization models I’ve seen in talks so far are mostly deterministic discrete optimization models (it’s worth pointing out that there are other flavors of optimization). The optimization models:
Locate resources (ambulances, clinics, etc.)
Assign spatially located patients to medical facilities
Route vehicles to optimally deliver healthcare resources
“Cover” patients with medical procedures ot healthcare access points
Schedule patients, appointments, doctors, and nurses
Determine the best mix of procedures to implement
As an optimizer, it has been great to see so many talks that apply optimization to healthcare problems. This conference helps to make the case that there is more to analytics than statistics.
I talked with a few people after a session from the medical comunity. One of the topics surrounded the objectives of our models: those of us in OR/MS tend to optimize class systems/operations/manufacturing metrics such as efficiency, coverage, and service times.
We tend not to optimize patient outcomes, and they are hypothetical when we do (for example, we may use survival curves from the literature).
We could collect patient outcomes after implementing the model in practice (and that would make for an excellent Interfaces article), but that takes a long time and isn’t necessary for publication in OR/MS journals. For “buy in” in medical journals, having demonstrated improvements in patient outcomes is necessary for changing policy. Not having patient outcomes can therefore delay positive changes to practice.
I understand why medicine takes a different perspective than OR/MS. I may be biased, but I’l like to see OR/MS get buy in from the medical community before implementation. Moreover, patient outcomes may not always make sense (e.g., when considering fairness or avoiding rare events that may not always have adverse outcomes). What is your take on this issue?
Post from Louise Wehrle, INFORMS Certification Manager
Certifications are everywhere. Really. You can buy certified pre-owned vehicles, certified diamonds, certified electrical cords. You can even build a business warehouse in a certified business park. You can earn certifications for everything from A (Ableton) to Z (Zentangle).
If there are so many and they are so prevalent, why did INFORMS develop the Certified Analytics Professional? Why not use something that’s already been created?
Here’s why: INFORMS is setting the standard for the analytics professional. It doesn’t matter where you work, or what courses you’ve attended, or what college or university you graduated from or what association you belong to. You can become a CAPTM only by passing an examination that is the same for all, regardless.
The CAPTM exam is based on a survey of analytics professionals who decided that what professional analysts do can be described as an end to end process that is common to all analytics projects. It starts with the business problem: what is it? Is the problem amenable to solution through analytics means? If so, what data are there and what data are needed? What’s the best methodology that will provide a solution to the problem? What model is best suited to this methodology? Once that’s been settled, you need to create, deploy and manage the model so it continues to provide answers to the problem.
Notice that we haven’t said anything about the specific tools that are used. That’s because the CAPTM is vendor neutral and software neutral. It assumes that if you know the key steps in the analytics process you can select the best software to solve your analytics problem and work with a vendor of your choosing. Many vendors have their own certifications: those are usually specific to the vendor software packages and may not be transferrable to another vendor or another software package.
What else distinguishes the CAPTM program? It requires the certificant or the person earning the CAPTM to maintain their level of knowledge. Every three years, certified persons must present evidence that they have used continuing education, or have taught, or created new information or have provided services to the analytics community that have enhanced the level of knowledge they displayed by passing the CAPTM exam.
Truly, if someone has earned the designation Certified Analytics Professional, he or she is among the elite practitioners of analytics.
And, if you see them, tell them INFORMS sends congratulations and best wishes. Add you own congratulations. Earn the CAPTM and join the elite. Practice saying ‘thank you’ modestly.
Today’s conference activities at the Marriott Magnificent Mile reminded me why I find smaller-scale conferences so valuable: great talks in close proximity to each other, real opportunities for meaningful discussions with colleagues, and excellent special events. I started the day with my session on healthcare finance – partial capitation schemes, robust risk scoring, and more – before the plenary talk on “IBM Watson and the Application of Unstructured Data Analysis to Healthcare”. The poster session in the afternoon was also highly informative – if you were unable to attend, you can check out the poster session interviews by the great Mary L. of INFORMS. The session on cancer therapy optimization chaired by Timothy Chan of the University of Toronto provided insights into the challenges and the potential of using advanced operations research techniques to orient radiation beams for treatment, determine optimal doses for radiotherapy, and prepare treatment planning for intensity-modulated radiation therapy. The conference reception was also a great success, with all the right ratios of food-to-attendees, tables-to-attendees, space-to-move-around-to-attendees, and plenty more networking opportunities.
I’m particularly looking forward to the networking lunch tomorrow – I’ve signed up to be a discussion facilitator at my table, so I hope to foster great conversations on healthcare, operations research and analytics. I also picked up a few brochures about the CAP (Certified Analytics Professional) exam to bring back to Lehigh. Perhaps some students will find that interesting. I took a look at the sample questions and found them excellent – allowing people who know what they’re talking about to shine at the test while leaving people who don’t no room to hide.
This has been an excellent conference so far! I’m really glad I came. The fantastic location doesn’t hurt either.
I had the opportunity to ask some of the poster presenters about their work, here is what they had to say about their work, along with the abstracts.
Using Health IT Tools to Improve Immunization Management in Community Health Centers,
Shannon Pohl and Nathan Taylor
Managing immunization care is an ongoing challenge for community health centers (CHCs). For CHCs, their patient population consists primarily of low-income, underinsured, uninsured, and other marginalized communities, where their access to health care is limited and their knowledge and compliance to obtaining immunizations is at increased risk. Maintaining up-to-date immunizations amongst this patient population, according to CDC recommendations , can directly help to improve public health and decrease the potential for infectious disease spreading across the community. The Alliance of Chicago Community Health Services (the Alliance), a health center-controlled network of 32 CHCs across 11 states, has focused on this challenge, with overarching goals of increasing the quality of immunization care provided and improving immunization rates. They have utilized health information technology (IT) tools for documentation, reporting and monitoring, using the electronic health record (EHR) as its primary database for immunization data. The Alliance will outline the key clinical and community health issues that CHCs have identified in immunization care, data strategies implemented to support clinical workflows, and reporting approaches to proactively target patients at risk.
Real-Time Decision Support System for Pandemic Influenza,
Sunderesh Heragu, Lihui Bai, Jose Bejarano, Ruth Carrico, Gail DePuy, Gerald Evans, Kari Kelton, Farshad Majzoubi, and Mehdi Zakerifar
The events of recent years have shown that preparation and informed decision-making are key factors in efficiently mitigating the effects of disaster situations, regardless of their nature. An effective decision making process for these situations requires tools that render precise information in a timely fashion such that operational capabilities can be assessed and action plans implemented or revised as necessary. The PanFlu Real-Time Decision Support System (RTDSS) supports logistics and planning by Commonwealth of Kentucky public health and emergency response personnel in the event of an influenza epidemic. The PanFlu RTDSS is composed of a suite of models that support state and local personnel in managing multiple aspects of pandemic response, including: –Distribution of the Strategic National Stockpile of antivirals throughout the state; –Management of a statewide network of mental health professionals to support emergency response personnel; –Dispatching Eeergency medical services (EMS) vehicles throughout a local jurisdiction; –Designing and simulating a mass vaccination clinic serving a local jurisdiction; –Optimizing the workflow of the emergency department of a hospital; and –Simulating the spread of the epidemic and the impact of control measures. This interactive poster presentation will focus on the EMS dispatch, mass vaccination clinic, and epidemic simulation models of the PanFlu RTDSS.
Integrated Medical Analytic Data Model for Evaluation of Unnecessary Laboratory Test Utilization,
Brad Brimhall and Bryan Clements
Conservatively 19% of clinical laboratory testing is unnecessary and is a large source of waste within the healthcare system. An integrated data model including not only laboratory data, but also other clinical data, more detailed financial data, and operational data would greatly facilitate identification of clinical resource utilization patterns and costs, thereby permitting healthcare systems to prioritize and focus efforts to reduce unnecessary testing. Integration of diagnostic and laboratory coding (LOINC) would facilitate comparisons between hospitals/clinics.
With all the interest in healthcare operations research, I expect that many of us OR professors are incorporating healthcare OR in the classroom. But exactly how much of that is happening? I see in the conference program that the Health Systems Engineering Alliance (http://catalyzecare.org/groups/hsea) includes 41 academic programs, which is impressive. How many healthcare-OR-related concentrations or programs are there in business schools? I know of one at the University of British Columbia and at the University of Ottawa, but I am sure that there are others that I don’t know about. Feel free to post responses to the blog with information about such programs.
If you are asked to teach a course on healthcare OR, then where do you go for help? INFORMS Transactions on Education, which I edit, is one resource. Take a look at the following, all of which are related to healthcare OR:
Sanjay Mehrotra welcomes attendees to the 2013 INFORMS Healthcare conference in Chicago. We would like to thank him and his committee for organizing this event, including:
Hari Balasubramanian, Cerry M. Klein, Steven Shechter, Vedat Verter, Dionne M. Aleman, Turgay Ayer, Mehmet A. Begen, Stephen E. Chick, Amy M. Cohn, Brian Denton, Diwakar Gupta, Eva K. Lee, Elisa F. Long, Omid Nohadani, Ronald L. Rardin, Burhaneddin Sandikci, and Kai Yang.
This post about the Monday plenary continues along the same idea. Basit Chaudhry from IBM gave the plenary talk entitled “IBM Watson and the Application of Unstructured Data Analysis to Healthcare.”
Healthcare analytics and OR/MS has done a wonderfu job using OR/MS methodologies using structured, clean data that has been collected (test results, appointment times, demographic information). A healthcare challenge is how to use unstructured data (e.g., doctor notes) to answer important questions.
IBM Watson was a computer program that used to play Jeopardy. IBM Watson has provide a set of reasonable answers (hundreds of them) to an open-ended English language question within 5 seconds. It does so by investigating a series of hypotheses about the question asked and returning a confidence associated with each potential answer. Context is an important issue when it comes to processing language.
IBM then trained Watson on medical data (e.g., medical texts, medical dictionaries, journals, resources) and posed medical questions to Watson for “Doctor’s Dilemma.” They have to weigh the knowledge according to how recent and reliable the knowledge is. Watson then generates a series of answers to the question with varying degrees of confidence.
The Jeopardy and healthcare environments are quite different. The objective isn’t just to get the answers right (as in Jeopardy), but to not be too far off base with wrong answers (in Jeopardy). The IBM model addresses this by generating a list of potential answers to the problem instead of one final answer (as would be given on a quiz show). This is a strength of the model that will be more beneficial in helathcare than in Jeopardy.
The Watson project has the potential to transform healthcare by supporting complex decision-making processes with analytics. It reminded me of the movie Robot and Frank (which I saw recently and highly recommend). Robot and Frank envisions a future where medical robots are widely used to take care of people with chronic medical conditions. Some of the robot’s duties are simple (such as ensuring that the patient takes medication at the right times) and other tasks are individualized and unclear (such as alerting the primary care doctor that the patient may have depression). Watson or a medical robot can help choose between several (or hundreds of) treatment paths.
The two hour tour focused on buildings from about 1890 to 1930 built in the Chicago style, which are tripartied construction, a base, shaft, and capital, similar to the construction of a column. Seen here is the Marquette Building, whose exterior was coated in terra cotta, a fire proofing effort in response to the Great Fire. There were some gorgeous mosaics in the lobby made of mother-of-pearl. The building is 17 floors, and it had 10 elevators, to avoid having an express elevator. Obviously built before queuing theory was popular.
Walking by the Art Institute, we saw its guard lions wearing Chicago Blackhawks helmets, honor of their Stanley Cup finals appearance. Apparently, the lions have names, the one on the left with the tail lowered is “On the Prowl,” and the lion on the right with the raised tail is “In the Attitude of Defiance.”
Another element in the Chicago Style is the Chicago window, which consists of a large fixed pane in the middle, flanked by two moveable windows on either side.
The Bank of America building, formerly the Fields building (of Marshall Fields fame) is an incredible example of Art Deco. The image above is of the post box and elevator light panel.
The Rookery building is on the corner of LaSalle St, which is Chicago’s version of Wall St. After the Great Fire, Queen Victoria of England wanted to show her sympathy for the people of Chicago, and sent many books to rebuild their library. The books were housed on the top floor of this building. It turns out that Chicago did not have a public library before the fire, so Queen Victoria was instrumental in the formation of the Chicago Public Library. 20 years after the buildings construction the lobby was renovated by Frank Lloyd Wright. (Might have to pop back to check it out, it was closed today.)
There were lots of other interesting facts, so check it out for yourself. What a great town!
Welcome to Chicago! It has been a while since I have blogged for INFORMS, we have so many exciting things happening to keep me busy. We had a great Annual Meeting in Phoenix last October, and the largest Analytics meeting to date, in San Antonio in April. The Edelman Competition and Gala are some of my favorite projects to work on, they are such incredible examples of O.R. and analytics in action. Congrats again to the Dutch Delta Programme Commissioner!
I’ve also been working on members-only social networking system for INFORMS, which is really going to be great. Members will be able to share, collaborate, and network within their INFORMS Communities as well as INFORMS as a whole.
Our journals are moving to a new hosting platform shortly, which will give a better online experience for mobile devices, so you can read your journals on the go! We’ve also given OR/MS Today a fresh new look.
We have also started two new programs geared towards the analytics crowd, certification and continuing education courses.
And to top things off, INFORMS is moving locations next week. Our new home will be the Research Park at UMBC, which is a beautiful, modern space. If you’re in the neighborhood, stop by!
I read an article skeptical of the benefits of data-driven medicine while preparing for the INFORMS Healthcare conference [Link]. The article correctly discusses the
A recent study at Johns Hopkins University indicated that hospital interns — physicians at perhaps their most formative stage of training — spend only about 12 percent of their time interacting with patients. By contrast, they spend 40 percent of their time — more than 3 times as much — interacting with hospital information systems. The flesh-and-blood patient is getting buried under gigabytes of data.
The article is a bit frustrating. It’s worthwhile to discuss both the benefits and costs associated with any medical intervention or healthcare policy. The author (Richard Gunderman) almost exclusively focuses on what doctors record in patient records. I agree that this potentially sounds problematic. However, doctors spending so much time recording patient data in medical records does not imply that data should not be recorded to be used later. There are often good ways and bad ways to do something. When someone chooses a bad way, maybe they should adopt a good way rather than abandon the game altogether. Maybe residents today need to learn the “optimal” amount/type of data to record for each patient? This is a new-ish issue, and there are probably some lessons to learn.
The main issue I have with the article is that Gunderman seems to get data confused with data-driven. Many healthcare policies are data-driven but don’t generate any (extra) data to implement. For example, women are no longer recommended to get an annual Pap smear to test for cervical cancer (the “Annual”). Recent changes in cervical cancer screenings are customized according to a patient’s sexual history and test results [Link]. The new recommendations are flexible enough that they maintain a high degree of sensitivity in detecting cancer/precancer while minimizing the number of actual tests performed. The new recommendations are data-driven in that they used large amounts of historical data to estimate the impact of a new testing policy, but no extra data needs to be collected for women for each medical examination. In fact, since women have fewer examinations, less data overall will be collected.
Other medical interventions can be data-driven but simple to implement. The first place winners of the 2012 INFORMS Doing Good with Good OR student paper competition — Jonathan Helm and Greggory Schell — explored a new data driven approach for monitoring chronic disease and for scheduling patient appointments according to lab results rather than according to a fixed schedule. They used a statistical approach to forecast when multivariate test results would no longer be classified as “normal.” The patients scheduled a visit when their was a higher chance that their disease had progressed, thus avoiding unnecessary appointments before then. Again, data is collected here, but no more than what was done before an improved patient scheduling approach was used.
The INFORMS Healthcare Conference will be an excellent place to discuss data-driven approaches to healthcare that offer great benefits with modest costs, and many of the proposed models would not add oodles of data to patient medical records when implemented. I’m looking forward to seeing talks that discuss in greater detail whether data-driven healthcare is a good idea.