Thursday, September 22, 2011

Structure, Process, Outcomes and CKM

Over 40 years ago, Donabedian described a model of assessing quality within a system.  The model had three components: structure, process and outcomes. AHRQ has a detail description of Donabedian’s Model. In short, it states that a system’s outcomes are based on its processes and its processes are bound by its structure. In order to have a long lasting effect on outcomes of interest you need to make changes to the system’s structure and processes.  

An understanding of this system is essential to clinical knowledge management because each component of the model represents a different knowledge type and often a different knowledge source. Knowledge about all three is needed to reach the threshold for actionable knowledge when solving system-based quality problems. If the threshold of actionable knowledge is not met, you will be guessing at which parts of the system to change. In my opinion, the guessing of which parts of the healthcare system to change explains the limited success there has been with improving quality in healthcare.

A stepwise process should be used to create actionable knowledge for system-based quality questions.

Step 1. Define, measure and benchmark the outcome of interest.

Step 2. Determine if the outcome of interest is at goal levels of performance. If not, go to step 3. Otherwise, spend you energy on a different issue.

Step 3. Define, measure and benchmark the processes driving the outcome of interest.

Step 4. Determine which process or processes are preventing the outcome from reaching goal levels.

Step 5. Define the structure of the processes identified in Step 4.

Step 6. Combine the knowledge developed in the previous steps to make actionable knowledge.

Step 7. Now that you have completed the first step of the AURI cycle (analysis), complete the AURI cycle to develop appropriate system changes and track their effect.

This process was used to reduce sepsis mortality at my previous hospital. It was first determined that sepsis mortality was above goal. Two main processes were identified as driving the mortality: the identification of the presence of sepsis and the time it took to deliver antibiotics. The structure around the diagnosis process was that physicians and nurses did not receive standard education on recognizing sepsis and there were no job aids to help them identify patients. With regards to the delivery of antibiotics, one of the most popular antibiotics ordered for these patients had to be mixed in a sterile hood by a pharmacist and thus administration was delayed. 

The structural changes made to improve the outcome were the standard education of nurses and physicians as well as the creation of posters that were posted in the emergency department on how to identify patients with sepsis. Additionally, a formulary change allowed an alternative antibiotic to be immediately available in the emergency department. These structural changes resulted in improved processes for identifying patients and delivering antibiotics more rapidly. The effect on the outcome has been a sustained 37% reduction in mortality for patients presenting to the emergency department with sepsis.

Thursday, August 11, 2011

Freakonomics and CKM

Asking the right questions and using the right analytics can lead to major knowledge breakthroughs. However, it is often difficult to know which questions are the right questions.  

Steve D. Levitt and Stephen J. Dubner, the authors of Freakonomics, would argue that you should not focus on asking one specific question based on what is known, but instead ask a variety of questions and use the right analytics to discover hidden connections between seemingly unconnected variables.  They reason that you will hit a number of dead ends and a number of questions won’t have clear answers, but some will be extremely revealing and meaningful.

The analytic approach they write about is based in economics with five major principles:
“Incentives are the cornerstone of modern life.”
“The conventional wisdom is often wrong.”
“Dramatic effects often have distant, even subtle causes.”

““Experts” – from criminologist to real-estate agents – use their informational advantage to serve their own agenda.”

“Knowing what to measure and how to measure it makes a complicated world much less so.”

Their book analyzes the drop in violent crime in the 1990s and cheating in sumo wrestling, among various other topics.  One large subject area that was mostly absent from the book was healthcare.  

I believe that their approach could be very revealing when applied to healthcare quality improvement, or the relative lack thereof.  Deep reliance on expert opinion rather than objective data, misaligned incentives that don’t support quality goals and limited ability to measure and understand the measurements of quality outcomes must play a role in the state of quality improvement. It would be fascinating to see the authors of the book explore this issue. With a robust health information system in place, many questions could be asked and efficiently answered, potentially revealing meaningful hidden connections to quality improvement in the healthcare setting.

Thursday, August 4, 2011

Better Questions, Better Answers, Better Solutions

Often, when we are struggling to make good decisions, the problem is not the quality of the data available… it is the quality of the question.  This is because the threshold for achieving actionable knowledge is dependent on the question being asked. The more simple and specific you make the question, the easier it is to gather enough information to reach the threshold for actionable knowledge. If the question being asked is too broad, the question itself may prevent the creation of actionable knowledge.

A perfect example of this in the medical field is the current discussion about salt intake. The main concern is the sodium component of salt. Diets high in sodium have been associated with increased rates of hypertension, which is clearly linked to an increased chance of heart attack and stroke. Because of these associations, there has been a public health campaign to advise adults in the US to reduce their salt intake. However, the campaign to reduce salt consumption has recently become controversial. This controversy stems from asking a question that is too broad to facilitate the creation of actionable knowledge. The question being asked is “What should we tell the US adult population about salt intake?”

A recent article in the Archives of Internal Medicine demonstrated a reduction in cardiovascular mortality associated with lower sodium diets. The study was statistically adjusted to be a representative sample of the US population. This article gives the appearance of creating sufficient actionable knowledge to tell the US population to eat less salt. However, two other recent studies create serious doubts about that conclusion. The first study, published in the New England Journal of Medicine, was also a study modeling the US and showed that the mortality benefit for a lower sodium diet had a significantly larger impact for blacks than for whites. The second article, published in the Journal of the American Medical Association, was a European study that only looked at a relatively young and healthy white cohort; this study demonstrated a large increase in mortality for the subjects with the lowest sodium diets.

These articles appear to disagree and there have been discussions about the methods of each study. No study is perfect and the conclusions from any of the three may be proven incorrect, but it is also possible that all three may be correct. It is possible that young and healthy whites in Europe may be harmed by a diet that restricts salt intake. Sodium is necessary for several biological functions. However, white, young, and healthy only represent a small group in a study that utilizing a sample representative of the adult US population. The potential harm to the white, young, and healthy cohort may be washed out by the benefits for other groups.

These studies create debate because the question being asked is too broad. If you believe that eating less salt will harm a cohort of people, it is unethical to tell them to do so. If you are then trying to take action based on the question “What should we tell the US adult population about salt intake?” you have quite a dilemma.

I suggest in these cases you don’t try to solve that dilemma with more data, information and knowledge. You solve it by changing the question. If some groups are helped by an action and others harmed, the question should not be “what advice do we give the whole population?” The question should be “how do I advise each cohort in the population?”

With regard to salt intake, we should stop asking “what should we tell the US adult population about salt intake?” and start asking "who in the US do we need to tell to eat less salt?"

Monday, August 1, 2011

Health Information Exchanges and CKM

According to SearchHealthIT, health information exchange is "the transmission of healthcare-related data among facilities, health information organizations (HIO) and government agencies..." Health information exchanges have been suggested as a tool to improve the delivery of care, especially in emergency departments and other acute care settings. However, there has been little evidence that proves they create actionable knowledge.

One study that indicates HIEs do produce actionable knowledge was presented at the Society of General Internal Medicine Annual Conference earlier this year. Dr. Lisa Mabry and colleagues presented a poster entitled "Does Health Information Exchange Use Improve Adherence With Evidence-Based Guidelines for Neuroimaging in the Emergency Evaluation of Headache?" Their study investigated whether ED staff use of the MidSouth e-Heatlh Alliance HIE, a health information exchange in the greater Memphis area, reduced the use of neuroimaging in patients presenting with repeat episodes of headache. Their results showed that although the HIE was only used in 21.8%  of encounters, when it was used, it was associated with a 76% decreased odds of neuroimaging. Additionally, adherence to evidence-based guidelines was improved.

These results indicate that, at least with regard to headache, the HIE does provide the user with actionable knowledge. If the HIE was used more frequently, unnecessary cost, testing, and radiation exposure could be avoided.

As creation and utilization of health information exchanges expands, it is important to design them in a fashion that assists users in reaching the threshold for actionable knowledge. Focusing on overcoming technical challenges to allow for the transmission of data within HIEs rather than insuring their ability to communicate actionable knowledge will limit the effectiveness and relevance of this technology.

Thursday, July 28, 2011

CKM and Positive Deviance

Positive deviance is a strategic approach to identifying top-performing individuals or groups, and disseminating their special knowledge to the remainder of the group. If you've read Malcolm Gladwell's book Outliers, you are familiar with the concept of positive deviants; positive deviance identifies these outliers for the basis of process change and improvement. Historically, positive deviance was used to improve conditions in the developing world, including malnutrition and peri-natal mortality. Recently, positive deviance has been applied to quality improvement in health care.

Positive deviance can play an important role in health care. The process not only identifies the positive outliers but also provides the framework for disseminating their knowledge. However, until recently, identifying individuals and groups with significantly better outcomes has been difficult in health care.

As organizations increasingly implement robust HIS systems, the systematic identification of positive deviants has become possible. From this starting point, the AURI cycle can be used to create actionable knowledge which can be disseminated and implemented.  This is why clinical knowledge management is an excellent platform for quality improvement.

To use positive deviance in the CKM process for a QI project, the "analyze" step of the AURI cycle is divided into three actions:

  1. Identify individuals or groups with significantly better performance in your outcome of interest
  2. Utilize quantitative techniques to determine differences in care delivery between positive deviants and others
  3. Utilize qualitative techniques to determine differences in care delivery between positive deviants and others
During the "understand" step of the AURI cycle, the differences in care delivery identified in the "analyze" step are examined to determine which are most significant. These significant differences are then incorporated in the "redesign" step. During the "implement" step, these differences in care delivery and their importance to the redesign are communicated to the group. The cycle continues by analyzing whether the remainder of the group has adopted the new care processes and if they have had the desired impact on outcomes. 

By using positive deviance in conjunction with the CKM process, quality improvement projects can demonstrate rapid improvement. 

Monday, July 25, 2011

CKM and Readmissions

The Journal of General Internal Medicine recently published my paper examining the effects of standardized discharge instructions on readmission. Readers may be surprised that standardizing our discharge instructions to meet consensus recommendations did not reduce readmissions.

However, in light of the necessity of providing actionable knowledge to improve decision making, these findings begin to make sense.

The standardization of the discharge instructions focused on insuring that a set of recommended components (ie: discharge medications, follow-up appointments, contact information, etc.) were always provided to the patient at the time of discharge. On the continuum from data to actionable knowledge, these components are information. Although patients can often synthesize these various information points into knowledge, good decision making is born of actionable knowledge. To move from information, to knowledge, to actionable knowledge, you first have to understand the question that discharge instructions are trying to answer: how should I take care of myself outside of the hospital so that I don't have to come back? 

In light of this question, actionable knowledge is based on an understanding of the patient's own health and disease processes as well as the actions necessary to maintain their health. The capacity to understand and perform these actions is highly variable among patients. Therefore, interventions to prevent readmission need to be customized to each individual patient rather than standardized.

Whether these customized interventions will be time efficient and cost effective has not yet been determined. However, in my opinion, discharge interventions that are not customized to the patient will continue to show lackluster results.

Wednesday, June 29, 2011

CKM and Serenity

Clinical informatics and electronic health records are often offered as the path to high quality medical care and reduced adverse events. However, technology can only help deliver the best care medical knowledge can achieve. Frequently, even with the best implementation, technology can only get adverse events and complications of care to an irreducible minimum. After that new medical knowledge is needed. This is where CKM can be extremely important. Even when CKM doesn’t provide new knowledge for medical advancement, it can provide a clear indication if an irreducible minimum has been meet.

When considering this, I often think of the Serenity Prayer.

God grant me the serenity 
to accept the things I cannot change;
courage to change the things I can;
and wisdom to know the difference.

CKM can help provide the wisdom to know the difference. Well designed CKM can determine if patients received ideal care for specific disorders, i.e. early goal directed therapy for sepsis. If 100 patients all received perfect care and 20 died then there is nothing to change. However, if 100 patients received less than perfect care and 30 died, we need to have the courage to change the system.  

Monday, April 25, 2011

American Airlines and CKM

The other morning I was sitting in a meeting and a question was raised. That question was "do we really know that understanding our data better, and using tools such as data visualization, will really improve healthcare?"

Only in healthcare would this question still be asked in 2011.

Industries of all types have learned over the last 20 years the power of developing knowledge based on their actions and the actions of their customers. Many of these industries are not as complicated as healthcare. But I believe that the idea that healthcare is the most complex industry is erroneous; however, it's been my experience that this idea is deeply rooted in the culture of medicine. It seems to be felt that since there are so many uncontrollable factors such as patient complexity, patient compliance, variations in disease presentation, and the intricate web of payers and delivery systems, it is not possible to understand the "healthcare system" with data. I do not believe this is true, and in support of my position, I'd like to present a case from an industry that is at least as complex as the healthcare system: the airline industry.

Within the airline industry, the uncontrollable variable range from mechanical problems to passenger behavior to natural disasters to terrorist attacks... to the most unpredictable of all- the weather! Despite all of these variables, there are many examples of how the airline industry has transformed data into knowledge, improved service, and remained profitable.

Take, for example, the recent case study reported by American Airlines. American Airlines had identified fraud as a major cost to their business. However, they had no data warehouse technology or knowledge management plan for addressing fraud in their system. It was originally estimated that an effective data analytics system would save the company $150,000 per year. Using an "off-the-shelf" data warehouse solution, great gains were immediately seen, and ultimately saved the company $5 million over 5 years. The success was credited to the new system's ability to identify forms of fraud that the company never knew existed and giving the company the ability to make changes to eliminate those causes.

While it is true that we may be many decades away from being truly knowledgeable about how the US healthcare system works, it is also true that there are many technologies available today that, if utilized in healthcare, could have immediate and meaningful impact. Targeted solutions can quickly exceed expectations when we focus on creating new actionable knowledge with current technologies.

Thursday, April 14, 2011

CKM and Spaghetti Sauce

One of the key components to clinical knowledge management is the discovery of actionable knowledge. The process of discovering knowledge is often more of an art than a science. A complex part of the art is asking the right question. The knowledge discovered and developed through data driven techniques such as data mining and statistical hypothesis testing are always framed by the questions being asked. Ask the wrong question, generate the wrong knowledge. Unfortunately, unless you know you are asking the wrong question you assume you are working with the right knowledge. 

Gains can be made when making decisions with the “wrong knowledge,” but they will be less than the gains made if decisions were made from the right knowledge. An excellent example of how knowledge is improved when you ask the right question is described by Malcolm Gladwell in his TED talk about the food industry and a spaghetti sauce breakthrough.

Gladwell describes how chunky spaghetti sauce revolutionized the food industry…  because companies stopped asking their research teams to find the perfect food and started to ask them to find the best food for a cluster of people. The right question was not “what is the perfect spaghetti sauce?”(or mustard or soda.) The right question was “which varieties of spaghetti sauce greatly appealed to large groups of people?” The result was more food options, happier customers and increased revenue.

Monday, April 11, 2011

Communication: Changing Behaviors

I just received my copy of the Journal of the American Medical Informatics Association. In it, I found an article that is very revealing with regard to communication: Actionable reminders did not improve performance over passive reminders for overdue tests in the primary care setting (abstract). In the article, El-Kareh et al. investigated the impact of altering a passive reminder to be a passive reminder that facilitated direct ordering of recommended tests. The reminders created highly sensitive and specific actionable knowledge from the clinic's electronic health records. The new "enhanced" reminders had absolutely no affect on the screening rates of bone density exams, HgA1C, and LDL monitoring; in fact, they may have decreased the rate of screening mammograms.

From a health IT standpoint, this outcome makes no sense whatsoever. The software provided streamlined functionality to place orders and reminders were based on reliable knowledge. However, a psychology major would have predicted these results. So, why didn't the enhanced alerts work?

They did not work because they did not effect the physician's intention to perform the preventative services. But in order to understand this, you must first understand the theory of planned behavior.

The theory of planned behavior, like most psychological theories, is fairly complex. I like how Wikipedia explains it here, if you are interested, but basically, what it comes down to is this: when a person has the option of whether or not to do an action, their choice to act or not act is dependent upon their intention to act.        

The intention to perform a behavior is the summation of three components: a person's attitude towards the behavior; the subjective norm; perceived behavioral control. Increasing someone's intention to perform the behavior increases the frequency of the behavior. So how do you increase a person's intention to perform a behavior? By changing their attitude towards the behavior, their perception of how others view the behavior, and the beliefs about their ability to complete the action.

The "enhanced" reminder studied by El-Kareh et al. did not alter any of the three components needed to increase intention. Both the basic reminder and the enhanced reminder were passive meaning the physician's perception of how colleagues felt about him/her completing preventative screening would not have changed. Streamlining the ability to place orders would not effect the physician's attitude towards providing screening to his patients. It is possible that the perceived ability to complete screening would have been increased due to increased ease of placing orders; however, the authors report that 79% of the physicians almost never used the system or were unaware of the functionality, despite receiving training on the new reminders. Thus, it is not surprising that the enhanced reminders did not result in improved care.

When you are sharing actionable knowledge with the intention of effecting behavior, you will be most effective when you keep the three tenets of intention in mind. For instance, in our VTE project, we purposely addressed each of these tenets in our interventions. We instituted a major, mandatory education program to change the attitude of providers. We used forcing functions and pop-up alerts to change the perception of VTE prophylaxis and reinforce the perceived importance of stratifying and prophylaxing patients. We embedded guidelines for risk stratification in the risk assessment tool and allowed order entry with a single click from the pop-up alerts to impact beliefs about the ability to complete the behavior.

Using the theory of planned behavior to optimize communication will only be effective if the health information system is designed to be functional for the provider. However, as El-Karah et al. demonstrated, improved functionality without improved communication is often ineffective.

Thursday, April 7, 2011

Communication: The Basics

One of the most fundamental elements of communication is using the same vocabulary so that each participant can understand the others. The best example I have of this comes from my 3 year old daughter.

She was walking all kinds of funky across the living room. I called to her, "Anna, do you have a wedgie?" She turned around, looked me square in the eye, and with all the will in her little body corrected me, "No! My panties are stuck in my butt!"

Clearly, this was a situation where we were not communicating well because we did not have a shared vocabulary. Whether evaluating information to make knowledge, sharing knowledge to effect behavior, or developing understanding in the AURI cycle, all parties involved must have a shared vocabulary. In my experience, this is a common pitfall in HIS implementations and data-driven quality improvement.

So, how do you establish this shared vocabulary in teams utilizing clinical knowledge management processes or quality improvement initiatives, such as with the AURI cycle? Defining terms and establishing metrics must be your first order of business in any of these projects; there's really no point in participating in these processes if the participants are not able to equally engage and speak with a common vocabulary.

As clinicians, we all understood what a VTE was, but in the first meetings of the VTE team, we had to define VTE clearly and establish which VTEs would be included in our metrics. For instance, would we count hospital-acquired VTEs in upper extremities that were associated with PICCs or other central lines as nosocomial VTEs and include them as a target of our interventions? We also had to decide if we would include VTEs discovered as outpatients and during readmissions or only those discovered during a single hospital admission as part of our intervention. Ultimately, we chose to include all VTEs, regardless of physical location or the clinical setting in which it was discovered.

As you can see, making another choice would have changed our evaluation of metrics, knowledge, and understanding. The interventions described in the VTE project were the result of the definitions and metrics we agreed upon as a team. If we had not established a common vocabulary, we would have experienced significant delays in the progress of our project, and possibly have been significantly less successful than we were.

Keep in mind that communication is dynamic and anytime a new member enters, you need to ensure you re-establish the shared vocabulary. For instance, when writing out the above story, my 8 year old daughter read it. Then she turned to her mother and asked, "Momma, what is a weed-ghee?"

In my next post, I will outline the science behind communicating with the goal of effecting volitional behavior.

Monday, April 4, 2011

Success Story: VTE Reduction

As part of the QualityBLUE Pay for Performance partnership between Highmark Blue Cross Blue Shield and Penn State Hershey Medical Center, hospital-acquired venous thromboembolism (VTE) was identified as an area for quality improvement.  This coincided with the release of the latest ACCP Guidelines on Antithrombolytic and Thrombolytic Therapy (8th Ed.) in the summer of 2008.

An interdisciplinary team of nurses, physicians, pharmacists, quality improvement specialists, and informatics specialists was assembled to determine how to implement the new guidelines at HMC. Through the use of Clinical Knowledge Management and the Analyze-Understand-Redesign-Implement cycle, we made tremendous gains in this QI project, including a sustained 25% reduction in nosocomial VTEs, reduced mortality associated with nosocomial VTEs, and cost avoidance estimated at $2-4 million annually. Let me walk you through the use of the CKM process and the AURI cycle in this QI initiative.

The CKM Process

Capturing & Classifying: At the time of the project initiation, HMC had been using an electronic medical record (EMR) with computerized physician order entry (CPOE) for greater than two years. This mean that in addition to the best practices outlined in the ACCP Guidelines, we also had data from greater than 50,000 inpatient visits available to us. This included: risk stratification data like demographics and clinical conditions; use of pharmocologic and non-pharmocologic prophylaxis; time elapsed from admission to first prophylaxis dose; rate of occurrence of nosocomial VTE.

Retrieving: Queries were developed to gather information about current VTE prophylaxis behavior from HMC's clinical database. Results from the queries were transferred into Excel spreadsheets.

Evaluating: Rates and timing of appropriate prophylaxis and rates of development of nosocomial VTEs were determined to identify gaps between current recommendations and current HMC practice. Underutilization of risk-scoring at admission as well as underutilization of pharmacologic and mechanical prophylaxis were identified. Inappropriate risk stratification was common as was inappropriate use of prophylaxis. 

Concurrently, the pharmacists and clinicians from both Medical and Surgical services condensed the ACCP Guidelines to an easy-reference pocket card that contained risk stratification guidelines and appropriate treatment options.

Sharing: Required education for all pharmacists, physicians, and nurses was provided along with the quick-reference pocket cards. The education reviewed the new ACCP Guidelines as well as required changes to HMC practice.

Action: Based on the education and availability of the pocket cards, there was a modest improvement in guideline compliance and a slight decrease in hospital acquired VTEs. 

A second cycle of the CKM process was then initiated. The data that was captured and classified after the roll-out of the education and pocket cards was retrieved and evaluated. It was determined that significant opportunities for improvement were as of yet untapped. It was also determined that a more structured and standardized approach was needed to accommodate the resident learning curve.

The sharing step was multifaceted. A clinical decision support (CDS) tool was created that included forcing functions at the time of admission that required VTE risk stratification on all patients. Residents were provided with the stratification criteria at the time of the risk assessment. Interactive alerts were developed to present providers with prophylaxis guidelines based on patient risk stratification at the time of order entry. Providers were required to either place appropriate prophylaxis orders or document contraindications.

The resulting actions from the providers were immediate. There was an increase in appropriate prophylaxis and a decrease in nosocomial VTEs. This reduction has been sustained for two years and through two intern classes. Partial results were presented at the Society of Medical Decision Making Annual Conference in October, 2010.

The AURI Cycle

As you can see from the above, the traditional Plan-Do-Study-Act (PDSA) cycle was not followed during this initiative. Most notably, no system changes were made until the data was carefully analyzed and understood by all team members. However, the AURI cycle was extremely effective in producing sustained behavioral change and improved outcomes. Let's examine how the steps of the CKM process fit into the AURI cycle.

Analyze: The retrieval and evaluation of both internal and external data to create new knowledge comprises this part of the AURI cycle. In this QI project, retrieving and evaluating data from HMC's EMR and studying and condensing the ACCP Guidelines represents the analyze component. The new knowledge gained from the analyze phase was three-fold: the correct prophylaxis choices were often clear to experienced clinicians but not to inexperienced residents; ideally, a single drug would be suggested for pharmacoprophylaxis; risk stratification was rarely performed at admission.

Understand: Sharing the new knowledge derived from the analysis phase with the quality improvement team comprises this part of the AURI cycle. The team identified barriers such as baseline resident knowledge, inability to use a single low-molecular weight heparin, insufficient availability of mechanical prophylaxis devices, and concerns for the feasibility of improvement with voluntary compliance.

Redesign: The redesign phase is a result of the CKM process rather than a step in the CKM process. The new standard of practice at HMC that involved risk stratification of all patients at the time of admission based on a standard risk stratification system as well as the testing of multiple models of mechanical prophylaxis devices by the Department of Nursing comprises this part of the AURI cycle. 

ImplementationSharing new knowledge derived from the analysis phase with all clinicians and other stakeholders comprises this part of the AURI cycle. The first barrier to success, baseline resident knowledge, was tackled during the roll-out of required education. Additionally, the Operations Department purchased an adequate number of machines and made them easily accessible. The EHR was modified to capture the use of mechanical prophylaxis devices and the use of pharmacoprophylaxis continued to be captured. We were not able to capture the timing of the risk assessment, which unfortunately, had to be done manually on a small sample of the patients.

Despite seeing modest gains, a second AURI cycle was needed to address the remaining two barriers. As noted above, the analysis phase identified the need for a more structured, standardized approach. The understanding phase led to the new knowledge that risk assessment needed to be required instead of voluntary, prophylaxis needed to be simplified, and appropriate guidelines needed to be shared in real-time with residents. The redesign phase resulted in the clinical decision support tool described above as well as a policy change that Pharmacy would substitute appropriate low-molecular weight heparin for patients with renal failure. The implement phase was the education and go-live of the CDS tool.

The time elapsed from the beginning of the first AURI cycle to the implementation of the second AURI cycle was only 9 months. As you can see, the CKM process and the AURI cycle can allow for rapid institutional improvement and identification of barriers to improvement. The success of this project highlights the way that efficient quality improvement methodologies result in significant financial gains as well as reduced morbidity and mortality.

Thursday, March 31, 2011

The AURI Cycle

Historically, quality improvement in healthcare has been done on the basis of expert consensus. One of the most supported models in that vein is the Plan-Do-Study-Act cycle endorsed by the IHI. The cycle begins when a quality improvement intervention is identified. Once the intervention is identified, you plan how you will study the effect of the intervention, including data collection methods. The intervention is then enacted and data collection begins in a small sample population. Once data collection is complete, it is analyzed and compared to predictions to identify lessons learned. The intervention is modified based on what was learned, and the cycle is repeated with the modified intervention.

This PDSA process can be extremely costly, time consuming and involve many false-starts. In an era of expanding health information systems, a new model based on Clinical Knowledge Management is possible. Over the last several years, I have developed a new data-driven model; the CKM process, when applied to quality improvement, produces the Analyze-Understand-Redesign-Implement cycle.

I have had many successes using the AURI cycle applied to various QI projects. However, the AURI cycle is only possible when you have a robust health information system in place, and it is continually capturing and classifying data.

Analyze: Retrieve and evaluate data and information to create actionable knowledge with regard to your QI question. This step employs multiple statistical methods and data-analytic approaches.

Understand: The "reality check" of the new knowledge. New knowledge is presented to stakeholders to determine what is feasible and to identify barriers to success.

Redesign: The alteration of current processes, both major and minor, based on the understanding of the new knowledge.

Implement: The education plan, roll-out of new processes, and modifications to the health IT applications to ensure the continued capturing and classification of relevant data. Due to the necessity of altering HIT applications, the QI initiatives often affect a large sample of patients.

The reason large scale interventions and subsequent extensive data collection and analysis is possible is the continuous capturing and classifying of data in robust health information systems. Incorporating the ability for the health IT applications to capture the redesign process into the implementation phase is critical to successful AURI cycles.

Where robust HIS is available, the AURI cycle of quality improvement is a more efficient and cost effective model than the PDSA cycle. However, due to the large number of patients affected, it must be implemented strategically.

Monday, March 28, 2011

Hans Roslings on Data, Information, and Knowledge

This TEDtalk by Hans Rosling first debuted in 2006, but the points he makes about data, information, and knowledge are priceless.

As you can see from this presentation, the ability to visually display data allows for in-depth contextualization, and the subsequent formation of new knowledge. Although the knowledge Dr. Rosling presents does not reach the threshold of actionable knowledge, it is easy to see how one could ask the right questions from a global health perspective to create actionable knowledge from the data available using his data visualization teachniques.

To further explore the new knowledge Dr. Rosling is creating, I highly recommend viewing his other TEDtalks. You can see his full bio and links to additional presentations here.

Thursday, March 24, 2011

Who is the End User?

It is generally accepted that information systems should be designed to benefit the end user. However, it is often unclear who the end user is. The answer to this question seems even more unclear in health information systems.

Thomas Goetz, executive editor or Wired and author of "The Decision Tree: Taking Control of Your Health in the New Era of Personalized Medicine" articulates a great argument that the end user of health information systems should be the patient.

I've held this same opinion for quite some time. Watching this TED talk causes me to think back to my studies of the Kimball Group's method of data warehouse implementation, outlined in their book, The Data Warehouse Lifestyle Toolkit. A large portion of this book was dedicated to end user identification and needs assessment. 

While there are many stakeholders in health care, and any one of them may be an end user for an individual application, the ultimate beneficiary of all health care information systems should be the patient. In the field of Clinical Knowledge Management, this means the focus must be maintained on ensuring that knowledge for improved decision making is made available not only to clinicians, but also to patients.

Monday, March 21, 2011

CKM and Central Line Infections

A recent article from the CDC reports that "Compared to 2001, approximately 58% fewer bloodstream infections occurred in 2009 in ICU patients with central lines." (see full Vital Signs article)

This improvement has largely been based on the acceptance and implementation of the "Central Line Bundle." According to the IHI, the bundle consists of:
  • Hand Hygiene
  • Maximal Barrier Precautions During Insertion
  • Chlorhexidine Skin Antisepsis
  • Optimal Catheter Site Selection
  • Daily Review of Line Necessity
This bundle may seem unrelated to clinical knowledge management, but it is an excellent example of the potential effects of clinical knowledge management, even when the process is unintentionally used.

This bundle was originally created during a research study at the Johns Hopkins University in 1998, published in 2004, performed by Peter Pronovost and others. The bundle compiled evidence-based practices for preventing central line infections into a single, digestible checklist. While the researched might not have set out to use the clinical knowledge management process, their success is a direct reflection on how well the CKM process was executed. Let me walk you through each step of the process, and how it was completed in this study.

Capturing and Classifying: The capturing and classifying of medical data, information, and even knowledge occurs on almost a daily basis through the publication of peer-reviewed medical journals and websites. The evidence on best practices to reduce central line infections was available to the researchers at the time of their literature review.

Retrieving: The identification and collection of articles that contained data, information, and knowledge about best practices to prevent central line infections represents the retrieval component of the CKM process.

Evaluating: The thoughtful analysis of which components in the literature review represented the greatest successes and were necessceary to include in their intervention embodies the evaluation step of the CKM process.

Sharing: The creation of the new knowledge represented by the checklist was not enough to initiate change; the researchers had to share their results in a meaningful way. They did this through a well-designed education and implementation plan.

Action: The change in practice by clinicians that occurred after the implementation of the checklist is the action component of the CKM process.

Just like the CKM process requires you to constantly be re-evaluating the effectiveness of your action, the researchers reevaluated their intervention. When the intervention was shown to be effective, the researchers shared this new knowledge with the medical community at large through the publication of their results. As others around the country implemented this checklist as the result of Johns Hopkins success, the results were analyzed and published. The communication of these results spurred even more hospitals and clinicians to implement the checklist and eventually led to endorsement by the CDC.

This continued utilization of the CKM process, although unknown to those using it, has greatly contributed to the 58% reduction in central line infections reported by the CDC. Imagine how much greater that reduction would be if the process had been used intentionally.

Saturday, March 19, 2011

The Process of CKM

The process of clinical knowledge management involves the capturing, classifying, retrieving, evaluating and sharing of clinical data and information in a way that creates actionable knowledge and provides the context for effective care decisions and therapeutic actions.

But what does this really mean?
Capturing: The process of gathering data while it is being used for a separate purpose
Joe Smith visits his PCPs office for an annual exam. While there, his weight is measured and entered into the electronic medical record.
Classifying: The policies and procedures that determine where, how and when data is stored
The office policy is to measure and record weight in kilograms to one decimal point. The weight is a mandatory field in the electronic intake form and must be completed to finalize the nursing documentation.
Retrieving: The act of collecting data in an organized way from wherever it has been stored
A researcher writes a program to query the electronic medical record and gather all recorded weights at the office of Joe Smith’s PCP.
Evaluating: The purposeful analysis of retrieved data to generate information and create knowledge
The researcher determines that the average weight of patients at the office of Joe Smith’s PCP increased 3 kg after the opening of a new ice cream shop.
Sharing: The communication of actionable knowledge in a manner that allows for effective care delivery
The researcher presents his findings at a staff meeting with all the physicians at the office along with the nutritional information provided on the ice cream shop’s website.
Action:  The purposeful activities resulting from the knowledge that has been shared.
The physician group meets with the owner of the ice cream shop, and agrees to sponsor a nutritionist to work with the owner to develop lower calorie options.
Since the weights will continue to be captured at each visit, the process of clinical knowledge management can be repeated to determine the effect of the physician’s intervention.

The above example may seem straightforward, easy, and even trivial. But that’s because it used a single data element. When examining actual clinical questions with dozens of data elements and millions of datum, the process is far from straightforward and easy. And the generation of actionable knowledge for a real clinical problem is rarely trivial.

IT, Informatics, Information Systems, & CKM

But what does this mean for clinicians and administrators who are trying to make decisions?
It means that health IT is not sufficient. Even clinical informatics is not enough. Decision makers need the right hardware, software, infrastructure and personnel trained in HIT, clinical informatics, and clinical knowledge management. Robust health care information systems are required.
Health information technology (HIT) is the hardware and software that is responsible for the collection and presentation of clinical data and information. HIT is continually occurring in the health care setting that utilizes an electronic medical record. HIT is the foundation of clinical knowledge management because it involves the capturing and classifying of data.

Clinical informatics is the process of contextualizing the data stored in health IT applications. The contextualization creates clinical information that is presented to clinicians or decision makers when interacting with software, i.e. displaying an abnormal lab value as red and a normal lab value as blue.  Clinical informatics is difficult to achieve without an electronic medical record and occurs when users retrieve stored data. Clinical informatics forms the supporting pillars of clinical knowledge management as it involves the retrieval and evaluation of data.

Health information systems are the combination of people, processes and technology that facilitate health care decision making by the creation, storage and sharing of clinical knowledge. Health information systems utilize HIT and clinical informatics, but only occur when organizations purposely set out to facilitate decision making based upon sufficient information to create actionable knowledge. Simply installing an electronic medical record application does not create a health information system. Health information systems are the complex structure that allow for the efficient execution of clinical knowledge management.

With robust health information systems in place clinical knowledge management can be leveraged to improve the delivery of health care.

Data, Information, and Knowledge

In order to understand the process of clinical knowledge management, you must first know the difference between data, information, knowledge, and actionable knowledge.
Data is isolated facts. This could be the time, temperature, heart rate or any other number of discrete measurements. Information is data put into context. For instance, is the time EST, CST, or PST? Is the temperature a body temperature or the ambient temperature? Was the heart rate measured on a neonate or an adult?
Once data is put into context to create information, it can be combined with other information to create knowledge. For instance, noon in Tokyo occurs thirteen hours before noon in New York City. While a body temperature of 90F is definitely low, an ambient temperature of 90F may be high or low.  A heart rate of 130bpm in a newborn is average while a heart rate of 130bpm in an adult is fast.
However, none of this knowledge allows you to make a decision, let alone a good decision. It has to be combined with more information to create actionable knowledge. The threshold for reaching actionable knowledge depends on the question.
Some questions are easy: when do I dial into the conference call? You need the time of the meeting, the time zone the meeting is in, your time zone and the time difference between the two time zones. Some questions are hard: is the patient’s heart rate a cause for concern? You need the patient’s current heart rate, their historic heart rate, their age, the normal values of heart rate for their age, their current activity level, the rhythm of their heart beats, the shape and pattern of the heart beats on an EKG and their clinical presentation. 
But whatever the complexity of the question, generating actionable knowledge will improve decision making.