Effect of In-Person Delivered Behavioural Interventions in People with Multimorbidity: Systematic Review and Meta-analysis

Interventions were coded for BCTs using the Behaviour Change Technique Taxonomy (v1) [14] by two researchers (MJä and GZ). The BCT taxonomy is a reliable method for specifying, interpreting, and implementing the active ingredients of interventions to change behaviours. The BCT Taxonomy v1 contains a cross-domain, hierarchically structured taxonomy of 93 distinct BCTs with labels, definitions, and examples [14], and it is a useful method for both research and practice. Each of the researchers coded all the interventions independently. Disagreements were resolved through discussion, and a third reviewer (MJo) mediated where a consensus could not be reached. MJä and GZ are trained in using the taxonomy and practised coding BCTs before this task via the online BCT community (https://​www.​bct-taxonomy.​com/​). All the intervention elements that contain specific BCT were coded. Only interventions (components) that closely correspond to the definitions of the BCTs provided in the taxonomy were coded. Authors were contacted if data was missing or unclear, and intervention protocols (or manuals) were requested to aid the BCT coding, if they were not included in the RCT publications or as additional materials.

The two reviewers (AB and LKH) independently assessed the internal validity of all included studies using the Cochrane ‘Risk of Bias Tool’ (version 2.0). This tool includes the following domains: (1) bias arising from the randomisation process; (2) bias due to deviations from the intended interventions; (3) bias due to missing outcome data; (4) bias in measurement of the outcome; and (5) bias in selection of the reported result. Within each domain, the two reviewers answered one or more signalling questions (e.g., Was the allocation sequence random? Were participants aware of their assigned intervention during the trial?) which led to judgments of ‘low risk of bias’, ‘some concerns’, or ‘high risk of bias’. The judgments within each domain lead to an overall risk-of-bias judgment for the assessed outcome [21]. Disagreements were resolved through discussion until consensus was reached. The overall quality of evidence for the estimates was evaluated using the GRADE (Grading of Recommendations Assessment, Development and Evaluation) approach [31]. The GRADE is a systematic approach to rate the quality of evidence across studies for specific outcomes. It is based on five domains that involve the methodological flaws of the studies (i.e., risk of bias), the heterogeneity of results across studies (i.e., inconsistency), the generalisability of the findings to the target population (i.e., indirectness), the precision of the estimates, and the risk of publication bias [31].

We performed meta-analysis to assess the average effect of behavioural interventions on the outcomes of interest using a random-effects model as heterogeneity was expected due to differences in interventions, outcome measures, etc. Statistical heterogeneity was examined as between-study variance and calculated as the I-squared statistic measuring the proportion of variation in the combined estimates due to between-study variance. An I-squared value of 0% indicates no statistical heterogenity between the results of individual studies, and an I-squared value of 100% indicates maximal statistical heterogentity. Standardised mean differences (SMDs) with 95% CIs were calculated for outcome measures of continuous data but measured in different ways (e.g., all studies measured physical activity, but they use different objective tools) and adjusted to Hedges’ g. On the other hand, for outcomes of continuous data measured in the same way (e.g., all studies measured weight loss assessing the BMI), the mean differences (MDs) with 95% CIs were calculated. The magnitude of the effect size of the pooled SMD was interpreted as 0.2 representing a small effect, 0.5 a moderate effect, and 0.8 a large effect [21]. For outcome measures where a meta-analysis was not possible, a narrative data synthesis of the results from individual studies was performed in line with the guidance from the Cochrane Handbook [21]. When several intervention groups were compared to one control group, the number of participants in the control group was divided by the number of intervention groups, and each was analysed as a separate study comparison [21]. Meta-analyses were performed in STATA (V.17.0) using the ‘meta’ command.

Prespecified meta-regression analyses [24] were performed to explain heterogeneity by exploring the association of different BCTs, participants, studies, and intervention characteristics with effect estimates. Given the explorative nature of such analyses, the most commonly reported (at least in 10 studies as per Cochrane Handbook guidelines) patient, intervention, and study characteristics were chosen as moderators, but no prior hypotheses were made on the possible associations. However, since too few studies were included in the meta-analyses for physical activity and weight loss, we did not perform meta-regression analysis for these outcomes according to the Cochrane Handbook [21]. Instead, we investigated the association between BCTs and these outcomes narratively, by calculating the effectiveness ratios (i.e., the ratio of the number of times each BCT was used in an effective trial divided by the number of times the BCT was used in all trials). This was not prespecified. An effective trial was defined as a trial reporting a statistically significant between-group difference (P < 0.05) or a SMD ± 0.2 [21] in favour of the intervention group. This method has been used in published systematic reviews of similar topics [32‐34], is deemed acceptable by the Cochrane Handbook [35], and was only used when at least three study comparisons were available to avoid overinterpreting the results.

We performed two sensitivity analyses to explore the robustness of the findings. First, given that physical activity and physical function are on the same continuum in the International Classification of Functioning, Disability and Health contextualisation, they were pooled together in one meta-analysis [36]. Second, the meta-analysis on health-related quality of life was repeated, including the mental component scores instead of the physical component scores of the SF-12 [37‐40]. This was done due to the fact that both the physical and mental component scores of the SF-12 can be used to measure health-related quality of life. Furthermore, as the majority of the studies included patients with depression and targeted depression symptoms in addition to lifestyle behaviours, we also assessed the effect of behavioural intervention on depression symptoms.

The MOBILIZE project is committed to patient involvement and has so far included patients living with multimorbidity in all aspects of the decision-making process in the project. Their experiences, needs, and preferences play an important role in developing a novel intervention (Collaborate level on the IAP2 Spectrum of Public Participation). For this systematic review, two patient partners of the MOBILIZE project were introduced to the review and provided feedback on what outcomes to include, before starting the review.

The search identified a total of 1226 unique publications, of which 95 individual RCTs were identified and full texts screened for potential eligibility. Ultimately, we included 14 studies (see Additional file 2 for an overview). The included studies were conducted in 7 countries: USA [37, 38, 40‐45], Croatia [46], Sweden [47], Iran [48], Turkey [49], Greece [50], and Taiwan [39] and were published from 2010 to 2019. The study authors of two studies [38, 50] were contacted for clarification on outcome data and for requesting additional data. Both authors replied, clarified, and provided the data requested. The characteristics of the included studies are reported in Table 1.

The overall mean age of the participants (n = 1,378) included in the studies was 58.1 (SD ± 4.7), 50.9% were female, and mean BMI was 32.5 (SD ± 4.6). The most common combination of conditions reported was type 2 diabetes and depression in 6 studies [37, 39, 40, 43, 45, 46], depression and heart failure in 5 studies [38, 42, 48‐50], type 2 diabetes and heart failure in 2 studies [41, 47], and hypertension and type 2 diabetes in one study [44].

All the interventions targeted lifestyle behaviours, including physical activity and healthy diet. The interventions were multifaceted and, in addition to usual care (e.g., counselling from their health care provider), the most commonly used components were exercise therapy in 8 studies [37, 42, 45‐50], cognitive behavioural therapy (CBT) in 4 studies [37‐39, 42], patient education in 3 studies [37, 42, 46], self-care in 2 studies [41, 43], and motivation enhancement therapy [39], pharmacology [43], and behavioural activation [45] in one study. Exercise together with patient education and CBT or behavioural activation was used in 3 studies [37, 42, 45]. The comparator groups included in meta-analyses were usual care (Table 1). Therefore, when several intervention groups were included in an RCT, the between-group difference was reported for all the interventions versus a comparator group. For example, when a study had two intervention groups (e.g., Exercise and CBT) and one comparator group (Usual care), we compared ‘Exercise’ versus ‘Usual care’ and ‘CBT’ versus ‘Usual care’, and reported the results as two separate study comparisons. This procedure is in accordance with the Cochrane Handbook [21]. The BCTs used in the included studies to target lifestyle behaviours such as physical activity and weight loss are reported in Additional file 3. Overall, the BCTs most commonly used were ‘Instructions on how to perform the behaviour’ (BCT 4.1) in all the studies but one [43], ‘Social support unspecified’ (BCT 3.1) in 11 studies [37‐39, 41‐45, 48, 49], and ‘action planning’ (BCT 1.4) in 9 studies [37, 38, 40, 42, 45, 47‐50]. The clusters of BCTs most commonly used were ‘Goals and planning’ and ‘Feedback and monitoring’ which were present 27 times in the 14 included studies.

Physical activity was reported in 8 studies [38, 40, 41, 43‐45, 49, 50], of which 5 used an objective assessment (e.g., accelerometer) [38, 40, 45, 49, 50] and 3 a self-reported tool [41, 43, 44]. Weight loss was reported in 6 studies [37‐39, 44, 45, 50] of which 5 studies reported data about the BMI of the participants and one as kg [44]. Physical function was reported in 7 studies [37, 38, 40‐42, 47, 48] of which 5 studies used an objective assessment (i.e., the 6 minutes walking test) [37, 38, 41, 42, 47] and two used a self-reported tool (i.e., the SF-12) [40, 48]. Health-related quality of life was reported in 10 studies [37‐43, 47, 49, 50]. Characteristics of the outcome measures are reported in Table 1.

Five studies were included in the meta-analysis on physical activity. At the end-of-treatment follow-ups (mean 16 weeks (SD ± 4)), on average behavioural interventions appeared to have little effect on objectively measured physical activity (k = 5; n = 548; SMD 0.38, 95% CI −0.12 to 0.87; I² = 83.6%) (Fig. 1); however, the evidence is uncertain. Only one study [45] reported data on long-term follow-up (24 weeks post randomisation), showing no difference on objectively measured physical activity between the intervention and comparator group (k = 1; n = 29; SMD 0.13, 95% CI −0.58 to 0.84).

Three studies assessed self-reported physical activity [41, 43, 44]. The results of these three studies were summarised narratively as no meta-analysis was deemed eligible due to large differences in reporting of the self-reported physical activity outcome measures. Overall, these three studies reported that the participants in the intervention groups were more physically active than the participants in the control groups at the end-treatment follow-up (mean 33 weeks, SD ± 16). One study [41] reported that the percentage of participants physically active (i.e., having a Community Healthy Activities Model Program for Seniors (CHAMPS) questionnaire score > 6) was 74.5% in the intervention group and 59.5% in the comparator group. Another study [43] reported that the percentage of participants physically active (two or more times per week) was 68.5% in the intervention group and 32.5% in the comparator group. While yet another study [44] reported that the participants in the intervention group improved their physical activity level (assessed with the CHAMPS questionnaires) more than the comparator group (P < 0.05).

Overall, 12 BCTs were reported in at least 3 study comparisons at the end-treatment follow-up, and effectiveness ratios were calculated. Ten of the 12 BCTs tested had an effectiveness ratio of more than or equal to 75%, with the BCT 3.2 ‘social support (practical)’ and BCT 1.4 ‘action planning’ having an effectiveness ratio of 100% (Fig. 2). At the follow-up closest to 12 months, we were unable to calculate effectiveness ratios due to insufficient data. Additional file 4 reports the raw data for calculating the effectiveness ratios.

Five studies were included in the meta-analysis on weight loss [37‐39, 45, 50] with end-of-treatment follow-ups (mean 18 weeks (SD ± 7)). It is uncertain whether on average behavioural interventions had an effect on weight loss (k = 6; n = 356; BMI mean difference −0.17, 95% CI −1.17 to 0.83: I² = 13.3%) (Fig. 3). The study not included in a meta-analysis reported that the intervention group lost 1.8 kg (95% CI −4.3 to 0.8) more than the comparator group [44]. Two studies were included in the meta-analysis with long-term follow-ups (24 months post randomisation) [39, 45] showing uncertainty for the effect of behavioural interventions on weight loss (k = 2; n = 86; BMI mean difference −0.54, 95% CI −2.70 to 1.62; I² = 0.0%) (Additional file 4).

Overall, 11 BCTs were reported in at least 3 study comparisons, and effectiveness ratios were calculated. Five of the 11 BCT tested had an effectiveness ratio of more than or equal to 75%, with the BCT 3.2 ‘social support (practical)’ and BCT 1.4 ‘action planning’ having an effectiveness ratio of 100% (Fig. 4). At the follow-up closest to 12 months, we were unable to calculate effectiveness ratios due to insufficient data. Additional file 4 reports the raw data for calculating the effectiveness ratios.

Ten studies were included in meta-analysis on health-related quality of life at the end-treatment follow-up (mean 17 weeks (SD ± 13)). On average, behavioural interventions improved health-related quality of life (k = 10; n = 1,042; SMD 0.29, 95% CI 0.17 to 0.42: I² = 0.0%) (Fig. 5). Three studies were included in the meta-analysis with long-term follow-ups (24 months post randomisation) [38, 39, 42] and one study was included in the narrative synthesis. Meta-analysis showed that behavioural interventions may improve health-related quality of life (k = 3; n = 233; SMD 0.20, 95% CI −0.05 to 0.46; I² = 0.0%). However, the evidence was uncertain (Additional file 5), and the study included in the narrative synthesis showed no difference between the intervention and comparator groups [46]. We did not conduct meta-regression analyses or effectiveness ratio for health-related quality of life due to the absence of statistical heterogeneity in the meta-analysis.

Eight studies were included in meta-analysis for physical function at the end-of-treatment follow-up (mean 12 weeks (SD ± 5)). On average, behavioural interventions improved physical function (k = 8; n = 734; SMD 0.42, 95% CI −0.12 to 0.73: I² = 69.5%) (Fig. 6). Meta-regression analysis showed that increasing age was associated with higher effect sizes (slope 0.07, 95% CI 0.02 to 0.13) explaining 65% (Adjusted R²) of the inconsistency of the findings. A higher proportion of female participants in the studies was associated with lower effect sizes (slope −0.02, 95% CI −0.04 to −0.01) explaining 36% (Adjusted R²) of the inconsistency of the findings. Meta-regression analysis also showed that studies using the BCT 2.1 ‘Monitoring of outcome of behaviour by others without feedback’ were associated with a lower improvement in physical function than studies not using this BCT. Additionally, meta-regression analysis showed that studies using a higher number of BCTs for ‘goal setting and planning’ were associated with lower effect sizes (slope −0.45, 95% CI −0.72 to −0.18) and this explained 87% of the variations in the results of the meta-analysis (Additional file 6). Finally, a subgroup analysis showed that behavioural interventions including structured exercise sessions reported a moderate improvement (k = 6; n = 219; SMD 0.56, 95% CI 0.08 to 1.04) compared to interventions without a structured exercise session (k = 3; n = 515; SMD 0.25, 95% CI −0.06 to 0.56); however, there was no statistically significant difference between the two subgroups (Additional file 7).

One study, including two study comparisons, was included in the meta-analysis with long-term follow-up (24 weeks post randomisation). The study assessed physical function with the 6 minutes walking test and showed that behavioural interventions improved physical function (mean difference in metres walked in 6 min: 74.9, 95% CI 0.01 to 149.9; I² = 0.0%).

Eleven studies were included in the additional analysis investigating the effect of behavioural interventions on depression symptoms. At the end-of-treatment follow-ups (mean 14 weeks (SD ± 6)) on average, behavioural interventions reduced depression symptoms (k = 11; n = 1,038; SMD −0.70, 95% CI −0.97 to −0.42: I² = 73.3%) (Fig. 7). At the long-term follow-up assessment, there was no effect of behavioural interventions on depression symptoms (SMD −0.38, 95% CI −1.02 to 0.26: I² = 89.9%). Meta-regression analysis showed that studies including people with a higher BMI (slope 0.9, 95% CI 0.04 to 0.15), and studies using a higher number of BCTs for ‘goal setting and planning’ (slope 0.31, 95% CI 0.04 to 0.58) and ‘Feedback and monitoring’ (slope 0.25, 95% CI 0.02 to 0.48) were associated with a lower reduction of depression symptoms. Depression severity at baseline was not associated with depression symptoms reduction (slope 0.01, 95% CI −0.02 to 0.03).

In the sensitivity analyses analysing physical activity and physical function together, 10 studies (12 comparisons) were included. At the end-treatment follow-ups (mean 14 weeks (SD ± 6)), behavioural interventions on average improved physical activity and physical function when combined (k = 12; n = 849; SMD 0.45, 95% CI 0.16 to 0.73: I² = 69.6%) (Fig. 8).

Ten studies (11 comparisons) were included in the sensitivity for health-related quality of life (i.e., including the mental component scale data instead of the physical component score data for the studies using the SF-12). At the end-of-treatment follow-up (mean 17 weeks (SD ± 13)) on average, behavioural interventions improved health-related quality of life (k = 11; n = 754; SMD 0.30, 95% CI 0.15 to 0.44: I² = 0.0%) (Fig. 9). These results are similar to the primary analysis results (Fig. 6).

The majority of the RCTs applied a proper randomisation process and reported and assessed the outcomes of interest correctly. Due to the nature of behavioural interventions, blinding of participants is challenging as patients receiving the intervention are also the outcome assessors of the patient-reported outcomes (Additional file 8). The overall quality of the evidence assessed using GRADE, including reasons for downgrading the quality of the evidence, is summarised in Table 2. Additionally, some of the included studies where possibly underpowered to detect a between-group difference due to their nature (i.e., pilot studies). However, there was no clear sign of publication bias from the visual inspection of the funnel plots suggesting no sign of small study bias (Additional file 9).

The small improvements for physical activity and weight loss observed are comparable to the short- and long-term improvements seen in behavioural interventions including people with single chronic diseases such as osteoarthritis [51], diabetes [52], heart disease [53], depression, [54] and chronic obstructive pulmonary disease [55]. A possible explanation for these findings is the lack of adherence to the intervention after the studies end. However, greater short-term effects on physical activity and weight loss may be achieved by using the BCT ‘action planning’ and the BCT ‘social support (practical)’, which may potentially have an impact on long-term benefits as well [56]. Nevertheless, the few studies included and the nature of the exploratory analysis prevented us from upgrading the confidence we have in these results. The benefits of behavioural interventions on physical and psychosocial outcomes observed in this systematic review are greater than the findings from a previous systematic review focusing on interventions in multimorbidity in general [17]. The focus on specific combinations of conditions, in our systematic review, may partially explain the differences in results between the two systematic reviews. However, direct comparisons of these findings should be interpreted with caution due to the different populations of the two systematic reviews.

Studies using exercise therapy as part of the behavioural interventions appeared to promote clinically relevant improvements in physical function. This is in line with another systematic review focusing on exercise therapy in people with multimorbidity [15], which found a clinically relevant improvement in physical function following exercise therapy. Furthermore, studies including a higher proportion of males or older people and studies focusing on one BCT for ‘goals and planning’ relative to studies focusing on two or three BCTs for ‘goals and planning’ reported lower improvements in physical function. Similarly, using a higher number of BCTs for ‘goals and planning’ and ‘feedback and monitoring’ may reduce the effect of behavioural interventions on depression symptoms. This may be partially explained by the fact that focusing on many goals and being monitored in many (multiple/various) aspects may be too burdensome for some patients. This is in line with the results of a systematic review investigating the association between BCTs and adherence to exercise in patients with persistent musculoskeletal pain, which is an issue that is also common in people with multimorbidity [57]. Finally, a higher reduction of depression symptoms was seen in people with lower BMI. However, since very few studies were included, this limits our confidence in these results.

It is unclear why interventions targeting lifestyle behaviours, including physical activity and weight loss, improve physical and psychosocial outcomes (e.g., HRQoL and depression symptoms) but not necessarily behavioural outcomes. In this systematic review, two studies did not report an improvement in physical activity [38, 45]. A possible explanation may be that either light intensity activities or sedentary time were not captured as they reported only the time spent performing moderate to vigorous activity [45]. By contrast, increasing physical activity, although being a targeted behaviour of the intervention, was not the primary goal of the study [38]. Physical activity may improve in people with multimorbidity when the intervention explicitly focuses on improving it [58]. Additionally, another possible explanation is that patients may have improved their HRQoL or depression symotoms not necessarily by being more physically active or by losing weight but by adhering to one or more of the other targeted behaviours of the intervention such as quitting smoking, medication adherence, and/or engaging with others. Finally, in dealing with multiple morbidities, patients’ mental representations of their health are more complex. As proposed by the Common-Sense Model of Self-Regulation [59] which is a theoretical model that explicates the processes by which individuals respond to and manage a health threat, the model proposes that individuals navigate affective responses by formulating perceptions of the threat and potential treatment actions, creating action plans for addressing the threat, and integrating continuous feedback on action plan efficacy and threat-progression. People with multimorbidity likely to deal with both the health threat that their conditions present, but also how the threat makes them feel. Our results suggest that more emphasis is put on the latter to improve psychosocial outcomes, including depression symptoms, rather than directing attention to only reducing the threat by engaging in more physical activity.

Behaviour change has been suggested to be contingent on both the capability, willingness, and readiness of the individual [60] and interventions that factor in all these, and recognise the equal status of intra-psychic and external factors in controlling behaviour may be more successful/effective. Therefore, when developing future interventions, a (socio)ecological theoretical approach that take this complexity into account by acknowledging an interplay between factors at the intrapersonal, interpersonal, organisational, community, and public policy levels should be applied [60]. Particularly, we suggest that future studies using behavioural interventions to improve physical activity should test the BCTs and clusters of BCTs that appear to be associated with greater improvements and focus on people with combinations of conditions linked by common risk factors and pathogenesis. Additionally, since the short-term benefits diminish over time, possibly due to lack of adherence to the interventions once the trial has ended. We suggest that future studies to focus on strategies that may help patients adhere to the effective interventions, as well as the investigation of interactions among BCTs, even after the intervention is finished (terminated/completed/discontinued). Similarly, attention should be paid to the mode of delivery of the intervention, which seems to play an important role in behavioural interventions [61‐63]. Furthermore, the content of the interventions received by the comparator groups was often not reported in sufficient details. This is unfortunately common [64], and we suggest that authors of future studies follow, for example, the TIDieR checklist also for reporting comparator groups interventions [30]. Also, we suggest that future studies also measure changes of light intensity physical activity as well as sedentary time, in line with the 2020 WHO guidelines for physical activity [65], and include follow-up assessment close to 12 months and beyond to assess the effect of behavioural intervention over time. Yet, people with multimorbidity experience more health issues than people with single chronic diseases, and this includes physical, psychosocial, and cognitive problems [66]. Finally, given that the majority of the included RCTs included White participants of high SES, we suggest future studies to also focus on a wider range of ethnic populations across different SES [67]. This may change the confidence we have in these results, because people with different ethnic backgrounds and SES may respond differently to a behavioural intervention [68‐71]. This should be considered when planning new interventions, and involving patients in the design of trials may help to improve feasibility and acceptability of the interventions.

To improve physical activity in people with multimorbidity, health care professionals should consider encouraging, educating, and planning together with the patients on what physical activity to do, when, and how (BCT ‘action planning’). Further, health care professionals should advise or provide them with practical social support (BCT ‘social support (practical)’, e.g., provide a membership to a fitness centre and support by a qualified professional trained to deliver exercise therapy such as a physiotherapist or exercise physiologist). This may also help to achieve weight loss. To achieve greater improvements on physical function, we suggest focusing on one of the BCTs for ‘goals and planning’ rather than two or three. Also, it is advisable to avoid observing or recording outcomes of behaviour (e.g., physical activity) without providing feedback which appears to be associated with lower improvements in physical function. Similarly, using a higher number of BCTs for ‘goals and planning’ and ‘feedback and monitoring’ may reduce the effect of behavioural interventions on depression symptoms. Finally, particular attention should be paid to people with higher BMI, as they seem to be the subgroup of people with multimorbidity who benefit the least from reducing depression symptoms from behavioural interventions [72].

The strengths of this systematic review are that we followed the Cochrane Handbook recommendations for performing it and the PRISMA guidelines for reporting it, contacted authors of the included studies to retrieve additional data about their studies, prespecified the main analyses, and followed a structured procedure to code BCTs. There are also limitations. Firstly, the scsarcity of studies matching our inclusion criteria is reflected in the inconsistency of the estimates of the meta-analyses and gave us low power for conducting the meta-regression analyses for physical activity and weight loss. However, we provided a narrative synthesis to investigate the associations between BCTs and these outcomes, thereby providing the readers with useful data applicable in clinical practice and research [32, 33, 73]. Secondly, among the studies reporting socioeconomic status and ethnicity included people of white ethnicity, with a high socio-economic status and with depression and heart failure, and very few studies with other common combination of conditions, limiting the generalizability of the findings to the entire multimorbid population [74, 75]. Finally, we potentially missed some of the BCTs used in the comparator groups who received usual care due to poor reporting of comparator interventions and due to not including digital health interventions, which however, is the focus of our current ongoing work [76].