Physical Therapy in Thin Slices…..When Less is More

One of Larry’s contrarian truths of physical therapy is that the last visit is the most important, not the first. It seems many PTs view it just the opposite as is evidence by conducting what usually amounts to a data collection safari on the initial visit. Hey, what are we looking for anyway?!

Cook County Hospital, 1996: Chicago's principal public hospital is in crisis, with a major contributing problem of indigent patients presenting to the ED with chest pain. The problem is that many of these people aren’t having a cardiac related issue at all (only 10% presenting actually do) yet all were being admitted the CCU for observation at the expense of $2000.00 a day. There was no rational, standardized way of making the decision of who goes to the CCU vs the observational unit. Enter Lee Goldman who collected data and developed what amounts to a clinical prediction rule (CPR) to determine which patients had urgent (relevant) risk factors predictive of major cardiac complications (and therefore needed to be admitted). Based on the ECG and 3 simple findings (unstable angina, fluid in lungs, and SBP <100mmHg) there was a whopping 70% improvement in identifying these patients (95% probability). Was this readily received? No way, after all how could a guide consisting of a few key indicators perform better than a trained physician?

In the book Blink (great read by the way), author Malcolm Gladwell discusses how we often make decision in just a few seconds based on pattern recognition and how when informed, these decisions are much more accurate than engaging in the intensive cogitation to which we often feel obligated; ie. thinking in thin slices. In fact, what was found to be the problem in the Cook County case was that physicians had too much information, which only confounded and confused their judgment, leading to less accurate judgments.

 

I believe this often the case in physical therapy practice as well. We have so many things we think are important to measure and collect so as to guide our decision making and treatment planning. Is all this information necessary and helpful or, as in the Cook County case, does it only serve to confuse in some cases and increase practice variation? The key to effective clinical decision making is not merely clinical reasoning based on exhaustive clinical examination findings, but rather knowing which data to seek and what single or combination of findings are most important for establishing a diagnosis, guiding treatment, or predicting outcome.

We have a long way to go in this regard but are making progress. The real question and concern is whether therapists are ready to embrace these data when available or stubbornly cling to standard operating procedure.

Remember, in many cases less is more. Are you ready to think in thin slices? Cook County eventually was and became the 1^st medical institution in the country to devote itself to the Goldman algorithm for chest pain.

 

Rob

 

 

Comments

[Noel Squires]

Rob great thoughts…I have been in a number of clinical examinations when I ask myself “why am I doing this? What impact will these data have on my evaluation, and intervention decisions?” Reflection upon my practice patterns has changed my thought process during my examination to a variety of different styles. Some examinations have consisted solely of ROM and pain measures, while others consist of multiple system screens, with tests and measures collected from everywhere. Too much information can certainly “cloud my vision.” For example, when a 70 y/o man with no red flags or UE syx presents with neck stiffness and loss of motion upon visual examination how many planes of cervical motion will someone collect from their inclinometer prior to engaging the treatment threshold?
Keeping a diagram of the diagnosis/treatment thresholds handy has kept me focused on more than one occasion while allowing me to adminster care in small slices.

[Andrew Guccione]

Thanks, Rob, for raising one of the most pressing issues we will face in the next five years, ie, what patient characteristics (including psychosocial and environmental factors)best sort patients on admission and can serve as predictors of outcome? As "pay for performance" and the possibility of an alternative payment system loom large, this will truly become the money question.
While your post emphasizes clinical prediction, we shouldn't miss the opportunity to mention that those variables which best predict a clinical outcome from a PT's perspective may not be the best predictors of utilization and cost, when the goal is to sort patients upon admission into payment groups - which is a very high priority. Furthermore, we should be very aware of the benefits of the type of statistical analysis (CART) Goldman used to develop a classification algorithm with the lowest misclassification rate. I see very little discussion of misclassification rates in our literature despite the emergence of sorting rules. Similarly, I notice a dependency on linear regression models which assume linear interaction effects (when interaction terms are even specified at all). One of the advantages of recursive partitioning methods like CART in diagnostic classification is that non-linear interactive effects between particular ranges of values of interaction terms are identified through the statistical process. This makes a lot of sense if, as has already been raised, ultimately we want to answer the question, "How much ROM, in relationship to other clinical varables, can my patient lose before I get concerned and determine that I treat it?"
Incidentally, Cook and Goldman first published on that topic in 1984, so the algorithm was available for 15 years before. Tells us something about scientific dissemination.

[Jon Newman]

Andrew,

I'm fascinated by your post. I'm wondering if you or anyone else has a good resource on recursive partitioning methods. I've looked around and found a few links but if someone has a text that explains things well I'd be interested.

When I first read Blink (ok, I listened to it on CD) I tried looking up that particular study and failed to find it. I tried again tonight but still came up empty. Does anyone have a specific reference?

I did find one study where they tried to implement the CPR in an actual clinical setting but it was only used on about half the patients. A follow up questionnaire revealed that the algorithm would have changed 1% of decisions made without it (according to the abstract--I didn't have access to the full text). Could this explain its low implementation?

jon

[Andrew Guccione]

Relative to CART, the original reference is Breiman L, et al. Classification and Regression Trees. 1984. Belmont CA: Wadsworth International Group. Be forewarned, this is a very mathematical treatment of the subject. A former colleague of mine, a statistician with a PhD in math before she became interested in statistics, told me she found it challenging to read. Some of the details of the algorithm, particularly preferences for splitting nodes, are more than a bit too technical for me. I generally recommend two approaches if you want to understand the technique: reading studies that have used it, and doing a CART analysis on a reasonably large and robust data set. Varying some of the choices for splitting nodes will quickly give you an end-user appreciation of the technique.
Some helpful readings (although not recent - shows you where I haven't kept up with the literature) include:
Bloch DA, Moses LE, Michel BA. Statistical approaches to classification: methods for developing classification and other criteria rules. Arthritis Rheum 1990; 33:1137-1144
Cook EF, Goldman L. Empiric comparison of multivariate analystic techniques: advantages and disadvantages of recursive partitioning analysis. J Chron Dis 1984; 37:721-731
Cook EF, Goldman L. Asymetric stratification: an outline for an efficient method for controlling confounding in cohort studies. Amer J Epidemiol 1988; 127:626-639
If I might also be so forward as to suggest my own work, you might find these interesting:
Guccione AA, Anderson JJ, Anthony JM, Meenan RF: The correlates of health perceptions in rheumatoid arthritis. J Rheumatol 1995; 22:432-439
Guccione AA: Patient classification, stratification and interaction: Methodological notes from epidemiology and the social sciences. In Proceedings of the 14th Eugene Michels Forum. Alexandria VA: American Physical Therapy Association, 1994, pp 11-18

One last thought. The original reference for the Goldman algorithm is: Goldman L, et al. A computer-driven protocol to aid in the diagnosis of emergency room patients with acute chest pain. N Engl J Med 1982; 307:588-596. Perhaps the title made people afraid in 1982 and nobody read it? Or perhaps, the reason that there was only a 1% difference when the algorithm was implemented was that most medical students had learned it, and what was once cutting edge science had faded into common wisdom. If it is the latter, there is hope for all of us.

[rick jemmett]

I read Rob’s post with interest as it touches on a topic that has become of great interest to me over the past few years. ‘Clinical reasoning’ was not a term I heard while in school in the early 90’s and didn’t come across until about 2000. Since then I have found a good amount of material on the topic, most of which exists in the medical education literature but some also from Mark Jones and Darren Rivett (Australia) as well as some very interesting thoughts from Olli Miettinen (Canada).

As Rob described, it turns out that in getting through the average day, humans function largely through a process of pattern recognition. In the midst of the enormous volume of data bombarding us at any given moment, we learn to recognize relevant relationships within and amongst things in order to function. We learn to recognize the salient features of ‘things’ and how not to become distracted by details and irrelevant differences. As Mark Jones and others have pointed out, in almost all walks of life, expert performance is characterized not so much by high levels of skill as it is by advanced pattern recognition. Experts in any field are able to accurately and quickly identify patterns while novices stumble through the process of understanding the problem at hand.

This stuff became of great interest to me when I realized that the physical therapy culture emphasizes a form of clinical reasoning equated not with expert performance, but with novice performance. Novices, as described in the medical education literature, engage in fairy stereotyped data collection exercises where the subjective and physical exams are used to gather as much information as possible. This is the ‘ask all questions and perform all known tests’ type of assessment that is still emphasized throughout much of the physical therapy world. These ‘retrospective’ exams generate large volumes of data, much of which is irrelevant to the specific patient. Further, they tend to do a poor job of identifying important tell-tale relationships amongst certain findings – the salient features of clinical or diagnostic patterns. The end result is usually a problem list rather than a deeper appreciation for the underlying issue at hand.

Alternatively, we have the ‘prospective’ examination. Here the emphasis is on searching out diagnostic possibilities from the get go. At the first sighting of the patient in the waiting room the clinician begins forming a differential diagnosis. The subjective examination is used to further form and challenge these diagnostic possibilities until such time as the clinician can state that they have two, three or four possible explanations for the patient’s symptoms. This list – the differential diagnosis - is rank ordered from most likely to least likely (docs use the term ‘index of suspicion’). The physical exam is then used to confirm this differential diagnosis. This requires much due diligence as the various items on the list are confirmed and negated. The physical exam must find reasonable confirming evidence for the most likely diagnosis as well as negating evidence of the other less likely diagnoses. At the heart of prospective clinical reasoning is, as Rob put it, ‘informed pattern recognition’. The clinician draws upon their broad knowledge and experiential base (within which reside their own clinical patterns) as the examination is in progress.

Interesting is the fact that retrospective reasoning is so culturally engrained in the PT world, while it is well known in all other realms of human experience as being characteristic of novice rather than expert functionality. One barrier to the widespread application of prospective reasoning in PT then is this cultural bias. Another barrier, one which a great many clinicians and researchers are actively working on removing, is an absence of valid clinical patterns. Pattern recognition works in the real world (ie, it is a valid and reliable means of functioning) because the patterns are indeed accurate and truly representative. We learn at a young age that a black station wagon and a green hatchback are indeed both ‘cars’. We learn that maple, palm and fir are indeed all forms of trees, and that these are fundamentally different from bushes. However, in orthopedics, the patterns that currently exist are not so valid. Try getting five clinicians in a room who can agree upon the salient features of frozen shoulder, tennis elbow and non-specific low back pain. So we currently attempt to work in a non-evidence based world where dogma and the opinion of guru’s is all that we have. Even efforts such as the recent ‘European Guidelines’ are little more than the somewhat arbitrary collective opinion of a group of potential experts. Until such time as we have a definitive understanding of the mechanisms underlying these neuromusculoskeletal problems, we will be unable to employ prospective reasoning to its full potential.

If there were any better reason to embrace clinically relevant research, to ‘buy into’ the need for truly evidence-based approaches to health care, I am not aware of it. Rob’s example of the Cook County experience in the specific setting of 'chest pain' is a great example. The world of clinical orthopedics needs to follow suit.

Rick Jemmett