Assessing the importance of predictors of adherence to a digital self‑management intervention for osteoarthritis

Osteoarthritis (OA), the most common joint disorder, is a disabling condition characterized by pain, stiffness, loss of function, and impaired quality of life [1‐3]. With no cure for OA, patient education, self-management, exercises, and weight control are recommended as core treatments [4]. However, implementation of these recommendations in clinical practice is suboptimal, with two recent systematic reviews reporting that only one in three persons with OA receives appropriate non-pharmacological care according to the guidelines [5, 6]. Lack of knowledge about treatment approaches and guidelines by healthcare providers and patients, limited provision of these treatments, and constraints related to access, time, and costs have been identified as major barriers to implementing these recommendations in clinical practice [7‐9].

Digital therapeutics, including smartphone apps, have emerged as an accessible solution to address these barriers with the potential to facilitate better patient-provider interactions and lead to more personalized care [7, 10, 11]. These interventions have shown efficacy in promoting physical activity and improving patients’ pain, function, and quality of life [7, 12, 13]. A major requisite to a digital intervention’s success is adherence [14, 15], defined as the extent to which a person’s behaviour matches agreed recommendations [16]. While clinical trials generally report relatively high adherence to digital self-management interventions for OA, these have mostly short study durations and may not reflect routine practices in real-world settings [10]. Research on adherence, including person-level factors influencing adherence in real-world settings, is very limited [15]. Better insight into factors influencing adherence may help to identify participants who benefit most from digital interventions, and to develop more targeted solutions for those at a higher risk of not adhering to these interventions. Previous research has mainly focused on average associations (e.g. regression coefficients, odds ratios) and statistical significance to identify the predictors of adherence. Such analyses are unable to provide information on the predictive ability of predictors and their relative importance in predicting an outcome [17]. We here explored the relative importance of person-level factors in adherence to a digital platform for OA self-management.

A total of 14,610 persons, mean (SD) age 64.1 (9.1) and 75.5% females, with data on adherence at 3-month follow-up were included (Table 1). Of these, 2682 persons had data on adherence at 12-month follow-up. These two groups had comparable baseline characteristics, with the main exception being the year of enrolment, which was expected (most persons enrolled in 2021 did not have 12-month follow-up due to the study time frame). We also excluded 2030 individuals with no follow-up responses. Compared to those included in the study, these persons were more likely to not report their place of residence, were less physically active, and had more comorbidities and anxiety/depression, as well as a larger proportion of wish for surgery at baseline.

Mean (SD) adherence at the 3-month follow-up (n = 14,610) was 86.3% (16.1), with 24.9% of participants having low adherence. The corresponding figures at 12-month follow-up (n = 2682) were 84.1% (16.7) and 29.6%, respectively. The predictors included in this study explained 5.6% (95% CI 5.1, 6.6) and 8.1% (7.3, 11.6) of variations in adherence at 3- and 12-month follow-ups, respectively (Table 2). Dominance analysis revealed that sociodemographic characteristics were the most important predictors explaining 3.3% (95% CI 2.8, 3.9) and 5.4% (4.0, 7.4) of variations in 3- and 12-month adherence, corresponding to 59.1% (95% CI 51.0, 63.9) and 66.3% (49.9, 70.4) relative contributions, respectively. Despite the differences in point estimates of the contributions of other factors, the bootstrapped 95% CI of their differences were wide and inconclusive (i.e. included 0), suggesting comparable contributions of other factors to adherence at both time points (Table 3). Using low adherence as the outcome generally yielded similar results with two main exceptions: the general health-related factors had more important contributions than OA-related factors at 3-month follow-up and the rank order of factors other than sociodemographic factors changed for 12-month follow-up.

The associations between adherence and single predictors suggested that adherence was generally positively associated with female sex, enrolment in more recent years, living outside three largest cities, lower education, having BMI < 25, higher physical activity, having walking difficulties, no fear of moving, no comorbidity, no anxiety/depression, and not being satisfied with symptoms (Additional file 1: Tables S1–S4). The associations between adherence and age, pain, overall health as well as physical function were generally nonlinear (e.g. 1-year increase in age was associated with 0.3% increase up to age 75 years and 0.5% reduction in ages older than 75 years in 3-month adherence). Among these single predictors, age was by far the most important predictor in both time points, accounting for over 40% of the explained variance in adherence/low adherence (Figs. 1, 2). For 3-month adherence, overall health and place of residence ranked as 2nd and 3rd important predictors, for 12-month adherence the year of enrolment in the programme and physical activity had these positions. For low adherence at 3 months, overall health and physical activity were the most important predictors following age, while education and the year of enrolment in the programme were the most important ones at 12 months. Among OA-related factors, fear of moving was the most important predictor of adherence, followed by pain and physical function.

This study is the first to apply dominance and relative weight analyses to investigate the relative importance of person-level factors in predicting adherence to a digital self-management programme for OA. Our results suggested that the broad array of person-level factors included in the study could explain only a modest amount (4–8%) of variations in adherence. Sociodemographic characteristics were the most important predictors of adherence with age being by far the most important factor accounting for over 40% of explained variance. Moreover, lifestyle, general health-, and OA-related factors had generally similar importance in predicting adherence.

A recent systematic review reported that the rates of adherence to digital self-management interventions for OA are generally over 70% with the highest adherence for app-based interventions [10]. In the present study, we observed an average adherence of around 85% for 3- and 12-month follow-ups, which are close to the highest adherence (88%) reported in previous studies [10]. Albeit it should be noted that around 2000 participants had missing adherence at 3 months meaning that these numbers are an overestimation of actual adherence. If we assume an adherence of 0% for those with missing adherence at 3 months, the average adherence drops to 76% which is still consistent with rates observed in previous studies (over 70%) [10]. This also implies that the actual average 3-month adherence in our sample should be between 76 and 85%. The high adherence rates for digital interventions might be due to the convenience, flexibility, and accessibility advantages of these interventions over face-to-face interventions [31]. Self-selection might be another explanation, as participants in digital interventions are possibly more motivated to take an active role in their disease management and are better informed about the benefits of self-management interventions [10]. In addition, and compared to face-to-face interventions, participants in digital interventions are more regularly monitored, which increases the possibility of the Hawthorne effect, i.e. change in one’s behaviour when being observed [32]. It should be noted that cross-study comparisons of the adherence rates and their influencing factors are complicated due to variability in how adherence is measured [22]. Despite high average adherence rates, around 25–30% of the participants in the present study failed to achieve a desirable level of adherence. This highlights the need for additional measures to stimulate adherence, such as group sessions, peer-monitoring, and reminders to promote adherence [33]. According to the information–motivation–behavioural skills (IMB) model, a commonly used model to explain adherence behaviour, to achieve a behavioural change, one should be: well informed about the behaviour, highly motivated to make necessary behavioural changes, and equipped with required skills to perform a specific health behaviour [34]. While the information aspect of the IMB model is covered by the digital self-management intervention, future attempts to promote adherence among the participants should pay more attention to motivation and skill aspects.

Previous research on predictors of adherence to mobile app-based self-management programmes for OA is limited and mainly investigated adherence to home- or web-based exercise programmes. Results suggested that person-level factors, including sociodemographic characteristics, physical activity, health status, fear of moving, baseline pain/function, and psychological disorders, are associated with adherence [31, 33, 35‐39]. While these studies assessed the average associations relying on statistical significance and hence were not directly comparable to the current study, their results are consistent with our findings, suggesting that person-level factors contribute to adherence to self-management interventions for OA.

However, the person-level factors analysed in the present study explained only a very modest portion (4–8%) of variance in adherence. In comparison, studies on non-OA cohorts explained 19–24% of variances in adherence to exercise [40] and 40% of adherence to a physiotherapy clinical guideline [41]. The modest explained variance in the current study suggests the presence of considerable heterogeneity in adherence which cannot be explained by the person-level factors included in the present study. This also suggests that any action to improve adherence to the digital intervention should be universal, rather than targeted at a specific subgroup defined based on the predictors investigated in the present study, but with a scale and intensity that is proportionate to the level of need (in line with “proportionate universalism” concept [42]). Adherence is a multidimensional complex behaviour influenced by many factors [16, 39] and the World Health Organization (WHO) grouped these factors into five categories: patient-related, condition-related, therapy-related, socio-economic, and healthcare system/team [16]. The predictors included in the present study mainly covered socio-economic and condition-related factors which explain in part the modest explained variance of adherence in the study. Therefore, there is a need for identifying and measuring/collecting data on other person-level and environmental factors that can improve the ability to predict adherence. For instance, intention to engage, self-efficacy, self-motivation, previous experience with exercise, and attitude towards exercise have been suggested as important predictors of adherence to home-based physical therapies [39, 43].

Large sample size, up to 12-month follow-up data on adherence, and using dominance/relative weight analyses coupled with bootstrapping to determine the importance of adherence predictors are the main strengths of the current study. We also acknowledge some limitations of our study. Data used in the study are self-reported and hence prone to biases. In particular, while adherence is recorded through log into the mobile app, its completion is reported by the participants, which might be overestimated. Indeed, potential errors in the measurement of adherence and predictors could have contributed to the modest explained variance in the present study. Self-selection into the digital intervention may limit the generalizability of the findings. For instance, the participants in the digital intervention are more highly educated compared with the general OA population (e.g. 56% had a college/university level education compared with 22% reported for persons with knee or hip OA in Southern Sweden [44]). Moreover, around 23% of participants in Swedish face-to-face self-management programme (“Better management of patients with OsteoArthritis”) had < 10 years of education [45], while in our sample only around 8% had < 12 years of education. The initial uptake of the digital self-management intervention requires Swedish literacy and access and ability to use a smartphone which limits the generalizability further. For most participants enrolled during the year 2021, the missing 12-month adherence was expected due to the study time frame. However, among those with 3-month adherence data enrolled during 2019–2020 (n = 4591), about 52% (n = 2393) did not report 12-month adherence. Although the baseline characteristics of respondents and non-respondents were generally comparable, there were some differences in terms of educational attainment (less non-respondents with college/university education), index joint (more non-respondents with hip OA), wish for surgery (higher proportion among non-respondents), and PASS (a lower proportion with “yes” answer among non-respondents). These differences call for caution in interpreting and generalizing the results for 12-month adherence. We observed high level of adherence over 12 months among the participants which might limit the generalizability to population/setting with low level of adherence. The digital intervention is covered by healthcare system in Sweden, and hence, the results might not be applicable to countries with different healthcare and insurance systems. Due to lack of data, we were not able to assess the importance of some potential determinants of adherence including self-efficacy, self-motivation, participants’ perception of OA and its management, social support, physiotherapist characteristics, quality of participant-physiotherapist interactions, and concurrent treatments (e.g. medications) [39, 43]. These unmeasured predictors may contribute to the unexplained variance and should be explored in future studies.

The findings of this study suggested that while person-level factors were associated with 3- and 12-month adherence to a digital self-management intervention for OA, these factors could only modestly explain the variations in adherence, with sociodemographic characteristics, mainly age, accounting for the greatest portion of the explained variance. These findings highlight the need for identifying other factors that influence adherence to the digital intervention, prior to developing targeted interventions to promote adherence in this population.