Development and internal validation of a field-based triage tool for lumbopelvic-hip dysfunction in collegiate athletes Hung-Chun Huang, Pei-Hsi Chou, Kung-Che Lee, I-Hua Chu, Ing-Jer Huang, Jing-Min Liang, Wen-Lan Wu ABSTRACT Because lumbopelvic–hip (LPH) dysfunction in athletes is commonly associated with subsequent low back pain (LBP), this cross-sectional derivation and internal validation study aimed to develop and internally validate a clinical triage scoring system (CTSS) for field-based identification of collegiate athletes requiring priority intervention-namely, those who, in the judgment of experts, are likely to warrant further assessment and management. A total of 864 collegiate athletes (mean age 21.3 ± 2.4 years; 80.8% male) were recruited from 10 universities. Participants underwent standardized assessments including demographic characteristics, clinical history, and LPH functional testing. Using an expert-adjudicated binary reference standard (priority intervention vs self-management), a multivariable logistic regression model was developed to derive the weighted CTSS. Model performance was evaluated using discrimination, calibration, and decision curve analysis (DCA), and internal validation was performed using 1,000 bootstrap resamples. Of the 864 participants, 463 athletes (53.6%) were classified as requiring priority clinical assessment. The final 14-factor CTSS comprised 12 positive-weight predictors, such as localized LPH pain, muscle weakness, and higher body mass index, and 2 negative-weight predictors, positive Lasegue’s sign and hamstring weakness, which functioned as safety-related modifiers. The model demonstrated acceptable discrimination (AUROC = 0.851, 95% CI: 0.824–0.876), with minimal optimism (optimism-corrected AUROC = 0.842) and excellent calibration (calibration slope = 0.996; calibration intercept = 0.0003). A total score of ≥9 was identified as the optimal threshold, yielding a sensitivity of 84.4% and specificity of 71.8%. DCA showed greater net benefit than treat-all and treat-none strategies across clinically relevant threshold probabilities (20%–50%), with a net benefit of 0.319 at a 50% threshold probability. The CTSS may provide a pragmatic field-based triage tool to support early identification of athletes who may benefit from priority intervention, although external validation is needed before broader implementation in sports medicine settings. Introduction The lumbopelvic-hip (LPH) complex plays an important role in force transfer between the trunk and lower extremities during athletic movement. Impairments in this region, including muscle imbalances, joint mobility restrictions, and altered neuromuscular control, have been associated with reduced movement efficiency and may contribute to musculoskeletal injury risk in athletic populations [ 1 , 2 ]. Low back pain (LBP) is one of the most common clinical manifestations associated with LPH dysfunction and remains highly prevalent among athletes exposed to repetitive loading demands [ 1 ]. However, LPH-related functional impairments may persist even in athletes without current LBP. Therefore, field-based assessment should not rely solely on the presence or absence of LBP when identifying athletes who may require further clinical evaluation. Importantly, a history of LBP remains a consistent predictor of recurrence [ 1 , 3 ], suggesting that symptomatic resolution does not necessarily reflect full restoration of neuromuscular function. Persistent alterations, such as impaired trunk muscle activation and increased spinal stiffness, have also been reported in asymptomatic athletes, potentially contributing to ongoing mechanical vulnerability [ 4 – 6 ]. In multidisciplinary sports medicine settings, some athletes presenting with LPH-related clinical manifestations that are suspected to exacerbate LBP may require priority intervention, activity modification, or self-management depending on clinical presentation [ 7 ]. In sideline and competition settings, sports medicine practitioners, including sports chiropractors, physiotherapists, and athletic trainers, are often required to rapidly distinguish athletes who require priority intervention for further clinical assessment and management from those suitable for self-management. However, many assessment approaches rely on imaging, specialist evaluation, or clinic-based resources, which may be impractical in time-constrained field environments [ 8 ]. As a result, triage decisions in these settings are often based on clinician judgment rather than a standardized framework. Although clinical prediction rules have been developed for identifying candidates for specific interventions in general LBP populations [ 9 , 10 ], their applicability to athletic field-based triage remains limited. Furthermore, while traditional single-item screening tests have been criticized for their limited predictive value in sports medicine [ 11 ], tools developed for established pain syndromes may be less suited to the dynamic functional impairments characteristic of high-load athletic participation. Although various functional screening instruments have been described, few have undergone rigorous internal validation, and even fewer have been explicitly linked to structured triage pathways in athletic populations [ 8 ]. This gap is particularly relevant in collegiate sports settings. Therefore, this study aimed to develop and internally validate a field-based Clinical Triage Scoring System (CTSS) for identifying collegiate athletes requiring priority intervention for LPH dysfunction. The CTSS was developed as a decision-support tool for triage and care prioritization, rather than as a diagnostic instrument. By integrating clinical history with objective functional indicators, including range of motion, muscle performance, and pain distribution, this study sought to derive a weighted scoring framework to support structured and practical triage in collegiate sports medicine settings. Materials and methods Participants and study design This cross-sectional study was conducted to develop and internally validate a CTSS and was carried out between August 2021 and December 2024. This study was reported in accordance with the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) Statement [ 12 ]. A total of 945 collegiate athletes were recruited from 10 universities in Taiwan, where standardized on-site assessments were performed by a trained sports medicine research team. The participant flow is detailed in Fig 1 . Inclusion criteria were: [ 1 ] active roster status on a varsity team; [ 2 ] age 18–30 years; and [ 3 ] a minimum training volume of 10 hours per week. Recruitment was conducted regardless of current symptom status in order to capture a broad spectrum of functional presentations, ranging from asymptomatic to symptomatic, within the collegiate athletic population. Exclusion criteria were defined to ensure participant safety and data validity. Athletes were excluded if they presented with: [ 1 ] “Red flags” indicative of serious pathology (e.g., suspected fracture, infection, or malignancy) [ 13 ], as determined by a supervising medical doctor; [ 2 ] current acute lower-extremity injuries that physically precluded the completion of functional assessments (i.e., unable to bear weight or perform dynamic tasks due to pain/instability); [ 3 ] history of lower-extremity surgery within the past year; or [ 4 ] incomplete data on primary outcome measures. To assess the impact of prior trauma, athletes with a history of recent minor injuries (e.g., Grade I/II sprains) were retained in the study, provided they had returned to full participation and could complete all protocols without pain exacerbation. Of the initial cohort (N = 945), 81 athletes were excluded based on these criteria (20 clinical contraindications: 8 red flags, 12 acute injuries; 61 missing data), which were mostly due to incomplete completion of the standardized on-site assessment protocol (e.g., scheduling or time constraints) rather than missing individual variables, yielding a final sample of 864 participants. This study was also reported in accordance with the TRIPOD guidelines [ 14 ]. All participants provided written informed consent, and the study was approved by the Kaohsiung Medical University Institutional Review Board (KMUHIRB-F(II)-20210113) in accordance with the Declaration of Helsinki. 10.1371/journal.pone.0348820.g001 Fig 1 Study Flowchart. Data collection Fifty-three candidate predictor variables were prespecified, guided by clinical rationale and existing literature, and categorized into four domains: (1) demographics and training characteristics; (2) retrospective injury history; (3) self-reported pain and function; and (4) objective physical examination findings. To assess these variables, a standardized, station-based assessment protocol was implemented. Examiners performing the physical assessments were blinded to participants’ demographic information, injury history, and questionnaire responses at the time of testing. The primary clinical outcome (priority intervention vs. self-management) was subsequently adjudicated independently by an expert panel after completion of data collection, thereby reducing the potential for incorporation bias. Demographic and injury history Data included sex (male/female), age (years), height (cm), weight (kg), body mass index (BMI, kg/m 2 ), years of sport participation, and average weekly training volume (hours/week). The point prevalence of LBP, 6-month, and 12-month prevalence were recorded. A history of sport-related musculoskeletal injury was recorded by anatomical region (low back, pelvis, hip, knee, and ankle). Detailed variable definitions and response formats are presented in Table 1 . 10.1371/journal.pone.0348820.t001 Table 1 Demographic Characteristics and Prevalence of Positive Screening Findings. Characteristics n (%) Mean ± SD Demographic Sex (Male/ Female) 698 (80.8)/ 166 (19.2) Age (year) 21.3 ± 2.4 Height (cm) 173.13 ± 7.53 Weight (kg) 71.72 ± 14.62 BMI (kg/m²) 23.86 ± 3.83 Sport experience (year) 9.7 ± 5.2 Training hours (hours/week) 20.4 ± 9.9 Low Back Pain Prevalence On day of evaluation 412 (47.7) In last 6-month 509 (58.9) In last 12-month 569 (65.9) Region History of Injury (Sport Practices or Competition) a Low back area 428 (49.5) Pelvic area 44 (5.1) Hip area 83 (9.6) Knee area 352 (40.7) Ankle area 509 (58.9) Pain Location in Lumbopelvic-hip region (Day of Evaluation) (1) Spinous process (L1-L5) 87 (10.1) (2) Sacrum (S1-S5) 83 (9.6) (3) Left lateral side 125 (14.5) (4) Left paraspinal 171 (19.8) (5) Right paraspinal 159 (18.4) (6) Right lateral side 112 (13.0) (7) Left gluteal 100 (11.6) (8) Right gluteal 83 (9.6) (9) Left ischial 20 (2.3) (10) Right ischial 24 (2.8) Lumbar and Hip Range of Motion Right lateral bending, mean degree ± SD 39.54 ± 9.81 Left lateral bending, mean degree ± SD 39.80 ± 9.85 Flexion, mean degree ± SD 81.07 ± 8.91 Extension, mean degree ± SD 40.02 ± 10.39 Hip area IR-R, mean degree ± SD 22.63 ± 10.64 ER-R, mean degree ± SD 40.39 ± 9.45 Total hip rotation-R, mean degree ± SD 63.02 ± 15.22 IR-L, mean degree ± SD 21.81 ± 10.33 ER-L, mean degree ± SD 38.23 ± 9.58 Total hip rotation-L, mean degree ± SD 60.03 ± 15.24 Manual Muscle Testing (Weakness) Gluteus maximus 482 (55.8) Gluteus medius 440 (50.9) Psoas major 317 (36.7) Piriformis 305 (35.3) Tensor fasciae latae 211 (24.4) Hamstring 208 (24.1) Quadriceps 122 (14.1) Positive Lumbopelvic-hip Provocation Tests Lumbar / Sciatic Hyperextension test 514 (59.5) Double leg raise 116 (13.4) Lasegue’s straight leg test 29 (3.4) Sacroiliac Joint (SIJ) Gaenslen’s test 252 (29.2) Advancement test 162 (18.8) Belt test 118 (13.7) Hip joint/ Muscle Single-leg hip extension test 471 (54.5) Bonnet’s Sign 386 (44.7) FABER test 321 (37.2) Internal/ External hip rotation test 274 (31.7) Ely’s sign 134 (15.5) Thomas test 108 (12.5) Ober’s test 100 (11.6) N (%) = Number and percentage of total athletes with a positive test result. Low back pain prevalence & pain area LBP was anatomically defined as the region bounded superiorly by the 12th thoracic rib, inferiorly by the inferior gluteal folds, and laterally by the posterior axillary lines. Symptom localization within the LPH region was documented using a standardized 10-zone anatomical mapping system ( S1 Fig ), consistent with established epidemiological protocols [ 15 ]. LBP case definitions: Current LBP was defined as pain or discomfort localized between the costal margin and the gluteal folds, self-reported at the time of assessment [ 15 ]. A history of LBP (6- and 12-month) was defined as an episode resulting in functional consequences, including reduced training volume, impaired performance, or medical consultation, consistent with IOC consensus criteria [ 16 , 17 ]. Specific assessment procedures Functional assessments were conducted using a standardized, station-based protocol adapted from established methodologies [ 18 , 19 ]. To minimize measurement error, all examiners completed a pre-study calibration workshop focusing on procedural standardization and scoring criteria. Testing was stratified by expertise: biomechanical measurements were led by a senior examiner (physical therapist or chiropractor), whereas all symptom-provocation tests were performed by a single experienced clinician (Doctor of Chiropractic with sports specialty certification) to eliminate inter-rater variability in symptom reproduction. A rigorous safety protocol was implemented under physician supervision, including a structured referral pathway for adverse events. All assessments adhered to symptom-limited criteria and were terminated immediately if: [ 1 ] pain intensity exceeded pre-test baseline levels significantly; [ 2 ] a > 20 mm increase on a 100-mm Visual Analog Scale (VAS) was reported [ 20 ]; or [ 3 ] observable protective neuromuscular responses (e.g., guarding, withdrawal) were noted. Bilateral Manual Muscle Testing (MMT) Bilateral manual muscle testing (MMT) assessed isometric strength bilaterally for eight muscle groups critical to lumbopelvic stability: quadriceps, hamstrings, adductors, tensor fasciae latae, psoas major, piriformis, gluteus maximus, and gluteus medius. Testing followed standardized positions and protocols [ 21 ]. For analytic purposes, strength outcomes were dichotomized to identify clinically relevant functional deficits. A score of “0” (Normal) was assigned if the participant maintained a strong isometric contraction against manual resistance. A score of “1” (Functional Deficit) was recorded if the contraction could not be sustained for ≥5 seconds (“break” phenomenon) or if marked bilateral asymmetry was observed. This binary coding was used to simplify model input and to identify clinically relevant functional deficits for logistic regression analysis. Assessment Lumbar Mobility Active lumbar range of motion was quantified using the dual-inclinometer technique. Two digital inclinometers (Hengshui Aohong Technology, China) were positioned over the T12 and S1 spinous processes. For flexion and extension, lumbar motion was estimated as the difference between the superior (T12) and inferior (S1) sensor readings to reduce pelvic contribution. Lateral side bending was measured using a single inclinometer at the T12 level, in accordance with established spinal flexibility assessment protocols. Hip Mobility Passive hip internal rotation (IR) and external rotation (ER) were measured bilaterally using a universal goniometer (Baseline®, Fabrication Enterprises Inc., USA). Participants were positioned prone with the knee flexed to 90°. To reduce pelvic compensation, a non-elastic strap was secured across the posterior superior iliac spines (PSIS). The examiner palpated the contralateral iliac crest throughout movement to monitor for pelvic rotation or tilt; trials showing compensatory motion were repeated [ 22 , 23 ] Lumbopelvic-hip Functional Screening A battery of fourteen standardized orthopedic maneuvers was selected to evaluate LPH function [ 23 , 24 ]. Although traditionally used in differential diagnosis, these tests were operationally defined in the present study as functional screening assessments to identify mechanical sensitivity, mobility restriction, and aberrant motor control, rather than to establish specific pathoanatomical diagnoses. Tests were categorized into three domains: [ 1 ] neural sensitivity (e.g., slump test, straight leg raise); [ 2 ] articular loading and sensitivity (e.g., FABER test, quadrant test); and [ 3 ] myofascial extensibility (e.g., Thomas test, Ober’s test). For model development and interpretability, outcomes were dichotomized (0 = negative; 1 = positive for reproduction of familiar pain significant restriction, or observable compensatory movement patterns). A complete list of tests and operational definitions is provided in S1 Table . Establishment of the clinical reference standard In the absence of a singular pathological gold standard, a two-phase procedure involving expert consensus and reliability verification was used to establish the clinical reference standard before predictive modeling. In Phase 1, a multidisciplinary panel of seven senior clinicians (three dual-credentialed chiropractors/physical therapists, two ICCSP-certified sports chiropractors, and two licensed chiropractors; mean clinical experience: 12 years) established a predefined consensus framework for clinical triage based on standardized symptom, functional, and orthopedic assessment findings. Rather than relying on item-by-item score comparison or any single abnormal finding, the framework incorporated integrated clinical judgment across multiple domains, including symptom-function concordance, training interference, and reproducibility of examination findings, with triage prioritization reflecting the overall clinical significance of these findings. In Phase 2, inter-rater reliability of the same predefined criteria was evaluated in a randomly selected subset of athletes (n = 41). Each panelist independently reviewed de-identified datasets comprising 53 variables and assigned triage classifications according to the prespecified rules, while remaining blinded to the classifications of the other raters. Inter-rater agreement for the expert-assigned 4-level triage classifications, rather than the individual CTSS assessment items, was quantified using Fleiss’ weighted kappa ( κ = 0.977, 95% CI: 0.936–1.000) [ 25 ]. The same criteria were then applied to the full analytic cohort (N = 864). Outcome classification for the analytic cohort was performed by a single calibrated examiner who had participated in the consensus process and applied the predefined criteria without modification. This approach was used to ensure consistent application of the predefined criteria across the full cohort. Participants were initially categorized into four triage grades reflecting clinical care needs: Grade 0 (healthy/asymptomatic), Grade 1 (low priority), Grade 2 (priority intervention), defined as requiring further comprehensive musculoskeletal assessment (e.g., more detailed history taking, neurological examination, and other relevant clinical assessments) followed by appropriate management, and Grade 3 (Red Flag). For predictive modeling, Grades 0 and 1 were combined as the self-management group (coded as 0), whereas Grade 2 was defined as the priority intervention group (coded as 1) [ 10 , 26 , 27 ]. Athletes meeting Grade 3 criteria were referred for immediate physician evaluation and were excluded from predictive modeling. Thus, the reference standard reflected predefined stratification of intervention priority within an operational framework rather than structural diagnosis or long-term clinical outcome. Model development & validation As prespecified, a complete-case analysis was performed. Missing data were not imputed because the proportion of incomplete cases was limited and the primary analysis was prespecified as complete-case analysis. Descriptive statistics were calculated for all demographic and clinical variables. Continuous variables were summarized as means and standard deviations (SDs), whereas categorical variables were summarized as frequencies and percentages. BMI was retained as a continuous variable during model development to preserve its linear risk gradient. To construct a parsimonious model suitable for clinical translation, a forward stepwise multivariable logistic regression analysis was performed to identify independent variables associated with priority intervention. This approach was selected to derive a simplified and interpretable model that could be translated into an integer-based clinical triage score for field-based use. To reduce overfitting risk, candidate predictors were prespecified based on clinical reasoning, model complexity was evaluated using the events-per-variable (EPV) criterion, and internal validation was performed using bootstrap resampling. Candidate predictors entered into the model were prespecified based on clinical reasoning and included demographic characteristics, LPH functional assessment findings, and orthopedic provocation test results. Associations were expressed as odds ratios (ORs) with 95% confidence intervals (CIs). Sample size adequacy was evaluated using the EPV criterion, with a minimum threshold of 10 events per predictor considered indicative of acceptable model stability [ 28 ]. Multicollinearity was assessed using variance inflation factors (VIFs), with values < 5 indicating no concerning collinearity. Overall model fit and performance were evaluated using the omnibus test, Nagelkerke R² , and the Brier score, with lower values indicating better overall accuracy [ 29 ]. Internal validation was performed with 1,000 bootstrap resamples to assess model optimism and coefficient stability. Bootstrap-corrected estimates of the area under the receiver operating characteristic curve (AUROC) and calibration metrics were obtained, and bias-corrected and accelerated (BCa) 95% CIs were derived for regression coefficients [ 18 , 19 ]. Calibration was assessed using the Hosmer-Lemeshow test, with p > 0.05 indicating acceptable model fit. A calibration plot was generated to compare observed event rates with predicted probabilities across ten decile-based groups, with 95% Wilson confidence intervals displayed for each group and the ideal 45-degree line serving as the reference [ 29 ]. The calibration slope was estimated by regressing the observed binary outcomes on the logit-transformed predicted probabilities, where a slope of 1.0 indicates perfect calibration. To further characterize calibration across the full range of predicted risks and to reduce the potential bias associated with grouping observations into deciles, a locally weighted scatterplot smoothing (LOESS) curve with a corresponding 95% confidence band was overlaid on the calibration plot. Construction of the weighted clinical triage scoring system (CTSS) To facilitate clinical implementation, the final multivariable logistic regression model was transformed into a weighted integer-based CTSS. Regression coefficients (β) were rescaled to preserve the relative contribution of each predictor while enabling straightforward point-based application [ 30 ]. The absolute value of the smallest statistically significant coefficient in the final model was used as the scaling reference. A common scaling factor was derived from this reference, and each predictor was assigned an integer weight proportional to its relative effect size. For categorical predictors, integer weights were obtained by multiplying the corresponding β coefficients by the scaling factor and rounding to the nearest integer. For continuous predictors, such as BMI, weighted contributions were initially modeled on their original continuous scale to reflect their cumulative contribution across the observed clinical range. These were subsequently categorized into discrete intervals, as presented in Table 2 , to improve interpretability and facilitate field-based use. An individual’s total CTSS score was calculated by summing the points assigned across all present predictors, with higher scores indicating greater priority for professional clinical evaluation and intervention. 10.1371/journal.pone.0348820.t002 Table 2 Multivariable Logistic Regression Model and Derived Weighted Scoring System for Identifying Athletes Requiring Priority Intervention (n = 864). Factors Odds Ratio (95%CI) β coefficient VIF Weighted Score Bonnet’s Sign 1.951 (1.362-2.795) 0.668 1.147 Yes = 2, No = 0 Internal/external hip rotation test 1.723 (1.171-2.534) 0.544 1.153 Yes = 1, No = 0 Thomas test 1.949 (1.127-3.370) 0.667 1.023 Yes = 2, No = 0 Hyperextension test 2.338 (1.620-3.375) 0.849 1.272 Yes = 2, No = 0 LBP on day of evaluation 2.374 (1.542-3.656) 0.865 1.945 Yes = 2, No = 0 LBP in last 6 months 2.423 (1.629-3.605) 0.885 1.659 Yes = 2, No = 0 Pain-now at [ 1 ] Spinal process (L1-L5) 2.567 (1.318-4.999) 0.943 1.141 Yes = 2, No = 0 Pain-now at [ 5 ] Right paraspinal 2.124 (1.233-3.659) 0.753 1.252 Yes = 2, No = 0 Pain-now at [ 7 ] Left gluteal 2.328 (1.197-4.527) 0.845 1.150 Yes = 2, No = 0 Gluteus medius muscle weakness 1.503 (1.065-2.122) 0.408 1.051 Yes = 1, No = 0 Piriformis muscle weakness 1.748 (1.199-2.548) 0.558 1.157 Yes = 1, No = 0 Body Mass Index (BMI, kg/m²) 1.061 (1.015-1.109) 0.059 1.020 < 17.5 = 2 17.5-24.0 = 3 24.1-31.0 = 4 31.1-38.0 = 5 38.1-44.8 = 6 ≥ 44.9 = 7 Lasegue’s straight leg test † 0.382 (0.152-0.959) −0.962 1.049 Yes = −2, No = 0 Hamstring muscles weakness † 0.636 (0.416-0.971) −0.453 1.120 Yes = −1, No = 0 † Factors inversely associated with the requirement for priority intervention (Odds Ratio < 1), resulting in negative weighted score values. Note: Point-based risk scores were derived by multiplying the β coefficients by a scaling factor of 2.45 and rounding to the nearest integer to facilitate clinical utility. For Body Mass Index (BMI), the score was calculated as: Score = round ( BMI × 0.145), yielding a range of 2–7 points. The total CTSS is the sum of all individual item scores (cumulative totals below 0 are truncated to 0), with a final range of 0–26 points. Evaluation and clinical utility assessment The performance of CTSS was evaluated in terms of discrimination, calibration, threshold-based classification, and clinical utility. Discriminative ability was assessed using the AUROC, reflecting the capacity of the scoring system to distinguish athletes requiring priority intervention from those suitable for self-management [ 18 ]. Sensitivity, specificity, and overall classification accuracy were calculated across score thresholds. The optimal operating cutoff was determined using the Youden Index (sensitivity + specificity − 1) [ 31 ]. To provide clinically interpretable risk estimates beyond dichotomous classification, the total CTSS score was also modeled as a continuous predictor within a logistic regression framework using the predefined binary reference standard as the dependent variable. Predicted probabilities of priority intervention were then derived for each possible score value, establishing a score–probability function for individualized probability estimation of priority intervention. Clinical utility was assessed using decision curve analysis (DCA) [ 32 ]. Net benefit was estimated across a prespecified range of clinically relevant threshold probabilities (20%–50%), and the model was compared against default strategies of “treat-all” and “treat-none.” This threshold range was selected a priori to reflect realistic trade-offs between unnecessary referrals and missed priority cases in field-based sports medicine settings. All statistical analyses were performed using SPSS version 25.0 (IBM Corp., Armonk, NY, USA; RRID: SCR_002865) with statistical significance set at p < 0.05. Calibration slope and DCA were performed in Python 3.11.3 (RRID: SCR_008394). Results Demographic characteristics and clinical classification A total of 864 collegiate athletes, primarily from basketball, badminton, rugby, and taekwondo were included in the final analysis. Based on the predefined clinical reference standard for early rehabilitation triage, 463 athletes (53.6%) were classified into the priority intervention group and 401 (46.4%) into the self-management group. The cohort was predominantly male (n = 698, 80.8%) with a mean age of 21.3 ± 2.4 years. Participants reported an average sports participation duration of 9.7 ± 5.2 years and a weekly training volume of 20.4 ± 9.9 hours. On the day of assessment, the mean pain intensity was 3.25 ± 1.84 on the visual analogue scale. The point prevalence of LBP was 47.7%, with 6-month and 12-month prevalence rates of 58.9% and 65.9%, respectively. A history of musculoskeletal injury was frequently reported, with the 3 most common sites being the ankle (58.9%), lower back (49.5%), and knee (40.7%). Regarding symptom distribution within the LPH zones, pain was most frequently identified in the left paraspinal (19.8%) and right paraspinal (18.4%) regions ( S1 Fig ). Detailed demographic characteristics and descriptive results of the LPH functional assessments, including range of motion, MMT, and orthopedic provocation tests, are presented in Table 1 . Multivariable logistic regression and internal validation Stepwise multivariable logistic regression identified 14 independent predictors from the predefined pool of 53 candidate variables ( Table 2 ). With 463 priority intervention cases and 14 retained predictors, the EPV ratio was 33.1, exceeding the recommended minimum threshold of 10 and supporting model stability. The final model demonstrated satisfactory overall performance (Nagelkerke R² = 0.470; omnibus test: χ² = 374.515, p < 0.001; −2 log likelihood = 818.791). Multicollinearity was minimal, with VIF ranging from 1.020 to 1.945 (mean VIF = 1.224), indicating low collinearity among retained predictors. As shown in Table 2 , factors positively associated with priority intervention included LBP within the previous 6 months, low back pain on the day of evaluation, localized pain at the L1–L5 spinous process, right paraspinal and left gluteal pain, positive hyperextension, Bonnet’s, Thomas, and internal/external rotation tests, gluteus medius weakness, piriformis weakness, and higher body mass index. In contrast, a positive Lasegue’s straight leg raise test and hamstring weakness were inversely associated with the outcome. Internal validation using 1,000 bootstrap resamples with replacement suggested good model stability. Bootstrap analysis ( S2 Table ) showed that 12 of the 14 retained predictors (85.7%) remained statistically significant (p < 0.05). Two predictors did not achieve statistical significance in the bootstrap analysis. Hamstring weakness showed a borderline effect (p = 0.077; BCa 95% CI: −0.959 to 0.024), whereas the Lasègue straight leg test was not significant (p = 0.143; BCa 95% CI: −2.300 to 0.365). Despite this, both variables were retained in the final model based on their significance in the original logistic regression (p < 0.05) and their established clinical relevance. All retained predictors demonstrated bootstrap 95% confidence intervals supporting their inclusion in the model. Calibration analysis demonstrated excellent agreement between predicted and observed probabilities ( Fig 2A ). The calibration slope was 0.996 and the intercept was 0.0003, indicating near-ideal calibration. The Hosmer–Lemeshow goodness-of-fit test showed no evidence of lack of fit (χ² = 15.005, p = 0.059), and the Brier score was 0.152. Consistent with these findings, the LOESS curve closely followed the ideal 45° reference line, with the 95% confidence band encompassing the reference line throughout most of the probability range, indicating good calibration throughout the full spectrum of predicted probabilities. 10.1371/journal.pone.0348820.g002 Fig 2 Predictive Performance and Clinical Utility of the Clinical Triage Scoring System (CTSS). (A) Calibration Plot: Observed versus predicted probabilities with the 95% confidence band and the ideal 45° reference line (dashed). (B) ROC Curve: Discriminative performance of the CTSS. (C) Score-Probability Curve: Relationship between CTSS total score and predicted probability. Orange dashed line indicates optimal cut-off (≥ 9 points). (D) Decision Curve Analysis: Net benefit of the CTSS (blue line) compared to “Treat All” and “Treat None” strategies across threshold probabilities. Construction of the weighted CTSS Regression coefficients from the final multivariable model were converted into an integer-based scoring system to derive the CTSS. Predictor weights were assigned by multiplying the β coefficients by 2.45 and rounding to the nearest integer. For BMI, which was retained as a continuous predictor, points were assigned using a graded scale to preserve usability in field-based scoring. The final CTSS ranged from 0 to 26 points, with higher scores indicating greater priority for further clinical evaluation and intervention ( Table 2 ). Discrimination and threshold performance The CTSS demonstrated good discrimination, with an AUROC of 0.851 (95% CI: 0.824–0.876) ( Fig 2B ). The optimal threshold, determined by the maximum Youden index (0.563), was > 8.5 points, corresponding to an operational threshold of ≥9 points ( Table 3 ). At this threshold, sensitivity was 84.4% and specificity was 71.8%. At lower thresholds, such as ≥5 points, sensitivity increased to 97.2%, whereas specificity decreased to 40.1%. At higher thresholds, such as ≥11 points, specificity increased to 85.0%, whereas sensitivity decreased to 67.4%. These findings illustrate the trade-off between maximizing sensitivity and minimizing false-positive referrals across candidate triage thresholds. 10.1371/journal.pone.0348820.t003 Table 3 Coordinates of the ROC Curve showing Sensitivity, Specificity, and Youden Index at various CTSS cut-off values. Cut-off value Sensitivity Specificity Youden index (Sensitivity+ Specificity-1) 0.0000 1.000 0.000 0.000 1.5000 0.998 0.000 −0.002 2.5000 0.998 0.007 0.005 3.5000 0.996 0.107 0.103 4.5000 0.985 0.227 0.212 5.5000 0.972 0.401 0.373 6.5000 0.937 0.521 0.459 7.5000 0.905 0.628 0.533 8.5000 0.844 0.718 0.563 9.5000 0.749 0.788 0.537 10.5000 0.674 0.850 0.524 11.5000 0.575 0.900 0.475 12.5000 0.464 0.920 0.385 13.5000 0.352 0.958 0.310 14.5000 0.205 0.965 0.170 15.5000 0.127 0.985 0.112 16.5000 0.054 0.990 0.044 17.5000 0.028 0.995 0.023 18.5000 0.019 0.998 0.017 19.5000 0.004 0.998 0.002 20.5000 0.000 0.998 −0.002 22.0000 0.000 1.000 0.000 Note: The optimal cut-off point was determined using the maximum Youden index (0.563), corresponding to a score of >8.5. For clinical application, this is operationalized as a threshold of ≥9 points. Score-probability relationship Logistic regression analysis demonstrated a positive association between CTSS score and classification into the priority intervention group. Each 1-point increase in CTSS score was associated with a 1.51-fold increase in the odds of priority intervention classification (OR = 1.510, 95% CI: 1.430–1.593). The derived score-probability relationship ( Fig 2C ) showed a graded, monotonic increase in the estimated probability of priority intervention classification across the observed score range. A CTSS score of 9 corresponded to an estimated probability of 49.6%. Clinical utility DCA was conducted to evaluate the clinical utility of CTSS ( Fig 2D ). Across the prespecified clinically relevant threshold probability range of 20%–50%, the model demonstrated consistently higher net benefit than both “treat-all” and “treat-none” strategies. At a 20% threshold probability, the CTSS yielded greater net benefit than the treat-all approach (0.454 vs. 0.420). This advantage persisted at the 50% threshold (0.319 vs. 0.072). These findings suggest that applying the CTSS within this probability range may improve referral efficiency by increasing the proportion of correctly identified priority cases while reducing unnecessary referrals. Discussion The present study developed and internally validated the CTSS, a 14-factor weighted scoring tool for field-based triage of LPH dysfunction in collegiate athletes. This clinical decision- support tool was intended to approximate expert clinical triage decisions, thereby facilitating the prioritization of athlete screening in field settings. In this context, the predefined triage category of priority intervention indicates that these athletes may require further multidisciplinary assessment to ensure safety, rather than representing an objective pathological or prognostic endpoint. Its purpose is neither to establish an independent pathological diagnosis nor to predict future injury, rehabilitation outcomes, or athletic performance. By integrating anthropometric, functional, and orthopedic provocation findings, the CTSS provides a structured framework for identifying athletes who may require priority intervention in field-based settings. In contrast to traditional clinical prediction rules targeting established LBP [ 9 ], the CTSS emphasizes the early recognition of movement-related and motor-control impairments relevant to athletic populations. The inclusion of LPH mobility and neuromuscular domains is consistent with prior evidence linking these factors to LBP in athletes [ 33 ]. The CTSS demonstrated good discriminative performance (AUROC = 0.851), with an operational threshold of ≥9 points. Notably, only a subset of clinically assessed variables was retained in the final multivariable model, suggesting that the model may have filtered out redundant information while approximating aspects of the expert-derived decision process. Details of the retained predictors are provided in S3 Table . Multicollinearity was minimal (VIF 1.02–1.95), and internal validation using bootstrap resampling suggested limited optimism and stable calibration. The small difference between the apparent and bootstrap-corrected AUROC was also consistent with the favorable events-per-variable ratio (33.1). While stepwise selection yielded a parsimonious model in this study, its known tendency to produce unstable variable selection and poor reproducibility across datasets highlights the need for external validation. When modeled as a continuous score, the CTSS showed a graded increase in the estimated probability of priority intervention classification, with a score of 9 corresponding to an approximate predicted probability of 50%. This supports its interpretation as a clinically meaningful decision boundary rather than an arbitrary dichotomous threshold. Although the CTSS comprises 14 items, its component variables are derived from routine clinical information and standardized field-based assessments, including BMI, pain history, range of motion, and MMT. This may support practical implementation in sports medicine settings without requiring advanced diagnostic hardware or complex software during initial triage. In the present cohort, 53.6% of athletes met the predefined criteria for priority intervention, including 19.5% who did not report spontaneous symptoms but demonstrated objective functional deficits. This finding suggests that the CTSS may identify clinically relevant impairment that is not evident from symptom reporting alone. Under the predefined reference standard, Grade 2 classification required multiple positive clinical findings in conjunction with objective impairment, a criterion intended to reduce over-classification based on isolated symptoms alone [ 16 ]. The retained predictors represent clinically relevant domains related to loading exposure, kinetic-chain function, and neuromuscular control. Higher BMI and a history of LBP may be associated with cumulative loading exposure and recurrent symptom burden, consistent with reports linking prior symptoms to recurrence risk [ 3 ]. The inclusion of hip-centered markers, including Thomas test positivity, rotational mobility restrictions, and Bonnet’s sign, emphasizes the regional interdependence of the LPH complex rather than isolated lumbar symptoms [ 23 , 24 , 33 , 34 ]. Similarly, the pattern of paraspinal and gluteal findings, combined with hip muscle weakness, may reflect altered lumbopelvic coordination and compensatory loading patterns [ 5 , 6 , 35 ]. The OR pattern further suggests that recent symptom burden (OR range 2.37–2.42), localized pain provocation, and clustered mechanical findings contributed more strongly to priority classification than isolated muscle weakness (OR range 1.50–1.75). BMI was the only continuous predictor retained and was converted into a graded point scale (2–7 points) to preserve its contribution while supporting usability in field-based scoring. The retention of multiple localized pain regions (L1–L5, paraspinal, and gluteal) further suggests that the CTSS may capture not only the presence of pain but also its topographical distribution, providing additional triage value beyond a binary history of LBP alone. Notably, the model assigned negative weights to a positive Lasegue’s sign (OR = 0.38) and hamstring weakness (OR = 0.64). Within this triage context, these findings may reflect posterior chain tightness or non-priority mechanical presentations rather than the specific functional profile captured by the CTSS. Because straight leg raise positivity in athletes may be influenced by musculotendinous stiffness and lumbopelvic movement characteristics rather than neural involvement alone, its isolated presence may be less informative for priority triage [ 36 , 37 ]. Thus, these negative weights may have helped reduce over-classification of athletes with non-specific flexibility limitations. However, any athlete demonstrating overt neurological deficits or progressive red-flag symptoms should undergo comprehensive medical evaluation. An important strength of CTSS is its capacity to identify functional abnormalities in athletes who may not report prominent symptoms, thereby addressing a recognized limitation of symptom-based triage approaches [ 4 ]. DCA further contextualized the clinical utility of the model ( Fig 2D ). Compared with treat-all and treat-none strategies, CTSS demonstrated consistently higher net benefit across the prespecified threshold probability range of 20%–50%. At the 50% threshold probability, corresponding to the operational score threshold of 9, the net benefit was 0.319, suggesting improved decision balance between identifying athletes requiring priority intervention and reducing unnecessary referrals. These findings suggest that structured triage may improve referral efficiency in field-based athletic settings, particularly where time and resources are constrained. However, CTSS is intended to support prioritization of further clinical evaluation and management, rather than to replace comprehensive clinical assessment, establish diagnosis, or make return-to-play decisions in isolation. Importantly, classification into the priority intervention group does not imply a single standardized treatment pathway. Rather, it indicates that the athlete may warrant timely professional assessment and targeted conservative management, while the specific treatment approach should remain individualized according to the full clinical presentation. Its application should be integrated with professional clinical judgment and appropriate medical evaluation when red flags or progressive symptoms are present. In this respect, CTSS aligns with broader clinical recommendations supporting early, targeted conservative care for LPH dysfunction when clinically appropriate [ 7 ]. Clinically, CTSS may serve as a structured decision-support framework for field-based triage. A score of ≥ 9 points identifies athletes with a higher probability of requiring priority intervention, whereas lower scores may support supervised self-management or continued monitoring, depending on the broader clinical context. The ≥ 9-point threshold should be interpreted as an operational threshold derived from the present cohort rather than a universally fixed cutoff. Notably, 19.5% of the priority-classified athletes reported no spontaneous symptoms at the time of assessment but were identified through objective functional findings, such as localized muscle weakness and positive provocation tests. This underscores the system's capacity to detect clinically relevant impairments that are not captured by symptom reporting alone [ 38 ]. By incorporating objective functional findings alongside symptom-related information, CTSS is consistent with rehabilitation approaches that emphasize movement quality and motor control in athletic populations [ 4 ]. Nevertheless, the tool is intended to complement, rather than replace, comprehensive clinical evaluation, and its application should remain integrated with professional clinical judgment [ 8 ]. Limitations Several limitations should be acknowledged. First, the cross-sectional design captured current triage priority rather than longitudinal outcomes; thus, causal inference and prediction of future injury-related outcomes cannot be established. Second, the predominantly male (80.8%) cohort may limit the generalizability of the CTSS, particularly to female athletes, professional or youth populations, and non-Asian groups [ 39 ]. Given reported sex-related differences in lumbopelvic control, future studies may require model recalibration for broader application. Third, self-reported injury history may have introduced recall bias. Fourth, the expert-consensus reference standard did not include imaging; however, this was consistent with the study’s focus on functional triage rather than structural diagnosis, particularly given the documented weak correlation between imaging findings and clinical symptoms [ 40 , 41 ]. Although inter-rater agreement was high (kappa = 0.977), the reference standard remained consensus-based rather than an objective gold standard. Fifth, because the expert-derived reference standard and many candidate predictors were based on the same standardized clinical assessment ( S3 Table ), complete separation between the candidate predictors and the reference standard was not possible. Consequently, incorporation bias cannot be entirely excluded, and the reported discrimination may represent an optimistic estimate of model performance. Sixth, the stepwise regression approach has recognized limitations regarding variable selection stability, coefficient estimation, and reproducibility across different datasets. These risks were partly addressed through a favorable events-per-variable ratio (EPV = 33.1), bootstrap validation (n = 1,000) for optimism assessment, and low collinearity among retained predictors (VIF < 2.0). Nevertheless, bootstrap validation cannot replace external validation. Finally, although the CTSS demonstrated good internal performance, external validation and formal evaluation of inter-rater reproducibility in real-world settings have not yet been conducted. Therefore, further studies in independent and diverse athletic populations, especially among female athletes and those engaged in specific sport discipline, are warranted to establish the generalizability, reproducibility, and clinical utility of the CTSS before broader clinical implementation. Conclusion In conclusion, this study developed and internally validated a 14-factor Clinical Triage Scoring System (CTSS) for the structured triage of LPH dysfunction in collegiate athletes. The model demonstrated good discrimination and satisfactory calibration, supporting its internal performance and stability as a clinical decision-support tool. Rather than functioning solely as a binary threshold-based tool, CTSS may also be interpreted on a continuous probability scale to support more individualized triage decisions in field-based athletic settings. The system is intended to complement, not replace, comprehensive clinical assessment. Future research should focus on external validation and on evaluating the impact of CTSS-guided triage on clinical resource allocation and athlete outcomes. Supporting information S1 Fig Map of lumbopelvic-hip pain zones. (TIF) S1 Table Clinical Assessment Protocols for Lumbopelvic-hip Screening. (DOCX) S2 Table Bootstrap Validation of Logistic Regression Coefficients (1,000 Resamples). (DOCX) S3 Table Overlap between candidate predictor categories, expert outcome adjudication, and variables retained in the final multivariable model. (DOCX)