The United States Navy had a problem. It was the early 1970’s and the height of the Cold War.
US Navy submarines were playing cat-and-mouse submerged reconnaissance with Soviet submarines in every ocean around the world. Nuclear submarines ran quieter, faster and longer than older, diesel subs and could submerge to new, record-setting depths. US Navy submariners were among the best trained, most highly motivated military men in the world.
Submariners, however, still got chest pain at the same rate as ordinary civilians.
Navy doctors had to decide, based on clinical findings, if the submariner’s signs and symptoms were serious enough to consider aborting the mission and seeking a friendly port.(Blink - Malcolm Gladwell)
Underwater heart attacks didn’t fare well outside of a hospitals’ intensive care unit.
That’s where Dr. Lee Goldman came in. Dr. Goldman was studying statistical rules – algorithms – that predicted when people were having a heart attack.
Dr. Goldman’s rules predicted the occurrence of a major cardiac event based on three predictor variables:
1. Is the patients’ pain unstable angina?
2. Do you hear rales above the base? (indicates fluid in the lungs)
3. Is the systolic blood pressure below 100mm Hg?
Combinations of these
predictor variables indicated different treatment options:
1. surface and give away your submarine’s position to the enemy but save your submariner’s life
2. sit tight and monitor your submariner’s vital signs or
3. send your submariner back to work with a bottle of Pepto-Bismol.
Navy doctors studied these treatment algorithms and used them in the care of their sick submariners. At one point, military physical therapists led their civilian cousins using evidence-based medicine in decision making, ordering radiographs and making referrals to other health care professionals.
While the US military lead the way in the early 1970’s in using clinical prediction rules the American health care community responded to Dr. Goldman’s work with deafening silence. (Gladwell)
American Doctors Make a DecisionIt wasn’t until 1995 that American doctors began to use decision rules to inform the care of their patients. The best example on record comes from Cook County Hospital in Chicago. (Gladwell)
This 700-bed urban teaching hospital is a century-old, publicly-funded institution that was seeing thirty new chest pain patients per day in its emergency room and
79% of them were getting a full work-up for chest pain.
Patients were admitted to one of two wards for hospitalized chest pain patients:
• eight coronary intensive care beds or
• twelve telemetry-monitored coronary beds.
The coronary intensive beds cost $2,000 per bed per day and the telemetry-monitored beds cost $1,000 per bed per day.
Ironically, only
5-10% of the patients admitted to the hospital suspected of having a heart attack progressed to a full-blown heart attack. The hospital’s problem was that they were spending expensive resources on patients who were not having a heart attack.
The hospital’s chairman of the Department of Medicine, Dr. Brendan Reilly, wasn’t worried about the quality of care – the quality was good. Dr. Reilly was worried about the cost of providing cardiac care to patients who weren’t having a heart attack. He began studying the decision-making processes used by the emergency room doctors caring for patients with chest pain.
Ironically, the initial response from the ER doctors was reluctance and resistance – how can Dr. Goldman’s algorithm allocate intensive care bed space better than ER doctors’ decisions? What about family history? What about weight, sex, race, smoking history, stress and many other factors considered important at the time in the diagnosis of heart attack?
What Dr. Reilly found out was that race, gender and lifestyle factors were less important than whether or not the doctors followed the algorithm. Not that these factors were unimportant in the overall care of the patient – just that the initial decision to allocate the expensive, intensive care bed was better made by adhering to the algorithm, not to the host of factors that, while important, were incidental to the initial decision.
Dr. Reilly studied the impact of using the Dr. Goldman’s CPR in the Cook County ER. He found that the efficiency, the rate at which patients not having a heart attack were sent to inexpensive observation or sent home, went from
21% to 36%. Dr. Reilly also found that safety, the rate at which patients having a heart attack were triaged to coronary intensive care, went from 89% to 94%.
Just as the submariner’s doctor had to make a quick, initial decision that balanced the risk of giving away the submarine’s position with the risk that the submariner would progress into a full-blown heart attack so too did the Cook County ER doctor have to make a decision that balanced the risk that Cook County would spend $2,000 per night for up to three nights on a patient with acid indigestion versus the risk that the patient was having a heart attack.
Classifying submariners was a clinical 'shortcut' that enabled the submarine to stay submerged in those cases that were not clearly a major event. Classifying chest pain patients in Cook County was a clinical 'shortcut' that prevented spending thousands of dollars on people with tummy gas.
Classification as a Resource Allocation ToolBoth the submarine and the cardiac beds examples treat classification as the solution to a resource allocation problem. Both scenarios were prompted by crises of scarcity. Dr. Reilly at Cook County finding fewer public funds to pay for critical care cardiac beds as emergency room admissions rose and the US Navy facing a trade-off between dying submariners and national security.
American health care is facing its own crisis of scarcity as rising rates of per-capita health care consumption, the tidal wave of aging baby boomers and budget constraints on increased health care spending impose resource allocation challenges on increasingly scarce physical therapy resources, like time and money.
Classification, however, is not the appropriate tool for every clinical decision faced by physical therapists. As noted, classification is probably appropriate only for the
initial treatment assignment and may not describe the exact treatment to be used. For example, the spinal traction classification is useful in cases of non-centralizing leg pain of radicular origin but the decision rule does not give information as to the parameters of spinal traction: force, total time, ramp time, or patient position.
Classification is probably most useful when one or more discrete alternative treatment possibilities exist, eg: lumbar manipulation or stabilization. Presumably, not both. Classification is probably not helpful in straightforward PT decision-making such as an
uncomplicated ankle sprain. There needs to be some risk that making the wrong choice will produce worse outcomes or a less efficient clinical process.
For example, if the Navy doctor incorrectly diagnoses a heart attack and the submarine captain decides to surface en route to a friendly seaport it reveals its position to enemy radar and US national security could be compromised.
The submarine and cardiac beds examples offer illustrations of risk that are far more clear-cut than physical therapists would typically encounter in the clinic. It seems obvious that clinical prediction rules developed by Dr. Goldman and others were utilized earlier in these environments because of increased risk and greater costs involved.
Classification as DiagnosisThe physical therapy profession is currently shifting towards Treatment Based Classification (TBC) using clinical prediction rules (CPR) for diagnostic and treatment decision-making.
Unlike the physician profession, the physical therapy community seems almost uniform in its acceptance and embrace of classification measures as an aid to clinical decision-making. (Gladwell, Groopman in How Doctors Think)
An understanding of probability is required to fully understand the use of statistically-derived predictor variables. For example, the Fear-Avoidance Beliefs Questionnaire (FABQ) is a predictor variable for the
manipulation classification while plausible findings like pelvic landmarks and sacroiliac region pain are not predictor variables. How can this be?
The
derivation studies that identified the original predictor variables tossed out biologically plausible tests and measures instead showing us the true predictors of patients likely to respond to lumbar spinal manipulation.
Not leg length inequality, not mechanism of onset, not MRI or x-ray findings, not pelvic landmarks or pelvic movement tests. Instead, some surprising findings turned out to show physical therapists who should be manipulated:
1. Time since onset (> 2 weeks)
2. Extent of distal leg pain (not past the knee)
3. Lumbar hypomobility
4. FABQ work sub-scale >19 points
5. No hip ROM asymmetry
If, on average, manipulating your patients is a coin flip (about 50% get better, 50% don’t get better), then application of the CPR improves your chances to 68% for patient who have any 3/5 of the predictor variables. Your chances improve to 95% if the patient has just one more of the predictor variables.
Classification as ProbabilityProbabilistic decision-making is consistent with the hypothetico-deductive model that is associated with physician decision-making, prescriptive medicine and the patient’s role emphasizing ‘compliance’ over ‘collaboration’. As such, classification seems to shift traditional physical therapist decision making way from its ‘collaborative’ roots.
Will this shift threaten the intimacy that physical therapists have come to treasure with our patients?
Is intimacy sacrificed when decisions are made quickly?
Will physical therapists continue to consider patient-centered factors such as culture, social class, age, experiences and goals when applying clinical prediction rules? Just like the Cook County ER doctors who felt that the chest pain CPR ignored too many important factors in the ongoing care of their patients so too can TBC ignore important aspects that impact the ultimate physical therapy outcome.
Will CPRs allow therapists to quickly deliver routine aspects of care that are best made by statistics, like initial group allocation? Then physical therapists can focus on face-to-face interactions that engage patients’ emotional involvement in their own care.
Classification SuccessNothing succeeds like success and classification has succeeded in capturing the imaginations of educators, researchers and clinicians within physical therapy because of clinical successes and because of several well-designed studies published in
prestigious medical journals.
Classification of spinal pain patients has crystallized an incoherent field of data into five or fewer examination findings per group. Classification has revolutionized physical therapy education and empowered students and experienced clinicians to become better decision-makers.
Questions remain:
1. Can classification change physical therapist behavior?
2. Can classification change physical therapy outcomes?
3. Are classification groups mutually exclusive and exhaustive? 75
4. Are some manipulation patients also candidates for stabilization?
5. Can some findings be treated that are not measured by classification predictor variables?
6. Can one patient fit the criteria for more than one diagnostic label?
Is classification good for documentation?Aside from the risk that classification will change the interaction of patient and physical therapist to a less intimate relationship that is more typical to that of patient and physician I have concerns that classification will be used as a panacea for documentation; the clinical ‘shortcut’ will become a note-writing shortcut that leaves the physical therapist exposed to a Medicare audit because she has not adequately expressed her skilled decision in writing at every follow-up visit.