Persistent neuropathic-like pain phenotype after arthroscopic repair of massive rotator cuff tears: occurrence, associated factors, and a preoperative four-domain risk score

Abstract

Background

Persistent neuropathic-like pain may limit recovery after arthroscopic repair of massive rotator cuff tears, but its postoperative occurrence, associated factors, and preoperative risk stratification remain insufficiently characterized. This study estimated the 6-month occurrence proportion, examined associated factors, and developed an internally validated preoperative four-domain risk score.

Methods

This single-center retrospective cohort screened 418 consecutive adults treated between January 2019 and December 2024; 346 met the eligibility requirements, including ascertainable 6-month primary-outcome and 12-month follow-up phenotype data. The primary outcome required persistent or recurrent pain in the operated shoulder and related region at 6 months, a Douleur Neuropathique 4 (DN4) questionnaire score ≥ 4, a painDETECT score ≥ 13, at least 1 documented abnormality on a prespecified bedside sensory examination, and exclusion of a more plausible cause. painDETECT scores of 13–18 provided supportive evidence within the composite definition, for which all criteria were required, whereas scores ≥ 19 defined the strict sensitivity-analysis outcome. The 4 domains prespecified before outcome modeling were the central sensitization-related symptoms–poor sleep composite phenotype (Central Sensitization Inventory [CSI] ≥ 40 and Pittsburgh Sleep Quality Index [PSQI] > 5), preoperative subthreshold neuropathic-like symptoms, a magnetic resonance imaging (MRI)-based degeneration–tension phenotype, and limited socioeconomic and rehabilitation resources. Each domain contributed 1 point. A logistic regression model with the total score as the sole predictor was evaluated using 1,000 bootstrap resamples.

Results

The phenotype occurred in 46 of 346 patients (13.3%) at 6 months, compared with 16.2% at 3 months and 8.1% at 12 months. Among patients positive at 3 months, 73.2% remained positive at 6 months. All 4 domains were more frequent in outcome-positive patients (all P ≤ 0.001). The observed 6-month occurrence proportions across scores of 0–4 were 4.7%, 5.8%, 14.9%, 35.5%, and 60.0%, respectively (P for trend < 0.001). Each 1-point increase was associated with an odds ratio (OR) of 2.57 (95% confidence interval [CI], 1.96–3.37). The apparent area under the receiver operating characteristic curve (AUC) was 0.785 (95% CI, 0.705–0.864), and the optimism-corrected AUC was 0.784. The corrected calibration intercept and slope were 0.03 and 1.01, respectively, and the corrected Brier score was 0.092. At a threshold of ≥ 3 points, sensitivity was 56.5%, specificity was 90.0%, positive predictive value was 46.4%, and negative predictive value was 93.1%. Included and excluded patients differed little in age, sex, and tear size, whereas excluded patients had higher preoperative pain intensity (P = 0.024; absolute standardized difference [ASD] = 0.30).

Conclusions

The score showed reasonable internal discrimination and calibration and may inform preoperative risk stratification in comparable settings; external validation remains required. The composite outcome represents a clinically identifiable phenotype with potentially neuropathic, nociplastic, or mixed mechanisms, supporting mechanism-oriented assessment for postoperative pain management.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12891-026-10157-9.

Introduction

Rotator cuff tears are among the most common shoulder disorders. Massive rotator cuff tears, defined as a maximum tear diameter ≥ 5 cm or involvement of ≥ 2 tendons, are characterized by extensive anatomic disruption, marked tendon retraction, and severe fatty infiltration of the musculature, and represent one of the most structurally and functionally complex subtypes of rotator cuff injury [1]. With continued advances in arthroscopic techniques, arthroscopic repair has become a principal surgical option for massive rotator cuff tears, and most studies have reported improvements in postoperative structural integrity and functional scores [2]. Nevertheless, even after successful anatomic repair, some patients experience persistent pain that is difficult to relieve, with neuropathic-like features such as burning pain, electric shock-like pain, hypoesthesia, hyperalgesia, or brush-evoked pain, accompanied by limited functional recovery and reduced quality of life [3].

Chronic postsurgical pain (CPSP) is a widely recognized adverse outcome after orthopedic surgery and is generally defined as pain that persists beyond the normal tissue-healing period (≥ 3 months) after surgery [4]. Among CPSP subtypes, persistent pain with neuropathic-like features has received substantial attention because of its complex pathophysiology and the mechanism-specific nature of treatment [5]. Clinical grading of neuropathic pain generally requires the coexistence of positive sensory symptoms, abnormal sensory signs in a neuroanatomically plausible distribution, and evidence of a corresponding neural lesion or disease [6]. In the clinical assessment of pain after rotator cuff repair, the Douleur Neuropathique 4 (DN4) questionnaire and painDETECT are widely used to screen for neuropathic-like symptoms [7].

Previous studies have focused mainly on overall pain outcomes and functional recovery after rotator cuff repair and have identified several factors associated with postoperative pain, including psychological symptoms, sleep quality, tear morphology, and social support [8, 9]. However, several important limitations remain. First, most studies have used a single pain-intensity scale as the primary outcome and have not specifically defined or identified pain phenotypes with neuropathic-like features [10]. Second, studies of factors associated with pain after rotator cuff repair have generally included all tear types. Massive rotator cuff tears have inherent anatomic complexity, greater technical difficulty of repair, and a background of chronic degeneration; consequently, their pain manifestations and risk profile may differ from those of rotator cuff tears overall. Studies specifically addressing postoperative neuropathic-like pain phenotypes in this subtype remain limited [11]. Third, existing risk-stratification tools often rely on multivariable regression after univariable screening; when event numbers are limited, this approach may introduce variable-selection bias and coefficient instability, thereby compromising the internal consistency of risk stratification [12].

The rationale for focusing on massive rotator cuff tears is that patients with this subtype commonly have more severe preoperative tendon degeneration and fatty infiltration, greater repair tension and risk of structural failure, and a longer pain history. These clinical and imaging features indicate that this population differs from patients undergoing repair of small-to-medium tears in preoperative pain burden, degree of structural degeneration, and rehabilitation complexity; the risk and presentation of the postoperative neuropathic-like pain phenotype may therefore also differ [13]. Accordingly, systematic characterization of the occurrence of this postoperative phenotype and development of an applicable preoperative risk-stratification tool in the specific setting of massive rotator cuff tears have independent clinical and methodological value.

Accordingly, this study included consecutive patients undergoing arthroscopic repair of massive rotator cuff tears, used a multidimensional composite definition to identify a persistent neuropathic-like pain phenotype at 6 months postoperatively, and developed an equally weighted preoperative four-domain risk score based on 4 clinical risk domains prespecified before outcome modeling, primarily as a reference for preoperative risk stratification using internal single-center data.

Materials and methods

Study population

This single-center retrospective cohort study included consecutive patients who underwent arthroscopic repair of massive rotator cuff tears at the Department of Orthopedics, The Affiliated Hospital of Xuzhou Medical University, between January 2019 and December 2024. The study protocol was approved by the hospital ethics committee, which waived the requirement for written informed consent for the use of historical medical-record data.

The inclusion criteria were as follows: (1) age ≥ 18 years; (2) a full-thickness massive rotator cuff tear confirmed by both preoperative magnetic resonance imaging (MRI) and intraoperative exploration, with a maximum tear diameter ≥ 5 cm or complete involvement of at least 2 rotator cuff tendons; (3) arthroscopic rotator cuff repair; (4) an ascertainable primary outcome at 6 months postoperatively; (5) availability of the key preoperative, imaging, and pain-assessment data required for the primary analysis; and (6) an ascertainable supplemental phenotype at the 12-month follow-up. The 6-month postoperative assessment was the primary outcome time point, whereas the 12-month assessment was used for the secondary pain-trajectory analysis; ascertainability at both time points was a prespecified cohort-eligibility requirement.

The exclusion criteria were as follows: (1) previous ipsilateral shoulder surgery; (2) a definite preoperative neuropathic-like pain phenotype, defined as preoperative DN4 ≥ 4, painDETECT ≥ 13, and corresponding regional sensory abnormalities, or a previous definitive clinical diagnosis; (3) cervical radiculopathy, brachial plexus injury, or diabetic peripheral neuropathy that could explain neuropathic-like upper-limb pain; (4) rheumatoid arthritis, fibromyalgia syndrome, or complex regional pain syndrome; (5) continuous regular preoperative opioid use for ≥ 3 months or substance misuse; (6) postoperative infection, revision surgery, or imaging-confirmed severe retear requiring retreatment; and (7) inability to complete the pain scales or missing key follow-up/outcome data that prevented ascertainment of the 6-month primary outcome or 12-month supplemental phenotype.

A massive rotator cuff tear was defined as a full-thickness rotator cuff tear supported by both preoperative MRI and intraoperative exploration and meeting either of the following criteria: a maximum anteroposterior or mediolateral diameter ≥ 5 cm, or complete involvement of at least 2 rotator cuff tendons. This composite definition incorporates both tear size and the number of involved tendons to reflect the anatomic complexity of massive tears [14, 15].

Surgical technique and rehabilitation protocol

All procedures were performed by the same shoulder arthroscopy team. Patients were placed in the lateral decubitus or beach-chair position under general anesthesia combined with an interscalene brachial plexus block, and standard posterior, lateral, and anterior portals were established. Tear size, degree of retraction, tendon quality, and repair tension were assessed intraoperatively. Single-row, double-row, margin-convergence, or partial repair was selected according to tear morphology and tissue condition. Concomitant lesions were treated as appropriate, including acromioplasty, tenodesis or tenotomy of the long head of the biceps tendon, and acromioclavicular joint resection. Postoperatively, the affected shoulder was immobilized in an abduction brace for 4–6 weeks, during which passive motion and pendulum exercises were performed; active-assisted exercises began at 6 weeks, and resisted strengthening began at 12 weeks. The standard postoperative analgesic regimen included nonsteroidal anti-inflammatory drugs, with short-term weak opioids used when necessary. Physicians and rehabilitation therapists individualized the rehabilitation protocol according to repair quality, pain severity, and shoulder range of motion.

Data collection and variable definitions

Data were collected in 4 categories. First, demographic and socioeconomic variables included age, sex, body mass index (BMI), involvement of the dominant arm, educational level, occupation type, preoperatively available medical or commercial insurance coverage and reimbursement for rehabilitation expenses, distance from the residence to a rehabilitation facility, and family rehabilitation support [16]. Second, pain, functional, psychological, and sleep-related variables included the preoperative visual analog scale (VAS) score [10], night pain, pain duration, the American Shoulder and Elbow Surgeons (ASES) score [17], Constant-Murley score [18], Central Sensitization Inventory (CSI) [19], Pittsburgh Sleep Quality Index (PSQI) [20], Pain Catastrophizing Scale (PCS) [21], preoperative DN4 questionnaire [22], and painDETECT [23]. Third, imaging variables included maximum tear diameter, number of involved tendons, Patte grade, Goutallier grade, supraspinatus occupation ratio, and acromiohumeral interval. Fourth, perioperative variables included operative time, repair technique, number of anchors, peak VAS score within 48 h postoperatively, duration of opioid use, actual out-of-pocket expenses during the first 3 postoperative months, and rehabilitation adherence during postoperative weeks 0–12.

Night pain was defined as difficulty falling asleep, nocturnal awakening, or the need to change sleeping position because of pain in the affected shoulder during the 4 weeks before surgery, occurring on ≥ 3 nights per week. Pain duration was defined as the number of months from the patient-reported first occurrence of pain in the affected shoulder to the date of surgery; durations < 1 month were recorded as 0.5 months. The number of involved tendons was determined jointly from preoperative MRI and intraoperative exploration; when MRI and intraoperative records were inconsistent, intraoperative findings were used as the reference. Night pain, pain duration, and number of involved tendons were reported as preoperative descriptive variables and were not included in the four-domain risk score.

The CSI was used to assess the burden of central sensitization-related symptoms [19]. Total CSI scores range from 0 to 100, with higher scores indicating a greater symptom burden; CSI ≥ 40 was used as an operational threshold for elevated symptom burden [24]. PSQI > 5 defined poor sleep quality [20]. CSI ≥ 40 together with PSQI > 5 defined the central sensitization-related symptoms–poor sleep composite phenotype, which served as the first component domain of the four-domain risk score. PCS ≥ 30 defined high pain catastrophizing; this threshold was prespecified on the basis of previous measurement evidence [25]. The PCS was used in descriptive and sensitivity analyses and was not included in the four-domain score.

The socioeconomic and rehabilitation resource index comprised 5 preoperatively available items: education at high-school level or below; manual labor or unemployment/retirement; insufficient insurance or rehabilitation-expense coverage; a one-way distance of ≥ 30 km to a rehabilitation facility; and absence of a regular caregiver or family rehabilitation support. Each item contributed 1 point, yielding a total score of 0–5; a total score ≥ 2 defined limited socioeconomic and rehabilitation resources [26]. This domain entered the risk score as a prespecified multidimensional construct. Its component items were not removed, reweighted, or threshold-optimized according to their individual between-group P values in this cohort; item-level comparisons were used to describe heterogeneity in between-group differentiation among the components. Insufficient insurance or rehabilitation-expense coverage was determined from preoperative registration, cost estimates, and the reimbursement scope of planned rehabilitation. Actual postoperative out-of-pocket expenses were used only as an exploratory perioperative variable.

Preoperative subthreshold neuropathic-like symptoms were defined as a preoperative DN4 score of 2–3 or painDETECT score of 13–18 without meeting the composite exclusion criteria for a definite preoperative neuropathic-like pain phenotype. This domain served as the second component of the four-domain risk score [22, 23].

The MRI-based degeneration–tension composite phenotype integrated tear extent, tendon retraction, fatty infiltration, muscle atrophy, and the acromiohumeral interval. Positivity was defined as meeting at least 2 of the following 5 criteria: (1) maximum tear diameter ≥ 6 cm; (2) Patte grade III; (3) highest Goutallier grade of the primarily involved muscle ≥ 3; (4) supraspinatus occupation ratio < 0.50; and (5) acromiohumeral interval < 7 mm. This composite phenotype served as the third risk domain and was used to reduce duplicate scoring and collinearity among related imaging indicators [27].

Rehabilitation adherence during postoperative weeks 0–12 was assessed from outpatient or rehabilitation attendance, home-exercise records, and therapist documentation [28]. Good adherence was defined as completion of ≥ 80% of planned therapy or training with no interruption lasting 2 consecutive weeks; moderate adherence was defined as completion of 50%–79%; and poor adherence was defined as completion of < 50% or an interruption lasting ≥ 2 consecutive weeks for nonmedical reasons. This variable was a postoperative exposure and was used only in exploratory perioperative analyses.

Outcome definition

The primary outcome was the presence of a persistent neuropathic-like pain phenotype at 6 months postoperatively, jointly defined by the pain trajectory, DN4, painDETECT, bedside sensory examination documented according to prespecified items, and exclusionary clinical judgment [29–31]. All of the following criteria were required: (1) pain in the operated shoulder and related region that was new or markedly worse than before surgery and persisted or recurred through 6 months postoperatively; (2) DN4 ≥ 4; (3) painDETECT ≥ 13, with scores of 13–18 considered uncertain and used only as supportive information within the composite definition, and scores ≥ 19 considered highly supportive; (4) at least 1 abnormality on the prespecified bedside sensory examination, including tactile hypoesthesia, abnormal pinprick sensation, brush-evoked pain, or localized hyperalgesia; and (5) exclusion of more plausible explanations, including infection, retear requiring retreatment, marked shoulder stiffness, complex regional pain syndrome, cervical radiculopathy, and diabetic peripheral neuropathy.

Sensory findings were considered only when clearly documented in the medical-record template and comparable with the corresponding contralateral site or an adjacent nonpainful region. Unclear or unverifiable records were treated as missing and were not classified as normal or abnormal. If DN4 < 4 or painDETECT < 13, the phenotype at that time point was classified as negative because at least 1 required questionnaire criterion was not met. If both questionnaire thresholds were met but the sensory-examination record was unverifiable, the phenotype at that time point was classified as indeterminate. Patients with an indeterminate 6-month primary outcome or 12-month supplemental phenotype were excluded according to the cohort-eligibility criteria; an indeterminate 3-month phenotype was recorded as missing supplemental time-point data. This composite definition identifies a clinical phenotype; assignment of a pain mechanism requires integration of pain distribution, a complete neurologic examination, evaluation of structural causes, and quantitative sensory or neurophysiologic assessment when necessary.

The 3-month postoperative phenotype was used as a supplemental time point, and the 12-month phenotype was used for the secondary pain-trajectory analysis. In addition to all other primary-outcome criteria, the strict outcome definition required DN4 ≥ 4 and painDETECT ≥ 19.

The 6-month postoperative assessment was selected as the primary time point because 3 months is the minimum temporal criterion for CPSP, and patients undergoing repair of massive tears remain in transition to active motion and strengthening at that stage; by 6 months, confounding related to early rehabilitation load is reduced [32, 33]. The 12-month assessment was used to characterize a later trajectory but may be affected by subsequent structural failure, secondary stiffness, other shoulder or cervical disorders, retreatment, rehabilitation resources, and medication adjustments [34].

Statistical analysis

Statistical analyses were performed using IBM SPSS Statistics version 26.0 (IBM Corp., Armonk, NY, USA) and R version 4.2.0. Continuous variables were summarized as mean ± standard deviation or median (interquartile range), as appropriate. Continuous variables with complete data were compared using Welch’s independent-samples t test or the Mann–Whitney U test. Categorical variables were summarized as frequencies and percentages and compared using the Pearson χ² test or Fisher’s exact test. All tests were two-sided, with α = 0.05.

To assess potential selection differences introduced by the screening process, patients included in the final cohort were compared with all excluded patients with respect to age, sex, maximum tear diameter, and preoperative VAS score. Continuous variables were compared using Welch’s independent-samples t test, whereas sex was compared using the Pearson χ² test; absolute standardized differences (ASDs) were also reported to describe the magnitude of differences. This comparison was not used for risk-domain selection or model estimation. Patients excluded for eligibility/clinical reasons and those with indeterminate follow-up/outcome data were described separately; no significance test was performed for the latter subgroup of only 6 patients.

Before outcome modeling, 4 scoring domains were prespecified on the basis of clinical and literature evidence: the central sensitization-related symptoms–poor sleep composite phenotype, preoperative subthreshold neuropathic-like symptoms, the MRI-based degeneration–tension composite phenotype, and limited socioeconomic and rehabilitation resources. Each domain was assigned 0 or 1 point, yielding an equally weighted total score of 0–4. Equal weighting was intended to improve simplicity, interpretability, and reproducibility and was not empirically optimized according to univariable P values, observed odds ratios, regression coefficients, or optimal cutoffs in this cohort [35].

The primary model used the total score as the sole continuous predictor in logistic regression and reported the risk equation, odds ratio (OR) per 1-point increase, and 95% confidence interval (CI). Observed occurrence proportions were reported at each score level and in low-, intermediate-, and high-risk strata, defined as 0–1, 2, and 3–4 points, respectively. Model performance included the area under the receiver operating characteristic curve (AUC), calibration intercept, calibration slope, Brier score, and net benefit from decision curve analysis (DCA). Optimism was estimated using 1,000 bootstrap resamples, with apparent and optimism-corrected performance reported separately. The 95% CI for the apparent AUC was calculated using the nonparametric DeLong method. The Hosmer–Lemeshow test was used only as a supplementary calibration indicator.

The primary outcome and all 4 binary scoring domains were ascertainable in the final analysis sample. Missing values in continuous scale totals and other nonscore variables were handled using multiple imputation by chained equations, generating 20 imputed datasets. The imputation model included the primary outcome, the 4 scoring domains, and preoperative auxiliary variables associated with missingness or the variable being imputed; variables available only after the primary outcome were not used as imputation predictors. Between-group comparisons of continuous variables used Wald tests pooled according to Rubin’s rules; descriptive means and standard deviations were estimated within and pooled across the 20 imputed datasets. Bedside sensory signs were not imputed. Complete-case analysis was used to assess stability.

Collinearity was assessed using correlation coefficients and variance inflation factors (VIFs). Sensitivity analyses included the strict outcome definition, postoperative DN4 and painDETECT continuous scores, a strict massive-tear definition, complete-case analysis, decomposed four-domain Firth penalized-likelihood logistic regression, a traditional four-variable alternative model, and 2 prespecified exploratory interactions. Firth regression reported penalized-likelihood ORs, profile-likelihood 95% CIs, and two-sided penalized-likelihood-ratio P values; all other regression effect estimates were reported with 95% CIs and two-sided P values.

Results

Patient screening, differences between included and excluded patients, and phenotype transitions

Between January 2019 and December 2024, 418 patients entered initial screening. According to the prespecified criteria, 72 patients were excluded: previous ipsilateral shoulder surgery (n = 14); definite preoperative neuropathic-like pain phenotype (n = 11); cervical radiculopathy, brachial plexus injury, or diabetic peripheral neuropathy (n = 19); rheumatoid arthritis, fibromyalgia syndrome, or complex regional pain syndrome (n = 7); continuous regular preoperative opioid use or substance misuse (n = 6); postoperative infection, revision surgery, or severe retear requiring retreatment (n = 9); and indeterminate follow-up/outcome data (n = 6). Of the latter 6 patients, 4 lacked ascertainable 12-month phenotype data, and 2 met both the DN4 and painDETECT thresholds at 6 months but lacked a verifiable bedside sensory-examination record. Ultimately, 346 patients entered the primary analysis, representing 82.8% of the initially screened population (Fig. 1).

Fig. 1.

Patient screening and analytic-cohort flow diagram. The diagram shows the 418 initially screened patients, each reason for exclusion, the composition of the 6 patients with indeterminate follow-up/outcome data, and the final primary-analysis cohort of 346 patients. n, number of patients; DN4, Douleur Neuropathique 4 questionnaire

Included patients and all excluded patients differed little in age, sex, and maximum tear diameter (ASDs, 0.08–0.13; all P > 0.05), whereas excluded patients had a higher preoperative VAS score (ASD = 0.30; P = 0.024). The eligibility/clinical-exclusion subgroup and the subgroup with indeterminate follow-up/outcome data are presented separately for descriptive purposes (Supplementary Table 1).

The occurrence proportion of the primary phenotype at 6 months was 13.3% (46/346), with a higher proportion at 3 months and a lower proportion at 12 months; the strict outcome occurred less frequently than the primary composite outcome. Most patients positive at 3 months remained positive at 6 months, whereas the remainder became negative; only a few patients negative at 3 months became positive at 6 months. More than half of patients positive at 6 months remained positive at 12 months, while the remainder improved; only isolated patients negative at 6 months newly or again met the definition at 12 months (Table 1).

Table 1.

Overview of the Primary Outcome and Pain-Phenotype Transitions (n = 346)

Outcome/transition	Cases	Occurrence/transition proportion	95% CI	Interpretation
Phenotype at 3 months	56/346	16.2%	12.7%–20.4%	Supplemental time point
Primary outcome at 6 months	46/346	13.3%	10.1%–17.3%	Primary time point
Strict outcome at 6 months	31/346	9.0%	6.4%–12.4%	Strict definition with painDETECT ≥ 19
Phenotype at 12 months	28/346	8.1%	5.7%–11.4%	Secondary follow-up time point
Positive at 3 months → positive at 6 months	41/56	73.2%	60.4%–83.0%	Persistently positive
Positive at 3 months → negative at 6 months	15/56	26.8%	17.0%–39.6%	No longer met the definition
Negative at 3 months → positive at 6 months	5/290	1.7%	0.7%–4.0%	Delayed fulfillment of the definition
Positive at 6 months → positive at 12 months	26/46	56.5%	42.2%–69.8%	Persistent through 12 months
Positive at 6 months → negative at 12 months	20/46	43.5%	30.2%–57.8%	Improved by 12 months
Negative at 6 months → positive at 12 months	2/300	0.7%	0.2%–2.4%	Newly or again met the definition

n number of patients, CI confidence interval. Occurrence/transition proportions were calculated by dividing each numerator by its corresponding denominator, and 95% CIs were calculated using the Wilson binomial method. DN4, Douleur Neuropathique 4 questionnaire. The strict 6-month outcome required DN4 ≥ 4, painDETECT ≥ 19, at least 1 abnormality on the bedside sensory examination documented according to prespecified items, and exclusion of a more plausible cause

Verifiable bedside sensory-examination records at 6 months were available for 94.8% of patients. All patients positive for the primary outcome had verifiable records, whereas records were unverifiable for 18 patients in the primary-outcome-negative group. Of these 18 patients, 15 met neither questionnaire threshold and 3 met only 1 threshold; none met all required components of the composite outcome. Because sensory-record availability was structurally related to the positive-outcome definition, these data are reported descriptively only (Supplementary Table 2).

At initial extraction, varying degrees of missingness were present for the CSI, PSQI, night pain, pain duration, central sensitization-related symptoms–poor sleep composite phenotype, PCS, socioeconomic and rehabilitation resource index, preoperative DN4/painDETECT, Goutallier grade, and supraspinatus occupation ratio; the number of involved tendons, four-domain total score, and 6-month primary outcome had no missing data. After supplemental verification, night pain, number of involved tendons, the central sensitization-related symptoms–poor sleep composite phenotype, socioeconomic and rehabilitation resource index, preoperative DN4/painDETECT, Goutallier grade, four-domain total score, and primary outcome were all ascertainable. Small numbers of values remained missing for the continuous CSI, PSQI, pain-duration, PCS, and supraspinatus occupation-ratio measures and were handled by multiple imputation (Supplementary Table 3).

Preoperative baseline, imaging, and biopsychosocial characteristics

The primary-outcome and non-primary-outcome groups did not differ significantly in age, sex, BMI, dominant-arm involvement, education at high-school level or below, manual labor or unemployment/retirement, hypertension, or diabetes (all P > 0.05); a history of anxiety or depression was more common in the primary-outcome group (P = 0.002).

The primary-outcome group had a higher preoperative VAS score, more frequent night pain, longer pain duration, and lower ASES and Constant-Murley scores (all P ≤ 0.002). The primary-outcome group also had higher CSI scores and a greater proportion with CSI ≥ 40, higher PSQI scores and a greater proportion with PSQI > 5, higher PCS scores and a greater proportion with PCS ≥ 30, and a more frequent central sensitization-related symptoms–poor sleep composite phenotype (all P < 0.001).

The socioeconomic and rehabilitation resource index and the proportion with limited resources were higher in the primary-outcome group (all P ≤ 0.001). Insufficient insurance or rehabilitation-expense coverage, a distance of ≥ 30 km to a rehabilitation facility, and absence of a regular caregiver or family rehabilitation support were more common (all P ≤ 0.004), whereas between-group differences in educational level and occupation type were not statistically significant (all P > 0.05).

The primary-outcome group had higher preoperative DN4 and painDETECT scores and more frequent preoperative subthreshold neuropathic-like symptoms (all P < 0.001). The group also had a larger maximum tear diameter, more involved tendons, and higher frequencies of maximum tear diameter ≥ 6 cm, Patte grade III, Goutallier grade ≥ 3, supraspinatus occupation ratio < 0.50, acromiohumeral interval < 7 mm, and the MRI-based degeneration–tension composite phenotype, together with a lower supraspinatus occupation ratio (all P ≤ 0.010) (Table 2).

Table 2.

Comparison of Preoperative Baseline, Imaging, and Biopsychosocial Variables (n = 346)

Variable	Primary-outcome group, n = 46	Non-primary-outcome group, n = 300	Statistic	P value
Age (years)	60.1 ± 8.9	58.6 ± 8.7	Welch t = 1.07	0.290
Male sex	25 (54.3)	172 (57.3)	χ² = 0.14	0.703
BMI (kg/m²)	25.7 ± 3.3	25.3 ± 3.1	Welch t = 0.77	0.444
Dominant arm involved	31 (67.4)	188 (62.7)	χ² = 0.38	0.536
High-school education or below	31 (67.4)	168 (56.0)	χ² = 2.12	0.146
Manual labor or unemployed/retired	23 (50.0)	120 (40.0)	χ² = 1.64	0.200
Insufficient insurance or rehabilitation-expense coverage	20 (43.5)	70 (23.3)	χ² = 8.41	0.004
Distance to rehabilitation facility ≥ 30 km	17 (37.0)	54 (18.0)	χ² = 8.79	0.003
No regular caregiver/family support	15 (32.6)	38 (12.7)	χ² = 12.23	< 0.001
Hypertension	18 (39.1)	95 (31.7)	χ² = 1.01	0.315
Diabetes	7 (15.2)	38 (12.7)	χ² = 0.23	0.632
History of anxiety/depression	15 (32.6)	42 (14.0)	χ² = 10.04	0.002
Preoperative VAS score	7.1 ± 1.4	6.3 ± 1.6	Welch t = 3.54	< 0.001
Night pain	38 (82.6)	176 (58.7)	χ² = 9.69	0.002
Pain duration (months)	16.0 ± 7.6	12.2 ± 6.4	Pooled Wald test according to Rubin’s rules	0.002
ASES score	38.9 ± 12.1	45.5 ± 14.3	Welch t = − 3.36	0.001
Constant-Murley score	42.0 ± 10.7	48.8 ± 12.1	Welch t = − 3.94	< 0.001
CSI score	44.3 ± 13.2	29.1 ± 12.8	Pooled Wald test according to Rubin’s rules	< 0.001
CSI ≥ 40	32 (69.6)	82 (27.3)	χ² = 32.20	< 0.001
PSQI score	8.8 ± 3.2	5.7 ± 3.0	Pooled Wald test according to Rubin’s rules	< 0.001
PSQI > 5	34 (73.9)	112 (37.3)	χ² = 21.88	< 0.001
Central sensitization-related symptoms–poor sleep composite phenotype	28 (60.9)	57 (19.0)	χ² = 37.73	< 0.001
PCS score	31.2 ± 8.6	22.7 ± 9.0	Pooled Wald test according to Rubin’s rules	< 0.001
PCS ≥ 30	28 (60.9)	73 (24.3)	χ² = 25.76	< 0.001
Socioeconomic and rehabilitation resource index	2.3 ± 1.2	1.5 ± 1.2	Welch t = 4.21	< 0.001
Limited socioeconomic and rehabilitation resources	26 (56.5)	92 (30.7)	χ² = 11.86	0.001
Preoperative DN4 score	2.6 ± 0.7	1.0 ± 0.9	Welch t = 13.85	< 0.001
Preoperative painDETECT score	14.1 ± 3.0	7.3 ± 4.0	Welch t = 13.63	< 0.001
Preoperative subthreshold neuropathic-like symptoms	30 (65.2)	64 (21.3)	χ² = 38.82	< 0.001
Maximum tear diameter (cm)	5.95 ± 0.65	5.62 ± 0.45	Welch t = 3.32	0.002
Number of involved tendons	2.7 ± 0.6	2.4 ± 0.5	Welch t = 3.22	0.002
Maximum tear diameter ≥ 6 cm	25 (54.3)	92 (30.7)	χ² = 9.99	0.002
Patte grade III	29 (63.0)	105 (35.0)	χ² = 13.22	< 0.001
Goutallier grade ≥ 3	31 (67.4)	110 (36.7)	χ² = 15.59	< 0.001
Supraspinatus occupation ratio	0.47 ± 0.12	0.55 ± 0.13	Pooled Wald test according to Rubin’s rules	< 0.001
Supraspinatus occupation ratio < 0.50	28 (60.9)	91 (30.3)	χ² = 16.48	< 0.001
Acromiohumeral interval < 7 mm	27 (58.7)	116 (38.7)	χ² = 6.60	0.010
MRI-based degeneration–tension composite phenotype	31 (67.4)	96 (32.0)	χ² = 21.50	< 0.001

pn number of patients. Continuous variables are presented as mean ± standard deviation (SD), and categorical variables as number (percentage). Welch t denotes the test statistic from Welch’s independent-samples t test; χ² denotes the Pearson chi-square statistic; pooled Wald test according to Rubin’s rules denotes a Wald test based on estimates obtained in each of 20 multiply imputed datasets and pooled according to Rubin’s rules; P denotes a two-sided P value. Between-group P values were used only for descriptive comparison and not for selection of scoring domains

BMI  body mass index, VAS  visual analog scale, ASES  American Shoulder and Elbow Surgeons score, CSI  Central Sensitization Inventory, PSQI  Pittsburgh Sleep Quality Index, PCS  Pain Catastrophizing Scale, DN4  Douleur Neuropathique 4 questionnaire, MRI  magnetic resonance imaging

Exploratory perioperative comparisons showed that the primary-outcome group had longer operative time, more anchors, a higher peak VAS score within 48 h postoperatively, longer opioid use, greater related out-of-pocket expenses during the first 3 postoperative months, and more frequent poor early rehabilitation adherence (all P < 0.05). The proportions undergoing double-row repair and partial repair did not differ significantly (all P > 0.05) (Supplementary Table 4).

Distribution and risk gradient of the preoperative four-domain risk score

As the four-domain total score increased, the observed occurrence proportion of the 6-month primary outcome increased stepwise: it was low at scores of 0 and 1, intermediate at a score of 2, and markedly higher at scores of 3 and 4 (P for trend < 0.001) (Table 3; Fig. 2).

Table 3.

Four-domain risk score and observed/predicted risk

Score	Patients	Events	Observed occurrence (95% CI)	Model-predicted probability/statistic
0 points	106	5	4.7% (95% CI, 2.0%–10.6%)	3.0%
1 point	137	8	5.8% (95% CI, 3.0%–11.1%)	7.3%
2 points	47	7	14.9% (95% CI, 7.4%–27.7%)	16.8%
3 points	31	11	35.5% (95% CI, 21.1%–53.1%)	34.3%
4 points	25	15	60.0% (95% CI, 40.7%–76.6%)	57.3%
Per 1-point increase	346	46	OR = 2.57 (95% CI, 1.96–3.37)	P < 0.001

n number of patients; CI, confidence interval; OR, odds ratio. logit denotes the log-odds of predicted probability Pr(Y = 1), and Y = 1 denotes occurrence of the 6-month primary outcome. Model-predicted probabilities were calculated as logit[Pr(Y = 1)] = − 3.488 + 0.946 × total score. The 95% CIs for observed occurrence proportions were calculated using the Wilson binomial method. P denotes a two-sided P value

Fig. 2.

Observed occurrence proportion of persistent neuropathic-like pain phenotype at 6 months according to the four-domain risk score. Bar height represents the observed occurrence proportion at each score level, labels show events/patients, and error bars represent Wilson 95% confidence intervals (CIs)

With the total score as the sole continuous predictor, the logistic regression equation was logit[Pr(Y = 1)] = − 3.488 + 0.946 × total score, where logit denotes the log-odds of outcome probability Pr(Y = 1) and Y = 1 denotes occurrence of the 6-month primary outcome. Each 1-point increase in score was associated with higher odds of the primary outcome (OR = 2.57; 95% CI, 1.96–3.37; P < 0.001). Model-predicted probabilities were generally consistent with observed occurrence proportions across score levels (Table 3).

Discrimination, calibration, and clinical utility of the risk score

VIFs for the 4 scoring domains ranged from 1.12 to 1.48 and did not reach the prespecified collinearity threshold. The apparent AUC was 0.785 (95% CI, 0.705–0.864), and the bootstrap optimism-corrected AUC was 0.784. The bootstrap-corrected calibration intercept was 0.03 (95% CI, − 0.45 to 0.73), and the calibration slope was 1.01 (95% CI, 0.74–1.36); the apparent Brier score was 0.091 and the corrected Brier score was 0.092. The Hosmer–Lemeshow test did not indicate appreciable lack of fit (P = 0.619) (Table 4; Fig. 3).

Table 4.

Performance and risk stratification of the four-domain risk score

Item	Value/definition	Statistic or interval	Interpretation
Risk equation	logit[Pr(Y = 1)] = − 3.488 + 0.946 × total score	—	Reproducible risk-probability calculation
Apparent AUC	0.785	95% CI, 0.705–0.864	Discrimination
Bootstrap-corrected AUC	0.784	1,000 resamples	Optimism correction
Bootstrap-corrected calibration intercept	0.03	95% CI, − 0.45 to 0.73	Close to 0
Bootstrap-corrected calibration slope	1.01	95% CI, 0.74–1.36	Close to 1
Apparent Brier score	0.091	—	Overall error of the original model
Bootstrap-corrected Brier score	0.092	1,000 resamples	Optimism-corrected overall error
Hosmer–Lemeshow test	χ² = 1.78; df = 3	P = 0.619	Supplementary calibration indicator
DCA net-benefit range	Threshold probability, 8%–40%	Higher than treat-all and treat-none	Exploratory clinical utility
Low risk	0–1 points; 243 patients; 13 events	5.3% (95% CI, 3.2%–8.9%)	Within-stratum NPV = 94.7%
Intermediate risk	2 points; 47 patients; 7 events	14.9% (95% CI, 7.4%–27.7%)	Intermediate-risk stratum
High risk	3–4 points; 56 patients; 26 events	46.4% (95% CI, 34.0%–59.3%)	High-risk enrichment
Threshold ≥ 3 points	TP = 26; FP = 30; FN = 20; TN = 270	Sensitivity 56.5%; specificity 90.0%	PPV = 46.4%; NPV = 93.1%

n number of patients, AUC  area under the receiver operating characteristic curve, CI  confidence interval; P, two-sided P value; bootstrap, internal validation by resampling with replacement. The Brier score is the mean squared error between predicted probabilities and binary outcomes, with lower values indicating lower overall prediction error. Hosmer–Lemeshow χ² is the grouped calibration goodness-of-fit statistic; df, degrees of freedom; DCA, decision curve analysis. TP, FP, FN, and TN denote true positive, false positive, false negative, and true negative, respectively. Sensitivity = TP/(TP + FN); specificity = TN/(TN + FP); PPV positive predictive value, NPV negative predictive value

Fig. 3.

Calibration plot for the four-domain risk score. The horizontal axis shows probabilities predicted by the logistic model, and the vertical axis shows observed occurrence proportions at each score level. The dashed line indicates ideal calibration, the connected line shows observed values for score levels 0–4, and error bars represent Wilson 95% confidence intervals (CIs)

Decision curve analysis showed that, across the prespecified threshold-probability range of 8%–40%, the score strategy had greater net benefit than the treat-all and treat-none strategies. This finding represents exploratory evidence of clinical utility within the development cohort (Fig. 4).

Fig. 4.

Decision curve for the four-domain risk score. The horizontal axis shows threshold probability and the vertical axis shows net benefit. The curves compare the score strategy with treat-all and treat-none strategies; the shaded area indicates the prespecified threshold-probability range of 8%–40%

Observed occurrence proportions increased across the low-, intermediate-, and high-risk strata (Table 4; Fig. 5). At the high-risk threshold of ≥ 3 points, sensitivity was 56.5%, specificity was 90.0%, positive predictive value was 46.4%, and negative predictive value was 93.1%. For the low-risk stratum of 0–1 point, the within-stratum negative predictive value was 94.7% (Table 4).

Fig. 5.

Observed occurrence proportion of persistent neuropathic-like pain phenotype at 6 months across low-, intermediate-, and high-risk strata. Low risk was defined as 0–1 point, intermediate risk as 2 points, and high risk as 3–4 points. Labels show events/patients, and error bars represent Wilson 95% confidence intervals (CIs)

Sensitivity analyses

In decomposed four-domain Firth penalized-likelihood logistic regression, the 95% CIs for the central sensitization-related symptoms–poor sleep composite phenotype, preoperative subthreshold neuropathic-like symptoms, and MRI-based degeneration–tension composite phenotype excluded 1. The estimate for limited socioeconomic and rehabilitation resources was directionally consistent, but its 95% CI included 1 (Table 5).

Table 5.

Sensitivity analyses

Analysis	Sample/outcome	Main result	Conclusion
Decomposed four-domain Firth regression	346 patients; 46 events	Central sensitization-related symptoms–poor sleep: OR = 2.73 (95% CI, 1.32–5.64; P = 0.007); subthreshold symptoms: OR = 2.95 (1.45–6.02; P = 0.003); MRI composite: OR = 2.06 (1.01–4.21; P = 0.047); limited resources: OR = 1.70 (0.82–3.52; P = 0.153)	The CIs for the first 3 domains excluded 1; the resource-domain estimate was directionally consistent but imprecise
Strict outcome definition	31/346 (9.0%)	OR per 1-point increase = 3.08 (95% CI, 2.20–4.31; P < 0.001); corrected AUC = 0.779	Direction consistent
DN4 continuous score	6-month postoperative DN4 total score	Each 1-point increase in score was associated with a 0.54-point higher DN4 score (95% CI, 0.38–0.70; P < 0.001)	Consistent continuous outcome
painDETECT continuous score	6-month postoperative painDETECT total score	Each 1-point increase in score was associated with a 1.72-point higher painDETECT score (95% CI, 1.20–2.24; P < 0.001)	Consistent continuous outcome
Strict massive-tear definition	301 patients; 42 events (14.0%)	Corrected AUC = 0.782; occurrence in the high-risk group = 38.7%	Results stable
Complete-case analysis	319 patients; 43 events (13.5%)	Apparent AUC = 0.793; corrected AUC = 0.786; Brier = 0.095	Consistent with multiple-imputation analysis
Traditional four-variable alternative model	History of anxiety/depression, preoperative VAS, Patte grade III, Goutallier grade ≥ 3	Apparent AUC = 0.746; corrected AUC = 0.721	Lower than the four-domain score
Prespecified interactions (exploratory)	Central sensitization-related symptoms × poor sleep; MRI composite × subthreshold symptoms	P = 0.176; P = 0.214	Neither met the prespecified P < 0.10 criterion

OR odds ratio, CI confidence interval, AUC area under the receiver operating characteristic curve, DN4 Douleur Neuropathique 4 questionnaire, MRI magnetic resonance imaging, VAS visual analog scale. The Brier score is the mean squared error between predicted probabilities and binary outcomes. Firth regression used penalized-likelihood estimation and reports profile-likelihood 95% CIs and two-sided penalized-likelihood-ratio P values. Continuous-outcome analyses report the regression coefficient and 95% CI corresponding to each 1-point increase in total score. All other regression results report two-sided P values

Under the strict outcome definition, the direction of the association per 1-point increase and the corrected AUC were consistent with the primary analysis. When continuous DN4 and painDETECT scores at 6 months were used as outcomes, higher total scores were associated with higher scores on both instruments (all P < 0.001). The strict massive-tear definition, complete-case analysis, and traditional four-variable alternative model did not change the direction of the main conclusions; discrimination of the four-domain score exceeded that of the traditional alternative model. Neither prespecified exploratory interaction met the P < 0.10 criterion (Table 5).

Discussion

This study focused on the persistent neuropathic-like pain phenotype after arthroscopic repair of massive rotator cuff tears. The occurrence proportion was 13.3% at 6 months, higher at 3 months, and lower at 12 months; all 4 preoperative risk domains showed between-group associations with the primary outcome; and the equally weighted total score demonstrated a clear risk gradient with reasonable internal discrimination and calibration. Estimates of CPSP vary with surgical procedure, outcome definition, and observation time, and the present findings should be interpreted within the specific context of massive rotator cuff tears and the composite phenotype definition [36]. Most patients positive at 3 months remained positive at 6 months, whereas the overall occurrence proportion declined by 12 months, suggesting that the phenotype has both persistent and reversible components. Functional recovery after rotator cuff repair may continue into later stages, and observation at multiple time points helps characterize the clinical trajectory [37].

The central sensitization-related symptoms–poor sleep composite phenotype was among the preoperative domains with the most pronounced between-group differences. CSI scores, the proportion with CSI ≥ 40, PSQI scores, the proportion with PSQI > 5, and positivity for the composite phenotype were all higher in the primary-outcome group. A greater burden of central sensitization-related symptoms is associated with more complex pain presentations in patients with shoulder disorders, supporting use of the CSI as risk-phenotype information rather than as a stand-alone indicator of structural injury [38]. Poor preoperative sleep has been associated with the postoperative pain burden after rotator cuff repair, and combined CSI and PSQI information may provide complementary information on symptom burden and sleep [39]. These findings represent risk-stratification associations; the temporal sequence and potential mediating roles of central sensitization-related symptoms and poor sleep in postoperative pain persistence require prospective evaluation.

Pain-mechanism heterogeneity is central to interpretation of the primary outcome. DN4 and painDETECT can identify neuropathic-like features such as burning, electric shock-like pain, hyperalgesia, and sensory abnormalities, but they are symptom-screening instruments with limited mechanism specificity [40]. We treated painDETECT scores of 13–18 as supportive information within the composite definition and used scores ≥ 19 to construct the strict sensitivity-analysis outcome. Potential mechanisms underlying this clinical phenotype include postoperative neuropathic pain, nociplastic pain, and a mixed phenotype in which both overlap. The higher CSI scores in the primary-outcome group indicate a greater burden of central sensitization-related symptoms and provide a clinical clue to mechanistic heterogeneity. Clinical management should integrate pain distribution, a complete neurologic examination, structural causes, sleep, and psychological factors, with quantitative sensory testing or neurophysiologic assessment when needed to guide pharmacologic, rehabilitative, psychological, and sleep-focused interventions.

Preoperative subthreshold neuropathic-like symptoms constituted the second risk domain. The primary-outcome group had higher preoperative DN4 and painDETECT scores and more frequent subthreshold symptoms, together with greater preoperative pain intensity, more night pain, longer pain duration, and greater functional limitation. Persistent pain after rotator cuff repair is multifactorial, and preoperative pain burden, sensory abnormalities, sleep, and psychological factors may jointly influence the postoperative pain experience [41]. Subthreshold symptoms may therefore provide preoperative risk information, with their association more appropriately interpreted as susceptibility to multiple pain mechanisms.

The MRI-based degeneration–tension composite phenotype integrated tear extent, tendon retraction, fatty infiltration, muscle atrophy, and the acromiohumeral interval. Maximum tear diameter, number of involved tendons, and several degeneration indicators were more pronounced in the primary-outcome group. More severe fatty infiltration is associated with poorer structural and functional outcomes after repair, indicating that muscle quality is an important component of prognosis in massive tears [42]. Fatty degeneration may continue to change after repair and is associated with clinical outcomes, supporting integrated characterization of muscle and tendon degeneration during risk assessment [43]. Grouping related imaging indicators into a composite domain reduces duplicate scoring and collinearity, but its operational thresholds require testing in independent cohorts.

The socioeconomic and rehabilitation resource domain showed substantial internal heterogeneity. Insufficient insurance or rehabilitation-expense coverage, longer distance to rehabilitation facilities, and lack of family support were more common in the primary-outcome group, whereas differences in educational level and occupation type were not statistically significant. Completion of short-term follow-up after rotator cuff repair may vary with insurance status and demographic factors, supporting inclusion of resource accessibility in preoperative assessment [44]. Rehabilitation behavior is also influenced by pain, knowledge, family support, and professional guidance, and resource limitations may affect the ability to sustain training and follow-up [45]. This prespecified 5-item construct spans multiple dimensions of resource accessibility, and the true contribution of each component may differ; the incremental information provided by the educational and occupational items remains uncertain in this cohort.

Equal weighting of the 4 domains was a prespecified simplification intended to improve the score’s simplicity, interpretability, and reproducibility. The decomposed Firth analysis showed that observed association strengths differed across domains: ORs were higher for the central sensitization-related symptoms–poor sleep and subthreshold-symptom domains, intermediate for the MRI domain, and the 95% CI for the socioeconomic and rehabilitation resource domain included 1. With a limited number of events, deriving weights directly from regression coefficients is vulnerable to sampling variation and overfitting, and a simple score must balance interpretability against empirical fit [46]. Each 1-point increment in this score denotes the presence of 1 additional risk domain; it does not imply equal underlying mechanisms or equal true prognostic contributions across domains.

Risk increased stepwise from scores of 0 to 4. The apparent AUC was 0.785, the bootstrap-corrected AUC was 0.784, the calibration intercept and slope were close to ideal values, and the Brier score was low. These findings indicate that the score differentiated relative risk in the current cohort and produced probability estimates that were generally consistent with observed risk. Evaluation of prediction models should jointly consider discrimination, calibration, overall error, and clinical net benefit [47]. Decision curve analysis suggested potential net benefit across the prespecified threshold range, but this estimate derives from the development cohort and constitutes exploratory evidence of clinical utility.

The current scope of the score is limited to patients at this center who underwent arthroscopic repair of massive rotator cuff tears and met the study definitions. External validation should retain the original scoring rule and risk equation and evaluate calibration, discrimination, and clinical utility across centers, health-care coverage environments, rehabilitation models, and patient populations, with recalibration when necessary [48]. Adequate external validation can further clarify the incremental value of the score for preoperative communication, follow-up intensity, and allocation of supportive resources.

This study has several limitations. First, as a single-center retrospective cohort, it remains susceptible to selection and information bias. Cohort eligibility required an ascertainable 6-month primary outcome and 12-month supplemental phenotype; 4 patients were excluded because of insufficient 12-month phenotype data and 2 because of insufficient 6-month sensory-examination records. Included patients and all excluded patients differed little in age, sex, and tear size, whereas excluded patients had greater preoperative pain intensity. Most exclusions arose from prespecified clinical or eligibility criteria; the overall comparison therefore reflects both restriction of the target population and follow-up selection and cannot separately quantify the direction or magnitude of bias attributable to missing data. Second, the primary outcome relied on symptom questionnaires, retrospectively documented bedside sensory findings, and exclusionary judgment. Verifiable 6-month sensory records were available for 46/46 (100.0%) patients positive for the primary outcome and 282/300 (94.0%) patients negative for the outcome; records were unverifiable for 18/300 (6.0%) negative patients. Because the positive-outcome definition required a verifiable abnormal sensory record, record availability was structurally related to outcome classification, creating a risk of verification bias and potential underclassification of cases with inadequate documentation. Consistency of documentation across examiners and calendar years and interobserver agreement could not be quantified retrospectively. Overlap between sensory-sign items in the DN4 and the additional bedside examination also limits interpretation of independent mechanistic evidence. Third, painDETECT scores of 13–18 are indeterminate and provided only supportive information within our composite definition. The clinical phenotype may include neuropathic, nociplastic, and mixed mechanisms; quantitative sensory testing, nerve-conduction studies, and electromyography were not routinely available, precluding estimation of the proportions of specific mechanistic subtypes. The CSI reflects the burden of central sensitization-related symptoms, and its higher values provide a clue to mechanistic heterogeneity. Fourth, neither the equally weighted score nor the operational thresholds of its component domains have undergone external validation, and the true contributions of the domains may differ. The educational and occupational items in the socioeconomic domain showed limited discrimination in this cohort; their independent contributions and incremental information require validation in external samples. Fifth, internal bootstrap validation can estimate optimism within the development cohort, but transportability across centers, recalibration requirements, and clinical net benefit still require external validation. Sixth, this was an observational analysis; between-group differences, the risk score, and exploratory perioperative associations represent association and risk-stratification information and do not support causal inference.

Conclusions

At 6 months after arthroscopic repair of massive rotator cuff tears, 13.3% of patients met the definition of a persistent neuropathic-like pain phenotype. A 0–4-point score comprising the central sensitization-related symptoms–poor sleep composite phenotype, preoperative subthreshold neuropathic-like symptoms, the MRI-based degeneration–tension composite phenotype, and limited socioeconomic and rehabilitation resources showed a reasonable risk gradient, discrimination, and calibration in this cohort. The score may inform preoperative risk stratification and allocation of perioperative supportive resources in comparable clinical settings; the identified phenotype may involve neuropathic, nociplastic, or mixed mechanisms, and clinical management should incorporate mechanism-oriented assessment. Multicenter external validation will further clarify transportability and clinical utility.

Abbreviations

ASD

Absolute standardized difference

ASES

American shoulder and elbow surgeons

AUC

Area under the receiver operating characteristic curve

BMI

Body mass index

CI

Confidence interval

CPSP

Chronic postsurgical pain

CSI

Central sensitization inventory

DCA

Decision curve analysis

df

Degrees of freedom

DN4

Douleur neuropathique 4 questionnaire

FN

False negative

FP

False positive

MRI

Magnetic resonance imaging

NPV

Negative predictive value

OR

Odds ratio

PCS

Pain catastrophizing scale

PPV

Positive predictive value

PSQI

Pittsburgh sleep quality index

RMB

Renminbi

SD

Standard deviation

TN

True negative

TP

True positive

VAS

Visual analog scale

VIF

Variance inflation factor

Authors’ contributions

Y.Z. and Y.C. contributed equally to this work and participated in data collection and database management. K.Z. performed the statistical analysis and assisted with manuscript preparation. J.C. contributed to data interpretation and manuscript revision. C.L. and K.G. conceptualized and designed the study, supervised the research process, and made major contributions to drafting and revising the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by the Jiangsu Provincial Basic Research Program (Natural Science Foundation) (Grant No. BK20201154). The funder had no role in study design, data collection and analysis, the decision to publish, or preparation of the manuscript.

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Ethics Committee of the Affiliated Hospital of Xuzhou Medical University (Ethics Approval No. XYFY2022-KL408-01). Because of the retrospective design, the requirement for written informed consent was waived by the Ethics Committee.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.