We conducted a focus group study using reflexive thematic analysis (realist orientation) to explore opinions on metrics for functioning in the Swiss neurorehabilitation landscape. Focus groups were held between December 2023 and July 2024. All participants received verbal and written information about the study and provided written informed consent prior to participation. Participants also consented to the audio and video recording of the focus group discussions for research purposes. Data were anonymized during transcription, and any potentially identifiable details were removed before analysis and reporting.

According to the Swiss Federal Act on Research involving Human Beings (HRA), the study did not require formal ethical approval because it involved health care professionals only and did not collect health-related personal data. A formal determination of competence (Req-2025-009-30) was obtained after data collection, serving as administrative confirmation that the project was outside the scope of the HRA.

Invitations were distributed via email, along with a recruitment flyer addressed to therapy leads and medical directors at 8 neurorehabilitation clinics. Clinics were asked to nominate at least 3 clinicians who treated neurological patients with upper limb impairments and had at least 3 years of clinical experience. Prior experience with technology-based assessment was described as beneficial but not required. Nonparticipation was mainly due to organizational constraints (ie, allocating staff for a 90-minute time commitment). Out of 8 invited clinics, 3 agreed to participate in a focus group discussion (one per site).

Focus groups, each lasting 90 minutes, were held in quiet meeting rooms at the participating clinics and guided by a moderator (JP, a physiotherapist with 10 years of experience in mixed clinical and research settings), who had no prior supervisory relationship with the participants. A co-moderator (LM, a physiotherapist) managed audio recordings, field notes, and whiteboard visualizations. Focus groups consisted of 3 to 4 participants. Smaller focus groups (3‐5 participants) are considered appropriate in health research when participants share specific expertise, as they allow more time per participant and facilitate open, more in-depth discussion [36,37].

Participants were familiarized with the principles of focus group discussions and the ICF as the conceptual framework for health assessments. The focus group discussions were conducted according to an interview guideline (Multimedia Appendix 1), which included key questions and practical patient examples, and were structured into 3 parts (Figure 1). Each part began with a discussion of parameters that are relevant and meaningful for quantifying the respective ICF domain, followed by participants indicating the importance of each parameter on individual handouts. Throughout this paper, the term “rating” denotes an indication of importance (yes/no selection) for each parameter.

The first part of the session focused on identifying key metrics within the ICF body function domain (32 items, including ROM, muscle strength, and somatosensory function; Table S1 in Multimedia Appendix 2). This section served as a warm-up to familiarize participants with the domain-specific reasoning process. To stimulate discussion on prioritization and underlying reasoning at the beginning, participants were limited to selecting a maximum of 5 metrics they considered most important in their clinical routines. As this exercise was intended to stimulate reasoning rather than generate interpretable quantitative rankings, the individual votes were not analyzed as outcomes. For transparency, the full table of items is provided in Table S1 in Multimedia Appendix 2.

The second part, focusing on kinematic metrics that quantify aspects of movement quality, began with a video clip showing a patient with moderate one-sided paresis performing the standardized drinking task from both frontal and sagittal perspectives. The drinking task was designed as a representative reach-to-grasp activity to provide a shared, standardized reference for discussing movement quality constructs; however, participants were encouraged to reflect on upper limb movement more broadly, beyond this specific task. Participants then discussed important movement characteristics, clinically relevant features, and requirements for clinical reasoning. The concept of “relevance” was intentionally not predefined or operationalized using explicit criteria. Instead, participants were invited to discuss digital metrics of functioning within clinical scenarios, allowing perceptions of relevance to be implied through how metrics were contextualized, interpreted, and integrated into clinical reasoning. In practice, participants appeared to associate relevance with aspects such as clinical interpretability, usefulness for decision-making, familiarity, feasibility in clinical workflows, and perceived patient-centered value.

Thereafter, participants indicated the perceived importance of a list of kinematic metrics for the drinking task [24,25], which are commonly categorized into the following domains: velocity and movement time, movement smoothness and coordination, movement strategy, and ROM (Table S2 in Multimedia Appendix 2).

The third part included key discussion points on upper limb performance during ADLs, prompted by video clips of the same patient performing the drinking task in a daily-life context. Metric importance was assessed across 10 performance metrics in the domains of arm use duration, arm use intensity, and reaching count and space. Metrics quantifying the duration and intensity of arm use have shown good measurement properties [29] and explain real-world performance well [28] (Table S3 in Multimedia Appendix 2). Despite the limited evidence base for validity, the metrics for reaching count and movement space were included based on previous reports of clinical professionals’ preferences [38].

The participants’ opinions were analyzed using a mixed methods analysis. In qualitative evaluations, reflexive thematic analysis was applied to transcriptions, while ratings were evaluated quantitatively by calculating, for each metric, the proportion of participants who selected it. In the ICF-function section, quantitative evaluations of ratings were excluded due to differences in rating logic.

The reflexive thematic analysis framework [39] enables an interpretative, flexible, and researcher-reflexive engagement with participants’ perspectives, making it particularly well suited to capturing how clinical experts construct clinical reasoning, decision-making, and value-based evaluation of upper limb functioning. Transcripts were imported into the text annotation software Taguette (v1.4.1) [40] and coded by 2 coders (JP and LM) from a realist perspective, assuming that the participants’ words represented their actual experiences and opinions. Codes were generated by adopting an inductive-deductive approach, combining data-driven coding with concepts from existing frameworks (Table 1). Both coders discussed the essential aspects, interpretations, and underlying theories of code labels in regular meetings. Codes were then structured into clusters, forming subthemes and superordinate themes, by 2 researchers (JP and LM), who collated coherent clusters. The meaning, reflexivity, and coherence of themes were then discussed between both researchers. Differences in coding or theme boundaries were addressed through deeper analysis of the coded data. The researchers explored how different interpretations reflected different aspects of participants’ opinions. Where divergent interpretations arose, both perspectives were discussed and revised as the analysis progressed.

Finally, verbatim German quotations and their English translations, together with contextual descriptions of the respective subthemes, were sent to participants for review. All participants confirmed that the translations accurately reflected the original meaning and context. This study was conducted and reported in accordance with the Standards for Reporting Qualitative Research (SRQR) [41]. The completed SRQR checklist, indicating where each item is addressed in the paper, is provided in Checklist 1.

Out of the 8 invited clinics, 3 participated in the focus group study, including a total of 11 participants across 3 focus groups (one conducted in each clinic). Participants presented a range of professional backgrounds, including physiotherapy, occupational therapy, movement science, and medical practice, and varied in clinical roles and experience levels (Table 2). Clinical experience ranged from 4 to 26 years, and research experience ranged from 0 to 20 years. Average durations of focus group discussions were 33 (SD 9.6) minutes for ICF functions, 25 (SD 1.5) minutes for kinematic metrics, and 15 (SD 3.5) minutes for real-world performance, respectively.

Five themes were identified in the analysis: (1) functional requirements to interpret movement quality and performance, (2) essential aspects of movement quality, (3) added value of real-world performance, (4) individualizing what matters, and (5) blending clinical eye and reference data. Each theme comprises between 2 and 4 subthemes, as shown in Figure 2. Illustrative quotations are referenced in the text using an identifier (eg, Q1). The corresponding quotations, including participant IDs and full wording, are provided in Tables S1-S5 in Multimedia Appendix 3. Quantitative ratings of kinematic and real-world performance metrics are presented in Figure 3 and are reported convergently within respective themes.

This theme outlines the functional requirements that clinicians considered essential for meaningful interpretation of movement quality, whether assessed through observational movement analysis or instrumented kinematic analysis. These prerequisites, typically assessed through standard clinical evaluations, provide the baseline context needed to interpret technology-based metrics. Participants highlighted 4 key aspects as particularly decisive for understanding upper limb performance: range of motion, muscle strength, selective muscle control, and grasping function (Table S1 in Multimedia Appendix 3).

This theme highlights movement smoothness, efficiency, and compensatory movements as the 3 key attributes that clinicians use to judge movement quality in relation to the drinking task (Table S2 in Multimedia Appendix 3). These aspects capture how well movements are performed, how effortful they are, and whether patients rely on compensatory patterns.

Clinicians emphasized that what patients do outside the therapy room ultimately determines functional recovery. Three interrelated aspects of real-world performance framed this discussion: the amount and patterns of arm use, efficiency and endurance, and contextual factors influencing ADL behavior (Table S3 in Multimedia Appendix 3).

This theme underscores the importance of individualized selection of assessments, as standardized protocols often fail to capture patient-specific differences. Clinicians highlighted tailoring measures to priority ADLs, balancing functional relevance with feasibility, and adapting to diagnosis-specific challenges (Table S4 in Multimedia Appendix 3).

This theme described how clinicians combine subjective observation, objective reference values, and reasoning across interacting constructs when interpreting movement analysis (Table S5 in Multimedia Appendix 3).

With respect to objective 1, clinicians emphasized that ROM, strength, selective muscle control, and grasping function form the foundation for functional health and for the interpretation of movement analysis. Testing functional requirements (eg, passive ROM) is recommended as critical before conducting standardized observation-based assessments, such as the Fugl-Meyer Assessment for Upper Extremity (FMA-UE) [42] and the Action Research Arm Test (ARAT) [43]. The FMA-UE and ARAT are core assessments for stroke [8,11]. They also serve as a reference for identifying benchmark kinematic [21,44] and real-world performance metrics [29,45] in individuals with neurologic conditions. Notably, impairment measures and digital metrics of functioning are not only highly correlated but also exhibit similar recovery trajectories [46,47]. These relationships are robust yet not interchangeable: task instructions (eg, maximal speed vs comfortable speed) and compensatory movements can alter kinematic profiles without corresponding changes in impairment. The reliance on basic impairment and capacity assessment reflects long-standing rehabilitation research. Our findings reinforce that digital metrics of functioning should complement rather than replace these clinical assessments.

Considering objective 2, the clinicians prioritized digital metrics of functioning that are simple, familiar, and intuitively interpretable. In our study, the drinking task duration, number of movement units, trunk displacement, and dynamic motion trajectories of the elbow and shoulder joints were the metrics consistently selected across professions. These metrics have previously shown good reliability [48] and responsiveness (ie, sensitivity to change over time) [47] in stroke populations, and also inform observation-based assessment of drinking tasks [49]. While these metrics were evaluated in the context of the drinking task, the underlying constructs they represent, such as movement efficiency, smoothness, and compensatory strategies, are relevant across a range of functional upper limb activities. Quantifying upper limb movement quality has been found to be critical for assessing and preventing compensatory movements [38], particularly abnormal interactions among shoulder, elbow, and trunk motion trajectories [25,49,50]. Clinicians in our sample found quantification of compensatory movements or movement intermittency/jerk particularly important, as these metrics could help explain a person’s limited real-world arm use outside the therapy setting.

Revealing patients’ activity performance outside inpatient therapy sessions, or even in habitual environments, is considered highly relevant by clinicians in this and other cohorts [51-54]. This insight into real-world arm use behavior enables goal setting and evaluation. This performance-centered clinical decision-making may be enhanced when therapists are provided with sensor-based measures [51,53,55], whereas therapists showed a preference for activity capacity measures [56].

Advances in measurement technology and data processing enable differentiation of activity types, such as walking [57,58], distinguishing functional from nonfunctional upper limb movements [59,60], and even movement primitives, such as reaching or object transportation [61,62]. About half of the therapists in this study selected more specific metrics, such as reaching count and movement space, which were also reported in other studies on clinicians’ preferences [38,63]. Therapists in our study emphasized daily arm use behavior, which is effectively captured by arm-use duration metrics, including symmetry, both of which are recognized as key indicators in stroke rehabilitation research [28,29,64]. Performance duration metrics are intuitive and straightforward because they directly measure time, making them easy to understand without additional context. Notably, therapists working in rehabilitation after SCI did not consider symmetry metrics important, which lowered the overall percentage of symmetry metrics. Consistent with theme 4 (individualizing what matters), this suggests that metric relevance is shaped by diagnosis-specific reasoning: in unilateral conditions such as stroke, between-limb symmetry is a meaningful comparison, whereas in bilateral conditions such as SCI, it loses interpretive value. Selection of real-world performance metrics should therefore be aligned with the clinical population rather than applied uniformly.

Revealing the meaning behind naked numbers can be challenging and requires complementary information, especially for clinicians who lack experience and knowledge about metrics and their psychometric properties. With respect to objective 3, clinicians expressed a need for visual references and normative data to interpret kinematic metrics, as well as contextual information and normative data for real-world performance metrics. Daily arm use performance can be highly variable, influenced by numerous factors, including health status, activity limitations, functional impairment, personal, and environmental factors [35]. To support interpretation and informed decision-making, clinicians in this study expressed a need for population-specific reference data, alongside cutoff values (eg, percentiles, visual benchmarks, or expected recovery trajectories), to determine whether observed changes are clinically meaningful. In clinical settings, aligning data with daytime and therapy schedules is a practical way to contextualize performance metrics to activity type [51,65]. For goal setting and evaluating therapy effectiveness, the minimal detectable change (MDC) and minimal important change (MIC) are important references for within-person change. Consequently, the successful integration of digital metrics of functioning into clinical reasoning depends on their linkage to patient-specific contexts, diagnostic-specific expectations, and usable normative data to inform treatment planning and progression. Benchmarking schemes for upper limb kinematic and real-world performance metrics [66], together with patient-centered benchmarking approaches [67], offer concrete frameworks that translate changes in digital metrics of functioning into actionable targets for neurorehabilitation practice.

Electronic health records are the backbone of information collection and sharing among clinical professionals. An integration of continuous monitoring of digital metrics of functioning could enable personalized rehabilitation, provide richer representations of a patient’s health status, and enable timely detection of recovery plateaus or declines in real-world performance. To ensure clinical interpretability across heterogeneous neurorehabilitation populations, robust reference datasets are needed that include healthy controls and relevant patient groups, stratified by diagnosis, age, hand dominance, and impairment level. For kinematic metrics, task-specific benchmarks for validated tasks, such as the drinking tasks [25,48,68], could provide z-scores or percentiles alongside MDC/MIC thresholds. For real-world performance metrics, normative data and MIC thresholds exist but remain limited to stroke populations [29].

Future research should explore the optimal convergent set of meaningful, intuitive information that is accessible through graphical illustration. Nearly a decade ago, Ploderer et al [69] proposed dashboard designs that bring together ROM, kinematic analyses, hourly activity patterns, and 3D functional workspace visualizations. The need for integrated, actionable displays remains, but infrastructural and implementation barriers continue to slow adoption [70,71]. Currently, measurement technologies that capture upper limb digital metrics of functioning, such as 3D trajectory analysis and ADL movement space and task detection, remain largely impractical for routine clinical settings. Advances in scalable sensor- or computer vision-based systems are therefore crucial for facilitating precise, clinically relevant visualizations of kinematic metrics and functional reaching space during ADL.

Building on the identified clinical reasoning priorities, Weikert et al [72] developed a learning health system embedded in clinical neurorehabilitation practice. This integrated learning health system enables an end-to-end evaluation of multimodal patient data capture and clinical data presentation. These systems aim to integrate kinematics [73], performance data, cognitive/mental status, patient-reported outcomes, speech, sleep, nutrition, and contextual information (including therapy schedules and temporal patterns) [72]. A key principle for clinical embedding is that clinician-facing outcomes must provide meaningful information to ensure adherence and acceptance.

Such systems directly address two priorities raised by our participants: blending objective metrics with contextual reference information (theme 5) and combining task-based capacity with real-world performance profiles (theme 3). Realizing this potential, however, requires systematic evaluation of the system components, the clinical utility [74], usability [75,76], and the real-world effectiveness of new digital outcome domains. Ultimately, a multipronged approach comprising clinical value systems [77], educational programs, and user-friendly data capture is needed to introduce precision neurorehabilitation at scale [78-80].

Training programs should align with recognized digital-health competency frameworks [81,82] to ensure clinicians acquire the knowledge, skills, and attitudes needed to interpret and act upon digital metrics of functioning. As demonstrated in electronic health record–based educational initiatives, targeted modules on dashboard use and metric literacy can improve clinician proficiency in real-world data-driven workflows [83]. In the present context, educational efforts should therefore focus on foundational metric interpretation and practical integration of digital metrics of functioning into routine decision-making. Such programs could serve as an educational scaffold for translating digital metric evidence into individualized, data-informed rehabilitation planning.

This study was conducted in 3 Swiss rehabilitation clinics, which may limit transferability to other health care systems, settings (inpatient vs. outpatient), and languages/cultures. The low participation rate (3/8 clinics, 38%) was primarily due to organizational constraints in releasing multiple clinicians simultaneously for a 90-minute focus group discussion. Recruitment through clinic nomination may have favored clinicians who are more open to digital assessments, potentially more optimistic in their perspectives, although skeptical views also emerged during the focus groups.

The small number of focus groups and the limited number of participants per group may have influenced group interaction dynamics and the breadth of perspectives discussed. Given the limited number and size of focus groups, thematic saturation cannot be assumed. Furthermore, the small group sizes and site-specific sampling limit the transferability of the findings to clinics with a broader mix of patient populations. Consistent with the exploratory qualitative design, we aimed to provide information-rich insights rather than comprehensive coverage. Accordingly, our findings should be interpreted as exploratory and context-dependent rather than representative or generalizable.

The concept of “relevance” was not operationalized using predefined criteria during the focus group discussions. This approach was chosen to allow clinicians to articulate their own interpretations and reasoning processes. However, participants may have interpreted the term differently, for example, in terms of clinical usefulness, familiarity, feasibility, or patient-centeredness. Such variability may introduce construct ambiguity or response-process bias. While the qualitative analysis allowed us to capture these perspectives through thematic interpretation, future studies could benefit from more explicit operationalization or structured rating criteria.

In addition, the variation in clinicians’ prior experience with digital metrics of functioning may have influenced interpretations. Future research should examine how perspectives differ across professional roles, experience levels, and clinical contexts and assess the impact of targeted training and exposure on the interpretation of digital metrics of functioning in clinical routine.

Moreover, participants were shown an example of a patient performing a drinking task and were evaluated using kinematic metrics validated in this context to provide a tangible, standardized reference and ensure a common frame of reference. While this approach supported understanding of abstract kinematic concepts, presenting a specific task and task-validated metrics may have introduced contextual bias and may limit the full generalizability of the findings.

Both moderator and co-moderator were clinician-researchers, introducing potential social desirability bias despite reflexive views. To mitigate this risk, analytic decisions and theme development were documented through saved versions of code structures, cluster maps, and theme definitions, allowing us to track the evolution of codes and themes.

Finally, we focused on the clinical significance of digital metrics of functioning while intentionally setting aside practical constraints. The feasibility, usability, and data-quality requirements of instrumented assessments (eg, sensor setup and maintenance, nonwear detection, calibration procedures, processing resources, and data privacy demands) warrant dedicated evaluation before routine deployment.

This exploratory study mapped how clinicians in Swiss neurorehabilitation interpret and prioritize digital metrics of functioning of the upper limb. Participants valued kinematic and real-world performance metrics when they were embedded in patient-specific reasoning and paired with contextual information. They consistently favored intuitive metrics, such as task duration, number of movement units, and ROM, over more complex measures whose interpretation requires specialized knowledge. Across the five themes, three priorities emerged for translating digital metrics of functioning into routine practice: (1) stratified normative and reference datasets and meaningful change thresholds, integrated into co-designed dashboards that link task-based kinematic metrics to daily activity profiles; (2) targeted training that builds metric literacy and supports interpretation alongside conventional clinical reasoning; and (3) scalable measurement technologies that integrate seamlessly into clinical workflows, evaluated pragmatically for decision impact, usability, and adoption. These findings should be interpreted as context-dependent rather than generalizable, given the small number of focus groups and the specific clinical and organizational settings sampled. Perspectives are likely to vary across health care systems, professional backgrounds, and levels of experience with digital assessments. Nonetheless, the convergence across professions and clinics on the need for interpretability, contextualization, and clinical embedding offers a clear direction for future work. Centering on clinician and patient preferences will be essential for meaningful integration of digital metrics of functioning into clinical practice and for narrowing the evidence-practice gap.