Page 93 - AIH-2-3

Artificial Intelligence in Health Organizational culture’s impact on burnout

working with colleagues. This finding aligns with several The small dataset may have led to model overfitting,
studies showing that OC influences burnout symptoms, meaning it might have learned patterns specific to
including depersonalization. 48-50 However, our approach the training data rather than those applicable to more
is innovative in its ability to use a complex data model generalizable data. The model could have overreacted to
to predict burnout based on OC. The aforementioned small variations in the data, such as outliers or the responses
studies did not employ a random forest model to predict of a few respondents who rated their OC as positive with
burnout scores among non-patient-facing and patient- a “1.” Moreover, the small dataset may not have captured
facing employees, highlighting the novelty of using a the complex, diverse perspectives of employees across
random forest algorithm for this purpose. Moreover, the different health systems, such as those who viewed their
models’ strong performance, despite the small sample size, OC negatively but did not exhibit significant burnout
suggests that this approach can be expanded to predict symptoms. As a result, the model may not have fully learned
burnout using OC and potentially other factors such as all the relationships between the features, potentially
workload. The model’s ability to predict burnout based on performing poorly on unseen data due to noise specific to
OC is supported by a previous study that used a decision the small dataset. Future studies should incorporate larger
tree model and Bayesian analysis on the same dataset. datasets with a broader representation of various roles
32
The Bayesian analysis, which, like a random forest model, within health systems to improve generalizability, reduce
is suitable for small datasets, corroborates the results of this bias, and prevent overfitting.
study.
6. Conclusions and policy implications
The AUC-ROC curves for Models 1 and 2 demonstrate
that the model classified, with some accuracy, whether This random forest model demonstrates that perceptions
employees perceived their organization’s culture positively. of OC can be used to predict specific burnout symptoms,
such as engagement with others and empathy. The
Furthermore, the model suggests that burnout symptoms model also reveals that the more positively employees
are more likely to be linked to perceptions of OC. This perceive their organization’s culture, the less likely they
finding indicates to leaders that, if employees exhibit are to exhibit burnout symptoms. Previous studies have
burnout symptoms, they may also perceive the OC as quantified the effects of OC on burnout, 50,51 and this
negative. Leaders could use the model’s identification of model further highlights that OC influences burnout more
specific burnout symptoms – such as depersonalization, significantly than internal factors, such as resilience and
as indicated by callousness toward others – to evaluate self-care. If employees begin to display burnout symptoms,
perceptions of OC. For instance, if leaders recognize leaders could assess and improve the OC to mitigate these
that their employees are treating others harshly and effects. Therefore, it is crucial for leaders in health systems
exhibiting other burnout symptoms identified by the to cultivate an OC where colleagues and employees are
model (e.g., viewing others as impersonal objects), they supported, workloads are manageable, and the resources
could begin to foster a more positive OC. Strategies needed to perform the job effectively and efficiently are
could include providing employees with the necessary provided. 52-55 As a result, perceptions of OC will improve,
equipment and resources to succeed, as well as offering and employees will be less likely to experience burnout.
flexible work schedules to promote work-life balance and
reduce burnout symptoms. The findings also suggest that policymakers could
invest in improving health systems’ work environments
While this exploratory study introduced a novel by providing flexible work options and reducing workload
random forest method for predicting burnout using OC, it through strategies such as increasing staffing. For example,
has several limitations. One notable limitation is the small given the shortage of healthcare workers, particularly
sample size. Although a previous study’s Bayesian analysis nurses and physicians, in the United States, hiring
on the same dataset supports the results of the random international employees could help alleviate the staffing
forest model, other ML methods suited for small sample crisis. Policymakers could ease the hiring process by
sizes, such as SVMs, could be explored in future studies. loosening visa requirements for international employees.
32
Nonetheless, this exploratory study provides a framework In addition, health system leaders and policymakers should
and methodology for demonstrating how OC influences focus on internal factors contributing to burnout, such as
burnout. Moreover, this study was cross-sectional, mental health. Policies could be implemented to invest
meaning causality could not be established. It was also more in mental health resources, such as crisis hotlines.
conducted solely in the United States. To determine the Leaders in health systems could also establish confidential
generalizability of the results, the study could be replicated peer support groups, enabling employees to discuss their
in other countries. mental health concerns without fear of their discussions

Volume 2 Issue 3 (2025) 87 doi: 10.36922/aih.5127