Presented at NACOB 98: North American Congress on Biomechanics Canadian Society for Biomechanics - American Society of Biomechanics University of Waterloo Waterloo, Ontario, Canada August 14-18, 1998

---

THE DEMANDS INDEX: A METHOD FOR WORK-RELATED INJURY RISK ASSESSMENT

D.M. Andrews ¹ , R.W. Norman ^2
1 Department of Kinesiology, McMaster University, Hamilton, Ontario, Canada, L8S 4K1
² Department of Kinesiology, University of Waterloo, Waterloo, Ontario, Canada, N2L 3G1

INTRODUCTION

A method was developed to combine the measured value of multiple physical exposure variables into a single number, a Demands Index (DI), which can be used to assess risk of injury. A single index simplifies the interpretation of overall physical stress on workers, making it easier for field practitioners to rate or compare different tasks on a common scale.

REVIEW AND THEORY

The risk of injury associated with work-related physical demands is multifactorial in etiology. Therefore, it seems that prioritization of tasks or jobs as 'safe' or 'hazardous' should be based on multifactorial approaches (Moore and Garg, 1995). A number of models based on multiple risk factors have been developed and used in the workplace to assess the risk associated with a given task or job (Moore and Garg, 1995; Kumar, 1994; NIOSH, 1981). These methods use an approach whereby the product of individual variable multipliers is compared to a threshold level of acceptable load in order to judge risk. Tissue level exposure variables, such as lumbar spine compression or shear forces, have not been used as multiplier terms in these models but have been shown to be indicators of risk of low back pain reporting (e.g. Norman et al., 1998; Kumar, 1990). Therefore, the purposes of this study were to: develop a method for combining multiple low back physical demands variables into a single value; illustrate the importance of justifying the number and type of input variables in terms of injury risk potential; and show that tissue level exposure variables are helpful in such models.

PROCEDURES

Subjects: Development of the Demands Index (DI) model was based on low back physical demands data of 233 autoworkers from a case/control epidemiological study of low back pain (LBP) reporting (Kerr et al., 1997; Norman et al., 1988). Cases (C; n=104) were workers who reported pain and Random controls (R; n=129) did not.

DI Model Components: The DI model components are outlined in Figure 1. Model input variables are processed in two main steps. Firstly, the value of each measured variable for a given subject is normalized with respect to an assigned 'tolerance maximum' from the literature, resulting in a number ranging from 0 to 1 for each variable. The normalized variables are then averaged, resulting in a DI, also a number ranging between 0 and 1 (min and max demand, respectively) which represents the physical demand of work on an individual. Based on the DI values for a group of subjects, a Risk Index (RI) is estimated at each decile of the DI range (0 to 1) using Equation [1], where the number of cases and random controls with DIs greater than the decile value (high demands), and less than the decile value (low demands) are counted. Equation [1] represents an odds ratio (OR) calculation for case/control studies.

Figure 1: DI model main components.

RI = (# Chigh x #Rlow) / (#Rhigh x #Clow) [1]

Input Variable Justification: Statistical analyses of 19 biomechanical variables were conducted to determine the best input variables for the DI model. This data driven (DD) approach, consisted of the following steps: univariable logistic regression, Pearson correlations (14 variables with significant odds ratios from the univariable regression), multiple logistic regression (backward elimination), final model testing ('best subsets', score statistic, forward regression). The initial set of 19 variables was established based on some from several key studies in the literature that have been found to be predictors of LBP (e.g. Marras et al., 1993; Punnett et al., 1991; Kumar, 1990).

Demands Index Analyses: The 19 variables were run through the DI model individually and in various combinations. The DIs and RIs associated with each analysis were compared. The final DD model of 3 input variables (see below) was also run through the DI model, the results of which were compared to those from 3 experimenter driven (ED) models also comprised of 3 variables from the initial subset of 19. The ED models were: ED1 (peak L4/L5 compression force (N), cumulative compression force (N·s/shift), peak trunk flexion angle (degrees)), ED2 (peak L4/L5 moment (N·m), peak hand load (kg), peak trunk flexion velocity (degrees/s)), and ED3 (%time severely flexed trunk, %time twisted trunk, %time laterally bent trunk). Variables were normalized in the DI model to an acceptable tolerance maximum. For example, a regression equation (Jäger et al., 1991) was used to estimate the maximum vertebral compressive strength for each individual, based on age, gender, endplate cross-sectional area, and level of the vertebral column. With this equation, a 25 year old male has an estimated maximum compressive strength at the L4/L5 disc of 9183 N. Several variables, such as %time severely flexed, were normalized to maximum values from this study sample (n=233).

RESULTS

Statistical Analysis: The statistical analysis resulted in a final data driven (DD) model of 3 statistically independent variables from the initial set of 19: average L4/L5 spine compression force (N) (OR 1.28; 95%CI 1.09-1.51), peak L4/L5 reaction shear force (N) (OR 1.30; 95%CI 1.07-1.61), and peak hand load (kg) (OR 1.12; 95%CI 1.02-1.25).

Multivariable Assessment: Risk (RI) associated with the 3 variable DD combination was at least 2 times greater than when the 3 variables were considered individually, despite having comparable demands (Table 1). This confirms that risk of LBP reporting is multifactorial, and that when variable contributions are summed, additional risk can be accounted for.

DD Variables	Mean DI (SD)	Mean RI (SD)
Individual -avg. compression -peak rxn. shear -peak hand load	0.20 (0.09) 0.59 (0.12) 0.22 (0.19)	3.04 (0.29) 2.86 (2.34) 3.24 (1.47)
Combined	0.33 (0.10)	7.18 (5.39)

Table 1: Multivariable assessment. Combining the 3 DD variables leads to an increased RI.

Variable Combinations Assessment: Estimates of risk resulting from the DD combination were much higher than any of the ED combinations (Table 2). This suggests that if no statistical considerations are made for the variables input into the DI model (and other models), risk which exists may not be accounted for.

Combinations	Mean DI (SD)	Mean RI (SD)
ED1	0.34 (0.12)	3.55 (1.34)
ED2	0.24 (0.12)	3.83 (1.67)
ED3	0.07 (0.08)	1.16 (0.22)
DD	0.33 (0.10)	7.18 (5.39)

Table 2: Variable combination assessment. The DD model resulted in a larger RI than the ED models.

DISCUSSION

The results indicate that tissue level exposure variables such as spine compression and shear forces, are important variables for assessing low back physical demands in terms of risk potential. This is highlighted by the fact that despite only having DIs of approximately one third of the maximum range for the DD model variables, the corresponding risk estimate was approximately 7. The DI model also confirmed that risk of LBP has a multifactorial etiology and that an additive model shows promise. However, DIs estimated by the model are sensitive to the tolerance maximums chosen for each variable. For example, the distribution of DI values for all workers will shift to the right (higher demands) with a decrease in magnitude of the tolerance maximum. The effect this has on risk, a relative measure, seems small in general (since a change of tolerance maximum is the same for both cases and controls in the sample), but needs to be quantified further. As a case in point, peak reaction shear force showed the highest DI value (0.59), but did not correspondingly produce the highest RI. The high DI may be a function of the limit set for it. Future work needs to be done to compare this additive approach with a multiplicative approach, to quantify the sensitivity of the method to different input variable weightings and tolerance limits, and to establish the method's generalizability by applying it to a variety of workplace populations and types of activities.

REFERENCES

Jäger, et al. Int.J.Ind.Erg., 8:261-277, 1991.

Kerr, et al. Proc. 13th IEA, 4:64-65, 1997.

Kumar, S. Human Factors, 36(2):197-209, 1994.

Kumar, S. Spine, 15(12):1311-1316, 1990.

Marras, et al. Spine, 18(5):617-628, 1993.

Moore, et al. Am.Ind.Hyg.Assoc.J., 56:443-458, 1995.

NIOSH, Tech. Report No. 81-122, 1981.

Norman, et al. Clin. Biomech. (accepted), 1998.

Punnett, et al. Scan.J.W.&En.Hlth., 17:337-346, 1991.

ACKNOWLEDGMENTS

The IWH and WSIB of Ontario for funding, Dr. Mickey Kerr for his statistical expertise, Drs. Richard Wells, Donald Chaffin and Harry Shannon, and Jack Callaghan, Patrick Neumann and Anne Moore, for their helpful comments and support.