The current hypothesis-generating study was designed to develop prediction models for identifying a reduction in the KAM impulse during gait with standalone or supported-LWI in individuals with medial TFOA. Using clinically-accessible and laboratory-derived measures of predictor variables in multivariable logistic regression, we were able to predict the biomechanical responder status to WEDG and WEDG + V-ARCH at a 2% response threshold with acceptable AUCs of the ROC curve (c ≥ 0.794). Faster gait speed emerged most frequently as a significant predictor variable, with additional significant predictor variables including female sex, and lower FFI pain. While not statistically significant at p < 0.05, other variables such as lower KL grade, higher NRS pain, and less varus alignment emerged as variables of interest at the 0.05 < p < 0.10 level of significance. By providing a comprehensive report on response likelihood across different KAM reduction thresholds, different insole conditions, and different methods of acquiring predictor data, we feel that information from this current work may be able to guide researchers and clinicians in their consideration for LWI prescription for medial TFOA management, and to inform future research regarding variables that may be worth investigation in refinements of models to predict LWI response biomechanically.
Individual predictor variables demonstrated varying abilities to distinguish between biomechanical responders and non-responders to LWIs. We found self-selected gait speed, regardless of its clinically-accessible or laboratory-derived origin, was the most influential in predicting a KAM impulse reduction, with significantly greater odds of experiencing a reduction in the KAM impulse with LWIs for every unit increase (1 dm/s) in gait speed across 9/12 AIC-selected models. Faster gait speeds are known to be associated with a larger magnitude of KAM peak (26, 27). Since gait speed was constrained between insole conditions, higher baseline magnitudes of KAM – from a faster self-selected gait speed – may potentially provide more opportunity for an LWI to impart a KAM-reducing effect. Our modelling suggests that the odds of reducing the KAM impulse with LWIs is greater in individuals who are female, older, and have less foot-related pain, less varus knee alignment, and a lower KL grade. While we did not investigate the mechanisms for KAM reduction with LWIs related to each of these predictor variables, this list may highlight variables that should be prioritized to obtain in clinical settings for predicting biomechanical response when resources are limited or that are worth future investigation in prospective biomechanical prediction models.
The predictive ability of clinically-accessible and laboratory-derived prediction models appeared to perform similarly, which suggests more resource intensive methods for obtaining relevant data may not always be necessary. Since values of the AUC of ROC curve were not statistically compared, we could not formally validate clinically-accessible prediction models against their laboratory-derived counterpart. However, using the AIC-selected model at the 2% response threshold for WEDG + V-ARCH as an example, the c [95%CI] values for the clinically-accessible (c = 0.829 [0.709, 0.949]) and laboratory-derived (c = 0.843 [0.729, 0.957]) appeared to be similar. Furthermore, the significant predictor variables selected by AIC for a given response threshold and insole condition were generally similar between clinically-accessible and laboratory-derived models; this provides confidence that the clinically-accessible predictors are appropriately representing the same constructs as their laboratory-derived counterparts. A similar trend was found for clinically-accessible and laboratory-derived models matched for the same combination of response threshold and LWI condition.
While speculative, our current findings may shed light on previous LWI clinical research. Our findings present a picture that those with knee OA who are more likely to respond biomechanically to LWIs are less impacted by the disease – those that are more functional (ie. walk faster), have less varus malalignment, and have lower radiographic severity. Given that the likely mechanical effect of a 5 degree LWI is relatively low, it stands to reason that those with more varus malalignment and structural degradation would require a larger mechanical intervention to produce the biomechanical realignment that would reduce the KAM. These findings may translate to improvements in symptoms as well. Indeed, when assessing pain improvement with the use of LWIs, Baker et al. showed in their sub-group analysis that individuals with KL < 4 improved their pain by 21 points (out of 500) on the Western Ontario and McMaster Univeisties Osteoarthritis Index, while those with KL = 4 improved by only 2 points (28) following six weeks of use. Clinical trials may be best operationalized by specifically recruiting those with earlier stage disease, though future research is needed to examine phenotypes of those who symptomatically respond to LWIs.
Predictor variables that were not selected into any AIC-selected models suggests that these metrics may not be useful for predicting biomechanical response to LWIs in the presence of all other chosen predictor variables. Static foot posture, represented by FPI, did not significantly contribute to AIC-selected models as a predictor. This finding aligns with a report that metrics of static foot structure, excluding FPI, cannot predict biomechanical responses to wedged footwear in individuals with medial TFOA (17). Ankle/subtalar eversion motion had previously been shown to predict biomechanical responders from non-responders to LWI treatment (13). However, the influence of frontal plane knee alignment on mediating the association between ankle/subtalar eversion and the KAM magnitude (29) may also be influencing the KAM response with LWIs observed in our study. Clinically-accessible and laboratory-derived frontal plane knee alignment was a significant predictor in two AIC-selected models, and was a predictor variable of interest in four others. Taken together, frontal plane knee alignment may prove to be more useful in predicting biomechanical response to LWI than ankle/subtalar eversion. Lastly, FPA was not a significant predictor in any model, which contrasts a previous report that larger reductions in the KAM with LWI occurred in healthy adults with a smaller natural FPA than those with a larger FPA (18). Our findings regarding FPA may differ from previous works because our data were sampled data from individuals with medial TFOA, and was also evaluated amongst the influence of a concert of predictor variables which may produce a different result than when FPA is studied in isolation.
The findings from this study should be interpreted with the following limitations in mind. Firstly, given the number of predictor variables that were explored for their predictive capabilities, a larger sample size may have been warranted to improve the confidence and generalizability of our prediction models. Particularly at higher response thresholds of KAM impulse reduction, the observed distribution of biomechanical responders to non-responders (e.g. 12 responders : 41 non-responders at 10% response threshold for WEDG + V-ARCH) was less proportional than the distribution at a lower threshold (e.g. 28 responders : 25 non-responders at 2% response threshold for WEDG + V-ARCH). Although our sample size (n = 53) was still larger than previous attempts at predicting biomechanical response with LWI, future studies would benefit from even larger sample sizes to obtain a greater spread of data to avoid model overfitting. Next, our selection of clinically-accessible measurements of gait, posture, and movement characteristics was driven by evidence from previous literature which suggested they could be relevant to KAM response with LWI. This is by no means an exhaustive list of possible biomechanical influences and other outcomes may emerge as relevant predictors in future research. However, since the clinically-accessible and laboratory-derived versions of predictor variables tended to be significant in the same iteration of response threshold and LWI condition, this gave us confidence that the clinically-accessible metrics adequately represented the same construct as their laboratory-derived counterparts. Future research to confirm the validity of clinically-accessible predictors as a surrogate for its corresponding laboratory-derived counterpart would strengthen the findings of the current study.