Diagnosis (Imaging)

Ultrasound
Six high quality studies evaluated the use of ultrasound for diagnosing rotator cuff tears (Cheng 2015, Day 2016, Gormeli 2014, Moosmayer 2007, Ok 2013, Waldt 2007).

Four of the studies tested for the presence of any rotator cuff tear (full/partial thickness) vs no tear (Day 2016, Gormeli 2014, Moosmayer 2007, Waldt 2007). A meta-analysis of these four studies was conducted (see figures 1 and 2 for ROC curve and forest plot in eAppendix 2). The pooled positive LR was 3.60 (2.10,6.00) indicating that a positive result produced a small, but sometimes important increase in probability of a tear. The confidence interval for this pooled effect crosses 5(the threshold for a moderate effect), so we cannot rule out the possibility that the test is of moderate strength for ruling in a tear.  The pooled estimate of the negative LR revealed substantial heterogeneity, with an I-squared of 50.10%. Therefore, only the range of negative LRs from the included studies are reported in the summary of findings tables. Negative LRs ranged from 0.07 (indicating a strong rule-out test) to 0.53 (indicating a poor rule out test).

Four studies evaluated the ability of ultrasound to distinguish between full thickness tears and either partial or no tear (Cheng 2015, Gormeli 2014, Ok 2013, Waldt 2007). A meta-analysis of positive and negative LRs produced consistent estimates, with I squared heterogeneities of 0 and 22.15%. The Meta-Analysis ROC curves and forest plots can be found in figures 3 and 4. A positive ultrasound test produced a moderate increase in probability of full thickness tears vs partial/no tears (pooled positive LR=5.20 (3.20,8.20)).  A negative test produced a small, but sometimes important decrease in the probability a patient did not have a full thickness tear, but instead either had a partial tear or no tear (negative LR=0.28 (0.20,0.38)).

Magnetic Resonance Arthrography (MRA)
12 high quality studies tested MRA for the diagnosis of rotator cuff tears. (Anbar 2015, Dae 2009, Duc 2006, Lee 2018, Magee 2014, Magee 2016, Ok 2013, Pfirrmann 1999, Probyn 2007, Schaeffeler 2012, Schreinemachers 2009, Waldt 2007).

Five studies used MRA to distinguish between any tear (full/partial thickness) and no tear (Dae 2009, Pfirrmann 1999, Probyn 2007, Schaeffeler 2012, Schreinemachers 2009). A meta-analysis was attempted, but pooled estimates of positive and negative LRs revealed substantial heterogeneity, with I squared statistics of 85.16 and 80.33% respectively. Therefore, the range of positive and negative LRs are presented, rather than the pooled estimates. The positive LRs ranged from 1.6 (poor test) to 53.57 (Strong test) for ruling in any tear vs no tear. The negative LRs ranged from 0 (strong test) to 0.65 (poor test) for ruling out any tear vs no tear. Given these results, MRA may provide a benefit for diagnosing any tear, but the exact strength of the test is unclear given widely heterogeneous results between studies.

Four studies tested the ability of MRA to distinguish between full thickness tears and partial or no tears (Duc 2006, Lee 2018, Ok 2013, Waldt 2007). A meta-analysis was conducted(see figures 5 and 6 for ROC curve and forest plot). The meta-analysis produced consistent estimates of the positive LR, indicating that a positive MRA test for full thickness tears produced a large increase in probability that a patient truly had a full tear instead of a partial or no tear (pooled positive LR= 19.55 (5.73,55.72)). Pooled estimates for the negative LR had high heterogeneity (I squared=91.15%), so the pooled result is not reported. The negative LR in the included studies ranged from 0 (strong rule out test) to 0.68 (poor rule out test).

Three studies tested the ability of MRA to distinguish between a full thickness tear and no tear (Anbar 2015, Magee 2014, Magee 2016). These could not be meta-analyzed because a minimum of four studies is required. A positive test produced a moderate to large increase in probability of a full thickness tear versus no tear(positive LR range=7.33 to 100). The negative LR was more variable, ranging from 0(a strong decrease in probability of a full tear with a negative test) to 0.25(a small but sometimes important decrease in probability).

Two studies evaluated MRA for diagnosing partial tears vs no tears (Dae 2009, Lee 2018). Positive LRs ranged from a small(but sometimes important) increase in probability of a partial tear with a positive test (positive LR=4.16) to a large increase in probability of tear(positive LR=13.42).   Negative LRs ranged from a strong decrease in probability of a partial tear with a negative test(negative LR=0.03), to a small but sometimes important decrease in probability of a partial tear(negative LR=0.44).

Magnetic Resonance Imaging (MRI)
Thirteen high quality studies evaluated MRI for the diagnosis of rotator cuff tears (Binkert 2001, Herold 2006, Lee 2016, Lee 2018, Magee 2014, Magee 2016, Mohtadi 2004, Razmjou 2016, Ryu 2016, Shellock 2001, Tuite 1994, VanBeek 2014, Yildiz2017).

Seven studies used MRI to test for any rotator cuff tear(full or partial thickness) versus no tear (Herold 2006, Lee 2018, Ryu 2016, Shellock 2001, Tuite 1994, VanBeek 2014, Yildiz 2017).  An attempt was made to meta-analyze these studies, but there was very high heterogeneity in estimates of the positive and negative LRs between studies (I squared=91.94% and 91.8% respectively). Therefore, only the range of estimates from the studies is reported.  The positive likelihood ratios ranged from 1.18(poor rule in test) to 97.06 (strong rule in test).  The negative LRs also ranged from a strong rule out test (negative LR=0.03) to a poor rule out test (0.79).

Six high quality studies used MRI to distinguish full thickness tears from no tear (Binkert 2001, Magee 2014, Magee 2016, Mohtadi 2004, Razmjou 2016, Tuite 1994). Again, meta-analysis results revealed high heterogeneity, and likelihood ratio ranges are presented here. The studies indicated that MRI was a moderate to strong test for ruling in a full thickness tear over no tear (positive LR range=6.82-100). The negative LR ranged from a large decrease in probability of a full tear with a negative test (negative LR=0) to a small, but sometimes important decrease in probability of a full tear(negative LR=0.47).

Five studies evaluated the ability of MRI to distinguish between partial tears versus no tear. A meta-analysis produced consistent estimates of the positive LR, but negative LRs were inconsistent between studies. The ROC curve and forest plot can be found in figures 7 and 8. The pooled positive LR was 3.40 (2.10,5.40), indicating that a positive MRI produced a small, but sometimes important increase in probability of a partial tear.  The negative LR ranged from 0.27 (a small but possibly important decrease in probability of a partial tear with a negative test) to 0.84 (a poor rule out test).