Walking ability has been investigated as a predictor in LSS patients in a relatively small number of studies. Sigmundsson et al. reported that the self-estimated walking distance can be a predictor for satisfaction with operative outcomes [
3]. Furthermore, preoperative walking capacity was found to be a good predictor of satisfaction with postoperative walking capacity after surgical intervention [
7]. Conway et al. conducted SPWT and motorized treadmill test (MTT) [
14], asking LSS patients to walk on a level ground or treadmill at a self-paced speed until they voluntarily stopped due to worsening of LSS-related symptoms or until they reached the predefined maximum time duration of 30 min. They reported a significant correlation (
p<0.01) between the walking distance of the SPWT and the ODI score with a Pearson correlation coefficient of −0.60. Rainville et al. compared the changes in walking time and distance against the changes in ODI score [
15]. A significant correlation (
p<0.05) was reported between the walking time using the MTT and the ODI score with a Pearson correlations coefficient of 0.48. Similarly, Tomkins-Lane et al. reported a significant correlation (
p<0.01) between the changes in walking capacity from the SPWT and changes in the ODI score with a Spearman coefficient of −0.70 [
16]. However, the aforementioned works focus on measuring walking capacity, which require patients to walk up to 30 min for each test. This time requirement may limit the use of these tests in the clinical setting, notwithstanding the burden placed on patients to perform such extensive testing. Comparatively, the test proposed in this study requires a total of 40 m walking distance (10 m walking test repeated four times) and takes approximately 6 min to complete, which makes it more accessible in the clinical setting and likely more appealing to patients. Furthermore, the spatiotemporal gait characteristics provided by the smart-shoes, such as the ability to fine control the distribution of the body weight during walking, may provide more comprehensive interpretation of the LSS patient’s functional level following decompression surgery as compared to the relatively uni-dimensional evaluations of walking capacity.
Table
3 summarizes the results for predicting the postoperative ODI scores. It is noteworthy that the ODI scores collected preoperatively did not show statistical significance to the postoperative ODI scores (
p<0.21). This may be explained by operative subjects improving non-linearly compared to their preoperative status or may be because of the response shift of the ODI; ODI is known to suffer from the response shift after surgical intervention in lumbar spinal cord disorder patients [
26]. Thus, accurate prediction requires additional information, which motivated our study. None of the clinical variables investigated in this study showed significant correlation to the postoperative ODI score. The results in Table
3 reveal that the top five gait measurements with the largest (absolute) Spearman correlation coefficient contain gait information collected from P
2 (near the heel) and/or P
3 (mid-lateral plantar). For example, StdTime-P
2P
3-Max showed a positive correlation to the postoperative ODI. This implies that patients with superior functional condition (low ODI score) showed more consistent weight shifting from P
2 to P
3, i.e. more consistent walking pattern. Moreover, the length of time taken to shift their weight from P
2 to P
3 (MeanTime-P
2P
3-Mean) was shorter in functional patients, and thus the cumulative pressure applied to P
2 was smaller (SumMag-P
2-Min).This also implies that patients with higher ODI score (non-functional patients) took longer time to shift from P
2 to P
3, which can be explained by the sensory deficit and pain inherent to LSS resulting in a less fluid walking pattern. Patients with low ODI score also showed consistent pressure pattern at P
2 for both limbs (AutoCorr-P
2-Mean) and across the two limbs (CrossCorr-P
2) during walking. When multiple predictors were considered to predict the postoperative ODI, the StdTime-P
2P
3-Max and SymIndex yielded the strongest (and statistically significant) correlation (
p< 5.32 ×10
−4). It is noteworthy that SymIndex was not one of the measurements that produced the best correlation when considered for the univariate analysis. The most likely reason that StdTime-P
2P
3-Mean were combined with SymIndex rather than one of the top five measurements shown in Table
3 is that the top five measurements quantify similar gait characteristics (as most of them were derived from P
2 and/or P
3), and thus the maximum information gain was achieved when StdTime-P
2P
3-Mean was combined with SymIndex, which provides different gait characteristics [
27]. The predicted improvement computed by subtracting the predicted postoperative ODI score from the preoperative ODI score also showed a strong correlation compared to the actual improvement with RMSE of 0.13,
r=0.93, and
p<3.82×10
−7. Thus, the ability of the smart-shoes to reliably predict changes in ODI in response to surgery based on preoperative data makes this technology a valuable tool for identifying which patients will derive the greatest benefit from surgical intervention. This will help the decision making process for both clinicians and patients when considering surgical versus medical management of LSS, since it can provide a reasonable prediction for the symptomatic improvement that a surgical intervention can reliably provide. Table
4 summarizes the predictability of the gait measurements and clinical variables for the postoperative VAS scores. None of the clinical variables showed clinically significant correlation to the postoperative VAS. However, the smoking status and the number of previous spine surgeries showed near significant correlations (
p<0.089 for the smoking status and
p<0.076 for the number of surgeries). This agrees with previous findings that non-smoking status [
10,
11] and fewer previous surgeries [
13] have a predictive influence for improved functional level and leg pain relief after surgical intervention. The VAS score that was collected preoperatively did not show significance to the postoperative VAS, which supports that predicting the improvement in pain level after surgery requires additional clinical information (other than just preoperative VAS score). Table
4 shows that the top five gait measurements with the most significant correlations measure similar gait characteristics: consistency of weight distribution during gait. However, unlike the predictors of the ODI, the predictors of the VAS included gait parameters that were extracted from different landmarks of the plantar, e.g., P
1, P
2, and P
5. AutoCorr-P
2-Mean, CrossCorr-P
2, and AutoCorr-P
5-Min quantified the consistency of weight distribution at P
2 and P
5 (or walking pattern in general). Furthermore, patients with less postoperative pain had a shorter time to shift the body weight from P
1 and P
2 (MeanTime-P
1P
2-Max) and smaller SumMag-P
2-Min. We believe that these results resemble the results of the ODI analysis in that patients with less perceived pain were more apt to shift their weight rapidly and homogeneously to the whole plantar foot to stabilize themselves during walking. Whereas patients with higher perceived pain were more hesitant to shift their weight between limbs for fear of causing discomfort and/or instability. When more than one predictor was used to predict the postoperative VAS, CrossCorr-P
2 and AutoCorr-P
3-Min produced the highest correlation (
r=0.83,
p<1.28×10
−4, and RMSE of 0.19). Again, one of the top individual predictors (CrossCorr-P
2) was combined with AutoCorr-P
3-Min, which was not included in the top five individual predictors in Table
4, i.e. a predictor that provides different dimension of gait characteristics [
27]. The predicted improvement in VAS also showed a significant correlation compared to the actual improvement. Figure
6 illustrates the relationship that yielded the RMSE of 0.20,
r=0.82, and
p<2.58×10
−4. It is noteworthy that, although the prediction results of postoperative score and the level of improvement in VAS were statistically significant, the results of the ODI were more significant and accurate. This agrees with prior studies that have found the ODI to be a more sensitive prognosticator as compared to the VAS score regarding surgical outcome in LSS patients [
28].
The work introduced in this paper has some limitations. All patients who participated in this study were operated on by a single neurosurgeon (DCL). Thus, the predictors found in this work may vary from those found in patient populations of other surgeons. Additionally, the number of subject participants is relatively small. Thus, the identification of predictors in a large population needs to be verified, and the statistical results reported herein (e.g., the effect of the unbalanced gender distribution) may not be generalized. The research team is continuing to collect data from LSS patients and future studies will address these issues. Some factors that were previously found to have prognostic value, such as depression and psychiatric illness [
3,
6], were not included in this study and their utility as predictors of outcome cannot yet be compared to the smart-shoe data. Additionally, patients were reevaluated three months after their scheduled surgeries in order to obtain postoperative ODI and VAS results. Amundsen et al. reported that while most patients experienced relief of pain approximately three months after surgical intervention, pain levels in these patients would continue to decrease over years [
29]. Therefore, longer-term follow-up may be necessary to discover predictors of more permanent postoperative clinical outcomes. It is worth noting, however, that Atlas et al. found that a patient’s baseline postoperative functional level was reached by three months after the surgical intervention, i.e. patients function level did not improve much after three months postoperatively [
30]. This supports our belief that the reported predictors for our three-month follow-up study should provide insights into the long-term outcomes of our LSS patient population.