From 2004 to 2008, we conducted a prospective cohort study of occupational risk factors for UEMSDs among 387 household appliance manufacturing workers in Iowa, United States (the Iowa Consortium Study). The participation rate was 52%. A brief description of the methods used to conduct the Iowa Consortium Study is provided later, and a detailed description of the methods is provided elsewhere (Gerr et al., 2014). All employees performing production work at a large appliance manufacturing facility were eligible to participate in the Iowa Consortium Study. Participants who were symptomatic at enrollment or who performed cyclic tasks with work cycles longer than 6 min were not included in the current analyses. Among 387 Iowa Consortium Study participants, 276 were included in the current analyses.

Demographic, personal, and occupational psychosocial information was collected on two self-administered questionnaires completed by participants at the time of enrollment. The Job Content Questionnaire (JCQ; Karasek, 1985; Karasek & Theorell, 1990) was used to estimate psychological job demands (i.e., demand), decision authority (i.e., control), coworker support, and supervisor support. A four-category job strain variable was created from JCQ results by splitting the demand and control score distributions at their respective median values (i.e., 1 = low demand, high control, 2 = high demand, high control, 3 = low demand, low control, and 4 = high demand, low control). Negative affectivity (a person’s tendency to experience unpleasant feelings) was assessed with the Positive Affectivity and Negative Affectivity Scales (Watson, Clark, & Tellegen, 1988).

We dichotomized participants into hand–arm symptom categories using information about hand–arm symptom quality, severity, and duration recorded by participants on weekly diaries. Participants who (a) reported hand or arm pain, numbness, tingling, or burning for at least 30 min during the previous week with either a severity level of at least 5 on a 0 to 10 visual analog scale or reported use of analgesic medication and (b) denied acute trauma as an immediate cause of the symptoms were categorized as symptom positive (Sx+).

On a weekly basis, participants recorded on preprinted logs information about (a) daily hours worked per task, (b) changes in work activities, (c) current work stress, (d) time spent performing non-work-related hand-intensive activities (e.g., gardening, playing video games), and (e) time spent working at a second job. In addition to the self-reported information, 10 to 20 min of simultaneous sagittal (side view) and frontal plane (anterior or posterior view) video was recorded of all study participants while performing each of their tasks. Because of time constraints, our research team did not collect video information for all tasks. The proportion of missing video data was relatively small (5% of total task hours) but affected 32% (n = 89) of the participants. For most participants with missing video data, SI physical and temporal task exposure ratings were imputed in accordance with the imputation procedures established for the Iowa Consortium Study (Gerr et al., 2014). Three representative work cycles were identified from the video of each cyclic task, and two trained observers, using standard SI methods (Moore & Garg, 1995), rated each task’s overall intensity of exertion, hand–wrist posture, temporal exertion requirements (exertions per minute and percentage duration of exertion), and speed of work.

To maximize sample size, modified SI procedures were developed for rating noncyclic tasks in this study. Specifically, for noncyclic tasks, trained observers watched 20-min video segments to identify the work element (e.g., remove the clear film protective coating from door, move a door from the conveyor to a rack) with the longest total duration (the most common work element). The observers then viewed the most common work element and rated the overall intensity of exertion, hand–wrist posture, and speed of work using standard SI methods (Moore & Garg, 1995). When exposure estimates across noncyclic tasks in a particular work area (e.g., crating, brazing) were similar (i.e., had low coefficient of variation), the mean values for percentage duration of exertion and efforts per minute were assigned to all noncyclic tasks in that work area; otherwise, facility-wide mean values were assigned. Several other exposure assessment approaches (e.g., the Hand Activity Level, surface electromyography) were used to estimate exposure to physical risk factors in the Iowa Consortium Study, but those data were not used for the current study.

A total of 886 tasks were observed among study participants. Because many participants performed the same task(s), repeated estimation of SI values for common tasks performed by multiple participants would have been redundant and was not feasible because of limited resources. Therefore, prior to data extraction, we collapsed the 866 tasks into 179 task similarity groups and 162 unique, solitary tasks. Tasks were grouped by similarity when they were located in the same area of the plant (e.g., basepan, crating, shelving), had the same function (e.g., assemble basepan, braze basepan, install gaskets, secure lids on crates), were characterized by the same intensity of exertion rating (using standard SI methodology), and were performed using the same dominant hand. Most participants with missing video data were assigned task similarity groups based on knowledge of the facility and task.

The two trained observers used a consensus approach to select the final SI physical and temporal exposure ratings (overall intensity of exertion, hand–wrist posture, exertions per minute, percentage duration of exertion, and speed of work). The remaining SI parameter, task duration per day, was estimated from daily task hours worked recorded by each participant on the daily task log. Procedures to calculate SI scores for multitask jobs vary somewhat among investigators (Bao et al., 2009; Garg, 2006). In this study, all SI scores for an entire shift were calculated for all participants according to Cumulative Strain Index (CSI) formulas developed by Arun Garg, codeveloper of the SI. SAS code used to calculate SI scores with the CSI method was adapted from Microsoft Excel macros developed at University of Wisconsin– Milwaukee by Jay Kapellusch and Arun Garg (personal communication, May 12, 2008). Task-level SI scores for single-task jobs calculated with the CSI method are the same as SI scores calculated using standard SI methodology. A participant’s peak daily SI score for the work week (Monday–Sunday) was used to assign weekly SI risk categories for each participant. All peak daily SI scores were calculated in SAS. More detailed documentation for the CSI calculation method used in this study is available elsewhere (Meyers, 2010).

As previously mentioned, numerous methods of parameterizing SI categories exist. For this study the cut points for the four ordinal “Original” risk categories were ≤3, >3 to <5, ≥5 to <7, and ≥7. In this study, an alternate set of SI score cut points was defined based on quartile values of the distribution of SI scores calculated for the week during which Sx+ participants (n = 97) first reported symptoms. In this way, approximately equal numbers of incident Sx+ events were distributed across the four Empirical SI risk categories (Category 1_EMP–Category 4_EMP), resulting in the greatest estimated precision and stability. The specific SI values for these empirically obtained categories were ≤8.72, >8.72 to <13.5, 13.5 to <18.56, and ≥18.56. The cut points for both the Original and Empirical SI risk categories were used to categorize weekly SI scores for each participant.

Time-independent and time-varying demographic, personal health, and occupational characteristics for participants (N = 276) in the current study are presented in Table 1. The study sample mean age was 43 years, and the sample was nearly half female. Participants had long tenure at the facility (average 16 years employment).

Compared to participants, the mean age of nonparticipants in the parent study was 2 years younger, the mean number of years worked at the facility was about 4 years less, and a greater proportion were men (61% vs. 49%) and those who worked on the second shift (50% vs. 32%; data not shown).

One or more categories of both the Original and the Empirical SI risk category classification systems were statistically significantly associated with incident hand–arm symptoms, indicating that both have utility for examining exposure–effect associations, especially among female participants.

Results from the current study provide some empirical evidence to support the Original SI highest exposure strata, Category 4_ORIG, as recommended by Moore and Garg (1995). In particular, we observed a monotonic increase in risk across the four Original SI risk categories. More than 70% of weekly SI scores observed in the current study were 7 or higher and were therefore assigned to Category 4ORIG using the Original SI cut points. Thus, when using the Original SI cut points it was not possible to explore differences in risk that might exist across more than two thirds of the exposure since it was collapsed into a single category.

In contrast, the Empirical SI cut points were all higher than 7, enabling more precise estimation of associations between SI risk category and hand–arm symptoms across the large proportion of exposure samples that were classified in the highest SI risk category under the Original categorization system. Among the participants in this study, elevated risks were observed among the higher Empirical SI risk categories relative to the referent category. But in contrast to the Original structure, a monotonic increase was not observed across the Empirical SI risk categories. Furthermore, the only HR for the Empirical structure that was statistically significant was a twofold increase in risk for the second highest category.

Direct comparisons with previous studies using the SI as an exposure metric are difficult because health outcomes, study design, industry sector, SI score calculation methods (for multitask jobs), and cut points vary across studies. Similar to those for the current study, investigators for several studies have reported associations (although not always statistically significant ones) between various SI metrics and distal upper extremity musculoskeletal symptoms (Knox & Moore, 2001; Moore & Garg, 1995; Moore, Rucker, & Knox, 2001; Moore, Vos, Stephens, Stevens, & Garg, 2006), wrist tendonitis (Harris et al., 2011), rotator cuff tendonitis (Hegmann et al., 2006), or carpal tunnel syndrome (Garg et al., 2012; Silverstein et al., 2006).

One unexpected observation of the current study was that the highest strata of Empirical SI scores did not have the highest risk of hand–arm symptoms. The same pattern of lower observed risk among higher SI scores was observed in a recent study using the SI as an exposure metric among manufacturing workers (Garg et al., 2012). A possible explanation for the lower HRs observed among the highest Empirical SI risk category is the selection bias called selective survival. Specifically, because participants in this study had been working at the facility for many years (M = 16 years), it is possible that the workers most susceptible to hand–arm symptoms were underrepresented among the highest exposure strata because, after experiencing intolerable musculoskeletal effects, they self-selected into jobs with lower exposure levels or left the facility entirely (i.e., selective survival).

For members of this study sample, the association between Empirical SI risk category and hand–arm symptoms was modified by gender. In particular, a statistically significant difference in the strength of the association between Empirical SI risk category and hand– arm symptoms was observed between men and women, especially for the highest exposure strata. Specifically, compared to the referent category, the risk of hand–arm symptoms among the highest strata of Empirical SI scores was doubled in comparison to the referent group among women, whereas a nonsignificant risk of less than unity was observed among men. Prior research has shown that men underreport upper extremity musculoskeletal symptoms compared to women, which could account for some of the observed gender effects (Silverstein et al., 2009).

The findings of this study suggest that the SI score category may predict incident hand–arm symptoms among workers performing single-task and multitask manufacturing jobs similar to those performed by the study sample. However, about 90% of tasks at this facility were rated as “light” or “somewhat hard” on the SI “intensity of exertion” task rating subscale (a subscale intended to capture the forcefulness of hand– arm exertions). Observed associations between hand–arm symptoms and the SI metrics may have been attenuated because of the limited range of exposure to forceful hand–arm work performed at this facility.

Nondifferential sources of error in SI scores may have affected the results of this study. One source of nondifferential error in the SI scores was the use of task similarity groups for assigning a common SI score to similar tasks rather than conducting separate assessments for each task. This practice introduced some error into both Original and Empirical SI risk categories by artifactually reducing the observed exposure variability in comparison to the true variability. Another source of nondifferential error was using the most common work element for noncyclic tasks to assign SI scores. In both cases, nondifferential error in exposure estimation will attenuate observed associations between exposures and health outcomes (Jurek, Greenland, & Maldonado, 2008), which would not have affected comparisons between the Original and Empirical multivariable survival analysis models.

Only manufacturing workers were included in this study; therefore, it is uncertain whether the Empirical SI risk category cut points used in this study can be generalized to other industries. However, compared to the Original SI cut points, using the higher Empirical SI cut points presented in this study may provide a better way to identify hazardous manufacturing jobs.

Another limitation of the current study was its sample size. Among published epidemiologic studies of musculoskeletal outcomes among manufacturing workers, a sample size of 276 is not considered small. Regardless, statistical power in this study was limited because of sparseness of information within some strata of categorical variables. Estimates of association for the Original SI cut points were unstable because of sparse numbers of Sx+ participants in Original SI risk Categories 2 and 3. In addition, analyses by gender were especially limited by sample size considerations.

The modest participation rate of 52% may have resulted in distortion of the study sample. It is possible that the associations between exposure and outcome among participants were different than among nonparticipants. However, a participation rate of 52% is consistent with the experience of many investigators conducting prospective occupational cohort studies. Furthermore, because differences in associations between the Original SI risk category structure and hand–arm musculoskeletal symptoms and the Empirical SI risk category structure and hand–arm musculoskeletal symptoms were observed among the same set of participants, we have no reasons to believe that they were an artifact of differential participation.

Future research is needed to test the Empirical SI cut points presented in this paper (a) when assessing single-task jobs, (b) when multitask SI scores have been calculated using alternate methods, and (c) among other populations of both manufacturing and nonmanufacturing industries.

In addition to future research, future software development could be another way to make the use of multitask SI computation methods such as the CSI more accessible to practitioners. Until CSI calculation software becomes available, widespread use of the CSI is unlikely because manual calculation of more than a few multitask SI scores may be too time-consuming.