The review team is multi-disciplinary and includes content experts, a reference librarian, clinician researchers, and systematic review experts. The review is registered with the PROSPERO international prospective register of systematic reviews (registration number CRD42012003428).
Search strategy for the review
Participants: Studies of non-institutionalized adults with one or more chronic conditions will be included. Chronic conditions are defined as conditions that last, or are expected to last, one year or longer and result in functional limitations and/or require ongoing medical care . A published list of common chronic conditions will be included with the study eligibility forms to guide reviewers .
Interventions: Original reports of randomized controlled trials published between January 2002 and August 2012 seeking to improve the adoption of and adherence to diet/nutrition modification, physical activity, or both. These modifiable behaviors were selected because they are the leading causes of chronic conditions and death in the USA . To focus on the role of healthcare practitioners in health promotion, we will exclude community and environmental interventions (for example media campaigns, legislative measures) and only include physical or virtual interventions delivered from the healthcare setting (primary care, hospital, specialty care, pharmacy, or public health clinic) by healthcare providers/practices for their patients.
Control interventions: Studies with either alternate interventions or control interventions (usual care or no intervention) will be included.
Outcomes: Outcomes of interest include measures of adherence to one or more of the selected health behaviors.
An expert reference librarian will design and conduct the initial search in relevant biomedical databases, including Ovid MEDLINE, Ovid EMBASE, Ovid PsycINFO, Ovid Cochrane CENTRAL, CINAHL and Web of Science in collaboration with content matter experts. Search terms will include controlled vocabulary and text-words (including truncations) for the following concepts: chronic disease, comorbidity, multimorbidity, adherence, compliance, health behaviors, physical activity, diet, weight loss, and behavioral and educational interventions. We will review the citation and reference sections of eligible studies and available reviews. We will also identify additional references through consultation with content experts, and hand searching of key journals and meeting proceedings.
All abstract and full text eligibility and data extraction procedures will be conducted using DistillerSR systematic review software (Ottawa, ON, Canada). Initially, the potential eligibility of each of the abstracts and titles that result from executing the search strategy will be reviewed in duplicate using a pre-defined abstract eligibility form detailing the selection criteria. Full text versions of all potentially eligible studies will be requested. Any disagreements by reviewers will also be retrieved in full text for evaluation. Full text articles (all available versions of each study) will also be independently reviewed in duplicate for eligibility. The reviewers will calibrate their judgments using a smaller set of reports. Subsequently, disagreements will be resolved by consensus; if not possible, by arbitration. Agreement will be measured using the kappa or phi statistics, as appropriate (the latter is appropriate when the distribution of agreement is extreme).
Data extraction will include full description of participants enrolled, eligibility criteria, behaviors targeted, interventions received, control or alternate interventions, and measures of behavior. To better understand the range of interventions and their effectiveness, characteristics of the interventions will be detailed, including use of behavior of change theories, modes of delivery, orientation of the intervention (towards patient or provider), length of intervention, and other common intervention techniques. Full descriptions of the interventions and outcomes will be collected for further analysis using the CCM (see Analysis section). To ensure the quality of data extraction, each reviewer will be trained on the extraction process and each will extract data from five studies in duplicate to ensure reviewer agreement. Conflicts will be resolved by consensus and this calibration process will be repeated until reviewers reach near perfect agreement. In addition to behavioral outcomes of interest, other significant results indicating improved patient capacity will also be recorded (for example quality of life, clinical outcomes).
To assess the methodological quality of randomized trials we will determine the following: how the randomization sequence was generated; how allocation was concealed, whether there were important imbalances at baseline; which groups were blinded (patients, care givers, data collectors, outcome assessors, data analysts); any monitoring for fidelity to the intervention, the loss to follow-up; whether the analyses were by intention to treat; and how missing outcome data was handled. Assessors of quality will work independently and their interobserver agreement optimized through training.
In order to apply the CCM to this analysis, intervention components will be assigned to the model in a two-step process. Initially, the pre-determined intervention components included in the data extraction form will be fit to the CCM a priori. Next, full descriptions of the interventions will be extracted and reviewed to determine how additional elements of the interventions fit into the conceptual model. To determine how intervention components fit the CCM, a review group consisting of physician-researchers, social-behavioral scientists and content matter experts will independently code each intervention component and outcome to the CCM. Inter-rater reliability will be assessed and reported. Disagreements will be resolved by group consensus. To avoid bias, the group categorizing intervention components will be blinded to the outcomes of the studies and will only be presented with the extracted relevant data needed for making the decision. Per the CCM, intervention components will be described as contributing to the patients’ treatment workload, bolstering their capacity, or neutral using the following criteria: workload consists of demands; demands, in turn, are actions that take up time, space, and effort. If intervention components add to these, in terms of traveling a distance, using an amount of time and expending effort (at least two of the three for the purposes of this project) beyond no intervention, then they add to workload. If components somehow decrease at least two of these three factors, then the components decrease workload. If they do not change at least two of these three factors in the same direction, then they will be listed as ‘neutral’ on workload. Intervention outcomes will be assessed the same way. Reviewers will also note whether any study outcomes are in fact related to workload.
Capacity consists of physical, psychosocial, interpersonal, financial, and healthcare-related abilities and resources. If intervention components directly increase or decrease any of these (for example providing diet/exercise counseling is adding to healthcare-related resources available to patients), then they will be rated as increasing or decreasing capacity. Study outcomes, at whatever time of follow-up, will be assessed the same way. Reviewers will also note whether any study outcomes are in fact related to capacity.
We aim to categorize interventions as increased, decreased, or neutral for workload and capacity for each intervention component and measured outcome. This conceptual exercise will not only provide insights into the work required of patients in behavior interventions, but also advance our understanding of how the initial work required of health behavior interventions may in turn increase patients’ capacity as evidenced in intermediate outcomes, and whether it thereby results in significant behavior changes or improved clinical outcomes.
For each study, we will estimate the odds ratio (OR) as the effect size establishing the association between the interventions and adherence to the health behaviors of interest. ORs will be pooled across studies using the random effects model  as implemented in Stata version 12 software (StataCorp, College Station, TX, USA). Heterogeneity across individual studies will be assessed using the I2 index and Cochran’s Q statistical test . Meta-regression will be used to test for interactions between the effect size and a priori determined covariates (subgroup analysis).
These subgroup analyses will evaluate interactions across these outcomes of intervention effects and: a. number of chronic conditions; b. presence of depression (if depression was required for study inclusion); c. the health behavior targeted; d. whether the intervention was provider-facing or patient-facing; e. whether it was a single- or multiple-risk factor behavior intervention; and f. the conceptual model characteristics of the intervention, including capacity-enhancing and workload-inducing components determined by the group consensus. We will test univariate and multivariate models. Data that are heterogeneous or inappropriate for meta-analysis will be evaluated using a meta-narrative approach.