Development of a UK core dataset for geriatric medicine research: a position statement and results from a Delphi consensus process

An initial exploratory literature review of the literature was conducted to identify common themes included within prospective clinical studies within geriatric medicine research, and measures available (DW, CW, TAJ). Prospective cross-sectional, cohort, and controlled studies within geriatric medicine research conducted in the UK in the previous 5 years (1st Jan 2014 – 31st Dec 2018) were identified from publications in high impact geriatric medicine journals. The approach to identification and exclusion of articles was non-systematic. The results of this and the proposed minimum dataset were presented at the National Institute for Health Research (NIHR) Newcastle Biomedical Research Centre (BRC) academic geriatric medicine event in May 2019 [1].

The results of the literature review, and discussion at the NIHR BRC Newcastle event were used to generate a questionnaire with broad responses (Questionnaire 1, Additional file 2). This questionnaire was distributed to all delegates who had attended the NIHR BRC meeting, and additional expert academic geriatricians. Participants were asked (via short answer questions with no prespecified responses) to comment on the proposed items to be included within the minimum dataset, and how these should be measured. They were also given the opportunity to suggest additional items that should be included, that had not been included in the survey.

In evaluating the responses given at this step, it became clear there were two separate elements to the proposed dataset:

Subsequent steps of the Delphi process were, therefore, designed considering both of these elements, as described below.

All measures that were suggested in Step 2 were included as options within the Step 3 focused questionnaire (Additional file 2). This survey was distributed to everyone who had been invited to participate in Step 2. Participants were asked to declare at this stage whether they considered each aspect of assessment to be “core” to geriatric medicine research, regardless of how it was to be measured. Participants were asked to select single responses for the most appropriate assessment tool and method from each of the options included for each aspect of assessment, regardless of whether or not they selected this item as “core”.

The responses from the Step 3 questionnaire were utilised to generate a further focused questionnaire in Step 4 (Additional file 2). If more than 80% of participants in Step 3 had considered an item core, then this item was proposed as a core item in Step 4, and participants were invited to express their agreement with this. Items that were considered not relevant to the minimum dataset (either core or extended option items) by more than 75% of participants in Step 3 were removed in Step 4. If a measure was proposed by more than 80% in Step 3, participants were invited to express their agreement with this in Step 4. If there was a divide in opinion between two measures, participants were asked to select one of these in Step 3. If there was a divide in opinion between more than two measures, participants were asked to rank these options. The mean ranking was calculated across participants. This questionnaire was distributed more widely, and responses were completely anonymous. As well as those invited to previous rounds, this survey was also distributed to members of the UK Association of Academic Geriatric Medicine, the British Geriatrics Society Research and Academic Development Committee, and via a national mailing list of research active clinicians in Geriatric Medicine, and early career researchers.

A consensus meeting was held virtually on November 5th 2020. Participants were invited via the same channels used to distribute the survey from Step 4. The results of earlier rounds were presented to all attendees, and each item was discussed until a consensus decision was made. All participants were given the opportunity to contribute. This included both senior academic geriatricians and early career researchers/ trainees.

During the consensus meeting, it was felt that the survey responses from the earlier rounds were insufficient to make a consensus decision about the recording of long-term conditions within the core dataset. An agreement was made for a wider distribution of a final survey, asking clinicians to rank their perceived importance of 34 pre-specified conditions from 1 = most important, to 34 = least important. The ten conditions with the lowest mean ranking score were then chosen to be included within the core dataset for geriatric medicine research (Additional file 1).

There is increasing evidence for multiple long-term conditions as a potential driver of poor health and frailty [13, 14] but there is no definitive preferred method of classification of multi-morbidity. Available methods, such as disease counts or co-morbidity indices, have limitations and fail to consider the complexities of accumulating multiple disease processes, all with different and potentially aggregated effects on an individual’s health [15, 16]. We have recorded long-term conditions as a count of the ten most highly rated conditions by the Delphi process. Each condition should be recorded separately as an individual binary variable.

We have also included a medication count. Guidance on what is considered a medication is shown in Table 1. The strict details of medications have not been included within the dataset for simplicity, but we do recommend that these are recorded and collected locally on CRFs. Dependent on the nature of the study, local CRFs may record drug names only, or may also record strength, dosing instructions, and indications.

Functional dependence is an excellent marker of global health and fundamental in the assessment of frailty [17]. We have included a choice of assessment of functional independence depending on the research cohort being assessed. The Barthel Index [18] is recommended for assessment of basic Activities of Daily Living (ADLs) and the Nottingham Extended Activities of Daily Living (NEADLs) [19] is recommended for instrumental ADLs. The chief investigator should choose the most appropriate assessment for the cohort being assessed. Where possible, both tools should be used to avoid ceiling and floor effects. However, we recognise that assessment of instrumental ADLs may be less applicable to research in very dependent populations, such as residents in long-term 24-hour care. The NEADLs consists of many overlapping data variables and can be time-consuming to complete. NEADLs has been shown to have less sensitivity to change than quality of life measures in orthopaedic surgery populations; using this tool as a dynamic measure was not discussed within this consensus process [20]. The Disability Assessment of Dementia (DAD) [21] is recommended specifically for dementia populations, and can be recorded alongside the Barthel Index and NEADLs.

Clinical Frailty Scale (CFS) should be included in all prospective geriatric medicine research studies. It is easy to assess with minimal training, has been validated in multiple settings, and is increasingly embedded into routine clinical care in the UK [22, 23]. Assessment of CFS can be conducted by all researchers, provided sufficient training and support is available [24]. Assessment of frailty by a Frailty Index and/or Fried phenotype are included within the extended dataset.

Assessment of cognition in some form is considered essential to all geriatric medicine research. However, how this is assessed will depend upon the population being examined. It was not possible to reach a consensus decision upon assessment of cognition appropriate for all populations. In hospital populations, all studies should incorporate delirium assessment. As a minimum, this should involve screening with the 4AT [25], and ideally full assessment according to the Diagnostic and Statistical Manual of Disorders 5 (DSM-5) [26]. The 4AT should be recorded separately where both screening and assessment are performed. In stable community settings, formal cognitive assessment should be performed using a recognised tool e.g. Stroop Test [27], Mini-Mental State Examination (MMSE) [28], Montreal Cognitive Assessment (MoCA) [29], Addenbrooke’s Cognitive Assessment III (ACE-III) [30], or Mini-ACE [31]. The Informant Questionnaire on Cognitive Decline (IQCODE) can be used to screen for pre-existent cognitive impairment when formal cognitive assessments are not appropriate, such as in patients with delirium [32].

We also recommend that all prospective geriatric medicine research studies should incorporate some form of patient-reported outcome measures (PROMs). The tools that are used will be dependent on the research population and the research questions, but may include assessment of quality of life (e.g. EQ-5D [33] or Short Form 36, SF-36 [34]), perceived physical function (e.g. Patient Reported Outcome Measures Information System, PROMIS®, Physical Function Short Form 10 [35]), or cognitive function (e.g. Patient-Reported Outcomes in Cognitive Impairment, PROCOG [36]).

We recognise that there are a number of limitations associated with this dataset, and the process of producing it. Firstly, the initial literature review did not take a systematic approach. Although the questions in the first questionnaire were sufficiently broad to overcome this, and responders were able to suggest additional aspects, it is possible that other important aspects were not considered. Secondly, although nurses, allied health professionals, and pharmacists were invited to participate, responders were predominantly doctors. Thirdly, it was not possible to reach a consensus for all items, therefore at times we have taken a pragmatic approach (e.g. we do not make any specific recommendation of tool to be used for cognition but instead suggest options). Considering the limitations of the dataset itself, the total number of items included is small. Whilst this improves efficiency, this may reduce the representativeness of data that is collected and collated if only the core dataset is frequently utilised. The Geriatric Expert Group of the EMA recommends measurement of walking speed, and ideally Short Physical Performance Battery, for baseline characterisation of frailty in clinical trials involving older adults. However, our Delphi process did not reach a consensus that this should be considered core to all research studies. It should be noted that our core dataset is designed to be very broadly applicable to all geriatric medicine research including studies of virtual/remote design, observational studies, and studies in hospitalised patients or nursed in bed. In such studies, assessment of physical performance and handgrip strength may represent unique challenges, although we also acknowledge that handgrip strength and physical performance can be measured remotely, this may not be possible for all research studies. We emphasise that aspects of the extended dataset are strongly encouraged where study design allows, and our guidance on assessment of these will promote standardised approaches.

Fourth, the Delphi process did not explore the concepts of dynamic change, or suggested intervals for measurement. Fifth, we have designed this dataset with prospective research studies in mind, and it is unlikely to be applicable to retrospective research studies. The use of routinely collected clinical data for secondary research analysis is an ever-expanding field; a separate process that considers what data should be extracted, and how this should be conducted would complement this dataset. Sixth, we appreciate that this dataset may have limited permanency, and may need to be continuously updated as new assessment tools become available and more widely utilised (e.g. body-worn sensor data) [60]. We encourage body-worn sensor data within future studies, but at present do not make any recommendations as to a standardised technology and this was not one of the aspects suggested during the Delphi process. Lastly, this dataset was derived from expert opinion only; the dataset has not been specifically trialled to assess how this will improve research conduct. Unless the dataset is widely utilised by researchers within geriatric medicine, it is unlikely to provide significant impact. We plan to disseminate this dataset through available channels including the British Geriatrics Society, the Geriatric Medicine Research Collaborative [1, 3], and the NIHR Ageing Clinical Research Network. Where possible, we recommend citation of this manuscript where our dataset is used to enable utilisation to be tracked. Factor analyses may provide value in determining which specific variables consistently provide the most predictive value against outcomes of value to older people.

These datasets have been designed to be general for all geriatric medicine research. Importantly, we involved researchers with different subspecialty research expertise at each stage of the Delphi process. However, we recognise that further challenges with standardisation of data collection exist within subspecialty areas. We encourage the formation of subgroups to enable the standardisation of specific measures within specific research fields (e.g. sarcopenia, dementia, continence).