Follow-up Study on the Sleep Status of Anti-Epidemic Staff: A Study Based on Wearable Sleep Trackers

Introduction

Since its emergence in December 2019, the COVID-19 pandemic has considerably influenced the mental wellbeing of healthcare workers.¹ Kang et al conducted a survey of 994 members who worked in Wuhan during the pandemic and found that 22.4% suffered from moderate to severe depression, anxiety problems, with 6.2% reaching a severe level. The situation not only posed additional risks to the healthcare staff, but also reduced the efficiency of assistance work.

Occupational stress among healthcare workers is a well-documented phenomenon, particularly during infectious disease outbreaks, and has been associated with adverse effects on sleep health. However, the longitudinal trajectory of sleep changes after such acute stress exposure remains less well-characterized. While our group previously published a cross-sectional analysis of this cohort focusing on stress and sleep apnea risk, the present study uniquely extends those findings by providing one-year follow-up data to examine how sleep patterns evolve over time after the acute stress phase. We hypothesized that total sleep time would decrease from the immediate post-deployment period to the one-year follow-up, reflecting a transition from a stress-recovery phase to a chronic adaptation phase, and sleep architecture disturbances would persist at follow-up, manifesting as altered proportions of sleep stages.²

In situations involving stress, individuals may experience sleep suppression and increased wakefulness, which can lead to sleep disorders such as insomnia (characterized by difficulty falling asleep, maintaining sleep, and early awakening), excessive daytime sleepiness, and nightmares.^3,4 In turn, insomnia, poor sleep quality, and excessive sleep duration may also be risk factors for developing COVID-19 infection.⁵ Despite the recognized impact of stress on sleep, there is a need for convenient and scalable methods for sleep screening in highly demanding occupational settings. Wearable devices, such as pulse oximeters, offer a potential solution due to their ease of use and ability to collect data in naturalistic environments, although their capability to accurately resolve sleep architecture compared to polysomnography (PSG) is limited and requires further validation.⁶

From complementary psychological and physiological perspectives, this study examines the impact of COVID-19-related stress on sleep and breathing in healthcare personnel, comparing findings from a post-deployment assessment with a one-year follow-up and a control group. We also integrated the assessment of sleep apnea-hypopnea syndrome (SAHS) risk, as stress-related sleep disturbances and sleep-disordered breathing may coexist and interact, compounding health risks in this population.

Methods

Study Design and Participants

This was a mixed-design study combining a longitudinal follow-up of an exposed cohort with a cross-sectional control group comparison. The anti-epidemic group included healthcare staff who 1) participated in the epidemic prevention and control and 2) had worked at designated hospitals or makeshift hospitals in Wuhan from January to March 2020. The subjects completed the first sleep monitoring and psychological assessment in April 2020, during the health observation period after completing COVID-19 control tasks in Wuhan (hereafter referred to as the “post-deployment assessment”), and completed the second follow-up in May 2021. The control group consisted of healthcare staff recruited from various hospitals in Shanghai from July 2020 to July 2021 who did not participate in the COVID-19 control work. Clinical assessments were completed by psychiatrists from the Shanghai medical team that provided psychological support to Wuhan, Hubei.

All subjects had not taken hypnotic medications within one month and signed an informed consent. Subjects who were not infected with COVID-19 during the study period (2020–2021) were included to exclude the direct physiological effects of the infection. Participants with the following conditions were excluded:

• Severe heart, liver, kidney, or hematopoietic system disease, severe brain organic diseases, and mental illness.
• A history of alcohol abuse.
• Women who were pregnant or breastfeeding.

Pulse Oximeter Device and Sleep Monitoring

Overnight sleep monitoring was performed at the participants’ residences to approximate natural sleeping conditions. A ring-shaped medical pulse oximeter (Wellue O2Ring, Viatom Technology Co., Ltd., Shenzhen, China) was worn on the finger throughout the night. The device measures peripheral oxygen saturation (SpO2) and pulse rate. The derived sleep metrics provided by the device’s proprietary algorithm were recorded. It is critical to note that this device estimates sleep stages (light, deep, REM) based on autonomic parameters such as heart rate variability and body movement, not from electroencephalography (EEG). Therefore, all sleep architecture findings in this study should be considered exploratory, device-estimated data. Validation references for this specific device’s sleep staging performance against PSG are needed; in their absence, results are interpreted with caution.⁶

Variables and Measurements

Pulse rate and blood oxygen level parameters included the oxygen desaturation index (ODI4), the mean percutaneous oxygen saturation (MSpO2), and the percentage of time with blood oxygen saturation below 90% during total sleep time (TS90%). Previous studies have demonstrated a positive correlation between ODI4 and apnea-hypopnea index (AHI).⁷ For example, using an ODI threshold of ≥10.4 times/hour to screen for sleep apnea-hypopnea syndrome (SAHS) with an AHI ≥5 times/hour as the diagnostic criterion, the sensitivity is 86.4% and the specificity is 82.4%.⁸ In this study, an ODI ≥ 15 events/hour was applied as the screening threshold for moderate-to-severe SAHS, as this cutoff is conventionally linked with higher clinical specificity and relevance. The discrepancy between this and the referenced validation threshold is acknowledged as a limitation.

Psychological Assessment

Insomnia Severity Index (ISI) is a self-rating tool developed by Morin to assess self-awareness of insomnia symptoms over the past two weeks.⁹ Compared with the PSQI and ESS, ISI specifically evaluates the severity of insomnia, which has been validated as a highly effective and widely used tool in clinical research. For logistic regression analysis, insomnia was defined using an ISI score ≥ 8. This threshold, while in the “subthreshold insomnia” range of the original Morin classification,^9–11 was chosen for its higher sensitivity in screening for clinically significant sleep disturbance in a high-stress population. The implications of using this cutoff are considered in the discussion.

The Self-reporting Questionnaire (SRQ-20) was designed specifically for developing countries with 20 items, totaling 40 points, with higher scores indicating more prominent symptoms of mental disorders. It has been used in post-disaster mental health surveys and shown to have good consistency and criterion-related validity.¹⁰

Patient Health Questionnaire¹¹ (PHQ-9) is a concise and effective self-rating tool for depression with nine items, totaling 27 points, and was widely applied in screening for depression symptoms with a sensitivity of 88%, specificity of 88% and positive detection rate of 31% to 51%.

Generalized Anxiety Disorder¹² (GAD-7) as a concise and effective self-rating tool for anxiety with seven items, and has been widely used in screening for anxiety symptoms with sensitivity of 86.8% and specificity of 93.4%.

Perceived Stress Scale¹³ (PSS-10) has 10 items with 40 points, including six negative items and four positive items and the higher scores indicate stronger stress perception. The negative items evaluate the sense of losing control and negative emotional reactions, while the positive items assess the ability to handle existing sources of stress. Previous studies have shown that the Chinese version of PSS-10 has a Cronbach’s α coefficient of 0.86, which supports its use in China.

Statistical Analysis

All data were analyzed by SPSS 19.0. Measurement data were presented as mean ± standard deviation (), and count data were expressed as percentages (%). The aim of the analyses was threefold: (1) to compare sleep metrics between groups and time points using independent-samples t-tests or one-way analysis of variance (ANOVA); (2) to examine relationships between ISI and other variables using correlation analysis; and (3) to identify independent factors associated with insomnia using logistic regression. Normality of continuous variables was assessed using the Shapiro–Wilk test. Shapiro–Wilk test results indicated that age, BMI, ISI, SRQ-20, PHQ, GAD, PSS, and average heart rate were normally distributed. For normally distributed variables, Pearson correlation was used; otherwise, Spearman’s rank correlation was applied. For post-hoc comparisons following ANOVA, Bonferroni correction was applied to control for multiple testing. Group incidence rates were compared using chi-square tests. Logistic stepwise regression analysis was conducted to identify independent factors associated with insomnia. The test standard was set as α=0.05. A post-hoc power calculation for the primary outcome (TST) was performed.

Results

General Demographic Data and Questionnaire Survey

Initially, 40 participants were screened in the anti-epidemic group. Five participants were excluded because of the use of benzodiazepines or non-benzodiazepines, whereas four participants dropped out because of diminished interest in the study (Figure 1). Thirty-one were eventually included in the evaluation and only 1 individual did not complete the sleep monitoring. Among them, there were 18 males and 12 females, with an average age of 39.83±6.89 years. Follow-up was conducted on 28 of the participants 1 year later. In the control group, a total of 40 participants were recruited and 34 received sleep monitoring and psychological assessment. Finally, 28 completed all assessments. Of these, 18 were males and 10 were females, with an average age of 35.53±9.89 years. No significant differences were observed in gender or age between the two groups (p > 0.05). Demographic characteristics are presented in Table 1.

Table 1 General Demographic Data and Questionnaire Survey ()

Figure 1 The study flow chart.

The results of the questionnaire survey are shown in Table 1. No apparent differences in total scores of each scale were observed between the two groups at the post-deployment assessment or at 1-year follow-up. However, pairwise comparisons revealed that the PHQ-9 scores of the anti-epidemic group at both time points were lower than those of the control group at a trend level (p = 0.060 and 0.081, respectively).

Pulse Oximetry Monitoring Using a Ring-Shaped Device

The results of the medical pulse oximeter monitoring are shown in Table 2. The omnibus ANOVA for the proportion of deep sleep was not statistically significant (F=2.456, p=0.092). Therefore, reported pairwise differences for this variable should be interpreted only as non-significant trends and exploratory findings. Among them, the difference in actual total sleep time (TST) between the post-deployment assessment, 1-year follow-up of the anti-epidemic group, and the control group was significant (F=9.942, p<0.001). Post-hoc pairwise comparison results showed that at the post-deployment assessment, TST of the anti-epidemic group was significantly higher than at the 1-year follow-up and control group (standard error = 15.60, p<0.001 and p=0.001, respectively). A post-hoc power calculation for the primary outcome (TST) was performed using G*Power 3.1.9.7, with parameters set as α=0.05, effect size f=0.35 (calculated from the ANOVA results: mean TSTs of 445.3, 402.8, and 397.1 min, with a pooled SD of approximately 68.2 min), and a total sample size of N=84 (28 per group). This calculation indicated that the study achieved a power of >80% to detect the reported difference in TST. However, the study was underpowered for detecting small-to-medium effects in secondary outcomes such as sleep stage proportions.

Table 2 Medical Pulse Oximeter Readout ()

Given the non-significant omnibus test for the proportion of deep sleep stage among the three groups, exploratory pairwise comparison results showed that this trend was mainly due to a numerical decrease of deep sleep in the anti-epidemic group at the end of follow-up (standard error = 1.213, p=0.034), and a relative numerical increase in light sleep compared to the post-deployment assessment (standard error = 2.428, p=0.042). Using ODI≥15 times/h as the threshold, it was found that 7 cases (23.3%) in the anti-epidemic group were diagnosed with SAHS at the post-deployment assessment, which decreased to 5 cases (17.9%) after 1-year follow-up, while 5 cases (17.9%) of SAHS were found in the control group, and the difference in incidence of SAHS between anti-epidemic group and control group was not significant.

Pearson correlation analysis was performed between the ISI of the anti-epidemic group and their general conditions, scores on various scales, and sleep indicators. The results showed ISI was positively correlated with SRQ, PHQ, GAD, PSS, and average heart rate (correlation coefficients were 0.544, 0.366, 0.360, 0.284, and 0.289; p values were <0.001, 0.005, 0.006, 0.031, and 0.028), and negatively correlated with BMI (correlation coefficient was −0.296, p = 0.024).

Logistic stepwise regression was performed for the anti-epidemic group with an ISI score ≥ 8 points as judgment criterion for insomnia. The insomnia criterion was as dependent variable, variables including age, gender, BMI, scores on various scales such as PHQ, GAD, PSS, SRQ20, and sleep quality were included in logistic multivariate stepwise regression analysis (Table 3). The results showed that SRQ20, age, BMI, and PHQ entered the regression equation. The chi-square test result was χ²=29.23, p<0.001, indicating that the regression equation was statistically significant. The analysis revealed that a higher SRQ-20 score (OR = 10.55, 95% CI 1.689–65.932, p = 0.012) and older age (OR = 1.667, 95% CI 1.060–2.623, p = 0.027) were independently associated with increased odds of insomnia. Conversely, higher BMI (OR = 0.376, 95% CI 0.149–0.945, p = 0.038) and higher PHQ-9 score (OR = 0.385, 95% CI 0.134–1.109, p = 0.077) were associated with decreased odds of insomnia, though the latter did not reach conventional statistical significance.

Table 3 Parameters for the Logistic Regression Model

Discussion

The mental health effects of anti-epidemic staff have been studied during previous outbreaks of SARS-CoV-1 (2003),^14–16 H1N1 (2009),^17,18 MERSCoV (2012)^19,20 and Ebola (2014).^21,22 There are high levels of post-traumatic stress, depression, anxiety, and absenteeism observed in the individuals who were in direct contact with an infected person.²³

Due to the strict control of COVID-19 infection in China, our subjects, although directly facing the infected person, were fortunately not infected in 2020–2021. This was very important to exclude the effects of COVID-19 infection in our study. Changes in sleep phase after stress are not unusual.²⁴ Sanford et al showed that acute stress may lead to the inhibition of rapid eye movement (REM) sleep and an increase in duration of non-REM sleep. However, as the stress event ends,²⁵ reported that individuals may exhibit a restorative sleep compensation, with increased sleep time and an increase in REM sleep. Consistently, we observed prolonged sleeping time of anti-epidemic staff due to sleep compensation after stress. Exploratory analysis of device-estimated data from Table 2 showed a non-significant numerical trend toward a decrease in deep sleep (21.18% to 18.56%) and an increase in light sleep (48.31% to 53.31%) at the 1-year follow-up, which should be interpreted with caution due to the lack of statistical significance and the inherent limitations of the device. Interestingly, although the REM time of anti-epidemic staff was trending higher, there was no statistical difference when compared with control group. In the 1-year follow-up, although the total sleep time had generally recovered, exploratory device-estimated data suggested a non-significant trend of altered sleep architecture. This can be attributed to continuous or repeated occurrence of chronic stress and may have negative effects on sleep.²⁶ It is important to emphasize that all sleep architecture findings are exploratory and require PSG confirmation due to the limitations of the pulse oximeter device.

We found that the ISI of anti-epidemic staff was positively correlated with SRQ, PHQ, GAD, PSS, and average heart rate, while the negative correlation between ISI and BMI index suggests that individuals with insomnia in this sample tend to have a relatively lower body weight. This result suggested that the sleep quality of anti-epidemic staff was affected by multiple factors, such as depression, anxiety, physical condition, and stress level. Further logistic regression analysis results showed that SRQ20, age, BMI, and PHQ score were independently associated with the presence of insomnia. Crucially, our correction of the initial misinterpretation reveals that while higher SRQ-20 and older age were associated with greater odds of insomnia, higher BMI and higher PHQ-9 score were associated with lower odds. This inverse association for PHQ-9 requires careful interpretation. It may suggest that the ISI cutoff used captured a group where somatic symptoms of depression (like weight loss) were prominent, or it may reflect collinearity between PHQ-9 and SRQ-20, with SRQ-20 capturing the primary distress variance. The inverse relationship with BMI contrasts with some previous reports hypothesizing obesity as a risk factor for insomnia through inflammatory or apneic pathways.^27–30 Our finding aligns more with an alternative hypothesis that in this specific high-stress context, lower BMI and insomnia may be concurrent manifestations of depression, rather than one causing the other.³¹

Wearable devices have improved the convenience of sleep monitoring devices, especially since subjects can perform sleep monitoring at home, ensuring a monitoring environment closer to natural sleep. In previous studies, wearable devices were found to quickly screen for sleep problems. For example, in 2020, reference⁴ used a wearable medical pulse oximeter to monitor medical staff with insomnia symptoms and stress in Wuhan, and found that insomnia-prone medical staff had obvious sleep apnea. SRQ-20 scores and gender were the major risk factors for insomnia co-morbid with SAHS. However, our study found that SRQ-20, age, BMI and PHQ were the primary factors associated with insomnia. These findings exhibit notable differences compared to previous research. Possible reasons for this discrepancy could include: the demographic and clinical profiles of the study populations might differ, influencing the risk factors identified; differences in study design, including sample size, selection criteria, and data collection methods, may contribute to the observed differences; cultural and regional variations in sleep patterns, stress responses, and health-seeking behaviors could play a role.

Previous report⁶ showed that consumer wearable sleep monitoring devices have better sensitivity and accuracy in evaluating TST and SE compared with portable EEG. However, wearable sleep monitoring devices still have limitations in accurately measuring sleep latency and sleep cycles. A critical limitation of this study is that the ring-shaped pulse oximeter estimates sleep stages from autonomic signals, not EEG. Therefore, findings on deep sleep, light sleep, and REM sleep are device-estimated and require confirmation by the gold-standard PSG. The absence of a published validation study for sleep staging by this specific device means our sleep architecture results must be viewed as preliminary. In this study, the incidence of SAHS in both anti-epidemic group and control group was higher than incidence of obstructive sleep apnea hypopnea syndrome³² (>17.9%, and reached as high as 23%). Nevertheless, the promotion and application of wearable devices make it feasible to conduct sleep screening in various special situations, making it more convenient to obtain reliable evidence to guide relevant personnel in sleep intervention.

Limitations

Several limitations of this study must be acknowledged. First, the “post-deployment assessment” was conducted after the acute stress of working in Wuhan had ended, not before it, and therefore does not represent a true baseline. All comparisons reflect changes from a post-stress recovery phase, not a stress-versus-recovery contrast. Second, the control group was recruited from a different city (Shanghai) and during a partially overlapping but distinct time frame, introducing potential geographic and temporal confounds that limit the validity of between-group comparisons. Third, the study lacks longitudinal follow-up data for the control group, preventing a robust parallel comparison of time-related changes. The mixed design limitations are discussed transparently. Fourth, the sample size was small (N=28 per group), and although the study was adequately powered for the primary TST outcome, it was underpowered for many secondary analyses, particularly the logistic regression which may yield unstable estimates. Fifth, the reliability of sleep structure analysis by the pulse oximeter has not been validated against PSG, so all sleep architecture findings are exploratory. Sixth, the use of an ODI threshold of ≥15 events/hour for SAHS classification deviates from the ≥10.4 threshold supported by the cited validation study; this choice limits comparability and may have underestimated the prevalence of milder SAHS. Seventh, the ISI threshold of ≥8 was used for defining insomnia, which is a subthreshold cutoff and may have influenced participant classification. Finally, only a single night of sleep data was recorded at each time point, which may not be representative of habitual sleep patterns. Future research should incorporate multi-night assessments, true baseline measurements, a parallel and contemporaneously-recruited control group with longitudinal follow-up, and validation against PSG.

Conclusion

In this exploratory study, greater stress exposure was associated with prolonged TST in the immediate post-deployment period, which partially normalized at one-year follow-up. Non-significant trends suggested persistent alterations in device-estimated sleep architecture. SRQ-20 score, age, BMI, and PHQ-9 score were independently associated with insomnia in this cohort, with higher SRQ-20 and older age linked to increased odds, and higher BMI and PHQ-9 score showing inverse associations. Given the observational and exploratory nature of this study and its significant methodological limitations, particularly concerning device validity and study design, causal conclusions cannot be drawn. The application of wearable pulse oximeters may offer a feasible approach for large-scale sleep health screening in challenging environments, but diagnostic conclusions, especially regarding sleep architecture, require confirmation by polysomnography.