The “Healthcare Workers’ Wellbeing [Benessere Operatori]” Project: A Longitudinal Evaluation of Psychological Responses of Italian Healthcare Workers during the COVID-19 Pandemic

Background: COVID-19 forced healthcare workers to work in unprecedented and critical circumstances, exacerbating already-problematic and stressful working conditions. The “Healthcare workers’ wellbeing (Benessere Operatori)” project aimed at identifying psychological and personal factors, influencing individuals’ responses to the COVID-19 pandemic. Methods: 291 healthcare workers took part in the project by answering an online questionnaire twice (after the first wave of COVID-19 and during the second wave) and completing questions on socio-demographic and work-related information, the Depression Anxiety Stress Scale-21, the Insomnia Severity Index, the Impact of Event Scale-Revised, the State-Trait Anger Expression Inventory-2, the Maslach Burnout Inventory, the Multidimensional Scale of Perceived Social Support, and the Brief Cope. Results: Higher levels of worry, worse working conditions, a previous history of psychiatric illness, being a nurse, older age, and avoidant and emotion-focused coping strategies seem to be risk factors for healthcare workers’ mental health. High levels of perceived social support, the attendance of emergency training, and problem-focused coping strategies play a protective role. Conclusions: An innovative, and more flexible, data mining statistical approach (i.e., a regression trees approach for repeated measures data) allowed us to identify risk factors and derive classification rules that could be helpful to implement targeted interventions for healthcare workers.


Introduction
Healthcare settings may represent a challenging workplace, characterized by long and undefined working hours, excessive workloads, competitiveness of training, high responsibility, and constant exposure to suffering, illness, and mourning [1,2]. However, lack of time, stigma, and concerns around confidentiality may prevent seeking psychological support [2,3].
Consistently, increasing evidence shows that healthcare workers around the world report high levels of depression, anxiety, stress, burnout, and post-traumatic stress disorder (PTSD) [1,4]. However, the literature is limited, and the samples analyzed are heterogeneous, resulting in a prevalence of psychiatric symptoms ranging from 30% to 60% for physicians [5][6][7] and from 11% to 73% for nurses [8].

Participants and Procedure
The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Ethics Committee of the University of Milano-Bicocca (protocol n. 0024531/20), the Ethics Committee of the IRCCS San Raffaele Scientific Institute (protocol n. 109/2020), and the Ethics Committee of the Parma Local Health Authority (protocol n. PG0019826_2020).
This study is part of a web-based longitudinal project to examine the psychological impact of COVID-19 on a sample of Italian healthcare workers involved in the management of the pandemic. After reading the informed consent, participants voluntarily completed an online survey, administered through Qualtrics and sent to the e-mail address provided, during the baseline assessment. We assessed participants' working conditions, individual perception of the COVID-19 situation, anxiety, depression, and insomnia symptoms, posttraumatic stress, state anger, and burnout levels. In the present study, we will also analyze the coping strategies and perceived social support measured during the baseline study.
In total, 344 healthcare workers participated in the second survey, ten of whom stated that they had not worked in the previous three months and were thus excluded from the analysis. Finally, statistical analyses have been carried out on a sample of 291 respondents with complete records on both demographic and psychological variables.

Measures
A self-report questionnaire was used to collect socio-demographic and work-related information from participants, including their age, gender, psychological/psychiatric history, ward in which they worked, and whether they had received emergency training.
The Depression Anxiety Stress Scale (DASS-21) [23,24] is a 21-item scale that assesses general distress using a tripartite model of psychopathology. This questionnaire includes three subscales: depression, anxiety, and stress. Each item is rated on a four-level Likert scale (0 = never; 3 = almost always). The total score is calculated by adding together the response values for each item. Higher scores suggest severe levels of depressive, anxiety, and stress symptoms. The original version of the questionnaire showed an internal reliability with Cronbach's alpha coefficient of 0.91 for the depression scale, 0.84 for the anxiety scale, and 0.90 for the stress scale [24]. The total score of the Italian version reported a Cronbach's alpha value of 0.90, with subscale values ranging from 0.74 to 0.85 [23].
The Insomnia Severity Index (ISI) [25,26] is a self-report questionnaire that assesses the nature, severity, and impact of insomnia using seven items rated on a five-level Likert scale (0 = "no problem"; 4 = "very severe problem"), with scores ranging from 0 to 28. The dimensions evaluated are severity of sleep onset, sleep maintenance, early morning awakening problems, sleep dissatisfaction, interference of sleep difficulties with daytime functioning, noticeability of sleep problems by others, and distress caused by the sleep difficulties. The original version of the ISI reported a Cronbach's alpha coefficient of 0.74 [26]. The Italian version showed a Cronbach's alpha coefficient of 0.75 [25].
The Impact of Event Scale-Revised (IES-R) [27,28] is a 22-item self-report questionnaire for evaluating the frequency of intrusive and avoidant thoughts and behaviors associated with a traumatic event. Items are rated on a five-point Likert scale (0 = "not at all"; 4 = "extremely"). The IES-R is divided into three subscales. Intrusion (8 items) assesses intrusive thoughts, nightmares, intrusive feelings, and imagery related to the traumatic event; Avoidance (8 items) evaluates avoidance of feelings, situations, and ideas; Hyperarousal (6 items) measures difficulty in concentrating, anger and irritability, psychophysiological arousal in response to reminders, and hypervigilance. The original version showed high levels of internal consistency (Intrusion: α = 0.87-0.94, Avoidance: α = 0.84-0.87, Hyperarousal: α = 0.79-0.91) [28]. The Italian version shows good (0.84) and acceptable (0.71) internal consistency for the intrusion subscale and the avoidance subscale, respectively [27].
The State-Trait Anger Expression Inventory-2 (STAXI-2) [29] is a 57-item self-report questionnaire that measures five domains of anger: State-Anger, Trait-Anger, Anger Expression-In, Anger Expression-Out, and Anger-Control. Responses are rated on a fourpoint Likert scale, ranging from 1 (not at all) to 4 (almost always). Cronbach's α coefficients range from 0.73 to 0.76, indicating high internal reliability for all the subscales except for the Trait Anger Scale/Angry Reaction [29]. In the present study, we only used the State-Anger subscale to assess healthcare workers' acute reaction to the pandemic.
The Brief Cope [32,33] is a 28-item questionnaire, divided into 14 subscales, measuring coping responses. Each item is rated on a four-level Likert scale (0 = "I have not been doing this at all"; 3 = "I have been doing this a lot"). Coping strategies can be grouped into problem-focused (strategies aimed at changing a stressful situation: active coping, use of instrumental support, positive reframing, and planning), emotion-focused (strategies to regulate emotions associated with a stressful situation: use of emotional support, venting, humor, acceptance, self-blame, religion), and avoidance coping strategies (physical or cognitive efforts to disengage from the stressor: self-distraction, denial, substance use, behavioral disengagement) [34,35]. The original version of the questionnaire showed Cronbach's alpha coefficients, ranging from 0.50 to 0.90 [32], while the Italian version of the questionnaire revealed omega coefficients for reliability, ranging from 0.439 to 0.959 [33].
The Multidimensional Scale of Perceived Social Support (MSPSS) [36,37] is a 12-item self-administered questionnaire evaluating social support perceived by family, friends, and significant others, rated on a 7-point Likert scale (1 = "very strongly disagree"; 7 = "very strongly agree"). Higher scores indicate higher perceived social support. The internal reliability of the questionnaire is good, with Cronbach's alpha coefficients ranging from 0.85 to 0.91 [38]. The Italian version shows good indices of reliability with Cronbach's alpha coefficients ranging from 0.81 to 0.98 [36].
Furthermore, we assessed how worried participants were about the possibility that themselves, their relatives, their friends, and their colleagues could contract COVID-19. Four items were rated on a five-point Likert scale (1 = "not at all"; 5 = "extremely"). A total score of worry was obtained by averaging items scores.
Finally, we evaluated participants' working conditions over the previous three months in several areas, including eating, sleeping, working shifts, isolation, and wearing appropriate protective equipment. Seven items were rated on a five-point Likert scale (1 = "not at all"; 5 = "very much"). A total score of working conditions was obtained by averaging item scores. Higher scores indicate worse working conditions.

Statistical Analysis
Median and interquartile range (IQR) were used as summary statistics to describe continuous variables, while categorical variables were expressed as frequencies and percentages. Radar plots were used to visualize differences between measurements collected during the two evaluations in the different healthcare worker categories.
Linear mixed-effects (LME) models [39] were applied to evaluate the changes in the psychological outcomes over time while accounting for respondent-specific heterogeneity through random effects specification. The variables included in the models were: time (categorical with two levels, T0 and T1, respectively, at baseline and during the second wave), gender, occupation, working or having worked in COVID-19 wards (time dependent variable), worry scores and the evaluation of working conditions, the presence of psychological or psychiatric symptoms in the past, having attended emergency training, perceived social support as measured by the MSPSS, and the three Brief COPE subscales. To highlight specific differences in the outcome variables over time for the different healthcare workers categories, we also entered in the model the interaction between occupation and time.
Standard transformations (square root, power, ordered quantile normalization) were applied to outcome variables to satisfy model regression assumptions.
To examine psychological measure dynamics over time, within a data mining framework, an extension of regression trees methodology accounting for correlation structure among observations was considered. This approach is suited for studies with repeated measures and longitudinal data. The tree-based estimation method, proposed and implemented in the R RE-EM tree package by Sela and Simonoff [40], was considered.
A regression tree [41] implements a binary recursive partitioning in which predictor variables that best discriminate among response profiles, along with the optimal cut-points, are automatically chosen. The splitting criterion is based on maximizing the reduction in the sum of squares, for the node, until convergence is reached. The approach is very flexible in handling missing values and both quantitative and qualitative predictors.
Following the branches of the resulting tree, hence considering rules derived from the variables' best cut-off values, it is possible to derive different patterns of longitudinal responses. Regression trees provide an effective tool for supporting decision-making in clinical practice and identifying relevant variables associated with different outcomes, while uncovering complex relationships among predictors. Trees were estimated using the same covariates included in the mixed models. All the analyses were performed using R statistical software (version 4.1.1, https://cran.r-project.org/index.html, accessed on 10 October 2021). The significance level was set at 0.05.

Results
Participants' characteristics are shown in Table 1 Among the sample, 23% reported having a psychological/psychiatric history, and only 16.8% reported having undergone emergency training.
Concerning their occupation, 31.3% of the participants were physicians, 33.3% were nurses, 27.9% were other healthcare workers, and 7.6% were clerks. At T0, 33.3% of the participants worked in a COVID-19 ward, and 15.5% worked in a COVID-19 ward at T1. Table 1 Table 2 shows the characteristics of the participants stratified by occupation, and radar plots in Figure 1 display the average score of the psychological constructs of interest for each occupation group at each time point. Overall clerks reported higher average scores on almost all the scales, and their psychological condition seems to worsen, at least at a descriptive level, at the second time point. State-anger, DASS-21 scales, and emotional exhaustion subscale are, on average, higher for all the healthcare workers at the second time point (complete descriptive statistics are reported in Table S1). Observed ranges for all the psychometric scales are reported in Table S2. exhaustion subscale are, on average, higher for all the healthcare workers at the se time point (complete descriptive statistics are reported in Table S1). Observed rang all the psychometric scales are reported in Table S2.    Table 3 displays the estimated models for the DASS-21 subscales. DASS-21 subscales are not significantly different at T1, with respect to the baseline values (T0). Higher levels of worry, worse working conditions, and having a psychological/psychiatric history significantly increase anxiety, depression, and stress levels. Having undergone emergency training significantly decreases all DASS-21 subscales. Higher levels of perceived social support decrease only participants' depression levels. Considering the effects of coping strategies, we found that the use of problem-focused coping significantly decreases all DASS-21 subscales, avoidant coping significantly increases all DASS-21 subscales, and emotional-focused coping significantly increases only depression and stress levels. Finally, nurses show higher anxiety levels than physicians. Figure 2 shows the estimated regression trees for the prediction of the DASS-21 subscales scores. For depression, among all the variables, the algorithm selected both avoidant coping and working conditions as the variables best discriminating among participants' response profiles. High levels of avoidant coping and worse working conditions predict the highest levels of depression, whereas low levels of avoidant coping and better working conditions predict lower levels of depression symptoms. For subjects with avoidant coping ≥ 2.2, the mean depression level is 4.4. For subjects having a working conditions score < 2.4 and avoidant coping scores lower than 1.7, instead, the predicted depression level is 2.

DASS-21-Depression, Anxiety, and Stress
When anxiety is considered as an outcome, in addition to avoidant coping and working conditions, the algorithm selected the worry scale.
Following the tree branches, participants with higher scores on the worry scale (≥3.4) and avoidant coping scores, greater than or equal to 2.2, show the highest average value equal to 3.9. Conversely, participants with scores on the worry scale lower than 3.4, and showing avoidant coping scores lower than 1.9, show the lowest average value equal to 1.5. In the tree for stress, psychiatric/psychological history and problem-focused coping strategy were selected, in addition to splitting variables characterizing the other two trees, to best discriminate among different response profiles.
Following the tree branches, participants with higher scores on the worry scale (≥ 3.4) and avoidant coping scores, greater than or equal to 2.2, show the highest average value equal to 3.9. Conversely, participants with scores on the worry scale lower than 3.4, and showing avoidant coping scores lower than 1.9, show the lowest average value equal to 1.5. In the tree for stress, psychiatric/psychological history and problem-focused coping strategy were selected, in addition to splitting variables characterizing the other two trees, to best discriminate among different response profiles. Depending on the answer, other branches may appear until the final node, which displays the average predicted outcome value for participants, satisfying all the conditions, leading to that node and the proportion of subjects falling in the node itself. For example, in the first tree for the depression scale, the top split assigns observations having avoidant coping scores greater than or equal to 2.2 to the right branch. The predicted depression level for these subjects is given by the mean response value for the individuals in the data set with avoidant coping ≥ 2.2. For such subjects, the mean depression level is 4.4. Among subjects who have avoidant coping scores < 2.2, the working conditions also affect depression level. For subjects with avoidant coping score < 2.2 and working conditions ≥ 2.4, the predicted depression level is 2.9. For subjects having working conditions < 2.4 and avoidant coping scores between 1.7 and 2.2, the predicted depression level is 2.7, whereas for subjects having a working conditions score < 2.4 and avoidant coping scores lower than 1.7, the predicted depression level is 2. The same logic applies to all the other trees. Abbreviation: CONDWORK, working conditions; Psych History, psychiatric history. Depending on the answer, other branches may appear until the final node, which displays the average predicted outcome value for participants, satisfying all the conditions, leading to that node and the proportion of subjects falling in the node itself. For example, in the first tree for the depression scale, the top split assigns observations having avoidant coping scores greater than or equal to 2.2 to the right branch. The predicted depression level for these subjects is given by the mean response value for the individuals in the data set with avoidant coping ≥ 2.2. For such subjects, the mean depression level is 4.4. Among subjects who have avoidant coping scores < 2.2, the working conditions also affect depression level. For subjects with avoidant coping score < 2.2 and working conditions ≥ 2.4, the predicted depression level is 2.9. For subjects having working conditions < 2.4 and avoidant coping scores between 1.7 and 2.2, the predicted depression level is 2.7, whereas for subjects having a working conditions score < 2.4 and avoidant coping scores lower than 1.7, the predicted depression level is 2. The same logic applies to all the other trees. Abbreviation: CONDWORK, working conditions; Psych History, psychiatric history.   Table 4 shows the estimated models for the MBI emotional exhaustion scale, the ISI score, the IES-R Intrusion score, and the State anger score. When comparing the two time points, the state anger score significantly increases at T1 with respect to the baseline values (T0). Only for clerks, the MBI emotional exhaustions score also significantly increases at T1 compared to T0. Focusing on the occupation, overall, nurses report higher ISI levels and IES-R Intrusion levels than physicians. Higher levels of worry and worse working conditions significantly increase all of the considered scales. A previous history of psychological or psychiatric symptoms significantly increases the MBI emotional exhaustion score, the ISI score, and the IES-R Intrusion score but not the state anger score. Older age significantly increases the ISI score and the IES-R Intrusion score. Having undergone emergency training significantly decreases the ISI and IES-R Intrusion scores. Higher levels of perceived social support decrease the MBI emotional exhaustion score. Finally, we found that using avoidant coping significantly increases all considered scales, and the use of problem-focused and emotion-focused coping strategies significantly influences the MBI emotional exhaustion scale and state anger scale. Figure 3 shows the estimated regression trees for the same outcomes. For emotional exhaustion, the algorithm selected working conditions, avoidant coping strategies, and survey administration time as the best discriminating variables. Worse working conditions and higher use of avoidant coping strategies will lead to the highest predicted levels of emotional exhaustion, whereas the lowest outcome score is predicted for participants with better working conditions (<2.4). Moreover, worse working conditions and avoidant coping scores lower than 2.3, depending on the survey administration time, will lead to different predicted outcome scores, with higher predicted values at the time of the second survey.

MBI Emotional Exhaustion, Insomnia, IES-R Intrusion, and State Anger
For insomnia, the algorithm selected working conditions and the worry scale as the best discriminating variables. Worse working conditions and higher levels of worry will lead to the highest predicted levels of insomnia, whereas the lowest outcome score is predicted for participants with better working conditions (<2.4). Participants with worse working conditions and a level of worry lower than 3.4 will have predicted values lying in the middle and between those found in the previous two scenarios.
With reference to intrusion, working conditions, avoidant coping strategies, and scores on the worry scale are selected as the best splitting variables to identify different response profiles. Higher use of avoidant coping strategies (≥1.8) will lead to higher predicted levels of intrusion.
When avoidant coping strategies are lower than 1.8 and associated with better working conditions and lower levels of worry, the lowest levels of intrusion are predicted.
When avoidant coping strategies are lower than 1.8 and associated with worse working conditions (>2.7), participant profiles, with levels of intrusion lying in the middle, between the best and the worst scenarios, are identified.
For state anger, out of all the variables, only the worry scale plays a role in discriminating among participants' response profiles, with higher levels of worry leading to the highest predicted state anger score.
Estimated LME models for depersonalization and professional realization MBI subscales are reported in Table S3. In Table S4, LME models for the IES-R subscales of avoidance and hyperarousal are reported.

Discussion
The present study is the second phase of a longitudinal study investigatin psychological consequences of the COVID-19 outbreak and their predictive factor sample of Italian healthcare workers. To our knowledge, this is one of the few longitu studies that monitored the mental health of healthcare workers during the COV outbreak [22,42,43].
In general, the scores obtained by our sample on the psychological scales durin second phase do not significantly differ from the scores obtained during the first p Overall, regardless of the category of healthcare workers, our sample only show increase in state anger levels. Furthermore, during the second wave, clerks seem experience higher levels of burnout symptoms.
These results are in line with the scant literature, showing a substantial invarian psychiatric symptoms among healthcare workers throughout the epidemic [22,42,43 However, working under chronic stress conditions and experiencing a poor q of sleep both contribute to increased feelings of anger among healthcare workers d the COVID-19 outbreak [44,45]. Moreover, during the second wave of the pand public opinion appears to have shifted, and healthcare workers now feel less sup appreciation, and trust from the general population, and they are viewed less favo than in the past [22]. Consistently, during previous epidemics, feelings of anger se to be frequent among healthcare workers, caused by both long-term stressful wo conditions and poor adherence to infection control guidelines, as well as rejection stigma experienced from relatives and public opinion [46].

Discussion
The present study is the second phase of a longitudinal study investigating the psychological consequences of the COVID-19 outbreak and their predictive factors in a sample of Italian healthcare workers. To our knowledge, this is one of the few longitudinal studies that monitored the mental health of healthcare workers during the COVID-19 outbreak [22,42,43].
In general, the scores obtained by our sample on the psychological scales during the second phase do not significantly differ from the scores obtained during the first phase. Overall, regardless of the category of healthcare workers, our sample only showed an increase in state anger levels. Furthermore, during the second wave, clerks seemed to experience higher levels of burnout symptoms.
These results are in line with the scant literature, showing a substantial invariance of psychiatric symptoms among healthcare workers throughout the epidemic [22,42,43].
However, working under chronic stress conditions and experiencing a poor quality of sleep both contribute to increased feelings of anger among healthcare workers during the COVID-19 outbreak [44,45]. Moreover, during the second wave of the pandemic, public opinion appears to have shifted, and healthcare workers now feel less support, appreciation, and trust from the general population, and they are viewed less favorably than in the past [22]. Consistently, during previous epidemics, feelings of anger seemed to be frequent among healthcare workers, caused by both long-term stressful working conditions and poor adherence to infection control guidelines, as well as rejection and stigma experienced from relatives and public opinion [46]. Furthermore, the increasing levels of burnout experienced by clerks during the second phase may be explained by the end of smart-working and the subsequent return to work in hospitals.
Concerning predictive factors, our results highlight that higher levels of worry, worse working conditions, a previous history of psychiatric illness, being a nurse, older age, and avoidant and emotion-focused coping strategies seem to be risk factors for healthcare workers' mental health. Conversely, high levels of perceived social support, the attendance of emergency training, and problem-focused coping strategies seem to be protective factors for healthcare workers' mental health.
Specifically, worse working conditions and worry about the infection represent risk factors for higher levels of depression, anxiety, stress, burnout, PTSD, insomnia, and state anger. Considering the regression trees, a score greater than or equal to 2.4 in working conditions is a cut-off for higher depression scores (for participants showing avoidant coping scores lower than 2.2), stress, burnout, and insomnia; a high worry score (≥3.9) is the only factor that distinguishes between the highest and lowest levels of state anger, whereas a score greater than or equal to 3.4 discriminates among participants with more extreme levels of anxiety and insomnia.
These findings are in line with the literature showing that difficult working environments, including poor supervision and organizational support, intense workloads, a lack of personal protective equipment or its continuous use for many hours, and fear and concern about becoming infected or infecting relatives or colleagues, increase the psychological distress' levels of healthcare workers [47][48][49].
Concerning demographic variables, older age seems a risk factor for higher levels of insomnia and intrusion symptoms. This result is consistent with the literature, which shows an association between increasing age and a physiological decline in sleep quality [50]. Additionally, the literature identifies older age as a risk factor for PTSD symptoms during the COVID-19 pandemic, probably due to the elderly being a high-risk category for infection and death [51,52].
Moreover, a previous history of psychiatric illness seems to be a risk factor for higher levels of depression, anxiety, stress, burnout, PTSD, and insomnia symptoms. The current literature highlights that a history of psychiatric symptoms seems to make healthcare workers more vulnerable and more likely to experience psychological distress during the COVID-19 emergency [53,54].
Furthermore, being a nurse seems to be a risk factor for higher levels of anxiety, PTSD intrusion symptoms, and insomnia. This result is consistent with the literature showing that being a nurse represents a risk factor for worse mental health, which is probably due to the longer time spent with patients in contact with their fears, suffering, and death [18,49,55,56].
Concerning protective factors, having undergone emergency training predicts low levels of depression, anxiety, stress, PTSD, and insomnia symptoms. This result is in line with the literature showing perceived adequacy of training as a protective factor for longterm psychiatric morbidity [57], post-traumatic stress [58,59], and burnout [58] in previous epidemics. This is understandable given that the goal of emergency training is to improve healthcare workers' skills and abilities to increase their sense of control and avoid being overwhelmed during a real emergency [60].
High levels of perceived social support from family and friends seem to be a protective factor for lower depression and burnout symptoms. In line with the literature, high perceived social support from family, colleagues, and friends is helpful to deal with workrelated stress and to increase self-confidence [18,47].
Concerning coping strategies, regression trees highlighted that higher avoidant coping scores lead to the highest scores in the depression, anxiety, stress, burnout, and PTSD subscales.
The detrimental role of avoidant coping strategies is consistent with current and past epidemic literature [17,[61][62][63]. Avoidance strategies, such as denial or self-distraction, may be helpful for short periods of time to allow the person to continue with their tasks while also giving them some time to think. However, in the long-term, these coping strate-gies are dangerous for mental health, provoking dysfunctional detachment and distance from the problem, while changing neither the situation nor the associated psychological distress [64,65].
Conversely, the use of problem-focused coping strategies seems to be a protective factor for lower levels of depression, anxiety, stress, burnout, and state anger. This result is in accordance with the literature, showing that these strategies can increase feelings of autonomy and self-efficacy while reducing psychological distress [17,61,66]. These coping strategies may help in changing the meaning of the event and focusing on a specific goal, thereby increasing the perception of control, and avoiding being overwhelmed by the stressful situation [64,65].
Finally, the use of emotion-focused coping strategies seems to be a risk factor for higher levels of depression, stress, burnout, and state anger. The literature concerning emotion-focused coping strategies is contradictory [17,64], which is probably due to the heterogeneous nature of the strategies included in this subscale (e.g., use of emotional support and self-blaming). Moreover, during an emergency, problem-focused strategies appear more effective and adaptive than emotion-focused strategies [67]. Making difficult decisions in a stressful and extraordinary situation, with little knowledge about the disease and poor support from family and friends due to isolation, may be overwhelming if emotions are prioritized over the problem. Indeed, because of the governments' social restrictions policy in response to the COVID-19 emergency, a typically positive and protective coping strategy, such as the use of emotional support [66,68], could be a risk factor for poor mental health among healthcare workers [63,66,67,69].
In general, our findings align with other Mediterranean studies on the psychological impact of the COVID-19 pandemic on healthcare workers. Indeed, studies conducted in Italy, Greece, Portugal, Spain, and France highlighted a more resilient attitude among healthcare workers compared to studies conducted in China, the United Kingdom, and the Middle East [70]. Moreover, worse working conditions, higher levels of worry about the infection, and a previous history of psychiatric illness seem to be the main risk factors for poor mental health among Mediterranean healthcare workers [71][72][73][74].

Limitations and Strengths
Among limitations to be acknowledged, a self-selection bias may have occurred, as only participants experiencing low levels of psychological distress may have taken part in this research. Then, as only a subset of participants completed the assessment twice, the study sample may not be representative of both the healthcare worker population and the initial study sample.
However, this longitudinal study allowed us to monitor the mental health of healthcare workers over time and investigate relationships among variables to identify potential target risk and protective factors. Another strength of our study is the inclusion of healthcare workers from various hospitals located throughout Italy. Finally, the use of an online survey may have encouraged participants to reveal sensitive aspects of their work without worrying about confidentiality [75].

Conclusions
Healthcare workers seemed to be resilient in the face of the pandemic. Their mental health and general wellbeing is a complex issue that the government should monitor. In the current context, much can be offered, such as virtual clinics and remotely delivered psychological therapies and psychoeducation. However, it is also necessary to reduce mental health stigma, as physicians appear generally reluctant to disclose their problems, even when they are experiencing significant psychological distress [76]. The proposed data mining approach allowed us to derive classification rules and to identify risk and protective factors for healthcare workers' psychological wellbeing, which should be monitored during emergency situations. The proposed statistical methodology could provide more insight into the psychological aspects which are leveraged when implementing training/intervention programs.

Supplementary Materials:
The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/jcm11092317/s1, Table S1: Descriptive statistics of investigated psychometric variables; Table S2: Observed ranges of investigated psychometric variables; Table S3: Estimates (standard-errors) of the models. Square root transformation was applied to MBI depersonalization scale, while scores on the MBI professional realization scale were raised to the power of two; Table S4: Estimates (standard-errors) of the models. Square root transformation was applied to scores on both IES-R Avoidance and Hyperarousal scales. Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.
Data Availability Statement: Data cannot be shared because participants did not provide written informed consent for it.

Conflicts of Interest:
The authors declare no conflict of interest.