Accuracy of the Thessaly Test when used by primary care clinicians

When the study tests were performed by primary care clinicians, significant differences were found between the tests with respect to sensitivity and specificity of the tests, but not with NPVs and PPVs (Table 8).

TABLE 8

Accuracy of physical test and recorded clinical history vs. MRI for diagnosis of meniscal tears by primary care clinicians for subjects with knee pathology (95% CI)

The most sensitive test was the joint line tenderness Test [0.77 (95% CI 0.68 to 0.84)]; however, this test also had the lowest specificity in the hands of primary care clinicians [0.26 (95% CI 0.18 to 0.36)]. Figure 9 shows a likelihood ratio graph for each test with the Thessaly Test set as the reference line (solid black). The graph shows slight superiority for McMurray’s Test over the Thessaly Test when used by primary care clinicians. No single test was identified as being an appropriate surrogate for MRI diagnosis of meniscal tears by primary care clinicians (see Table 9). All tests including clinical history had a diagnostic accuracy of 55% or lower (Table 9).

FIGURE 9

Likelihood ratio graph, primary care clinicians vs. MRI (knee pathology patients only).

TABLE 9

Implications of primary care clinicians using the Thessaly Test in routine practice as an assessment screening tool for meniscal tears to differentiate those patients with knee pathology who need a MRI scan/referral to secondary care for meniscal tears (more...)

For guidance on the interpretation of likelihood ratio graphs, see Chapter 2, Guide to likelihood ratio graphs.

We examined the influence of subject characteristics (age, sex and BMI), arthritis, presence of knee injury, ACL injury on the sensitivity and specificity of the Thessaly Test as performed by primary care clinicians. When examining the entire cohort, only the presence of a knee injury influenced sensitivity of the Thessaly Test (Table 10). Specificity was influenced by both the presence of an injury and the report of a sports injury. These results are intuitive and fit with the accepted pathogenesis of meniscal tear, at least in younger patients.

TABLE 10

Influence of subject characteristics on sensitivity and specificity of the Thessaly Test as performed by primary care clinicians: ORs (with 95% CI)

The STARD diagram for the Thessaly Test when carried out by primary care clinicians (see Appendix 2) shows that in 35 patients the clinician was unable to perform the Thessaly Test. This was due to pain in the knee at the time of the examination. In order to ensure that this effect has not biased the overall results we have compared the rate of meniscal tears diagnosed by MRI for the group of patients in whom the test was unable to be performed with the remaining study cohort. Eight of the 35 patients did not attend for a MRI scan and we therefore have no knowledge of their meniscal status. The results for the remaining 27 patients are detailed in Table 11. A statistical comparison of the two groups of patients, using Fisher’s exact test, revealed no significant difference between the two groups (p = 0.224).

TABLE 11

Comparison of rate of meniscal tears diagnosed by MRI between patients where the Thessaly Test was not performed and was performed

Accuracy of the Thessaly Test when used by musculoskeletal specialists

When the study tests were performed by specialist musculoskeletal clinicians, significant differences were found between the tests with respect to sensitivity and specificity and NPV of the tests, but not PPVs (Table 12).

TABLE 12

Accuracy of physical tests and recorded clinical history vs. MRI for diagnosis of meniscal tears by musculoskeletal clinicians for knee pathology subjects only (95% CI)

The most sensitive test was the joint line tenderness Test [0.83 (95% CI 0.75 to 0.89)]; however, this test also had the lowest specificity [0.39 (95% CI 0.29 to 0.49)]. The test with the highest sensitivity was Apley’s Test [0.72 (95% CI 0.61 to 0.81)], but Apley’s Test had low specificity [0.43 (95% CI 0.34 to 0.52)]. Figure 10 shows a likelihood ratio graph for each test with the Thessaly Test set as the reference line (solid black). The graph shows slight superiority for both clinical history and McMurray’s Test over the Thessaly Test when used by musculoskeletal clinicians.

FIGURE 10

Likelihood ratio graph, musculoskeletal clinicians vs. MRI (knee pathology patients only).

No diagnostic physical test was better at diagnosing a meniscal tear than a well-trained musculoskeletal clinician taking a clinical history (Figure 10 and Table 13). McMurray’s Test was a better overall test than the Thessaly Test with a better compromise of sensitivity and specificity (see Figure 10).

TABLE 13

Implications of musculoskeletal clinicians using the Thessaly Test, McMurray’s Test and clinical history in routine practice as an assessment screening tool for meniscal tears to differentiate those who need a MRI scan

There were only subtle differences in the ability of each test to diagnose medial and lateral tears (Table 14).

TABLE 14

Comparison of diagnostic accuracy for medial and lateral meniscal tears (95% CI)

Interestingly, the specificity and NPV was greater for all tests in the control group when analysed separately (Table 15). This indicates that the tests actually work well in differentiating a normal knee from a knee with a meniscal tear. However, all of the physical tests have limited ability to differentiate between a painful knee with a meniscal tear and painful knee due to any other cause. Table 16 lists the diagnosis and clinical findings in patients who were positive for the Thessaly Test, but did not have a meniscal tear on MRI scan. This demonstrates the range of common conditions that affect the knee which are poorly differentiated by both the Thessaly Test and the other physical tests assessed in this study.

TABLE 15

Specificity and NPV in control patients assessed by musculoskeletal clinicians (95% CI)

TABLE 16

Diagnosis and clinical findings in patients who had a positive Thessaly Test (when performed by a musculoskeletal clinician), but were negative for a meniscal tear by MRI

For guidance on the interpretation of likelihood ratio graphs, see Chapter 2, Guide to likelihood ratio graphs.

Comparison between primary care clinicians and specialist musculoskeletal clinicians using physical tests and clinical history to diagnose meniscal tears

Comparison between the two clinician groups (specialist musculoskeletal clinicians and primary care clinicians) showed significant differences in the results obtained using both the Thessaly Test and Apley’s Test to diagnose meniscal tears (Table 17). Better agreement was found with McMurray’s Test and the joint line tenderness Test. Note that this analysis only assesses whether or not the two clinicians agreed and if they were correct. Assessment of diagnostic accuracy shows that specialist musculoskeletal clinicians were consistently better at diagnosing meniscal tears based on both physical tests and on clinical history (Table 18).

TABLE 17

Comparison of musculoskeletal clinicians and primary care clinicians use of physical tests and clinical history to diagnose meniscal tears (McNemar’s Test)

TABLE 18

Comparison of diagnostic accuracy between clinician groups using the Thessaly Test, McMurray’s Test and clinical history

Influence of the presence of osteoarthritis and other patient factors on the accuracy of the Thessaly Test (and other physical tests)

We have examined the influence of the following patient characteristics on the sensitivity and specificity of all four physical tests in the hands of both musculoskeletal and primary care clinicians: sex, age, BMI, osteoarthritis, sports injury, knee injury, ACL tear and previous surgery on the affected knee.

The sensitivity of the Thessaly Test when performed by primary care clinicians was influenced by age and presence of sports injuries (Table 19). The sensitivity of Apley’s Test was influenced by age and other knee injuries. The specificity of McMurray’s Test was also influenced by age and other injuries.

TABLE 19

Influence of subject characteristics, arthritis and medical history on sensitivity of physical tests in knee pathology patients (p-value of logistic regressions)

The presence of an injury influenced the sensitivity of all tests performed by musculoskeletal clinicians. In addition, the joint line tenderness Test was also influenced by age (Table 20). No factors influenced specificity.

TABLE 20

Influence of subject characteristics, arthritis and medical history on specificity of physical tests (p-value of logistic regressions)

The presence of osteoarthritis of the knee did not significantly influence the sensitivity or specificity of any of the tests, performed by either group of clinicians.

Comparison of patient subgroup recruited directly from a general practitioner practice compared with patients recruited from within an orthopaedic department

Eleven patients were recruited directly from their GP (PD) in order to allow us to verify that the cohort of patients recruited through the orthopaedic department at Glasgow Royal Infirmary were representative of the overall target population (patients attending GP with knee pathologies). The patient groups were well matched for age and BMI (Table 21), with only minimal differences that were not of statistical clinical relevance. There was a lower proportion of females in the group recruited directly from the GP practice (18% vs. 37%), but this was not statistically significant (p = 0.338). The proportion of patients in each group that had a meniscal tear diagnosed by MRI was almost the same, 57% for the group recruited directly from the GP practice and 55% for the group recruited through the orthopaedic department (p = 1.0).

TABLE 21

Age and BMI comparison between patients recruited directly from a GP practice and those recruited through an orthopaedic department

Although the overall proportion of patients recruited directly from their GP was smaller than we had anticipated (4% vs. 10%), the group demographics and final diagnoses are very similar indicating that our main cohort is representative of the target population.

Do combinations of physical tests provide better specificity and sensitivity than a single physical test?

Analysis of individual tests for primary care clinicians is given in Table 22. Only McMurray’s Test, when performed by a primary care clinicians, was predictive of MRI outcome, no other test or the clinical history taken by a primary care clinician was a significant predictor. Nor was the order that the tests were performed in significant (see Table 22). In contrast, all physical tests and the clinical history taken by musculoskeletal clinicians were significant predictors of MRI (see Table 23). The order that the tests were performed in was not found to be significant (Table 23).

TABLE 22

Logistic regression: four models (one model per test) with the effect of the test and its order (p-values and AUC), for primary care clinicians

TABLE 23

Logistic regression: four models (one model per test) with the effect of the test and its order (p-values and AUC), for musculoskeletal clinicians

The simple analysis of individual test fails to take account of any potential covariates and therefore logistic regression was undertaken. We have used a logistic regression model including all relevant factors and their interactions with a stepwise procedure to remove unimportant variables. Logistic regression analysis including all four tests, the clinical history and relevant covariates (BMI, age, sex, osteoarthritis and previous surgery) reveal a different result to the individual analyses. When all relevant factors are taken into account the only factors for primary care clinicians that are predictive of a MRI result are age and a past history of osteoarthritis (Table 24). Although this may be beneficial in aiding diagnosis of meniscal tears in the elderly that are associated with degenerative disease of the knee, neither of these factors are helpful in diagnosing acute traumatic injuries in young individuals participating in sports. As degenerative tears tend to be associated with knee conditions such as osteoarthritis, few degenerative tears are ever treated. In contrast, treatment is regularly offered for young patients with acute traumatic or sporting injuries.

TABLE 24

Multivariable logistic regression with all factors included and stepwise logistic regression for MRI outcome (tests performed by primary care clinicians)

A variety of different clinicians of participated in this study and we had originally intended to include experience of the clinician undertaking the assessment within our models. However, classifying or grading clinician experience is not a simple process. It cannot simply be based on years since qualification as this takes no count of part-time working, maternity leave, training quality, or the number of appropriate cases each clinician has been previously exposed to. We were unable to produce a reliable method of classifying clinicians to include in our models. Instead, within the logistic regression models we included clinician as a random effect to take account of the variability between clinicians. However, this could not be specified with the stepwise regression models, where this factor was left unspecified and therefore included by default in the residual variability not accounted for by the model.

The results for musculoskeletal clinicians are potentially more clinically relevant than those observed for primary care clinicians (Table 25). In this clinician group the predicative factors are clinical history (history taken from the patient regarding recent symptoms and, if relevant, injury and mechanism of injury) and age.

TABLE 25

Multivariable logistic regression with all factors included and stepwise logistic regression for MRI outcome (tests performed by musculoskeletal clinicians)

The area under the curve (AUC) for the stepwise logistic regression (where clinician variability is accounted for in residual variability) is nearly the same as the logistic regression model with all the effects included and clinician specified as random effect (AUC 0.762 vs. 0.761), suggesting that there was nearly no difference between the different musculoskeletal clinicians. This was not the case for the primary care clinicians (AUC 0.722 vs. 0.651), suggesting a higher degree of variability between clinicians in this group.

Odds ratio for musculoskeletal clinicians diagnosis of meniscal tear based on taking a clinical history, and also for the influence of age, have been calculated and are presented in Table 26. A graphical representation of these probabilities for standard adult age range is shown in Figure 11.

TABLE 26

Odds ratio estimates for clinical history and age

FIGURE 11

Influence of age on the probability of a meniscal tear diagnosis by MRI when a musculoskeletal clinician diagnoses a meniscal tear based on a clinical history.

Validation of magnetic resonance imaging diagnosis for meniscal tears using knee arthroscopy

Seventy-seven patients had an arthroscopy as part of their routine clinical care. When arthroscopic diagnosis of meniscal tears is taken as the gold standard and compared with results obtained with MRI, the sensitivity of MRI was 1.0, whereas the specificity was 0.53 (Table 27).

TABLE 27

Magnetic resonance imaging vs. arthroscopy (95% CI)

The specificity of 0.53 (95% CI 0.28 to 0.77) is lower than reported in the literature for MRI diagnosis of meniscal tears. Only eight patients had a disagreement between their MRI diagnosis and their arthroscopic findings. In three patients the radiologist’s report of the MRI scan reported a ‘possible’ meniscal tear. We have included in the analysis any patient with these ‘possible’ meniscal tears, because this reflects the real life situation where diagnosis, even on MRI scan, is not clear cut. Diagnosis on MRI is an interpretation of scan findings and although the results tend to be reported using a binary approach, MRI itself is certainly not binary and is highly dependent on the experience of the radiologist reporting the scan.

In a further three cases the patients had previously received meniscal knee surgery. We believe that the explanation for the discordant results in these patients is that the MRI scan still shows an abnormal signal in the areas of the knee that previously underwent surgery; however, on arthroscopic visualisation any previous meniscal damage appears to have been repaired and therefore a result of no meniscal tear was recorded. Removing patients who have had previous knee surgery improved the specificity from 0.53 to 0.60 (95% CI 0.32 to 0.84). Similarly, including only those patients who were classified as not having a meniscal tear based on diagnosis by a musculoskeletal clinician improved the specificity to 0.67. This suggests that indiscriminate use of MRI to diagnose meniscal injuries is unlikely to be beneficial and a more targeted approach to use of MRI will yield more accurate and beneficial data.

Patient and public involvement

Patients were not directly involved in the design or analysis of the study. We had intended to recruit a suitable patient to the trial management group prior to initiation of the study. However, we were unable to recruit a candidate with appropriate patient feedback, and opinion was sought on the individual physical tests used within the study. Each patient who underwent examination was asked to give their view on the tests (the Thessaly Test, McMurray’s Test, Apley’s Test and the joint line tenderness Test), particularly in relation to the pain experienced during the test.

Overall, 72% of patients said that one or more tests were painful when performed. In particular the joint line tenderness Test (57.1% of patients) and the Thessaly Test (50.5%) were found to cause pain during the test procedure. In comparison McMurray’s Test and Apley’s Test were reported as causing pain less frequently, 37.7% and 33.3% respectively.