**4. Discussion**

The results of this study show a strong association between visiting an on-site occupational health center and subsequent termination of employment within 60 days from the date of hire among a large sample of newly hired manufacturing workers. Specifically, the risk of termination was more than double among those who visited the on-site occupational health center compared to those who did not. The magnitudes and precisions of the risk estimates were also consistent across different injury classifications and mostly consistent across different body parts/areas affected. Associations were also elevated regardless of the reason for termination, although of somewhat greater magnitude among those with terminations classified as involuntary compared to those with terminations classified as voluntary.

Few longitudinal studies are available to which the results of the current study can be compared. Most recently, Okechukwu et al. [18] examined the association between self-reported occupational injury and both involuntary and voluntary "job loss" among a sample of 1331 nursing home workers. Injury data were collected at the time of enrollment and again after six and 12 months of follow-up, and then linked to employers' administrative records to identify those with job loss occurring in the subsequent six months. Overall, 24.2% of the sample experienced job loss within the 18 months of

observation, which is much lower than the frequency of employment termination observed in the current study (in the current study, 31.2% of new hires were employed fewer than 60 days from the date of hire). Similar to the current study, statistically significant associations were observed between occupational injury and both voluntary and involuntary job loss, and the magnitude of the association was also greater for involuntary job loss. However, important differences in the study sample (nursing home workers with an average of 6.3 years of experience at the time of enrollment vs. newly hired manufacturing workers) and differences in the nature of work between the cohorts limit comparisons between the results of our study and those reported in Okechukwu et al. [18].

The incidence rates of early OHC visits were approximately 50–60% greater than the incidence rates observed across the full study period, suggesting an increase in the risk of occupational injury during the earliest stages of employment. Gerr et al. [24] reported results from a prospective study of physical risk factors and upper extremity musculoskeletal outcomes among 386 workers at the same facility as that of the current study. In contrast to the current study, Gerr et al. [24] ascertained incident musculoskeletal symptoms with a weekly self-reported survey and incident musculoskeletal disorders via clinical evaluation (following a self-report of symptoms). Compared to the employees included in the current study, participants of Gerr et al. [24] were experienced (average of 15.8 years at the facility at the time of entry vs. all new hires), older (mean age 43.1 years vs. 33.9 years), and less frequently male (48.1% vs. 68.0%). Incidence rates were reported as 58/100PY for hand/arm symptoms, 19/100PY for hand/arm disorders, 54/100PY for neck/shoulder symptoms, and 14/100PY for neck shoulder disorders. Analogous incidences rates in the current study (by combining the nature of injury categories with the body part/area affected categories) were 8.5/100PY for OHC visits classified as either acute sprain/strain or repetitive strain and affecting the wrist/hand and 8.4/100PY for injuries classified as either acute sprain/strain or repetitive strain and affecting the neck/shoulder. The difference in incidence rates might appear to contradict the evidence suggesting that the risk of occupational injury is greatest during the earliest stages of employment. However, the active case-finding approach used by Gerr et al. [24] is expected to result in greater observed incidence rates than the use of passive surveillance sources, such as the OHC database used in the current study [25]. It is possible that only those experiencing the greatest levels of musculoskeletal discomfort reported to the OHC. In addition, it is possible that some employees, upon experiencing musculoskeletal discomfort, elected to terminate employment but did not report to the OHC.

Error in the ascertainment of dates of employment termination (of any type) was unlikely given the use of human resources data and inclusion of all newly hired employees in the study sample. While it is possible that some terminations were recorded one or more business days following the actual event, we have no way of validating the accuracy of the termination dates. Regardless, any error was unlikely to have differed systematically between those who visited the OHC and those who did not. However, the classification of each termination as involuntary or voluntary relied on our interpretation of the information included in the human resources database. We discussed our termination classification strategy with the employer prior to analyses. The only heterogeneity of opinion occurred for the "3-day no-call/no-show" reason for termination (*n* = 216), which we classified as voluntary. Ultimately, we believe our choice was appropriate since the employee made an active decision both to not report to work and to not communicate an explanation for the absence.

Errors in the ascertainment of exposure may have occurred. First, it is possible that some employees experienced an occupational injury and did not report to the OHC (despite employer policy). If employment termination during the probationary period were more likely as a result of the unreported occupational injury, then the observed hazard ratios would have been attenuated. We believe it is likely that the dates of events classified as general occupational injuries (e.g., chemical exposures and foreign objects in the eye) were captured accurately given their acute nature and the policy requiring employees to immediately report to the OHC. However, the date of an event classified as "acute sprain/strain" or "repetitive strain" does not necessarily reflect the date of symptom onset. Any lag between the onset of symptoms and the OHC visit date would increase the number of

unexposed days and decrease the time to termination following the OHC visit date, and therefore inflate the observed hazard ratios. However, we have no reason to believe that the frequency or duration of reporting lags were of sufficient magnitude to cause meaningful bias of the estimated hazard ratios. Finally, the OHC was staffed by multiple occupational health nurses and so some (inter-observer) misclassification may have occurred of the nature of the injury/event which brought the employee to the center and/or the body part/area affected.
