The diagnostic accuracy of clinical tests for anterior cruciate ligament tears are comparable but the Lachman test has been previously overestimated: a systematic review and meta-analysis

A search was conducted for relevant studies without restriction on date of publication using the bibliographic databases PubMed, Scopus, MEDLINE and Web of Science (Supplemental Table 1).

Results from bibliographic databases were combined and duplicates removed. Studies obtained through screening previous systematic reviews were also considered and PRISMA flow diagram and checklist followed to screen literature and report selection of relevant studies (Fig. 1 and Supplemental Fig. 1) [51].

Inclusion and exclusion criteria for the study are defined in Supplemental Table 2. Bibliographic database search results were uploaded into Rayyan QCRI web application (https://​rayyan.​qcri.​org/​cite) [56] and the titles and abstracts of every citation screened to exclude clearly irrelevant studies. Remaining citations were independently reviewed by PAS, RAO and RN for eligibility based on title, abstract and full text and conflicts resolved by consensus.

Standardised assessment of the quality of reporting by primary diagnostic accuracy studies and risk of bias regarding applicability of results was determined using QUADAS-2 tool [78] (Supplemental Table 3).

Values of 2 × 2 contingency tables [number of true positives (TP), false positives (FP), true negatives (TN) and false negatives (FN)] were extracted from each study. If this information was not provided, the values were calculated from descriptive statistics presented within the study. If this was not possible for at least one of the two diagnostic properties of sensitivity [TP/(TP + FN)] or specificity [TN/(TN + FP)] the studies were excluded from meta-analysis. A cut-off of 3-weeks post-injury was used to differentiate acute (< 3 weeks) and post-acute (> 3 weeks) presentations.

Bivariate diagnostic random-effects meta-analysis was performed where studies reported both sensitivity and specificity as this approach is considered more valid than univariate analysis [19, 65]. A univariate random effects model meta-analysis was used where studies reported only sensitivity or specificity and to enable meaningful comparison with previous systematic reviews [65]. The sensitivity and specificity were analysed by subgroups of the time since injury (acute or post-acute), injury type (partial or complete tear) and reference used (arthroscopy or MRI). Positive (LR +) and negative (LR−) likelihood ratios were calculated from the values of sensitivity and specificity to determine the pre- to post-test shift in probability of an ACL tear [48]. The analysis was performed in R Studio (https://​rstudio.​com; supplemental information) [61] using mada [19] and meta [65] packages.

The total number of citations retrieved from the bibliographic databases was: PubMed 7369, Scopus 17125, MEDLINE 5094 and Web of Science 4329. Screening of previous systematic reviews identified 24 additional citations. Screening titles and abstracts of the citations against the inclusion and exclusion criteria identified 80 studies for potential inclusion with in-depth scrutiny of each article generating 24 studies for final inclusion in this review [6, 8, 9, 11, 12, 18, 20, 22, 25, 26, 33, 38, 39, 41‐45, 49, 54, 57, 62, 72, 75]. Selection of studies is summarised in the PRISMA flow diagram (Fig. 1) and details of excluded studies with justifications provided in supplemental information.

Comprehensive description of each study and patient cohorts is provided in Tables 1 and 2. Only two studies [6, 8] were not considered at risk of bias (Table 3). One study was an RCT [6], seven studies had a prospective design [8, 42, 43, 49, 54, 72, 75] and 14 assessed patients retrospectively [9, 12, 18, 20, 22, 26, 33, 38, 39, 41, 44, 45, 57, 62]. Arthroscopy and/or arthrotomy was used as a gold standard in 17 studies [6, 8, 9, 11, 12, 18, 20, 22, 25, 38, 39, 43, 44, 54, 57, 62, 72] with MRI reported as a reference standard in seven studies [26, 38, 41, 42, 45, 49, 75]. Isolated ACL tears were analysed in seven studies [6, 9, 11, 20, 25, 42, 49], whereas ten studies recruited patients with an isolated ACL tear or combined ACL tear and meniscal injury [8, 18, 20, 22, 26, 39, 41, 62, 72, 75]. The spectrum of conditions was not specified or unclear in eight studies [12, 18, 33, 38, 43‐45, 57]. Patient cohorts with complete ACL tears only were analysed by ten studies [6, 8, 11, 12, 20, 25, 39, 42, 44, 75], partial ACL tears only were analysed in two studies [22, 42], combined complete and partial ACL tears in nine studies [9, 18, 22, 26, 38, 43, 45, 49, 75]. Six studies did not specify the type of ACL tear [33, 41, 54, 57, 62, 72]. The number of patients in cohorts extracted from studies ranged from six [22] to 217 [22]. The average age of patients ranged from 21 [8] to 45 [6, 39] with seven studies including patients below 16 years of age [8, 9, 20, 33, 45, 49, 62]. Time from injury was not defined in three studies [38, 45, 72] while acute, subacute and/or chronic injuries were included in the same cohort in six studies [9, 33, 41, 43, 57, 75]; three studies described patients as acute yet did not specify time from injury [8, 49] or provided only the average time [62].Each study has been screened for risk of bias based on 4 criteria of patient selection, index test, reference standard and flow and timing. Bias screening is presented as Low Risk (if the answer is “yes” to all signalling questions for a domain) High Risk (if the answer is “no” to all signalling questions for a domain), ? Unclear Risk (insufficient data reported precludes judgement)

Two studies performed arthrocentesis before examination [26, 62] and two studies did not aspirate the joint, while the remaining studies make no mention of such interventions. Six studies implemented measures to blind the examiner to patient history [6, 11, 25, 33, 54, 75] and available imaging [6, 11, 25, 33, 42, 54, 75] or to prevent facial recognition of a previously seen patient [11], whilst other studies did not report and/or did not use such strategies. Random order of testing was applied in four studies [6, 11, 25, 54].

Bivariate random effects model meta-analysis was possible for data from 12 studies (Table 4 and Supplemental Tables 4, 5, 6) [6, 11, 25, 26, 33, 38, 41, 45, 49, 54, 72, 75]. For all studies combined, the pivot shift test had the highest specificity (94%) and LR + (10.70), producing a large shift in the positive post-test probability. However, this test also had the lowest sensitivity (55%) and highest LR− (0.48), producing the smallest shift in negative post-test probability. The Lever sign produced moderate shifts in the positive post-test probability and was the only test to produce a moderate shift in the negative post-test probability. The anterior drawer and Lachman tests produced moderate shifts in the positive post-test probability and small shifts in the negative post-test probability. The post-test probability of an ACL tear for all tests is illustrated in Fig. 2 using a 36% pre-test probability [63].

Based on studies using gold-standard arthroscopy and/or arthrotomy for reference [6, 8, 9, 11, 12, 18, 20, 22, 25, 38, 39, 43, 44, 54, 57, 62, 72], the anterior drawer was the most accurate test with the highest LR+ (25.10) and joint lowest LR− (0.17) alongside the Lever sign, but LR+ confidence intervals were wide. The pivot shift test produced large shifts in the positive post-test probability, with the Lachman and Lever sign producing moderate shifts. The anterior drawer and Lever sign tests produced moderate shifts in the negative post-test probability, with the Lachman and pivot shift tests producing small shifts.

Where MRI was used as the reference standard (Supplemental Table 4) [26, 38, 41, 42, 45, 49, 75], the Lever sign was the most accurate test with the highest LR+ (13.50) and lowest LR− (0.20). The specificity and LR + of the anterior drawer and LR + of the pivot shift were considerably lower than values reported using arthroscopy and/or arthrotomy.

For specific presentations, bivariate random effects model meta-analysis was only possible for the Lachman test in post-acute injuries and complete ACL tears. In both presentations, the Lachman test produced small shifts in the positive and negative post-test probability (Supplemental Tables 5 and 6).

Pooled diagnostic accuracy estimates were determined with univariate analyses of data from 23 studies [6, 8, 9, 11, 12, 18, 20, 22, 25, 26, 33, 38, 39, 41‐45, 49, 54, 57, 72, 75] (Table 4 and Supplemental Tables 4, 5, 6). In acute presentations (< 3 weeks since injury), the Lever sign was the most accurate test with the highest sensitivity (100%) and joint highest specificity (94%) alongside the pivot shift. In post-acute presentations (> 3 weeks since injury), the Lever sign had the highest sensitivity (100%), with the anterior drawer and pivot shift tests demonstrating considerably higher sensitivity values than in acute presentations. The Lachman test’s specificity was also higher than in acute presentations; insufficient data were available to calculate specificity values for the anterior drawer or Lever sign tests.

For studies reporting complete ACL tears only (Supplemental Table 5), the pivot shift test demonstrated the highest specificity (96%) but lowest sensitivity (48%). The Lever sign and Lachman tests had the highest sensitivity, with Lachman also demonstrating high specificity; insufficient data were available to determine the Lever sign test’s specificity in complete tears. For partial ACL tears, sensitivity values only were available for all four tests (Supplemental Table 5). With exception of the Lever sign, all tests demonstrated inferior diagnostic ability for ruling out partial ACL tears compared with complete ruptures.

Benjaminse et al. [5], the highest quality meta-analysis with the most studies included [14], employed a univariate approach and describe Lachman as the most valid test overall (Sn:85%, Sp:94%). An equivalent univariate analysis performed in the current study demonstrates almost identical values (Sn:85%, Sp:93%) (Table 4) but bivariate analysis, which provides a more accurate estimate of pooled effect sizes, indicates lower diagnostic accuracy. Since Benjaminse et al. [5] included studies with concomitant knee ligament injury and the current review excluded such studies, concomitant ligament injury may not necessarily affect the validity of the Lachman test, but its diagnostic accuracy may have been previously overestimated. Future studies should directly compare the diagnostic accuracy of tests in ACL-injured patients with and without additional ligament injury.

The findings of the present study support recommendations that the pivot shift should be used to rule in an ACL tear when positive [5, 14] but the Lachman test did not demonstrate superior validity when compared with the anterior drawer or Lever sign; therefore, the hypotheses were rejected. Specifically, the diagnostic accuracy of the Lachman test is considerably lower than previously reported in post-acute presentations [5] and complete tears [40] based on bivariate analysis values.

The lever sign appears to be an accurate test regardless of time since injury or tear type, somewhat refuting the interdependence between sensitivity and specificity. However, it is worth noting that the only study with moderate methodology reported a sensitivity of 68% [33] and most studies demonstrate limited quality [2, 59]. Reiman et al. [59] reported two diagnostic accuracy values, one including and the other excluding data from the original study [42]. With the original data omitted (400 tests with no false positive or false negative results), the LR + decreased from 128.0 to 13.1 indicating verification and case–control bias [59]. Other studies reporting on the Lever sign test are also at risk of verification bias [9, 18, 26, 33, 39, 43, 49, 72]; therefore, results based on time since injury and tear type should be interpreted with caution.

The original Lever sign study data [42] were included in univariate analyses but excluded from bivariate analyses; the latter demonstrates inferior diagnostic values for the Lever sign test, yet it is still comparable with the other tests. Although bivariate analysis of MRI studies identifies the Lever sign as the most accurate test for diagnosing ACL tears, the diagnostic accuracy of MRI is dependent on magnetic field strength [58, 66] and arthroscopic assessment remains the gold standard; therefore, results should be interpreted accordingly. Given that the Lever sign is the only test to produce a moderate shift in the negative post-test probability, other modalities (i.e., arthrometry, MRI or arthroscopy) should be considered when the history is suggestive of an ACL injury, but clinical tests are negative.

To ensure this review was as clinically relevant as possible, only data for clinical tests performed on non-anaesthetised (awake) patients, without additional measuring equipment (e.g., arthrometry), were included. PRISMA guidelines and QUADAS-2 risk of bias analysis were used to promote methodological quality of the study. Where possible, a bivariate random effects model meta-analysis was performed as this method provides the most accurate estimate of pooled effect sizes and is recommended for meta-analysis of diagnostic test accuracy [27]. However, only 12 studies qualified for bivariate analysis and a lack of sufficient data precluded comparison between tests based on time since injury and tear type. Further research is required to establish more accurate estimates for the Lachman test in acute presentations and partial ACL tears using bivariate analysis.

Univariate analyses were performed where bivariate analysis was not possible and to allow comparison with previous systematic reviews [2, 5, 59], but the difference between methods should be acknowledged. For example, univariate analyses in post-acute presentations and complete ACL tears demonstrate higher sensitivity and specificity values for the Lachman test than bivariate methods, thereby overestimating the test’s diagnostic accuracy. The likelihood ratios calculated by univariate analysis also cast doubt on the accuracy and reliability of these values; therefore, results should be interpreted with caution.

The methodological quality of many studies was compromised by numerous factors including a retrospective design and lack of examiner blinding from clinical information that could bias the test outcome. With specific reference to the Lever sign test, studies should clearly report the landmarks chosen for hand placement, surface used (hard or soft), fist size, calf size and softness, all of which could affect test outcomes [33, 39, 42, 47]. Future studies should comply with the STARD guidelines [10] for completeness and transparency of reporting.

For this review, arbitrary time frames were used to differentiate acute and post-acute injuries. Whilst the terminology and time frames are a subject of debate, these categories were applied based on the most frequently reported thresholds amongst included studies. In addition to time since injury, future studies should report other covariables that could influence the outcome of a test. For example, a patient that is examined 4 weeks after injury may no longer be categorised as ‘acute’ but can still present with impairments that impact the examiner’s ability to perform a test unequivocally (e.g., pain, effusion, protective guarding). It is proposed that presentations should be differentiated not only by the time since injury, but also the presence or absence of associated impairments. This has previously been suggested to improve patient care following injury [36].

Studies with concomitant ligament injury were excluded from this systematic review but those reporting non-obstructive meniscal tears were included. Meta-regression analysis demonstrated increased sensitivity of the Lachman test with a concomitant meniscal tear, but no difference for the pivot shift or Lever sign tests. No studies reported meniscal root tears or ramp lesions, which have been shown to increase anterior and rotational laxity in an ACL deficient knee [16, 17, 23, 53, 68]. However, the awareness and understanding of these associated lesions has improved over the last decade and they may, therefore, have been overlooked in older studies. Future studies should compare ACL test findings with and without concomitant meniscal injury, to determine their impact on diagnostic accuracy.