This Rule Doesn’t Rule

Abstract & Commentary

Synopsis: A clinical prediction rule helps identify patients with knee pain who might benefit from orthopedic surgery.

Source: Solomon DH, et al. Arch Intern Med. 2004;164:509-513.

Solomon and colleagues set out to devise a clinical prediction rule to identify patients with knee pain who would benefit from orthopedic surgery other than arthroplasty. A research assistant examined patients presenting for the first time to an academic orthopedic surgery department. The patients had been referred by primary care physicians or rheumatologists or were self-referred. The examinations consisted of a structured history and physical and an adaptation of the Western Ontario and McMaster Universities Osteoarthritis (WOMAC) Index that rated pain and disability on 1-to-5 scales. A second evaluation by an orthopedic surgeon established the likelihood that the patient would benefit from knee surgery (excluding knee replacement). The likelihood was recorded on a 100-mm visual analog scale with 0 representing "definitely not likely to benefit," 50 representing "unsure," and 100 representing "definitely likely to benefit." After 6 months Solomon et al contacted the patients and completed a follow-up questionnaire that addressed knee pain and what therapy had been prescribed. Twenty percent (20%) of patients were lost to follow-up; their office and hospital records were reviewed to identify what, if any, knee surgery they may have had.

Of 174 eligible patients, 103 were recruited. They were an average age of 48 years, 56% female, and had knee pain for an average of 7 months (range 3-24 months). Forty-nine percent (49%) had preceding knee trauma. Locking or catching, buckling or giving away, and knee popping sounds were the most common history. Knee effusion, limited flexion (< 120º of passive flexion), limited extension (unable to fully straighten leg), joint line tenderness, positive McMurray test, and anterior cruciate ligament (ACL) laxity were the most common physical findings. The most common diagnoses were degenerative arthritis, meniscal lesion (medial or lateral), patellofemoral syndrome, and ACL lesion, comprising 88% of all diagnoses. The patients’ average WOMAC scores for pain and disability were both 2.4.

Several cut points of the orthopedists’ scores were tested for sensitivity and specificity. A score of 75 gave the best combination (71% sensitivity, 77% specificity).

In unadjusted analysis, a history of sports-related trauma, disability score < 2.4, positive knee effusion, limited knee flexion or extension, positive McMurray test, joint line tenderness, and ACL laxity were all statistically significant. Knee effusion dropped out in multivariate analysis.

The adjusted odds ratios of the 6 remaining variables were calculated. ACL laxity had the highest odds ratio (52.7), and its presence almost always identified a patient who would benefit from surgery. The other 5 variables were assigned whole number points that represent rounding of their odds ratios. Sports-related trauma, limited flexion or extension, joint line tenderness, and positive McMurray test all received 2 points; disability score < 2.4 received 4. All patients were then scored and placed into 3 groups, low- (= 3), medium- (4-8), and high-risk (=9) for benefiting from surgery. Of those patients who eventually underwent surgery, 13% were in the low-risk group, 13% in the medium-risk group, and 53% in the high-risk group.

Comment by Allan J. Wilke, MD

I like clinical prediction rules; they can be very useful in guiding clinical decision-making. The Ottawa Knee Rule comes immediately to mind.1 This rule does not join that esteemed company—yet. First this rule must be tested prospectively in a primary care setting. The site of the study was an academic orthopedic department. The patients who find their way there are not representative of patients you and I see. My second concern is in the design of the rule. One of its variables, disability score < 2.4, is itself based on WOMAC, a scale that is not readily available in most primary care offices. My other concerns lie in the design of the study. The research assistant who performed the physical examinations had 5 days of training in knee anatomy and examination. Between medical school, residency, and practice, most primary care physicians spend much more time learning to do a good knee exam. The rule tries to predict which patients orthopedic surgeons predict will need their services.

While I expect that these surgeons made their decisions based on their clinical acumen and not their pocketbooks, why not eliminate the middleman and use surgery, and not the estimation of surgery, as the outcome of interest? My final concern is in the rule’s utility. Presumably, I would use it to determine who needs an orthopedic referral. If a patient scores in the high-risk group, the odds of that patient receiving surgery are about 50-50. This isn’t so bad, and arguably reasonable, provided that the orthopedic surgeon’s consultation fee isn’t exorbitant. On the other hand, if a patient scored in the low-risk group, I wouldn’t be sending him/her to ortho, and I would be wrong a sixth of the time. I have no idea what to do with those folks in the middle-risk group. The problem lies in the rule’s relatively poor sensitivity and specificity. Really useful tools have sensitivity and specificity > 90%. In a recent systematic review the Ottawa Knee Rules had 98.5% sensitivity and 48.6% specificity.2

Dr. Wilke, Assistant Professor of Family Medicine, Medical College of Ohio, Toledo, OH, is Associate Editor of Internal Medicine Alert.


1. Stiell IG, et al. JAMA. 1996;275:611-615.

2. Bachmann LM, et al. Ann Intern Med. 2004;140:121-124.