Deviations from the published protocol

To be inclusive of data from LMICs, we intended to search the African Index Medicus, Africa Portal Digital Library; Index Medicus for the Southeast Asia Region; IndMed; Latin American and Caribbean Health Science Literature Database (LILACS). However, we decided to omit these after a scoping exercise revealed a lack of controlled trials in these databases, and we considered it was very unlikely that any publications would fulfil our inclusion criteria.

We intended to use the Grading of Recommendations Assessment Development and Evaluation (GRADE) to assess the weight of evidence of the reported outcomes from the included studies [16]. However, there was too much missing information to use GRADE. We have, therefore not presented the GRADE assessment in the paper (see Table S1 in the Online Supplementary Document).

Search strategy

We searched 10 electronic databases (Table 2). The search strategy used medical subject headings (MeSH) and text words related to health literacy, asthma, self-management and controlled trial. The initial search (January 1990 to June 2018) was updated in October 2019. We conducted forward citation on included studies and contacted experts in the field to identify related trials. We did not perform manual searches as no journal(s) emerged as having a particular interest in this topic. There was no language restriction, though we did not find any non-English publications. We searched the databases using PICOS criteria (Table 2). We used the definitions in Table 1 to confirm eligibility.

Study selection and data extraction

After training and quality control, two authors (HS and SNR) independently screened the de-duplicated titles and abstracts. We obtained the full text of potentially relevant studies, and both reviewers independently assessed for eligibility. Disagreements or uncertainties at any stage were resolved by discussion within the team (HP, IY, SGS or PYL).

Studies which had multiple publications (eg, a protocol, trial findings, process evaluations, qualitative studies, translations) were treated as one study, and reference made to the different publications.

We piloted a data extraction form adapted from the Effective Practice and Organisation of Care (EPOC) recommendations for describing interventions [19] and the Template for Intervention Description and Replication (TIDieR) checklist [20]. Two reviewers (HS and SNR) independently extracted data. We contacted authors for any information which was not found within the included paper(s).

Quality assessment

We used the Cochrane Risk of Bias tool [15], and the guidance from the EPOC group [19], to assess selection, performance, detection, attrition, reporting and other potential sources of bias [15]. The risk of bias for each domain was classified as ‘low', ‘high' or ‘unclear' based on the information available [15]. We generated ‘risk of bias‘ summary graphs and figures using Review Manager 5.3 [21].

Outcomes

Outcomes are described in Table 2. We were primarily interested in health outcomes (eg, asthma control; acute attacks) and implementation outcomes (eg, adoption of intervention). Secondary outcomes included intermediate self-management measures (eg, knowledge improvement), health literacy outcomes and impact indicators (eg, cost-effectiveness).

Data synthesis.

We conducted two analyses to answer the two objectives of our systematic review. First, we considered the effectiveness of asthma self-management interventions which addressed health literacy needs compared with the control group. From scoping work, we anticipated that the studies included in this review would vary substantially in design, target populations, outcomes measured and duration of follow-up precluding meta-analysis. We, therefore, conducted a narrative synthesis of the data.

Second, we described and characterised the included interventions using the Behaviour Change Wheel (BCW) framework (Figure 1), which provides a systematic way to describe and characterise the techniques used in the interventions in this review [22,23]. The BCW has three layers; its core components consist of the COM-B system (Capability, Opportunity and Motivation); interactions between these components determine Behaviour [22]. Capability is the individual’s psychological and physical capacity to engage in the behaviour. It includes having the required knowledge and skills. Motivation is defined as processes that contribute towards both reflective and automatic mechanisms that activate or inhibit behaviour. Opportunity includes aspects of the physical and social environment that lie outside the individual that prompt or make behaviour possible. The second layer of the BCW describes the nine functions of interventions that are designed to change behaviour. The intervention functions are; education, persuasion, coercion, training, enablement, modelling, environmental restructuring and restrictions. The third layer of the BCW identifies seven types of policies (eg, legislation, fiscal measures, etc.) that can be applied to deliver these intervention functions [22].

Figure 1.  The Behaviour Change Wheel (BCW). Reproduced with permission from BioMed Central Ltd [22].

It is proposed that specific intervention functions are likely to influence change in the specific target behaviour. This underpins a matrix (Table 3), produced through a consensus exercise amongst behaviour change experts [22], that enables gaps in intervention functions required to impact on the three core components (capability, opportunity and motivation) that govern behaviour change [22].

We plotted the components of interventions in this review onto the matrix. In the mapping process, which was completed independently by two reviewers (HS and KM), we first identified the core components of behaviour that were targeted, and also the intervention functions used in each included study. Through a consensus approach (see Table S2 in the Online Supplementary Document), we plotted our findings within the matrix (Table 3).

Characteristics of included studies

The randomised control trials were conducted from 2011 to 2017; four studies were conducted in high-income countries [24,25,27,28] (three in the United States (US); one in Canada) and one in Turkey (a middle-income country) [26]. Table 4 summarises population characteristics (see detailed descriptions in Table S3 in the Online Supplementary Document).

Participants characteristics: The three US studies included majority and minority populations [24,27,28]. Yin et al. (2017) included mainly Latin Americans (Hispanics); Apter et al. (2011) included mainly African-Americans, and the majority of the population in the study by Macy et al. (2011) was White American. The trial conducted in Canada by Poureslami et al. (2012) included participants from minority Chinese and Punjabi ethnic groups [25]. The study conducted in Turkey by Ozyigit et al. (2014) did not specify the ethnicity of the population [26]. Participants’ asthma status was described as uncontrolled [26,28]; mild intermittent, persistent or moderate-severe asthma [27]; mild asthma [24]. One study did not describe the participants’ level of asthma control [25].

Study setting: Two studies were conducted in primary care settings [26,28]. Three studies were conducted in secondary/tertiary care settings (specialist paediatric [27] or emergency department [24], university-based pulmonary medicine clinic [25]).

Geographical area and socioeconomic status: Four studies were described as set in an urban environment [24,25,27,28]; three described their population as of low socioeconomic status [24,27,28], the fourth had less than a third in the ‘working-class group’ [25]. The non-urban study described the population as living in the most socio-economically under-developed province in the country [26].

Health literacy status of the population: Only three studies measured the level of health literacy of their participants. One study, which used the validated Newest Vital Sign (NVS), estimated that 70% of the study population had limited health literacy level [27]. Two other studies measured the health literacy level of the study population using the Short Test of Functional Health Literacy in Adults (sTOFHLA) (stating that the mean reading comprehension score was ‘adequate’[28]) or the Rapid Estimate of Adult Literacy in Medicine (REALM) (reporting that ‘two-thirds of the study population had an ‘adequate’ level of health literacy’) [24]. Two studies included ‘immigrants’ [25] or ‘illiterates’ [26] as their study population.

Intervention characteristics: Table 4 summarises the interventions (see detailed descriptions in Table S4 in the Online Supplementary Document). All studies had one intervention and one control group [24,26-28] except Poureslami [25], which had three intervention groups [25].

All interventions included education delivered through various methods; one used a face-to-face personalised problem-solving approach [28], two used video-based education [24,25], and two used education with pictorial asthma action plans [26,27] although only one of these explicitly tailored its action plan to low-literacy level [27]. Three interventions were delivered by research assistants [26-28] and one by a respiratory physician [26]. Four studies specified the language used to deliver the intervention; English or Spanish [27,28], ‘native language’ [26], Punjabi or Mandarin [25]. Only two studies specified the duration of the intervention: 20-minute video [24] or four 30-minute problem-solving sessions [28]. Length of follow-up ranged from five weeks to a year [24-26,28]. One study assessed the immediate understanding of a pictorial asthma action plan [27] rather than longer-term outcomes

Quality assessment of the included studies

Only one study was at an overall low risk of bias [27] (Figure 3). The high risk of bias in the other four studies was typically due to no description of random sequence generation or blinding of outcome assessment. Other biases included no specified sample size [26,28] and use of non-validated tools to measure outcomes [25].

Figure 3.  Risk of bias summary: judgement about each risk of bias item for each included study. Green – low risk, red – high risk, yellow – unclear.

Quality of description and replication

All the studies described the rationale for the essential elements included in the intervention, but none were explicitly guided by a theoretical framework. Three studies lacked descriptions of how the intervention was provided:[24,25,27] for example; one report was unclear whether the video-based intervention was provided individually or in a group [24]. Brief descriptions of the interventions are in Table 4; (see detailed descriptions in Table S4 in the Online Supplementary Document).

Effectiveness of interventions on primary and secondary outcomes

The study at low risk of bias did not report any of our primary outcomes. [27]. Two studies (at high risk of bias) reported health outcomes [26,28], one of which reported a positive outcome for unscheduled care [26]. None of the five studies reported on implementation outcomes (such as uptake/completion of the intervention).

Findings are detailed in Table 4 and the key points described below.

Identification of intervention components in relation to the behaviour change

Limited reporting and the lack of effectiveness in the included studies meant that it was not possible to map the components of BCW to effectiveness. The core components of behaviour and the intervention functions used in the included studies based on reported information are provided in Figure 4. Reports were sometimes limited: for example, one intervention described providing ‘patient skills’ in its education video [24], with no further description of what was taught.

Figure 4.  The use of the core components of behaviour in COM-B model in the included studies.

In terms of the use of the BCW core components of behaviour (COM-B), three studies only addressed ‘capability’ in their interventions [24,26,27]. Two studies, at high risk of bias, addressed a combination of capability and motivation [25,28].

In Table 3, we used the published matrix [22] to plot the included studies according to the core components of behaviour change and intervention function. The low risk of bias study used only one intervention function (enablement) [27]. For the high risk of bias studies; two used three intervention functions [25,26], and two studies used two intervention functions [24,28].

Michie et al. (2011) suggest that the core components of behaviour can be linked to the interventions in more than one way. As an example, the use of a pictorial action plan by Ozyigit et al. [26] is ‘education’ as it increases the capability to understand asthma self-management. A pictorial action plan is also a form of ‘enablement’ as it reduces barriers (eg, lack of knowledge/cognitive skills) to self-management of asthma in the event of deterioration. Most of the interventions concentrated on capability components of the behaviour model, and these interventions used functions such as education, training, persuasion and enabling interventions (action plans) to produce behaviour change.

Summary of findings

This review reports the synthesised findings from five randomised control trials. Four studies, at high risk of bias, concluded that their interventions were ineffective; the only study at low risk of bias did not report on health outcomes. The paucity of studies, limitations in study design and diversity of the interventions meant we are unable to draw conclusions about overall effectiveness on any of our outcomes of interest.

Most studies [24,26-28] included in this review did not describe any theoretical framework underpinning the intervention development, although one conducted prior exploratory work to understand the impact of health literacy in the targeted population [25]. Education, training and enablement are the intervention functions used in these interventions, and the content and the method of delivery varied, including video-based [24,25] and a pictorial action plan [27]. All the interventions used components of behaviour change primarily directed at individuals’ (physical and psychosocial) capabilities; two addressed self-motivation; none targeted opportunity).

Interpretation of the findings and comparison with previous findings.

Strengths and limitations of this study

We followed Cochrane methodology to search systematically for trials of interventions addressing health literacy in the specific context of asthma self-management. All the stages in the review were duplicated, including the selection of papers, risk of bias assessment and data extraction. Our decision not to search some LMIC-focused databases may mean we missed some relevant studies, though our initial scoping exercise in discussion with a medical librarian suggested this was unlikely. All the included studies were RCTs though we would have accepted other designs of controlled trials. We defined our outcomes with care, ensuring we looked for standardised measures of asthma symptom control and risk of attacks [11] and we included trials based on an evidence-based definition of limited health literacy [8,9].

We used the BCW, a validated framework to describe each of the intervention functions, and interpretation of the findings was conducted by a multidisciplinary team to ensure accuracy. The primary studies have small sample size and diverse in populations which makes it challenging to draw a conclusion from the reported results. Four studies did not use health literacy definitions or framework to map its interventional design. Unfortunately, less than half of interventions in this review reported on asthma control [26,28] or unscheduled care, [24,26,28] limiting the conclusions we could draw. For example, there were insufficient data to present our findings graphically (eg, in a Harvest plot [41]) or to use the GRADE [16] approach to assess the quality of evidence. There was limited description of some of the interventions. We could not, for example be certain whether the ‘patient skills’ described as being included in educational videos in one trial, [24] covered behaviour change techniques such as demonstration of behaviour and/or instruction how to perform the task.