The protocol for our process evaluation follows a systematic approach for the design and conduct of the evaluation .
Planning the process evaluation
· Working with intervention developers and implementers
A process evaluation requires a degree of independence to appraise the intervention team’s delivery of the trial . Thus, we delegated oversight of the evaluation to an expert not involved in the day-to-day conduct of the RCT but on the wider SENSE-Cog team (CA). Regular process evaluation reports will be made at Trial Steering Committee meetings (see ), and specific meetings with the study chief investigator (IL), the methodologist (EF) and the process lead (CA).
· Overlap of the process and outcomes, and cost-effectiveness evaluations
Due to the complexity of the study, we have embedded the process evaluation within the daily conduct of the outcomes and cost-effectiveness evaluations. Thus, data for all three purposes are being collected concurrently, and some measures may be used for both process and outcomes’ evaluations. During team training, we emphasized the multi-purposing of data and the need to maintain ‘researcher equipoise’.
· Description of the intervention
We have previously detailed the SI components and implementation . In summary, the SI, which is delivered as 10-session programme by a trained ‘sensory support therapist’ (SST) and audiologists/optometrists, is outlined in Figure 1. It involves several components, notably: (1) Identifying and correcting any vision or hearing impairment; (2) supporting adherence with the hearing and/or vision devices, through advice and training in correct use and care; (3) enhancing communication between the PwD and their companion; (4) demonstrating environmental aids and sensory devices; and (5) accessing relevant support services and social networks. The non-intervention group receives ‘care as usual’ (CAU) accesses the services and interventions normally available to PwD and their companions in their respective countries and sites.
· Causal assumptions about how change will be produced
To understand fully the impact of the intervention on the outcomes, and to generalise the findings in the pan-European context of SENSE-Cog, an exploration of causal mechanisms and unanticipated pathways is needed . We will do this by extracting pre-specified mediating variables and qualitative data (from a sub-sample of 30 participant dyads), and evaluating adverse events and unexpected consequences of the intervention. We will also undertake semi-structured interviews with the SSTs at each of the five study sites.
The principal aim of the intervention is to improve quality of life and functional ability by improving sensory function through devices and behavioural change. Dementia-related quality of life comprises the domains of daily activities (activities of daily living and self-care), physical health and wellbeing, cognitive functioning and social relationships . To influence these domains, we have adopted the COM-B component of the Behaviour Change Wheel  as our framework for how the intervention might work. According to this model, behavioural change (‘B’) results from: capability (’C’), the individual’s psychological and physical capacity to engage in the activity concerned; opportunity (‘O’), the external factors that support behavioural change; and motivation (‘M’), the conscious and sub-conscious processes that direct decision making . In Table 1 and Figure 1, using a logic model, we outline how each COM-B element aligns with specific components of the SI, and how these might hypothetically lead to improvements in quality of life, our primary outcome.
Briefly, for capability, the sensory aids (i.e. hearing aids, glasses, and sensory environment modification in the home) and SST adherence support will improve hearing and vision (physical capability), which will enhance sensory-cognitive function, overall functional ability, and reduce neuropsychiatric symptoms [24-28]. The SST will train communication skills and improve knowledge of dementia and sensory impairment (psychological capability). This will increase opportunity by decreasing dependency on companions, enhancing social interactions and reducing loneliness. Adherence support for PwD with sensory devices will enhance hearing , vision , or both . Furthermore, social opportunities will be enhanced through signposting outside the home, thus addressing social isolation, improving social relationships, and providing respite for companions. Regarding motivation, higher sensory-cognitive function and improved neuropsychiatric symptoms will improve self-efficacy, self-esteem, and mental wellbeing [31, 32]. However, motivation may be reduced in dementia, particularly if apathy is present. Thus, the SI will address this through goal setting. We will measure apathy to take account of motivation as a potential moderator, or even mediator, of the intervention’s impact. Greater independence and communication ability in the PwD will reduce companions’ burden and stress, which will also impact positively the overall wellbeing. Attitudes and knowledge training will support change maintenance and relationship quality .
[*Please insert Table 1 here*]
[*Please insert Figure 1 here*]
Figure 1 How the elements of the ‘COM-B Behaviour Change’ model and components of the sensory intervention link together
*Key: BADLs: Bristol Activities of Daily Living Scale ; Bangor: The Bangor Goal-Setting interview ; DEMQoL: Dementia Quality of Life ; FCS: Family Caregiving Role Scale ; HADS: Hospital Anxiety and Depression Scale ; HASK: Hearing Aid Skills and Knowledge Test ; HHIE-25: Hearing Handicap Inventory for the Elderly ; GHQ-12: General Health Questionnaire ; MoCA: Montreal Cognitive Assessment ; NPI-12: Neuropsychiatric Inventory ; RSS: Relationship Satisfaction Scale ; SF-12: Short Form Health Survey ; VA LV‐VFQ-20: Veterans Affairs Low vision Visual Functioning Questionnaire .
· Identification of key uncertainties, and developing a framework for the process evaluation
In Table 2, we identified key uncertainties to address for each SI. We will use a mixed method approach to capture the data for the evaluation. This will involve a variety of instruments, as outlined in Table 3.
[*Please insert Table 2 here*]
[*Please insert Table 3 here*]
· Exploration of delivery
To ascertain whether the SI is delivered (i.e. ‘how’) and enacted (i.e. ‘what’) as intended , we will examine the fidelity and dose (i.e. duration, number and frequency of SI visits) of the delivered intervention. Due to potential burden on participants, we have chosen not to include an external evaluation of fidelity (i.e. independent observer during sessions). Instead, we will rely on the proxy measure of ‘fidelity’ as determined by thoroughness of SST training and supervision, use of therapist manual, and SST logbook recordings of sessions. We will document the nature of the support offered by the intervention, including the type of corrective devices, the environmental changes to support sensory function, the number and types of referral or signposts to extra-trial services. These data will be captured in through participant diaries (the PwD and their companion), and the SST logbook.
Specifically, the PwD diaries will contain Likert style ratings  of acceptability and tolerability of SI visits, including measures of helpfulness, effort, fatigue, understanding and motivation; and how acceptable the corrective sensory devices are. The companions´ diaries will capture data relating to how the PwD engages with the visit, how the PwD is adapting to their sensory aids, and how confident the companion feels in supporting the PwD in using the aids. The SST logbooks will contain details of each visit, the components of the SI delivered, participant response to the intervention and skill in managing their aids. Additionally, the SST logbooks will detail how the SI is specifically tailored to the dyad.
We will assess reach through the representativeness of the sites, the recruitment process (refusal rate, attrition rate) and the representativeness of the study population according to the target population .
· Evaluation of contextual issues
The SENSE-Cog RCT will take place in several different countries and involve three languages (English, French and Greek). Thus, contextual issues, which are external to the intervention itself, need to be carefully considered. These include differences in language, culture, access to services, and the health and social care economy. Context may influence the SST’s ability to foster change in the participant dyad’s circumstances. For example, for social isolation, the SST may recommend attendance at a local lunch club; however, if transportation is not suitable for individuals with sensory and cognitive impairment, the opportunity to take up the offer will be hampered. Likewise, communication training with companions may be differently received in diverse cultural and linguistic contexts. Thus, the same intervention may have divergent outcomes according to the setting in which it is delivered . The dyadic relationship (between the PwD and their companion) should also be considered because the level of support and quality of relationship may vary among dyads. To capture contextual data, we will collect information from the demographic and outcome measures, the participant dyad diaries, the SST logbook, and in-depth qualitative interviews of a sub-sample of participant dyads (n=30 dyads across the sites), as detailed in the SENSE-Cog Trial protocol .
· Sampling and timing of data collection
We will collect characteristics of each participant dyad, including gender, age and support structure at the screening visit, and at baseline, week 18 and week 36 (Table 3 & Table 4). Following each SI visit (for the active arm), participant diaries and SST logbooks will be completed. The sub-sample qualitative interviews will take place within two weeks following the SI (details described in ). Training logs for the SSTs were collected prior to study start. SST supervision logs and fidelity checks of the SST logbooks are being collected throughout the trial. SST interviews will be held within two weeks after the last intervention visit of the last randomised dyad in each site. Briefly, these interviews will explore the experience of having received the intervention, from the perspective of each member of the dyad. The sample size for this sub-sample was selected to achieve theoretical and data saturation. The interviews from all sites will be analysed by using conventional qualitative content analysis  and a grounded theory approach .
[*Please insert Table 4 here*]
To avoid biased interpretation, as recommended by the MRC’s guidance, we will analyse and explore process data arising from the qualitative interviews and contextual factors before the unblinded trial outcomes are known . We will use the process data to generate specific hypotheses (pre-trial explanation before trial outcomes are revealed) regarding factors that moderate and/or mediate the effect of the SI on outcomes, notably quality of life. This will minimise the risk of “fishing” for relationships and falsely significant findings due to multiple testing.
Moderator analyses will be undertaken, with appropriate caution, to investigate any influence of the baseline characteristics of the dyads (e.g. age, gender, type of sensory impairment, level of cognitive impairment, type of companion), and country/site effect on the strength and/or direction of the relationship between the SI and the outcomes. We will undertake mediation analyses to assess the degree to which the impact of the SI on the stated outcomes is a direct effect, or is indirect via the hypothesised mediating factors which will be modelled as latent variables in a structural equation modelling framework.
We will conduct the moderator/mediator analysis only after the final RCT analysis has been completed and the dataset has been un-blinded. We will do this regardless of whether the SI has a significant direct impact on the primary outcome.
We will apply a regression framework, using newer methods and statistical models [54, 55] that improve on traditional approaches (e.g. ). These models can become complex, particularly when controlling for multiple covariates. Thus, depending upon the number and complexity of the hypotheses to be tested, we will assess whether it will be better to analyse each factor separately, or to combine sets of moderators and/or mediators into a ‘conditional process analysis’ . We will conduct separate analyses for those variables available in both trial arms, and those available in the intervention group only (e.g. related to the SI). The latter analyses will help to identify process measures that are part of the SI and may moderate its efficacy (for e.g., number of SST visits, SST experience, and fidelity), using appropriate techniques .