The protocol for this study design has been published (27). A semi-randomized, longitudinal study design was used including three different study groups: traditional IRENA, digIRENA (digital IRENA), and a control group. All participants met the legal criteria for post-rehabilitation IRENA eligibility. The allocation of participants to the groups took place during their initial consultation with the responsible doctor at a rehabilitation clinic. Those rejecting IRENA participation were randomly allocated to either the digIRENA or control group. Sealed envelopes were used for the randomization procedure and the participants knew their group only after opening the envelopes. These envelopes were evenly distributed across the clinics to achieve a balanced 1:1 ratio between the digIRENA and control groups. It is important to note that, due to the transparent nature of the aftercare interventions, neither participants nor staff could be blinded to group allocation. Additionally, it was observed that some participants opted not to participate in the study upon learning they had been assigned to the control group rather than the digIRENA group, indicating a preference for the digital rehabilitation aftercare option. To increase the motivation to participate in the study, the participants that completed all four measurements received a 50 Euro Amazon voucher. Participants were initially recruited from three multiple rehabilitation clinics belonging to DRV Knappschaft-Bahn-See and the plan was to recruit from January 2020 until October 2020. However, recruitment was interrupted due to COVID-19 and also after the interruption there was a reduced admission in the rehabilitation clinics. For this reason, in August 2020 two more rehabilitation clinics joined and a one-year recruitment extension was authorized until October 2021. Because the number of sufficient participants was reached, the recruitment ended in August 2021. Although the initial plan was to recruit 1,150 participants at baseline (27), the dropout was less than expected and therefore the recruitment stopped when 1,060 participants were recruited.

The study comprised four measurement time points: pre-aftercare (T0), at 13 weeks post-initiation of aftercare (T1), at 26 weeks or the assumed end of aftercare (T2), and a 17-week post-aftercare follow-up (T3). These measurement intervals were strategically selected to capture the midpoint (T1), endpoint (T2), and a four-month follow-up (T3). It is critical to acknowledge that this schema represents an idealized timeline as deviations occurred due to varying starting time points and inconsistent adherence to the aftercare programs. Data collection included in-clinic questionnaires at T0 and subsequent mail or email sending of questionnaires at T1, T2, and T3, accompanied by a two-week non-response follow-up reminder. The study has been formally registered with the German Register of Clinical Studies under the identifier DRKS00022467. Ethical approval and data protection compliance have been obtained from the Ethics Committee and the Data Protection Officer at the Karlsruhe Institute of Technology.

Power calculations indicated the necessity for 573 participants to assess telerehabilitation aftercare effectiveness, assuming a small effect size (Cohen’s f = 0.07) due to limited robust empirical studies in this area (22). With an alpha error set at 0.05 and a desired test power of 0.80, this sample size is required when using analysis of variance with repeated measurement. Anticipating a dropout rate of about 30% per measurement point, the study initially aimed to enroll approximately 1,150 patients. Finally, because drop-out was lower than expected, 1,060 orthopedic rehabilitation patients were recruited at the baseline measurement (T0). In our study, we specifically targeted patients undergoing orthopedic rehabilitation, reflecting our primary aim to assess the effectiveness of digital and traditional aftercare within a cohort primarily affected by musculoskeletal disorders. Participants were recruited from five distinct rehabilitation clinics, ensuring a diverse sample representative of individuals undergoing orthopedic treatment. During their rehabilitation stay, the study’s objectives and procedures were outlined to potential participants in a medical consultation, where study information and consent forms were provided. Those who agreed to participate completed the questionnaire either during their hospital stay or directly after discharge. A key inclusion criterion was a sufficient command of the German language, ensuring that all participants could complete the questionnaire independently without requiring external assistance. Attrition occurred over the course of the study, with four participants withdrawing consent for data utilization (two from digIRENA, one from traditional IRENA, and one from the control group). Thus, at T0 the sample consisted of 405 individuals in the digIRENA intervention cohort (145 females, representing 35.8%), 352 in the traditional IRENA cohort (120 females, 34.1%), and 299 in the control cohort (79 females, 26.4%). With relevance for the analysis of the present study, there were 805 participants who filled out the questionnaires at all four measurement occasions. From these, 323 were in the digIRENA group, 252 in the IRENA group, and 230 in the control group.

All participants in the study were offered IRENA intervention, and those subjects who declined to participate in IRENA were randomly assigned to either the digIRENA group or the control group. Two intervention programs were implemented in this study: Traditional IRENA and digIRENA. Participants in both intervention programs were under the continuous supervision of accredited therapists. In the case of adverse outcomes, participants were referred for additional treatment. Additionally, a control cohort was established, without a prescribed aftercare intervention.

The acquired dataset underwent preliminary data processing prior to analysis. In a first step, missing data were analyzed separately for each measurement time point using Little’s MCAR test to determine the extent to which systematic missingness was present (35). Missing values at the item level were imputed utilizing an Expectation–Maximization algorithm (36), an iterative procedure consisting of expectation and maximization steps that converge upon a local maximum likelihood estimate (37). Post-imputation, values were rounded to the nearest integer. In cases where fewer than 50% of the required values were available for a given scale score, missing value imputation was precluded, and the corresponding sum score was reported as ‘missing,’ thereby excluding the data from subsequent analyses. The differences between the three groups in terms of dropout rate between t0 and t3 were analyzed using the chi-squared goodness of fit test. One participant who self-identified as “diverse” in gender was excluded from analyses using age and gender as covariates.

An intention-to-treat analysis was employed, encompassing all individuals who consented to participate in the study and for whom data were available. Descriptive statistics were presented with regard to the distribution of gender, age, nationality, employment status, participation in aftercare, and motivation. To examine the differences between the digIRENA group and the IRENA group in terms of motivation and attendance, t-test and Welch test were used. Thereafter, repeated measures analyses of variance (ANOVA) were conducted to interrogate longitudinal trends in dependent variables, specifically subjective physical, and subjective mental health metrics as assessed by the SF-12 questionnaire (38). ANOVA was used across all four measurement occasions to assess the effects of the aftercare intervention and its durability. The sphericity assumption was tested using the Mauchly test and, in case of violation, the Greenhouse–Geisser adjustment was used to correct for violations of sphericity. To accommodate the challenges posed by unbalanced sample sizes in our ANOVA analysis, we employed both Type II and Type III ANOVA, ensuring the robustness of our findings across the digIRENA, IRENA, and control groups. The similarity of results between these analyses suggests that our conclusions are reliable and unaffected by the choice of analytical method. Initial group allocation biases based on age and gender variables were examined through logistic regression. In a separate ANOVA, the study examined the interaction of time, group, and age, based on the study by Schmidt et al. (39), where such an interaction suggested that younger participants benefited more from digital prevention programs, while older participants benefited more from traditional approaches. For this analysis, median split was performed for the variable age. For all three groups, the mean curves over the four measurement time points were presented graphically, with the 95% confidence interval for each mean value. In case an interaction was significant, additional figures were presented. All analyses were performed at the 5% significance level using SPSS version 28.

Analysis of missing values for the SF-12 at the first measurement time point revealed that there were 51 missing values (0.40%) across 33 (3.13%) of all 1,056 participants, affecting all 12 items. Little’s MCAR test yielded a non-significant result (χ² = 164.2; df = 166; p = 0.53), suggesting that the missing data were completely random. At the second measurement time point, there were 26 missing values (0.24%) across 16 (1.76%) of the 907 participants, affecting 9 out of 12 items. Little’s MCAR test yielded a non-significant result (χ² = 91.7; df = 133 p = 1.00), supporting the hypothesis of missing completely at random data. At the third measurement time point, there were 18 missing values (0.17%) across 12 (1.39%) of the 866 participants, affecting 9 out of 12 items. Little’s MCAR test showed a non-significant result (χ² = 87.319; df = 75; p = 0.16), supporting the MCAR assumption. At the fourth measurement time point, there were 32 missing values (0.31%) across the 21 (2.46%) of the 853 participants, affecting 10 out of the 12 items. Little’s MCAR test was non-significant (χ² = 118.2; df = 132; p = 0.80), also supporting the MCAR hypothesis. The dropout rate did not significantly differ between digIRENA group (16.3%), IRENA group (22.8%), and the control group (19.1%) (χ² = 5.1; df = 2; p = 0.08).

The demographic profile of the sample showed that approximately two-thirds (67.3%) were male, with the control group having the highest proportion at 73.6% (see Table 1). The mean age of the participants was 54.2 years, with the oldest participants in the control group (M = 55.7 years) and the youngest in the digIRENA group (M = 53.0 years). A vast majority of participants were German nationals (97.6%). More than three quarters (77.7%) were employed full-time. At the last measurement time point (t3), the participants in the digIRENA group (M = 1836 min; SD = 548 min) participated significantly more in the aftercare (t = 5.1; df = 317.6; p < 0.01) than the participants in the IRENA group (M = 1,492 min; SD = 826 min). A one-way ANOVA revealed no significant differences in the duration of structured physical activities outside of rehabilitation aftercare among the groups (F = 0.2; df₁ = 2; df₂ = 834; p = 0.78). Specifically, the mean duration for the digIRENA group was 80.6 min per week (SD = 74.3), for the IRENA group it was 85.3 min per week (SD = 120.4), and for the control group it was 85.4 min per week (SD = 87.9). There were also no significant differences between the digIRENA group (M = 11.84; SD = 3.88) and the IRENA group (M = 12.39; SD = 3.97) when comparing motivation in terms of the relative autonomy index before the start of the aftercare (t = 1.90; df = 746; p = 0.06).

The analysis incorporated data from a total of 805 participants who completed the SF-12 questionnaire at all time points. This included 323 participants from the digIRENA group, 252 from the IRENA group, and 230 from the control group (see Table 2). Figure 1 illustrates that the baseline mean scores were similar across the three groups, showing no significant differences (F = 2.7; df₁ = 2; df₂ = 1,052; p = 0.07). A significant main effect of time was observed (see Table 3), accounting for approximately 7.5% of the within-subject variance (F = 65.4; df₁ = 2.4; df₂ = 1955.0; p < 0.01; η² = 0.075). Both the digIRENA and IRENA groups demonstrated a continuous increase in mean scores from T0 to T2, followed by a plateau (see Figure 1). In contrast, the control group exhibited an increase in mean scores from T0 to T1, after which a marginal decline was observed. A significant interaction effect for time * group was noted (F = 2.5; df₁ = 4.9; df₂ = 1955.0; p = 0.03; η² = 0.006), with the digIRENA group displaying a steeper incline in mean scores from T0 to T2 compared to the other two groups (see Figure 1). However, the effect size was rather small, explaining only 0.6% of the variance in physical health.

Upon incorporating age as a covariate and gender as an additional factor in the ANOVA model (see Table 4), the previously significant time * group interaction was no longer statistically significant (F = 1.3; df₁ = 3.7; df₂ = 1526.3; p = 0.29; η² = 0.003). In contrast, a significant time * age interaction emerged, explaining 0.9% of the variance in physical health (F = 7.6; df₁ = 1.8; df₂ = 1526.3; p = 0.00; η² = 0.009). Using the median split for the age variable, the follow-up analyses showed a more continuous increase for the younger rehabilitants (≤55 years) across all four measurement time points, while for the older rehabilitants (>55 years) there was a decrease in values from the second measurement time point (see Figure 2). In a separate ANOVA, we probed the three-way interaction effect of time * age * group, which yielded no significant results (F = 0.4; df₁ = 4.9; df₂ = 1936.4; p = 0.87; η² = 0.001).

Data were analyzed for a total of 805 participants who completed the SF-12 questionnaire at all measurement occasions. This included 323 participants from the digIRENA group, 252 from the IRENA group, and 230 from the control group (see Table 5). The results indicated that at baseline (T0), participants in the control group had significantly higher scores compared to both the digIRENA and IRENA groups, with digIRENA participants scoring higher than those in the IRENA group (F = 4.0; df₁ = 2; df₂ = 1,052; p = 0.02). All three groups exhibited an increase from T0 to T1, reflected in a significant main effect of time that accounted for approximately 1.8% of within-subjects variance (see Table 6). Participants in the digIRENA group demonstrated the steepest rise from T0 to T2 (see Figure 3). However, a decline in scores was observed for all groups beyond T2. Overall, the time * group interaction was not statistically significant (F = 1,1; df₁ = 4.7; df₂ = 1890.7; p = 0.34; η² = 0.003).

When age was included as a covariate and gender as an additional factor in the ANOVA, the main effect of time was not significant (F = 1.0; df₁ = 2.3; df₂ = 1874.2; p = 0.39; see Table 7). Notably, only the interaction of gender * group exhibited a significant effect (F = 4.6; df₁ = 2.4; df₂ = 1874.2; p = 0.01; η² = 0.006). Subsequent analyses revealed that this effect was attributable to female participants having significantly lower baseline scores compared to males. During the follow-up period, the mean scores for males and females converged. In a separate ANOVA, it was determined that the three-way interaction of time * group * age had no significant effect on mental health outcomes (F = 1.9; df₁ = 4.7; df₂ = 1871.1; p = 0.09; η² = 0.006).

The present study has several strengths. First, the large sample size supports the robustness of our findings, increasing their generalizability and reducing the margin of error inherent in smaller samples. Second, the use of validated measures to assess health outcomes was a strength. Moreover, the longitudinal design allows to assess the development of the individual indicators across time. Third, our study design included two intervention groups (digIRENA and IRENA) and a control group, which allowed us to estimate the effectiveness of digital aftercare. These methodological strengths allow for a comprehensive comparison of the digIRENA, IRENA, and control group.

However, several limitations need to be taken into account. Primarily, the adoption of partial randomization, necessitated by ethical considerations, introduces the possibility of self-selection bias. This bias may manifest as participants with limited digital proficiency opting for traditional IRENA. Secondly, as previously mentioned, the observed discrepancy in initial participant numbers between the digIRENA and control groups may indicate an attrition bias related to group allocation. These factors collectively highlight the challenges in achieving a perfectly balanced experimental design and necessitate a cautious interpretation of the findings within these constraints. Thirdly, the study design may not have captured the true intervention period, given the variability in the commencement of aftercare measures. Additionally, our exclusive reliance on self-report measures might have introduced biases. Incorporating objective metrics, such as actual days of work incapacity, could enrich future studies. Furthermore, we did not systematically collect data on the specific reasons participants declined participation in the traditional IRENA program. This oversight limits our ability to fully understand the range of factors influencing individuals’ decisions against engaging with this aftercare option. Moreover, the unprecedented conditions of the COVID-19 pandemic introduced significant extraneous variables that may have impacted the study’s outcomes. The pandemic’s influence extended beyond the mere availability of traditional IRENA courses, potentially affecting participants’ willingness to engage in face-to-face interventions due to health concerns. This could have led to a preferential decline in IRENA participation or suboptimal engagement during sessions, motivated by fears of contracting the virus. Moreover, the pandemic’s pervasive impact on public health measures and personal well-being might have influenced both mental and physical health outcomes, introducing motivational deficits that hinder proper participation in the aftercare programs (52). In addition, a notable limitation of our study is that the control group independently engaged in structured physical activities, effectively creating their own intervention. This unforeseen variable may have diminished the apparent effects of the structured IRENA and digIRENA programs, as the control group’s self-directed activities contributed to health improvements that could mask the distinct benefits of the formal interventions. Finally, this study did not account for the varying contextual conditions of the participants, such as their living environments, availability of informal care, and social support systems. These factors could significantly influence rehabilitation outcomes by affecting patients’ ability to engage with aftercare programs and adhere to recommended practices.

The findings from this study have several implications for health rehabilitation research and its applied domains. One of the observations concerns the effectiveness of the digIRENA program, which highlights the potential of digital interventions in the rehabilitation sector (22). In today’s digital age, healthcare professionals have the opportunity to seamlessly integrate traditional rehabilitation methods with digital alternatives, allowing them to expand and improve aftercare services (27). This is particularly relevant when addressing the rehabilitation needs of younger cohorts, who are often more adept and receptive to technological interfaces (53). Moreover, the nuances in rehabilitation responses, as reflected in age and gender differences, underscore the urgency of individualized intervention designs. Relying on generalized, standardized approaches may inadvertently attenuate the efficacy of interventions failing to take into account the heterogeneity of rehabilitation recipients (54). Furthermore, it becomes imperative to delve deeper into understanding how digital aftercare influences different forms of motivation, including intrinsic and extrinsic motivation. Such exploration could unveil critical insights into how motivational factors contribute to the adherence and overall success of rehabilitation programs, offering guidance for the design of more effective, engaging, and personalized aftercare strategies.

Despite the observed benefits of the digIRENA program in our study, it is important to recognize the inherent limitations of telerehabilitation, which have been highlighted by existing research. Challenges such as technological barriers, the necessity for patient and provider digital literacy, potential reductions in the therapeutic relationship due to the absence of in-person interactions, and concerns regarding data security and privacy remain significant (19, 20). Moreover, the impersonal nature of digital interventions may not suit all individuals, especially those who derive substantial therapeutic benefit from direct human contact. While our findings suggest a role for telerehabilitation in expanding access to aftercare services, especially under constraints such as those imposed by the COVID-19 pandemic, these limitations underscore the need for a nuanced approach to integrating digital solutions into the rehabilitation continuum. It is essential to balance the benefits of digital aftercare with the value of traditional, face-to-face interactions, ensuring that rehabilitation programs are accessible, effective, and respectful of patient preferences and needs (55).