An online survey was developed in VA REDCap [19] to measure all outcomes and covariates of interest, excluding facility-level metrics. The survey was designed to be completed online in 15 minutes.

We assessed moral injury risk using the 9-item Moral Injury Events Scale (MIES) with minor modifications to ensure applicability to healthcare workers [20]. The MIES is not a direct measure of moral injury, but a proxy thereof; it assesses exposure to potentially morally injurious events and distress caused by that exposure. We defined a positive MIES screen as moderate, strong, or very strong agreement with two or more items (indicative of both exposure and, where applicable, current distress) in any one MIES domain—betrayal, witnessing, and/or perpetration. We considered those who screened positive on MIES to be at risk for moral injury. Other studies have dichotomized the MIES in a similar manner but have not required multiple items for a positive screen [cf. 17]. We chose to require multiple items within a single domain (including, where applicable, items indicating present distress) in order to use the MIES as a surrogate measure of risk for moral injury.

We assessed posttraumatic stress using the 6-item Primary Care PTSD Screener for DSM-5 [21]; positive screen was defined as confirmation of exposure to a traumatic event plus agreement with two or more subitems. We assessed depression using the 2-item Patient Health Questionnaire-2 [22]; positive screen was defined as a score of at least 3/6. We assessed burnout using a single-item burnout measure found to be highly correlated with the longer Maslach Burnout Inventory [23]; positive screen was defined as endorsing persistent symptoms that won’t go away on the single-item burnout measure. Across these measures, positive screens do not necessarily indicate clinically meaningful symptoms or the presence of any disorder. However, we can consider those who screened positive as at risk for the measured condition.

We assessed pandemic exposures using questions developed by the study team to measure level of exposure to patient deaths, coworker deaths, denial of family visits to critically ill patients, perceived short-staffing and duration thereof, and lack of supplies and equipment. Individual covariates were assessed using standard demographic questions (age, gender, race/ethnicity) and questions from the VA’s All Employee Survey (occupation, supervisory status) [24]. Workplace culture covariates (e.g., management health/safety support, supervisor support, coworker support, decision authority) were assessed using selected scales from the Centers for Disease Control and Prevention’s National Healthy Worksite Program Health & Safety Climate Survey: INPUTS [25,26].

We assessed facility-level covariates (care quality, patient safety, patient satisfaction, and staff satisfaction metrics) by extracting VA administrative data from multiple sources, including VA’s Strategic Analytics for Improvement and Learning; the VA All Employee Survey; and VA Human Resources data [27]. See Table 1 for a description of all instruments and measures.

Aided by local contacts at participating facilities and facility directories, we assembled a roster of e-mail addresses for all employees (regardless of occupation) working in inpatient units, emergency departments, and nursing homes at each facility. In assembling e-mail rosters, we prioritized inclusiveness over accuracy, which resulted in the inclusion of some healthcare workers whose primary work location was outside of the targeted high-risk departments.

We e-mailed survey invitations and up to three reminders to potential participants. To minimize volunteer bias, the e-mail invitation described the study as an evaluation of healthcare workers’ experiences during the COVID-19 pandemic without mentioning specific outcomes of interest. Survey administration dates ranged from 12/10/2021 to 6/1/2022, roughly corresponding with the Omicron wave of the COVID-19 pandemic (after widespread vaccine availability). Of 10,563 staff invited, 2,810 (26.6%) participated. Complete data were obtained for 2,004 participants.

Demographic characteristics and occupations of the 2,004 survey participants who were included in the analysis are presented in Table 2. Nurses constituted a majority of the sample with other healthcare occupations adequately represented. The sample was diverse with regard to race/ethnicity, gender, and age. Among participants included in the analysis, 80% (1,605) endorsed working in one of the high-risk work locations targeted for inclusion (i.e., intensive care unit, other inpatient unit, emergency department, or nursing home).

We calculated unadjusted relative risks and chi-square tests to describe how all survey covariates were associated with risk of positive screens on the MIES. We then fit generalized linear mixed models using the modified Poisson method to estimate adjusted relative risks, with a log-link, robust error variance, and random intercepts for each VA facility [28]. Because a central purpose of our study was to understand how modifiable workplace factors can affect risk for moral injury, we specified a priori multiple stages of modeling guided by our original conceptual model (Fig 1). We added covariates in stages, with the initial stages focusing on factors conceptualized as modifiable (e.g., workplace/organizational culture; right side of Fig 1) and subsequent stages moving toward factors conceptualized as intrinsic (e.g., demographics; left side of Fig 1). This approach allowed us to evaluate how the relationships between more modifiable factors and moral injury risk might vary as more intrinsic factors were added to the model. For each stage of modeling, we classified covariates into three categories: mandatory inclusion, possible inclusion, or exclusion based on observed associations (or lack thereof) in unadjusted analyses and based on qualitative interview themes (qualitative methods are explained below). Covariates were included if unadjusted analyses or qualitative interview themes suggested a potential relationship; covariates were excluded if neither unadjusted analyses nor interviews indicated a relationship. When variables did not meet a priori criteria for inclusion or exclusion, we prepared versions of the model with and without them and then evaluated fit using the Bayesian Information Criterion (BIC) statistic. We also used the BIC statistic to assess overall model fit at each stage of modeling, moving from more modifiable to more intrinsic variables.

Unadjusted analyses and modeling techniques were repeated with secondary outcomes to examine their relationship to the primary outcome and to all covariates. In modeling secondary outcomes, we retained the covariates derived from the moral injury model so the relative risks estimated in each model would be comparable.

Finally, to assess moral injury risk at the facility level, we stratified facilities into quartiles based on their average scores on each facility-level metric (see Table 1) over the time period when the survey was administered. (For employee satisfaction [VA AES] measures, we weighted average scores by occupation to match the occupational distribution of survey respondents for each facility.) We then examined whether the prevalence of positive screens for moral injury was associated with metric performance across the four quartiles. We repeated this analysis for all secondary outcomes.

The survey questionnaire included an optional follow-up form in which respondents could indicate their willingness to participate in the qualitative phase of the study. The follow-up form was recorded separately from survey responses but forms were auto-sorted into separate batches based on participant responses to the MIES. Interview invitations were then sent in small batches to respondents who screened as at risk for moral injury, with purposive inclusion of all participating clinical occupations and all participating facilities. Invitations were sent until preliminary analysis suggested that interviews had reached saturation. Ultimately, 46 of 413 volunteers were interviewed. The final sample included 29 nurses, 8 physicians, 3 respiratory therapists, and 6 other clinicians.

Interviews lasted approximately 60 minutes and were conducted by telephone by research analysts with training and experience in qualitative methods, supervised by the principal investigator. Interviews were audio-recorded with permission of participants.

Interviews were guided by a semi-structured interview guide. The interview guide was organized by a priori domains aligned with the conceptual model, including what specific conflicts they experienced, what could/should have been done differently, what was done well, and protective factors/prevention strategies at the individual, unit, and organizational levels.

Interviews were analyzed in three phases. In phase one, interviewers completed a free-form memo summarizing central takeaways immediately after each interview. Memos highlighted observations pertinent to risk for moral injury. Phase two used a template-based rapid analysis technique designed to be time- and resource-efficient and to yield results that are comparable to traditional qualitative methods [29-31]. Guided by interview audio-recordings, qualitative team members prepared structured summaries of each interview using a template organized by topical areas drawn from the interview guide. This entailed summarizing participant responses for each domain and populating relevant quotations into the analysis template. To ensure consistency among analysts, more than one team member analyzed a subset of completed interviews (n=17, or 37%). The second analyst reviewed the original templated summary and added notes documenting any questions, observations, or conflicting interpretations. Regular team meetings and written correspondence allowed the team to discuss questions, identify differences in interpretation, and achieve consensus in analytic approach. The final phase of analysis involved identification and refinement of themes through an iterative, collaborative process that combined inductive and deductive analysis. Team members began by carefully reading all templates. A lead analyst then mapped preliminary themes and supporting evidence onto a spreadsheet-based matrix pre-organized by interview domain. All qualitative team members then collaborated to review, refine, and finalize the matrix.

Interviewed workers described feeling betrayed and abandoned by healthcare leaders and managers as an important risk factor for moral injury. This occurred when leaders’ decisions, actions, and/or inactions were seen as causing harm to the participants, their coworkers, and/or their patients. Examples included leaders’ failing to listen to and learn about the resource and staffing needs of frontline staff, failing to be visible to or present with frontline staff during the pandemic, or establishing policies and procedures that impacted working conditions or patient care without adequately consulting frontline staff.

This betrayal was felt most acutely when it negatively impacted patient care. In the words of an ICU RN, “I was desperately trying to do what I felt was right and best for my patients and not only did I feel unheard and unvalued by my leadership, but I felt handicapped by them, and I think that was particularly difficult… It felt like… my coworkers and I were betrayed and abandoned and left to [do] more with less and less.” Another inpatient RN said, “We felt like we were left to our own devices…. We got a lot of lip service and no actual action… It’s demoralizing and it's disheartening.”

This qualitative theme aligns with survey data, where 30% of respondents endorsed two or more items in the betrayal domain of the MIES. Notably, one of the strongest predictors of moral injury risk was perceived management support for workplace health and safety. Surveyed workers who strongly disagreed with the statement that “management considers workplace health and safety to be important” were greater than five times more likely to be at risk for moral injury compared to those who strongly agreed. In the final model, the adjusted relative risk (the contribution of this covariate over and above the contribution of all others in the model) was 2.70 (p<0.001).

Interviewed workers described following hospital policies that violated their own values as a significant contributor to moral injury. The most common example of this was enforcing rules requiring isolation of critically ill patients from loved ones or long-term isolation of nursing home residents from families. For example, an inpatient nurse practitioner found it extremely distressing when she was unable to allow a dying patient time with their young child: “We had a family member who had a 4-year old. This patient was dying and they wanted to see their baby one last time… They couldn't see him, was the bottom line, they couldn't see him.” Similarly, an ICU RN described the moral distress created by pandemic-era visitation restrictions: “We were taking care of them and they were dying without their family…. After they're dead, they're dead, and those policies I felt were wrong… That family member, if they want to suit up and wear the same protective gear that we have to wear… they should be given that choice.”

Survey data showed a significant correlation between moral injury risk and denial of family visits to critically ill patients. Workers who denied family visits to critically ill patients more than four times were 73% more likely to be at risk for moral injury than those who did not deny family visits. In the final model, the adjusted relative risk was 1.21 (p<0.001).

Interviewed workers experienced intense moral distress when they felt unable to provide the standard of care that patients deserved due to inadequate staffing and inadequate time to spend with each patient, or due to lack of needed supplies and resources (e.g., personal protective equipment). An emergency room nurse described feeling distress at “watching these people just be stranded or stuck in an emergency department” when the facility ran out of beds: “sometimes, just to get them moved, people would place them in the wrong level of care.” An ICU nurse felt intense moral conflict due to short-staffing in her department that prevented her from monitoring intubated COVID-19 patients with the frequency she felt was required: “We were expected to monitor those patients like once every four hours. And that felt extremely uncomfortable… You can’t appropriately paralyze somebody and not check them in four hours… They couldn’t protect themselves, you couldn’t protect them, that just wasn’t safe.”

This qualitative theme aligns with survey findings, which showed a significant correlation between moral injury risk and duration of short-staffing during the pandemic. Workers who reported that their team lacked personnel required for patient care for three months or longer were more than twice as likely to be at risk for moral injury compared to those who reported more transient short-staffing. In the final model, the adjusted relative risk was 1.62 (p<0.001).

Interviewed workers described feeling powerless to help dying patients as a significant contributor to moral injury: "We're losing people left and right, and there's nothing we can do.” An ICU RN acknowledged “the distress of knowing with almost certainty that every patient you cared for was probably going to die…. I remember experiencing a significant amount of moral and ethical conflict.” Interviewees’ distress was particularly acute when their patients were suffering. They felt that patients were relying on them and yet they could not deliver care that would mitigate their suffering or prevent their deaths. “When the patients are asking you to save them… that's the most distressing part,” explained another ICU RN.

This qualitative theme aligns with and helps to explain survey findings, which showed a significant correlation between moral injury risk and the deaths of patients under one’s care during the COVID-19 pandemic. Workers who cared for three or more COVID-19 patients who died were 89% more likely to be at risk for moral injury than those who did not experience a patient death from COVID-19. In the final model, the adjusted relative risk was 1.28 (p=0.002).

Yet, interviews suggested that not all deaths were equal in the eyes of healthcare workers; some were far more distressing than others, and many interviewees described at least one patient loss that haunted them. Failure to provide patients a death with dignity or to pause and honor those who died could be especially distressing. An LVN, for example, was deeply troubled that, when “your patient dies… you just got to throw ‘em in a body bag and clean the room and there’s another one waiting in the ER for the bed.”

Interviewed workers felt that conflicts between their duties at work and their obligations to family and loved ones contributed to moral injury. Especially during the COVID-19 pandemic, workers feared putting their own family at risk of contagious illness. An inpatient nurse explained, "We all understood that, on the one hand, we wanted to take care of patients like we always had… One part of our mind was focused on the job…. Another part of our mind, for every nurse I know, was also worried about 'what if I bring this home' to whoever was their special immunocompromised person… Every single person had somebody in their circle they were especially worried about bringing it home to."

Interviewed workers also reported moral distress related to neglecting the needs of loved ones because work demanded all of their time and energy. They felt they had nothing left to give at home. Another inpatient RN shared that her “parenting was severely affected… I just wanted to be in my room when I wasn't at work… I'm just exhausted and mentally not there.”

This qualitative theme aligns with survey findings, which showed a significant correlation between moral injury risk and work-family conflict. Workers who reported experiencing conflict most or all of the time were more than twice as likely to be at risk for moral injury than those who reported less frequent work-family conflict. In the final model, the adjusted relative risk was 1.59 (p<0.001).

A significant source of moral injury described by interviewees was feeling bound to deliver care that is futile and causes suffering. This happened often in the context of COVID-19 care but was relevant in other contexts as well. “It was really hard to watch painful procedures, painful care be done to people in the name of 'oh, well maybe we can get them better,'” shared an ICU RN, “The futile care was something I really struggled with.” Another ICU RN who cared for COVID-19 patients said she “felt like [we] were torturing them [patients] by proning them, then being on blood pressure medication and being intubated and sedated for more than two weeks and we knew the patient wasn't going to be able to come off the ventilator.” The word “torture” was used by multiple interviewees in the context of delivering futile care: “At a certain point, you're just literally torturing them and you're a part of that. Emotionally, it was draining, still is to this day,” shared a respiratory therapist.

There were no survey questions linked with this theme, but it emerged during qualitative interviews as a novel theme and a significant contributor to moral injury.

A related source of moral injury involved high-stakes provider-family communication failures. Specifically, interviewees reported feeling distressed that patients with dire prognoses and their families were sometimes not adequately and honestly informed of those prognoses. They felt that false hope affected their treatment decisions and that healthcare providers were not always doing enough to deliver difficult-to-hear, honest information. An ICU RN described the case of a COVID-19 patient she cared for: “Those two docs, the first time they saw him, knew there's no chance. But we, for three and a half months, we had this guy, our doc could never tell her there's no chance… It's just wrong, it's torture.” Another ICU RN was similarly distressed at these communication failures: “When you go to work, you feel like you're failing families because they're not being cared for or told information that you would want to know.”

Again, there were no survey questions linked with this theme, but it emerged during qualitative interviews as a novel theme and a significant contributor to moral injury.

Table 6 integrates qualitative and quantitative data to summarize the types of experiences and events that contributed to moral injury.

Across interviews, participants felt that the presence of a supportive workplace community provided some protection against moral injury. When distressing events were not faced alone and responsibility was meaningfully shared, workers felt that morally distressing events were less likely to happen or less likely to result in lasting moral injury. By contrast, there was little buffer against moral injury where there was lack of community, connection, and belonging – where, instead, individuals felt alienated from others and burdened by responsibility. A morally injured inpatient RN described “feeling like left-overs… Nobody wants to come to us, but we have to show up every time and do the job.” Another RN lamented that “nobody came to talk to us to make sure that we were mentally okay.”

This qualitative theme aligns with survey findings, which showed a significant correlation between moral injury risk and perceived coworker support. Workers who reported that their co-workers could not be relied on to help were 59% more likely to be at risk for moral injury compared to those who reported that they could rely on their co-workers. In the adjusted model, the relative risk was 1.18 (p=0.004).

Interviewees felt that moral injury was less likely when leaders, managers, and decision-makers were present and in communication with those on the frontlines. They emphasized the importance of meaningful communication in both directions so frontline workers feel genuinely heard and appreciated. When leaders, managers and decision-makers seemed disengaged from the frontlines, interviewees described feeling greater and more frequent moral distress. This was particularly true when leaders lacked understanding of the experiences of workers and made important decisions without listening carefully for their input. Unfortunately, this was a common experience. One ICU RN shared, "My manager was not really there. She was not involved, she did not glove up, put PPE on, and actually work. She has not actually stepped into a COVID room but yet she's representing her staff and she has no idea what she's asking her staff to do.” Similarly, an ER physician explained that his feelings of betrayal would have been eased "if they [hospital leaders] would appear to understand how hard this has been for us, or acknowledge that…but, like I said, we don't see them.”

In alignment with this qualitative theme, survey data showed a significant correlation between moral injury risk and perceived supervisor support. Workers who disagreed that their supervisors were helpful in getting the job done were twice as likely to be at risk for moral injury compared to those who agreed. In the final model, the adjusted relative risk was 1.15 (p=0.026).

When interviewees saw that it was possible to make meaningful changes to their workplace and to address the factors that caused moral distress, they experienced this as a mitigating factor that could reduce moral injury risk. By contrast, workers’ distress and frustration were amplified when they lacked agency to improve the situation or to do things differently in the future. For example, a respiratory therapist explained that his own distress would have been alleviated “if management had only asked… ‘what could be learned from this, what could be changed?’” But this did not happen.

This qualitative theme helps to explain survey findings, which showed a significant correlation between moral injury risk and perceived empowerment in the workplace. Workers who strongly disagreed they had the authority to make decisions that impacted them at work were more than twice as likely to be at risk for moral injury than those who agreed or responded neutrally. In the final model, the adjusted relative risk was 1.24 (p=0.006).

Across interviews, the presence of meaningful breaks that allowed for reflection and recovery was identified as a factor that could provide some protection against moral injury. Workers felt that morally distressing events were less likely to result in lasting moral injury when there was time to pause and opportunity to process and discuss these events. Morally injured workers described feeling compelled to ignore or suppress their feelings so they could continue to operate under high-stress conditions. For them, this could feel unrealistic and inhumane. One ICU RN shared, 'I took care of [Tom] for four weeks. I took care of [Chris] for two weeks. I took care of [Steve] for three weeks. I took care of [Andy] for four. You don't get to take care of these people every day and then put them in body bags and it doesn't impact you.” Those losses, she felt, had to be acknowledged and grieved.

There were no survey questions linked with this theme, but it emerged during qualitative interviews as a novel theme and a notable factor impacting moral injury risk.

Although different jobs carry different risks and burdens, interviewed workers appreciated it when the organization made an effort to share the heaviest burdens, meaningfully acknowledge workers who carried more than their share of burden, and take action to make things more balanced when there were pronounced inequities. By contrast, interviewees felt that moral injury was more likely when certain workers were consistently asked to bear a disproportionate share of workplace risk and burden over a long period of time. A common example was the routine delegation of inpatient COVID-19 care to a very small number of nurses in order to protect other employees from any contact with COVID-19 inpatients. “We bore the brunt of this pandemic,” shared an exhausted inpatient nurse. Workers also felt frustrated and betrayed when their sacrifices were acknowledged in only superficial ways and there were no meaningful efforts to ensure a fairer distribution of risk and burden. An emergency room nurse described feeling “abandonment by higher up people.”

There were no survey questions linked with this theme, but it emerged during qualitative interviews as a novel theme and a notable factor impacting moral injury risk.

Table 7 integrates qualitative and quantitative data to summarize workplace conditions affecting moral injury risk.