Current issue: problems interpreting behaviour-change interventions during pregnancy

Clinical outcomes are often the major focus of interventions, which can mean that changes in the dietary and physical activity behaviours targeted to achieve the primary clinical outcomes are sometimes overlooked⁽ Reference Phelan, Phipps and Abrams ^{4 –} Reference Vítolo, Bueno and Gama ^{6 )}. This results in unanswered questions such as: what behaviour-change theories and techniques were used; were the desired behaviours and respective influencing factors adequately targeted; and did the target behaviours improve following the intervention? Intensive lifestyle interventions have successfully changed lifestyle behaviours in overweight and obese pregnant women⁽ Reference Wolff, Legarth and Vangsgaard ^{7 ,} Reference Guelinckx, Devlieger and Mullie ^{8 )}. However, financial, time and labour constraints limit the feasibility of intensive interventions in clinical practice⁽ Reference Birdsall, Vyas and Khazaezadeh ^{9 )}. Thus, determining the most successful and cost-effective components of interventions is an important aspect of translational research in this field⁽ Reference Birdsall, Vyas and Khazaezadeh ^{9 )}.

Researchers should be cognisant of the specific behaviour warranting change, the factors that influence the behaviour, and ways to systematically design, implement and evaluate interventions that can change behaviour⁽ Reference Michie, Richardson and Johnston ^{10 )}. This systematic approach is achievable using appropriate evidence, tools and guidance, and facilitates the translation of behaviour-change research into clinical practice. The current paper proposes a decision tree to assist the design, implementation and evaluation of behaviour-change interventions among overweight and obese pregnant women.

Understanding behaviours and behaviour change: addressing contextual influences among pregnant women

Formative qualitative research has identified obese women’s perceptions of barriers and facilitators to changing diet and physical activity during pregnancy, which may be categorised as: physiological; emotional; environmental; cognitive; and interpersonal factors⁽ Reference Sui, Turnbull and Dodd ^{3 )}. Perceived barriers often outnumber facilitators among women, resulting in reduced self-efficacy to initiate change⁽ Reference Sui, Turnbull and Dodd ^{3 )}. Additionally, barriers and facilitators may differ depending on a woman’s environment. Thus, identification of contextually relevant barriers to behaviour change is a critical step in the design of effective interventions.

Some trials have performed pilot or feasibility testing to gain participant feedback on planned interventions⁽ Reference Poston, Briley and Barr ^{11 –} Reference Thomas, Vieten and Adler ^{14 )}. Qualitative data and pilot testing during developmental stages of an intervention can refine methods for large-scale trials, increasing the likelihood of success. Two recent qualitative studies have explored pregnant women’s thoughts, ideas and perceptions on the design of trials to encourage healthy GWG⁽ Reference Goodrich, Cregger and Wilcox ^{13 ,} Reference Thomas, Vieten and Adler ^{14 )}. In the first study⁽ Reference Goodrich, Cregger and Wilcox ^{13 )}, to aid the development of an ongoing lifestyle intervention among overweight and obese African-American pregnant women, participant interviews based on the social ecological model⁽ Reference McLeroy, Bibeau and Steckler ^{15 )} were used to assess women’s perceptions of a healthy GWG and any barriers or enablers they perceived to healthy eating and physical activity⁽ Reference Goodrich, Cregger and Wilcox ^{13 )}. In the second study among low-income, overweight and obese women, focus groups were conducted to inform the design of an ongoing stress-reduction intervention to target the root cause of lifestyle behaviours which influence GWG⁽ Reference Thomas, Vieten and Adler ^{14 )}.

Assessing behaviour change: measuring how an intervention leads to behaviour change

To assess whether an intervention changes behaviour, we must measure any changes in target behaviours as well as behavioural constructs. In previous randomised controlled trials among overweight and obese pregnant women, subjective and/or objective measures of dietary and physical activity behaviours targeted by interventions to improve pregnancy outcomes have not always been reported⁽ Reference Phelan, Phipps and Abrams ^{4 –} Reference Vítolo, Bueno and Gama ^{6 ,} Reference Ong, Guelfi and Hunter ^{16 )}. Trials aiming to reduce GWG via a diet and exercise intervention⁽ Reference Phelan, Phipps and Abrams ^{4 )} and via a dietary intervention alone⁽ Reference Vítolo, Bueno and Gama ^{6 )} did not report changes in diet and physical activity – the behaviours directly influencing GWG. The former study employed behaviour-change theory and techniques⁽ Reference Phelan, Phipps and Abrams ^{4 )}. Similarly, an intense dietary–behavioural intervention aiming to reduce GWG, infant birth weight and the incidence of gestational diabetes, which also employed behaviour-change techniques, reported only the frequency of certain foods consumed the day before the first and final study visits during pregnancy⁽ Reference Quinlivan, Lam and Fisher ^{5 )}. Although this provides some knowledge of the participants’ diet, it is not sufficient to provide a reliable estimate of changes in energy and nutrient intakes which may affect GWG. Some trials measure adherence to intervention recommendations for these behaviours, but not the overall improvement, if any, in dietary or nutrient intakes or activity levels⁽ Reference Thornton, Smarkola and Kopacz ^{17 –} Reference Vinter, Jensen and Ovesen ^{19 )}. The tide may be shifting, however, as recently published antenatal intervention trials⁽ Reference Dodd, Turnbull and McPhee ^{20 ,} Reference Poston, Bell and Croker ^{21 )} have reported changes in dietary and physical activity behaviours and clinical outcomes among overweight and obese pregnant women, namely the incidence of infants born large-for-gestational-age⁽ Reference Dodd, Turnbull and McPhee ^{20 )} and gestational diabetes⁽ Reference Poston, Bell and Croker ^{21 )}. Contemporary trials have also measured changes in psychological behavioural constructs, such as self-efficacy, well-being and depression⁽ Reference Nascimento, Surita and Parpinelli ^{18 ,} Reference Poston, Bell and Croker ^{21 )}. Researchers are also now designing patient-informed interventions which have a strong psychological focus⁽ Reference Thomas, Vieten and Adler ^{14 )}. Measuring alterations in psychological behavioural constructs, or performing mediation analyses, might provide insight to the underlying mechanisms by which an intervention may change behaviour. For example, a recent study protocol details how it will statistically measure the mediating effect of social support, self-efficacy, self-regulation and motivation – as measured by individual questionnaires – among women enrolled in a dietary and physical activity intervention to manage GWG⁽ Reference John, Cassidy and Playle ^{22 )}.

The use of implicit/explicit theory and techniques to change diet and/or physical activity behaviours within antenatal interventions has varied. Although it is possible that behaviour-change theories and techniques are used in all published lifestyle interventions, reporting of the same is insufficiently detailed, as some trials have described neither⁽ Reference Vítolo, Bueno and Gama ^{6 ,} Reference Wolff, Legarth and Vangsgaard ^{7 ,} Reference Ong, Guelfi and Hunter ^{16 )} or described techniques only⁽ Reference Dekker Nitert, Barrett and Denny ^{23 )}. More recently published and ongoing trials have been more detailed in their reporting; detailing theory and defined behaviour-change techniques⁽ Reference Poston, Bell and Croker ^{21 ,} Reference Dodd, Newman and Moran ^{24 )} to successfully change diet and physical activity in overweight and obese pregnant women. Furthermore, intervention mapping protocols, which use an iterative path to map out an intervention and its hypothesised effect on behaviour change, could be of benefit when planning antenatal interventions⁽ Reference John, Cassidy and Playle ^{22 ,} Reference Bartholemew Eldredge, Markham and Ruiter ^{25 )}. The recent detailed reporting of specific behaviour-change theories, techniques and protocols from behavioural science in antenatal interventions is promising⁽ Reference Poston, Bell and Croker ^{21 ,} Reference Dodd, Newman and Moran ^{24 )}, encouraging further research.

Where can we go from here?

To improve the quality of research and create a better evidence base for clinical practice, we recommend researchers universally adopt systematic methods for constructing behaviour-change interventions during pregnancy; incorporating behaviour-change theories and techniques along with formative qualitative research and trial piloting. Recognising the need for systematic reporting of behaviour-change interventions across all disciplines, the Medical Research Council (MRC)⁽ Reference Moore, Audrey and Barker ^{26 )} and the National Institute for Health and Clinical Excellence (NICE)^{( 27 )} in the UK have developed generic protocol guidelines for formulating complex behaviour-change interventions. Another approach to behavioural intervention design is the Multiphase Optimisation Strategy (MOST)⁽ Reference Collins, Nahum-Shani and Almirall ^{28 )}. The MOST approach involves a three-phase development structure in which behavioural interventions are optimised prior to commencement of randomised controlled trials to maximise the suitability of the intervention and its impact within the target population⁽ Reference Collins, Nahum-Shani and Almirall ^{28 )}. Interventions are optimised using any experimental design, but favourably a Sequential Multiple Assignment Randomised Trial (SMART). This differs from classic pilot interventions in that several treatments are tested to identify and select one with optimal components for implementation in the randomised controlled trial phase⁽ Reference Collins, Nahum-Shani and Almirall ^{28 )}.

Additionally, the afore-mentioned intervention mapping, involving logic models, can be used to systematically construct interventions⁽ Reference John, Cassidy and Playle ^{22 ,} Reference Bartholemew Eldredge, Markham and Ruiter ^{25 )}. Intervention mapping protocols and programme impact pathways (PIP) allow researchers to determine the impact of behaviour-change components within interventions on outcomes⁽ Reference John, Cassidy and Playle ^{22 ,} Reference Avula, Menon and Saha ^{29 ,} Reference Mbuya, Jones and Ntozini ^{30 )}. These pathways and protocols can take contextual factors as well as behaviour-change theory into account. Thus, performing intervention mapping or PIP analysis can help researchers establish what behaviour-change components worked, and why and how they worked⁽ Reference John, Cassidy and Playle ^{22 ,} Reference Avula, Menon and Saha ^{29 )}.

Regarding the approaches discussed above, researchers with limited knowledge and experience of behaviour change may benefit from structured support when designing antenatal interventions.

Structured support for developing antenatal behaviour-change interventions

The decision tree proposed in the present paper (Fig. 1) is a step-by-step guide through the design, implementation and evaluation stages of antenatal dietary and physical activity interventions. This framework was developed using information drawn from scientific and grey literature to collate information about the use of behaviour-change theories and techniques and developing behavioural interventions. The decision tree also highlights intervention mapping techniques and demonstrates how these can be used in intervention research. Recent systematic reviews of interventions incorporating diet⁽ Reference Tanentsapf, Heitmann and Adegboye ^{31 )} or diet and physical activity⁽ Reference Thangaratinam, Rogozinska and Jolly ^{32 –} Reference Agha, Agha and Sandell ^{40 )} to improve GWG and the adverse pregnancy outcomes associated with maternal obesity were examined. Additionally, recent Cochrane reviews of dietary and physical activity⁽ Reference Bain, Crane and Tieu ^{41 )} and dietary-only⁽ Reference Tieu, Shepherd and Middleton ^{42 )} antenatal lifestyle interventions to reduce the incidence of gestational diabetes were examined. The most recent studies included in these reviews, among women with a higher BMI, were further reviewed by the authors to gather more information on their specific use of behaviour-change theories and techniques and their reporting of behavioural and health outcomes⁽ Reference Phelan, Phipps and Abrams ^{4 –} Reference Wolff, Legarth and Vangsgaard ^{7 ,} Reference Rhodes, Pawlak and Takoudes ^{12 ,} Reference Ong, Guelfi and Hunter ^{16 –} Reference Vinter, Jensen and Ovesen ^{19 ,} Reference Poston, Bell and Croker ^{21 ,} Reference Dekker Nitert, Barrett and Denny ^{23 ,} Reference Dodd, Newman and Moran ^{24 ,} Reference Callaway, Colditz and Byrne ^{43 –} Reference Renault, Nørgaard and Nilas ^{46 )}. Following this, the behavioural science literature was reviewed for more information on behaviour-change theories and techniques⁽ Reference Michie, Atkins and West ^{47 ,} Reference Fisher, Fisher and Harman ^{48 )}, in addition to published guidelines for the development of lifestyle interventions in non-pregnant populations⁽ Reference Moore, Audrey and Barker ^{26 ,} Reference Craig, Dieppe and Macintyre ^{49 )}. These data were collated and amalgamated among the team. The decision tree was agreed upon through multiple meetings with authors K.M.A., K.L.L. and E.J.O. Thus, we designed a comprehensive tool that can direct researchers to various sources of peer-reviewed information to construct evidence-based antenatal behaviour-change interventions. The procedure for implementing behaviour-change theories and techniques, and the various options for choosing both theories and techniques, are presented in this decision tree (points I–IX) and are based on the work of published authors in the behavioural science field⁽ Reference Michie, Richardson and Johnston ^{10 ,} Reference Michie, Atkins and West ^{47 ,} Reference Fisher, Fisher and Harman ^{48 ,} Reference Michie, West and Campbell ^{50 )}. The decision tree was based on MRC guidelines for designing a well-structured, antenatal lifestyle intervention (largely points IX–X and overarching structure of decision tree)⁽ Reference Moore, Audrey and Barker ^{26 ,} Reference Craig, Dieppe and Macintyre ^{49 )}. References are provided throughout the tree.

Fig. 1 A decision tree – support for developing behaviour-change interventions in overweight and obese pregnancy (GI, glycaemic index; COM-B, Capability, Opportunity and Motivation to change Behaviour; APEASE, Affordability, Practicality, Effectiveness and cost Effectiveness, Acceptability, Side Effect/Safety, Equity; CAL-ORE, Coventry, Aberdeen and London Refined taxonomy)

The MRC guidelines discuss the development of complex interventions in stages, from design through to evaluation⁽ Reference Craig, Dieppe and Macintyre ^{49 )}. Foremost, the design phase of an intervention involves understanding behaviours during pregnancy⁽ Reference Michie, Atkins and West ^{47 )}. The corresponding points on the decision tree are: (I) define the health problem of interest, e.g. the impact of excess GWG on maternal and infant health outcomes; (II) identify the specific behaviour(s) that require change, e.g. frequent consumption of high-fat foods; and (III) identify specific target behaviours using ‘who, what, when’ criteria. The next step is: (IV) to identify barriers and facilitators to adopting new behaviours in a real-world context from the scientific literature. Behaviour-change experts describe this series of steps as key when designing behaviour-change interventions⁽ Reference Michie, Atkins and West ^{47 )}. Ideally, conducting an initial qualitative study (V) would identify barriers and facilitators for selecting appropriate behaviour-change theories and techniques. If a substantial body of qualitative literature already exists to inform the intervention, this may be used to choose behaviour-change theories and techniques. However, in support of MRC guidelines⁽ Reference Moore, Audrey and Barker ^{26 ,} Reference Craig, Dieppe and Macintyre ^{49 )} we suggest that if an intervention previously described in the literature is to be conducted in a new context (e.g. different countries, ethnicities, age groups), then further qualitative work should be carried out.

The next phase (VI) of intervention design is the selection of behaviour-change theories. Researchers may have identified a behaviour-change theory that supports the intervention a priori; for example, Social Cognitive Theory has been used in pregnancy interventions to encourage behaviour change⁽ Reference Poston, Bell and Croker ^{21 )}. Alternatively, researchers can select a theory from a comprehensive list of theories⁽ Reference Michie, West and Campbell ^{50 )}, or use a general model such as the COM-B⁽ Reference Michie, Atkins and West ^{47 )} or the Information, Motivation Behavioural Skills Model (IMBSM)⁽ Reference Fisher, Fisher and Harman ^{48 )}. The COM-B was derived from established theory and encompasses three key behavioural constructs that form the acronym: Capability, Opportunity and Motivation to change Behaviour⁽ Reference Michie, Atkins and West ^{47 )}. COM-B eliminates the need for a specific theory to fit an intervention. Similarly, the IMBSM is a generalised model devised on a variation of these constructs: motivation, information and skills to change behaviours. There are guides available on how to use the COM-B and IMBSM as alternatives to specific theories within interventions⁽ Reference Michie, Atkins and West ^{47 ,} Reference Fisher, Fisher and Harman ^{48 )}.

Once a theory or theoretical basis has been chosen, and target behavioural constructs have been identified, the next step is to: (VII) link behavioural constructs to intervention functions, i.e. the ways in which the intervention is delivered to work⁽ Reference Michie, Atkins and West ^{47 )}. For example, if the function of the intervention is to educate or incentivise, researchers can create materials to convey the message of a particular functional component. The COM-B model centres on the Behaviour Change Wheel⁽ Reference Michie, Atkins and West ^{47 )}, a tool which maps behavioural constructs to a list of intervention functions such as ‘enablement’ or ‘education’. If using the Behaviour Change Wheel, appropriate intervention functions may be systematically selected using APEASE (Affordability, Practicality, Effectiveness and cost Effectiveness, Acceptability, Side Effect/Safety, Equity) Criteria⁽ Reference Michie, Atkins and West ^{47 )}.

Delivery of intervention functions is facilitated through: (VIII) behaviour-change techniques, according to a chosen taxonomy of choice⁽ Reference Michie, Richardson and Johnston ^{10 ,} Reference Michie, Ashford and Sniehotta ^{51 )}. Taxonomies are structured lists of behaviour-change techniques to act as a universal language for their reporting⁽ Reference Michie, Richardson and Johnston ^{10 )}. This universal language eliminates ambiguity that can arise if techniques are described differently between papers, allowing researchers to compare interventions and conduct meta-analyses⁽ Reference Michie, Richardson and Johnston ^{10 )}. At this point, constructing an intervention mapping protocol or devising a PIP (IX) would be useful to hypothesise mechanisms of impact between intervention components, women’s behaviours in context and intervention outcomes⁽ Reference Bartholemew Eldredge, Markham and Ruiter ^{25 ,} Reference Avula, Menon and Saha ^{29 )}. Once ready for implementation, it is strongly suggested that the intervention undergo: (IXi) pilot/feasibility testing⁽ Reference Nascimento, Surita and Parpinelli ^{18 ,} Reference Poston, Bell and Croker ^{21 )} to refine behaviour-change theories and techniques and PIP/intervention mapping protocol, if necessary, prior to implementing a larger trial (IXii).

Evaluation is generally the final step when constructing an intervention⁽ Reference Moore, Audrey and Barker ^{26 ,} Reference Craig, Dieppe and Macintyre ^{49 )}. As an exception, process evaluation should run concurrently with the intervention and be performed by a team independent to the researchers conducting the intervention⁽ Reference Moore, Audrey and Barker ^{26 )} to minimise bias (IXiii). Process evaluation is a type of intervention monitoring that focuses on implementation fidelity and could include the collection of both qualitative and quantitative data. A process evaluation can help determine whether the intervention is delivered appropriately and whether intervention functions work to instigate change⁽ Reference Moore, Audrey and Barker ^{26 )}. In conjunction with intervention mapping, or with the use of a PIP, process evaluation can help establish a mechanism of action between the intervention, behaviours and health outcomes. As proposed by the MRC, planning and conducting a process evaluation ensures the intervention is carefully conceived with respect to its context, the various modes through which it will be implemented and its mechanisms of impact⁽ Reference Moore, Audrey and Barker ^{26 )}. Alternatively, a PIP can be used as a tool to carry out process evaluation, in addition to measuring the mediating effects of an intervention on desired outcomes and the moderating effects of participant characteristics on behavioural outcomes⁽ Reference Mbuya, Jones and Ntozini ^{30 )}. Once process evaluation has been arranged, it is possible to proceed with the full-scale intervention (IXiv).

More traditional intervention evaluation occurs post-intervention, and allows researchers to determine whether the intervention: effectively improves diet and physical activity behaviours; is cost-effective; and is sustainable. Thus, the final stage of the decision tree is: (X) evaluate the intervention economics and clinical or behaviour-change outcomes using published guidelines⁽ Reference Denford, Abraham and Smith ^{52 ,} Reference Glasgow, Bull and Gillette ^{53 )}.

Implications for research and practice

Using established guidelines⁽ Reference Moore, Audrey and Barker ^{26 , 27 ,} Reference Craig, Dieppe and Macintyre ^{49 )} along with the proposed decision tree could help improve the design, implementation, evaluation, reporting and reproducibility of complex behaviour-change interventions in overweight and obese pregnancy.