The current study comprised an online three-round Delphi procedure [17, 18] among experts on measurement instruments used in the field of ABI.
Initially, potential experts were identified based on the professional network of the authors (criterion sampling). Invitees were then requested to propose supplementary potential experts within their own network (snowball sampling). Inclusion criteria were: being experienced in developing and/or evaluating measurement instruments that can be used to assess adults with ABI and being or having been employed in rehabilitation, neurology/neurosurgery, neuropsychiatry, elderly care, or disability care in the Netherlands.
Potential experts were asked by email to participate, before being admitted to the actual procedure. A briefing letter concerning the project, providing the outline and exact planning of the procedure was included with the e-mail. Potential participants were asked to judge whether they met the inclusion criteria and were willing to participate. Upon enrolment, participants provided information on their age, gender, occupation and experience with developing/evaluating measurement instruments. Participants received no compensation.
In the preparation phase, a literature study into outcome domains and measurement instruments relevant to measuring the consequences of ABI was conducted. First, identification of potentially relevant domains was done by the authors prior to consulting the experts and was based on the ICF model  and the ICF core sets for stroke  and traumatic brain injury . Because we anticipated that some domains (such as mental functions) are more applicable to ABI than others, these domains were further divided into second-level categories (such as emotional functioning). Furthermore, since health-related quality of life is an important outcome measure in ABI research, but is not yet a confirmed domain in the ICF model , it was decided that this domain be proposed as a separate factor. Last, the ICF categories ‘activities’ and ‘participation’ were merged, since they have proven to be difficult to differentiate in outcome measurement . A discussion among members of an advisory group consisting of persons with ABI and their informal caregivers (n = 17) confirmed the relevance of the domains that were identified by the researchers.
Second, we looked into measurement use in the field of ABI. In order to improve compatibility with other data collection initiatives and current clinical practice, the authors determined that the minimal dataset should be composed of existing measurement instruments. Accordingly, we made an inventory of measurement use in large studies and other data-shaping initiatives [15, 23,24,25] and benchmarks  in the field of ABI. Furthermore, the systematic review by Tate  and her compendium of tools used for measuring the outcomes of ABI  were consulted and served as a guide for classifying measurement instruments within ICF categories. Finally, evidence-based guidelines of several healthcare disciplines and sectors were checked for recommendations regarding the use of measurement instruments.
A list of requirements for analysing the suitability of measurement instruments for the MDS-ABI was composed through a survey study among members of the knowledge network that the current project is part of (n = 11). This network is comprised of Dutch healthcare professionals and researchers aiming to increase knowledge, cooperation and communication within the field of ABI. The survey study involved a digital form containing proposed requirements for measurement instruments in the MDS-ABI. Respondents reviewed the suggested requirements (necessary/preferable/unnecessary), leading to a set of requirements and preferences, as displayed in Table 1.
All measurement instruments that were identified in the literature study were checked with our criteria and, when meeting all requirements, were entered into the first Delphi round.
The executive phase of the study comprised a three-round Delphi procedure. Experts who agreed to participate received a personal invitation by e-mail containing an anonymised web link to the first Delphi round. In this round, participants were asked to indicate whether every proposed domain was important for the outcome measurement of persons with ABI, with three response options (‘yes’, ‘no’ and ‘this is not my area of expertise’). When this first question was answered affirmatively, the respondent was asked to indicate whether the measurement instruments proposed are suitable for the concerned domain (‘yes’, ‘no’ and ‘no opinion’). The first round was aimed at identifying potentially suitable measurement instruments for the MDS-ABI. Accordingly, respondents were not asked to express a preference for a particular measurement instrument until the second round, but rather were asked to indicate whether a proposed instrument would be suitable for the MDS-ABI.
The aspects that healthcare interventions focus most on, such as reducing symptoms, minimising disability and improving quality of life, can only be assessed by patient-reported outcome measures. Moreover, patient-reported outcome measures avoid observer bias and reduce the administrative burden of clinicians . Therefore, respondents were stimulated to select patient-reported measurement instruments when appropriate. When respondents felt that a particular domain should be measured subjectively as well as objectively (by means of a test or observation measure), multiple instruments could be selected per domain in round one. Domains and instruments that reached a consensus of ‘no’ (≥51% of the respondents) were not taken into the next round.
Throughout all rounds, respondents were given the opportunity to elaborate on their decisions and to recommend additional or alternative domains and measurement instruments. Domains or instruments that were not in our list but proposed ≥ four times (i.e. > 10% of the sample) were included with the proposed domains and instruments in the next round. Measurement instruments that did not meet the requirements, (e.g. by not being freely available) were not presented in further rounds. In order to guide decision making on measurement instruments for the MDS-ABI, the requirements for measurement instruments and samples of the actual measures could be consulted using a hyperlink. Collective responses were fed back to the participants anonymously in the next round, using an information letter that was sent to the participants by e-mail. Participants who did not complete one of the three rounds were excluded from further participation.
The first Delphi round yielded consensus on the inclusion of all proposed outcome domains. Despite the fact that consequences of ABI can occur in all of these domains, measuring the full range of applicable ICF domains would be beyond the scope of a minimal dataset. Therefore, identification of the core domains for the MDS-ABI was carried through in round two. In order to narrow the selection of key domains, we sorted the domains that reached consensus in round one ascendingly by the percentage of ‘Yes’ answers, and asked respondents to reassess the proposed domains, keeping in mind that a minimal dataset needs to be compact and can be composed only of domains that are applicable to all adults with ABI.
Regarding domains for which no consensus was reached on the level of measurement instrument, participants were asked to indicate their preferences, by putting the proposed instruments in order of preference (first place = most preferred, last is least preferred). For some domains (such as emotional functioning), the multiple instruments that were regarded as suitable for the MDS-ABI by respondents in the first round measured different constructs (behaviour and depression/anxiety, respectively). In these cases, respondents had to indicate whether they wanted both, one or none of the instruments to be included in the MDS-ABI. Domains and instruments for which consensus was reached were entered into the concept MDS-ABI.
For some domains, no suitable instrument was identified by the expert panel. Therefore, the use of a screening question was reviewed by the respondents in the third round. Moreover, the third Delphi round contained questions to help clarify the last issues regarding overlap between selected instruments.
The surveys were conducted using Qualtrics software . All analyses were conducted using SPSS statistical software, version 24 . Descriptive statistics were used for responder characteristics. Frequencies were calculated for ratings on multiple choice questions, excluding respondents who indicated they had no opinion on the particular subject. For ranking questions, mean ranks, standard deviation of the mean ranks, sum of ranks and the number of times the instrument was most preferred (first ranks) were calculated to indicate group preferences.
The level of agreement in order to reach consensus in Delphi procedures is not clearly defined . The research team set this level a priori on the majority of respondents (51%) in the current study for multiple choice questions. Consensus for rating questions was defined as an item that has the lowest mean rank, combined with the highest number of first ranks. Any discrepancies between these two outcomes meant that consensus was not reached.