All projects on the programme, apart from the exact form of the final economic model (see Reflections on what was and what was not successful in the programme), were undertaken as originally intended. However, several limitations emerged, with respect to both the methodology and the execution of the research.

Robust methods, both quantitative and qualitative, were used throughout, which was a strength of the programme. However, methodologies offered by particular studies do raise limitations in the findings from other parts of the programme. For example, from the reviews, the home support models containing education, social support and behaviour management appeared most effective. In addition, the DCEs did not point to memory aids as the most preferred intervention by people with dementia and their carers. It is possible, therefore, that the intervention evaluated in our trial was not what people wanted or needed, which may have limited its efficacy.

In our analysis of costs to people with dementia and carers and their relationships to formal care (see Workstream 3, Analysis of costs to people with dementia and carers and their relationship to formal care), the sample of 28 people consulted to provide feedback on the type and hours of their own and formal care was small. However, our decision to focus on people with moderate to severe dementia was appropriate, as a group with milder dementia would be likely to have minimal care needs. However, it would have been interesting and important to explore wider personal and societal costs that were not part of the original study design; for example, the costs of informal care, such as informal carers being unable to work or have their own leisure time, and reduced hospital admissions for patients with dementia.

The outcome measures chosen for both the trial and the observational study were ones used previously in other dementia care studies, and for which there were data already available to assist in sample size calculations. Particularly for ADL (i.e. BADLS, the primary outcome in both studies), there are data available on the minimum clinically important difference to estimate expected effects for trial sample sizes. There are no such data available for other outcomes that may be important, such as quality of life (DEMQOL), and certainly none for other outcomes of value that have been developed more recently. This is a current challenge in trial science for dementia studies. Although improvements in, or at least maintenance of, ADL is an appropriate outcome measure to investigate effectiveness of care in later-stage dementia, it may not have been sensitive to potential effects of the trial intervention in early stages. In addition, in a wider sense, a measure of functioning like BADLS may not reflect an outcome that is viewed as important by people with dementia and their carers. BADLS is also limited in being an informant-rated measure. One of our inclusion criteria for entry to the trial was that the person with dementia should have a carer in touch with them who could comment on their functioning, using the BADLS. This carer need not have been living with the participant and in our trial a large proportion (37%) were not. Nevertheless, using BADLS as an outcome would exclude participants with dementia in particular circumstances, such as those without any contact with friends or family. This could compromise future trial designs for more vulnerable and isolated people with dementia and other outcome measures may be more appropriate.

There were challenges in executing the research. Most of the participants in the trial were white. It is possible that there was recruitment bias here in that people with dementia from certain cultural backgrounds may not access mainstream services, such as memory clinics, which was the access point for the trial. Non-white participants may therefore have been harder to recruit, but could have possibly benefited from the intervention. There were delays and difficulties in recruitment to both the trial and the observational study. These led to delays in data collection and some limitations in the data available for the final analyses. For example, the DESCANT did not recruit to its target sample size of number of patients randomised or followed up. There was significant attrition (26%), slightly more than anticipated, and so there were missing follow-up data for outcome measures. For the observational study, attrition was 25%, which was in line with expectations in this more vulnerable and frail population. Although the follow-up period for both studies was relatively short (6 months), this can be a long time in the lives of people with dementia and their carers. Circumstances change and the condition fluctuates. Participants did drop out or refuse follow-up interviews, as is their right to do. These delays and difficulties are common in dementia research, particularly in trials, where recruitment is often difficult. Nevertheless, we achieved large target sample sizes, particularly for the observational study, which recruited from a more vulnerable population. However, to achieve this necessitated extensions to the programme. For the trial, this also needed negotiations around excess treatment costs, initially with Clinical Commissioning Groups and then local CRNs, as governance arrangements changed during the life of the programme.

The research was mainly undertaken in England (with one trial site being in Wales), which may limit the generalisability of the findings to other countries with different health and social care systems. However, the reviews examined international literature that were designed to achieve similar aims to those explored in this programme, as a whole, and had findings relevant to international developments in home support for dementia.