What are effective program characteristics of self-management interventions in patients with heart failure? An individual patient data meta-analysis

Background— To identify those characteristics of self-management interventions in patients with heart failure (HF) that are effective in influencing health-related quality of life, mortality, and hospitalizations. Methods and Results— Randomized trials on self-management interventions conducted between January 1985 and June 2013 were identified and individual patient data were requested for meta-analysis. Generalized mixed effects models and Cox proportional-hazard models including frailty terms were used to assess the relation between characteristics of interventions and health-related outcomes. Twenty randomized trials (5624 patients) were included. Longer intervention duration reduced mortality risk (hazard ratio 0.99, 95%CI 0.97–0.999 per month increase in duration), risk of HF-related hospitalization (hazard ratio 0.98, 95%CI 0.96–0.99), and HF-related hospitalization at 6 months (risk ratio 0.96, 95%CI 0.92–0.995). Although results were not consistent across outcomes, interventions comprising standardized training of interventionists, peer contact, log keeping, or goal-setting skills appeared less effective than interventions without these characteristics. Conclusion— No specific program characteristics were consistently associated with better effects of self-management interventions, but longer duration seemed to improve the effect of self-management interventions on several outcomes. Future research using factorial trial designs and process evaluations is needed to understand the working mechanism of specific program characteristics of self-management interventions in HF patients.


INTRODUCTION
Heart failure (HF) is a major health concern. Its prevalence is steadily increasing and presently over 10% of the people 85 years and older have been diagnosed with HF. 1 Patients suffering from HF are faced with lifestyle adjustment to prevent deterioration, daily medication intake, and monitoring symptom changes. 2 Interventions to support patients' self-management behavior generally aim to equip patients with skills to actively participate in the management of their chronic condition, through stimulating symptom monitoring and enhancing problem-solving and decision-making skills for medical treatment management and healthy lifestyle. 3 Self-management interventions have received increasing attention as they have been shown to affect a range of outcomes, including all-cause hospitalization and HF-related hospitalization. 4,5 Despite favorable pooled effects, several recent large randomized trials have shown inconclusive results, [6][7][8] raising new questions regarding the effectiveness of those interventions.
A possible explanation for the ambiguous findings across trials can be sought in the diversity of interventions being evaluated. Self-management interventions vary widely in terms of intensity, duration, content, and delivery. 9 Analysis of multiple studies in a meta-analysis or meta-regression may provide insight into the program characteristics that are associated with better outcomes. This knowledge may contribute to the optimal design of effective selfmanagement interventions.
Previous meta-analyses have tried to identify essential program characteristics by focusing on delivery of the intervention to patients. Interventions using face-to-face communication 10 and a multidisciplinary team of interventionists 10,11 were found to be more effective than interventions without these strategies. However, only a small selection of program characteristics was analyzed, isolated from other characteristics, thereby ignoring the possible impact of other characteristics on the outcome. 12 Although aggregated data of studies allow for estimating global effects of program characteristics, using individual patient data (IPD) enables a uniform imputation of missing values and computation of treatment effects across studies. 13 Analytic assumptions, such as uncertainties regarding program characteristics, can be checked with principal investigators, leading to a more reliable analysis. 13 Our IPD meta-analysis aims to identify program characteristics of self-management interventions in patients with HF that affect HF-related quality of life (HF-QoL), mortality, all-cause, and HF-related hospitalization.

Search strategy and study selection
This IPD meta-analysis only included studies of self-management interventions. All individual studies had received approval from their local ethics committees, and this IPD meta-analysis was exempted from the Medical Research Involving Human Subjects Act of the Netherlands by the Medical Ethics Research Committee of the University Medical Center Utrecht. To identify randomized trials on self-management interventions in patients with HF, the electronic databases of PubMed, EMBASE, Cochrane Central Register of Controlled Trials, PsycINFO and CINAHL were searched from January 1985 through June 2013 (for search syntax in PubMed see Supplementary material online, Table S1), as well as reference lists from systematic reviews.
Studies were selected by two independent researchers (N.H.J. and H.W.). Discrepancies were resolved through consensus with a third researcher (J.C.A.T.). Self-management interventions were defined as interventions providing HF-related information to patients and including at least two of the following characteristics: (1) stimulation of sign/symptom monitoring, (2) education on problem solving skills (i.e., self-treatment, stress/symptom management), and improvement of (3) medical treatment adherence, (4) physical activity, (5) dietary adherence, or (6) smoking cessation. Studies were included in the IPD meta-analysis if they (1) fulfilled the requirements of the definition of self-management intervention, (2) had a randomized trial design, (3) included patients with a confirmed diagnosis of HF, (4) compared the self-management intervention to usual care or another self-management intervention, (5) reported data on one or more of the selected outcomes, (6) reported outcome assessment for at least 6 months follow-up, and (7) were reported in English, Dutch, French, German, Italian, Portuguese, or Spanish.

Data collection
The principal investigators of selected studies were invited to participate in this IPD metaanalysis and share their de-identified trial data. The complete list of all requested variables and details on collaboration with principal investigators are reported in the published study protocol. 14 Data from each trial were checked on range, extreme values, internal consistency, missing values, and consistency with published reports.

Outcomes
To identify characteristics of effective self-management interventions across different health outcomes, this study focused on several main outcomes: HF-QoL at 6 and 12 month followup (as measured with Heart Failure Symptom Scale, 15 Kansas City Cardiomyopathy Questionnaire, 16 MacNew Heart Disease Health-related Quality of Life Instrument, 17 or Minnesota Living With Heart Failure Questionnaire 18 ), mortality (time to event, at 6 months, at 12 months), all-cause hospitalization (time to first event, at 6 months, at 12 months), and HF-related hospitalization (time to first event, at 6 months, at 12 months).

Program characteristics
A selection of program characteristics was identified as potential determinants of effective self-management interventions based on literature on self-management and behavior change and their presence across included studies:

1.
Intensity: number of planned contacts between person who delivered the intervention and patient, including planned telephone contacts 19

8.
Problem-solving skills: teaching patient problem-solving skills for management of the condition (yes/no) 21,22

9.
Seeking support: teaching patient skills for seeking support in social network, from caregivers, or from healthcare professionals (yes/no). 21 Information on program characteristics was extracted for the intervention and control arms of all included studies, and confirmed by the principal investigators.

Statistical analysis
Original data from individual studies were merged to create one database. Missing values for baseline variables and outcomes were imputed within studies only using multiple imputation by chained equations (25 imputations), 24 for an overview of missing values per study see Supplementary material online (Table S2). The imputed datasets were used for the primary analysis and results of imputed datasets were pooled using Rubin's rules. 25 All analyses were performed according to the intention-to-treat principle. Studies were analyzed using a one-stage approach, i.e., simultaneously analyzing all observations while accounting for clustering of observations and preserving randomization within studies. 26 The continuous outcomes (HF-QoL at 6 and 12 months) were rescaled to ensure all scales were in similar direction. Effects were quantified by standardized mean differences (SMD) between intervention and control arms and analyzed using linear mixed effects models. Binary outcome data (mortality, all-cause, and HF-related hospitalization at 6 and 12 month follow-up) were analyzed with log-binomial mixed effects models, which estimated risk ratios (RRs). All mixed effects models included a random intercept and random slope for the treatment effect to take clustering within studies into account. For time-to-event endpoints, effects of self-management were quantified by estimating hazard ratios (HRs) using Cox proportional-hazard models, which included a frailty term for each study to account for clustering within studies. This frailty term was assumed to follow a normal distribution. The Cox proportional-hazard models were fitted using the frailty command from the R package survival.
As an intermediary step in the analysis, we estimated the main effects of the selfmanagement interventions in general (i.e., without focusing on specific program characteristics). Main effects have been reported elsewhere, 27 but are presented to enable a comparison of the effects of specific program characteristics with the overall effects.
The primary analysis comprised the identification of program characteristics of effective self-management interventions. Characteristics were evaluated one-by-one in separate analyses. One trial had two intervention arms 5 and these were included as separate interventions in the analysis. To identify the effect of intensity and duration of interventions, the aforementioned models were repeated with the covariate for treatment (and random slope) being replaced by either intensity or duration of the intervention. Hence, the effects of intensity and duration were estimated irrespective of intervention arm. A different approach was applied for analyzing the binary program-specific characteristics. The studies were grouped according to the presence or absence of a binary program characteristic. Two regression models were then applied in parallel to estimate the treatment effect of self-management within both sets of studies. Differences between the two estimated effects from the two sets of studies were tested using a Q-test for heterogeneity. 28 Modification of the effects of program characteristics on clinical outcomes was considered statistically significant if this test yielded P<0.05. Only statistically significant findings from the primary analysis are presented to enable a direct comparison across the different endpoints.
Several sensitivity analyses were performed to assess robustness of findings from the primary univariable analysis (see Supplementary material online for details). All analyses were performed in R for Windows version 3.

Primary analysis of program characteristics
None of the program characteristics in self-management interventions was effective for all endpoints considered. However, several program characteristics showed an effect on one or more endpoints (Table 4). Figure 1

Sensitivity analysis
Observational analysis of the data in a multivariable model confirmed the direction of effects, except for the effects of standardized training (HR 0.55, 95% CI 0.29-1.08) and multidisciplinary teams (HR 0.91, 95% CI 0.64-1.29) on time to HF-related hospitalization, now both appearing advantageous (see Supplementary material online, Table S3). The sensitivity analyses, consisting of a complete-case analysis, repeating the analyses by excluding the largest trial, 6 Table S5).

DISCUSSION
This IPD meta-analysis contributes to the discussion on the optimal design of selfmanagement interventions for patients with HF. Even analyzing 20 trials representing 5624 patients, we could not identify program characteristics that showed a consistent pattern in modifying the effects of self-management interventions across all outcomes considered. However, longer duration of self-management interventions reduced the risk on mortality and HF-related hospitalization with 1-4% for each increasing month of the intervention. Unfavorable associations were observed for standardized training of interventionists, log keeping, goal-setting, and peer contact, but only on specific outcomes.
Meta-analyses of similar interventions have shown that the use of multidisciplinary teams 10,11 and face-to-face contact 10 improved outcomes in patients with HF. Our primary analysis suggested only that a longer duration of self-management interventions was more effective. It is likely that sustained contact over time with a healthcare professional who helps identify signs and symptoms of decompensation may support the patient's selfmanagement. A similar finding was reported by a meta-analysis of HF disease management programs, which found an association between longer follow-up of programs and reduced risk of mortality. 59 In contrast to the previous meta-analyses, 10,11 we found a less favorable effect of multidisciplinary teams compared to a single interventionist on time to first HF-related hospitalization. This effect disappeared after adjustment for other program characteristics, suggesting a correlation with the presence of other characteristics. Since this effect disappears, we do not believe our study contradicts the favorable association reported by previous meta-analyses. The other beneficial characteristic revealed by the prior metaanalysis, face-to-face contact, 10 could not be analyzed in our study since it was known a priori that nearly all eligible interventions used face-to-face contact. Overall, the selfmanagement interventions elicited favorable main effects on HF-QoL and HF-related hospitalization. These effects could not be attributed to any of the binary characteristics considered in our study. The face-to-face contact present in nearly all intervention arms might be a critical characteristic in explaining the favorable effects of self-management interventions and this possibility deserves attention in future research.
From earlier work on social cognitive theory 21,22 and meta-regressions on effective behavior change techniques, 20,23 we assumed that standardized training of interventionists, keeping logs for symptom-monitoring, goal-setting, and contact with peers would positively influence the effect of self-management interventions. However, our findings were counterintuitive and showed that self-management interventions comprising those characteristics resulted in less favorable outcomes than interventions without those characteristics. It may be that studies had commonalities on methodological aspects or on other characteristics which confounded our results, 60 for example, additional (medical) care provided along with the self-management intervention. Inspection of other study characteristics and aggregate baseline variables in tables 61 revealed that there was a tendency for the self-management program characteristics to be particularly present in more recently conducted studies. We hypothesized that treatment effects may have decreased over time, because usual care has evolved due to insights from research (i.e., new treatments, more comprehensive care protocols). Although the post hoc sensitivity analyses did not confirm this hypothesis, differences in usual care given to control patients or additional care given in the intervention arms might still be confounding factors for the observed effects.
The information on usual care was limited and we could not appropriately adjust for the wide diversity in usual care in our analysis.
Without a clear explanation for the unfavorable effects, it would be unjustified to recommend that self-management interventions should not comprise specific program characteristics. The large number of program characteristics analyzed increases the chance of false positive findings and any observed effect therefore should be considered explorative rather than confirmative. 61 Considering the complex nature of self-management interventions, we might even question the extent that researchers should look at isolated program characteristics of complex self-management interventions in a meta-regression analysis, since the interventions were designed as a cohesive compilation instead of separate characteristics. 12 Our findings support the notion that effectiveness of self-management interventions may not be attributable to specific program characteristics, but rather that certain types of interventions show a pattern of effects which is dependent on the context in which the intervention is delivered. 62 From this perspective, this IPD meta-analysis should be considered the first large effort towards identifying characteristics of effective self-management interventions in patients with HF. It applied a careful data collection and analysis, and the causal nature of effect modifiers was addressed by checking the primary findings on confounding factors. Nevertheless, several limitations are worth discussing. First, despite the inclusion of twenty studies and data on 5624 patients, the number of studies was too restricted for multivariable analysis using mixed effects models, limiting causal interpretation of our findings. Second, the use of meta-regression techniques required a categorization of program-specific features. This may have left room for interpretation of categories and may have created large, still heterogeneous sets of studies being grouped together (i.e., goal-setting in one study may have differed from that in another study). Underreporting of intervention details prevented us from creating detailed categories following existing taxonomies like the behavior change technique taxonomy, 63 which deserves attention by future trials. Finally, fidelity to study protocols and adherence to interventions by patients was unknown in a majority of included studies. Process evaluations of behavioral interventions such as self-management interventions have shown that fidelity to study protocols is often compromised, 64,65 consequently patients in the intervention groups might have actually received different program characteristics than assumed. The unavailability of these data prevented assessment of the impact of treatment compliance on the outcomes. 66 This IPD meta-analysis highlights the need for incorporating the complexity of this type of intervention in the study design, e.g. through carefully defining intervention components, planning feasibility studies and process evaluations of intervention delivery alongside trials. This may contribute to a thorough understanding of how the intervention exerts its effects.

CONCLUSIONS
Despite the large numbers of patients included in this IPD meta-analysis, no specific program characteristics could be identified that were clearly associated with better outcomes of self-management interventions. There were indications that a longer duration positively modified the effects of self-management interventions on several outcomes, supporting sustained contact over time between healthcare professionals and patients with HF. Advances in usual care for patients with HF over time may have confounded the effects observed. Future research using factorial trial designs and process evaluations is needed to assess adherence to self-management interventions and understand the mechanism whereby self-management interventions enhance clinical outcomes in patients with HF.

Supplementary Material
Refer to Web version on PubMed Central for supplementary material.  Table 1 Baseline characteristics of heart failure patients in control and self-management intervention arm included in the individual patient data meta-analysis.  Table 2 Description of trials on self-management in heart failure patients included in the individual patient data meta-analysis.  Table 4 Effects of self-management interventions and characteristics in patients with heart failure included in the individual patient data meta-analysis.