Modeling Work Intensity in a Pediatric Training Program – a Pilot Study

Background: The Accreditation Council for Graduate Medical Education requires training programs to monitor "scheduling, work intensity, and work compression that impacts resident well-being."(1) A prior study quantified resident workload using order entry and clinical documentation but did not find an association between residents’ workload and their self-perceived work intensity. (2) We hypothesized that looking more broadly at patient and clinical factors within our electronic systems would identify factors associated with increased work intensity as self-reported by residents.

Methods: This study occurred at an urban, quaternary, free-standing children’s hospital and was deemed exempt by our institutional review board. We defined work intensity as a measure of activity related to a trainee’s job over their shift. We identified measurable factors related to resident work intensity within our electronic health systems. Factors were extracted for a pilot period between 12/16/2020 – 5/04/2021. We surveyed four interns on general pediatrics rotations each weekday (excluding holidays) during the study period. The survey asked interns to rank work intensity on a scale from -100 (bored) to +100 (exasperated). Residents who rated >90 were sent a follow-up email encouraging them to utilize available support resources. We used multiple linear regression to identify patient and clinical factors that correlate with subjective scores of work intensity.

Results: We received 102 survey responses (28.33% response rate) for 19 unique interns during the study period. Bucketed work intensity scores ranged from -2 to +2 (mean= 0.2, SD= 1.09). Three surveys prompted referral for support services. We identified 22 candidate variables for the regression model, including 8 patient factors and 12 clinical factors (Table 1). Five variables were transformed due to abnormal distributions. The two most significantly associated variables from our initial univariate regression model were texts (p < 0.0009) and orders (p < 0.0002). Stepwise regression produced a reduced model (R2 =0.2179) with nearly all variables reaching statistical significance (Table 2).

Conclusion: This pilot demonstrates that various patient and clinical variables are available in electronic health systems and can be used to model resident perceptions of work intensity. Orders placed had the highest correlation with work intensity. Our model accounted for about 20% of the variability seen in subjective work intensity scores, but notably we were unable to assess trainee knowledge base and other external factors (personal life events) that undoubtedly contribute to self-perceived work intensity. Future efforts should expand data collection to include more trainees, other rotations, and additional variables. Larger studies are needed to determine whether the model can be used across clinical settings and training programs. Ultimately, training programs can utilize such models to tailor staffing and curriculum to support trainee wellbeing.