|
North American Congress on Biomechanics Canadian Society for Biomechanics - American Society of Biomechanics University of Waterloo Waterloo, Ontario, Canada August 14-18, 1998 |
|
Regression models were created to determine the three-dimensional loading on the lumbar spine during an awkward barrel lift. The equations were constructed from both a full set of predictor variables and a reduced set, both of which were collected during a laboratory investigation of the lift.
Many biomechanical models have been developed to predict the spinal loading during lifting tasks. These models are often highly complex, making it difficult to implement them in industrial settings without severely compromising the motion of the lifter. Reducing the complexity of the calculations required for these models is essential to be able to calculate similar spinal loadings in workers as they perform the lifting tasks on site. Regression models enable the prediction of a dependent variable, the spinal loading, from a set of independent or predictor variables. Several investigators have used regression models to predict the spinal loading but these models have either been severely reduced in complexity or contain predictor variables that are difficult to collect. Potvin et al. (1) used regression models to predict the peak L4-5 compressive force during dynamic, sagittal plane lifting. Although potentially useful to the practicing ergonomist, the restrictions to the sagittal plane reduces the model's applicability. McGill et al. (2) later developed a simple polynomial to predict L4-5 compression during a wide variety of tasks. The predictor variables for this regression model were the 3-D external moments acting about L4-5, which may be complicated to collect in some situations. Furthermore, both groups only investigated the compressive forces, ignoring the shear forces which may aid in risk prediction.
The purpose of this investigation was to create simple regression models to predict the three-dimensional lumbar spinal loading during a hazardous lifting task.
For this study, a hazardous lift was identified from an ergonomic evaluation of several industrial sites (N=8). This task involved lifting a large barrel filled with recycled paper (150 - 355N) and emptying it into a tall (~103 cm), non-standard cardboard box.
The lift was then simulated in a laboratory setting and a detailed biomechanical model, including a muscle modeling technique involving EMG activation and optimization, was developed to calculate the lateral shear, compressive and A/P shear (3-D) forces acting on L3-4 and L4-5 for four different periods during the lift. Linear regression models were then created from the 16 subject's data to predict the 3-D forces on L3-4 and L4-5 using a stepwise inclusion criteria in the software package SAS. The predictor variables were identified from the simulation of the lift. These variables included the subject's demographics: height, weight, trunk depth and width, the physical characteristics of the lift: barrel weight and lift height, the subject's motion during the lift: planar angles and angular velocities of the ankles, knees, and hips and 3-D angles and angular velocities of the spine between S1 and T6, and also the normalized EMG activity of the right and left erector spinae, latissimus dorsi, rectus abdominis, and the external oblique. Linear regression models were also created from a reduced set of predictor variables which excluded the EMG measures and the angles of the legs, leaving a set of feasibly collected variables.
The six worker's who perform this recycling lift were tested using the regression models created from the reduced set of variables. Each subject was provided an information summary concerning the experiment and signed a consent form. The B-Tracker(tm), a three axis goniometer, was fitted on each subject's back during the lift and was sampled at 10Hz. using a portable data collection device. The lift was also videotaped in order to identify the different phases of the lift such as the initial lifting and the emptying phases. The subject's demographics along with the physical characteristics of the lift were measured. The angular velocities were calculated from the B-Tracker angles. The 3-D forces on L3-4 and L4-5 during the different phases of the lift were then calculated from the regression equations.
The full regression models typically included the subject's demographics, the physical characteristics of the lift, the 3-D angles and angular velocities of the spine, and also the normalized EMG activity of the muscles. The angles and angular velocities of the ankles, knees, and hips did not pass the stepwise criteria for inclusion in the models. Therefore, the difference between the full and the reduced models was only the inclusion of EMG measures. The R 2 values for the lifting position and the emptying position for each motion segment level, for both the full and reduced models are presented in Table 1.
Although the amount of paper in each barrel was not controlled, two subjects lifted barrels of similar weights (155.6 N) to similar heights (103.9 cm). Both workers were 60 years old and weighed approximately 88 kg on the day of the test. The first worker was 180 cm tall and his trunk depth and width were 26 cm and 36 cm, respectively. The second worker was 173 cm tall, with a trunk depth and width of 24 cm and 34 cm, respectively. The B-Tracker angles for the lifting and emptying position for the first worker were 13.0, -1.0 of flexion, 9.6, -1.4 of lateral bending, and -5.0, 1.9 of rotation, respectively. The B-Tracker angles for lifting and emptying positions for the second worker were 29.0, 13.1 of flexion, 11.3, 7.6 of lateral bending, and -10.2, -6.4 of rotation, respectively. A plot of the 3-D forces on L3-4 for both workers is shown in Figure 1.
| Lifting Position | Emptying Position | |||
|---|---|---|---|---|
| L3-4 | L4-5 | L3-4 | L4-5 | |
| Full Model | ||||
| Lat. Shear | 0.923 | 0.940 | 0.820 | 0.840 |
| Comp. | 0.925 | 0.921 | 0.908 | 0.909 |
| A/P Shear | 0.898 | 0.923 | 0.834 | 0.828 |
| Reduced Model | ||||
| Lat. Shear | 0.930 | 0.945 | 0.629 | 0.685 |
| Comp. | 0.758 | 0.759 | 0.746 | 0.745 |
| A/P Shear | 0.821 | 0.819 | 0.740 | 0.783 |
Table 1: The R 2 values for the full and reduced models.
Figure 1: A bar chart comparing the 3-D forces on L3-4 for both workers.
The regression models created from the full set of predictor variables accounted for a large percentage of the variability in spinal forces. The reduced set was less predictive due to the exclusion of the EMG measures. For both models, the predictability was higher for the lifting position than the emptying due to the contrast in posture between these two phases. The models were applied to a subject's data from the laboratory experiment to demonstrate the predictability of these regression models. The L3-4 lateral shear, compressive, and posterior shear forces calculated during the experiment were 334.9, 5546.7, and 1077.3N, respectively. The forces predicted from the full regression model were 398.1, 5518.5, and 1375.0N, respectively.
Although the two workers presented here were similar in physical characteristics, the first worker had much higher forces despite having a less flexed trunk. For the lifting position, the difference in spinal forces was mainly due to the differences in height between the two workers. If the second worker's height were 180 cm then the compressive force would have been 7340.1 N instead of 6866.2 N. For the emptying posture, having the back in extension, as the first worker did, increased the 3-D forces on the spine. Similarly, because the second worker was somewhat flexed and bent to the right side during the emptying, the compressive and lateral shear forces were not as great, whereas the posterior shear forces were increased with this posture.
Regression models are a unique method to reduce the complexity of a spinal loading analysis but must be based on a comprehensive model to ensure accurate predictions. These models have great practical utility and therefore, may be used to better understand why one person may become injured while another never is.
Future work will focus on expanding the regression models to apply to a wide range of lifting tasks.
1. Potvin JR, Norman RW, Eckenrath ME, McGill SM, Bennett GW. Regression models for the prediction of dynamic L4-5 compression forces during lifting. Ergonomics, 35: 187-201, 1992.
2. McGill SM, Norman RW, Cholewicki J. A simple polynomial that predicts low-back compression during complex 3-D tasks. Ergonomics, 39: 1107-1118, 1996.
