Enhancing mental toughness among healthcare leaders to support employee engagement

1 Introduction

Workplace stress remains a critical organizational concern, contributing to an estimated 120,000 deaths and $190 billion in excess healthcare costs annually (Goh et al., 2019). Healthcare is consistently cited as one of the most stressful sectors, with nearly half of employees experiencing burnout (Rotenstein et al., 2023). Senior healthcare leaders, in particular, operate under persistent strain, as they are responsible for organizational performance, workforce stability, and patient outcomes (Shanafelt et al., 2017; Singh et al., 2024; Lee et al., 2019). In such contexts, sustained performance depends on psychological resources that help employees manage pressure, stay engaged, and lead effectively (Joshi et al., 2022; Shanafelt and Noseworthy, 2017).

Effective leadership is one of the most important factors for organizational success (Alloubani et al., 2014). High levels of stress may erode employees' cognitive, emotional, and behavioral capacity, directly influencing how they show up at work (Shuck et al., 2017b). These psychological demands shape employee engagement (Eng)—a work-related state defined by the intensity, direction, and maintenance of employees' cognitive, emotional, and behavioral energies (Shuck et al., 2017a; Lee and Shuck, 2024). Rooted in Kahn's (1990) original conceptualization, engagement reflects how fully employees invest themselves in their work roles—physically, cognitively, and emotionally (Kahn, 1990; May et al., 2004). Engagement matters; disengaged healthcare workers report higher rates of medical errors (Janes et al., 2021; Scott et al., 2022), whereas engaged employees demonstrate improved performance, innovation, citizenship behaviors, and reduced absenteeism (May et al., 2004; Shuck, 2019; Opoku and Boateng, 2024).

The healthcare environment presents persistent structural and emotional barriers to employee engagement. Heavy patient loads, exposure to suffering and loss, inadequate staffing, extended work hours, and limited recovery opportunities constrain employees' capacity to remain fully invested in their work over time (Greenberg, 2020; Rotenstein et al., 2023; Wang et al., 2023). These stressors contribute to burnout and turnover, with the average hospital turnover rate reaching 18.3%, and each 1% shift costing organizations nearly $289,000 annually (NSI Nursing Solutions Inc, 2025). Post-pandemic trends indicate continued turnover linked to stress, trauma, and depleted psychological resources (Poon et al., 2022; Rotenstein et al., 2023). Without mechanisms to replenish or strengthen employees' internal capacities, organizations may face disengagement, declines in performance, and potential risks to patient safety (Collini et al., 2015; Singh et al., 2024). Although research has extensively documented stress and burnout in healthcare settings (Shanafelt et al., 2017; Solms et al., 2023), far less attention has focused on the internal psychological resources that may enable senior leaders to remain engaged and effective under persistent demands (Lee and Kim, 2023). This gap is significant, as leader engagement is central toward shaping organizational culture and performance (Shanafelt and Noseworthy, 2017).

One promising psychological resource is mental toughness (MT). Mentally tough individuals demonstrate composure under pressure, quick recovery from setbacks, and clarity in decision-making even in volatile contexts (Lee and Kim, 2023; Lin et al., 2017). We acknowledge that the term ‘mental toughness' may carry unintended negative connotations, particularly if interpreted as implying that effective leadership requires stoicism or that individuals should simply endure stress without support. In this study, however, mental toughness is conceptualized as a set of developable psychological skills related to attention, emotion, and behavior regulation under pressure. We, therefore, use the term to describe trainable capacities that may support leadership functioning alongside, not instead of, organizational support and healthy working conditions. Conceptually, MT reflects the ability to remain committed to multiple higher-order objectives that may be in tension, regulate thoughts and emotions under stress, and view challenges as opportunities for growth. Such attributes are commonly studied within high-performance sport, but increasingly relevant to other high-stakes domains (Gucciardi et al., 2015; Gucciardi, 2017; Stamatis et al., 2020a). In today's healthcare landscape, with rapid technological change, shifting workforce expectations, and high stakes for patient outcomes, but also financial underperformance, high labor costs, reimbursement delays/denials, and thin cash, MT may help leaders maintain both individual and organizational performance (Ruparel, 2020; Tawfik et al., 2019). Despite conceptual alignment, few studies have examined whether strengthening MT can influence employee engagement in organizational settings, and almost none have done so among senior healthcare leaders (De Kock et al., 2022). Addressing this gap is critical as healthcare organizations seek strategies to improve well-being, morale, and patient care through evidence-based psychological interventions (Albott et al., 2020; Ho et al., 2024; Rosa et al., 2022). However, the dose-response relationship for psychoeducation interventions has not been systematically assessed in the current literature, though existing research suggests program duration may influence outcomes (Payne et al., 2025).

Therefore, the study's purpose was to examine whether participation in a structured MT intervention was associated with changes in employee engagement among senior healthcare leaders. We aimed to determine whether engagement increased across sessions, and whether greater participation (“dose”) predicted larger within-person gains. This pilot investigation sought to provide preliminary, exploratory, hypothesis-generating evidence for the potential of MT skills training initiatives as a viable mechanism to enhance employee engagement in high-pressure organizational contexts. The study focuses on two primary questions:

RQ1: How did employee engagement change over the course of the mental toughness intervention?

RQ2: What was the association between session attendance (dose) and within-person change in employee engagement?

2 Materials and methods

2.1 Participants

A purposive sample of 12 University of Louisville Physicians senior healthcare leaders employed within University of Louisville Health (UofL Health) were enrolled in a six-session MT program. Prespecified eligibility requirement included (a) a current formal leadership role (e.g., senior director, director), (b) employment at UofL Health, (c) ability to attend the six sessions and complete brief session surveys, (d) age ≥18, and provision of informed consent. No additional exclusion criteria were applied. Initially, 20 leaders were contacted via email by senior organizational leadership (independent of the intervention and manuscript team) and 12 agreed to participate. Participants represented 12 Departments/Units (e.g., Operations, Communications, Patient Access, Quality, Finance). The target size reflected pilot capacity/availability (≈10–15) rather than formal power calculations. All participants contributed baseline data. Participants were assigned study IDs, no Protected Health Information (PHI) was used in analysis. Demographic characteristics were not collected to minimize identifiability and encourage honest participation. The study was IRB-approved by University of Louisville, protocol (IRB 25.0027). Written consent was obtained electronically prior to the first session and reaffirmed presession. No compensation was provided.

2.2 Instruments

The English version Mental Toughness Index (MTI) was administered at every session (Available for free here: http://www.danielgucciardi.com.au/questionnaires.html). The scale has 8 items, each rated 1–7 (1 = “False, 100% of the time”; 7 = “True, 100% of the time”); we used the sum score (range 8–56), with higher scores equaling greater MT. There are no reverse-coded items. We evaluated internal consistency at each session using Cronbach's α and McDonald's ω. The MTI was chosen because the psychoeducation was based on Gucciardi's MT conceptualization (Gucciardi et al., 2015; Gucciardi, 2017). Additionally, employee engagement was measured at each session using the Employee Engagement Scale (EES; Shuck et al., 2017a). The 12-item scale assesses cognitive (CE), emotional (EE), and behavioral engagement (BE), with responses recorded on a 5-point Likert scale (1 = strongly disagree, 5 = strongly agree). Higher scores reflect greater engagement. Item-level responses (Q1–Q12) were collected by the project team and used to compute subscale and total scores. For the analyses, we used the provided totals. Internal consistency (Cronbach's α and McDonald's ω) was computed at each session for the total and subscales.

2.3 Procedure

Leaders attended 6 bi-weekly, 2-h group sessions. Session content followed a standardized agenda aligned with the MT key dimensions: generalized self-efficacy, context knowledge, facing adversity, emotion regulation, attention regulation, buoyancy, optimistic style, and success mindset. All sessions were facilitated by the second author using brief didactics, guided discussion, and applied skill practice. Attendance was logged at each session and served as the exposure/dose metric. Surveys (MTI and EES) were administered at each session in paper or Qualtrics formats. If a participant was absent or a total score was not recorded, that session was treated as missing for that measure. No protocol deviations were recorded in the available documentation. Data were centrally merged and cleaned by first author and analyzed using anonymized study IDs only by second author. Across sessions, content progressed from foundational conceptualization (e.g., defining mental toughness within an applied leadership context) to specific skill domains including adversity appraisal and buoyancy, emotion and attention regulation, belief-based mechanisms (self-efficacy, optimism), and goal-directed persistence and contextual decision-making. Sessions combined brief didactic instruction with guided reflection and application to participants' real-world leadership challenges. A detailed session-by-session curriculum, including core content and evidence base, is provided in Supplementary Appendix A.

2.4 Statistical analyses

2.4.1 Overview and estimands

Primary questions were (RQ1) how engagement changed over the course of the MT intervention, and (RQ2) the association between session attendance (dose) and within-person change in employee engagement. Given small N, analyses emphasized within-person change and multiple robustness checks. In more detail, we focused on within-person change over time and modeled training as cumulative “reps” from prior time points. This matches how skills typically develop, and helps reduce concerns that changes in the outcome might be driving later training exposure. Secondary, conditional on the primary results, exploratory questions addressed two additional subsets of aims: (A) RQ1a–c: drivers (which engagement subscales changed most), distribution/minimal clinically important difference (MCID; whether the session-wise distribution shifted upward and what proportion achieved ≥+3 points from baseline), and mechanism (whether session-level MT_t preceded next-session engagement Eng_{t+1}), and (B) RQ2a–d: the nature of dose (recency-weighted exposure, consecutive streaks, practical thresholds, and baseline moderation). To assess stability, we implemented a robustness portfolio including: leave-one-out, permutation, cluster bootstrap, outlier-robust estimators, Bayesian priors, IP-weighted sensitivity, and non-parametric trend tests. Consistent with pilot studies in healthcare research (Arnold et al., 2009; Leon et al., 2011; Troy et al., 2023), all findings are interpreted as feasibility assessments and hypothesis-generating.

The research questions guiding this analysis included:

RQ1: How did employee engagement change over the course of the mental toughness intervention?

Hypothesis 1 (directional): Employee engagement increased linearly across intervention sessions.

RQ2: What was the association between session attendance (dose) and within-person change in employee engagement?

Hypothesis 2 (directional): Higher cumulative session attendance was positively associated with greater within-person increases in employee engagement.

The secondary research questions, contingent upon conclusions regarding the overall increase in engagement and the presence of a dose-response relationship, include the following sub-research questions:

RQ1: How did employee engagement change over the course of the mental toughness intervention?

RQ1a: Which subscales (CE, EE, BE) drive RQ1?

RQ1b: Does the distribution shift upward; what percentage meets ≥+3 points?

RQ1c: Is mental toughness at session t associated with subsequent engagement at session t+1?

RQ2: What was the association between session attendance (dose) and within-person change in employee engagement?

RQ2a: Are results robust to recency-weighted dose, streaks, and attendance rate?

RQ2b: Does attending the immediately prior session give an extra short-term lift?

RQ2c: Is ≥4 sessions a practical threshold for meaningful gains?

RQ2d: Does baseline engagement/mental toughness modify the dose effect?

2.4.2 Variables

Time = session (1–6). Dose = number of sessions with valid MT and Engagement recorded prior to time t (cumulative lagged dose; primary). Alternative dose definitions included a recency-weighted dose (geometric decay 0.7), consecutive attendance streak (up to t-1), and attendance rate (total/6; between-person). Outcomes were (a) engagement level and (b) change from baseline, ΔEng_t = Eng_t – Eng_baseline. Covariates included session (factor or linear term) and concurrent MT_t; baseline engagement was included where appropriate.

2.4.3 Primary models

Time trend (RQ1): linear mixed-effects model (random intercepts for person, random slopes for time where supported).

Dose-response (RQ2; primary estimand): ΔEng_t regressed on cumulative lagged dose + session + MT_t. We used Linear Mixed-Effects Model or Ordinary Least Squares regression with cluster-robust Standard Errors at the person level. The main specification used all available sessions. A tightened specification excluded baseline rows (sessions 2–6).

2.4.4 Secondary analyses

RQ1a (drivers): subscale models for CE, EE, BE trajectories and Δ scores.

RQ1b (distribution/MCID): distributional shifts and proportion achieving ≥+3 points.

RQ1c (mechanistic, exploratory): MT_t predicting Eng_{t+1} with person/session effects.

RQ2a (alternative dose specs): recency-weighted dose, streak length, attendance rate.

RQ2b (sequentiality/recency): immediate change Eng_t – Eng_{t−1} on attended prior session?

RQ2c (threshold): ≥4 vs. ≤ 3 sessions for achieving MCID.

RQ2d (moderation): dose × baseline engagement and dose × baseline mental toughness.

2.4.5 Robustness portfolio

Leave-one-participant-out (12 re-fits), cluster-aware permutation tests (1,000–2,000 circular shifts within person, two-sided p); cluster bootstrap (2,000 participant resamples) to obtain percentile CIs and tail probabilities P(β > 0), P(β >0.5), P( β >1.0), outlier-robust fits (winsorization at 5^th-95^th percentiles, Huber M-estimation); non-parametric trend tests (Kendall's τ, Spearman's ρ, Kruskal–Wallis across dose groups); Bayesian hierarchical model (random intercepts; skeptical prior on the dose effect: Normal[0, 0.5²] points/session; weakly informative elsewhere) with posterior means, 95% credible intervals, probability of direction, and P(β > 0.5); and an IP-weighted marginal structural model using stabilized weights from pooled logistic completion models (SWt = P(completet |session)P(completet|session,priorEng,priorMT,baseline)) with weighted regression of engagement on cumulative lagged dose + session + baseline.

2.4.6 Missing data

Mixed models used all available rows under a Missing At Random (MAR) assumption. Sensitivities included last-observation-carried-forward (LOCF) for ΔEng, with re-estimation of the primary models.

2.4.7 Multiplicity

For subscale families (CE, EE, BE), p values were adjusted using Holm (or False Discovery Rate, FDR, as specified in Results). Robustness checks addressing the same estimand were not multiplicity-adjusted because they serve to assess estimator stability rather than to test independent hypotheses.

2.4.8 Assumptions and diagnostics

We inspected residual QQ plots and residuals-vs.-fitted, estimated ICC from a null LMM to justify random effects, and probed influence via leave-one-out and robust fits. All tests were two-sided α = 0.05; effect sizes with 95% intervals were emphasized. Analyses were conducted in Python. Reproducible code is available on request.

2.4.9 Internal validity appraisal (NIH tool)

We applied the NIH Quality Assessment Tool for Before-After (Pre-Post) Studies With No Control Group to document risks of bias and design safeguards (National Heart, Lung, and Blood Institute (NHLBI), n.d.).

3 Results

Overall, engagement rose across sessions (≈+3 points), and each additional prior session was associated with ≈+1.3 points higher within-person engagement at a given session. The dose signal was consistent across resampling, randomization, and robustness checks.

Twelve healthcare leaders enrolled. Six sessions were scheduled per leader (72 planned person-sessions). All 12 contributed baseline data. Across sessions 2–6 (60 post-baseline person-sessions), totals were intermittently missing: ~32% for MT and 36% for Engagement. Put differently, there were about 41 analyzable MT totals and 38 Engagement totals post-baseline. Missingness appeared to be intermittent rather than progressive attrition, as several leaders contributed data after a missed session. A random-intercept null model for Engagement yielded a high intraclass correlation (ICC ≈ 0.84), indicating substantial within-person clustering and supporting longitudinal mixed-effects modeling (Chen et al., 2018; Liljequist et al., 2019; Nakagawa et al., 2017). Internal consistency was acceptable for MT across sessions (Cronbach's α and McDonald's ω ≈ 0.60–0.80) and moderate for Engagement and for the Employee Engagement Scale total, with Cronbach's α ranging from 0.68 to 0.88 and McDonald's ω ranging from 0.76 to 0.90 (Session 1: α = 0.70, ω = 0.76; Session 2: α = 0.68, ω = 0.78; Session 3: α = 0.88, ω = 0.83; Session 4: α = 0.86, ω = 0.90; Session 5: α = 0.75, ω = 0.77; Session 6: α = 0.81, ω = 0.80). Analyses used all available observations, with missingness addressed through model-based inference and prespecified sensitivity checks described in Table 1.

Table 1. Sample and missingness.

“Dose” is defined as the count of sessions per leader where both mental toughness and Engagement totals were present. Missingness proportions are calculated over post-baseline opportunities (sessions 2–6; 60 rows). ICC was estimated from a null mixed model with a random intercept for person on Engagement totals.

3.1 RQ1: How did employee engagement change over the course of the mental toughness intervention?

Linear mixed-effects models (random intercepts and session effect) showed a significant upward trend in Engagement (standardized β ≈ +0.12 per session; 95% CI [+0.01, +0.23]; p = 0.036). On the raw scale, this corresponds to ≈ +0.6 points per session, or ≈ +3 points from Session 1 to Session 6. Because the Engagement total ranges from 12 to 60, a ~3-point increase across the program represents about 6% of the full scale range (approximately 0.25 points per item across the 12 items), suggesting a small-to-moderate practical improvement in self-reported engagement (Figure 1).

3.2 RQ1a: Which engagement dimensions contributed most?

To parallel RQ1, we estimated within-person time trends for each engagement subscale using fixed-effects regression (demeaned outcome on demeaned session), with cluster-robust SEs by person. In other words, does each subscale increase across sessions when controlling for all stable person characteristics? As shown in Table RQ1a (supplementary), Behavioral Engagement (BE) showed the clearest upward movement over time, with a small positive slope that reached conventional significance before multiplicity adjustment. Emotional (EE) and Cognitive (CE) slopes were positive but imprecise.

CE: +0.07 points per session, 95% CI [−0.06, 0.19], p = 0.280

EE: +0.11 points per session, 95% CI [−0.08, 0.30], p = 0.272

BE: +0.14 points per session, 95% CI [0.00, 0.27], p = 0.049

Standardized effects (SD units per session) were small (≈ 0.04–0.06), BE again the most consistent. With a Holm correction across 3 subscale tests, BE's p value would be trend-level (Holm-adjusted p ≈ 0.15), subscale inferences were therefore treated as exploratory.

3.3 RQ1b: Distribution shift and minimal clinically meaningful change

We assessed whether the distribution of Engagement shifted upward across sessions using session-wise descriptives (n, mean, median) and a distribution plot. For clinical interpretability, we defined MCID as ≥ +3 points improvement from each leader's baseline and evaluated the percentage achieving MCID at their last observed session (last non-missing value), irrespective of attendance. The session-wise distributions showed an upward shift over the program (see Supplementary Table 2). Overall, 4 of 12 leaders (33%) achieved the MCID at endpoint, the exact 95% binomial (Wilson) interval was 14% to 61%. Individual baseline, last score, change, and MCID status are listed in Supplementary Table 3.

Independent of attendance, about one-third of leaders achieved a clinically meaningful improvement in engagement by their last observation, consistent with the positive time effect reported under RQ1. Because endpoint analyses use one observation per person, we treat RQ1b as supportive evidence. The primary inference remains the within-person time trend.¹

3.4 RQ1c (exploratory mechanism): Does mental toughness at session t precede engagement at t + 1?

We tested a lagged, within-person association between MT at session t (MT_t) and subsequent engagement change at session t+1. The primary specification regressed ΔEng_{t+1} = Eng_{t+1} – Eng_t on MT_t, including person fixed effects (to remove all time-invariant differences between leaders) and session indicators (to absorb program-wide shocks). Standard errors were cluster-robust by person. Analyses used only observed adjacent session pairs (i.e., where both Eng_t and Eng_{t+1} were available), yielding n transitions = 16. Higher MT at session t predicted greater next-session engagement gains, though estimates were imprecise given the small number of observable transitions.

Primary lag model: ΔEng_{t+1} on MT_t = +0.30 points per 1 MT point, 95% CI [−0.09, +0.69] (n transitions = 16). Sensitivity (level-on-lag with adjustment): Eng_{t+1} on MT_t + Eng_t (plus person and session effects) = +0.32, 95% CI [−0.39, +1.04] (n transitions = 16).

Point estimates suggest that a 1-point higher MT at session t is followed by roughly +0.3 points higher engagement at t+1 (e.g., a +5-point MT rise would correspond to ~+1.5 engagement points next session). The estimate points in the expected (positive) direction, but uncertainty is large and the confidence interval includes zero; so, the effect is not definitive.

Regarding context and limitations, the lagged design strengthens temporal ordering relative to concurrent associations. However, (a) the analysis is underpowered (only 16 usable transitions due to intermittent missingness), (b) it assumes no unmeasured time-varying confounders that affect both MT and next-session engagement, and (c) the one-session lag may be short to fully capture transfer of MT skills. Therefore, our results encourage larger studies with planned cross-lagged or dynamic models and higher completion rates.

3.5 RQ2: What was the association between session attendance (dose) and within-person change in employee engagement?

Within-person change from baseline at each session (ΔEng_t = Eng_t – Eng_baseline) was regressed on cumulative prior sessions completed (lagged dose), adjusting for session and concurrent MT. We fit (a) the main specification using all available sessions and (b) a tightened specification using sessions 2–6. The results indicated dose was positively associated with engagement gains (Table 2).

Table 2. Main dose-response (ΔEng from baseline; cluster-robust SEs).

Moving from two to five sessions is associated with ~+4 points higher engagement at a given session; a moderate, operationally meaningful lift in this leadership context. On the 12–60 Engagement scale, this corresponds to roughly one-third of a point per item across the 12 items, indicating a noticeable but not large difference associated with greater program exposure (Figure 2).

3.6 RQ2a: Alternative operationalizations of dose

To test whether the dose-response signal depends on how “dose” is defined, we re-estimated the primary within-person model for ΔEngagement (ΔEng_t = Eng_t – Eng_baseline) substituting three dose metrics, each adjusted for session and concurrent MT with cluster-robust SEs by person. Recency-weighted dose (geometric decay 0.7): at time t, dose = 1.0 for the immediately prior session attended, 0.7 for two sessions prior, 0.7² for three prior, etc. Consecutive attendance streak (0–5) is the number of sessions attended in a row immediately before session t, and attendance rate (0–1) is the total sessions attended divided by 6. Redefining dose preserved the positive association with engagement gains, and the two time-varying specifications were statistically reliable. The results are shown in Supplementary Table 4.

Findings are interpreted as: (a) Recency, (b) Continuity, and (c) Between-person rate is directional. (a) Recency mattered. The recency-weighted index gives 1.0 credit for the immediately prior session, 0.7 for two sessions ago, etc. A one-unit increase in this index (e.g., attending the most recent session vs. not) is associated with +3.3 points higher ΔEng_t. Over the plausible observed range (~0–2.8 by Session 6), moving from no recent attendance to attending the last three sessions (1 + 0.7 + 0.49 ≈ 2.19) corresponds to a predicted gain of ~+7 points (consistent with a strong recency signal). (b) Continuity and accumulation. Beyond recency, engagement gains also increased with uninterrupted participation. Each additional consecutive prior session is associated with +1.09 points higher ΔEng_t. Compared with having no consecutive prior sessions, having a three-session attendance streak was associated with about a 3.3-point larger engagement gain, and a five-session streak with about a 5.5-point larger gain. Finally, (c) between-person rate is directional only. The attendance-rate coefficient is positive but imprecise (N = 12), reflecting limited between-person information. Primary inferences rely on within-person time-varying metrics above.

Overall, the dose-response pattern is not an artifact of a single definition of dose; it is strongest when dose captures recency and continuity, suggesting consistent, back-to-back attendance is especially consequential for engagement gains.

3.7 RQ2b: Sequentiality/recency

To test whether immediate continuity boosts engagement, we analyzed the first difference in engagement between adjacent sessions, Δ(Eng)_t = Eng_t – Eng_{t−1}, as the outcome and included a binary predictor indicating whether the prior session (t−1) was attended (i.e., both MT and Engagement totals recorded). This within-person specification removes all time-invariant person factors. Standard errors were cluster-robust by person. As a robustness check, we re-estimated the model including session fixed effects and concurrent MT; results were materially unchanged. When the prior session was attended, the immediate gain was +1.60 points larger (95% CI [+0.74, +2.46]). Leave-one-participant-out re-fits ranged from +1.37 to +1.96, indicating the result was not driven by any single leader. Back-to-back attendance yields an additional short-term lift of about 11/2 points on the engagement scale at the next measurement, over and above each person's own typical week-to-week change. This complements the cumulative dose models: attendance appears to matter both in aggregate (more sessions overall) and sequentially (recent continuity).²

3.8 RQ2c: threshold/minimal clinically important difference

To complement the continuous dose-response analyses, we examined a clinically interpretable threshold: achieving a minimal clinically important difference (MCID) defined a priori as ≥ +3 points improvement in Engagement total from each leader's baseline. We used each participant's last observed session as the endpoint (last non-missing value), and split participants into higher dose (≥ 4 total sessions) vs. lower dose ( ≤ 3 total sessions) based on total attendance across the program.

Due to small cell counts, we compared MCID attainment rates with Fisher's exact test and summarized the risk difference (RD) and risk ratio (RR) descriptively. At the endpoint, 43% of higher-dose leaders achieved the MCID vs. 20% of lower-dose leaders (Fisher's p = 0.58). The point estimates imply RD = +23 percentage points and RR ≈ 2.15 (0.43/0.20), but the precision is limited (wide 95% CIs encompassing the null), reflecting the small N and dichotomization.

The MCID analysis is directionally consistent with the primary dose-response models (leaders with more sessions were more likely to achieve a clinically meaningful improvement), but the study is underpowered for definitive binary comparisons. The endpoint uses only one observation per person and discards within-person information. Hence, we treat it as exploratory and supportive of the primary continuous analyses.³

3.9 RQ2d: Baseline moderation

We tested whether the attendance (dose) effect differed by baseline status. For each baseline variable [(a) baseline Engagement (Eng_baseline) and (b) baseline MT (MT_baseline)], we added a dose × baseline interaction to the primary within-person model: ΔEngt = αi + β1Doset−1 + β2Sessiont + β3MTt + γ(Doset−1 × Baseline) + εit, estimated with cluster-robust SEs (participant clusters). Because of small N, we obtained cluster-aware permutation p-values for the interaction by circularly shifting dose sequences within person (baseline values fixed) to generate the null. There was no evidence of moderation by baseline:

Dose × baseline Engagement: interaction not significant (permutation p ≈ 0.27).

Dose × baseline MT: interaction not significant (permutation p ≈ 0.27).

Simple associations between baseline and total attendance were essentially zero (Eng_baseline ↔ dose r ≈ −0.03; MT_baseline ↔ dose r ≈ +0.02), indicating little selection into higher attendance by starting level. Within the observed baseline range, the dose-response remained positive and of similar magnitude. We found no differential benefit restricted to initially low- or high-engagement, or low/high MT leaders. Given the small sample, wide interaction CIs were expected. While strong moderation is unlikely, subtle moderation cannot be ruled out and should be reassessed in a larger study.

All robustness checks targeted the same estimand as the primary RQ2 model (dose effect on within-person ΔEngagement adjusting for session and concurrent MT), under alternative assumptions and estimation strategies. Results were consistent across multiple inferential lenses (Table 3):

Table 3. Robustness checks.

All p values two-sided. Bootstrap/LOPO refit the same covariate structure as the primary model. Permutation nulls generated by within-person circular shifts of dose, holding outcomes and covariates fixed. Huber and winsorized models use ΔEng_t as the outcome with session and MT_t included. IPW used stabilized weights from pooled logistic models (denominator: Pr[complete_t | session, prior Eng, prior MT, baseline]; numerator: Pr[complete_t | session]).

Overall, the above findings reveal that across orthogonal strategies (resampling, randomization-based inference, robust M-estimation, Bayesian priors, and IP weighting), the dose effect remains positive and practically similar in magnitude. The only approach that meaningfully attenuates the point estimate (Huber) is intentionally conservative and directionality holds. A consolidated forest of all dose estimates is shown in Supplementary Figure 1.

3.10 NIH internal validity appraisal

Using the NIH Quality Assessment Tool for Before-After (Pre-Post) Studies With No Control Group, we rated overall internal validity as fair. Internal validity is supported by clear eligibility/enrollment, complete follow-up, prespecified and consistently assessed outcomes, and appropriate longitudinal analyses. However, confidence is limited by the small pilot sample, self-reported outcomes without blinded assessment, and a single baseline measurement (no pre-intervention trend). Item-level justifications and mitigation strategies are summarized in Supplementary Table 5.

4 Discussion

In high-pressure healthcare environments where senior leaders must remain engaged despite persistent operational, financial, and ethical demands, greater participation in the MT program was associated with larger improvements in engagement. Although MT and employee engagement are conceptually distinct, the pattern of results suggests strengthening MT may be associated with leaders' engagement at work. Our findings are consistent with this pattern, as engagement increased by approximately three points from baseline to the final session; a moderate improvement in a leadership population operating under constant pressure. Improvements to engagement appeared to vary with session attendance. Each additional session attended was associated with higher engagement. Leaders completing five sessions demonstrated approximately four points higher engagement than those completing only two. The findings were consistent generally consistent across analytic approaches, and engagement gains were not explained by differences in leaders' starting levels of engagement or attendance patterns. Importantly, observed improvements in engagement cannot be attributed solely to intervention content. The cohort-based format may have introduced additional mechanisms, including peer interaction, shared reflection, and increased psychological safety, which could independently or synergistically contribute to engagement gains. In practical terms, these effects were modest in absolute size, but they may still be meaningful for a brief leadership intervention in a high-pressure healthcare setting; the observed changes suggest a small improvement in how consistently leaders brought cognitive, emotional, and behavioral energy to their work over time. Given the small sample and single-group design, these findings should be interpreted as preliminary and hypothesis-generating. We also note that the term ‘mental toughness' is used here to reflect a set of developable psychological skills, rather than a normative expectation that leaders must be stoic, suppress distress, or function without support (See Supplementary Appendix A; Session 6 “dark side discussion”). Our framing is intended to emphasize trainable self-regulatory capacities that may complement, but not replace, organizational resources and structural efforts to reduce unnecessary stress.

Our findings extend healthcare scholarship by moving beyond documenting the pervasive stress inherent in healthcare work (Goh et al., 2019), toward developable internal psychological resources (Lee and Kim, 2023; Lin et al., 2017; Ruparel, 2020). Existing research has emphasized workload, burnout, and organizational constraints as drivers of disengagement in healthcare settings (Shanafelt et al., 2017; Solms et al., 2023), while engagement itself has been linked to patient safety, performance, and retention outcomes (Janes et al., 2021; Lowe, 2012). Much less work has examined whether leadership-focused mental skills training can influence engagement over time, particularly among senior healthcare leaders. To our knowledge, this study is among the first to demonstrate participation in a structured MT intervention is associated with engagement gains, positioning MT as a relevant psychological resource in persistently high-pressure healthcare organizational contexts.

Mental toughness may be especially significant for senior healthcare leaders because their roles require sustained engagement amid unavoidable and often complex trade-offs (Janitra et al., 2025; Mathew et al., 2024). Conceptually, MT reflects the capacity to remain committed to competing higher-order objectives, regulate thoughts and emotions under stress, and persist in goal-directed behavior despite adversity (Gucciardi et al., 2015; Gucciardi, 2017). In practice, senior leaders routinely confront situations in which financial constraints, staffing shortages, and operational pressures directly conflict with commitments to patient care and workforce wellbeing. Leaders may be required to reduce staff or services to preserve organizational viability, even when such decisions limit patient access or increase strain on remaining employees. Mental toughness may help leaders manage competing demands, regulate emotional responses, and remain engaged in high-stakes, complex situations (Gucciardi, 2017; Roslan et al., 2022). Mental toughness does not remove stressors from leadership roles, but may alter how leaders perceive them and respond to them, supporting continued functioning under persistent stress (Guraya et al., 2025; McClafferty et al., 2022; Turkington et al., 2023).

Engagement gains were most pronounced in BE, suggesting MT development may be related to how leaders perform their roles under pressure. Behavioral engagement showed the clearest and most consistent improvement across sessions, whereas cognitive and emotional engagement trended positive but were less precise. This pattern may align with the applied, skill-based nature of the intervention, which emphasized attention/emotional regulation, and persistence under stress. For senior healthcare leaders, sustaining BE may be particularly consequential in environments characterized by continuous accountability and high-stakes decision-making (Singh et al., 2024).

The magnitude of engagement gains depended not only on participation, but also on the consistency and recency of attendance. Leaders who attended sessions consecutively, or who had participated most recently, demonstrated larger engagement gains than those with more sporadic attendance. Such a pattern suggests MT skills may require reinforcement and continued practice to translate into sustained engagement, particularly in environments where stressors are ongoing (Albott et al., 2020; Gucciardi, 2017; Solms et al., 2023). The findings highlight the importance of program structure, indicating isolated or one-off interventions may be less effective than approaches emphasizing continuity and cumulative exposure (Davis et al., 1999; Rugulies et al., 2023; Sawyer et al., 2023; Schwarz et al., 2019).

Exploratory analyses further suggest MT may function as a psychological mechanism supporting continued engagement over time. Higher MT at one session tended to be followed by slightly higher engagement at the next session. Although imprecise due to limited complete session pairs, the direction of this association aligns with theoretical expectations that psychological resources support continued investment in work roles under stress (Joshi et al., 2022; Wang et al., 2023). The findings are suggestive rather than conclusive, but do provide preliminary justification. Future studies should use designs that better test how changes in MT and engagement influence one another over time.

The results suggest strengthening internal psychological capacities may help leaders remain engaged in environments where external demands cannot be readily reduced (Solms et al., 2023; Wittmers and Maier, 2023). Rather than focusing exclusively on stress reduction via structural change (i.e., focusing on the “environment”), MT development may offer a complementary approach by supporting leaders' ability to sustain engagement amid persistent pressure and competing demands (i.e., focusing on the “individual” and their interaction with the “environment”).

4.1 Practical implications

The findings provide a preliminary basis for considering several practical implications for healthcare organizations and leadership development initiatives. First, MT training may serve as a proactive complement to existing wellbeing and engagement strategies. Rather than focusing solely on reducing burnout or stress exposure, organizations may benefit from developing leaders' internal capacities to positively navigate persistent demands (Albott et al., 2020; Ho et al., 2024). Such approaches may be especially relevant for senior healthcare leaders, whose roles involve constant decision-making under pressure and accountability for organizational outcomes. Beyond developing psychological skills, group-based interventions may offer opportunities for leaders to engage in open dialogue with peers who face similar challenges, fostering psychological safety (Edmondson et al., 2016). When these experiences are supported by the organization, they may function as visible signals of care and investment, reinforcing perceptions of trust and meaningfulness that encourage engagement in demanding environments (O'Donovan et al., 2021; Opoku and Boateng, 2024). Importantly, this framing should not be interpreted as placing responsibility for engagement solely on individual leaders; mental skills training is best understood as one complement to broader organizational efforts to improve support, workload, and working conditions (Chen et al., 2025).

Second, structured MT interventions may represent a relatively low-cost and scalable option for leadership development. Integrating psychological skills training into existing leadership programs, coaching initiatives, or professional development curricula (including train-the-trainer approaches and hybrid or digitally augmented delivery formats) may support leaders in maintaining engagement amid ongoing organizational challenges (Janitra et al., 2025; Solms et al., 2023; Lee et al., 2019). This may be particularly valuable in healthcare settings facing staffing shortages, high turnover, and increasing performance expectations (Mangurian et al., 2023; Rosa et al., 2022). For healthcare organizations, embedding MT development into leadership programs represents a proactive investment in human capital signaling organizational support (Ho et al., 2024). By equipping leaders to model their psychological MT training, such approaches may contribute to broader organizational effects through structural reforms. Such effects (e.g., improvements in leader effectiveness and team morale) may have possible downstream implications for patient care outcomes (Albott et al., 2020). However, the present study did not assess downstream, employee-level outcomes, and we do not assume that leaders will actively promote or transmit mental toughness to their teams. Any potential cascading effects should therefore be considered speculative and require direct examination in future research.

Third, focusing on psychological MT skills may help organizations move beyond reacting to burnout, and toward supporting leaders in managing ongoing demands more effectively. Viewing engagement as malleable and fluid may encourage organizations to invest equitable ongoing dedicated resources in leadership practices that strengthen meaning, safety, and access to the cognitive, emotional, and physical resources needed to fully engage (Kahn, 1990; Mangurian et al., 2023).

Fourth, the structured and manualized six-session cohort (see Supplementary Appendix A) provides a foundation for sustainability and scalability, but effectiveness likely depends on maintaining continuity and fidelity (Cox et al., 2019; Páez et al., 2025). The observed dose-response pattern, particularly the importance of consecutive attendance and recency; (e.g., Erekson et al., 2015; Gutner et al., 2016), suggests benefits arise from repeated, reinforced practice rather than isolated exposure (e.g., Cheng et al., 2018; Walsh et al., 2023). Accordingly, sustainable delivery may require embedding the program within recurring leadership cycles (e.g., quarterly cohorts) with opportunities for between-session application (e.g., Bodkin and Hakimi, 2020).

Scaling may be achieved through train-the-trainer models and integration into existing organizational learning systems (e.g., leadership academies, onboarding; e.g., Klaic et al., 2022), though attention to fidelity is critical, as reduced continuity may attenuate effects (e.g., Tiemens et al., 2019). Digital and AI-enabled approaches offer additional scalability: personalized mobile applications could deliver adaptive prompts, real-time reflections, and context-sensitive skill reinforcement, while AI systems tailor content based on engagement and prior responses. Hybrid models combining cohort delivery with digital augmentation may extend reach while preserving the relational elements central to the intervention (Sriharan et al., 2024; Tsekouropoulos et al., 2025). These approaches were not evaluated here but represent a key direction for future research.

4.2 Theoretical implications

This exploratory, hypothesis-generating study brings Shuck et al. (2017a) multidimensional model of employee engagement into dialogue with Gucciardi (2017) unidimensional model of mental toughness. Both conceptualize optimal functioning through integrated cognitive, emotional, and behavioral processes, and our findings suggest these frameworks may be mutually reinforcing. Engagement describes the psychological state through which individuals invest energy in their roles (Lee and Shuck, 2024), while MT represents the capacity to sustain that investment when faced with stress, uncertainty, or fatigue (Gucciardi et al., 2015). Together, they may form a complementary system in which MT is related to the maintenance of engagement under strain. This relationship advances theoretical understanding by identifying a trainable psychological resource that may buffer disengagement and foster improved performance.

The present findings also extend engagement scholarship by illustrating how a structured psychological skills intervention may activate the core psychological conditions underlying employee engagement. Kahn (1990) proposed individuals fully engage when they experience their roles as meaningful, feel psychologically safe to express themselves, and possess sufficient cognitive, emotional, and physical resources to invest in their roles. This framework provides the conceptual foundation for Shuck et al. (2017a) multidimensional model. Our findings suggest MT interventions may be related to all three of Kahn's (1990) psychological conditions. Meaningfulness may be strengthened as leaders align goals and values with daily behaviors, assuming organizational alignment is reached (McClafferty et al., 2022; Rosa et al., 2022). Safety may be fostered through structured group sessions, normalizing shared challenges, and supporting open expression (Edmondson et al., 2016; O'Donovan et al., 2021). Availability may be enhanced by developing skills related to emotion/attention regulation, and persistence under pressure, increasing access to the resources required for engagement, assuming those physical and emotional resources are at an adequate level (Turkington et al., 2023).

5 Limitations

Several limitations should be considered when interpreting these findings. First, the small sample size limits statistical power and the generalizability of the results. As such, the findings should be interpreted as exploratory rather than confirmatory. Second, the absence of a control or comparison group precludes causal inference, as improvements in engagement may have been influenced by unmeasured factors such as organizational changes, time effects, or external events. Third, the study design did not allow us to distinguish whether changes in engagement were associated with the intervention content, peer interaction, or a combination of both, as participant interactions were not controlled for within the program structure. Fourth, reliance on self-report measures introduces the potential for response bias, including social desirability due to participation in the intervention. Additionally, attendance varied across sessions, introducing potential bias in the observed dose-response effects and contributed to missing data. In more detail, because session non-completion may have been related to workload, availability, or unmeasured engagement, some selection bias remains possible; for example, if leaders who were more available or responding more favorably to the program were also more likely to complete later sessions, the observed time and dose effects may be somewhat overstated, although sensitivity analyses were directionally consistent with the primary results. Furthermore, the study focused exclusively on senior healthcare leaders, limiting the extent to which findings can be applied to other healthcare roles (e.g., mid-level management) or occupational contexts. Finally, self-reported outcomes without blinded assessment, and a single baseline measurement (no pre-intervention trend) affected further the internal validity of our study.

6 Future research directions

Future research should build on these preliminary findings by employing larger, multi-site samples to enhance generalizability. Experimental designs, including randomized controlled trials, would strengthen causal inference regarding the effects of MT interventions on engagement (Leon et al., 2011). Qualitative approaches might explore how participants perceive and apply MT strategies in their daily leadership behaviors. Longitudinal research is also needed to confirm durability and generalizability of effects (Calo et al., 2024; Kunzler et al., 2020).

Although the study focused on senior healthcare leaders, future research may examine how the development of mental toughness may extend to broader workforce populations and contribute to resilience across organizational levels. Subsequent research may benefit from incorporating objective performance indicators—such as turnover, retention, and patient satisfaction scores—to better understand how psychological resource development relates to organizational and care-related outcomes. Exploring downstream outcomes such as team engagement, staff retention, and organizational performance would clarify the practical impact of MT interventions in healthcare. Furthermore, future studies may explore moderators such as role type, workload intensity, or organizational support, as well as compare MT interventions with other psychological resource-based approaches, including other positive psychology constructs, such as self-compassion or grit (Stamatis et al., 2020b, 2024).

Finally, addressing both non-blinded, self-reported outcomes and a single pre-intervention baseline would strengthen internal validity and shift interpretation from associational, within-person improvements consistent with the intervention to more credible causal evidence that the intervention produced quantifiable changes in engagement (i.e., an estimate of effect magnitude above and beyond baseline trends and other time-varying influences).

7 Conclusion

The pilot study is the first to provide preliminary evidence of an association between participation in a MT intervention and improved engagement among senior healthcare leaders. Engagement increased over time, and leaders with greater and more consistent participation showed the largest gains. Improvements were strongest in BE, highlighting the role of MT in supporting leaders' ability to act and persist under pressure. Although the sample was small, the same dose-response pattern appeared across different analyses, increasing confidence the association was not driven by analytic choices. The findings suggest MT may offer a practical, scalable approach to supporting engagement in healthcare leadership. While exploratory and hypothesis-generating, these findings justify continued investigation in larger, controlled, and multi-site studies and support integrating psychological skills into leadership development and wellbeing initiatives. In healthcare systems where demands are persistent and trade-offs unavoidable, supporting leaders' capacity to remain engaged may be as critical as efforts to reduce stress itself.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. Portions of the manuscript were revised with the assistance of AI-based editing tools (ChatGPT versions 5, 5.1, and 5.2) to improve clarity, grammar, and organization. All conceptual content, theoretical framing, interpretations, and arguments are the author(s)' own. The author(s) verify and take full responsibility for the use of generative AI in the preparation of this manuscript. Generative AI was used solely for language refinement and editorial support and did not contribute to the study design, data collection, data analysis, interpretation of results, or generation of original scientific content.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/forgp.2026.1787997/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Gabriele Giorgi, European University of Rome, Italy

Reviewed by: Zi Jian Oh, Kolej Vokasional Batu Lanchang, Malaysia

Cheryl Connors, Johns Hopkins University, United States

Footnotes

Because anchor-based criteria were unavailable in this pilot, we adopted a distribution-based MCID. The baseline SD of engagement in this cohort was 4.19; we defined MCID as ≥ +3 points, corresponding to ≈0.72 SD. This exceeds common 0.5-SD heuristics used in patient-reported outcome research, providing a conservative threshold (Mouelhi et al., 2020; Ousmen et al., 2018; Revicki et al., 2008). Directional conclusions were similar when varying the cut point by ±1 point.

Because Δ(Eng)_t requires Eng_{t−1}, analyses used only session pairs with both observations available; this is a within-person restriction but does not change the estimand. Results are descriptive/associational and interpreted as hypothesis-generating.

Results were similar in direction when (a) varying the MCID cut point (e.g., ±1 point), and (b) using last-observation-carried-forward in a conservative sensitivity, though absolute rates changed as expected. Baseline engagement and mental toughness were not associated with total attendance, and dose × baseline interactions were non-significant, reducing concerns that baseline severity explains the observed pattern.