The methodology of the study was previously described in the frame of the prospective randomized controlled trial, published in a study protocol by Ochmann et al [40]. The protocol adheres to the guidelines from the SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials) and the CONSORT (Consolidated Standards of Reporting Trials)–eHEALTH checklists (Multimedia Appendix 1). With the outbreak of the COVID-19 pandemic, the lockdown measures introduced in Germany in March 2020 prevented the continuation of the regular study procedure as well as further enrollment of participants. To still be able to provide sports therapy for the participants included at the beginning of 2020, the initial study protocol had to be adapted, and in the following sections, mainly these amendments are described.

The study design, illustrated in Figure 1A, included examinations before the intervention (T0: psychological assessments and T1: physiological assessments), an intervention period, and final examinations (T2: psychological assessments and T3: physiological assessments). Healthy untrained (less than 120 minutes per week being physically active) participants aged between 18 and 45 years were recruited via flyers distributed at the Johannes Gutenberg–University of Mainz and the University Medical Center campuses (for a detailed overview of inclusion and exclusion criteria, see Ochmann et al [40]). A total of 35 participants who provided written informed consent were randomly assigned to an intervention group (IG) or CG and completed the assessments before the intervention. This study includes data from a pre–COVID-19 cohort (2019) and a lockdown cohort (2020). Their difference in methodology relates mainly to the longer intervention period for the lockdown cohort, as shown in Figure 1B. A total of 7 participants had to be excluded from all analyses: 3 of them were excluded due to incomplete data sets and 4 participants dropped out of the study (for more details, see the Results section). This leads to a final sample of 13 participants in the pre–COVID-19 cohort (IG: n=7 and CG: n=6) and 15 participants in the lockdown cohort (IG: n=9 and CG: n=6). Further descriptions of the final study population can be found in the Results section.

Weekly training in the IG consisted of progressive continuous and interval runs with the goal of gradually increasing aerobic capacity (for details, see Ochmann et al [40]). The trainer individually adapted the training to the weekly feedback provided by the participant via web-based support. As we did not receive any reports regarding major technical problems from the study participants, we assume that the technical use of web-based support was not a challenge from the perspective of the test participants. However, if more intensive clarification was required, a telephone consultation was arranged. After verifying that the participant complied with the training recommendations, the 2 factors of perceived ailment and load mainly determined the adjustments. Since the original training progression had been designed for an 8-week study period, this design could not cover an indefinite period. Therefore, the trainer mainly focused on the continuation of the participants’ sporting activities. As originally planned, the principle of training adjustment was also applied to the lockdown cohort. Thus, depending on the perceived ailment and load, the training was reduced, maintained, or intensified.

The German version of the Brief Symptom Inventory-18 (BSI-18) [41,42] was applied to measure psychological distress. The BSI-18 contains 3 subscales, that is, depression, anxiety, and somatization, which are assessed as sum scores of 6 items each, as well as a Global Severity Index calculated as a sum across all 18 items. All participants worked twice on the BSI-18, once before the intervention period (T0) and once after the period (T2). In accordance with our hypotheses, we limited our main analyses to the depression subscale of the BSI-18.

VO_2peak was determined as a main indicator of aerobic capacity. It was chosen over VO_2max as the main parameter since we did not expect the occurrence of a plateau in testing untrained participants [43]. VO_2peak was measured during a stepwise incremental running test starting at 4 km/h and increasing velocity every 3 minutes by 1.5 km/h (for further details, see the initial study protocol by Ochmann et al [40]). The determination took place once before the intervention period (T1) and once after the period (T3). Training adherence was calculated by dividing documented training sessions by recommended training sessions.

Participants in the lockdown cohort were tested for SARS-CoV-2 seroconversion to exclude an undetected infection via the enzyme-linked immunosorbent assay described by Notz et al [44]. Venous blood was drawn in tripotassium-ethylenediaminetetraacetic acid–covered Monovettes (Sarstedt) and was processed within 5 minutes after blood collection. Platelet-free plasma was prepared by 2 rounds of centrifugation for 15 minutes at 2500×g and room temperature. Plasma was aliquoted and stored at –80 °C until enzyme-linked immunosorbent assay measurement.

At T1, all participants showed no sign of seroconversion, as expected prior to the pandemic situation. At T3, seroconversion of 2 participants, who had already stated a SARS-CoV-2 infection during the intervention phase, was confirmed. Medical anamnesis as well as pulmonary function and spiroergometric tests ensured that these participants did not experience any aftermaths affecting their ability to participate in the study. Besides this, no seroconversions or stated SARS-CoV-2 infections were reported. Therefore, no participants were excluded due to the SARS-CoV-2 infection.

SPSS (version 23; IBM Corp) was used for data analysis. Aligned rank transform (ART) ANOVAs [45] were conducted for each cohort separately to test our hypotheses, with time (before vs after the intervention) as a within-subject factor, group (IG vs CG) as a between-subject factor, and VO_2peak and depression scores as dependent variables. Follow-up Wilcoxon tests were performed to contrast specific time- and group-related effects.

The study protocol follows the standards of the Declaration of Helsinki of the World Medical Association, and ethics approval was obtained by the medical association Rhineland-Palatinate (July 29, 2019; IRB approval 2019-14305). The original study was registered with the German Clinical Trials Register (DRKS00018078; October 2, 2019). All participants provided informed consent before participation and were informed about the ability to opt out of the study at any time. Furthermore, all data were pseudonymized. As compensation for complete participation, 15 smartwatches (M430, Polar Electro Oy) were raffled among all participants.

Due to the unexpected occurrence of the COVID-19 pandemic and its related restrictions, we have the unique opportunity to report on the feasibility of the conduction of a randomized controlled remote exercise intervention under these circumstances. Whereas a pre–COVID-19 cohort was able to proceed through the study protocol as planned, the intervention period and the final examinations of the lockdown cohort could not be conducted as planned. On March 16, 2020, all participants were notified about an indefinite extension of the intervention period. The local restrictions, in the form of the first lockdown, in Germany started on March 27, 2020. Figure 1B shows all participant timelines combined with a visualization of the weekly COVID-19 cases. Table 1 provides additional measurements describing the study population. Participants in the IG and CG did not show significant differences regarding VO_2peak (pre–COVID-19 cohort: P=.22 and lockdown cohort: P=.52) or depression scores (pre–COVID-19 cohort: P=.43 and lockdown cohort: P=.70) before the intervention.

The duration of the intervention for the lockdown cohort between T1 and T2 increased on average to 17.3 (SD 1.0) weeks for the CG and 17.7 (SD 2.0) weeks for the IG. The intervention times for the pre–COVID-19 cohort in comparison lasted a mean of 8.3 (SD 0.5) weeks for the CG and 8.3 (SD 0.5) weeks for the IG. The mean number of training sessions per week, on the other hand, showed no difference (pre–COVID-19 cohort: mean 2.3, SD 0.4 and lockdown cohort: mean 2.3, SD 0.6). However, in the lockdown cohort, the training became more individualized the longer the training intervention lasted. While the framework of the original training progression was retained, the structure of the training sessions was adjusted to the individual demands. This led to a slightly higher effort for the trainer without increasing communication with the participants. For example, there were individuals who preferred to maintain a frequency of 3 units per week during the lockdown, while others preferred 5 units per week. Furthermore, in addition to endurance runs and interval units, combinations of moderate and vigorous intensities as well as Fartlek runs (individual pace and intensity variations during unstructured interval training) were included. Depending on how the participants accepted these training methods, they were recommended more or less frequently. However, the overall weekly volume was always maintained. Finally, the mean adherence, that is, the quotient of training sessions performed and recommended, did not differ between the cohorts (84% for both cohorts) but differed in the SD (pre–COVID-19 cohort: 5.5% and lockdown cohort: 11.6%; Table 1).

There were low dropout rates in both cohorts in the IG (pre–COVID-19 cohort: n=0, 0% and lockdown cohort: n=2, 16.7%) and the CG (pre–COVID-19 cohort: n=0, 0% and lockdown cohort: n=2, 20%). No participants in the pre–COVID-19 cohort and 4 participants in the lockdown cohort dropped out from the study. However, there was no study dropout specifically induced by the prolongation of the intervention period. In total, 2 participants dropped out of the CG due to a change of personal situation, and 2 participants dropped out of the IG due to medical reasons not related to the study or COVID-19. One of the latter participants dropped out 16 weeks after T1, and the other dropped out 4 weeks after T1 and before the lockdown started.

Taken together, the adaptation of the training intervention to the pandemic situation was well feasible, and the training adherence as well as the dropout rate were adequate. Despite the small sample sizes induced by introducing a pre–COVID-19 cohort and a lockdown cohort, the study populations in the IG and the CG for both cohorts were highly similar.

Since the conduction of our remote physical exercise intervention turned out feasible and the study population was highly comparable at T1, specifically regarding VO_2peak and depression scores, we investigated a potentially beneficial effect of the physical exercise intervention on both health variables in the next step. Since the interventions in the pre–COVID-19 and lockdown cohorts differed especially regarding training durations, we analyzed both cohorts separately. We performed nonparametric analyses to account for the small sample sizes in both cohorts. More specifically, ART [45] was conducted to investigate the main and interaction effects of group (intervention vs control) and time (before vs after the intervention) on depression and VO_2peak. Follow-up 1-sample Wilcoxon signed rank tests were performed to investigate changes in both health parameters within groups.

Regarding the pre–COVID-19 cohort, the ART ANOVA with VO_2peak as a dependent variable revealed nonsignificant main effects of time (F_1,11=0.98; P=.34; η²=0.08) and group (F_1,11=0.01; P=.92; η²<0.01) but a significant time×group interaction (F_1,11=5.27; P=.04; η²=0.32). However, a second ART ANOVA on depression did not indicate significant main effects of time (F_1,11=0.45; P=.52; η²=0.04) and group (F_1,11=0.18; P=.68; η²=0.02) nor a significant time×group interaction effect (F_1,11=0.82; P=.39; η²=0.07). One-sample Wilcoxon tests for each group separately indicated a significant increase in VO_2peak (before the intervention: mean 34.99, SD 8.43 and after the intervention: mean 37.94, SD 7.48; P=.04) in the IG (Figure 2).

Regarding the lockdown sample, the ART ANOVAs revealed significant time×group interaction effects on VO_2peak (F_1,13=11.44; P=.01; η²=0.47) and a marginally significant interaction effect on depression (F_1,13=3.90; P=.07; η²=0.23). The ART ANOVAs revealed no significant main effects of time or group on VO_2peak (time: F_1,13=0.04; P=.85; η²<0.01 and group: F_1,13=0.09; P=.77; η²=0.01) or depression scores (time: F_1,13<0.01; P=.98; η²<0.01 and group: F_1,13=0.23; P=.64; η²=0.02). As indicated in Figure 3, follow-up Wilcoxon tests indicated a significant decrease of VO_2peak (before the intervention: mean 37.58, SD 7.84 and after the intervention: mean 35.38, SD 8.16; P=.03) and a marginally significant increase in depression (before the intervention: mean 1.83, SD 1.72 and after the intervention: mean 3.67, SD 2.42; P=.08) in the CG.

Taken together, our findings indicate that while participants in the CG experienced a decrease in aerobic capacity and an increase in depression levels during the first COVID-19–induced lockdown, participants in the IG remained stable on both parameters. On the other hand, under normal conditions (pre–COVID-19 cohort), only the aerobic capacity increased significantly in the IG.

In this study, we had the unique chance to investigate the feasibility of a remote physical exercise intervention and, furthermore, its potentially beneficial effect on physical and mental health during a stress-inducing COVID-19 lockdown situation. With a low dropout rate and high adherence to the training sessions, the intervention was well feasible. Moreover, our results hint at the stress-buffering potential of the exercise intervention, although the statistical power was limited due to the small sample size.

The continuation of our study was well feasible in the prevailing pandemic situation mainly due to the use of a web-based training intervention. All training support structures implemented in this study procedure could be maintained as planned, and high adherence values and low dropout rates in the lockdown cohort were achieved. Most importantly, there was no need for physical contact, while the continuation of a program including such an approach would not have been possible. The training could simply continue as before since the trainer and participant were in contact exclusively via digital means. This remote contact allowed for fast support in case of problems, as there was no need to wait for the next training session. Changes that were necessary to keep the intervention running included the extension of the training plans from originally 8 weeks to a longer, undefined period, stronger exchange between participant and trainer, and more individualized training plans. Heart rate and other parameters were tracked continuously, and the trainer had a very detailed insight into the physiological variables during the workout, which facilitated personalization processes. Moreover, the individuals’ training preferences and personal motivation could be easily considered through frequent feedback via chat messages and training protocols completed by the participants. However, this increased the trainer’s workload to a certain level; but in interventions that keep contact with the participants exclusively via remote, it is important for the trainer to follow a certain regularity, provide frequent feedback, and engage with the participant on a personal level with genuine interest anyway [46]. Finally, the adaptations were feasible, and we had the opportunity to further evaluate the effect of our physical exercise intervention on physical and mental health.

Next, we investigated the stress-buffering potential of our remotely supervised physical exercise intervention facing a severely restrictive life situation in terms of daily activity and social interactions. Specifically, we were interested in the potentially beneficial effects of regular physical exercise on both VO_2peak and depressive symptoms in the IGs. The results of the pre–COVID-19 cohort showed the expected significant increase in aerobic capacity for the IG but no changes in the CG. The increase in VO_2peak in the healthy pre–COVID-19 cohort is similar to changes in patient cohorts with different diseases in comparable intervention studies [35,37]. However, the exercise intervention did not significantly decrease depression scores, most likely due to only minimal depressive symptoms before intervention. In the lockdown cohort, time×group interactions revealed a significant deterioration of VO_2peak and a marginally significant decrease of depression in the CG but not in the IG. While the results in the pre–COVID-19 cohort indicate the overall effectiveness of the remote training intervention, the IG of the lockdown cohort maintained their physical and mental health in the face of a naturalistic stressor, that is, the pandemic-related lockdown. The latter finding is in line with previous studies on the stress-buffering effects of physical exercise [28,29]. Accordingly, the training intervention has likely compensated for the detrimental effects of a sedentary routine due to COVID-19–related restrictions.

Our findings corroborate the stress-buffering potential of physical exercise previously indicated by multiple cross-sectional studies [47]. However, only a limited number of randomized controlled trials, especially in real-life stressor settings, have been conducted so far. For example, 20 weeks of aerobic exercise training were shown to have beneficial effects on the emotional stress reactivity of healthy students [48] for the prevention of stress-related mental health impairments. In line with the current findings, future studies should focus on real-life stressors when investigating the stress-buffering potential of physical exercise interventions.

The results hint at a stress-buffering effect of physical exercise after ~18 weeks. It remains unclear whether fewer training sessions, for example, 8 weeks, as originally planned, would yield similar stress-buffering effects. Most studies on physical exercise use intervention periods of 8-12 weeks, whereas a meta-analysis by Bruner et al [48] shows stronger effects for different factors of positive youth development in studies exceeding 10-week intervention periods. Interventional studies with overfat individuals show the most beneficial effects on cardiorespiratory characteristics in interventions that last at least 12 weeks [49]. The longer intervention period may thus also have enhanced effects on aerobic capacity. Further randomized controlled trials should define the modes, volumes, and lengths of physical exercise that elicit the stress-buffering effect of routine physical exercise.

To our knowledge, this is the first study to investigate the benefits of a training intervention at the onset of the COVID-19 pandemic in healthy untrained participants, which measures aerobic capacity by VO_2peak. Specifically, participants started the intervention shortly before pandemic-related restrictions were implemented. The significant improvements in VO_2peak observed in the pre–COVID-19 cohort were not achieved under lockdown conditions. Instead, aerobic capacity in the IG of the lockdown cohort did not change significantly, while it decreased significantly in the CG. In line with findings from similar studies, participants in the CG of the lockdown cohort may have experienced detraining or deconditioning. Ammar et al [50] showed a detraining effect after an initial 8-week fitness-dance training in older adults with mild cognitive impairment. While detraining in general is defined as the partial or complete loss of training-induced adaptations [51], our observations show a similar effect even without initial training in untrained participants. This deconditioning effect is most likely induced by the overall lowered physical activity [52-54], which different studies show during the COVID-19 lockdowns [4,6]. To conclude, training interventions may stop or at least buffer lockdown-induced deconditioning even in untrained healthy participants.

However, as external circumstances introduced unforeseen differences between the first and second cohorts, small sample sizes limited the robustness of our findings, which should consequently be interpreted with caution. While the results indicate large effect sizes regarding the intervention on aerobic capacity in the pre–COVID-19 cohort (η²=0.32) and on both health parameters in the lockdown cohort (VO_2peak: η²=0.47 and depression: η²=0.23), a follow-up study in a more representative sample would enable a more conclusive interpretation of these findings. Along these lines, the limited statistical power did not permit analyses of potential mediator variables. Future studies in larger samples should analyze resilience mechanisms affected by the training intervention, for example, improved hypothalamic-pituitary-adrenal axis activity or increased self-efficacy. Still, the coincidental (or serendipitous) nature of our findings renders an equivalent replication study rather impossible. Nevertheless, a follow-up study may apply a somewhat similar design by implementing the training intervention in a population that experiences a more predictable real-life stressor with similar characteristics, for example, increased sedentary behavior and less social interactions. A population of interest may, for example, consist of students facing an examination period.

The outbreak of the COVID-19 pandemic and the implemented lockdown periods led to an increase in sedentary behavior associated with decreased health and well-being. We had the unique opportunity to investigate the feasibility and the stress-buffering potential of a remote physical exercise intervention with healthy untrained participants under these circumstances. Our results indicate that a lockdown cohort, which started a remotely supervised physical exercise intervention shortly before pandemic-related restrictions, significantly benefited from regular physical exercise in terms of aerobic capacity and depressive symptoms. Low dropout rates and high adherence to the training, despite the prevailing lockdown-related restrictions, underscore the feasibility and importance of remote physical exercise training approaches in such scenarios. Furthermore, our findings highlight the importance of physical exercise to cope with stressful life situations and its integration into daily life, for example, on a personal level or via occupational health management. Future randomized controlled trials regarding further relevant real-life stressors may reveal the broad stress-buffering potential of physical exercise.