1 Introduction
Lumbar spondylolysis is a bony defect located between the posterior elements of the vertebral body and the articular processes and is typically bridged by fibrous, cartilaginous, or osseous tissue.[
1,
2] Approximately 80% of individuals with lumbar spondylolysis remain asymptomatic, and the condition is often identified incidentally or during postmortem examinations; when symptoms occur, low back pain is the predominant clinical manifestation.[
3,
4] Epidemiological studies have reported an overall prevalence of approximately 6% in the general population, with most estimates ranging from 3% to 10%.[
5,
6] The incidence varies substantially across ethnic groups and occupational populations, reaching up to 50% among Inuit populations and between 11.0% and 52.6% in young athletes.[
7,
8]
The L5 vertebra is the most frequently affected level, accounting for more than 90% of reported cases.[
9,
10] When bilateral spondylolysis occurs at L5, approximately 70% of patients subsequently develop lumbar spondylolisthesis, underscoring the clinical importance of early and accurate diagnosis.[
11]
In routine practice, lumbar radiography, including anteroposterior, lateral, and dynamic views, remains the most commonly used initial imaging modality for suspected lumbar spondylolysis.[
12,
13] However, its diagnostic accuracy is often compromised by facet joint hypertrophy and obscuration by the iliac crest, particularly at the L5 level, which can lead to missed or delayed diagnosis.[
14] Computed tomography (CT), which provides multiplanar visualization in coronal, sagittal, and axial planes, is widely regarded as the diagnostic reference standard.[
1] Nevertheless, CT is associated with relatively high radiation exposure, longer examination times, and increased cost.[
15]
In clinical practice, patients with L5 spondylolysis often appear to require a greater pedicle screw insertion angle during surgical fixation, suggesting the presence of distinct morphological characteristics. Therefore, this study aimed to elucidate radiographic and morphological features associated with bilateral L5 spondylolysis and to develop an X-ray-based screening model capable of improving radiographic risk stratification and supporting decisions regarding further CT evaluation.
2 Methods
2.1 Study design and patient selection
This cross-sectional diagnostic screening study systematically analyzed patients hospitalized between 2018 and 2025. The study was reported in accordance with relevant observational and diagnostic-accuracy reporting principles where applicable.
The inclusion criteria were as follows: availability of lumbar spine radiographs, including anteroposterior, lateral, and dynamic views; and completion of lumbar CT examination. The exclusion criteria were as follows: non-L5 spondylolysis; rheumatoid or immune-related disease; history of malignancy; multilevel spondylolysis or unilateral L5 spondylolysis; spinal scoliosis, infection, or previous fracture; prior lumbar surgery; and age younger than 18 years.
Patients with unilateral L5 spondylolysis, multilevel spondylolysis, or non-L5 defects were excluded to maintain a morphologically homogeneous bilateral L5 cohort. Defect laterality and involved vertebral level may have different biomechanical patterns and radiographic manifestations; therefore, the resulting model was intended for bilateral L5 spondylolysis and should not be directly generalized to unilateral or multilevel disease.
Based on CT findings, patients were classified into two groups. Individuals with CT-confirmed defects of the pars interarticularis were assigned to the spondylolysis group, whereas those without such defects were classified as controls.
2.2 Ethics approval and consent to participate
This study was approved by the Institutional Review Board and Ethics Committee of The Seventh Affiliated Hospital, Sun Yat-sen University, Shenzhen, China (Approval No. KY-2025-563-01, Date: December 10, 2025). The requirement for informed consent was waived by the Medical Ethics Committee of the Seventh Affiliated Hospital of Sun Yat-sen University. All procedures were conducted in accordance with applicable institutional guidelines and the Declaration of Helsinki.
2.3 Data collection and imaging measurements
Demographic variables included age, sex, body mass index (BMI), and histories of diabetes mellitus and hypertension.
Radiographic parameters measured on lumbar X-rays are illustrated in Figure 1 and included intervertebral joint space (IJS) on the anteroposterior view, lamina inclination angle (LIA) on the lateral view, sacral inclination angle (SI), sacral slope angle (SS), L5 lumbar index (a/b ratio), degree of spondylolisthesis, and lumbar instability when assessable on dynamic radiographs. The candidate variables for the X-ray screening model were restricted to parameters obtainable from standard radiographs.
CT-based parameters were measured using the Magerl method and included the presence of L5 spondylolysis, pedicle width, intended screw depth, and pedicle internal inclination angle (Fig. 2).[
16] These CT-based variables were analyzed as morphological and surgically relevant features but were not included in the final X-ray-only screening nomogram.
All measurements were independently performed by two observers, with repeated measurements obtained at a 1-week interval. Measurement reliability was assessed using intraclass correlation coefficients (ICCs).[
17] ICC values greater than 0.75 were considered indicative of good reproducibility.[
18]
2.4 Propensity score matching
To minimize confounding, 1:1 propensity score matching (PSM) was performed.[
19] Propensity scores were calculated using logistic regression models incorporating age, sex, BMI, diabetes mellitus, and hypertension. Matching was conducted with a caliper width of 20% of the standard deviation of the propensity score, with random selection among individuals sharing identical scores.[
20]
2.5 Statistical analysis
Continuous variables were analyzed using Student's t test or the Wilcoxon rank-sum test, as appropriate. Categorical variables were compared using the chisquare test. Variables significant in univariable analysis were entered into multivariable logistic regression models. Model discrimination was evaluated using receiver operating characteristic (ROC) curves and the area under the curve (AUC). Optimal cutoff values were determined using the Youden index.
Nomograms, calibration plots, ROC curves, decision curve analysis (DCA), and clinical impact curves (CICs) were constructed using R packages including rms, regplot, and rmda. Internal validation of discrimination and calibration was performed using bootstrap resampling with 1,000 iterations, as described previously.[
21]
For model reproducibility, the intercept and beta coefficients of the final X-ray screening model were reported. Multicollinearity among SI, SS, L5 LIA, and the L5 lumbar index was assessed using variance inflation factors (VIFs). Sensitivity analyses were performed by fitting the final X-ray model in the unmatched development cohort and in a subgroup without radiographic spondylolisthesis (grade 0). Threshold-based exploratory metrics included sensitivity, specificity, observed positive predictive value, observed negative predictive value, missed cases, and the proportion of patients classified below the threshold. Because PSM and case–control sampling alter disease prevalence, model-estimated probabilities, calibration, decision-analytic indices, and CT-reduction estimates were interpreted cautiously.
2.6 External validation
An independent external validation dataset was assembled from an independent tertiary referral center. Patients enrolled between January 2024 and January 2025 who underwent both standard lumbar radiography and lumbar CT within a defined time interval were screened. For external validation, a case–control dataset was constructed according to CT-confirmed outcome status and consisted of 200 patients, including 100 patients with L5 spondylolysis and 100 controls without pars interarticularis defects on CT.
The external validation dataset was entirely independent of the development dataset and was not involved in variable selection, model fitting, or threshold optimization. All variables required for the nomogram were complete, and no imputation was required. External validation followed a strict model-application-only approach: the original development-cohort regression coefficients were applied without modification, and no variable reselection, model refitting, or model updating was performed.
Because the external validation dataset was assembled using case–control sampling, external model performance was assessed primarily in terms of discrimination using the AUC, which is equivalent to the C-index for a binary outcome. Calibration, decision-analytic measures, positive and negative predictive values, and CT-reduction estimates in this setting were considered exploratory and sample-specific and were not used as the primary basis for model validation.
3 Results
3.1 Study population and propensity score matching
After screening, 151 eligible patients were included in the development dataset. Before PSM, the spondylolysis and control groups differed significantly in age (p = 0.017) and hypertension prevalence (p = 0.034), whereas sex, BMI, and diabetes mellitus did not differ significantly between groups.
Following 1:1 PSM, 114 patients were included in the matched cohort, with 57 patients in each group. Baseline demographic and clinical characteristics were well balanced after matching (Table 1). Measurement reliability was excellent, with ICCs ranging from 0.855 to 0.959 across all radiographic and CT-based parameters.
3.2 Radiographic and morphological comparisons
In the matched cohort, univariable analysis demonstrated that patients with L5 spondylolysis had significantly larger SS, L5 LIA, and pedicle internal inclination angle, as well as significantly smaller SI, IJS, and L5 lumbar index values than controls (Table 2).
Multivariable logistic regression further identified smaller SI and L5 lumbar index, together with larger SS, L5 LIA, and pedicle internal inclination angle, as factors independently associated with L5 spondylolysis (Table 2). These associations should not be interpreted as evidence of temporal or causal effects.
3.3 Diagnostic performance of individual radiographic parameters
ROC analysis was performed to evaluate the discriminative ability of individual radiographic parameters. L5 LIA demonstrated the strongest discrimination, with an AUC of 0.879 (95% CI, 0.813–0.944). This was followed by the L5 lumbar index (AUC = 0.771), SS (AUC = 0.637), and SI (AUC = 0.611) (Fig. 3).
3.4 Development and internal validation of the nomogram
Based on the multivariable analysis and the need to maintain an X-ray-only tool, the final screening nomogram incorporated the L5 lumbar index, SI, SS, and L5 LIA (Fig. 4). CT-based pedicle internal inclination angle was analyzed separately as a morphological feature and was not included in the nomogram. The final logistic equation was: logit(p) = -12.415 - 0.151 × SI + 0.114 × SS + 0.194 × L5 LIA - 10.673 × L5 lumbar index, where p represents the model-estimated probability of bilateral L5 spondylolysis in the study sample.
The VIFs for SI, SS, L5 LIA, and the L5 lumbar index were 1.66, 1.85, 1.47, and 1.25, respectively, indicating no substantial multicollinearity. Internal validation using bootstrap resampling demonstrated good model performance. The bootstrap-corrected concordance index was 0.911, and ROC analysis of the nomogram yielded an AUC of 0.916 (95% CI, 0.865–0.967), indicating strong discriminative ability. DCA and CIC analysis showed greater apparent net benefit than individual parameters within the development sample (Fig. 5), but these decision-analytic findings should be interpreted cautiously because matching alters the underlying prevalence.
At the development-cohort Youden threshold of 0.623, the X-ray screening model showed a sensitivity of 0.789 and specificity of 0.912, with 12 missed cases among 57 CT-confirmed cases and 56.1% of patients classified below the threshold. In sensitivity analyses, the model retained strong discrimination in the unmatched development cohort (AUC = 0.927; 95% CI, 0.885–0.968) and in the subgroup without radiographic spondylolisthesis (grade 0; AUC = 0.896; 95% CI, 0.806–0.986).
3.5 External validation results
The external validation dataset consisted of 200 patients from an independent institution, including 100 patients with CT-confirmed L5 spondylolysis and 100 controls. All variables required for the nomogram, including L5 LIA, SI, SS, and L5 lumbar index, were complete, and no missing data were identified (Fig. 6).
In the external validation dataset, the nomogram demonstrated strong discriminative performance when the original development-cohort coefficients were applied without refitting. ROC analysis yielded an AUC of 0.949 (95% CI, 0.922–0.977). Because the outcome was binary, the C-index was identical to the AUC (0.949), indicating the strong ability of the model to distinguish patients with L5 spondylolysis from controls in an independent external dataset.
At the development-cohort Youden threshold of 0.623, the external dataset showed a sensitivity of 0.830 and specificity of 0.910, with 17 missed cases among 100 CT-confirmed cases and 54.0% of patients classified below the threshold. At a lower threshold of 0.20, sensitivity increased to 0.970, specificity was 0.570, three cases were missed, and 30.0% of patients were classified below the threshold. These threshold-based estimates are exploratory because the external validation dataset used case–control sampling and therefore does not represent the disease prevalence or CT-use patterns of an unselected clinical population. Exploratory calibration and decision-analytic plots are shown in Figure 7 for descriptive completeness only and were not used as primary external validation endpoints.
4 Discussion
This study evaluated LIA as a radiographic marker associated with bilateral L5 spondylolysis. Patients with L5 spondylolysis had significantly larger L5 LIA values than controls, consistent with prior findings.[
22] This phenomenon may be attributable to altered pedicle-facet geometry and mechanical loading at the pars interarticularis; however, the present cross-sectional design cannot determine whether these features precede the pars defect or represent adaptive remodeling after spondylolysis has developed. Biomechanical studies have shown that bilateral pars defects increase motion at L5–S1 while altering kinematics at adjacent segments, supporting the possibility that compensatory adaptations may occur away from the defect site.[
23–
25]
The surgical implications are also noteworthy. Anatomical distortion due to osteophyte formation and pseudoarthrosis may complicate pedicle screw placement, and navigation technologies have been developed to improve screw placement accuracy in challenging anatomy.[
26,
27] In the present study, patients with L5 spondylolysis had a significantly larger pedicle internal inclination angle. This finding should be interpreted as an associated morphological difference and potential adaptive consequence rather than a proven causal risk factor, but it may still be useful during preoperative planning.
Alterations in pelvic sagittal parameters play a critical role in spinal loading.[
28–
32] Increased sacral slope has been associated with spondylolisthesis progression, whereas lower SI and related sacral table morphology have been linked to spondylolysis risk.[
33–
37] These findings are consistent with the present results and underscore the biomechanical relevance of spinopelvic alignment in L5 spondylolysis.
Radiographic diagnosis of spondylolysis remains challenging, with limited sensitivity even when oblique views are added.[
38–
41] The nomogram developed in this study integrates readily obtainable radiographic parameters to provide individualized screening and risk stratification. Clinically, this model may help identify patients who warrant CT confirmation, but it should be used as a triage aid rather than as a replacement for CT when a definitive diagnosis is clinically required. Any claim regarding the reduction of unnecessary CT examinations requires prospective validation in consecutive clinical populations.
This study has several limitations. First, its cross-sectional design precludes causal inference; therefore, the radiographic and CT-based variables should be interpreted as associated features rather than proven etiologic risk factors. Second, only bilateral L5 spondylolysis was included. This restriction reduced heterogeneity and matched the intended model target, but it limits generalizability to unilateral, multilevel, or non-L5 spondylolysis. Third, PSM created a balanced analytical cohort but also altered the natural disease prevalence, limiting interpretation of model-estimated probabilities, calibration, positive and negative predictive values, and clinical utility. Fourth, although external validation was performed in an independent institution, the external validation dataset was assembled using a case-control sampling framework rather than a naturally occurring clinical cohort. Therefore, external validation primarily supports the discriminative performance of the model, whereas calibration, decision-analytic results, and possible reduction of CT use should be interpreted with caution. Fifth, spondylolisthesis may influence sagittal parameters and local morphology. Although sensitivity analysis in patients without radiographic spondylolisthesis showed preserved discrimination, future studies should prospectively evaluate the influence of spondylolisthesis severity and dynamic lumbar instability in consecutive populations.
5 Conclusions
This study identified key radiographic and morphological features associated with bilateral L5 spondylolysis and developed an X-ray-based screening nomogram for radiographic risk stratification. By integrating LIA, SI, SS, and the L5 lumbar index, the model demonstrated strong discriminative performance in the development cohort and retained strong discriminative performance in an independent external dataset when the original development-cohort coefficients were applied without refitting.
The proposed nomogram is a noninvasive, low-cost, and readily applicable radiographic screening aid for bilateral L5 spondylolysis. Its clinical role should be considered supportive: it may help stratify patients for further imaging evaluation, but definitive diagnosis and any CT-reduction strategy require prospective validation in consecutive clinical cohorts.
The Author(s) 2026. Published by Wolters Kluwer Health, LLC. on behalf of Higher Education Press. This is an open access article distributed under the Creative Commons Attribution License 4.0 (CCBY), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.