National Evaluation of the Association Between Resident Labor Union Participation and Surgical Resident Well-being

Key Points Question Is the presence of a resident labor union associated with improved well-being at US surgical training programs? Findings In this cross-sectional survey study of 5701 residents, unionized programs were more likely to offer housing stipends and 4 weeks (instead of 2-3 weeks) of vacation time to residents; however, no difference in burnout, suicidality, job satisfaction, duty hour violations, mistreatment, salary, or the educational environment were found between residents at unionized and nonunionized programs. Meaning These findings suggest that resident labor unions do not appear to improve resident well-being.


eMethods 1. Selected Items From Survey Administered to Residents After the 2019 ABSITE INSTRUCTIONS
Please complete the following survey to share your thoughts about your residency experience. The information will be used to inform future research and policy decisions to improve the learning environment and culture of residency. It is estimated that it will take approximately 5 to 8 minutes to complete the questions. The survey responses are never associated with your personal identity. All data are de-identified for analyses and reporting. Your program will not have access to your individual responses.
1. Are you currently enrolled in a clinically active year of your residency? This is defined as the PGY1-5 years that count toward your board eligibility. An instrumental variable modeling approach was employed in this study to account for suspected endogeneity of the variable denoting whether a program was unionized. Sources of endogeneity were hypothesized to include unmeasured confounding program characteristics which may influence the decision to unionize, and simultaneity, where outcomes related to poor working conditions or well-being may have contributed to the decision to form a union. The unionization rate of non-healthcare employees within a given region was selected as an instrumental variable for this study because it meets the following criteria: 1. It is strongly associated with the unionization status of residency programs within the region, irrespective of other confounding variables.
2. It is not itself associated with the outcomes of interest though any mechanism except program unionization status, as it does not directly affect outcomes (the instrument does not measure healthcare worker unions) and other regional confounding factors (geographic region, urban-rural classification, and where pertinent, income and housing costs) are included in the models.
The causal graph of the model is depicted below: Included within the error term in the causal graph are several measured covariates which are incorporated as adjustments in models. After adjusting for these covariates, no association between the instrument (regional unionization rate) and the outcomes is believed to be present except through the primary exposure variable (program union status). The instrumental variable models used in this study can theoretically identify the direct influence of program unionization status on the outcomes independent from other unmeasured confounders (such as individual program characteristics that may confound union status and outcomes) and simultaneity, which the naïve regression models included in the manuscript are unable to do. Program did not take wellness seriously, No. (%) 67 (11.9) 393 (7.7) 1.39 (0.94-2.07) a Each row represents a separate model assessing the association of unionization status with each outcome. Residents with missing responses for the following outcomes were excluded from the model for that outcome: thoughts of attrition (n = 16), dissatisfaction with decision to become a surgeon (n = 12), dissatisfaction with time for rest (n = 9), duty hour violations (n = 118), discrimination (n = 414), bullying (n = 127), sexual harassment (n = 165), dissatisfaction with educational quality (n = 10), inadequate time for patient care (n = 33), protected educational time (n = 64), inadequate time in operating room (n = 63), inadequate autonomy in operating room (n = 64), inadequate autonomy in clinical decisions (n = 47), lack of effective support staff (n = 34), program non-responsiveness to resident concerns (n = 51), and program did not take wellness seriously (n = 37).

Regional unionization rate
b Logistic regression models estimated odds ratio for outcome among residents at programs unionized for more than 3 years compared with programs that are not unionized or have been unionized for less than 3 years. Covariates included gender, race, Hispanic ethnicity, relationship status, parental status, census region, urban-rural classification, and program size. Odds ratios greater than 1 indicate worse outcome at unionized programs.
c Includes discrimination based on gender, gender identity, sexual orientation, race, ethnicity, and religion. b Linear regression model of salary for programs which have been unionized for more than 3 years compared with programs which are not unionized (n = 170) or have been unionized for 3 or fewer years (n = 4) Covariates included program size, program type, census region, urban-rural classification, and county median household income as covariates.
c Logistic regression models estimated odds ratio for residents at programs which have been unionized for more than 3 years compared with programs which are not unionized (n = 170) or have been unionized for 3 or fewer years (n = 4). Covariates included program size, program type, and urban-rural classification as covariates. For the housing stipend model, county-level median rent was also included. a Each outcome is assessed as using a linear probability model to permit comparison of coefficients between naïve and IV models. CI denotes confidence interval.
b Absolute and relative difference reflects difference in estimates from IV regression models versus naïve models. NE denotes not able to be estimated (due to a naïve model estimate of 0). c Includes discrimination based on gender, gender identity, sexual orientation, race, ethnicity, and religion. d Remains significant after Bonferroni correction for multiple comparisons.