The surgeons marked on a visual analog scale—where 0 indicated "definitely not likely to benefit"; 50 indicated "unsure"; and 100 indicated "definitely likely to benefit"—their confidence that knee surgery would benefit each of their patients. Patients' ratings were converted to a numerical response based on the mean ± SD measurement on the visual analog scale (mean, 45 ± 33; median, 50; interquartile range, 11-76).
Clinical prediction scoring system to determine which patients are likely to be juged to benefit from nonarthroplasty knee surgery. The likelihood to benefit from surgery was determined from the orthopedic surgeon's rating of each study patient's knee condition immediately after initial consultation. The scoring system is based on the adjusted relative risk of each patient characteristic. *For patients with anterior cruciate ligament laxity, 52 points would have been added based on the adjusted relative risk. Since it was such a strong predictor, it was left out of the 12-point system.
Relationship between clinical prediction score and likelihood to benefit from surgery. Three referral score categories were determined. The proportion of patients juged likely to benefit from surgery was 8% in the low-score category and 84% in the high-score category (P for trend = .001). Patients with anterior cruciate ligament laxity were included in the high-score group.
Solomon DH, Avorn J, Warsi A, Brown CH, Martin S, Martin TL, Wright J, Burgener M, Katz JN. Which Patients With Knee Problems Are Likely to Benefit From Nonarthroplasty Surgery?Development of a Clinical Prediction Rule. Arch Intern Med. 2004;164(5):509-513. doi:10.1001/archinte.164.5.509
We examined factors associated with the clinical presentation of patients judged likely by orthopedic surgeons to benefit from nonarthroplasty knee surgery.
Consecutive patients presenting to orthopedic surgeons were eligible for the study and 103 were recruited. A trained research assistant took histories, performed physical examinations, and administered a standardized questionnaire. Surgeons provided diagnoses and, in each case, rated their confidence that nonarthroplasty knee surgery would help the patient. We assessed the relationship between patient characteristics and the orthopedic surgeon's level of confidence that the patient was likely to benefit from knee surgery.
In multivariable logistic regression models, the following characteristics were associated with a surgeon's judgment that a patient would likely benefit from knee surgery: a history of sports-related trauma, low functional status, limited knee flexion or extension, medial or lateral knee joint line tenderness, a click or pain noted with the McMurray test, and a positive Lachmann or anterior drawer test (c statistic from model, 0.83). These items were combined into a clinical prediction score, and low-, medium-, and high-risk categories were identified. Independent evaluation by surgeons indicated that only 8% of patients in the low-risk category but 84% of patients in the high-risk category were judged likely to benefit from surgery (P for trend <.001).
Using a small group of easily accessible patient characteristics, we developed a clinical prediction score that clearly differentiated patients who were viewed by experienced orthopedic knee surgeons as likely or not likely to benefit from nonarthroplasty knee surgery.
Many patients with knee pain present initially to a primary care physician. For some of these patients, management includes referral to an orthopedic surgeon. Orthopedic surgeons provide diagnosis and, although they propose nonsurgical management of patients with knee pain, consideration of surgery is an important aspect of the surgical consultation. Little information, however, is available to assist the primary care physician in identifying patients likely to benefit from nonarthroplasty knee surgery.
Orthopedic surgery textbooks describe a complex decision-making process with respect to knee surgery that involves an interplay of the patient's diagnosis, prior and current functional status, selected symptoms, and social factors.1 While primary care physicians may not always be able to arrive at a diagnosis for a patient's knee pain, they should be able to take a careful history and perform a physical examination as well as assess the level of pain and function. We conducted a prospective study to examine the association between items that a primary care physician could assess and the orthopedic surgeon's likelihood of recommending nonarthroplasty knee surgery. These items were then tested in aggregate as a clinical prediction rule.
Eligible patients were those scheduled for a first consultation with any one of 5 orthopedic surgeons specializing in lower extremity disorders. Patients were referred by primary care physicians and rheumatologists, or they were self-referred. All surgeons were affiliated with Brigham and Women's Hospital, Boston, Mass. We recruited patients who understood spoken and written English while they were waiting to be seen by the orthopedic surgeon. The study was approved by the appropriate institutional review board.
Structured histories were taken and physical examinations conducted by a trained research assistant (the assessment tool is available upon request from the authors). The research assistant's training consisted of approximately 5 days of studying knee anatomy and examining patients with knee pain with experienced orthopedic surgeons and a rheumatologist. The physical examination was conducted by the research assistant using standard techniques and criteria for test results.2 The research assistant also administered an adaptation of the Western Ontario and McMaster Universities Osteoarthritis (WOMAC) Index, which measures pain and functional status on scales of 1 to 5 for persons with knee conditions.3,4 After these evaluations, the patient was seen by the orthopedic surgeon who also completed a brief assessment of the patient's likelihood of benefiting from nonarthroplasty knee surgery on the menisci or ligaments and indicated the likelihood of various diagnostic possibilities.
Approximately 6 months later, we attempted to contact each patient by mail or telephone to complete a follow-up questionnaire. The questionnaire consisted of the adaptation of the WOMAC Index for knee pain as well as questions about the management of the patient's knee problem. We reviewed the hospital and physician's records of the patients whom we were unable to contact (20%) to determine whether they underwent nonarthroplasty knee surgery on the menisci or ligaments.
We compared the age and sex of the patients we recruited with the age and sex of those we did not recruit, and we created a dichotomous variable from the orthopedic surgeons' ratings of the patients' likelihood of benefiting from nonarthroplasty knee surgery. Surgeons were asked to "mark on the line how confident [they were] that surgery for an internal derangement (arthroscropic or open) would benefit this patient." The surgeons rated the likelihood of benefit on a 100-mm visual analog scale (VAS) where 0 was labeled "definitely not likely to benefit," 50 was labeled "unsure," and 100 was labeled "definitely likely to benefit." Each rating was converted to a numerical response based on the VAS measurement and several cut points were tested. Patients with VAS measurements at or above the given cut point were considered to be likely to benefit from nonarthroplasty knee surgery. The sensitivity and specificity of these cut points were calculated by comparing the patients' actual surgical status at 6 months as the reference standard with the predicted status based on the VAS cut point.5 Then, we determined the optimal cut point by comparing the sensitivity (or proportion of patients who had surgery and who were predicted to have surgery) and specificity (or proportion of patients who did not have surgery but who were predicted to have surgery) of each potential cut point. We also performed analyses using the continuous VAS measurement. The same variables remained in the multivariable model.
The dichotomous end point of a surgeon's judgment that a patient was or was not likely to benefit from surgery was used as the dependent variable in logistic models. The association between the dependent variable and patient characteristics such as historical items, functional status, pain index, and physical examination were examined. Predictors with P values less than .2 were tested in multivariable models. To reduce the number of variables in the final model, we created several composite clinical examination items. For example, limited knee flexion and limited knee extension were combined into a single item.
Patient characteristics with P values less than .05 or adjusted odds ratios of 2 or greater were included in the final multivariable model. The odds ratios from the multivariable model were then used as a weighting ("scoring") system to create a clinical prediction score. To simplify the scoring, weights were assigned as follows: adjusted odds ratios were assigned 2 points when they were between 2.0 and 2.9, 3 points when they were between 3.0 and 3.9, and so on. Based on these weights, a score was calculated for each patient. We aggregated patient scores into low-risk (0-3 points), medium-risk (4-7 points), and high-risk (8-12 points) categories, and then calculated the proportion of patients in each category deemed likely by the surgeon to benefit from knee surgery. Alternative groupings (0-4 points, 5-8 points, and 9-12 points) were tested and the results were similar; these results are not presented. Finally, based on self-report and a review of the medical records, we calculated the proportion of patients in each risk category who actually underwent nonarthroplasty knee surgery.
We recruited 103 (59%) of 174 eligible patients. The eligible patients who were not recruited either refused or were seen by an orthopedic surgeon before the research assistant could recruit them. With respect to age (48 vs 45 years; P = .3) or sex (56% vs 53% women; P = .7), there were no statistically significant differences between the study patients and those who did not consent to participate.
The baseline characteristics of the 103 study patients are displayed in Table 1. The mean ± SD patient age was 48 ± 17 years, a majority of patients were women, and the median duration of knee pain was 7 months (interquartile range, 3-24 months). Most patients had unilateral complaints, and almost half of them reported some trauma prior to the knee pain. Approximately half of the traumas were sports related, and the remainder resulted from a blunt injury, a fall, or a twist. Almost half of the patients had used a cane, crutch, brace, or elastic wrap; 22% reported prior surgery on the index knee; and a similar proportion had tried physical therapy or home exercises.
The distribution of diagnoses given by the orthopedic surgeon are reported in Table 2. The most frequent diagnoses were degenerative arthritis (25%), medial meniscal lesions (24%), patellofemoral syndrome (25%), anterior cruciate ligament (ACL) lesions (7%), and lateral meniscal lesions (7%). The findings from the standardized questionnaire and the physical examination performed by the research assistant are reported in Table 3. With respect to physical history items, about one third of patients reported locking or catching, one third reported buckling or giving way, and one third described a popping sound around the knee. The mean scores on the pain and functional status subscales suggested a mildly to moderately affected group of patients. Relatively common physical examination findings included a knee effusion, limitation in flexion, limitation in extension, joint line tenderness, pain or a click on the McMurray test, and ACL laxity demonstrated by a Lachmann test or an anterior drawer test. Limited flexion was defined as less than 120° of passive flexion and limited extension was defined as the inability to fully straighten the leg.
The orthopedic surgeons' ratings are plotted in Figure 1, which shows a bimodal distribution with a peak at less than 10 and another around 75. Table 4 demonstrates sensitivity and specificity at several potential cut points. A cut point of 75 or higher produced a high sensitivity and high specificity, and it demarcated the upper peak. Based on these considerations the cut point of 75 was used as the threshold for dichotomizing the ratings plotted in Figure 1. Thirty-four patients (33%) were deemed likely to benefit from surgery using this cut point.
The dichotomous assessment of likelihood to benefit from knee surgery was then used as the dependent variable in logistic analyses shown in Table 5. In the unadjusted (crude) analyses, factors that had P values less than .2 were considered for multivariable analyses. These factors included a history of sports-related trauma, disability scores less than than 2.4 (the median) on the functional status scale, the presence of a knee effusion, limited knee flexion or extension, a click or pain on the McMurray test, medial or lateral joint line tenderness, and the presence of ACL laxity on an anterior drawer or Lachmann test. Other than sports-related trauma, no patient demographic or historical findings were significant predictors in crude bivariate analysis. With the exception of the presence of a knee effusion, all the variables tested in multivariable models had adjusted odd ratios of 2.0 or greater and were included in the clinical prediction score.
Figure 2 illustrates the scoring system that was based on the adjusted odds ratio. For patients with ACL laxity, 53 points would have been added based on the adjusted odd ratio of 52.7. Since this was such a strong predictor of who would be deemed likely to benefit from nonarthroplasty knee surgery, and 8 of 9 patients with ACL laxity were judged likely to benefit from surgery, it was left out of the 12-point scoring system. All patients with ACL laxity were given the maximum score. We then calculated a clinical prediction score for each patient and grouped these into low-, medium-, and high-risk categories. The proportion of patients in each category who were deemed likely to benefit from surgery is plotted in Figure 3. It illustrates that 8% of patients in the low-risk category, vs 84% of patients in the high-risk category, were judged by their orthopedic surgeon likely to benefit from knee surgery (P for trend <.001). In secondary analyses, we assessed the proportion of patients in each risk category who actually underwent nonarthroplasty knee surgery within 6 months of the initial consultation. Five (13%) of the 38 patients in the low-risk group, 6 (13%) of the 46 patients in the medium-risk group, and 10 (53%) of the 19 patients in the high-risk group underwent surgery (P = .006).
We sought to determine whether easily definable characteristics could identify patients whom orthopedic surgeons would judge likely to benefit from nonarthroplasty knee surgery. A few items—a history of sports-related trauma, poor functional status, limited flexion or extension, medial or lateral joint line tenderness, pain or a click on the McMurray test, and evidence of ACL laxity—were found to be highly associated with that judgment. When these factors were weighted by their adjusted odds ratio and combined into a clinical prediction score, the score clearly discriminated between patients judged likely or not likely to to benefit from knee surgery. Further, patients in the high-risk category were much more likely to have undergone nonarthroplasty knee surgery within 6 months of the surgical consultation.
We also sought to give primary care physicians a tool that would be useful in assessing which patients with knee pain might be good candidates for surgery on their menisci or ligaments. Since such tools can translate results from complex analyses into simple practical algorithms, and this approach has been useful for other orthopedic disorders,6 we aggregated the obtained data into a clinical prediction score. Other investigators have considered developing similar multivariable scores for unstable meniscal lesions in patients with degenerative joint disease of the knee.7 While the decision to operate is not the sole contribution made by skilled orthopedic surgeons, such a tool might be useful for the primary care physician when determining which patients should be considered for referral to an orthopedic surgeon.
The patient characteristics that form the basis of the clinical prediction score were obtained by a trained research assistant who was not a health care professional and who had no clinical experience, which suggests that primary care physicians could obtain the same information readily. The decision to undertake surgery should not be based on this score, but the decision to refer a patient to an orthopedic surgeon might be aided by such a tool. Additionally, the physical examination items found to be the most important might form the basis for future musculoskeletal education programs. Currently, many physical examination maneuvers are demonstrated in educational CD-ROMs and Internet-based programs.8
These findings must be interpreted within the limits of the method. The small size of the study sample limits the precision of our estimates; since the study was based at a referral center, all patients had already been referred to an orthopedic surgeon and this may limit its generalizability; and the research assistant may have incorrectly assessed patients for specific items and thereby introduced misclassification. The assistant, however, was rigorously trained to conduct the physical examination and had observed several clinicians take the histories of many patients with knee complaints. Moreover, such possible misclassification would tend to weaken the model rather than cause any specific bias. In any case, it will be important to test this clinical prediction score in other populations before it can be translated into a validated rule. The likelihood of benefiting from surgery was judged by 5 academic orthopedic surgeons and may not represent the opinion of all surgeons. However, the absence of a universally agreed-upon criterion for which patients should undergo meniscal or ligamentous knee surgery makes the opinion of academically trained orthopedic surgeons a valuable standard.
In conclusion, we have developed a clinical prediction score that is strongly associated with patients judged likely by experienced orthopedic surgeons to benefit from nonarthroplasty knee surgery. The score consists of easily accessible patient information and fulfills many of the methodological standards described by McGinn and colleagues in their Users' Guide to Clinical Decision Rules.9 If the score is considered valid after testing in other settings, it may help primary care physicians in deciding which patients to refer for consideration of orthopedic knee surgery.
Corresponding author: Daniel H. Solomon, MD, MPH, Division of Pharmacoepidemiology, 1620 Tremont St, Suite 3030, Boston, MA 02120 (e-mail: firstname.lastname@example.org).
Accepted for publication November 7, 2002.
This work was supported by a grant from the Arthritis Foundation, Atlanta, Ga, and grants K23 AR48616, K24 AR02123, and P60 AR47782 from the National Institutes of Health, Bethesda, Md.