Differences Between Men and Women in Stress Hormone Changes, Stress Levels, and School Performance in Quarantined Foreign Medical Students During COVID-19 Lockdown

Salivary cortisol is frequently used as a stress biomarker. Under physiological conditions, it exhibits a circadian pattern of secretion, with a peak in the morning—corresponding to waking—and serves an essential function of synchronizing peripheral clocks with the central molecular clock in the suprachiasmatic nucleus of the hypothalamus [14]. Salivary cortisol measurement is a common non-invasive way to determine daily fluctuations [15]. There is a high correlation between salivary cortisol levels and free cortisol levels in plasma and serum, which remains high throughout the circadian cycle and during dynamic tests [16]. Salivary cortisol and α-amylase are also associated with cardiovascular and cardiometabolic risks, with dynamic changes and altered secretion patterns serving as subclinical indicators of stress [17].

The SARS-CoV-2 pandemic (COVID-19) and lockdown which included isolation and social distancing were significant stressors that brought people to novel, unpredictable and uncertain living situation [18,19]. According to the World Health Organization (WHO), the COVID-19 outbreak has resulted in the death of more than 380,000 people worldwide by the first days of June 2020. Final numbers for 2 November 2023, are 6,934,072 deaths [20]. Social isolation, loss of job and everyday life routine, an increase in conflicts inside the family due to 24-h staying together, separation from close people and even their death, and excessive consumption of alcohol are just several factors that affected people's behavior and physiology during lockdown. All these factors, including the constant fear for the health of oneself and loved ones, strained mental health, and during COVID-19 pandemics significantly higher prevalence of depression, anxiety, insomnia, PTSD, and psychological distress were found regardless of sex, group or region [21]. The COVID-19 pandemic has affected women in unique sex-specific ways as home managers. The COVID-19 lockdown has intensified women's domestic workload, and remote work has contributed to severe challenges [22].

Due to the worldwide COVID-19 outbreak, the Ukrainian government declared a state of emergency on 12 March 2020, and a few days later, strict lockdown measures were implemented to prevent the spread of the virus. The measures included a ban on gatherings in public places and across regional borders, the closure of schools, universities and all non-essential businesses, and the instruction to work from home with very few exceptions. At that time two hundred foreign medical students of I. Horbachevsky Ternopil National Medical University, Ternopil, Ukraine (TNMU) (first to sixth year of study) was isolated within dormitory facilities and were not allowed to exit, including for walks or other outdoor activities. Self-isolation and "stay at home" measures were implemented nationwide and communicated through public officials and various communication channels.

This cross-sectional study aimed to investigate sex-specific patterns in daily salivary cortisol fluctuation and perceived stress, including physical and behavioral symptoms, among foreign students of TMNU related to the COVID-19 lockdown. Given the unique circumstances in which all participants were confined to the same dormitory in a foreign country, without access to external support systems (such as immediate family support), the sample provided a controlled setting for examining stress-related physiological and psychological responses. We hypothesized that sex would be associated with differences in both physiological (salivary cortisol) and psychological (perceived stress and related symptoms) responses to the lockdown environment. In a subsequent analysis, we explored the academic performance of these students during pre-COVID (2018/2019), COVID (2019/2020), and post-COVID academic year (2020/2021).

A cross-sectional study was conducted in the second part of the May 2020, during academic year 2019/2020 and Ukrainian lockdown that had already lasted 2 months, with no indication of when it might end. The study lasted over a whole day, from waking up until bedtime, during which participants collected saliva samples and completed questionnaires (self-assessments of stress experiences and health issues). The study was approved by the Bioethics Committee of TNMU (Protocol No. 59 from 29 April 2020). The study was performed based on the principles of Declaration of Helsinki (2013). Informed consent was received from every participant.

The study primarily involved 200 foreign medical students enrolled in the first through sixth years at TNMU, located in Ternopil, Ukraine. Inclusion criteria for participants were as follows: compliance with lockdown requirements, signed informed consent, agreement to saliva collection, and the presence of all five requested samples. Exclusion criteria were acute illness including COVID-19 or exacerbation of chronic disease. Nationality was recorded for all participants, but it was not included in the subsequent analyses. For female participants, menstrual cycle phase was not controlled during sampling.

Saliva was self-sampled by participants according to precise and detailed guidelines provided by the research team—in written form and via online meeting, ensuring compliance with the study protocol and standardized procedures across the sample. Sampling was performed 5 times over 1 day using Salivette (Sarstedt, Nümbrecht, Germany). Sample collection was conducted from 07:00–09:00 (t₁) and continued in 4-h increments (11:00–13:00 (t₂), 15:00–17:00 (t₃), 19:00–21:00 (t₄), 23:00–01:00 (t₅)). All participants were instructed to adhere to dietary and activity restrictions prior to sampling. They were carefully advised to gently chew on the provided cotton roll to stimulate saliva flow rate and to follow the time frames for saliva collection; those who did not provide a sample during the morning collection window, when participants were expected to be awake due to scheduled academic obligations, were not taken into consideration for the research. All students who provided samples were recruited in this manner, irrespective of their habitual circadian rhythms. Samples were analyzed for cortisol content (ng/mL) using the Cortisol ELISA Kit (Abcam, Cambridge, UK; ab154996) and microplate photometer (Multiscan FC-357, Thermo Fisher Scientific, Waltham, MA, USA).

The second part of the study included the self-reported questionnaire "Perceived Stress Scale-10" (PSS-10) [31] and an unvalidated complement questionnaire based on the The Kaiser Family Foundation (KFF) Health Tracking Poll—July 2020 [32] to assess the impact of chronic stress on lifestyle and health issues during the previous two months.

The PSS-10 is a widely used psychological instrument for measuring perceived stress. Developed by Sheldon Cohen and colleagues in 1983 [31], it evaluates the extent to which individuals perceive their lives as being unpredictable, beyond their control, and excessively demanding [33,34,35,36]. It is validated across diverse populations and cultures and correlates with health outcomes, such as cortisol levels, sleep quality, and mental health [37]. The PSS-10 contains 10 items rated on a 5-point Likert scale (0 = never to 4 = very often). Items reflect feelings and thoughts during the last month. The PSS-10 total score ranges from 0 to 40, with higher values reflecting greater perceived stress. Based on the total score, stress levels can be categorized as follows: 0–13 (Low), 14–26 (Moderate), and 27–40 (High). The data completeness for the PSS-10 was 99.6%, with only 7 individual items missing across the entire dataset of 177 participants. Specifically, missing responses occurred on item 5 (n = 2), item 6 (n = 1), item 7 (n = 1), and item 8 (n = 3). To maintain the full sample size (N = 177), subscale-based pro-rating was applied, where missing values were replaced by the mean of the available items within the respective subscale (Perceived Helplessness for item 6; Self-Efficacy for items 5, 7, and 8).

The analysis also covered the two subscales of the PSS-10: Perceived helplessness (PH) (items 1, 2, 3, 6, 9, 10)—measuring an individual's feelings of a lack of control over their circumstances or their own emotions or reactions; and Lack of self-efficacy (LSE) (items 4, 5, 7, 8)—measuring an individual's perceived inability to handle problems.

The questionnaire based on the KFF Health Tracking Poll was used as a complement exploratory tool to the PSS-10. It is a non-validated, expert-designed questionnaire developed for the U.S. public opinion research context [32]. It consists of 10 binary (yes/no) items designed to capture self-reported physical and behavioral symptoms potentially related to stress, such as sleep disturbances, appetite changes, headaches, difficulties with emotional regulation, and increased substance use. Although not standardized for clinical use, it provides several useful insights into population-level trends in perceived health and well-being, and offers a broader perspective on how stress may manifest through somatic and behavioral symptoms in everyday life [38]. Its inclusion enables the identification of symptom patterns that might otherwise go undetected, especially in non-clinical populations.

Academic performance data (grade point averages) were obtained directly from the official records of the TNMU, ensuring that the results reflect verified institutional outcomes. The grades were obtained for each participant for three academic years: the year preceding the study—2018/2019 (pre-COVID), the year during which the study was conducted—2019/2020 (COVID), and the subsequent academic year—2020/2021 (post-COVID). It should be noted that academic performance data for the 2018/2019 academic year were not available for all participants, as some students were in their first year of study during that period. This was only exploratory analysis, and we find that any correlation (i.e., Spearman's) to cortisol daily fluctuation pattern is not justified due to cross-sectional nature of the study, and we got insight into cortisol daily fluctuation during only one day.

Statistical analysis of survey results was performed using Prism software, version 10.6.1 (GraphPad Software LLC, Boston, MA, USA). Quantitative parameters were expressed according to their distributions, assessed using the Shapiro–Wilk test—variables with normal distribution were presented as mean ± standard deviation (Mean ± SD), whereas those with non-normal distribution were presented as median (lower quartile; upper quartile) [Me (Lq; Uq)]. Qualitative parameters were summarized as frequencies and percentages. Regarding evaluation of cluster validity, construct validity was first established on the entire aggregated sample using EFA, providing a robust common structure. As the study's primary objective is descriptive, cluster-based sensitivity was addressed by comparing the mean scores derived from this validated factor structure across our defined clusters using ANOVA, rather than pursuing complex measurement invariance modeling. Comparisons among three or more normally distributed groups were performed using ANOVA with Tukey's post hoc test. For non-normally distributed variables across three or more groups, the Kruskal–Wallis test was applied. Two independent groups with normal distribution were compared using the t-test, whereas those with non-normal distribution were compared using the Mann–Whitney U test. Two dependent groups with non-normal distribution were compared using Wilcoxon matched-pairs signed rank test. Associations between categorical variables were assessed using χ² test for comparisons between sexes and Fisher's exact test in comparisons between clusters due to small sample size of clusters. Correlations were assessed using Spearman's nonparametric correlation. Statistical significance was set at p < 0.05. Effect sizes were reported for all significant comparisons: Cohen's D for t-tests, Vargha-Delaney A for Mann-Whitney U tests, Cramer's V for Fisher's exact tests, and rank-biserial correlation (r) for Wilcoxon matched-pairs tests. Post-hoc power analysis revealed that the achieved power (1 − β) for primary sex-based comparisons ranged from 0.65 to 0.75, while for the cluster-based Kruskal-Wallis analysis, it exceeded 0.99. For within-group academic success comparisons with smaller effect sizes (r approximately 0.2), the achieved power was in the range of 0.3 to 0.4.

Salivary cortisol was sampled five times throughout the day, starting at awakening (at 07:00–09:00) and ending at bedtime (at 23:00–01:00). The characteristics of the cortisol curve for all participants, including its AUC, amplitude, slope, and diurnal variation, were analyzed. Furthermore, sex-specific analyses were performed to examine potential differences in cortisol curve characteristics between male and female participants. There were no statistically significant sex-specific differences in AUC, amplitude or slope of the salivary cortisol curve (Figure 2). When observing the cortisol curve of all participants together, the highest salivary cortisol concentration was recorded at the second measurement point (at 11:00–13:00), which represents a shift from the typical cortisol pattern, where the peak concentration occurs in the early morning before awakening (corresponding to the first measurement point or even earlier) and gradually decreases throughout the day. When examining cortisol levels by sex, higher concentrations were observed in males. Statistically significant differences between males and females were found at the third and fourth measurement points, where males had higher cortisol levels (t₃: p = 0.021, U = 3130, t₄: p = 0.009, U = 3025) (Figure 2).

The construct validity of the PSS-10 scale was assessed using Exploratory Factor Analysis (EFA). KMO and Bartlett's test of Sphericity confirmed adequacy of the data for EFA. KMO was 0.787, and Bartlett's Test of Sphericity was significant (χ² = 496, df = 45, p < 0.001). EFA was conducted using PAF with Varimax rotation. Two factors were identified (factor 1 including items from subscale PH and factor 2 including items from subscale LSE) and their loadings are shown in Table 1. Those two factors together explained 44.3% of the total variance.

The average total PSS-10 score was significantly higher for women than men (Table 2). 46.7% of males and 63.2% of females had total PSS-10 scores which correspond to moderate levels of perceived stress. Low perceived stress was reported by 51.1% of men and 32.2% of women, whereas high perceived stress was reported by 2.2% of men and 4.6% of women. Total PH-score of females was significantly higher than the one in males, while LSE did not differ between sexes (Table 2). The overall score of items 2 and 9, which was significantly higher in females, contributed to the sex-specificity in the PH subscale (Table 3).

After salivary cortisol-based clustering, mentioned stratification in magnitude of perceived stress was altered—these scores indicated that respondents from male clusters with predicted daily cortisol pattern (π₁ and π₂) and from male π₅-cluster were perceiving low stress levels and all female clusters were perceiving moderate levels of stress (Table 2). Sex-specificity in PH score was lost after clustering.

When the distribution of responses to the PSS-10 items was compared between males altogether and females altogether, no sex-specific difference was found (Figure A2). After salivary cortisol-based clustering, few sex-specific differences in the distribution of responses to PSS-10 were observed, localized within the PH subscale (items 1, 2, and 9) where females had significantly differently distributed responses resulting in higher scores (Table 3). Cluster-based approach elucidated a minor sex-specific difference within the LSE subscale; specifically, males from π₂ scored higher on item 5 (tending toward "often") than females from the same cluster (tending toward "sometimes") (Table 3).

In the context of stress experience research, another sex-specific difference emerged among the clusters: cortisol AUC significantly correlated with PH scores within female clusters exhibiting aberrant daily cortisol patterns. Specifically, this correlation was moderate positive for the π₃ cluster (Spearman's r = 0.56, p = 0.039, r² = 0.31), whereas it was strong negative for the π₅ cluster (Spearman's r = −0.79, p = 0.048, r² = 0.62). Conversely, no such associations were observed among male clusters. Due to its small sample size, the female π₄ cluster was excluded from this analysis.

The KFF Health Tracking Poll served as a complement exploratory tool to the PSS-10 to assess perceived health and well-being of the participants, and it yielded several noteworthy insights. There was no statistically significant difference in the frequency of answers between sexes or clusters; therefore, answers were analyzed together for all participants (Table 4). The table shows the frequency of symptoms related to stress, emotional, and physical difficulties. Most participants do not report symptoms, but there is a large proportion who still report challenges (around one third of all participants).

The overall academic success of respondents was obtained from the official records of the TNMU. It was analyzed in the academic year preceding this study (2018/2019, pre-COVID), in the academic year during which the study was conducted (2019/2020, COVID), and in the year after (2020/2021, post-COVID). The data is shown in Table A1. Data for the pre-COVID academic year was not available for respondents who were in their first year of study at that time. The goal of this analysis was to assess potential changes or trends in academic performance over time that could have been driven by lockdown-induced stress.

Our findings reveal a notable upward trend in grades during and after the pandemic period. Females showed a significant improvement in grades in COVID year compared to pre-COVID (p = 0.022), with further gains observed in the post-COVID year (vs. COVID year: p = 0.025, vs. pre-COVID year: p < 0.001). Males had significantly better grades in post-COVID year compared to pre-COVID year (p < 0.001).

After cortisol-based clustering of the participants, significant changes were observed only within π₁-clusters: in comparison pre-COVID vs. COVID year, only females showed significantly better grades (p = 0.009), suggesting sex-specificity in academic success; in comparison COVID vs. post-COVID year, both males and females within the π₁-cluster showed significant improvement in grades (males: p = 0.033, females: 0.004); and comparison of pre-COVID vs. post-COVID year revealed that both males and females within the π1-cluster had significantly better grades (males: p = 0.008, females: p < 0.001).

Our comparative analysis revealed significant sex-specific differences in physiological stress markers. Specifically, male participants exhibited significantly higher cortisol levels compared to females, with statistical significance reached at the third (15:00–17:00 h) and fourth (19:00–21:00 h) sampling points. This finding is in accordance with previous reports [12]. While initial observations of the aggregate data showed an unexpected cortisol peak between 11:00 and 13:00 h for all participants, mathematical clustering was employed to further investigate this phenomenon. This modeling revealed five distinct secretory patterns, suggesting that while two-thirds of the cohort (around 70% of participants from clusters π₁ and π₂) followed a conventional diurnal rhythm [2], the remaining one-third displayed atypical patterns that distorted the group's mean curve. These findings highlight that robust underlying variability exists within both sexes—a phenomenon well-documented in the literature regarding sex-specific HPA axis responses [11,42,43], which may be further influenced by methodological factors such as the lack of control for individual 'morningness' [44] or the menstrual cycle phase in female participants [45]. Furthermore, given the cross-sectional nature of this study, it is not possible to determine pre-lockdown cortisol dynamics or the specific longitudinal stressors to which these individuals were exposed. While our clustering revealed that approximately one-third of the cohort exhibited atypical cortisol fluctuations, it remains unclear how these proportions compare to pre-pandemic conditions or whether these patterns represent a transient or a chronic adaptation to the lockdown environment.

Analysis of the PSS-10 subscales provided further insight into the psychological architecture of the students' stress. The total PSS-10 score falls within the expected range based on age-related norms reported in the literature [46]. Both sexes operated within the moderate stress range, which may reflect the preserved structure and cohesion of the student community during lockdown. The availability of virtual communication and continuity of daily routines likely mitigated the psychological impact, preventing the disintegration of social structure typically observed in more chaotic scenarios [47]. However, male participants reported significantly lower total PSS-10 scores and lower Perceived Helplessness (PH) compared to females. This sex-specific pattern may reflect differences in stress appraisal or coping mechanisms. Epidemiological studies consistently show that women are more frequently diagnosed with anxiety and depression than men, with female-to-male prevalence ratios often reaching 2:1. Sex differences in coping strategies may partly explain this disparity, as women tend to rely more on emotion-focused coping—such as rumination and self-blame—which is associated with higher levels of psychological distress. In contrast, men are more likely to engage in problem-focused coping [48,49,50]. The magnitude of these differences was supported by small-to-medium effect sizes, suggesting that the sex-specific divergence in emotional vulnerability is a robust feature of this population's response to lockdown. These findings indicate that despite their higher physiological cortisol output, male students maintained a significantly stronger sense of control. In contrast, the absence of significant differences in Lack of Self-Efficacy (LSE) across sexes and clusters suggests that the pandemic exerted a uniform pressure on the students' confidence in their personal coping abilities, regardless of their biological profile or perceived helplessness.

To provide a broader context for the students' wellbeing, an exploratory KFF Health Tracking poll was used to screen for common somatic and behavioral symptoms across the entire sample. The descriptive results for all participants revealed that approximately one-third of them experienced challenges. These observations suggest that the moderate psychological stress captured by the PSS-10 was accompanied by tangible physical and emotional toll for a significant portion of the student population, reinforcing the need for holistic support during prolonged periods of isolation.

A further exploratory analysis within the female clusters suggested divergent associations between cortisol output and psychological distress. Specifically, in the π₃ cluster, cortisol output was found to correlate positively with PH, whereas the π₅ cluster showed a strong inverse relationship. While these results must be interpreted with extreme caution due to the limited number of participants in these subgroups, they offer an interesting insight into possible individual differences in stress adaptation. It is well-recognized that these differences arise from the complex interplay genetic predispositions and individual life histories, which 'program' the HPA axis and shape how an individual perceives and biologically integrates environmental challenges [51]. Furthermore, since the phase of the menstrual cycle was not controlled, it remains a source of non-modeled biological variability that could influence individual cortisol output and its correlation with perceived stress scores. Nevertheless, this divergence might point toward a 'decoupling'—a separation of physiological and psychological stress responses—within the π₅ subgroup. These specific associations were notably absent in all male clusters, suggesting the possibility that this biological-psychological interaction might be more pronounced in females under the specific conditions of the lockdown.

The longitudinal look at academic performances (which were verified institutional outcomes, should be considered exploratory, as academic success was not directly correlated with other psychological or physiological variables in this study. The observed upward trend in grades—most evident in the larger π₁ clusters—suggests a generalized resilience, though it may also reflect altered assessment criteria during the transition to online education, as suggested by Gonzales et al. [52]. Furthermore, while the cortisol-based clustering provided a unique framework for analysis, it is important to acknowledge that our one-day sampling protocol may not fully capture the complete diurnal cortisol patterns throughout the prolonged lockdown period. Despite these limitations, the fact that students maintained or improved their academic standing amidst reported somatic symptoms and reported stress scores points toward a notable capacity for adaptation. This highlights the complex relationship between biological stress profiles and functional outcomes in high-pressure environments like medical school [53].

This study has limitations that should be acknowledged. First, its cross-sectional design limits the ability to draw causal inferences about the relationship between stress-related variables and academic performance. Longitudinal data on stress biomarkers and psychological scores would be necessary to determine the directionality and stability of these relationships over time. Second, the lack of pre-lockdown cortisol measurements prevents direct comparison with baseline physiological stress levels, making it difficult to assess the specific impact of the lockdown period. Third, although the sample size was sufficient for the analyses conducted, the sample was particular, comprising foreign students at a single university during a unique global event, which limits the generalizability of the findings to broader student populations. Future research should aim to include more diverse and representative samples, as well as longitudinal designs that capture stress dynamics before, during, and after major stressors. And, lastly, menstrual cycle phase was not controlled among female participants. The luteal phase is known to influence cortisol secretion [43], and its variability could have contributed to interindividual differences observed in the data. However, the aim of this study was not to examine phase-specific effects but rather to identify overall patterns of cortisol variation. Future research should consider controlling or recording cycle phase to reduce potential bias and improve interpretability.