Epidemiological studies have shown that low back disorders are associated with extreme and repetitive trunk postures, high moments of force, and heavy lifting while twisted. Other exposure variables including lumhar spine compression and shear forces are suspected to be related to low back injury risk in industrial workers. Traditional biomechanical methods such as computerized video, and electromyography (EMG), are usually cost prohibitive for quantifying exposure from risk factors such as these on the scale necessary for epidemiological studies. Questionnaires have been widely used for large scale physical exposure assessment, but have not compared well with more instrumented methods.

Physical exposure data from a large case/control epidemiological study on the reporting of low back pain, and from several laboratory experiments, are included in this thesis. Three papers are presented whose purpose was to determine how well the magnitudes of peak and repetitive low back physical exposure variables resulting from worker reports on a questionnaire (QR) compared to the same variables determined using video analyses (VID) and trained observation (OBS). Also included is a description and evaluation of a model which was developed, using an additive approach, to combine multiple exposure variables into a single index estimate of work-related low back physical demands (demands index = DI) and injury risk (risk index = RI).

The magnitude of peak low back exposure variables (spine compression and shear forces, moment, trunk angle and hand load) estimated from QR self-reports from 99 industrial workers, did not compare well with those from VID (correlations from r=0.1 to 0.4). The agreement was better (correlations from r=0.6 to r=1.0) for spine compression forces when simple laboratory tasks were analyzed (n=27), suggesting that investigator assistance with the questionnaire may improve the accuracy of self-reports, and should be studied in the future.

The agreement between OBS and QR methods for individual workers was also poor in general (correlations from r=0.01 to 0.61) for 9 repetitive loading variables expressed on a per shift basis (# trunk twists, # trunk lateral bends, # moderate and severe trunk flexions, # trunk extensions, # arms overhead, # heavy lifts, # squats, and static postures (Y/N)). On average, workers reported a higher number of repetitions for 7 of the variables than were recorded by trained observers. Differences in how precisely some variables were defined for workers and observers may explain the generally poor agreement between the methods.

A select list of 19 variables was analyzed independently using the Di model. In addition, 4 combined analyses of 3 variables each were performed to give job estimates of overall low back exposure and risk of injury. The variables of one combined analysis were determined using a data driven approach (logistic regression), and the remaining 3 were determined by the experimenter. The APDFs of 17 of the 19 variables, and all 4 combined analyses, indicated that the low back physical demands of workers who reported pain (n=104) were higher than for workers who did not report pain (n=129). In general, physical loading variables such as peak compression and moment, resulted in higher mean DIS (0.41, and 0.34, respectively) than postural variables such as %time severely flexed, and %time twisted (0.08, and 0.07, respectively). Mean and peak risk indices for the data driven variable combination (7.18 and 15.3, respectively) were at lease twice as large as for the experimenter driven combinations and for all individual variables. This illustrates that the additive effect of multiple variables can produce higher estimates of risk than single variables, and that a statistical approach like logistic regression, may be the better method in terms of risk, than an entirely experimenter driven approach, for determining input variables to the model.