Psychological problems and burnout among healthcare workers: Impact of non-pharmacological lifestyle interventions

Abstract

Objective

To evaluate role of rajyoga meditation (RYM) versus stress management counselling (SMC) in addressing burnout syndrome and resultant improvement in electrocardiogram (ECG) so as to automate burnout prediction from raw ECG data with machine learning (ML).

Methods

Healthcare providers were assigned to two groups: RYM (n = 100) or SMC (n = 102). Subjects in RYM received rajyoga for 3 months including one week offline and thereafter, virtual mode. SMC group received counselling for 1 day in offline mode and thereafter, received positive thoughts on a weekly basis. All subjects were assessed for psychological (depression, anxiety, stress scale-21 (DASS-21) and burnout syndrome (Mini Z questionnaire) along with 12-lead ECG at baseline after 4 weeks, and after 12 weeks. Based on response on question 3 of the Mini-Z questionnaire, participants were classified either as burnout or satisfied.

Results

RYM group showed significant reduction in depression, anxiety, and stress in comparison to SMC group. Burnout results display significant reduction in the RYM group in comparison to SMC group. Reduction in burnout and enhancement in satisfaction from visit-1 to visit-3: burnout visit-1 (27.2 %), visit-2 (23.8 %), visit-3 (19.3 %) and, satisfaction visit-1 (72.8 %), visit-2 (76.2 %), and visit-3 (80.7 %). ML algorithms could identify burnout patients using the raw ECG data with time-series features based classifier performing better than Ultra Short HRV features based ML classifier model.

Conclusion

AI based early diagnosis of heart's healthy status using ECG analysis may prevent development of cardiovascular disorder in the long run.

1. Introduction

Burnout in healthcare workers (HCWs) has been increasingly recognized as a critical factor affecting well-being of healthcare professionals and the quality of care they provide.¹ HCWs are most vulnerable to develop psychological stress and other mental health symptoms. Burnout has been considered as an outcome of long-term exposure to occupational stress.^2,3 Data suggest that psychological issues and burnout among medical professionals are often associated with more medical errors and poor patient outcomes.³

In the current era, meditation, yogasana and pranayama are considered to be helpful in managing psychosocial issues. Breathing techniques such as yogasana and pranayama support in reducing palpitations, regulation of breathing pattern, and releasing the toxins to improve body fitness.⁴ Rajyoga meditation (RYM) is a mediation technique shown to have clinical benefit in conditions such as coronary artery disease, tension headache, depression and anxiety in patients undergoing coronary artery bypass surgery, substance abuse, and diabetes mellitus.^5,6 Early detection of stress and burnout can have significant impact on mental health of HCWs. Since stress and burnout has an impact on cardiac and autonomic system, Ultra Short heart rate variability (HRV) serve as an important marker of stress levels.⁷

The present study assessed psychological problems like burn out in HCWs and evaluated impact of RYM versus stress management counselling (SMC) among HCWs in India. The study also aimed to analyze 12-lead ECG data using artificial intelligence-based ML algorithms to evaluate impact of RYM versus SMC on heart rate variability in HCWs.

2. Methods

This was a prospective observational study at a tertiary care medical center in North India where 203 consecutive healthcare providers ≥18 years of age including residents, physicians, nurses and paramedical staff (n = 202, of which 100 in RYM and 102 in SMC) were recruited after obtaining written informed consent. Out of these, 100 were in RYM group and 102 were in SMC group. Healthcare providers with history of or currently suffering with neurological or psychiatric impairment, or cognitive dysfunction or suffering with severe comorbidity such as uncontrolled hypertension, uncontrolled diabetes mellitus, myocardial infarction or heart failure within 30 days of recruitment or subjects already practicing meditation or any other relaxation practices were excluded. The study was approved by the IEC (F.1/IEC/MAMC/(84/02/2021/No 389).

2.1. Assessment

Complete demographic profile of all participants were noted. Psychological assessments were performed using Depression, anxiety, stress scale-21 (DASS-21)⁸ and the Mini Z questionnaire⁹ by trained psychologist blinded to the recruitment process and statistical analysis. DASS-21 has been shown to distinguish between symptoms of depression, anxiety and stress in clinical and non-clinical samples.⁸ DASS-21 comprises of three scales: (1) depression scale: measures hopelessness, low self-esteem, and low positive affection; (2) anxiety scale: evaluates autonomic arousal, musculoskeletal symptoms, situational anxiety and subjective experience of anxious arousal; and (3) stress scale: measures tension, agitation, and negative affection. Each item's score ranges from zero to three with the zero score stands for the statement “did not apply to me at all”, one score implies “applied to me to some degree, or some of the time”, two score implies “applied to me to a considerable degree or a good part of time”, and three score implies “applied to me very much or most of the time”.⁸ The Mini-Z instrument was developed as a tool for efficiently measuring physician burnout.⁹ On the Mini-Z questionnaire, the participant's response to question 3 indicates if the person is feeling burnout or not. All subjects were assessed at the beginning, after 4 weeks and after 12 weeks (total 3 visits). In each of these visits, all the enrolled patients underwent 12-lead ECG as well as psychological assessments using DASS-21 and Mini Z questionnaire.

2.2. Intervention

HCWs in RYM group were taught a structured module of raj yoga for three months by qualified raj yoga teacher in a group of five people, one-week offline mode, thereafter, online everyday up to 30 min. In order to ensure compliance among study subjects, social media group was created and regular alerts were posted twice a day and telephonic calls were made when needed. In SMC group, subjects were introduced to a dedicated trainer. The trainer counselled and taught seven steps for managing stress one day in the offline mode for 30 min, thereafter received positive affirmations every day for a week and then once a week for three months. SMC includes seven steps: (i) Reprogramming Thoughts: make your mind your best friend; (ii) Changing Perspective: life is all about perspective; (iii) Positive Outlook: choose goodness in everyone; (iv) practice digital detox; (v) Embracing Gratitude and Peace: (vi) energize your food with vibrations of God's love, peace, purity and gratitude for him; and (vii) sleep hygiene. All the subjects were given message reminders and video/audio calls every week to ensure proper practice and compliance.

2.3. Sample size calculation

Based on meta-analysis by Brown and colleagues,¹⁰ taking the minimum effect size of Hedge's(g) = 0.25 between the two groups (RYM vs. counselling) in analysis of covariance with a significance level of 0.05, statistical power = 0.80, a total of 128 participants were needed (64 per group). Adding a follow-up loss of 20 %, 160 participants were needed (80 per group) for this study.

2.4. Data for AI workflow

A 12-lead ECG at 500Hz of 10s duration was recorded for all subject using the Vesta 301i machine (Recorders and Medicare Systems (P) Ltd.) and the ECG data was stored in a raw format. In the ML model, the burnout group is represented by class label 1, while the rest (‘not burnout’) are labelled by class label 0. Raw ECG data is generally corrupted by noise that makes the feature extraction and classification less accurate. To mitigate this issue, a fifth-order high pass digital Butterworth filter of critical frequency 300 Hz was designed using Python Scipy library to remove the noise. Two types of features were extracted for each patient: Ultra Short HRV and Time Series. Python sklearn library was used for developing the ML models. The NeuroKit and the Python hrv-analysis tool were used to extract the Ultra Short HRV features in time and frequency domain. A total of 30 Ultra Short HRV features were extracted for each lead of a patient. For the time series features, python package tsfresh was utilized for extracting time and frequency features. All the extracted features were scaled to z-scores with mean to zero and variance to one. Overall, a total of 30 top features for the complete 12-channel ECG data were selected from each of the Ultra Short HRV and the time-series based features using the Scikit-learn's SelectKBest function that uses the ANOVA F-value between the feature and label.

Feature Selection: Overall, a total of 30 top features for the complete 12-channel ECG data were selected from each of the Ultra Short HRV and the time-series based features using the Scikit-learn's SelectKBest function.⁴ This algorithm identifies the highest-performing features based on the given number of features (k) and a designated scoring function. Specifically, the ‘f_classif’ scoring function was employed that calculates the ANOVA F-value between the feature and label. Multiple values of k were tested in conjunction with the classifier and finally top 30 features were utilized to generate the final results.

2.5. Design of ML classifier

Random Forest (RF) classifier was built using both types of features. Visit-3 data was used to build the classifiers and also validated on the visit-1 and visit-2 dataset. A 70-30 train-test split was performed on the visit-3 data. Prior to training the classifier, an up sampling of the minority class (burnout) was performed in the training set using the Synthetic Minority Oversampling Technique (SMOTE) method because there was a class imbalance of approximately 1:5. Subsequently, the performance of the trained model was evaluated separately on the test set and the visit-1 & visit-2 dataset of both the groups.

2.6. Assessment of heart rate variability using statistical analysis on ML results

The likelihood of belonging to the ‘not burnout’ class was computed for every sample and every visit. This was computed as the probability of belonging to the ‘not burnout’ class as predicted by the trained random forest classifier. The Wilcoxon Signed Rank test and the Wilcoxon Unsigned Rank test were conducted to evaluate the statistical significance of the changes in probability as visits progressed within each RMY and SMC groups. The analysis was done to check the hypothesis that Rajyoga meditation elicits relatively better positive change in heart health as compared to the traditional stress management counselling.

2.7. Statistical analysis and evaluation metrics

Continuous data were tested for normality using the Shapiro–Wilk test. For baseline comparison between groups, t-test for two independent sample mean was used for continuous variables and chi-square test was used for categorical variables. Analysis of repeated measures (RMANOVA) was performed with baseline assessment as covariate and intervention group as between factors for comparing the post assessment. Data were analysed using the Jamovi software (version 2.3). A number of evaluation metrics were used to evaluate the performance of the classifiers. In this work, burnout was the positive class or class 1 and ‘not burnout’ was the negative class or class 0. When a model correctly predicts positive and negative classes, the outcome is a true positive (TP) and true negative (TN), respectively.

A detailed discussion on the AI workflow has been highlighted in the Supplementary appendix S1.

3. Results

A total of 250 participants were screened of whom 202 were enrolled with 100 subjects being in RYM group, while 102 subjects were randomised to SMC group. Baseline demographic details have been compared in Table 1. Practice of RYM for 12 weeks, resulted in significant decrease in DASS-21 score (−45.00 (Baseline); - 24.2 (at 12 week, p < .001) and the burnout in question no. 3 of Mini-Z questionnaire (1.87 (baseline) vs 1.34 (at 12 weeks); p < .001). Data regarding the class distribution in terms of burnout or satisfied has been highlighted in Fig. 1. There was no significant decrease observed in the DASS-21 and Burnout scores in the SMC group at 12 weeks of follow-up (Supplementary Table 1 and Fig. 2). No correlation could be established between DASS-21 and burnout scores with gender, age, marital status, education and family type between two groups (Supplementary Table 2).

Table 1.

Baseline comparison data between groups.

Variables	RYM (n = 100)	SMC (n = 102)	P
Age in years (mean ± SD)	38.0 ± 11.0	31.8 ± 10.8	<0.001
Sex
Male	28 (28 %)	34 (33.3 %)	0.411
Marital status
Married	57 (57 %)	49 (48 %)	0.202
Single	43 (43 %)	53 (52 %)
Family structure
Nuclear	70 (70 %)	64 (62.7 %)	0.275
Joint	30 (30 %)	38 (37.3 %)
Highest education
Graduate	57 (57 %)	46 (45.1 %)	0.245
Post Graduate	43 (43 %)	56 (54.9 %)
DASS-21 (mean ± SD)	45.0 ± 21.9	36.4 ± 22.6	0.002
Burnout (mean ± SD)	1.87 ± 0.925	1.94 ± 1.042	0.002

Abbreviations: RYM: rajyoga meditation; SMC: stress management counselling; DASS-21: Depression, anxiety, stress scale-21.

Fig. 1.

(a) Class distribution in visit 1, (b) Class distribution in visit 2, (c) Class distribution in visit 3, as per mini z burnout score.

Fig. 2.

(a) Pre-post intervention data for DASS-21; (b) pre-post intervention data for Burnout.

For ECG data, performance of the RF classifiers 1) built using Ultra Short HRV features [Classifier-1] and 2) built using time-series analysis features (Classifier-2) were compared. The evaluation metrics were calculated separately for the 30 % test data of Visit-3 and for Visit-1 & Visit-2 data combined. The accuracy, F1 score, precision and recall of both the classifier models has been represented in Supplementary Table 3. The time-series features based classifier performed better than the Ultra Short HRV features based ML classifier model. The performance of classifier-2 was very good on all metrics including precision, recall, F1-score and accuracy on the test data of Visit-3 compared to that of Visit-1 & Visit-2 combined. The statistics (mean, median, and standard deviation) of the probabilities of belonging to the ‘not burnout’ class in the three visits of both the RMY and SMC groups as predicted by the trained RF classifiers has been highlighted in Table 2. Results showed that the probability of being ‘not burnout’ significantly improved in the mean and median from visit-1 to visit-3 for the RYM group and it declined for the SMC group with both the classifiers. Improvement in burnout status indicated a healthier cardiac status of subjects in RYM group despite work pressure. This was not observed in the SMC group.

Table 2.

Statistics computed on the probabilities of being ‘not burnout’ as predicted by Classifier-1 and Classifier – 2 respectively (using Ultra Short HRV features).

Classifier - 1
	RYM Group (Treatment)			SMC Group (Control)
	Visit 1	Visit 2	Visit 3	Visit 1	Visit 2	Visit 3
Mean	0.56	0.61	0.68	0.59	0.54	0.51
Median	0.56	0.59	0.71	0.58	0.52	0.48
Standard Deviation	0.17	0.19	0.2	0.18	0.16	0.22
	RYM Group (Treatment)			SMC Group (Control)
	Visit 1 - Visit 2	Visit 2 - Visit 3	Visit 1 –Visit 3	Visit 1 –Visit 2	Visit 2 –Visit 3	Visit 1 – Visit 3
Classifier-1 (using HRV features)b	0.03a	0.016a	<0.0001a	0.97	0.86	0.99
Classifier-1 (using HRV features)c	0.06	0.03a	<0.0001a	0.04a	0.28	0.001a

Classifier - 2
	RYM Group (Treatment)			SMC Group (Control)
	Visit 1	Visit 2	Visit 3	Visit 1	Visit 2	Visit 3
Mean	0.61	0.63	0.7	0.63	0.6	0.5
Median	0.61	0.64	0.74	0.65	0.62	0.54
Standard Deviation	0.14	0.14	0.2	0.13	0.14	0.23
	RYM Group (Treatment)			SMC Group (Control)
	Visit 1 - Visit 2	Visit 2 - Visit 3	Visit 1 - Visit 3	Visit 1 - Visit 2	Visit 2 - Visit 3	Visit 1 – Visit 3
Classifier-2 (using time-series based features)b	0.14	0.0018a	<0.0001a	0.93	0.99	0.99
Classifier-2 (using time-series based features)$	0.29	0.003a	<0.0001a	0.13	0.0002a	<0.0001a

^aStatistically significant result at a p-value <0.05.

^bWilcoxon signed rank test.

^cWilcoxon unsigned rank test.

4. Discussion

In the present study, we compared the impact of RYM and stress management counselling in reducing depression, anxiety, stress and burnout among healthcare workers. The interesting findings of present study were 1) HCWs practicing RYM are likely to have lesser scores on DASS 21 than those in the SMC arm and 2) RYM had significant impact as compared to SMC in reduction of burnout among healthcare workers. Using ML models, we were able to categorize participants as either burnout or ‘not burnout’ based on 12-lead ECG data analysis. Using the trained classifier, cardiac health status was predicted on a scale of 0–1 via measuring the probability predicted by the trained RF classifier for belonging to the class of being ‘not burnout’. Results indicated that the cardiac status of the RYM group improved significantly, while such an improvement was not observed in the SMC group. These important findings suggest role of RYM as one of the potential tools in addressing the mental health status of HCWs.

The challenges faced by the HCWs while providing quality healthcare to patients is unsurmountable. The unpredictable traumatic challenges include constant fear of getting infection, transmitting infection, feeling of guilt and often losing self-confidence. Loneliness in first line HCW frequently led to depression. In addition, stress perceived and experienced by HCWs during patient care, difficulty in coping with life threatening situations, long and tiring work hours and adverse outcomes frequently led to emotional exhaustion and burnout.¹¹ In a previous study among HCWs, overall prevalence of burnout was 16.2 %. Burnout rates were shown to be higher among second-line workers 20.5 % than frontline 14.9 %.⁷ Practicing relaxation techniques are well-known interventions to enhance mental well-being. Yoga has previously been shown to be an important modality for dealing with stress.¹² RYM can possibly help the HCWs in achieving a stable psychological state that would enable them to maintain physical and emotional balance within their work environment. Li and colleagues in a randomized trial among staff nurses reported that regular yoga led to a marked reduction in work stress with improved health outcomes.¹³ RYM, a specialized form of yoga, has been shown to reduce stress, anxiety and depression among healthy adults. A comparative evaluation of the DASS-21 score among RYM (n = 802) and non-meditators (n = 357) reported marked lower scores in the RYM group.¹⁴ The present study also showed marked reduction in the DASS-21 scores in RYM group as compared to the SMC arm.

Results for psychological constructs among HCWs in our study indicated significant reduction in burnout in RYM group in comparison to SMC group. The analysis of survey data of 202 participants reported reduction in burnout and enhancement in satisfaction from first visit at baseline to third follow-up visit at 12 weeks. Similar findings were reported in a study among 40 nurses where an eight-week yoga intervention was shown to reduce burnout.¹⁵ In another study utilizing mindfulness-based yoga practice among healthcare professionals, significant improvement was noted in terms of perceived stress, burnout, vitality and sleep quality as compared to the control population.¹⁶ In our study, we used ML based algorithms that identified burnout patients using the raw ECG data. Previous study using ML model to predict burnout among HCWs reported that tree based extra tree classifier had the highest sensitivity of 84 %, AUC Score of 84 % and an accuracy of 77 %.⁷ In our study, RF based ML model reported the highest accuracy, precision, recall, and F-1 score to be 90 %, 100 %, 67 %, and 80 % respectively. Present study is unique as it uses classifier to make a discrete decision of change of state from ‘burnout’ (class 1) to ‘not burnout’ (class 0). This study therefore introduces the concept of measuring the cardiac status on a scale of 0–1 instead of the traditional ML classifier metrics. Thus, measuring the probability of being ‘not burnout’ as predicted by the classifiers would be a better metric to track over a period of time over multiple visits while a therapy is in progress.

The study has several limitations. It is a single centre, open label study with limited sample size. The effect of RYM has been demonstrated over a short follow up of 3 months only. Longer duration follow is necessary to see the effect over long term. Further, practice of RYM and SMC at home was ensured but compliance and individual learning effort is different for all subjects. The findings of this study needs to validated in larger multicentre studies in view of the observational study design and the small sample size.

5. Conclusion

In conclusion, integrating non-pharmacological lifestyle interventions such as yoga into healthcare workers’ routine can significantly improve their mental health outcomes, reduce burnout, and increase their capacity to continue providing high-quality patient care. Institutions should prioritize these interventions to promote sustainable mental health solutions within the healthcare workforce.

Funding

This study was funded by Delhi Science and Technology vide grant no: DST/SATYAM/COVID19/2020/328(G).

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: