In 2005, the field produced just 37 publications per year; by 2025, that figure had climbed to 321, with an exponential curve fitting the annual counts with a coefficient of determination of 0.9425. Growth was not perfectly smooth — output dipped from 248 papers in 2022 to 236 in 2023 — but the overall trajectory is unmistakable. Publications from 2025 alone accounted for more than 10 percent of the entire two-decade corpus, underscoring how rapidly the literature continues to expand.

To build the corpus, the researchers searched the Web of Science Core Collection and Scopus on July 26, 2026, retrieving 3,786 records. After removing 551 duplicates and excluding 125 records with missing keywords or abstracts, they were left with 3,110 unique publications. Two investigators independently screened titles, abstracts, and keywords, with disagreements resolved by a third senior investigator.

The team then standardized author names, affiliations, and keywords — consolidating singular and plural forms, expanding abbreviations, and merging synonyms — before running their analyses in R using the bibliometrix package. The geographic picture is heavily skewed. The United States led in both publication output (900 papers, 28.9 percent of the corpus) and total citations (40,710).

China ranked second in output with 355 papers but only fifth in citations with 4,800. Australia, the United Kingdom, and Germany showed the highest rates of international collaboration, with multiple-country publication percentages of 38.9, 36.9, and 28.0 respectively. At the institutional level, the University of Melbourne was the most productive with 140 publications, while Boston University accumulated the most citations at 8,554.

Melbourne's dominance extends to individual researchers: Bennell, Hinman, and Wrigley — all Melbourne-affiliated — topped the author productivity rankings with 69, 62, and 50 publications respectively. Author productivity followed a long-tail distribution. Roughly three-quarters of authors contributed only a single publication to the corpus, while a small group published repeatedly.

The fitted Lotka-type exponent was 2.38, and the Kolmogorov–Smirnov test did not reject this fitted distribution, whereas the classical fixed exponent of 2 was rejected. The team also identified nine core journals under Bradford's law, with Osteoarthritis and Cartilage standing out as both the most productive (207 papers) and most cited (15,525 total citations) journal in the field, followed by Gait & Posture and the Journal of Orthopaedic Research. Citation-burst analysis revealed how the field's intellectual center of gravity has shifted over two decades.

From 2006 to 2013, burst references established the biomechanical framework: Zhao and colleagues used direct in vivo measurements to validate the link between the knee adduction moment and medial-compartment contact force, while Astephen and colleagues showed distinct lower-extremity kinetics across osteoarthritis severity levels. Attention then turned to whether abnormal loading predicts structural progression — Bennell's work on dynamic medial knee loading and cartilage loss showed a citation burst of strength 27.20 from 2012 to 2016, and Chehab's five-year study linking adduction and flexion moments to cartilage change burst at 28.88. After 2017, bursts increasingly clustered around early-onset and post-traumatic osteoarthritis following anterior cruciate ligament injury, with one reference sustaining a burst through the end of the analysis period in 2025.

The methodological centerpiece of the study is its use of non-negative matrix factorization (NMF), a machine-learning technique that identifies latent thematic components in large text collections. The researchers concatenated titles, abstracts, and standardized keywords, transformed the text using term frequency–inverse document frequency weighting, and tested candidate models containing between 8 and 20 topics. A 19-topic model emerged as the best compromise between coherence and reproducibility, achieving a mean normalized pointwise mutual information of 0.4225.

Reproducibility was tested rigorously: 30 full-corpus refits with random initialization and 50 year-stratified refits using 80 percent subsamples, with topic–term vectors aligned via cosine similarity and Hungarian matching. Fourteen of the 19 topics met both reproducibility thresholds; five were flagged as exploratory and interpreted cautiously. What did the 19 topics reveal?

Gait and lower-limb kinematics claimed the largest overall share of the literature, followed by exercise rehabilitation, pain, and physical function, and then chondrocyte mechanotransduction and cartilage matrix biology. But the temporal analysis told a more dynamic story. Comparing mean annual topic shares between 2016–2020 and 2021–2025, the largest increases involved stem cells and regenerative cartilage repair, chondrocyte mechanotransduction, and meniscal tears, extrusion, and root repair.

High tibial osteotomy and lower-limb realignment, along with adiposity, infrapatellar fat pad, and inflammation, also gained ground. Meanwhile, the relative shares of gait kinematics, footwear orthoses and lateral-wedge insoles, and body mass and obesity research declined — though the authors stress that because annual topic shares sum to 100 percent, these shifts represent redistribution within an expanding corpus rather than absolute declines in publication counts. The authors were careful to distinguish recent rebounds from longer-term trends.

Chondrocyte mechanotransduction, for instance, showed a recent increase in share despite a negative slope across the full 2005–2025 period, whereas stem cell research showed both a positive long-term slope and a recent increase. A sensitivity analysis using normalized soft topic assignments correlated strongly with the primary hard-assignment results (Spearman's ρ = 0.926), with concordant directions for 17 of 19 topics. A separate refit restricted to Web of Science records confirmed high alignment for 18 of 19 topics, though the team acknowledged that database heterogeneity and English-language restrictions may have shaped the observed patterns.

The study has clear limitations. It excluded conference proceedings, which may have underrepresented computational and engineering developments, and non-English publications, which may have omitted region-specific research priorities. Citation counts, the authors note, should not be read as direct measures of methodological quality, since recent publications have had less time to accumulate citations.

The authors suggest that future work should prioritize longitudinal, externally validated multiscale models that combine gait measurement, imaging, and computational modeling — and, crucially, direct tests of whether modifying the mechanical environment can deliver durable clinical and structural benefits for patients.

Subject of Research: Bibliometric and topic-modeling analysis of knee osteoarthritis biomechanics research from 2005 to 2025

Article Title: Evolution of knee osteoarthritis biomechanics research, 2005–2025: bibliometric mapping and machine-learning-assisted topic modeling

Article References: Lu, F., Liu, L., Yang, P., Gao, H., Song, H., Wang, G., & Wang, Z. (2026). Evolution of knee osteoarthritis biomechanics research, 2005–2025: bibliometric mapping and machine-learning-assisted topic modeling. Frontiers in Medicine, 13, Article 1968099. https://doi.org/10.3389/fmed.2026.1968099

Image Credits: AI Generated

DOI: 10.3389/fmed.2026.1968099

Keywords: knee osteoarthritis, biomechanics, bibliometrics, topic modeling, non-negative matrix factorization, gait analysis, mechanobiology, cartilage repair, meniscus, high tibial osteotomy, citation analysis, research trends