Development and validation of a nomogram to predict the achievement of minimal clinically important difference following percutaneous endoscopic lumbar discectomy

Abstract

Background

This study is devoted to developing a nomogram predicting the achievement of minimal clinically important difference (MCID) in patients with lumbar disc herniation (LDH) following percutaneous endoscopic lumbar discectomy (PELD).

Methods

The patients involved were followed up for at least 2 years. Univariate logistic analysis and multivariable logistic regression were applied for identifying factors significantly influencing the achievement of MCID. Based on the selected factors, a nomogram was developed using R (v4.4.2). Discriminative ability, calibration, and clinical utility of the nomogram were evaluated by receiver operating characteristic curve (ROC), calibration curve, and decision curve analysis (DCA), respectively.

Results

The study involved 442 patients, of which 23 (5.20%) failed to achieve the MCID. The statistical analysis identified the baseline Visual Analog Scale (VAS) score, presence of Lasegue’s sign, and fat infiltration rate (FIR) of the multifidus muscle as independent risk factors. The nomogram showed good discrimination in development (AUC 0.84, 95% CI 0.73–0.95) and internal validation (AUC 0.88, 95% CI 0.69–0.98). Calibration was assessed by Hosmer–Lemeshow test and calibration curves, with p-values of 0.536 (development) and 0.369 (internal validation), and mean absolute errors (MAE) found (development) and 0.026 (internal validation). Decision curve analysis suggested positive net benefit across threshold probabilities of ~ 52–98% (development) and > 68% (internal validation).

Conclusions

Baseline VAS scores, presence of Lasegue’s sign, and FIR of multifidus muscles are predictive factors for achieving the MCID in patients undergoing PELD.

Significance

The study developed and validated a nomogram that can predict the achievement of MCID following PELD by assessing preoperative risk factors in patients with lumbar disc herniation.

Introduction

Lumbar disc herniation (LDH) is one of the most common spinal ailments, typically presenting as lower back pain (LBP) and/or sciatica, which can significantly impair patients' quality of life. For patients who fail to respond to conservative treatments, surgical interventions such as standard discectomy and microendoscopic discectomy (MED) have traditionally been used to alleviate clinical symptoms [1]. However, with the advancements in endoscopic techniques, the past several decades have witnessed a marked increase in the use of percutaneous endoscopic lumbar discectomy (PELD) [2]. This minimally invasive technique is favored for its numerous advantages, including reduced surgical trauma, shorter operative time, shorter length of hospital stay and faster return to work [3–6].

Among the various approaches to evaluating surgical outcomes, patient-reported outcome measures (PROMs) have gained prominence. These tools, which directly capture patients' perspectives, are highly valued for assessing the efficacy of surgical interventions. Commonly used PROMs in spine surgery include the Visual Analog Scale (VAS), the Oswestry Disability Index (ODI), and the Patient Satisfaction Index (PSI) [7]. Nevertheless, despite their convenience and clinical relevance [8, 9], no definitive cutoff has been established that links changes in these scores to a “satisfactory” surgical result [10, 11]. To address this limitation, the conception of the minimum clinically important difference (MCID) has been proposed [12]. MCID establishes a clinically meaningful threshold for changes in PROM scores before and after surgery, facilitating a standardized and statistically robust evaluation of the procedure.

Nomograms, as emerging predictive models, have gained widespread application in the prognosis of cancer and cardiovascular diseases, as well as in orthopedics and sports medicine. Several studies have developed predictive models for residual symptoms [13, 14] and recurrence of LDH following PELD [15]. However, to date, to our knowledge, few studies have specifically developed models to predict PROMs after PELD.

Although a growing body of research has examined unfavorable outcomes following PELD, most studies focus on identifying risk factors associated with poor surgical results. In this context, to better support clinical decision-making and improve communication with patients, this study aims to develop a predictive model. By analyzing preoperative risk factors, the model seeks to estimate the probability of patients achieving MCID after undergoing PELD, thereby offering a novel instrument for personalized care planning.

Methods

Study design

The retrospective study was a cohort investigation conducted in a single medical institution. Clinical data were collected from consecutive patients diagnosed with LDH who had undergone PELD at the Department of Orthopedics in local hospital. The study protocol was approved by the hospital’s research ethics committee.

Inclusion and exclusion criteria

The inclusion criteria for study were as follows: patients diagnosed with LDH who underwent PELD, with a postoperative follow-up period of at least 2 years and complete preoperative as well as 2-year postoperative clinical data available.

The exclusion criteria included: severe cardiopulmonary dysfunction or other systemic diseases that could potentially impact postoperative recovery; severe neurological dysfunction (such as cauda equina syndrome or foot drop) present prior to surgery; and severe osteoporosis, spinal tumors, or spinal infections.

Data collection

The collected data encompassed baseline characteristics, clinical manifestations, physical examinations, and radiological features. Baseline characteristics included gender, age, smoking history, comorbidities and alcohol consumption history. Radiological assessments involved calcification of the intervertebral disc, the affected segments, Modic changes, the fat infiltration rate (FIR) of multifidus muscle, and Pfirrmann classification.

Modic changes were assessed based on the classification proposed by Modic et al., which categorizes endplate degeneration into three types [16]. FIR of paravertebral muscles was graded on a scale of 1–4 [17]. Based on previous studies [18–21], we specifically collected data on multifidus muscle fat infiltration in the L4/5 segment. The Pfirrmann classification, introduced by Pfirrmann et al., evaluates the severity of intervertebral disc degeneration on a scale from mild to severe [22].

Missing data were addressed by multiple imputation using SPSS (IBM SPSS 23.0, SPSS Inc.).

Evaluation of surgical outcomes

Patients were categorized into two groups based on whether they achieved the MCID postoperatively [12]. Pain severity was assessed using the VAS, and the change in VAS scores from preoperative to postoperative follow-up was calculated. Although a definitive MCID for PELD has not yet been established, prior studies of minimally invasive lumbar surgery have typically reported VAS-based MCIDs of approximately 2–3 points. Consistent with prior endoscopic lumbar discectomy studies, we defined a ≥ 3-point decrease in the VAS score as the MCID threshold [23–25].

Grouping and rationale

Because 70/30–80/20 splits are commonly used for development and validation in previous studies [13, 26, 27], we randomly divided the cohort 70/30 into a development cohort and a validation cohort using SPSS with stratified random sampling by outcome (MCID achieved vs. not achieved), thereby preserving the event/non-event ratio in both cohorts. The development cohort was used to build the model, and the validation cohort to assess its performance.

Identification of independent risk factors

Data from the development cohort were analyzed using SPSS. For analysis purposes, two variables were transformed into binary categories: the affected segment (classified as involvement or non-involvement of the L4–L5 segment) and the fat infiltration rate (FIR) of the multifidus muscle (categorized as mild [< 3] or severe [≥ 3]).

Potential risk factors were determined through univariate logistic regression analysis. For continuous variables with normal distribution, a t-test was applied, while non-normally distributed continuous variables were analyzed using nonparametric tests. Categorical variables were examined using Chi-square tests. Variables with p-values below 0.1 in the univariate analysis were considered as potential risk factors and included in the multivariate logistic regression. In the multivariate analysis, variables with p < 0.05 were classified as independent risk factors.

Development and validation of the predictive model

The predictive model was developed using R (version 4.4.2, https://www.r-project.org). Model performance was evaluated through several methods: discrimination was assessed by plotting the receiver operating characteristic (ROC) curve and calculating the area under the curve (AUC), while calibration was evaluated using calibration curves and the Hosmer–Lemeshow test. Finally, decision curve analysis (DCA) was conducted to assess clinical utility and the benefits of decision-making. These evaluations were conducted for both the development and validation cohorts to ensure comprehensive development and internal validation of the model.

Results

Patient selection process

Data from 456 consecutive patients admitted to the hospital between June 2014 and December 2021 at the Department of Orthopedics were collected. Following the selection process, 14 patients were excluded, resulting in the final study population of 442 patients. Among these, 312 patients were assigned to the development cohort, while 130 patients were designated for the validation cohort. The selection process is shown in Fig. 1. Of all patients, 23 (5.2%) did not achieve the MCID. In the development cohort, 17 patients (5.4%) failed to achieve the MCID, while in the validation cohort, this figure was 6 patients (4.6%). Detailed demographic and clinical characteristics of the patients are presented in Table 1.

Fig. 1.

Patient selection flowchart

Table 1.

Baseline characteristics of the included patients

Variables	Development cohort n = 312		Validation group n = 130
	MCID n = 295 (94.6%)	Non-MCID n = 17 (5.4%)	MCID n = 124 (95.4%)	Non-MCID n = 6 (4.6%)
Age	45 ± 14.21	53 ± 14.41	43.96 ± 14.85	50.5 ± 10.6
Gender
Female	126	8	48	2
Male	169	9	76	4
Smokers
No	227	12	91	5
Yes	68	5	33	1
Alcohol consumption
No	222	13	96	4
Yes	73	4	28	2
Diabetes
No	279	14	115	6
Yes	16	3	9	0
Symptom
LBP or LN	86	5	38	2
LBP and LN	209	12	86	4
Paresthesia
Yes	153	10	67	2
No	142	7	57	4
Lasegue sign
Positive	208	5	90	1
Negative	87	12	34	5
Fiber annular calcification
Yes	92	7	48	2
No	203	10	76	4
FIR of multifidus muscle
≤ Grade 2	232	7	99	3
> Grade 3	63	10	25	3
Modic change
Normal	193	9	84	5
Type 1	13	1	2	0
Type 2	80	6	31	0
Type 3	9	1	7	1
Baseline VAS score	8 (7.0, 9.0)	7 (6.0, 8.0)	8 (7.0, 9.0)	6.5 (5.5, 8.3)
Pfirrmann grade
Type 1	0	0	0	0
Type 2	18	1	5	0
Type 3	118	4	59	3
Type 4	148	10	53	3
Type 5	11	2	7	0
Dysreflexia
No	205	14	81	6
Yes	90	3	43	0
Affected segment
Involvement of L4–L5	180	14	77	2
None involvement of L4–L5	115	3	47	4

MCID clinically important difference, VAS visual analogue scale, FIR fat infiltration rate, LBP low back pain, LN leg numbness

Univariate logistic analysis and multivariable logistic regression

Univariate logistic regression was applied for identifying potential risk factors (p < 0.1). The significant factors identified included age (p = 0.048), the presence of a positive Lasegue sign (p < 0.001), affected segment (p = 0.078), baseline VAS scores (p < 0.001), and the FIR of multifidus muscles (p = 0.001).

After performing multivariable logistic regression analysis, 3 independent risk factors associated with the failure to achieve the MCID were identified: baseline VAS scores (p < 0.001), a positive Lasegue sign (p = 0.028), and the FIR of the multifidus muscle (p = 0.038). Detailed information is presented in Table 2 below.

Table 2.

Results of the statistical analysis

Variables	Univariate logistic analysis		Multivariable logistic regression
	P	OR (95%CI)	P	OR(95%CI)
Age	0.048	7.63 (0.65, 14.61)	0.750	1.00 (0.96, 1.05)
Gender	0.725	1.19 (0.45, 3.18)
Smokers	0.758	0.72 (0.25, 2.11)
Alcohol consumption	0.910	1.07 (0.34, 3.38)
Diabetes	0.127	0.27 (0.07, 1.03)
Symptom	0.982	1.10 (0.35, 2.96)
Paresthesia	0.576	0.75 (0.28, 2.54)
Lasegue sign	< 0.001	5.74 (1.96, 16.78)	0.028	3.77 (1.15, 12.31)
Intervertebral disc calcification	0.389	0.65 (0.24, 1.75)
FIR of multifidus muscle	0.001	0.19 (0.07, 1.52)	0.038	0.28 (0.08, 0.93)
Modic change	0.739
Normal		–
Type 1		0.95 (0.92, 0.98)
Type 2		0.69 (0.24, 1.97)
Type 3		0.47 (0.05, 4.05)
Baseline VAS score	< 0.001	0.45 (0.30, 0.68)	< 0.001	0.42 (0.27, 0.67)
Pfirrmann grade	0.279	1.80 (0.83, 3.90)
Dysreflexia	0.260	2.05 (0.58, 7.31)
Affected segment	0.078	0.25 (0.07, 0.89)	0.160	0.35 (0.08, 1.51)

FIR fat infiltration rate, VAS Visual Analog Scale

Development of the nomogram

Based on the identified independent risk factors, a nomogram was developed. Each independent risk factor was assigned a specific score that reflects its contribution to the model. The total score, calculated by summing the individual scores for each patient, was used to predict the probability of achieving the MCID (Fig. 2).

Fig. 2.

Nomogram for predicting the achievement of MCID Following PELD. Each independent risk factor is assigned a specific score (determined by projecting its value onto the "Points" axis; e.g., patients with a positive Lasegue's sign receive a score of approximately 27.5 points). The total score for an individual patient is calculated by summing the scores of all independent risk factors. This total score is then projected onto the "Probability of Achieving MCID" axis to determine the probability of that patient achieving the Minimal Clinically Important Difference postoperatively

Evaluation of the nomogram

The discriminative ability of the model was assessed by ROC curve for both the development and validation cohorts. The model demonstrated good discrimination, with an AUC of 0.84 (95% CI 0.73–0.95) for development and 0.88 (95% CI 0.69–0.98) for internal validation (Fig. 3).

Fig. 3.

ROC of the nomogram for development and internal validation. a The ROC curve in development reports the AUC 0.84 (95% CI 0.73–0.95); b in internal validation, the AUC of ROC curves is 0.88 (95% CI 0.69–0.98), which suggests the good discrimination of the model

The calibration of the model was evaluated by Hosmer–Lemeshow test and calibration curves. In development, the mean absolute error (MAE) of the calibration curve was 0.018, and the p-value from the Hosmer–Lemeshow test was 0.536 (p > 0.05). In the internal validation, the MAE was 0.026, and the p-value from the Hosmer–Lemeshow test was 0.369 (p > 0.05). These results indicated that the model has good calibration (Fig. 4).

Fig. 4.

Calibration curve of the nomogram for development and internal validation. In development, both the result of the Hosmer–Lemeshow test and calibration curve (with MAE = 0.018) indicate the good calibration of the curve; b similarly, in internal validation, both the Hosmer–Lemeshow test results and the calibration curve indicate that the model has good calibration

DCA curve was employed to evaluate the clinical utility of the nomogram. In the development cohort, decision curve analysis suggested positive net benefit across threshold probabilities of ~ 52–98% (development) and > 68% (internal validation) (Fig. 5).

Fig. 5.

Decision analysis of the nomogram for development and internal validation. In development, the DCA suggested it was beneficial to adopt this prediction model to predict the MCID after PELD when the probability threshold is ranged from 52 to 98%; b in internal validation, a net benefit was noted for threshold probabilities exceeding 68%

Discussion

This study developed and validated a nomogram intended to predict the likelihood of patients achieving the MCID following PELD. The nomogram incorporates three key factors: the Lasegue sign, FIR of multifidus spinae muscles, and baseline VAS score.

The model demonstrated good discrimination and acceptable calibration in internal validation. However, in internal validation, the model showed a small mean absolute calibration error (MAE = 0.026). Nevertheless, because only six patients in the validation cohort failed to achieve the MCID, calibration was suboptimal in the lower predicted-probability range.

Although significant research has been conducted on surgical outcomes following PELD, studies specifically focusing on the MCID after PELD remain relatively scarce. Several trials observational have investigated risk factors associated with residual symptoms in patients with LDH following PELD. For instance, another retrospective study conducted by Jitpakdee et al. involves 194 patients treated by PELD [28]. After a minimum follow-up period of 1 year, 32 patients reported experiencing “incomplete clinical improvement”. The study identified overweight status, significant preoperative disability or motor weakness (paresis), and a history of previous surgery as independent risk factors for incomplete clinical improvement. In contrast, our research involved a larger sample size. Additionally, the criteria used to define 'incomplete clinical improvement' in their study were relatively broad, which may have led to a higher proportion of cases being classified as having “unfavorable outcomes” [28]. Wang et al. found that patients exhibiting higher preoperative VAS scores or positive Lasegue signs were less likely to experience postoperative residual symptoms, such as low back pain or leg numbness [29]. Furthermore, Hu et al. identified severe baseline VAS scores and significant fatty infiltration of the paraspinal muscles as independent risk factors for residual low back pain following PELD [30]. These findings closely align with our research results.

This result is consistent with previous research indicating that patients experiencing greater preoperative pain are more likely to achieve significant pain relief following surgical intervention [31, 32]. Patients with higher preoperative VAS showed greater odds of achieving MCID; this may reflect a greater burden from more severe compressive pathology. Furthermore, patients with mild preoperative pain may have heightened expectations for surgical improvement, which could also influence their subjective assessment of postoperative outcomes.

Additionally, we found that a positive straight-leg raise was associated with higher MCID attainment, which may indicate more targetable nerve root compression; this interpretation is hypothesis-generating. One of the primary objectives of surgical treatment for LDH is to alleviate nerve root compression by excising the herniated tissue. Consequently, patients exhibiting a positive Lasegue sign may experience more substantial symptom improvement postoperatively, thereby facilitating their attainment of MCID.

Numerous studies have demonstrated a correlation between paravertebral muscles and the prognosis of LDH [33]. Specifically, atrophy of the paravertebral muscles is closely associated with postoperative residual LBP [34]. Additionally, the FIR of paravertebral muscles significantly influences the exacerbation of postoperative LBP. Research indicates that patients with paravertebral muscle FIR exceeding 40% face a heightened risk of worsening LBP after surgery [33].

Several previous studies have suggested that fat infiltration of the multifidus muscle may be associated with surgical outcomes. Sun et al. found that severe multifidus fat infiltration increases the incidence of recurrence in patients with LDH after PELD [35]. Similarly, the study by Tang et al. identified multifidus fat infiltration as an independent risk factor for recurrent LDH [36]. Zhu et al. reported that multifidus fatty atrophy leads to unfavorable surgical outcomes [37]. Although these studies employed different outcome measures than our own, they all reached a similar conclusion: the degree of multifidus fat infiltration is negatively correlated with surgical effectiveness.

Both paravertebral muscle atrophy and fat infiltration contribute to a decline in muscle strength, which reduces the muscles' capacity to effectively support the spine. This may increase the load on the intervertebral disc exacerbates intervertebral disc degeneration and postoperative pain [38].

This study has certain limitations. As a single-center study, the model has not undergone external validation and may therefore be subject to bias; moreover, the study lacked a prospective design. Finally, because non-achievement of the MCID was infrequent, calibration was unstable in the low-probability region, reflecting class-imbalance-related sparsity. Future work should employ larger, multi-center cohorts and incorporate prospective methodologies to more rigorously establish the reliability and generalizability of these predictors.

Conclusion

We developed a preoperative model that estimates the probability of not achieving the minimal clinically important difference (MCID). In our cohorts, lower baseline VAS was associated with higher predicted risk, suggesting that conservative management may be reasonable for selected patients with low baseline pain. Paraspinal muscle quality also showed an association with postoperative outcomes, consistent with the growing interest in sarcopenia. These are associations rather than causal effects and warrant prospective, multi-center validation and evaluation of clinical utility.

Abbreviations

MCID

Minimal clinically important difference

LDH

Lumbar disc herniation

PELD

Percutaneous endoscopic lumbar discectomy

ROC

Receiver operating characteristic curve

DCA

Decision curve analysis

VAS

Visual Analog Scale

FIR

Fat infiltration rate

LBP

Low back pain

Author contributions

This study was designed by Yijie Kong, Weimin Huang. The data involved were collected by Lei Wang, Wenqiang Xing, and Xiaoduo Xu. The data were analyzed by Yijie Kong. and the results were critically examined by all authors. Yijie Kong had a primary role in preparing the manuscript, which was edited by Jingming Wang and Weimin Huang All authors have approved the final version of the manuscript and agree to be accountable for all aspects of the work.

Funding

The research was not funded.

Data availability

The datasets analysed during the current study are not publicly available due to their containing information that could compromise the privacy of research participants. but are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

Ethical approval and informed consent were obtained. And our study was conducted in accordance with the "Declaration of Helsinki".

Consent for publication

No patient data was published and consent for publication does not apply to this article.

Competing interests

The authors declare no competing interests.