Methodological refinement in bibliometric mapping of wearable sensors for stroke rehabilitation

To the Editor,

The article “Global trends in wearable sensors for stroke motor rehabilitation: A bibliometric analysis” by Li et al., recently published in DIGITAL HEALTH, ¹ caught my attention. The authors deserve praise for providing a comprehensive overview of twenty years of research integrating wearable technology with motor recovery after stroke. Their systematic use of CiteSpace and VOSviewer offers valuable insight into evolving research areas, interdisciplinary collaboration, and the growing role of artificial intelligence in neurorehabilitation. For researchers and clinicians navigating this rapidly expanding field, this contribution is both relevant and timely.

However, a few methodological aspects may warrant additional clarification.

First, concerns arise regarding conceptual accuracy when heterogeneous constructs such as wearable technology, accelerometers, smart textiles, inertial measurement units, and sensor systems are grouped under the single term “wearable sensors.” Bibliometric mapping is highly sensitive to the operationalization of the search strategy. As noted by Zupic and Čater, ² the intellectual structure derived from bibliometric analysis largely depends on construct clarity and careful term selection. Combining semantically distinct technological entities may risks artificially inflating clusters and distorting thematic relationships, potentially reflecting methodological aggregation rather than genuine knowledge convergence.

While keyword normalization is a recognized step in bibliometric analysis, the level of aggregation applied may influence the resulting intellectual structure.

Second, although the second-order polynomial regression model produced a high R² value, applying it to cumulative publication growth may create an impression of predictive stability that is mathematically inevitable rather than theoretically justified. Bornmann and Mutz ³ demonstrated that cumulative publication curves naturally generate high goodness-of-fit values regardless of model appropriateness, and that scientific growth typically follows exponential or logistic patterns. Consequently, polynomial fitting may overinterpret structural continuity while obscuring the underlying growth dynamics.

Third, reliance on raw citation counts, total citations, and H-index metrics without field or time-normalization may introduce systematic bias. Citation practices vary considerably across disciplines and publication years. Waltman et al. ⁴ showed that unnormalized citation indicators are unsuitable for cross field comparisons and may distort impact rankings. Similarly, Ioannidis et al. ⁵ emphasized the importance of standardized citation indicators to ensure fair comparison across scientific domains. Without normalization techniques such as field-weighted citation impact or time-adjusted metrics, comparisons among countries, institutions, and authors may inadvertently favor older publications and citation-dense fields.

Future bibliometric studies in wearable neurorehabilitation research would benefit from more refined search frameworks that clearly distinguish clinically integrated wearable systems from engineering-oriented sensor innovation. In addition, incorporating theoretically grounded growth modelling and normalized citation metrics would enhance analytical robustness and strengthen comparative validity.

In conclusion, while Li et al. ¹ provide a valuable and well-structured overview of global research trends, further addressing issues related to conceptual aggregation, growth modelling interpretation, and citation normalization would further improve the methodological rigor of bibliometric mapping in this interdisciplinary domain.