記住我
Burnout is a physical and mental exhaustion syndrome that includes phenomena such as a negative self-concept and work attitude and loss of interest in patients. This is a form of stress response that occurs when stress can no longer be tolerated. Burnout occurs simultaneously in 3 dimensions [], that is, personal, work-related, and client-related dimensions []. The World Health Organization has defined burnout as a work-related risk factor that affects health and has included it in the 11th edition of the International Classification of Diseases in 2019 [].
Nurses are known to experience high levels of burnout. The prevalence of nurse burnout has been approximately 30% worldwide in the last 10 years []. With the COVID-19 pandemic, nurses’ workload has increased leading to increased burnout. Nurse burnout is a serious issue in the health care system as it leads to an increase in the nurse turnover rate []. With increased turnover, the nursing workforce has to be replaced with new nurses, resulting in reduced work productivity and an increased burden on hospital finances []. In addition, burnout negatively influences the quality of nursing services and leads to frequent absences [,].
Acknowledging the significance of nurse burnout, various programs have been developed and tested. In a study that reviewed burnout reduction interventions for nurses and doctors, interventions such as yoga, meditation, and mindfulness reportedly alleviated burnout []. However, the program was conducted in person, and interventions for medical personnel can be burdensome due to time and location constraints []. Moreover, as nurses work in shifts, there are time and space restrictions for interventions. However, web-based interventions do not require a facilitator and are, therefore, better than in-person interventions in terms of efficiency [].
In previous studies, interventions through mobile apps to confirm the effect of burnout were based on voice programs [,]. Previous studies on burnout interventions introduced one-size-fits-all programs. However, nurses experience varying dimensions of burnout. Artificial intelligence (AI)–based recommendation systems based on similarity are widely used in various fields. Netflix and YouTube recommend similar media based on individual content preferences [,]. In addition, in the marketing field, similar products are recommended by analyzing the commonalities of products that consumers have purchased []. As the accuracy and diversity of recommendations increase, user satisfaction increases []. Recommendation systems commonly use a user-based collaborative filtering system, which recommends items or content based on similarities between users’ interest profiles []. Digital Me, an AI-based product service system, measures a user’s state through user behavior analysis, predicts it, and makes recommendations. The recommendation system maximizes a specific score or finds similar but superior existing users, thus attaining its goal []. Therefore, a similar system could be used to provide a tailored program for nurse burnout.
Nurses’ burnout is influenced by various factors, such as their demographic and job-related characteristics [-]. Burnout patterns differ depending on coping strategies, whether working night shifts, working at a tertiary hospital, age, and department [-]. Therefore, when recommending burnout reduction programs for nurses, we could consider that similar burnout experiences are shared among individuals with similar general and job-related characteristics, rather than simply relying on user preferences or uniform delivery. The system we have developed measures the similarity between users based on demographics, research variables (job stress, stress response, and coping strategy), and burnout dimension scores.
Literature ReviewCognitive behavioral therapy (CBT)–based interventions are evidence-based interventions to reduce burnout and stress in health care professionals and improve mental health, healthy lifestyle beliefs, and job satisfaction []. Several programs have been suggested to reduce and prevent nurse burnout. One of these, mindfulness meditation, is a well-known technique for managing stress and burnout. Mindfulness is the attitude of paying attention to present moment events and accepting feelings, ideas, and sensations without passing judgment []. Mindfulness-based cognitive therapy is a group psychotherapy program that combines mindfulness and CBT and has been proven to effectively improve depression and prevent its recurrence []. Mindfulness-based training is effective in reducing burnout not only in nurses [] but also in other health care professionals [,]. In addition, mindfulness meditation has been reported to enhance positive cognitive retraining by increasing positive mood and brain activity []. A recent study demonstrated that a website and a smartphone app for mindfulness were effective in reducing burnout among health care professionals. [].
Another therapeutic intervention is laughter therapy. Laughter therapy improves healing and coping abilities in physical, psychological, and cognitive aspects by inducing laughter, smiles, and pleasant emotions through physical and intellectual interactions with the body and the mind []. It is well known that laughter therapy can reduce psychosocial conditions, such as stress, anxiety, and depression, in people with health issues, such as patients with cancer, middle-aged women, and mothers [-]; it has also recently been reported as an effective way to improve life satisfaction and reduce burnout in nurses []. Using humor to deal with stressful and burnout situations can increase happiness and life satisfaction in individuals by fostering more positive evaluations [].
Storytelling, one of the most fundamental and traditional forms of communication, can influence behavior by allowing people to express human identity and self-identify, share similar life experiences, and explain situations []. A form of CBT called “narrative CBT” uses storytelling []. The use of storytelling among nurses provides a space for nurses to talk about, listen to, perceive, and evaluate their experiences []. In previous studies, nurses who took part in storytelling workshops gave storytelling a positive evaluation []. In addition, expressive writing has been shown to improve mood, cognitive function, and mental health problems [,]. In a recent study that used storytelling and reflective writing as an intervention for oncology nurses, self-awareness and self-compassion reportedly increased, and insomnia and loneliness decreased []. Therefore, storytelling and reflective writing can improve personal well-being by providing skills to recognize one’s emotions and cope with work-related emotions.
Acceptance and commitment therapy (ACT) is part of the third wave of CBT []. It is a comprehensive psychological therapeutic strategy that can be used to treat a variety of problems and disorders. It incorporates acceptance and mindfulness processes, active engagement, and behavioral change processes []. Recently, it was provided to a group of employees in the workplace [] and was verified as an intervention for stress and burnout management for various medical staff, including nurses []. It has been suggested that ACT may have the potential to reduce burnout among various occupational groups [].
ObjectiveThe purpose of this study was to develop and optimize an AI-based tailored mobile intervention, Nurse Healing Space, to reduce nurse burnout. Specifically, we designed an AI algorithm that provides tailored burnout reduction programs using user similarity scores and assessed the usability of the intervention. After optimizing the AI algorithm, we tested the effectiveness of the program for burnout, job stress, stress response, coping strategies, and user satisfaction.
We developed the mobile app, Nurse Healing Space, from December 2021 to July 2022 and then conducted a pilot test in August 2022 to evaluate the usability and test the operation of the app. From September 2022 to February 2023, 300 people were recruited to accumulate data and optimize algorithms. Before starting the program, we instructed participants on how to participate. This was necessary to prevent dropout as the program’s duration was 4 weeks (pretest, program 1 first and second weeks, posttest 1, program 2 first and second weeks, and posttest 2), with a total of 7 stages. A total of 325 potential participants showed interest in the study, of whom, 300 completed the intervention.
Nurse Healing SpaceOverviewNurse Healing Space is a 4-week intervention that consists of two 2-week programs to reduce burnout (). Users fill out questionnaires (pretest), and the AI recommends 1 of 4 programs (program 1) based on the similarity of users. After the completion of program 1, users are tested for burnout and program satisfaction (posttest 1). The AI recommends a second program out of the 3 remaining programs (program 2) based on the similarity of users from posttest 1. After completion of program 2, users fill out another questionnaire (posttest 2).
Each program consists of 6 sessions of 10-15 minutes each. The animated female character in the video guides participants (). All the programs were developed by the research team in consultation with experts in each field. Below are the descriptions of the four 2-week programs.
Figure 1. Nurse Healing Space operation process. ACT: acceptance and commitment therapy. 
Figure 2. Scenes from the program. Mindfulness Meditation ProgramMindfulness meditation is conducted by the fictional nurse character guiding the program through narration and demonstration of movements. The contents for mindfulness meditation were organized based on Kabat-Zinn’s [] mindfulness meditation by a researcher (CC) who had previously developed mindfulness programs and reviewed them by an expert. Each session consists of an intro, mindfulness meditation, and wrap-up. The intro explains the mindset before starting the mindfulness meditation. Mindfulness meditation is composed of a mixture of the following 2 in each session: body scan, sitting up straight meditation, relaxation meditation, yoga meditation, and self-compassion meditation. The wrap-up explains ways to apply mindfulness meditation to everyday life.
Storytelling and Reflective Writing ProgramIn storytelling, a fictional nurse narrates her story of burnout and coping strategies during each session. The topics in the sessions consist of “Work is piling up,” “I am a person too,” “There aren’t enough hours in the day,” and “Am I an emotional outlet?” The work environment is poor, with not enough nurses to take care of patients. After the storytelling, the fictional nurse asks participants to reflect on their experiences and write about them. Participants can freely share their reflective writings on the bulletin board of the app.
Laughter Therapy ProgramFor laughter therapy, a fictional nurse guides the laughter exercise with background music. Each session consists of an intro, laughter exercise, and wrap-up. The contents were organized based on the laughter therapy program of the Korean Society of Laughter Clinic and were reviewed by an expert. The introductory part consists of a brief explanation of laughter therapy and prelaughter exercises. Each laughter exercise consists of 2 exercises with exciting music and the laughter follows the rhythm. The wrap-up consists of self-expression and laughter meditation.
ACT ProgramIn ACT, a fictional nurse guides the ACT by narration and showing images for cognitive change by using various techniques, such as the breathing technique and music meditation. The contents were organized based on the ACT founder’s book [] and reviewed by an expert. The ACT consists of “Hello my heart” (introduction), “Thinking is just thinking” (cognitive diffusion), “This is the moment” (being present in the moment), “Look back at me” (self as context), “Worthy life” (the best 3 moments of my life), and “Let’s do it together” (committed action and willingness).
AI-Based Program Recommendation SystemWhen users fill out the pretest, their responses are calculated and stored as scores. One of 4 programs that were effective for a previous user who is most similar to the new user, is recommended to the new user. We define similar users as having similar demographic and work-related characteristics and burnout dimensions. After the completion of program 1, users are tested for burnout dimensions and program satisfaction. Thereafter, users are matched with program 2, which is 1 of 3 remaining programs that were evaluated as effective for burnout by the most similar users based on similarity scores for demographic and work-related characteristics and burnout dimensions. After the completion of program 2, users fill out posttest 2. The operation of the Nurse Healing Space app is shown in .
Figure 3. Nurse Healing Space operation. Measuring Similarity Between UsersWe used a pilot test with 10 nurses to collect initial data using the preallocated program according to the highest subdimension recommended. With 10 sets of existing data stored through the pilot test, when new users are entered, the algorithm compares demographic and work-related characteristics, job stress, stress response, and coping strategies with the stored data to find the most similar existing users with a matching subdimension of burnout. The search method is as follows.
In the recommendation system, the existing user is denoted as i and the new user as j. Zero is assigned to the element if the value of the k variable of the user i and the k variable of the user j matches, and 1 is assigned to the inconsistent element. Specifically, the similarity of each element of the categorical variable can be determined.
(1)dijkx refers to the similarity of each element of the categorical variable; i is an existing user, j is a new user, and x is an element included in the categorical information. Thereafter, the recommendation system calculates the similarity of each categorical variable by adding the similarity of each element.
The recommendation system calculates the similarity between the users and the existing data by applying the weight of each element to the numerical variable. The numerical type is calculated by dividing the absolute value of the difference between the value of the k variable of user i and the value of the k variable of user j by the range of the k variable. Specifically, the similarity for each element in the numerical type of information is determined.
(2)indicates the similarity of each element in a numerical variable, i is an existing user, j is a new user, x is an element included in the numerical information, and R may indicate the range of the variable. Thereafter, the recommendation system calculates the similarity of the numerical variable by adding the similarity for each element.
shows that when a new user is entered, a similarity between the new user and the existing users A, B, and C is calculated. The recommendation system determines the similarity between the new user and the existing users by calculating the similarity scores from 5 categorical variables and 17 numerical variables. Similarity is based on the sum of each value. The smaller the sum of the similarity level, the higher the similarity with the new user. The possible ranges of similarity scores are between 0 and 22. For example, in , user B has the lowest similarity scores and is determined as the user that is most similar to the new user.
Table 1. Measuring similarity between users.VariableRaw scoreNew userSimilarity calculationaNot applicable.
MeasurementBurnout was measured using Copenhagen Burnout Inventory (CBI) [], which was translated into the Korean version. Cronbach α at the time of translation was 0.93 []. The CBI consists of 19 items that measure 3 subdomains: personal burnout (6 items, ranging between 0 and 31.5), work-related burnout (7 items, ranging between 0 and 37), and client-related burnout (6 items, ranging between 0 and 31.5). Possible scores range between 0 and 100, with higher scores indicating higher burnout. In this study, Cronbach α was 0.94.
Job stress was measured with an instrument developed for Korean nurses []. This instrument consists of 23 items. The scores range from 23 to 115, with a higher score indicating greater job stress. In this study, Cronbach α was 0.92. Stress response was measured using the Stress Response Inventory Modified Form [], which consists of 22 items. Possible scores ranged between 0 and 88, with higher scores indicating higher levels of stress responses. In this study, Cronbach α was 0.95.
Coping strategy was measured using the Coping strategy indicator [], which was translated into the Korean version. Cronbach α at the time of translation was 0.84 []. The instrument consists of 33 items. The possible range of scores was 33-99. In this study, Cronbach α was 0.89.
A rating scale was used to evaluate the usability of apps []. This measure consisted of 23 items in 5 areas—engagement, functionality, aesthetics, information, and subjective mobile app quality. The possible range of scores was 26-130, with higher scores indicating higher usability and quality. App satisfaction was measured with 1 item using a 5-point Likert-type rating scale (1: very dissatisfied; 5: very satisfied).
Valid DataData are evaluated as valid when burnout dimension scores after program completion decrease by at least 5 points on a 0-100 scale based on the criterion suggested in a previous study []. Having valid data implies that the program effectively lowers burnout and data are stored in the AI algorithm.
illustrates an example of the process of determining valid data and the program provided to each user with the change in burnout scores after program completion. For example, user A achieved a 5-point reduction in personal burnout after completion of program 1; this was classified as valid data and included in the algorithm. However, after the completion of program 2, the burnout scores of user A reduced for all burnout dimensions by less than 5 points; this was considered invalid and not stored in the algorithm. None of user B’s data were stored because user B did not have valid data. For user C, burnout scores for personal and work-related burnout decreased by 5 points after completion of program 1. For 2 burnout subdimensions the score reduced by more than 5 points. In this case, it was considered that burnout had decreased as a result of other causes, and the data were regarded as invalid. For user D, the personal burnout score reduced by 5 points after completion of program 1 and work-related burnout score by 6 points after completion of program 2; thus, their data were stored as valid data.
Table 2. Valid data selection criteria.aNot applicable.
AI AlgorithmAs valid data are continuously accumulated, the recommendation system learns and updates the artificial neural network. The recommendation system determines the most similar user from the user information using the learned artificial neural network. The artificial neural network determines similar users using all the similarity scores of the valid data, and programs that have been effective in reducing burnout for similar users are provided to new users. Thereafter, the artificial neural network compares the degree of change in the pre- and posttest burnout scores of the new user to determine whether the program provided was effective for the new user and stores data about the user if it is determined to be valid. The recommendation system updates the artificial neural network by continuously updating valid data. Thus, the artificial neural network can acquire valid data for more users over time, and by determining users that are similar to new users, it provides more effective programs to new users over time. To ensure that the algorithm operates as intended, we extracted valid data from 10 pilot tests and 299 optimizations. Subsequently, we manually calculated the similarity score of the 300th participant using Excel and checked the suggested program and compared it with what was suggested to the user.
Data Collection and SettingWe used 2 large web-based nursing communities with 94,000, and 170,000 members for data collection. We posted an advertisement on 2 large web-based nursing communities in August 2022 to recruit participants for a pilot test. For the pilot test, we adopted a minimum of 12 participants recommended by Whitehead et al []; however, only 10 participants were able to complete the pilot test. After the pilot test, we posted advertisements on 2 large web-based nursing communities between September 2022 and February 2023 to recruit participants for AI algorithm optimization. The number of participants required for measuring the intervention effect was calculated based on the paired 2-tailed t test for measuring the effect of burnout, a research variable of this study. Using G-power (version 3.1.9.4) with an effect size of 0.2, significance level of .05, and power of 0.95, a total of 272 people were calculated. Due to the insufficiency of previous studies on developing mobile interventions for nursing burnout, the effect size was conservatively considered at a low threshold to achieve treatment effects. A total of 325 potential participants showed interest in the study, of whom, 300 completed the study. Their data were used for AI algorithm optimization. The inclusion criteria for the participants were nurses who had been providing care independently at hospitals for the past month at the time of data collection.
Ethical ConsiderationsThis study was reviewed and approved by the institutional review board (ewha-202208-0005-01) at the principal investigator’s institution. Informed consent was obtained from all participants. Participants received information about the study’s goals, anonymity, confidentiality, and their right to withdraw from the study at any time before signing the online informed consent form. As compensation for their participation, participants were provided with mobile coupons valued at US $3.60. The collected data were encrypted and stored on a computer accessible only to the researchers.
Statistical AnalysisWe used software SPSS (version 23.0; IBM Corp) to facilitate data analysis. Descriptive statistics were used to report sociodemographic and work-related characteristics and study variables. All variables met the assumption for normal distribution: skewness of 1 or less and kurtosis of 5 or less []. A paired t test was conducted on burnout, job stress, stress response, and coping strategies before and after the program. Spearman correlation was performed to examine changes in research variables and burnout reduction based on data accumulation. Nonparametric analysis (Spearman correlation) was used because data accumulation is considered an ordinal variable in that it follows the order of participation. ANOVA was conducted on satisfaction change by grouping participants into 3 groups: the first 100 participants, the second 100 participants, and the third 100 participants; the Scheffe test was performed for post hoc analysis.
The 300th participant had the highest work-related burnout and client-related burnout among the burnout dimensions, and laughter therapy was recommended. The similarity scores between the 300th participant and all the accumulated valid data ranged between 1.17 and 8.14. It was confirmed that the program completed by the participant with the lowest similarity was laughter therapy, which was recommended for the 300th participant. The highest burnout subtype for the most similar participant was work-related burnout, and the participant’s data were classified as valid data because of the effect of laughter therapy. The calculated similarity with the participant was 1.17. To make it more indisputable, we checked whether the similarity system worked in the same way for the 299th participant, and it was confirmed that the program recommended to the most similar participant among the valid data was the same.
Valid DataOf the total data, 45.7% (274/600) were selected as valid data. Personal burnout was the highest subdimension with 57.4% of valid data and work-related burnout was the next highest subdimension with 25.9% of valid data, followed by client burnout with 40% of valid data including duplicates. In terms of programs, laughter therapy accounted for the largest proportion of valid data, with 32.8%, followed by 28.1% for mindfulness meditation, 20.4% for ACT, and 18.6% for storytelling and reflective writing.
Usability AssessmentUsers scored ease of use, good visual design, and high quality of content graphics more than 4 out of 5. However, users felt most uncomfortable with the absence of reminders and notification functions and the inability to interact with the app. Moreover, they were not satisfied with the fact that the app could not be set up to suit their preferences. The overall score for this app was 3.4 out of 5 on a Likert scale. Furthermore, the total mean score for each question was 3.4 out of 5 on a Likert scale.
Intervention EffectsThe mean age of the participants was 32.8 (SD 5.75) years. Of these, 97.7% (293/300) were female, 87% (261/300) were staff nurses, and 62.3% (187/300) worked in hospitals with more than 500 beds ().
The decrease in burnout was statistically significant, from 62.36 (SD 18.0) at pretest to 55.22 (SD 18.79) at posttest 1 (t=7.012; P<.001) and 52.70 (SD 20.15) at posttest 2 (t=8.692; P<.001). Reduction in burnout scores from posttest 1 to posttest 2 was also statistically significant (t=2.811; P=.01). Moreover, burnout scores decreased significantly before the program and after the first and second programs (t=8.692; P<.001). Personal burnout decreased statistically from 20.40 (SD 6.39) at pretest to 18.15 (SD 6.38) at posttest 1 (t=6.134; P<.001) and 17.30 (SD 7.33) at posttest 2 (t=7.535; P<.001). In addition, the reduction in personal burnout from posttest 1 to posttest 2 was statistically significant (t=2.597; P=.01). Work-related burnout decreased statistically from 22.22 (SD 6.39) at pretest to 19.83 (SD 6.49) at posttest 1 (t=6.816; P<.001) and 19.06 (SD 6.68) at posttest 2 (t=8.286; P<.001). The reduction in work-related burnout from posttest 1 to posttest 2 was also statistically significant (t=2.306; P=.02). Client-related burnout decreased statistically from 19.72 (SD 6.63) at pretest to 17.22 (SD 6.91) at posttest 1 (t=6.439; P<.001) and 16.32 (SD 7.15) at posttest 2 (t=8.329; P<.001). Reduction in client-related burnout from posttest 1 to posttest 2 was also statistically significant (t=2.704; P=.01).
Job stress, stress response, and coping strategy were measured twice before the program and after the first and second programs. After the completion of the two 2-week programs, the participants reported a statistically significant reduction in job stress (t=6.765; P<.001) and stress response (t=5.864; P<.001) ().
Table 3. General characteristics of participants (N=300).CharacteristicValuesAge (years), mean (SD)32.81 (5.754)Gender (female), n (%)293 (97.7)Marital status, n (%)aICU: intensive care unit.
bSpecial units include emergency rooms, operating rooms, recovery rooms, and delivery rooms.
Table 4. Effectiveness of intervention by variables (N=300).VariablesPretest score, mean (SD)Posttest 1 score, mean (SD)Posttest 2 score, mean (SD)Pretest vs posttest 1Posttest 1 vs posttest 2Pretest vs posttest 2aBurnout total score range is between 0 and 100.
bPersonal burnout score range is between 0 and 31.5.
cWork-related burnout score range is between 0 and 37.
dClient-related burnout score range is between 0 and 31.5.
eJob stress score range is between 23 and 115.
fNot applicable.
gStress response score range is between 0 and 88.
hCoping strategy score range is between 33 and 99.
For program 1, the most frequently recommended program was laughter therapy (46.3%), followed by storytelling and reflective writing (20%), ACT (18.3%), and mindfulness meditation (15.3%). For program 2, the most frequently recommended program was mindfulness meditation (41.3%), followed by laughter therapy (23.7%), storytelling and reflective writing (21.3%), and ACT (13.7%). The effect of each program in reducing burnout was as follows: laughter therapy (8.7%), storytelling and reflective writing (8.7%), ACT (12.3%), and mindfulness meditation (4.5%). Thus, among the 4 programs, ACT reduced burnout the most. Client-related burnout was decreased the most by ACT (18.2% decrease). Laughter therapy reduced personal burnout by 10.4%, which was the highest reduction. Work-related burnout was reduced the most through storytelling and reflective writing and ACT (both resulted in a reduction of 10.5%) ().
shows the correlation with the order of participation and burnout reduction. The correlation of burnout differences between posttest 1 and pretest and the order of participation was 0.060 (P=.30). The correlation of burnout differences between posttest 2 and posttest 1 and the order of participation was 0.138 (P=.04). The correlation of job stress differences between pretest and posttest 2 and the order of participation was 0.134 (P=.03). The correlation of stress response differences between pretest and posttest 2 and the order of participation was 0.094 (P=.13). The correlation of coping strategy differences between pretest and posttest 2 and the order of participation was 0.140 (P=.08).
Table 5. Details for each program (N=300).aACT: acceptance and commitment therapy.
Table 6. Correlation with the order of participation and research variables.StagesOrder of participationBurnout differences between posttest 1 and pretestWhen programs were compared, the satisfaction score for storytelling and reflective writing (4.69 points) was the highest, followed by mindfulness meditation (4.61 points), laughter therapy (4.58 points), and ACT (4.52 points) (). The satisfaction scores increased over time from participant 1 in October 2022 to participant 300 in January 2023 by optimization (). By grouping participants into groups of 100, we were able to calculate the average satisfaction of programs 1 and 2 to understand how satisfaction changed. For program 1, satisfaction scores for the 3 groups increased from 4.46 to 4.60 and 4.69 (F=3.493; P=.03). The post hoc test showed statistically significant differences between the first 100 and the third 100 (P=.03). For program 2, the satisfaction score was 4.49 points for the initial 100 participants, it was 4.66 points for the next 100 participants, and the score was 4.70 for the last 100 participants; thus, the satisfaction for program 2 also gradually increased (F=3.911; P=.02). In addition, the difference in satisfaction with the second program was statistically significant as a result of the post hoc analysis between the first 100 and the third 100 participants (P=.03).
Figure 4. Changes in participant satisfaction. Group 1 included the first to the 100th participants, group 2 the 101st to the 200th, and group 3 the 201st to the 300th. In this study, we developed an AI-based tailored mobile app and optimized the algorithm for nurse burnout. As a result of organizing and optimizing the system with an algorithm that recommends programs that have been effective for highly similar users, burnout decreased and satisfaction scores increased as data accumulated. Furthermore, when we examined whether the algorithm was being applied correctly, we verified that the system recommends the most si
留言 (0)