Abstract
Introduction
Full-thickness rotator cuff tear (FTRCT) often requires surgical treatment. Magnetic resonance imaging (MRI) is the diagnostic gold standard; however, it is costly and resource-intensive. We aimed to develop and validate a clinical prediction model for FTRCT in patients with RCI, with the goal of identifying those at high risk at the initial visit.
Methods
This prospective study consecutively enrolled patients clinically diagnosed with RCI from May 2025 to September 2025. All patients underwent shoulder MRI to confirm the presence or absence of FTRCT. Potential predictors involving demographic characteristics, clinical history, and physical examination findings, were collected. Predictor selection was performed using least absolute shrinkage and selection operator (LASSO) regression, followed by multivariable logistic regression to construct a risk prediction model, and then a nomogram was developed.
Results
357 patients were included, MRI confirmed FTRCT in 166 patients (46.5%). Patients were randomly divided into a training set (n = 249) and a test set (n = 108). LASSO regression identified 10 candidate predictors, and multivariable logistic regression identified physical labor, history of trauma, age, daily pain score, pain duration, and physical examination shoulder strength as independent predictors of FTRCT. The nomogram achieved an AUC of 0.910 (95% CI, 0.88–0.95), with 88% sensitivity, 82% specificity, and a C-index of 0.890 (95% CI, 0.83–0.95). Calibration and decision curve analyses supported its accuracy and clinical utility.
Conclusions
This study developed a preliminary clinical nomogram for predicting FTRCT in patients with RCI, which showed good performance in internal validation and may help support early risk assessment and MRI triage. Its generalizability should be confirmed in external cohorts.
Level of evidence
Level III
Supplementary Information
The online version contains supplementary material available at https://doi.org/10.1007/s00402-026-06521-y.
Keywords: Rotator cuff injury, Nomogram, Magnetic resonance imaging, Clinical decision-making
Introduction
Rotator cuff injury (RCI) is one of the most common causes of shoulder pain and dysfunction and represents a substantial health burden worldwide [1]. Its prevalence increases markedly with age, affecting approximately 40%–50% of individuals older than 60 years [2–6]. Given its high frequency and its impact on daily activities, work capacity, and quality of life, RCI imposes considerable personal and socioeconomic costs.
Management of RCIs depends largely on the severity of the structural damage. Full thickness rotator cuff tears (FTRCTs) generally require surgical repair [7, 8], whereas partial-thickness tears and tendinopathy are usually managed conservatively at the initial stage [9]. Although RCIs can often be suspected on the basis of history and specialized physical examination, magnetic resonance imaging (MRI) remains the reference standard for confirming the presence and severity of tendon injury and for guiding treatment decisions [7]. However, routine MRI for every patient with suspected RCI is neither practical nor efficient. MRI is costly, time- consuming, and often limited by appointment availability, which may increase the burden on both patients and healthcare systems. At the same time, failure to obtain MRI in patients with FTRCI may delay surgical referral, potentially resulting in poor outcomes [8, 10]. Delayed surgery for full-thickness tears has been associated with inferior outcomes, as untreated tears may progress in size and fatty infiltration over time [8, 11]. Therefore, clinicians face a common dilemma at the initial visit: how to identify patients who are more likely to harbor FTRCT and should be prioritized for MRI evaluation. Previous studies on the severity assessment of RCIs have mainly focused on identifying individual risk factors associated with RCT [12–14]. In addition, imaging-based parameters such as acromial spur morphology, acromiohumeral interval, and acromioclavicular angle have also been explored as predictors of rotator cuff tears [15–18]. However, risk factors alone do not provide clinicians with an intuitive or quantitative estimate of an individual patient’s probability of having a full-thickness tear. A practical clinical prediction tool may therefore be useful for early risk stratification and more selective MRI utilization.
Accordingly, this study was to develop and validate a clinical prediction model for estimating the probability of FTRCT in patients with RCI using information that is readily available at the initial clinical visit. We sought to provide a simple and clinically applicable tool to support early decision-making, facilitate more appropriate MRI referral, and improve the efficiency of care for patients with suspected rotator cuff injury.
Materials and methods
Study design and patient population
This was a prospective study conducted at our shoulder joint specialty clinic. From May 2025 to September 2025, we consecutively enrolled patients clinically diagnosed with a rotator cuff injury (RCI) by a senior shoulder specialist based on a comprehensive physical examination. The inclusion criteria were: (1) age ≥ 18 years; (2) clinical diagnosis of RCI confirmed by a senior shoulder specialist through standardized physical examination maneuvers (including the Jobe test, external rotation resistance test, and abdominal compression test); (3) willingness to participate and provide written informed consent. The exclusion criteria were: (1) a history of shoulder surgery on the affected side; (2) combined with other conditions causing shoulder pain (e.g., shoulder fracture, infection, rheumatoid arthritis); (3) neurological disorders affecting shoulder strength (e.g., cervical radiculopathy, stroke sequelae); (4) inability to undergo MRI (e.g., claustrophobia, non-MRI compatible implants). All procedures followed the Declaration of Helsinki.
Data collection and predictor variables
Demographic and clinical history include: age, sex, dominant or non-dominant side, whether a direct trauma leading to shoulder pain or not, engagement in physical labor (PL), history of hypertension, diabetes, coronary heart disease, hyperlipidemia, and smoking. PL was further classified as heavy (e.g., construction workers, miners, farmers without automation) or mild (e.g., photographers, musicians, nurses, teachers) [19]. Daily pain score (DPS) was the average shoulder pain level of daily activity, Pain intensity was measured using a 10-cm visual analog scale (VAS), ranging from 0 (no pain) to 10 (worst imaginable pain). Patients were asked to rate their average shoulder pain during daily activities [20]. Pain time (PT, in months) was defined as the period from the first occurrence of shoulder pain (including intermittent pain with transient relief) to the clinic visit. The duration was recorded in months.
Specialized physical examination was performed by an experienced shoulder specialist. Standardized tests included the jobe test (supraspinatus assessment) [21, 22], abdominal compression test (subscapularis assessment) [23], and external rotation resistance (infraspinatus/teres minor assessment) [24, 25]. Results of the above tests are classified as positive or negative, for more precise quantification of specialized physical examinations, we introduced two quantitative indicators for each specialized test: physical examination shoulder strength (PESS) and physical examination pain score (PEPS). For each of the three specialized test, the patient was asked to rate the intensity of pain using VAS, which was described previously. PEPS was defined as the highest VAS score recorded across the three specialized tests. PESS was assessed by a shoulder specialist during physical examination, who compared the strength of the affected side with the healthy contralateral side. For each of the three standardized physical examination maneuvers, PESS was defined as the strength of the affected shoulder expressed as a percentage of that of the contralateral shoulder. The patient-level PESS used for analysis was defined as the lowest percentage among the three test-specific measurements. All PESS measurements were completed and recorded before MRI results were available and were therefore independent of the subsequent MRI findings.
All patients underwent standardized shoulder MRI using a 3.0 Tesla Siemens scanner with a dedicated shoulder coil. The protocol included T1-weighted, T2-weighted, and proton density-weighted fat-suppressed sequences in oblique coronal, oblique sagittal, and axial planes, slice thickness was 3 mm with a 0.3 mm interslice gap. A FTRCT was defined as complete discontinuity of tendon fibers from the articular to the bursal surface [26]. MRI images were independently reviewed by two experienced shoulder specialists (with > 5 years of musculoskeletal MRI experience) blinded to clinical data. Disagreements were resolved by a third specialist, with the majority opinion adopted as final.
Statistical analysis
Data were analysed using SPSS version 26.0 (IBM Corp, Armonk, NY, USA) and R software (Version 4.4.1, R Foundation for Statistical Computing, Vienna, Austria). Continuous variables were expressed as mean ± standard deviation (SD) or median (interquartile range, IQR) as appropriate; categorical variables were presented as numbers and percentages (%).
Variable selection was performed using least absolute shrinkage and selection operator (LASSO) regression. Variables with non-zero coefficients were selected as candidate predictors and further analysed by univariable and multivariable logistic regression to identify independent risk factors for FTRCT. According to guideline recommendations for clinical prediction model development [27], the required sample size is at least ten times the number of predictor variables. With 15 candidate variables, the minimum required sample size was 150, our sample of 357 patients was therefore adequate. A nomogram was then constructed based on the selected predictors. Model performance was evaluated by the concordance index (C-index), receiver operating characteristic (ROC) curve, calibration plot, and decision curve analysis (DCA). The C-index and area under the ROC curve (AUC) ranged from 0.5 to 1.0, with higher values indicating better discriminative ability. Calibration plots assessed agreement between predicted and observed probabilities. DCA evaluated clinical utility based on net benefit across a range of threshold probabilities. A two-sided P < 0.05 was considered statistically significant. To further evaluate the internal validity and stability of the model, bootstrap resampling with 1,000 replications was performed. The optimism-corrected AUC was calculated to assess potential overfitting.
Result
Patient characteristics
As shown in the study flow diagram (Fig. 1), a total of 476 consecutive patients with shoulder pain were initially screened, among whom 390 patients clinically diagnosed with RCI were enrolled and 33 were subsequently excluded according to predefined criteria, with 357 included in the final analysis.
Fig. 1.
Flowchart of patient selection. A total of 476 patients with shoulder pain were included in this study. 390 of them were clinically diagnosed with rotator cuff injury (RCI). According to inclusion and exclusion criteria, 357 patients were included in the final analysis. MRI confirmed full-thickness rotator cuff tear (FTRCT) in 166 patients (46.5%)
The mean age of participants was 53.5 ± 11.2 years (range, 21–80 years); 143 (40.1%) were male and 213 (59.9%) were female. MRI confirmed FTRCT in 166 patients (46.5%). Among the remaining 191 patients, 103 (28.9%) had partial-thickness rotator cuff tears,82 (23.0%) had tendinopathy, while 6 (1.7%) were misdiagnosed as RCI based solely on physical examination, but were subsequently confirmed by MRI to have other shoulder pathologies (e.g., adhesive capsulitis, subacromial bursitis, glenohumeral osteoarthritis). A detailed breakdown of the MRI diagnoses is presented in Table 1. Patients were randomly divided into a training set (n = 249) and a test set (n = 108) at a 7:3 ratio. Baseline characteristics were comparable between the training and test sets (Table 2).
Table 1.
MRI diagnosis distribution among 357 patients clinically diagnosed with RCI
| MRI Diagnosis | n (%) |
|---|---|
| Full-thickness rotator cuff tear (FTRCT) | 166 (46.5%) |
| Partial-thickness rotator cuff tear (PTRCT) | 103 (28.9%) |
| Tendinopathy without tear | 82 (23.0%) |
| Other (e.g., osteoarthritis, labral tear) | 6 (1.7%) |
| Total | 357 (100%) |
FTRCT full-thickness rotator cuff tear, PTRCT partial-thickness rotator cuff tear
Table 2.
Patient demographics and baseline characteristics(7: 3) Incorporate cases into the training and validation sets in a 7:3 ratio. DPS Daily pain score, PT Pain time, PEPS Physical examination pain score, PESS Physical examination shoulder strength (percentage of the healthy side), PL Physical labor (0 represents no physical labor in work, 1 represents mild physical labor, and 2 represents heavy physical labor). There is no significant difference in baseline characters between test and train set
| Variables | Total (n = 357) | test (n = 108) | train (n = 249) | Statistic | P |
|---|---|---|---|---|---|
| Age, Mean ± SD | 53.48 ± 11.19 | 54.76 ± 11.35 | 52.93 ± 11.10 | t = 1.42 | 0.157 |
| DPS, M (Q₁, Q₃) | 4.00 (3.00, 5.00) | 4.00 (3.00, 5.00) | 4.00 (3.00, 5.00) | Z=−0.02 | 0.980 |
| PT, M (Q₁, Q₃) | 4.00 (2.00, 8.00) | 4.00 (2.00, 7.00) | 4.00 (2.00, 10.00) | Z=−0.08 | 0.935 |
| PEPS, M (Q₁, Q₃) | 6.00 (5.00, 7.00) | 6.00 (5.00, 7.00) | 6.00 (5.00, 7.00) | Z=−0.47 | 0.639 |
| PESS, M (Q₁, Q₃) | 60.00 (40.00, 75.00) | 60.00 (40.00, 70.00) | 60.00 (40.00, 75.00) | Z = 0.00 | 1.000 |
| Sex, n(%) | – | – | – | χ²=0.59 | 0.443 |
| Male | 143 (40.06) | 40 (37.04) | 103 (41.37) | – | – |
| Female | 214 (59.94) | 68 (62.96) | 146 (58.63) | – | – |
| Dominantside, n(%) | – | – | – | χ²=0.39 | 0.534 |
| Non-dominant | 163 (45.66) | 52 (48.15) | 111 (44.58) | – | – |
| Dominant | 194 (54.34) | 56 (51.85) | 138 (55.42) | – | – |
| PL, n(%) | – | – | – | χ²=6.80 | 0.33 |
| None | 242 (67.79) | 65 (60.19) | 177 (71.08) | – | – |
| Mild | 82 (22.97) | 27 (25.00) | 55 (22.09) | – | – |
| Heavy | 33 (9.24) | 16 (14.81) | 17 (6.83) | – | – |
| Trauma, n(%) | – | – | – | χ²=0.14 | 0.707 |
| No | 263 (73.67) | 81 (75.00) | 182 (73.09) | – | – |
| Yes | 94 (26.33) | 27 (25.00) | 67 (26.91) | – | – |
| Diabetes, n(%) | – | – | – | χ²=0.82 | 0.364 |
| No | 312 (87.39) | 97 (89.81) | 215 (86.35) | – | – |
| Yes | 45 (12.61) | 11 (10.19) | 34 (13.65) | – | – |
| Hypertension, n(%) | – | – | – | χ²=0.18 | 0.675 |
| No | 249 (69.75) | 77 (71.30) | 172 (69.08) | ||
| Yes | 108 (30.25) | 31 (28.70) | 77 (30.92) | ||
| Hyperlipidemia, n(%) | – | – | – | χ²=0.39 | 0.821 |
| No | 225 (63.03) | 67 (62.04) | 158 (63.45) | – | – |
| Yes | 82 (22.97) | 24 (22.22) | 58 (23.29) | – | – |
| Not reported | 50 (14.01) | 17 (15.74) | 33 (13.25) | – | – |
| Smoke, n(%) | – | – | – | χ²=0.25 | 0.619 |
| No | 292 (81.79) | 90 (83.33) | 202 (81.12) | – | – |
| Yes | 65 (18.21) | 18 (16.67) | 47 (18.88) | – | – |
t t-test, Z Mann-Whitney test, χ² Chi-square test, SD standard deviation, M Median, Q₁ 1 st Quartile, Q₃ 3 st Quartile, DPS Daily pain score, PT Pain time, PEPS Physical examination pain score, PESS Physical examination shoulder strength
Variable selection and independent predictors
LASSO regression eliminated three variables (diabetes, hyperlipidemia, smoking) due to near-zero predictive ability (Fig. 2), retaining 10 candidate predictors: sex, age, dominant side, PL, trauma history, hypertension, DPS, PT, PEPS, and PESS (Table 3).
Fig. 2.
LASSO regression for variable selection. A Tenfold cross-validation for tuning parameter selection. The optimal lambda value (λ) is indicated by the dotted vertical line. b LASSO coefficient profiles of the 15 candidate variables. Ten variables with non-zero coefficients were selected as potential predictors of FTRCT
Table 3.
The coefficients of Lasso regression analysis The coefficients of diabetes, hyperlipidemia, smoke is infinitely close to 0, so they have almost no impact on the prediction of the outcome, the other ten variables were retained for further logistic analysis
| Variable | Coefficient |
|---|---|
| Sex | −0.008380642 |
| age | 0.004089709 |
| Dominant side | 0.044904785 |
| PL | 0.080321068 |
| trauma | 0.109094746 |
| diabetes | 0.000000000 |
| hypertension | 0.067967356 |
| hyperlipidemia | 0.000000000 |
| smoke | 0.000000000 |
| DPS | 0.037745758 |
| PT | 0.004739977 |
| PEPS | 0.002132156 |
| PESS | −0.011251260 |
LASSO Least absolute shrinkage and selection operator, PEPS Physical examination pain score, PESS Physical examination shoulder strength
Multivariable logistic regression identified six independent risk factors for FTRCT: physical labor (OR = 2.4; 95% CI, 1.4–4.1; P = 0.002), trauma history (OR = 3.1; 95% CI, 1.8–5.5; P < 0.001), age (per 10-year increase, OR = 1.9; 95% CI, 1.3–2.7; P = 0.001), DPS (per 2-point increase, OR = 2.2; 95% CI, 1.5–3.2; P < 0.001), PT (per 6-month increase, OR = 1.5; 95% CI, 1.1–2.0; P = 0.01), and PESS (per 20% decrease, OR = 3.5; 95% CI, 2.1–5.8; P < 0.001) (Table 4). PESS had the strongest predictive ability, followed by DPS, trauma, age, PL, and PT.
Table 4.
Univariable and multivariable logistic analysis We conducted regression analysis on these 10 potential predictors variables, 6 variables were ultimately identified as independent risk factors for the results, these 6 variables are bolded in the table
| Variables | Univariable Logistic Regression | Multivariable Logistic Regression | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| β | S.E | Z | P | OR (95%CI) | β | S.E | Z | P | OR (95%CI) | ||
| Sex | |||||||||||
| 1 | – | – | – | – | 1.00 (Reference) | – | – | – | – | 1.00 (Reference) | |
| 2 | 0.04 | 0.26 | 0.15 | 0.883 | 1.04 (0.63 ~ 1.72) | 0.15 | 0.42 | 0.34 | 0.732 | 1.16 (0.50 ~ 2.66) | |
| Dominantside | |||||||||||
| 0 | – | – | – | – | 1.00 (Reference) | _ | – | – | – | 1.00 (Reference) | |
| 1 | 0.48 | 0.26 | 1.85 | 0.064 | 1.61 (0.97 ~ 2.68) | 0.52 | 0.38 | 1.4 | 0.162 | 1.69 (0.81 ~ 3.52) | |
| PL | |||||||||||
| 0 | – | – | – | – | 1.00 (Reference) | – | – | – | – | 1.00 (Reference) | |
| 1 | 1.01 | 0.32 | 3.15 | 0.002 | 2.74 (1.46 ~ 5.13) | 0.92 | 0.45 | 2.06 | 0.04 | 2.52 (1.04 ~ 6.08) | |
| 2 | 1.05 | 0.53 | 1.99 | 0.047 | 2.87 (1.01 ~ 8.12) | 1.47 | 0.74 | 1.99 | 0.047 | 4.36 (1.02 ~ 18.62) | |
| Trauma | |||||||||||
| 0 | – | – | – | – | 1.00 (Reference) | – | – | – | – | 1.00 (Reference) | |
| 1 | 1.28 | 0.3 | 4.2 | < 0.001 | 3.59 (1.98 ~ 6.52) | 1.33 | 0.49 | 2.72 | 0.007 | 3.77 (1.45 ~ 9.82) | |
| Hypertension | |||||||||||
| 0 | – | – | – | – | 1.00 (Reference) | – | – | – | – | 1.00 (Reference) | |
| 1 | 0.87 | 0.28 | 3.11 | 0.002 | 2.40 (1.38 ~ 4.16) | 0.36 | 0.43 | 0.85 | 0.394 | 1.44 (0.62 ~ 3.31) | |
| Age | 0.05 | 0.01 | 4.17 | < 0.001 | 1.06 (1.03 ~ 1.08) | 0.04 | 0.02 | 1.95 | 0.049 | 1.04 (1.00 ~ 1.08) | |
| DPS | 0.71 | 0.13 | 5.61 | < 0.001 | 2.03 (1.59 ~ 2.60) | 0.56 | 0.25 | 2.3 | 0.022 | 1.76 (1.09 ~ 2.85) | |
| PT | 0.01 | 0.01 | 1.02 | 0.309 | 1.01 (0.99 ~ 1.03) | 0.04 | 0.01 | 2.62 | 0.009 | 1.04 (1.01 ~ 1.07) | |
| PSDPE | 0.77 | 0.11 | 6.88 | < 0.001 | 2.15 (1.73 ~ 2.68) | 0.04 | 0.2 | 0.2 | 0.838 | 1.04 (0.70 ~ 1.54) | |
| PESS | −0.08 | 0.01 | −8.19 | < 0.001 | 0.92 (0.90 ~ 0.94) | −0.07 | 0.01 | −5.61 | < 0.001 | 0.93 (0.91 ~ 0.95) | |
OR Odds ratio, CI Confidence interval, DPS Daily pain score, PT Pain time, PESS Physical examination shoulder strength
Nomogram development and performance
A nomogram incorporating the 10 candidate predictors was constructed (Fig. 3). In the training set, the model achieved an AUC of 0.910 (95% CI: 0.88–0.95), with a sensitivity of 88% and specificity of 82% (Fig. 4), indicating excellent discriminative ability. The C-index was 0.890 (95% CI: 0.83–0.95). Bootstrap internal validation with 1,000 resamples yielded an apparent AUC of 0.907 (95% CI: 0.878–0.937) and an optimism-corrected AUC of 0.892 (95% CI: 0.861–0.921), indicating good internal validity with minimal overfitting (Fig. 5). The calibration plot showed good agreement between predicted and observed probabilities (Fig. 6a, b). Decision curve analysis demonstrated a positive net benefit over a wide range of threshold probabilities (Fig. 7a, b). Similar performance was observed in the test set, confirming internal validity.
Fig. 3.
Nomogram for predicting the probability of full-thickness rotator cuff tear (FTRCT). To use the nomogram, locate each predictor on its corresponding axis, draw a vertical line to the “Points” axis to obtain the score for that predictor, sum the individual scores to obtain a total point value, and project the total points onto the “Risk of FTRCT” axis to obtain the predicted probability
Fig. 4.
Receiver operating characteristic (ROC) curves of the nomogram. The area under the curve (AUC) was 0.910 (95% CI: 0.88–0.95) in the training cohort, with a sensitivity of 88% and specificity of 82%
Fig. 5.
Bootstrap internal validation of the prediction model. The apparent AUC and optimism-corrected AUC with 1,000 resamples are shown
Fig. 6.
Model performance in the training cohort. a Calibration curve showing the agreement between predicted and observed probabilities of FTRCT. The diagonal dashed line represents ideal prediction. b Decision curve analysis (DCA) demonstrating the net benefit of the nomogram across a range of threshold probabilities
Fig. 7.
Model performance in the validation cohort. a Calibration curve showing the agreement between predicted and observed probabilities of FTRCT. The diagonal dashed line represents ideal prediction. b Decision curve analysis (DCA) demonstrating the net benefit of the nomogram across a range of threshold probabilities
Nomogram usage
As shown in Fig. 3, the value of each predictor is located on its corresponding axis, and a vertical line is drawn to the “Points” axis to obtain the score for that predictor (e.g., sex: 0 for male, 2.5 for female). The individual scores are summed to obtain a total point value, which is then projected onto the “Risk of FTRCT” axis to obtain the estimated probability of FTRCT.
Discussion
In this prospective cohort of patients with clinically suspected RCI, we developed and internally validated a clinical prediction model for estimating the probability of full-thickness rotator cuff tear. The model incorporated six independent predictors—age, physical labor, trauma history, daily pain score, pain duration, and physical examination shoulder strength—and showed good discrimination, calibration, and potential clinical utility. Among these, reduced shoulder strength during specialized physical examination (quantitative evaluation using PESS) was the most informative predictor. These results indicate that routinely available clinical information may provide a useful basis for early risk stratification and more selective use of MRI in this population.
The clinical rationale for developing this prediction model is supported by evidence that early surgical repair for full-thickness rotator cuff tears yields better outcomes than delayed surgery after failed conservative treatment [8]. Furthermore, untreated full-thickness tears tend to progress over time—with a reported progression rate of 54.9% at approximately 3-year follow-up [11]—and progressive fatty infiltration may compromise the success of later repair [28, 29]. These findings underscore the importance of identifying patients at high risk for full-thickness tears at an early stage, so that timely MRI and surgical referral can be considered.
Previous studies applied machine learning to predict rotator cuff tears using imaging-derived anatomical parameters [15–18]. In contrast, our model specifically predicts full-thickness tears using clinical variables available at the initial visit, addressing a distinct decision point in the clinical pathway. Previous studies evaluating the performance of specialized physical examination tests have uniformly classified results as positive or negative based on the presence of pain or subjective weakness [22–26]. In our study, we quantify specialized physical examination findings as continuous variables, and we demonstrate its dose-response relationship with FTRCT risk. This initiative could provide a reference for future studies require RCI quantification, clinicians should move beyond binary “weak/normal” assessment and adopt a graded evaluation. Notably, PESS emerged as the strongest predictor of FTRCT (OR = 3.5 per 20% decrease), outperforming other clinical variables. The strong predictive value of PESS may be explained by the fact that weakness on resisted testing reflects more severe tendon compromise, which is characteristic of full-thickness tears [30]. In contrast, PEPS showed poor predictive ability. This discrepancy likely arises because: (1) pain is less specific, whereas profound weakness is more specific to complete tendon discontinuity; (2) PEPS is more susceptible to transient confounders such as analgesic use, anxiety, or pain tolerance; (3) true mechanical weakness may outweigh pain as the dominant signal for FTRCT. Clinically, a patient with severe pain but near-normal strength (high PEPS, high PESS) is unlikely to have an FTRCT, whereas disproportionate weakness relative to pain (low PESS, moderate PEPS) should raise suspicion. Our findings suggest that quantifying strength deficits, not pain severity, should be prioritized during physical examination for FTRCT triage.
DPS and PT reflect the severity of RCI, and were independently associated with 2.2-and 1.5-fold increased risk of FTRCT respectively, in the present study. DPS captured the patient’s average pain level during daily activities, and patients with higher DPS experience greater mechanical irritation and inflammatory response around the torn tendon [31], therefore representing more severe tendon lesion. PT adds a temporal dimension to severity assessment. Longer symptom duration may indicate delayed care-seeking, failed natural healing, and continuous tendon degradation, allowing a partial tear to progress to full-thickness. Our findings showed that each 6-month increase in PT was associated with a 1.5-fold increase in the odds of FTRCT, underscoring the need to avoid prolonged “watchful waiting” beyond several months without diagnostic imaging, particularly in patients with additional risk factors.
Age, trauma, and heavy physical labor are well-established risk factors for rotator cuff tears [13, 14, 31–33]. Our study confirmed each as an independent predictor of FTRCT, with effect sizes comparable to previous reports: age (OR = 1.9 per 10-year increase vs. OR = 1.76–5.07 in prior meta-analyses [32]), trauma (OR = 3.1 vs. OR = 2.74–5.20 [31, 32]), and heavy physical labor (OR = 2.4 vs. OR = 1.57–2.51 [32]). Notably, we further distinguished heavy from mild physical labor, demonstrating a graded relationship between occupational workload and FTRCT risk. These findings reinforce the multifactorial nature of rotator cuff tears, wherein biological degeneration (age), acute overload (trauma), and chronic microtrauma (heavy labor) converge to compromise tendon integrity.
The primary clinical challenge in patients with suspected RCI is determining whether immediate MRI is warranted or whether an initial trial of conservative management is appropriate. The nomogram developed in this study provides a quantitative, individualized estimate of FTRCT probability based on clinical variables and demonstrated good discrimination.To facilitate clinical translation, we defined risk thresholds based on the nomogram’s predicted probability: low-risk (< 20%), intermediate-risk (20%–70%), and high-risk (> 70%). These cutoffs were chosen to balance sensitivity and specificity based on the ROC analysis. In the high-risk group (> 70%), where model specificity is relatively high, early MRI evaluation and timely surgical referral may be considered, given the potential impact of delayed intervention on outcomes [8, 10]. In the intermediate-risk group (20%–70%), the model provides modest net benefit, and clinical decision-making may be guided by patient preferences, symptom severity, and functional impairment. In the low-risk group (< 20%), where the sensitivity is relatively high (88%), an initial period (often 8–12 weeks) of conservative management without immediate MRI may be reasonable, with imaging reconsidered if symptoms persist or worsen. Overall, this risk-stratified approach may help support more selective use of MRI and improve resource allocation, while complementing—rather than replacing—clinical judgment.
Several limitations should be acknowledged. First, the model was developed and internally validated using a single-centre cohort of 357 patients. Although internal validation showed promising performance, external validation in diverse populations and healthcare settings is necessary to assess generalizability. Second, the model’s sensitivity (88%) and specificity (82%) are not perfect; a small number of FTRCTs may be missed (false negatives), and some patients without FTRCT may be referred for MRI unnecessarily (false positives). Therefore, the nomogram should be used as a decision support tool, not as a replacement for clinical judgment. Third, the study did not assess inter-rater reliability of physical examination findings, which may vary among clinicians. Future large-scale, multi-centre prospective studies are needed to validate and refine the model. Fourth, the high prevalence of FTRCT in this cohort (46.5%) reflects the enriched patient population at our institution, which is a regional orthopedic referral center. Some patients seen at this clinic are typically referred with more advanced pathology, representing the target population for this prediction model. Thus, the model’s findings are most applicable to specialist settings, validation in lower-prevalence primary-care populations is needed before broader use. Fifth, PESS was assessed as a semi-quantitative percentage of contralateral strength by a single examiner, without formal inter- or intra-rater reliability testing. While the use of a single experienced specialist reduced variability, the reproducibility of this measure across different examiners remains unestablished and should be addressed in future studies.
Finally, while our nomogram provides a quantitative estimate of FTRCT risk, decisions regarding MRI referral and surgery should remain individualized, taking into account patient symptoms, functional demands, age, activity level, tear chronicity, and patient preferences. The model is intended to support, not replace, clinical judgment.
Conclusion
We developed and internally validated a nomogram that accurately predicts the probability of FTRCT in patients with RCI using simple, readily available clinical information. This tool may assist clinicians in identifying high-risk patients who should be prioritized for MRI, while supporting initial conservative management in low-risk patients, thereby facilitating more efficient and cost-effective use of MRI resources. External validation is required before widespread clinical adoption.
Supplementary Information
Below is the link to the electronic supplementary material.
Author contributions
T.C.W., Q.P.T., and Y.B.Z. conceived and designed the study. T.C.W., Q.P.T., and J.L.G. collected the data. T.C.W. and B.L. performed the statistical analysis and data interpretation. T.C.W. and Q.P.T. drafted the manuscript. B.L., B.Y., and Y.B.Z. revised it critically for important intellectual content. All authors read and approved the final manuscript.
Funding
This study was supported by the Health Commission of Hebei Province (Grant No. 20270425).
Data availability
The data that support the findings of this study are available from the corresponding author upon reasonable request. Due to ethical restrictions imposed by the Institutional Review Board of the Third Hospital of Hebei Medical University (approval No. K2024-126-1) and patient confidentiality requirements, the raw data are not publicly available.
Declarations
Conflict of interest
The authors declare no competing interests.
Footnotes
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.Tashjian RZ (2012) Epidemiology, natural history, and indications for treatment of rotator cuff tears. Clin Sports Med 31:589–604. 10.1016/j.csm.2012.07.001 [DOI] [PubMed] [Google Scholar]
- 2.Smith KL, Harryman DT 2nd, Antoniou J, Campbell B, Sidles JA, Matsen FA 3rd (2000) A prospective, multi-practice study of shoulder function and health status in patients with documented rotator cuff tears. J Shoulder Elbow Surg 9:395–402. 10.1067/mse.2000.108962 [DOI] [PubMed] [Google Scholar]
- 3.Lawrence RL, Moutzouros V, Bey MJ (2019) Asymptomatic rotator cuff tears. JBJS Rev 7:e9. 10.2106/JBJS.RVW.18.00149 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Yamamoto A, Takagishi K, Osawa T et al (2010) Prevalence and risk factors of a rotator cuff tear in the general population. J Shoulder Elb Surg 19:116–120. 10.1016/j.jse.2009.04.006 [DOI] [PubMed] [Google Scholar]
- 5.Tempelhof S, Rupp S, Seil R (1999) Age-related prevalence of rotator cuff tears in asymptomatic shoulders. J Shoulder Elb Surg 8:296–299 [DOI] [PubMed] [Google Scholar]
- 6.Yamaguchi K, Ditsios K, Middleton WD, Hildebolt CF, Galatz LM, Teefey SA (2006) The demographic and morphological features of rotator cuff disease: a comparison of asymptomatic and symptomatic shoulders. J Bone Joint Surg Am 88:1699–1704 [DOI] [PubMed] [Google Scholar]
- 7.Pandey V, Jaap Willems W (2015) Rotator cuff tear: a detailed update. Asia Pac J Sports Med Arthrosc Rehabil Technol 2:1–14. 10.1016/j.asmart.2014.11.003 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Moosmayer S, Lund G, Seljom US et al (2019) At a 10-year follow-up, tendon repair is superior to physiotherapy in the treatment of small and medium-sized rotator cuff tears. J Bone Joint Surg Am 101:1050–1060. 10.2106/JBJS.18.01373 [DOI] [PubMed] [Google Scholar]
- 9.Andrews JR (2005) Diagnosis and treatment of chronic painful shoulder: review of nonsurgical interventions. Arthroscopy 21:333–347 [DOI] [PubMed] [Google Scholar]
- 10.Shah NS, Suriel Peguero E, Umeda Y, Crawford ZT, Grawe BM (2022) Long-term outcomes of massive rotator cuff tear repair: a systematic review. HSS J 18:130–137. 10.1177/15563316211008137 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Garcia MJ, Caro D, Hammerle MV, Villarreal JB, DeAngelis JP, Ramappa AJ, Nazarian A (2024) Disparities in rotator cuff tear progression definitions and rates: a systematic review. JB JS Open Access 9:e24.00097. 10.2106/JBJS.OA.24.00097 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Zhao J, Pan J, Zeng LF, Wu M, Yang W, Liu J (2021) Risk factors for full-thickness rotator cuff tears: a systematic review and meta-analysis. EFORT Open Rev 6:1087–1096. 10.1302/2058-5241.6.210027 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Preziosi Standoli J, Candela V, Gumina S (2025) Rotator cuff tear and associated risk factors: the mutual role of medical conditions. Sports Med Arthrosc Rev 33:54–60. 10.1097/JSA.0000000000000401 [DOI] [PubMed] [Google Scholar]
- 14.Khadour FA, Khadour YA, Alharbi NSK, Alhatem W, Albarroush D, Dao X (2025) Risk factors for rotator cuff tear in Syrian adults: a cross-sectional study. Sci Rep 15:5837. 10.1038/s41598-025-89878-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Hou Z, Wan Q, Wan L (2026) Machine learning–based prediction of rotator cuff tears using anatomical parameters: a retrospective cohort study. BMC Musculoskelet Disord 27:378. 10.1186/s12891-026-09765-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Kongmalai P, Apivatgaroon A, Chernchujit B (2017) Morphological classification of acromial spur: correlation between Rockwood tilt view and arthroscopic finding. SICOT J 3:4. 10.1051/sicotj/2017001 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Sakdapanichkul C, Chantarapitak N, Kasemwong N, Suwanalai J, Wimolsate T, Kongmalai P et al (2024) Transcending patient morphometry: acromiohumeral interval to glenoid ratio as a universal diagnostic tool for massive rotator cuff tears. Clin Orthop Surg 16:578–585. 10.4055/cios23261 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Saengpetch N, Bamrungchaowkasem J, Chitrapazt N, Kongmalai P (2025) Predicting bursal-side supraspinatus tendon tears with the acromioclavicular angle. Asia Pac J Sports Med Arthrosc Rehabil Technol 39:15–21. 10.1016/j.asmart.2024.11.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Feltri P, Monteleone AS, Marbach F, Filardo G, Candrian C (2023) Arthroscopic rotator cuff repair: patients with physically demanding work have significantly worse time to return to work, level of employment, and job loss. Knee Surg Sports Traumatol Arthrosc 31:153–160. 10.1007/s00167-022-07172-3 [DOI] [PubMed] [Google Scholar]
- 20.Huskisson EC (1974) Measurement of pain. Lancet 2:1127–1131 [DOI] [PubMed] [Google Scholar]
- 21.Jobe FW, Moynes DR (1982) Delineation of diagnostic criteria and a rehabilitation program for rotator cuff injuries. Am J Sports Med 10:336–339 [DOI] [PubMed] [Google Scholar]
- 22.Jobe FW, Jobe CM (1983) Painful athletic injuries of the shoulder. Clin Orthop Relat Res 173:117–124 [PubMed] [Google Scholar]
- 23.Gerber C, Krushell RJ (1991) Isolated rupture of the subscapularis tendon: results of operative repair. J Bone Joint Surg Am 73:389–394 [DOI] [PubMed] [Google Scholar]
- 24.Hertel R, Ballmer FT, Lombert SM, Gerber C (1996) Lag signs in the diagnosis of rotator cuff rupture. J Shoulder Elb Surg 5:307–313 [DOI] [PubMed] [Google Scholar]
- 25.Walch G, Boulahia A, Calderone S, Robinson AH (1998) The “dropping” and “hornblower’s” signs in evaluation of rotator cuff tears. J Bone Joint Surg Br 80:624–628 [DOI] [PubMed] [Google Scholar]
- 26.Aliprandi A, Messina C, Arrigoni P et al (2017) Reporting rotator cuff tears on magnetic resonance arthrography using the Snyder’s arthroscopic classification. World J Radiol 9:126–133 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Riley RD, Ensor J, Snell KIE et al (2020) Calculating the sample size required for developing a clinical prediction model. BMJ 368:m441. 10.1136/bmj.m441 [DOI] [PubMed] [Google Scholar]
- 28.Keener JD, Galatz LM, Teefey SA, Middleton WD, Steger-May K, Stobbs-Gucchi C, Patton R, Yamaguchi K (2015) A prospective evaluation of survivorship of asymptomatic degenerative rotator cuff tears. J Bone Joint Surg Am 97:89–98. 10.2106/JBJS.N.00099 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Gladstone JN, Bishop JY, Lo IK, Flatow EL (2007) Fatty infiltration and atrophy of the rotator cuff do not improve after rotator cuff repair and correlate with poor functional outcome. Am J Sports Med 35:719–728. 10.1177/0363546506297539 [DOI] [PubMed] [Google Scholar]
- 30.Murphy RJ, Daines MT, Carr AJ, Rees JL (2013) An independent learning method for orthopaedic surgeons performing shoulder ultrasound to identify full-thickness tears of the rotator cuff. J Bone Joint Surg Am 95:266–272 [DOI] [PubMed] [Google Scholar]
- 31.Longo UG, Franceschi F, Ruzzini L et al (2008) Histopathology of the supraspinatus tendon in rotator cuff tears. Am J Sports Med 36:533–538. 10.1177/0363546507308549 [DOI] [PubMed] [Google Scholar]
- 32.Seidler A, Romero Starke K, Freiberg A et al (2020) Dose-response relationship between physical workload and specific shoulder diseases: a systematic review with meta-analysis. Int J Environ Res Public Health 17:1243. 10.3390/ijerph17041243 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Karas V, Wang VM, Dhawan A, Cole BJ (2011) Biomechanical factors in rotator cuff pathology. Sports Med Arthrosc Rev 19:202–206. 10.1097/JSA.0b013e318225cc99 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request. Due to ethical restrictions imposed by the Institutional Review Board of the Third Hospital of Hebei Medical University (approval No. K2024-126-1) and patient confidentiality requirements, the raw data are not publicly available.