Development and Validation of a Nomogram for Predicting Recurrence in Patients Previously Treated Conservatively for Cervical Spondylosis Authors: Hanze Mao, Guangqi Lu, Shuaiqi Zhou, Minghui Zhuang, Xin Xiu, Jing Li, Xinyue Sun, Yakun Liu, Mingming Ma, Jiaming Hu, Jie Yu, Liguo Zhu Published: 2026 ABSTRACT BackgroundCervical spondylosis frequently recurs after conservative treatment, but its predictors remain unclear. This study aimed to identify factors associated with recurrence and develop a predictive nomogram.MethodsThis retrospective analysis used data from a multicenter cross-sectional survey (six Chinese cities, March–November 2025). We identified 668 patients who reported remission after systematic conservative treatment and retrospectively collected their post-treatment recurrence status via self-report. Potential predictors included demographics, lifestyle, clinical history, and radiographic findings. The sam ple was randomly split into training (70%) and validation (30%) sets. Multivariable logistic regression identified independent predictors, and a nomogram was constructed. Model performance was evaluated using AUC, calibration, and decision curve analysis. Supplementary analyses examined episode frequency and time to recurrence.ResultsSix independent predictors were identified: exercise duration per session (OR=0.992), neck length (OR=0.850), disc space narrowing (OR=0.478), disease duration (overall P<0.001), pre-treatment attack frequency (overall P<0.001), and PCS score (OR=0.943). Compared with disease duration <3 months, 3–11 months increased risk (OR=2.691); pre-treatment attacks ≥6/month raised risk (OR=4.608). The nomogram showed good discrimination (training AUC=0.760) and net benefit. Longer exercise was associated with delayed recurrence, while longer disease duration and frequent prior episodes increased recurrence risk; disc narrowing paradoxically predicted lower recurrence.ConclusionPost-treatment recurrence is associated with multifactorial factors, including modifiable elements like exercise and physical health, as well as disease chronicity. Imaging changes may prompt protective behaviors. These findings support individualized risk stratification, but causal inference requires prospective validation. Introduction Global Burden of Disease studies have shown that musculoskeletal disorders have become the leading cause of disability in many countries.1 Cervical spondylosis, as a common musculoskeletal disorder, is a group of clinical syndromes based on degenerative changes of the cervical spine and can involve the vertebral bodies, intervertebral joints, ligaments, and surrounding neurovascular structures.2 With population aging, changes in work patterns, and increased use of electronic devices, the burden of cervical spondylosis continues to rise.3 Patients often present with neck and shoulder pain, stiffness, and limited mobility, and severe cases may be accompanied by upper limb numbness, dizziness, and neurological dysfunction, which can further affect quality of life, work efficiency, and social functioning.2,4 Currently, treatment for cervical spondylosis mainly includes health education, postural correction, pharmacotherapy, physical therapy, exercise-based rehabilitation, and, when necessary, surgical intervention.2,4 Although symptoms are relieved in most patients following standard treatment, recurrence remains a substantial challenge. Previous reports have shown that recurrence after conservative treatment is common, with over 70% of patients experiencing a subsequent episode of neck pain.5 This not only increases medical and economic burden but may also reduce treatment adherence and contribute to disease chronicity and complexity.6,7 Therefore, investigating the risk factors for cervical spondylosis recurrence is of significant clinical importance for optimizing long-term management and developing individualized prevention and treatment strategies. Previous studies have suggested that the onset of cervical spondylosis is associated with age, occupation, prolonged poor posture, static loading, psychological factors, lack of physical activity, and obesity, among others.8–10 However, current evidence remains heavily focused on pathogenesis, diagnostic classification, and short‑term treatment outcomes. Crucially, to our knowledge, dedicated prognostic models or nomograms specifically designed to predict post‑treatment recurrence have not yet been developed and validated. Based on this gap, the present study aimed to comprehensively investigate factors associated with recurrence in patients with cervical spondylosis, including demographic characteristics, occupational and lifestyle factors, disease history, and imaging findings, and to develop a simple, clinically applicable predictive nomogram, with the goal of providing theoretical guidance and a practical reference for stratified management, recurrence warning, and long-term follow-up interventions. Materials and Methods Subjects This study was approved by the Institutional Review Board of Wangjing Hospital, China Academy of Chinese Medical Sciences (Approval No.: WJEC-KT-2024-069-P002). All participants provided informed consent. This is a retrospective analysis of cross-sectional survey data derived from a large multicenter survey conducted from March to November 2025 across six Chinese cities (Beijing, Shenzhen, Changchun, Xi’an, Changsha, and Suzhou), in which recurrence status was self-reported and retrospectively recalled by patients after their remission. The original survey was designed to characterize cervical spine health among Chinese adults. Community-dwelling adults aged 18 years or older were recruited through a multistage random sampling method. The survey initially planned to enroll 7000 participants, with 2000 from Beijing and 1000 from each of the other five cities. Trained investigators collected data via face-to-face interviews using a standardized questionnaire. Study Participants A total of 7250 individuals were initially surveyed. After excluding 281 participants due to incomplete data, 6969 provided valid responses and were included in the present analysis. The inclusion criteria for cervical spondylosis in this study were: (1) age ≥18 years; (2) a previous diagnosis of cervical spondylosis based on both radiographic evidence of structural changes (such as disc space narrowing, osteophyte formation, and cervical curvature abnormalities) and characteristic clinical symptoms corresponding to each subtype (including axial neck pain, cervical radiculopathy, and cervical myelopathy). The exclusion criteria were: (1) severe cervical myelopathy requiring surgical intervention; (2) a history of cervical spine surgery; (3) cervical trauma or fracture; (4) inflammatory or infectious spinal diseases; (5) spinal tumors; (6) other spinal disorders such as ankylosing spondylitis or ossification of the posterior longitudinal ligament; and (7) pregnancy or lactation (due to contraindications for radiographic examination). Based on these criteria, 1033 patients were identified as having been previously diagnosed with cervical spondylosis. Among these, 766 had received systematic hospital-based conservative treatment, and 668 achieved clinical remission following treatment. These 668 participants were subsequently enrolled in this study, of whom 561 experienced disease recurrence following treatment, and 107 did not experience recurrence (Figure 1).Figure 1Flowchart of participant selection and study enrollment.Part icipant screening diagram showing exclusions, treatment outcomes and recurrence vs non-recurrence counts.A boxed, arrow-linked diagram summarizes how a study sample was narrowed from a large survey group through multiple exclusion steps. It shows removal of incomplete responses, then exclusion of people who did not meet eligibility criteria, yielding a cervical spondylosis cohort. Additional exclusions remove those without systematic conservative treatment and those who failed to reach clinical remission. The remaining participants who achieved remission are split into two outcome groups: those with no recurrence and those with recurrence, with counts displayed in each box. Flowchart of participant selection and study enrollment. Data Collection Data were collected through face-to-face interviews using a standardized electronic questionnaire. The questionnaire was developed based on a comprehensive review of relevant literature and underwent multiple rounds of expert consultation to reach a consensus, ensuring its scientific rigor and relevance to the study objectives. Prior to the formal survey, a pilot test was conducted with a small sample (n=50) to evaluate the clarity, comprehensibility, and reliability of the items; the questionnaire was subsequently refined based on the feedback received. These development and validation steps were implemented to minimize potential subjectivity and inconsistency in self-reported data. During data collection, trained investigators administered the questionnaire. To enhance data accuracy, participants were provided with clear instructions and recall periods for key variables were deliberately kept short to reduce memory-related biases. Furthermore, upon completion of each interview, another researcher independently reviewed the questionnaire for completeness and logical consistency, with any missing or ambiguous entries clarified by follow-up contact with the participant. All predictor variables—including exercise duration, PCS score, lifestyle factors, and radiographic findings—were measured at the time of the survey, while recurrence status was retrospectively recalled by patients. The complete raw dataset used for this analysis is available in Supplementary File 1. Outcome Measures The primary outcome of this study was disease recurrence following systematic hospital-based conservative treatment for cervical spondylosis. We defined “systematic hospital-based conservative treatment” as patient-reported receipt of at least one course of conservative treatment in a hospital setting, which may include physical therapy, manual therapy, medications, or other non-surgical interventions. Remission was defined as patient-reported symptom disappearance or substantial relief, with no need for further treatment following the course of conservative therapy. Recurrence was strictly defined as the reappearance of symptoms that necessitated repeated medical consultations or additional treatment, or that were objectively confirmed by a physician through clinical examination, following initial remission. For further supplementary analyses, recurrence frequency, categorized as no recurrence, occasional recurrence, or frequent recurrence, and time to recurrence, measured in days from symptom remission to the first recurrence, were also assessed to analyze the association between predictors and recurrence timing. Selection of Predictors Potential factors associated with cervical spine disorder recurrence were systematically assessed through a combination of questionnaire-based and radiographic measurements. The questionnaire encompassed eight domains, all representing potential predictors: Basic demographic information: age, sex, and educational attainment;Physical information: height and weight were measured using standardized instruments, and body mass index (BMI) was calculated as weight (kg) divided by height squared (m2); Neck length was measured from the external occipital protuberance to the spinous process of the seventh cervical vertebra (C7) along the midline of the back, with the neck in a neutral position.11Occupational factors: daily working hours, time spent with the head in a downward position (both total daily duration and duration per continuous session), mental workload intensity, physical work intensity, and duration of maintaining the same working posture per session;Electronic device usage: daily duration of computer and smartphone use for work and entertainment, as well as screen height relative to eye level;Transportation and load carriage: primary mode of transportation, daily travel time, and bag-carrying habits including bag type and estimated bag weight;Sleep-related factors: daily sleep duration, sleeping positions, pillow height, and mattress firmness;Physical activity: exercise frequency per week and duration per exercise session;Assessed using the 12-item Short Form Health Survey (SF-12),12 including the Physical Component Summary (PCS) and Mental Component Summary (MCS), to reflect overall health status, disease burden, and functional limitations. Basic demographic information: age, sex, and educational attainment; Basic demographic information: age, sex, and educational attainment; Physical information: height and weight were measured using standardized instruments, and body mass index (BMI) was calculated as weight (kg) divided by height squared (m2); Neck length was measured from the external occipital protuberance to the spinous process of the seventh cervical vertebra (C7) along the midline of the back, with the neck in a neutral position.11 Physical information: height and weight were measured using standardized instruments, and body mass index (BMI) was calculated as weight (kg) divided by height squared (m2); Neck length was measured from the external occipital protuberance to the spinous process of the seventh cervical vertebra (C7) along the midline of the back, with the neck in a neutral position.11 Occupational factors: daily working hours, time spent with the head in a downward position (both total daily duration and duration per continuous session), mental workload intensity, physical work intensity, and duration of maintaining the same working posture per session; Occupational factors: daily working hours, time spent with the head in a downward position (both total daily duration and duration per continuous session), mental workload intensity, physical work intensity, and duration of maintaining the same working posture per session; Electronic device usage: daily duration of computer and smartphone use for work and entertainment, as well as screen height relative to eye level; Electronic device usage: daily duration of computer and smartphone use for work and entertainment, as well as screen height relative to eye level; Transportation and load carriage: primary mode of transportation, daily travel time, and bag-carrying habits including bag type and estimated bag weight; Transportation and load carriage: primary mode of transportation, daily travel time, and bag-carrying habits including bag type and estimated bag weight; Sleep-related factors: daily sleep duration, sleeping positions, pillow height, and mattress firmness; Sleep-related factors: daily sleep duration, sleeping positions, pillow height, and mattress firmness; Physical activity: exercise frequency per week and duration per exercise session; Physical activity: exercise frequency per week and duration per exercise session; Assessed using the 12-item Short Form Health Survey (SF-12),12 including the Physical Component Summary (PCS) and Mental Component Summary (MCS), to reflect overall health status, disease burden, and functional limitations. Assessed using the 12-item Short Form Health Survey (SF-12),12 including the Physical Component Summary (PCS) and Mental Component Summary (MCS), to reflect overall health status, disease burden, and functional limitations. Cervical spine imaging was performed using a mobile X-ray unit, including anteroposterior, lateral, and functional views, for all participants except those who were pregnant or lactating. Radiographic variables were considered potential structural predictors and included: Cervical curvatureIntervertebral disc space narrowingOsteophyte formationCervical range of motion Cervical curvature Cervical curvature Intervertebral disc space narrowing Intervertebral disc space narrowing Osteophyte formation Osteophyte formation Cervical range of motion Cervical range of motion To ensure measurement reliability, all radiographs were independently read by two trained radiologists following standardized protocols. In cases of disagreement, final decisions were reached through discussion or consultation with a senior radiologist to guarantee the consistency and accuracy of the recorded findings. Statistical Analysis A retrospective analysis was conducted among patients who had completed treatment for cervical spondylosis. Recurrence status was determined based on patient-reported symptom reappearance after initial remission. The primary outcome was disease recurrence, defined as a binary variable. For baseline characteristics, continuous variables with a normal distribution were presented as mean ± standard deviation, and group comparisons were performed using the t-test. Continuous variables with a non-normal distribution were presented as median and interquartile range (IQR), and group comparisons were performed using the Kruskal–Wallis test. Categorical variables were presented as counts (percentages) [n (%)], and group comparisons were performed using the chi-square test. The dataset was randomly divided into a training set and a validation set at a ratio of 7:3. Univariate analysis was conducted on the training set to screen for variables potentially associated with recurrence. Variables with a P-value < 0.1 were entered into a multivariable logistic regression model. Variables with a P-value ≥ 0.05 were then sequentially removed, and the remaining independent predictors were retained. A nomogram was subsequently constructed based on these predictors to enable individualized prediction of recurrence risk. The selected variables and the resulting model were applied to the validation set. Model performance was evaluated in both the training and validation sets using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). To examine the robustness of the findings, two supplementary analyses were performed. First, the outcome was redefined as frequency of episodes (no recurrence, occasional recurrence, frequent recurrence). An ordinal logistic regression model was used to assess the associations between the independent variables and the ordinal outcome. The proportional odds assumption was evaluated using the Brant test. Because the overall test rejected the parallel regression assumption, this indicated that the effects of the independent variables on the outcome were inconsistent across different cut-points of the ordinal outcome (Supplementary Table S1). Therefore, a generalized ordinal logistic regression model was used to relax the parallel regression assumption, allowing the effects of covariates to vary across different outcome categories. Second, in the subgroup of patients who experienced recurrence, linear regression was used to explore the associations between each predictor and time to recurrence (in days), with time to recurrence as the dependent variable. Both supplementary analyses included the independent predictors that were ultimately identified in the main analysis. These supplementary analyses aimed to improve the reliability and generalizability of the findings. All statistical analyses were performed using R software (version 4.5.1), and a P-value < 0.05 was considered statistically significant. Results Baseline Characteristics of the Study Participants Baseline characteristics of patients with and without recurrence are presented in Table 1. No statistically significant differences were observed between the two groups across the majority of variables assessed. However, several distinct differences emerged: patients with recurrence exhibited significantly shorter exercise duration per session and lower Physical Component Summary (PCS) scores, suggesting poorer physical health status. In addition, recurrence was closely associated with longer disease duration and higher pre-treatment attack frequency, and a greater proportion of patients in the recurrence group engaged in moderate physical workload intensity (35.6% vs 28.3%, P = 0.071, approaching statistical significance). Collectively, these findings suggest that physical health status, disease chronicity, and higher pre-treatment episode frequency may serve as important factors associated with post-treatment recurrence.Table 1Baseline Characteristics of Study ParticipantsVariableOverallNon-RecurrenceRecurrenceP valuen668107561Age (mean (SD))50.22 (15.58)50.18 (14.95)50.22 (15.71)0.977BMI (mean (SD))24.29 (4.55)24.56 (5.24)24.24 (4.41)0.498Neck length (mean (SD))13.00 (1.90)13.17 (1.90)12.97 (1.90)0.313Neck circumference (mean (SD))33.99 (3.65)34.57 (3.38)33.88 (3.69)0.077Daily working hours (mean (SD))6.36 (3.76)6.47 (3.62)6.33 (3.79)0.725Daily head down time (mean (SD))4.33 (2.63)4.48 (2.69)4.30 (2.62)0.536Daily work related computer use time (mean (SD))2.51 (3.31)2.78 (3.43)2.46 (3.28)0.365Daily recreational computer use time (mean (SD))0.72 (1.47)0.64 (1.28)0.73 (1.50)0.546Daily work related mobile or tablet use time (mean (SD))2.33 (2.82)2.28 (2.73)2.35 (2.84)0.828Daily recreational mobile or tablet use time (mean (SD))2.11 (2.08)2.21 (2.09)2.09 (2.08)0.572Travel duration (mean (SD))42.43 (35.87)46.44 (38.54)41.66 (35.33)0.207Weekly exercise frequency (%)0.9250141 (21.1)20 (18.7)121 (21.6)1≤X≤2 times/week238 (35.6)39 (36.4)199 (35.5)3≤X≤4 times/week129 (19.3)21 (19.6)108 (19.3)≥5 times/week160 (24.0)27 (25.2)133 (23.7)Exercise duration per session (mean (SD))39.73 (36.73)50.38 (46.40)37.70 (34.26)0.001PCS (mean (SD))43.81 (7.72)46.99 (7.78)43.20 (7.57)<0.001MCS (mean (SD))48.73 (9.98)49.31 (8.76)48.62 (10.20)0.509Gender = Female (%)458 (68.6)66 (61.7)392 (69.9)0.119Occupation (%)0.513Office worker300 (44.9)52 (48.6)248 (44.2)Student48 (7.2)7 (6.5)41 (7.3)Manual laborer62 (9.3)12 (11.2)50 (8.9)Freelancer225 (33.7)29 (27.1)196 (34.9)Other33 (4.9)7 (6.5)26 (4.6)Transportation mode (%)0.155Driving155 (23.2)31 (29.0)124 (22.1)Riding in a vehicle178 (26.6)33 (30.8)145 (25.8)Cycling115 (17.2)14 (13.1)101 (18.0)Walking220 (32.9)29 (27.1)191 (34.0)Supine position = Yes (%)361 (54.0)56 (52.3)305 (54.4)0.779Lateral position = Yes (%)381 (57.0)55 (51.4)326 (58.1)0.239Prone position = Yes (%)71 (10.6)15 (14.0)56 (10.0)0.284Backpack use = Yes (%)361 (54.0)64 (59.8)297 (52.9)0.23Double shoulder backpack = Yes (%)113 (16.9)19 (17.8)94 (16.8)0.91Single shoulder backpack = Yes (%)135 (20.2)25 (23.4)110 (19.6)0.45Crossbody bag = Yes (%)101 (15.1)15 (14.0)86 (15.3)0.842Handbag = Yes (%)51 (7.6)9 (8.4)42 (7.5)0.895Cervical curvature = Impaired (%)535 (80.1)83 (77.6)452 (80.6)0.562Intervertebral space = Impaired (%)189 (28.3)38 (35.5)151 (26.9)0.091Osteophyte formation = Yes (%)369 (55.2)61 (57.0)308 (54.9)0.767Range of motion = Impaired (%)148 (22.2)31 (29.0)117 (20.9)0.084Physical work intensity (%)0.088Light402 (60.2)66 (61.7)336 (59.9)Moderate230 (34.4)31 (29.0)199 (35.5)Heavy36 (5.4)10 (9.3)26 (4.6)Mental workload intensity (%)0.659Light270 (40.4)39 (36.4)231 (41.2)Moderate275 (41.2)47 (43.9)228 (40.6)Heavy123 (18.4)21 (19.6)102 (18.2)Backpack weight (%)0.4080307 (46.0)43 (40.2)264 (47.1)≤1 kg175 (26.2)30 (28.0)145 (25.8)>1 kg186 (27.8)34 (31.8)152 (27.1)Pillow height (%)0.466No pillow26 (3.9)6 (5.6)20 (3.6)5–10 cm467 (69.9)69 (64.5)398 (70.9)10–20 cm158 (23.7)28 (26.2)130 (23.2)>20 cm17 (2.5)4 (3.7)13 (2.3)Bed firmness (%)0.943Soft49 (7.3)7 (6.5)42 (7.5)Medium427 (63.9)69 (64.5)358 (63.8)Hard192 (28.7)31 (29.0)161 (28.7)Disease duration (%)<0.001<3 month199 (29.8)50 (46.7)149 (26.6)3–11 month354 (53.0)41 (38.3)313 (55.8)≥12 month115 (17.2)16 (15.0)99 (17.6)Pre-treatment frequency of onset (%)<0.001<3 times/month330 (49.4)70 (65.4)260 (46.3)3–5 times/month169 (25.3)25 (23.4)144 (25.7)≥6 times/month169 (25.3)12 (11.2)157 (28.0)Outcome freq (%)<0.001Never107 (16.0)107 (100.0)0 (0.0)Occasionally370 (55.4)0 (0.0)370 (66.0)Frequently191 (28.6)0 (0.0)191 (34.0) Table 1Baseline Characteristics of Study ParticipantsVariableOverallNon-RecurrenceRecurrenceP valuen668107561Age (mean (SD))50.22 (15.58)50.18 (14.95)50.22 (15.71)0.977BMI (mean (SD))24.29 (4.55)24.56 (5.24)24.24 (4.41)0.498Neck length (mean (SD))13.00 (1.90)13.17 (1.90)12.97 (1.90)0.313Neck circumference (mean (SD))33.99 (3.65)34.57 (3.38)33.88 (3.69)0.077Daily working hours (mean (SD))6.36 (3.76)6.47 (3.62)6.33 (3.79)0.725Daily head down time (mean (SD))4.33 (2.63)4.48 (2.69)4.30 (2.62)0.536Daily work related computer use time (mean (SD))2.51 (3.31)2.78 (3.43)2.46 (3.28)0.365Daily recreational computer use time (mean (SD))0.72 (1.47)0.64 (1.28)0.73 (1.50)0.546Daily work related mobile or tablet use time (mean (SD))2.33 (2.82)2.28 (2.73)2.35 (2.84)0.828Daily recreational mobile or tablet use time (mean (SD))2.11 (2.08)2.21 (2.09)2.09 (2.08)0.572Travel duration (mean (SD))42.43 (35.87)46.44 (38.54)41.66 (35.33)0.207Weekly exercise frequency (%)0.9250141 (21.1)20 (18.7)121 (21.6)1≤X≤2 times/week238 (35.6)39 (36.4)199 (35.5)3≤X≤4 times/week129 (19.3)21 (19.6)108 (19.3)≥5 times/week160 (24.0)27 (25.2)133 (23.7)Exercise duration per session (mean (SD))39.73 (36.73)50.38 (46.40)37.70 (34.26)0.001PCS (mean (SD))43.81 (7.72)46.99 (7.78)43.20 (7.57)<0.001MCS (mean (SD))48.73 (9.98)49.31 (8.76)48.62 (10.20)0.509Gender = Female (%)458 (68.6)66 (61.7)392 (69.9)0.119Occupation (%)0.513Office worker300 (44.9)52 (48.6)248 (44.2)Student48 (7.2)7 (6.5)41 (7.3)Manual laborer62 (9.3)12 (11.2)50 (8.9)Freelancer225 (33.7)29 (27.1)196 (34.9)Other33 (4.9)7 (6.5)26 (4.6)Transportation mode (%)0.155Driving155 (23.2)31 (29.0)124 (22.1)Riding in a vehicle178 (26.6)33 (30.8)145 (25.8)Cycling115 (17.2)14 (13.1)101 (18.0)Walking220 (32.9)29 (27.1)191 (34.0)Supine position = Yes (%)361 (54.0)56 (52.3)305 (54.4)0.779Lateral position = Yes (%)381 (57.0)55 (51.4)326 (58.1)0.239Prone position = Yes (%)71 (10.6)15 (14.0)56 (10.0)0.284Backpack use = Yes (%)361 (54.0)64 (59.8)297 (52.9)0.23Double shoulder backpack = Yes (%)113 (16.9)19 (17.8)94 (16.8)0.91Single shoulder backpack = Yes (%)135 (20.2)25 (23.4)110 (19.6)0.45Crossbody bag = Yes (%)101 (15.1)15 (14.0)86 (15.3)0.842Handbag = Yes (%)51 (7.6)9 (8.4)42 (7.5)0.895Cervical curvature = Impaired (%)535 (80.1)83 (77.6)452 (80.6)0.562Intervertebral space = Impaired (%)189 (28.3)38 (35.5)151 (26.9)0.091Osteophyte formation = Yes (%)369 (55.2)61 (57.0)308 (54.9)0.767Range of motion = Impaired (%)148 (22.2)31 (29.0)117 (20.9)0.084Physical work intensity (%)0.088Light402 (60.2)66 (61.7)336 (59.9)Moderate230 (34.4)31 (29.0)199 (35.5)Heavy36 (5.4)10 (9.3)26 (4.6)Mental workload intensity (%)0.659Light270 (40.4)39 (36.4)231 (41.2)Moderate275 (41.2)47 (43.9)228 (40.6)Heavy123 (18.4)21 (19.6)102 (18.2)Backpack weight (%)0.4080307 (46.0)43 (40.2)264 (47.1)≤1 kg175 (26.2)30 (28.0)145 (25.8)>1 kg186 (27.8)34 (31.8)152 (27.1)Pillow height (%)0.466No pillow26 (3.9)6 (5.6)20 (3.6)5–10 cm467 (69.9)69 (64.5)398 (70.9)10–20 cm158 (23.7)28 (26.2)130 (23.2)>20 cm17 (2.5)4 (3.7)13 (2.3)Bed firmness (%)0.943Soft49 (7.3)7 (6.5)42 (7.5)Medium427 (63.9)69 (64.5)358 (63.8)Hard192 (28.7)31 (29.0)161 (28.7)Disease duration (%)<0.001<3 month199 (29.8)50 (46.7)149 (26.6)3–11 month354 (53.0)41 (38.3)313 (55.8)≥12 month115 (17.2)16 (15.0)99 (17.6)Pre-treatment frequency of onset (%)<0.001<3 times/month330 (49.4)70 (65.4)260 (46.3)3–5 times/month169 (25.3)25 (23.4)144 (25.7)≥6 times/month169 (25.3)12 (11.2)157 (28.0)Outcome freq (%)<0.001Never107 (16.0)107 (100.0)0 (0.0)Occasionally370 (55.4)0 (0.0)370 (66.0)Frequently191 (28.6)0 (0.0)191 (34.0) Baseline Characteristics of Study Participants Logistic Regression In univariate logistic regression analysis, the two groups differed in pre-treatment attack frequency, disease duration, exercise duration per session, and PCS scores (Figure 1, P < 0.05), while cervical length, neck circumference, and intervertebral disc space showed marginal significance (P < 0.1) (Figure 2).Figure 2Forest plot of the univariate logistic regression model. This figure displays the odds ratios (ORs) with 95% confidence intervals (CIs) for each predictor variable in the univariate logistic regression analysis of post-treatment recurrence in patients with cervical spondylosis. Variables with P < 0.05 were considered statistically significant. The horizontal lines represent 95% CIs, and the squares indicate point estimates of the ORs.A forest plot of odds ratio estimates for multiple predictors, with most near 1 and a few elevated.Forest plot listing predictor variables with odds ratio and 95 percent confidence interval, plus P Value. The x-axis label is, Odds Ratio (95 percent CI), with no unit. Scale runs from 0.10, 0.20, 0.50, 1.0, 2.0, 5.00, to 10.00. Each row shows a square point estimate and a horizontal 95 percent confidence interval line. Values shown: Age 1.00 (0.98 to 1.02), P 0.982; BMI 1.00 (0.95 to 1.06), P 0.920; Backpack use 0.76 (0.45 to 1.25), P 0.279; Backpack weight 0.71 (0.39 to 1.31), P 0.520; Bed firmness 1.03 (0.35 to 2.66), P 0.993; Cervical curvature 1.45 (0.79 to 2.59), P 0.223; Crossbody bag 1.03 (0.52 to 2.24), P 0.940; Daily head down time 0.97 (0.89 to 1.07), P 0.567; Daily recreational computer use time 1.02 (0.86 to 1.25), P 0.820; Daily recreational mobile or tablet use time 0.99 (0.88 to 1.12), P 0.824; Daily work related computer use time 0.98 (0.91 to 1.05), P 0.510; Daily work related mobile or tablet use time 1.01 (0.93 to 1.11), P 0.815; Daily working hours 1.00 (0.93 to 1.06), P 0.905; Disease duration 3.62 (2.07 to 6.45), P less than 0.001; Disease duration num 1.01 (0.99 to 1.05), P 0.420; Double shoulder backpack 1.07 (0.56 to 2.18), P 0.849; Exercise duration per session 0.99 (0.99 to 1.00), P 0.041; Gender 1.54 (0.91 to 2.57), P 0.110; Handbag 0.67 (0.31 to 1.63), P 0.362; Intervertebral space 0.61 (0.36 to 1.04), P 0.067; Lateral position 1.46 (0.88 to 2.41), P 0.140; MCS 0.99 (0.96 to 1.01), P 0.319; Mental workload intensity 1.22 (0.59 to 2.67), P 0.704; Neck circumference 0.94 (0.88 to 1.00), P 0.060; Neck length 0.88 (0.77 to 1.01), P 0.071; Occupation 0.59 (0.24 to 1.61), P 0.262; Osteophyte formation 0.80 (0.48 to 1.33), P 0.402; PCS 0.94 (0.90 to 0.97), P less than 0.001; Physical work intensity 1.46 (0.85 to 2.58), P 0.319; Pillow height 1.83 (0.58 to 4.92), P 0.615; Pre treatment frequency of onset 4.84 (2.26 to 12.02), P less than 0.001; Prone position 0.58 (0.29 to 1.21), P 0.139; Range of motion 0.63 (0.37 to 1.13), P 0.122; Single shoulder backpack 0.86 (0.48 to 1.62), P 0.622; Supine position 1.07 (0.65 to 1.77), P 0.792; Transportation mode 1.49 (0.66 to 3.53), P 0.429; Travel duration 1.00 (0.99 to 1.01), P 0.767; Weekly exercise frequency 0.91 (0.46 to 1.75), P 0.961. Forest plot of the univariate logistic regression model. This figure displays the odds ratios (ORs) with 95% confidence intervals (CIs) for each predictor variable in the univariate logistic regression analysis of post-treatment recurrence in patients with cervical spondylosis. Variables with P < 0.05 were considered statistically significant. The horizontal lines represent 95% CIs, and the squares indicate point estimates of the ORs. Subsequently, the seven variables identified in the univariate analysis (P < 0.1) were included in the multivariate logistic regression analysis (Table 2). The results indicated that pre-treatment attack frequency, disease duration, exercise duration per session, PCS score, neck length, and intervertebral disc space were independent risk factors for recurrence of cervical spondylosis. The generalized variance inflation factor (GVIF) assessment and detailed information for all variables are provided in Supplementary Table S2.Table 2Multivariable Logistic Regression ModelVariableLevelOR (95% CI)P valueExercise duration (per session, min)—0.992 (0.985–0.999)0.033PCS—0.943 (0.907–0.979)0.002Neck length (cm)—0.850 (0.729–0.990)0.027Neck circumference (cm)—0.952 (0.883–1.026)0.200Pre-treatment frequency of onset (times/month)<33–51.467 (0.742–2.906)0.265≥64.608 (1.941–10.969)0.001Disease duration (month)<33–112.691 (1.461–4.950)0.001≥122.150 (0.941–4.913)0.074Intervertebral spaceImpairedNormal0.011Impaired0.478 (0.270–0.846) Table 2Multivariable Logistic Regression ModelVariableLevelOR (95% CI)P valueExercise duration (per session, min)—0.992 (0.985–0.999)0.033PCS—0.943 (0.907–0.979)0.002Neck length (cm)—0.850 (0.729–0.990)0.027Neck circumference (cm)—0.952 (0.883–1.026)0.200Pre-treatment frequency of onset (times/month)<33–51.467 (0.742–2.906)0.265≥64.608 (1.941–10.969)0.001Disease duration (month)<33–112.691 (1.461–4.950)0.001≥122.150 (0.941–4.913)0.074Intervertebral spaceImpairedNormal0.011Impaired0.478 (0.270–0.846) Multivariable Logistic Regression Model Nomogram Based on the six independent risk factors identified through multivariate logistic regression analysis, a dynamic interactive nomogram was developed to predict the risk of recurrence in patients with cervical spondylosis. As shown in Figure 3, a participant from the training set was randomly selected to illustrate the prediction of disease recurrence. This patient engaged in 60 minutes of exercise per session, had a PCS score of approximately 50.46, a pre-treatment attack frequency of ≥6 times per month, a disease duration of ≥12 months, a neck length of 12.5 cm, and intervertebral disc space narrowing. According to the predictive model, the patient’s total score was 341, corresponding to a predicted recurrence probability of 0.89. The model indicates a high risk of recurrence for this patient, which is consistent with the patient’s actual clinical outcome.Figure 3Nomogram for predicting recurrence of cervical spondylosis. The nomogram visualizes the contribution of six independent risk factors identified from the multivariate logistic regression model: exercise duration per session, neck length, intervertebral disc space, disease duration, PCS score, and pre-treatment attack frequency. Each predictor corresponds to a point scale at the top, and the total points map to the predicted probability of recurrence at the bottom. Higher total points indicate a higher predicted risk of recurrence. Asterisks indicate the level of statistical significance in the logistic regression model: * P<0.05, ** P<0.01, and *** P<0.001. The number of asterisks reflects increasing significance.A nomogram mapping six predictors to points, total points and recurrence probability for cervical spondylosis.Nomogram predicts cervical spondylosis recurrence using a points system. Predictors include exercise duration, neck length, intervertebral space, disease duration, PCS and attack frequency. Each predictor is mapped to points, contributing to total points, which indicate recurrence probability. Exercise duration ranges from 240 to 0, marked near 40. Neck length ranges from 19 to 7, marked near 13. Intervertebral space is categorized as Impaired or Normal, marked Impaired. Disease duration is categorized as less than 3, 3 to 11 and 12 or more, marked at 12 or more. PCS ranges from 65 to 20, marked at 50. Attack frequency is categorized as less than 3, 3 to 5 and 6 or more, marked at 6 or more. Total points range from 250 to 450, marked at 341. Probability scale runs from 0.2 to 0.995, marked at 0.89. Nomogram for predicting recurrence of cervical spondylosis. The nomogram visualizes the contribution of six independent risk factors identified from the multivariate logistic regression model: exercise duration per session, neck length, intervertebral disc space, disease duration, PCS score, and pre-treatment attack frequency. Each predictor corresponds to a point scale at the top, and the total points map to the predicted probability of recurrence at the bottom. Higher total points indicate a higher predicted risk of recurrence. Asterisks indicate the level of statistical significance in the logistic regression model: * P<0.05, ** P<0.01, and *** P<0.001. The number of asterisks reflects increasing significance. Evaluation and Validation of the Predictive Model Receiver Operating Characteristic (ROC) curves were plotted separately for the two datasets, and the area under the ROC curve (AUC) was calculated to evaluate the discriminative ability of the predictive model. The AUC of the training set was 0.760, exceeding the clinically acceptable threshold of 0.7, confirming that the model has good discriminative performance (Figure 4). Calibration curves indicated that, in the training set, the model closely followed the ideal line in both the low-to-moderate risk range (predicted probability 0.3–0.7) and the high-risk range (>0.8), suggesting high confidence and low bias in predicting low- and high-probability events. Slight overestimation was observed in the intermediate range (0.7–0.8), but the overall impact on calibration was minimal. The validation set showed some systematic deviations: slight overestimation in the low-to-moderate risk range and slight underestimation in the moderate-to-high risk range. Nevertheless, the calibration curve remained approximately parallel to the ideal line, without severe deviations (Figure 5). Decision curve analysis further demonstrated that, within a recurrence risk threshold range of approximately 0.1–0.9, using the model for risk-stratified management yielded a significantly higher net benefit than either the “treat all aggressively” or “treat all conventionally” strategies (Figure 6).Figure 4ROC curves of the training set (a) and validation set (b). The ROC curves plot the true positive rate (sensitivity) against the false positive rate (1-specificity). The left panel shows the ROC curve of the training set, and the right panel shows that of the validation set. The diagonal dashed line represents the performance of a random classifier. Both curves approach the top-left corner, indicating good discriminative ability.Two line graphs showing receiver operating characteristic curves for training and validation sets.Figure 5Calibration curves of the training set (a) and validation set (b). The calibration curves assess the agreement between predicted probabilities (x-axis) and observed probabilities (y-axis). The diagonal dashed line (Ideal) represents perfect calibration. The Apparent curve is directly estimated from the data, while the Bias-corrected curve is obtained via bootstrap resampling to reduce overfitting.A line graph pair showing calibration curves comparing apparent, bias corrected and ideal probabilities.Image A displays a line graph with Predicted Probability (0.3 to 1.0) on the x-axis and Observed Probability (0.3 to 1.0) on the y-axis. The legend includes Apparent, Bias corrected and Ideal curves. The Ideal curve is a diagonal from (0.3, 0.3) to (1.0, 1.0). The Bias corrected curve closely follows the diagonal, rising from (0.3, 0.3) to (0.78, 0.83), dipping to (0.82, 0.82), then increasing to (0.9, 0.9) and (1.0, 1.0). The Apparent curve is similar, with a rise near 0.75 to 0.8. Image B shows a line graph with Predicted Probability (0.5 to 1.0) on the x-axis and Observed Probability (0.5 to 1.0) on the y-axis. The Ideal curve is a diagonal from (0.5, 0.5) to (1.0, 1.0). The Bias corrected curve starts at (0.55, 0.56), rises through (0.6, 0.6) and (0.7, 0.75), peaks at (0.8, 0.83), dips near (0.84, 0.82), then increases to (0.9, 0.88) and approaches (1.0, 0.97). The Apparent curve is below the Bias corrected curve from 0.55 to 0.7, then aligns closer from 0.75 to 1.0.Figure 6Decision curve analysis (DCA) for the training set (a) and validation set (b). Decision curve analysis evaluates the clinical net benefit of a predictive model across different high-risk thresholds. The x-axis represents the high-risk threshold, and the y-axis shows the standardized net benefit. Three curves are displayed: the Model curve (net benefit of the proposed prediction model), the All curve (net benefit of intervening on all patients), and the None curve (net benefit of intervening on no patients, always zero).Two line graphs showing standardized net benefit versus high risk threshold for Model, All and None.Image A and B both depict line graphs with legends: Model, All, None. The x-axis is labeled High Risk Threshold (0.0 to 1.0) and the y-axis is Standardized Net Benefit (0.0 to 1.0). Below the x-axis, a secondary label shows Cost:Benefit Ratio with ticks: 1:80, 1:4, 2:3, 3:2, 4:1, 80:1. In both images, the None series is a horizontal line at y=0.0 from x=0.0 to 1.0. The Model and All curves begin near y=1.0 at x=0.0, decreasing towards y=0.0 as x nears 1.0. Image B shows jagged fluctuations in the Model and All curves near high risk threshold values of 0.8 to 1.0. ROC curves of the training set (a) and validation set (b). The ROC curves plot the true positive rate (sensitivity) against the false positive rate (1-specificity). The left panel shows the ROC curve of the training set, and the right panel shows that of the validation set. The diagonal dashed line represents the performance of a random classifier. Both curves approach the top-left corner, indicating good discriminative ability. Calibration curves of the training set (a) and validation set (b). The calibration curves assess the agreement between predicted probabilities (x-axis) and observed probabilities (y-axis). The diagonal dashed line (Ideal) represents perfect calibration. The Apparent curve is directly estimated from the data, while the Bias-corrected curve is obtained via bootstrap resampling to reduce overfitting. Decision curve analysis (DCA) for the training set (a) and validation set (b). Decision curve analysis evaluates the clinical net benefit of a predictive model across different high-risk thresholds. The x-axis represents the high-risk threshold, and the y-axis shows the standardized net benefit. Three curves are displayed: the Model curve (net benefit of the proposed prediction model), the All curve (net benefit of intervening on all patients), and the None curve (net benefit of intervening on no patients, always zero). Supplementary Analyses Analysis of Episode Frequency Generalized ordinal logistic regression was used to analyze factors affecting symptom frequency (never recurrence, occasional recurrence, frequent recurrence) (Table 3). The results showed that longer exercise duration reduced the risk of disease recurrence (OR = 0.992), but had little effect on the transition from no recurrence or occasional recurrence to frequent recurrence. Higher PCS scores were associated with lower risk of both types of recurrence (OR ≈ 0.93), indicating that better physical condition reduces the likelihood of recurrence. Higher pre-treatment attack frequency was associated with greater recurrence risk, particularly for patients with ≥6 episodes per month, whose risk of “frequent” recurrence was more than five times that of the reference group. Longer disease duration also increased the likelihood of recurrence, with patients having a disease duration ≥12 months exhibiting a 2.6-fold higher risk of “frequent” recurrence compared to those with shorter disease duration. Interestingly, abnormal intervertebral disc space was associated with a higher likelihood of not experiencing recurrence. Neck length, however, showed no association with recurrence frequency.Table 3Generalized Ordinal Logistic Regression ModelVariableLevelComparison GroupOR (95% CI)P-valueExercise duration (per session, min)—Never vs ≥Occasionally0.992 (0.987–0.997)0.004—≤Occasionally vs Frequently0.998 (0.993–1.003)0.499PCS score—Never vs ≥Occasionally0.936 (0.907–0.966)<0.001—≤Occasionally vs Frequently0.931 (0.908–0.953)<0.001Neck length (cm)—Never vs ≥Occasionally0.927 (0.825–1.041)0.202—≤Occasionally vs Frequently1.050 (0.953–1.156)0.322Pre-treatment frequency of onset (times/month)<33–5Never vs ≥Occasionally1.221 (0.724–2.060)0.453≤Occasionally vs Frequently2.096 (1.327–3.312)0.002≥6Never vs ≥Occasionally3.215 (1.638–6.311)0.001≤Occasionally vs Frequently5.145 (3.305–8.010)<0.001Disease duration (month)<33–11Never vs ≥Occasionally2.072 (1.278–3.360)0.003≤Occasionally vs Frequently1.936 (1.217–3.080)0.005≥12Never vs ≥Occasionally1.968 (1.015–3.815)0.045≤Occasionally vs Frequently2.585 (1.453–4.600)0.001Intervertebral spaceReference: NormalImpairedNever vs ≥Occasionally0.566 (0.351–0.914)0.020≤Occasionally vs Frequently0.964 (0.639–1.456)0.862Note: The model was specified with reverse = TRUE, modeling the cumulative probability P(Y≥j). Therefore, OR > 1 indicates that an increase in the predictor is associated with higher odds of being in a higher-frequency category (occasional or frequent recurrence). The reference comparisons are cumulative: “Never vs ≥Occasional” compares “no recurrence” with “occasional or frequent recurrence”, and “≤Occasional vs Frequent” compares “no or occasional recurrence” with “frequent recurrence”. Table 3Generalized Ordinal Logistic Regression ModelVariableLevelComparison GroupOR (95% CI)P-valueExercise duration (per session, min)—Never vs ≥Occasionally0.992 (0.987–0.997)0.004—≤Occasionally vs Frequently0.998 (0.993–1.003)0.499PCS score—Never vs ≥Occasionally0.936 (0.907–0.966)<0.001—≤Occasionally vs Frequently0.931 (0.908–0.953)<0.001Neck length (cm)—Never vs ≥Occasionally0.927 (0.825–1.041)0.202—≤Occasionally vs Frequently1.050 (0.953–1.156)0.322Pre-treatment frequency of onset (times/month)<33–5Never vs ≥Occasionally1.221 (0.724–2.060)0.453≤Occasionally vs Frequently2.096 (1.327–3.312)0.002≥6Never vs ≥Occasionally3.215 (1.638–6.311)0.001≤Occasionally vs Frequently5.145 (3.305–8.010)<0.001Disease duration (month)<33–11Never vs ≥Occasionally2.072 (1.278–3.360)0.003≤Occasionally vs Frequently1.936 (1.217–3.080)0.005≥12Never vs ≥Occasionally1.968 (1.015–3.815)0.045≤Occasionally vs Frequently2.585 (1.453–4.600)0.001Intervertebral spaceReference: NormalImpairedNever vs ≥Occasionally0.566 (0.351–0.914)0.020≤Occasionally vs Frequently0.964 (0.639–1.456)0.862Note: The model was specified with reverse = TRUE, modeling the cumulative probability P(Y≥j). Therefore, OR > 1 indicates that an increase in the predictor is associated with higher odds of being in a higher-frequency category (occasional or frequent recurrence). The reference comparisons are cumulative: “Never vs ≥Occasional” compares “no recurrence” with “occasional or frequent recurrence”, and “≤Occasional vs Frequent” compares “no or occasional recurrence” with “frequent recurrence”. Generalized Ordinal Logistic Regression Model Note: The model was specified with reverse = TRUE, modeling the cumulative probability P(Y≥j). Therefore, OR > 1 indicates that an increase in the predictor is associated with higher odds of being in a higher-frequency category (occasional or frequent recurrence). The reference comparisons are cumulative: “Never vs ≥Occasional” compares “no recurrence” with “occasional or frequent recurrence”, and “≤Occasional vs Frequent” compares “no or occasional recurrence” with “frequent recurrence”. Analysis of Time to Recurrence In the subgroup of patients who experienced recurrence, linear regression analysis was performed with time to recurrence (in days) as the dependent variable, as shown in Table 4. Among the predictors, exercise duration was significantly positively correlated with time to recurrence, with each additional minute of exercise per session delaying recurrence by an average of approximately 0.55 days. Disease duration of 3–11 months and ≥12 months was significantly negatively correlated with time to recurrence, indicating that longer disease duration led to earlier recurrence. PCS scores showed a positive trend but did not reach statistical significance. pre-treatment attack frequency, neck length, and intervertebral disc space status were not significantly associated with time to recurrence in the recurrent subgroup (P > 0.05), although the directions were generally consistent with the main analysis. Notably, the effect direction of pre-treatment attack frequency differed from that expected in the main analysis, possibly due to the small sample size of the subgroup or the influence of extreme values. Neck length was a significant protective factor in the main analysis but showed no association with time to recurrence.Table 4Linear Regression ModelVariableLevelβ (95% CI)P-valueExercise duration (per session, min)—0.55 (0.13, 0.98)0.011PCS score—1.75 (−0.18, 3.69)0.076Neck length (cm)—–0.50 (–8.18, 7.19)0.899Pre-treatment frequency of onset (times/month)<33–531.25 (–4.51, 67.00)0.087≥61.99 (–33.30, 37.30)0.912Disease duration (month)<33–11–40.56 (–75.20, –5.92)0.022≥12–49.21 (–94.50, –3.93)0.033Intervertebral spaceNormalImpaired24.38 (–9.30, 58.10)0.156 Table 4Linear Regression ModelVariableLevelβ (95% CI)P-valueExercise duration (per session, min)—0.55 (0.13, 0.98)0.011PCS score—1.75 (−0.18, 3.69)0.076Neck length (cm)—–0.50 (–8.18, 7.19)0.899Pre-treatment frequency of onset (times/month)<33–531.25 (–4.51, 67.00)0.087≥61.99 (–33.30, 37.30)0.912Disease duration (month)<33–11–40.56 (–75.20, –5.92)0.022≥12–49.21 (–94.50, –3.93)0.033Intervertebral spaceNormalImpaired24.38 (–9.30, 58.10)0.156 Linear Regression Model Discussion This study identified associations rather than causal determinants of post-treatment recurrence in patients with cervical spondylosis, providing new evidence for individualized risk prediction and long-term management. It is important to distinguish between risk factors for initial onset and predictors of post-treatment recurrence, as these may involve different mechanisms. Many previously identified factors, such as age, occupation, and poor posture, are primarily associated with the occurrence of neck pain rather than relapse after conservative treatment. Recurrence is more likely influenced by disease chronicity, prior episode frequency, and post-remission behaviors. In this study, we specifically focused on predictors of recurrence after remission, which may require different preventive strategies from those targeting initial onset. Our results demonstrated that exercise duration per session, neck length, intervertebral disc space status, disease duration, pre-treatment attack frequency, and physical health (PCS score) were independently associated with recurrence. These findings suggest that recurrence is associated with multiple factors, including patients’ physical condition, disease chronicity, and prior disease activity, rather than being determined by any single cause. The overall recurrence rate in this cohort was 84.0% (561/668), substantially higher than previously reported in the literature. This high rate may be attributed to several factors. First, our definition of recurrence was broad, encompassing any symptom reappearance requiring medical consultation or physician confirmation, without requiring a minimum symptom-free interval or severity threshold, which may have included mild or transient episodes not counted as recurrence in other studies. Second, retrospective self-reported data may have introduced recall bias, as patients with more severe or recent symptoms may have been more likely to report past episodes, leading to overestimation. Third, our study population may represent a subgroup with more active disease—all participants had received hospital-based conservative treatment and volunteered for a cervical health survey, potentially enriching the cohort with more symptomatic or health-conscious individuals. Despite this high rate, the model maintained good discriminative ability (AUC=0.760) and calibration in the training set, suggesting reliability even in this high-risk population. Future prospective studies with standardized recurrence definitions and objective follow-up are warranted to confirm our findings. Among these factors, exercise duration per session was significantly associated with reduced recurrence risk and prolonged time to recurrence, consistent with previous studies suggesting that regular exercise may enhance cervical spine stability and musculoskeletal health.13–16 However, this finding should be interpreted with caution. Exercise duration was self-reported and measured at the time of the survey, potentially after recurrence had already occurred. Therefore, we cannot conclude that longer exercise duration directly reduces recurrence risk; it may instead serve as a marker of better baseline health, health literacy, socioeconomic status, treatment adherence, or access to rehabilitation, rather than representing an independent therapeutic factor. The observation that exercise frequency was not significant in the model, whereas duration per session was retained, warrants further investigation and may reflect measurement error, collinearity, categorization, or chance. Notably, the effect of exercise on progression from occasional to frequent recurrence was modest, suggesting that additional interventions may be needed for patients with high baseline disease activity. Future studies using objective physical activity monitoring are warranted to clarify these relationships. Furthermore, longer disease duration and higher pre-treatment attack frequency were associated with increased recurrence risk and earlier relapse. These findings suggest that chronic and recurrent episodes of cervical spondylosis, through prolonged mechanical stress and repeated symptoms, may induce structural and functional changes in cervical tissues, exacerbate degenerative changes, and reduce cervical spine resilience, thereby increasing susceptibility to subsequent episodes, which is consistent with previous studies.9,17–19 Clinically, these results underscore the importance of early intervention and proactive monitoring, particularly for patients with long-standing or frequently recurring disease. Overall physical health status also emerged as a factor associated with recurrence. Lower PCS scores, which reflect poorer physical condition including reduced muscle strength, flexibility, and postural control, were associated with a higher risk of recurrence.17,20,21 However, this association should be interpreted with caution. PCS scores were measured at the time of the survey, which may have been after recurrence had already occurred. It is possible that lower PCS scores partly reflect current symptoms and functional limitations related to the recurrence episode itself, rather than representing a pre-existing risk factor. This creates a potential risk of outcome-predictor overlap or reverse causation. Nevertheless, the observed association highlights the need for comprehensive rehabilitation programs that address not only cervical symptoms but also general physical fitness and functional capacity to improve long-term outcomes. Interestingly, intervertebral disc space abnormalities were found to be negatively associated with recurrence, suggesting that structural degeneration does not necessarily translate into a higher risk of relapse. One possible explanation is that the presence of obvious radiographic abnormalities may serve as a “teachable moment”, prompting patients to proactively adopt protective behaviors—such as increasing exercise, optimizing sitting and working postures, or using cervical support measures—thereby indirectly reducing recurrence risk.22 However, this hypothesis is not directly supported by the data presented in our study and requires further validation. Alternative explanations, such as unmeasured confounding, reverse causation, or selection bias, cannot be excluded. Neck length was protective in the primary analysis but showed no significant association with time to recurrence, suggesting that anatomical factors may influence recurrence indirectly, potentially through biomechanical effects on cervical load distribution and posture habits.23,24 In addition, we examined the associations of electronic device use and head-down duration with recurrence after cervical spine treatment and found no significant associations. Several explanations may account for these null findings. First, risk factors for disease onset may differ from those for post-treatment recurrence; although forward-neck posture and excessive screen time are important for initial development,25 their impact on relapse after remission may be weaker, possibly due to differences in underlying pathophysiological mechanisms. Second, self‑reported exposure measures are subject to recall bias, which may attenuate true associations. Third, our broad definition of recurrence included many mild and transient episodes, potentially diluting the influence of posture or device use. Finally, categorization of exposure variables may have reduced statistical power and obscured dose–response relationships. Future studies using objective posture monitoring and a stricter recurrence definition are warranted to further clarify these relationships. Limitations Several limitations should be acknowledged. First, recurrence was self-reported without a predefined symptom-free interval, standardized diagnostic criteria, or uniform follow-up, which may introduce recall bias and outcome misclassification, and the unequal time at risk across participants was not accounted for in binary logistic regression. Second, predictor variables were measured after remission while recurrence was retrospectively recalled, so temporal sequencing cannot be established and only associations can be inferred. Third, self-reported lifestyle factors, disease duration, and pre-treatment attack frequency are subject to recall bias and imprecise categorization, which may have masked dose–response relationships. Fourth, the small number of non-recurrent participants may have affected subgroup analyses. Fifth, conservative treatment details were not systematically collected, limiting assessment of specific treatment effects. Sixth, stepwise variable selection may overestimate model performance, and external validation is needed. Finally, imaging was limited to radiographs; MRI could provide additional soft-tissue evaluation. Prospective studies with objective measures and standardized protocols are warranted to validate our findings. Conclusion In conclusion, this study identified several factors associated with post-treatment recurrence in patients with cervical spondylosis. Exercise duration, disease chronicity, pre-treatment attack frequency, overall physical health, and certain anatomical characteristics were found to be associated with recurrence risk. However, as discussed above, these associations should not be interpreted as causal determinants. In particular, exercise duration, although associated with reduced recurrence risk in our analysis, cannot be confirmed as an independent protective factor based on the current data, given the potential for reverse causation, confounding, and the self-reported nature of the exposure. Intervertebral disc space abnormalities showed a negative association with recurrence in our analysis, which may be related to patient behavioral responses to radiographic findings, but this finding is speculative and requires further validation. These findings underscore the importance of individualized risk stratification, early intervention, and comprehensive rehabilitation to reduce recurrence and improve long-term cervical spine health. Prospective studies with objective measurements and standardized definitions of recurrence are warranted to confirm these associations and establish causal relationships.