A valid model for predicting responsible nerve roots in lumbar degenerative disease with diagnostic doubt

Lumbar degenerative disease (LDD) often displayed as multilevel degeneration and stenosis occurs due to compression or ischemia, or both, of the lumbosacral nerve roots as a consequence of osteoarthritic thickening of the articulating facet joints, infolding of the ligamentum flava, and degenerative bulging of the intervertebral discs [1‐3]. It is the main cause of chronic low back pain in old people, leading to spine surgery among individuals older than 65 years [4, 5]. With an increase in aging population, the number of people who suffer from this condition is expected to grow exponentially and this will have a significant effect on healthcare resources in the near future. Surgical decompression of the responsible compression sites remains as a widely accepted therapy of LDD currently [6‐10].

Interestingly, although most patients with LDD exhibit a typical painful experience or present obvious degenerative changes on computed tomography or magnetic resonance imaging (MRI) scans, there still exists a group of patients with LDD whose diagnosis are uncertain or who have an ambiguous compressive region. In other words, when the responsible nerve roots are vague, or the pain source does not correspond to typical classical dermatomal patterns [11, 12], it is very difficult to select the decompression site and make reasonable surgical plans for such patients with diagnosis doubt (DD). Moreover, to date, there are no studies proposing a predictive method to determine responsible nerve roots in patients with DD. Nonetheless, studies on this topic are still in progress. Recently, an increasing amount of evidence has demonstrated selective nerve root block (SNRB) may play a role in predicting the responsible compression nerve roots [13‐18]. However, LDD usually reported a complicated progress, involving multiple factors, such as stenosis levels [19], stenosis locations [20], neurological deficit [21], and preoperative Oswestry disability index (ODI) score [22], that makes it quite complex to distinguish the responsible nerve roots.

Therefore, in this case, we planned to use relevant parameter of SNRB combined with some risk factors screened out from the baseline patient-related factors and clinical characteristics to establish a scoring system through multivariable logistic model. After that, the utility of this new predictive model was examined in an external subpopulation of a validation sample. Ultimately, we hope this new predictive model will play a role in decision making of which segments should to be decompressed and how many decompression segments should be conducted in such patients with DD.

To evaluate the responses of last JOA score after a minimum of 2 years follow-up, questionnaires were prepared to determine the percentage of patients with ECO or non-effective clinical outcome (NECO) relative to their initial questionnaire values. The clinical outcomes were divided into the following four types according to different JOA recovery rate which was calculated by the Hirabayashi method [25]: (postoperative score − preoperative score)/(29 − preoperative score) × 100 %. The four types of recovery rates were graded as follows: >75 %, excellent; 50–74 %, good; 25–49 %, fair; and <25 %, poor. The first two results were considered as ECOs.

Totally, of the 191 patients included in the study, 163 cases finally fulfilled the inclusion criteria, representing a loss of 14.7 % (n = 28), of whom, had at least one of the exclusion criteria. The 163 patients with DD were assigned to the construction sample (n = 106) or validation sample (n = 57) according to different admission times to hospital.

The baseline patient-related factors between the ECOs (excellent or good) and NECOs (fail or poor) on follow-up for no less than 2 years are compared in Table 1. The results showed no significant difference between the two groups (P = 0.08–0.87) (Table 1).

Additionally, a comparison of clinical characteristics between the ECOs and NECOs is presented in Table 2. The significant risk factors were stenosis level (L4/5, P = 0.02; L5/S1, P = 0.03), stenosis locations (neuroforamen, P = 0.03), neurological deficit (Reflexes, P = 0.03; Sensory, P = 0.02; Motor, P = 0.02), higher VAS score before operation (P = 0.01), and more VAS improvement rate after SNRB (P = 0.01). Our results of risk factors were mainly tallying with the previous report except for the higher VAS score before operation. This is a newly found risk factor in LDD patients with DD that may reflex the compression nerve roots to a certain extent.

After all risk factors were screened out, a multivariate logistic regression model was used with clinical outcome as the dependent variable, and the 5 risk factors as explanatory variables in the construction sample (Table 3). Thus, the respective odds ratio (OR) of risk factors were higher VAS score before operation (OR = 1.56, 95 % CI: 1.08–2.65, P < 0.01), stenosis levels of L4/5 or L5/S1 (OR = 1.44, 95 % CI: 1.10–1.89, P = 0.04), stenosis locations of neuroforamen (OR = 1.95, 95 % CI: 1.32–3.51, P = 0.01), neurological deficit (OR = 1.62, 95 % CI: 1.02–2.79, P = 0.01), and VAS improvement after SNRB (OR = 3.42, 95 % CI: 1.27–7.64, P = 0.02).

Once the logistic regression model was constructed, this was transformed into a scoring system according to the OR of each risk factor (Table 4). The key points that defined the risk groups were as follows. 1) Discard: value, 5; sensitivity, 0.98 (95 % CI: 0.95–1.00); specificity, 0.09 (95 % CI: 0.05–0.14); PLR, 2.03 (95 % CI: 1.93–2.77); NLR, 0.43 (95 % CI: 0.24–0.65). 2) Optimum: value, 11; sensitivity, 0.86 (95 % CI: 0.82–0.93); specificity, 0.67 (95 % CI: 0.56–0.75); PLR, 2.46 (95 % CI: 2.13–2.85); NLR, 0.46 (95 % CI: 0.22–0.57). 3) Confirmation: value, 16; sensitivity, 0.13 (95 % CI: 0.06–0.21); specificity, 0.99 (95 % CI: 0.97–1.00); PLR, 3.39 (95 % CI: 2.64–3.93); NLR, 0.61 (95 % CI: 0.42–0.79).

After the system score was established, we firstly tested it in the construction sample as an internal validation and the result of the area under curve (AUC) was 0.85, which demonstrated it to be a good model (Fig. 1). Moreover, the risk factors in the construction and validation samples were analyzed and the results were similar in both samples (P = 0.085–0.889) (Table 3). The ECOs of DD patients were around 76 % in the two groups (76.4 and 75.5 % respectively). On this basis, the ROC curve for our scoring system in the validation sample are reasonable and the AUC was 0.72 (Fig. 2). Additionally, to evaluate the calibration plot of this model, the data were also tested in the validation sample and the predicted probability showed good linear relationship with the actual probability, which exhibited as an appropriate calibration plot (Fig. 3). Finally, we once again compared the predicted and actually observed outcomes of this scoring system, and the analyzed result show no significant difference within the two samples (P = 0.532, Fig. 4).

This study constructed and validated a predictive model to determine the responsible nerve roots in DD patients. This model was constructed by transforming complex factors into a simple scoring system to enable a rapid calculation. To the best of our knowledge, this is the first predictive model in clinical application. In this model, both the internal and external AUC were > 0.7 and the calibration plot of prediction accuracy were tested as a good linear relationship. In addition, the predictive and actual outcomes showed no significant difference. Hence, this model was applicable and valid. As is known to all, LDD often displays several segmental pathological changes without exact localizing signs on physical examination, because of its elusive symptoms and missing standards on imaging analysis [26]. When counseling a patient with DD on which segments to be decompressed, or how many decompression segments to be conducted, a predictive model such as this, is of paramount importance.

In addition, a predictive model like this is also beneficial when considering risk factors. In our study, the univariate logistic regression model suggested 5 risk factors including higher VAS score before operation, stenosis levels of L4/5 or L5/S1, stenosis locations of neuroforamen, neurological deficit, and VAS improvement after SNRB. This will also play some role in some other kinds of lumbar spine diseases like failed back syndrome. Moreover, this model is a useful adjunct in predicting the clinical outcome after decompression surgery [13, 27, 28]. In our analysis, the OR of VAS improvement after SNRB was 3.42, which played a major role in the model. At this point, whether the pain relief after surgical decompression is good could be forecasted by this test to a certain extent because of the evidence that pain originating from nerve root compression can be effectively treated by surgical decompression [29‐31]. Nevertheless, our model also combines together several other related risk factors in order to improve the predictive accuracy, because only SNRB is not a cost-effective method for identifying the symptomatic nerve roots [27].

Our model was built upon JOA recovery rate with a minimum following-up of 24 months. Although this cannot replace long-term follow-up results and ultimate outcomes, our conclusions are based on the curative effect, and this model is supported by comprehensive evidence of credible outcomes in clinical trials. Meanwhile, this model also cannot draw any definite conclusion. At least, when the score in our model is >16 points or <5 points, we may get a rational and objective reason about whether it should be considered as a responsible segment or not. Additionally, this model could be used as a reference index in patients with DD for arriving at a diagnosis and for treatment purposes.

Since this model was based on the SNRB test, we would like to recommend the following guidelines: 1). Surgeons should be familiar with the anatomy so that he or she could accurately determine the precise nerve root of the test; 2). It is still important to preliminarily identify the possible responsible segment by combination of detailed physical examination and radiological results before SNRB; 3) In a possible liability gap, the most likely responsible segment should be tested first rather than one by one. If symptoms were relieved by >50 %, it could be judged as the responsible gap, or else taking order from the lowest nerve roots, because the block of upper nerve root is prone to defuse to the lower one, and thus, interferes with the result. 4). Needle should be introduced gradually under fluoroscopic guidance to avoid unnecessary nerve root injury and 5). The single dose should not be too much, generally 1 % lidocaine 0.5–1 ml, otherwise it will also cause other nerve roots.

As with any study, there are limitations to the present study. First, a great number of variations exist and we possibly did not identify all significant variables to predict the result. Future studies of this model may consider the effect of a more detailed database that contains more input variables (such as electromyography and the walking distance). Second, the number of patients was relatively small and this may have prevented significant correlation between the two groups. Finally, many subjective grading scores were not performed by the same surgeon on the same patient, and that may introduce some errors between the groups. However, we did attempt to minimize the weaknesses by using strict criteria for inclusion and exclusion. Although we were also very strict while performing the case inclusion criteria, these differences might be reduced but not abolished. Nevertheless, our model was validated, so that precise predictions are possible.

In summary, this study constructed and validated a predictive model that can be used to determine responsible segments or pain source of patients with DD. This tool is of substantial value in the preoperative counseling of patients, shared surgical decision making, and ultimately improving safety in spine surgery. Second, as we progress into an era of quality metrics and performance assessment, a tool like this can be beneficial in risk adjustment. Future predictive models are recommended for further risk stratification and modification.