*2.9. Statistical Analyses*

The duration of employment was dichotomized as ≤60 days (i.e., termination of employment during the probationary period) and >60 days (i.e., working at least to end of the probationary period). The demographic variables (gender, age, race/ethnicity) and job characteristic variables (shift, and nature of work) were then stratified and summarized by duration of employment for descriptive purposes. Age was reported using the mean and standard deviation; all other variables were reported using observation frequencies and proportions.

We calculated incidence rates of OHC visits to provide additional context to the analysis. Specifically, incidence rates were calculated as the number of first-time OHC visits divided by the total person-time at risk of visiting the OHC across the full duration of the study period (i.e., 2 January 2012 to 25 November 2016). For employees with an OHC visit, person-time at risk was censored on the date of the first OHC visit recorded in the OHC database. For employees with no OHC visit, person-time was censored on either (i) the date at which employment was terminated or (ii) the end of the study period. In addition, we calculated incidence rates of first-time OHC visits that occurred during the 60-day probationary period (i.e., early OHC visits). The same censoring criteria were used for these calculations, although each employee contributed a maximum of 60 days of person-time at risk of visiting the OHC.

Hazard ratios (HRs) of the crude associations between early OHC visits and time to termination during the 60-day probationary period were estimated using Cox regression methods [21,22]. Nine models were constructed, each representing one of the combinations of one OHC visit definition (any early OHC visit, nature of injury, or body part/area affected) and one time to termination definition (any, involuntary, or voluntary). In all models, the referent category was limited to those employees without an early OHC visit. Time-varying measures of the three OHC visit variables (listed immediately above) were created to capture person-time both before and following an OHC visit [23]. Specifically, among those with an early OHC visit, the number of days from the date of hire to the date of the early OHC visit was analyzed as unexposed time (but time while still at risk of employment termination), while the number of days from the date of the early OHC visit to the date on which person-time was censored was analyzed as exposed time. Among those without an early OHC visit, all person-time of observation was analyzed as unexposed time. Also, in models with involuntary termination as the outcome event, we included person-time accumulated among those with voluntary or other terminations because these employees were still at risk of involuntary termination prior to the time of voluntary or other termination. Similarly, in models with voluntary termination as the outcome event, we included person-time accumulated among those with involuntary or other terminations

because these employees were still at risk of voluntary termination prior to the time of involuntary or other termination.

For all crude analyses, the proportional hazards assumption was tested by including interaction terms between each predictor variable and time; no statistically significant interaction terms were observed, indicating that the proportional hazards assumption was not violated. Cox regression methods were also used to estimate crude associations between each demographic and job characteristic variable and time to each of the termination types (any, involuntary, and voluntary).

Adjusted associations between early OHC visits and time to termination were also estimated using Cox regression. As in the crude analyses, models were constructed for each combination of one OHC visit definition and one time to termination definition (i.e., nine total models). Demographic and job characteristic variables crudely associated with the risk of employment termination (with *p* < 0.10) were included in all multivariable models. No further model specification criteria were applied (e.g., backward elimination) given the small number of available demographic and job characteristic variables relative to the sample size.

To assess bias of the adjusted hazard ratios between early OHC visits and time to involuntary termination potentially introduced by retaining in the analyses person-time accumulated among those with voluntary or other termination (and vice versa), additional analyses were conducted with datasets (i) restricted only to those with involuntary or no termination (i.e., excluding those with voluntary or other termination) and (ii) restricted only to those with voluntary or no termination (i.e., excluding those with involuntary or other termination). All statistical procedures were performed using SAS (version 9.4, SAS Institute, Inc., Cary, NC, USA).

#### **3. Results**

A summary of demographic and job characteristic variables is provided in Table 1. The mean age of the 3752 employees was 33.9 ± 10.9 years, and 68.0% were male. 58.4% of the employees identified as white/Caucasian, 39.1% as black/African-American, 2.7% as Hispanic/Latino, and 3.4% as other racial/ethnic designations. Larger proportions of new hires were placed into the first (34.6%) and second (43.7%) shifts compared to third (18.7%) and premium (3.0%) shifts. A substantial majority (87.2%) was assigned to assembly work.

1172 (31.2%) of the new hires were employed 60 days or less. Of these, 793 (67.7%) were terminated for involuntary reasons, 345 (20.4%) for voluntary reasons, and 34 (2.9%) for other reasons (Table 1). Differences in the distributions of demographic and job characteristic variables between employment duration strata (i.e., employed >60 days vs. employed ≤60 days) were generally small. However, those employed ≤60 days were less frequently male, more frequently white/Caucasian, and more frequently assigned to assembly work. In addition, a smaller proportion of those employed ≤60 days was assigned to the premium product line, although the number of new hires in this category was relatively small.

Employees contributed a total of 856,845 person-days (2348 person-years (PY)) of observation during the study period (2 January 2012–25 November 2016). During this time, a total of 1090 first-time OHC visits were recorded, yielding an overall incidence rate (IR) of 46.4/100PY. Of the 1090 first-time visits, 453 (41.6%) were classified as general occupational injuries (IR = 19.3/100PY), 429 (39.3%) as repetitive strain (IR = 18.3/100PY), and 208 (19.1%) as acute sprain/strain (IR = 8.9/100PY). The body part/area affected most commonly was the wrist/hand (29.3%), followed by the shoulder/arm (25.2%) and the low back (10.9%), with other body parts/areas accounting for the remainder (34.6%).



Asian/Pacific Islanders, Alaska natives, and American Indians; employees assigned to the premium product line worked four, 10-hr

> Notes: 1 Age

shifts per week.

Of the 1090 first-time OHC visits, 339 (31.1%) occurred during the 60-day probationary period (i.e., early OHC visits). The median number of calendar days from the date of hire to the date of any early OHC visit was 20 (interquartile range: 10–32 days), with a range zero days (i.e., employee reported to the OHC on the first day of employment) to 60 days. Employees contributed a total of 173,816 probationary days (476 PY) of observation, resulting in an incidence rate for early OHC visits of 71.2/100PY. Of the 339 early OHC visits, 142 (41.9%) were classified as general occupational injures (IR = 29.8/100PY), 129 (38.1%) as repetitive strain (IR = 27.1/100PY), and 68 (20.0%) as acute sprain/strain (IR = 14.3/100PY). The body part/area affected most commonly was the wrist/hand (28.3%), followed by the shoulder/arm (24.5%) and the low back (12.1%), with other body parts/areas accounting for the remainder (35.1%).

Crude estimates of association between the demographic and job characteristic variables and time to termination are provided in Table 2. Statistically significant reductions in the risk of any employment termination were observed among (i) males (HR = 0.79, 95% CI = 0.70–0.89), (ii) those identifying as black/African-American (HR = 0.77, 95% CI = 0.68–0.87) and other racial/ethnic categories (HR = 0.70, 95% CI = 0.49–1.00), and (iii) those assigned to the premium shift (HR = 0.26, 95% CI = 0.14–0.48). In contrast, the risk of any employment termination during the probationary period was elevated among those assigned to assembly positions (HR = 3.69, 95% CI = 2.79–4.87). This pattern was consistent for involuntary and voluntary terminations. However the precision of many HR estimates was reduced as a consequence of small cell sizes. For example, the risk of voluntary terminations among those assigned to assembly positions was highly elevated (HR = 7.0, 95% CI = 3.60–13.56), but the referent category (i.e., those assigned to non-assembly positions) included only nine (2.6%) of the 345 voluntary terminations. A few differences were observed when examining the associations among those with involuntary and voluntary employment termination. Most notably, among those assigned to second shift, the risk of involuntary employment termination was reduced (HR = 0.77, 95% CI = 0.66–0.90) while the risk of voluntary employment termination was elevated (HR = 1.72, 95% CI = 1.72–2.21).


**Table 2.** Crude associations between demographic and job characteristic variables and employment termination within the 60-day probationary period.

Notes: <sup>1</sup> includes Asian/Pacific Islanders, Alaska natives, and American Indians; <sup>2</sup> employees assigned to the premium product line worked four, 10-h shifts per week.

Crude and adjusted estimates of association between early OHC visits and employment termination during the probationary period are provided in Table 3. Any instance of an early OHC was associated with a statistically significant increase in risk of any employment termination prior to the end of the probationary period (adjusted HR = 2.58, 95% CI = 2.12–3.15). The effect magnitude was similar when analyzing OHC visits by the nature of the injury, with statistically significant adjusted HRs of 2.38 for repetitive strain, 2.61 for acute sprain/strain, and 2.78 for general occupational injuries. Associations were also statistically significant when analyzing early OHC visits by the body part/area affected, although of somewhat greater magnitude for the shoulder/arm (adjusted HR = 3.58) and low back (adjusted HR = 3.12) than for the wrist/hand (adjusted HR = 1.95) and other body parts/areas (adjusted HR = 2.26). The magnitude of effect was generally greater for involuntary termination than for voluntary termination, but substantial overlap of the 95% confidence intervals around the HR estimates suggests minimal statistical difference (if any).


**Table 3.** Crude and adjusted <sup>1</sup> associations between occupational health center (OHC) visits and employment termination within the 60-day probationary period. <sup>2</sup>

Notes: <sup>1</sup> All models adjusted for gender, race/ethnicity, shift, and nature of work based on associations with *p* < 0.10 in crude analyses; <sup>2</sup> referent category in all models includes those with no OHC visit and no termination of the specified type within the 60-day probationary period.

Results of the additional analyses with restricted datasets showed minimal change in the adjusted HRs. Specifically, HRs in the additional analyses were, on average, 6% greater than those presented in Table 3. In no case was a HR reported in Table 3 greater than the analogous HR estimated in the analysis of the restricted dataset.
