Chapter 2. RESEARCH OBJECTIVES
Rationale for the study
As discussed in the previous chapter, back pain is a prevalent disease in our society. Not only is its frequency considerable, but treatment is expensive as well. Despite decades of extensive scientific study of the spine system, occupationally induced back diseases are still frequent. It can be argued that the scientific community has made little progress in gaining knowledge about the spine system with the necessary degree of applicability (Leamon, 1994). The number of muscles included in models has increased, the detail of the anatomy used has increased, and the number of parameters (e.g. vertebrae stiffness, ligament passive forces, vertebrae displacements) included has increased. Have those advances brought us any closer to understanding the mechanism of injury in that body region? Although some theories regarding injury mechanisms on the lumbar spine are available, we are really not much closer to confirming or disproving them than we were ten years ago. The question is, then, what needs to be done?
Even though they don’t seem to have been completely successful so far, the predictive models described before may provide one piece of the puzzle. These models are needed due to the unavailability of practical (i.e. low cost, high ruggedness, and low intrusiveness) instruments to measure torso muscle recruitment patterns on the field. If we had this ability, and assuming spine compression (as calculated from the muscle recruitment patterns measured) is related to LBP, we could directly redesign the field tasks to reduce the risk of LBP. Since we don’t have this ability, and it is unlikely we will have it in the near future, we need predictive models to produce realistic sets of muscle recruitment patterns that might occur in a given task. If we assume the models are a fair representation of these patterns, then the spine compression can be calculated, and the redesign process completed. The main limit on this approach is that some predictive models, it has been argued (Nussbaum, et al., 1997), may have reached a practical predictive limit with the use of current technology under static postures and loads. However, for this to be truly the case, predictive models have to be compared against EMG data obtained over a bigger range of loading magnitudes, loading directions (including axial twisting), and a larger number of subjects than the data currently available.
Even more importantly, intra- and inter- subject variability has to be measured and published. As difficult as the analysis of such variability may be, it is very possible that therein lies the link between biomechanics and epidemiology. If we are able to link particular groups of individuals that are different from the norm to low back disease, then we might be able to locate what aspects of or changes in behavior provoke injury. If an individual significantly changes their recruitment patterns from one trial to another, is it representative of random variation? Could that variation be caused by another mechanism? Can we accurately model these variations? Data is needed in order to answer these questions.
The weakest point of any of the models described before, and most biomechanical models for that matter, is the issue of model validation. Seldom can we take direct measurements of physiological processes in the human body. If this is the case, how can lumbar muscle recruitment models be validated? Two well thought out answers seem to be provided in Cholewicki and McGill (1996) and Hughes (1991). Cholewicki and McGill (1996) propose a validation approach that consists of component validation, internal validity checks, sensitivity analysis, and judgmental evaluation. Hughes (1991) proposes validation by subjecting a model to the greatest amount of testing possible. As the model continues to pass tests that it has a chance of failing, confidence in the model increases. If the model fails a test, reasons for this failure can be discovered and adjustments can be made.
ANN’s, DMH, and SCI represent three recent and very different attempts to model human behavior. Based on the published data for each, they also appear to be well correlated with experimental EMG data. Nevertheless, a direct comparison between these models based on the published data may confound the results with other artifacts, such as the anatomy assumed. Therefore, the question whether one of these models is better remains to be answered. The importance of this comparison, especially when EMG data resulting from significantly different loading conditions is used, is that it provides for the indirect validation of the models. If the models pass tests with “flying colors”, then confidence in the models increases. Models that fail tests can be studied for the cause of failure and improvements may result. In the long run (i.e. over several experiments), and as models evolve, we might come closer to finding a method that accurately models the actual body mechanisms responsible for lumbar muscle activation patterns.
In summary, this experiment attempts to increase the scientific body of knowledge on lumbar mechanics with three main contributions. First, EMG data was collected from several lumbar muscles using larger loading magnitudes than those generally reported in the literature. Second, EMG data was collected for the same set of lumbar muscles over biplanar loading conditions that include torsional moments, instead of just the usual sagittal and frontal moments. The data collected is used in the assessment of the relative performance (third contribution) of ANN, DMH and SCI models under different sets of loading conditions
The present work statistically tests a set of EMG data collected experimentally against the muscle activation patterns predicted by the models under various experimental loading conditions. The experimental question tested is:
Across all the loading conditions tested, across all participants, and across experimental replications, a single model, either ANN, DMH, or SCI, will emerge as the best predictor of lumbar muscular recruitment patterns as quantified using correspondence with the collected EMG data
The research presented here consists of two different phases. The first phase involved human participants that statically resisted several loads applied to their upper torso. EMG data measured from several lumbar and abdominal muscle groups were collected. This data was used as inputs in the second phase of the research. This second stage involves the statistical comparison of model predictions against the actual EMG values, as well as comparisons between models in terms of their predictive ability.
Phase I: EMG Data Collection.
The first phase obtained EMG data from anterior and posterior torso muscles in human participants. The goal of this phase was to obtain muscle recruitment data to be used in the evaluation of the predictive models.
A total of 8 participants were selected from a University student population. None of the participants had a previous history of back pain, back injury, or other disorders that would prevent them from resisting torques applied to the torso. These conditions were self-reported in a pre-experimental health questionnaire, shown in Appendix B, which also recorded age and gender. Weight and height were recorded by the experimenter before the experimental protocol is started. To avoid the process of scaling the models to account for subject variation, only participants falling within a ±10% range of the 50th male height and weight percentiles were selected. Although accurate scaling models do exist (e.g. Nussbaum and Chaffin, 1996a), their use would introduce an additional and undesired confounding factor in the experimental process. The male height and weight are considered because the anatomical studies used in the models are based on the male anatomy. Therefore, all measures reported in those studies should approximate the 50th percentile (average) male physique. It is assumed here that women with the same height and weight measures will also be described by the anatomical measures used if they fall in the same height and weight range. The height and weight values required range from 173.9-177.3 cm and 75.3-81.7 kg, respectively (calculated from data reported by Gordon, Churchill, Clauser, Bradtmiller, McConville, Tebbets, and Walker, 1989
Chapter 1. INTRODUCTION
1.1 Epidemiological Motivation
1.3 Lumbar Muscle Recruitment
1.4 Low Back Muscle Activation Models
Chapter 2. RESEARCH OBJECTIVES
2.1 Rationale for the Study
2.2 Research Question
Chapter 3. EXPERIMENTAL METHODS
3.1 Experimental Design
Chapter 4. RESULTS
4.1 Phase I: EMG Data Collection
4.2 Phase II: Model Data Collection
Chapter 5. DISCUSSION
5.1 EMG Data Collection
5.2 Model Data Collection
Chapter 6. CONCLUSIONS
GET THE COMPLETE PROJECT