Negative Pressure Wound Therapy Research: A Bibliometric and Visualized Analysis of the Web of Science Core Collection

Summary

This bibliometric and visualized analysis maps the development of negative pressure wound therapy (NPWT) research from 1991 to 2026. It summarizes publication patterns, collaboration networks, influential sources, and thematic transitions from mechanistic studies to wound care, wound dressings, infection prevention, and clinical outcome evaluation.

Abstract

Negative pressure wound therapy (NPWT) is widely used in surgical and wound care, but its long-term intellectual structure and collaboration patterns remain incompletely characterized. We screened 5,957 records retrieved from the Web of Science Core Collection and conducted a bibliometric and visualized analysis of 4,063 included articles and reviews. Records were screened independently by two researchers, and disagreements were resolved by consensus. VOSviewer, CiteSpace, and complementary statistical tools were used to evaluate publication trends, collaboration networks, keyword co-occurrence, citation bursts, and country-level patterns. Publication activity increased markedly after 2004. The United States and China were the most productive countries, whereas European and U.S. institutions occupied central positions in collaboration networks. Thematic emphasis evolved from subatmospheric pressure, perfusion, and mechanistic research toward wound care, wound dressings, surgical-site infection, diabetic foot ulcers, and efficacy. These findings define the development of NPWT research and identify priorities for multicenter clinical studies, mechanistic investigation, material innovation, and international collaboration.

Introduction

Negative pressure wound therapy (NPWT) is a widely used clinical modality that promotes wound healing by applying localized subatmospheric pressure to reduce tissue edema, remove exudate, and support infection control. It has been applied to diabetic foot ulcers^1,2, open abdominal wounds^3,4, traumatic soft-tissue defects, and orthopedic wounds^5,6. Landmark experimental and clinical reports published in 1997 established the biological basis and clinical feasibility of vacuum-assisted closure^7,8. Proposed mechanisms include microdeformation-induced cell proliferation⁹, angiogenesis¹⁰, and improved local perfusion. Over the past three decades, the clinical scope of NPWT has expanded toward individualized treatment and device-based innovation, including NPWT with instillation¹¹.

Despite its widespread use, uncertainty remains regarding the strength of the clinical evidence for NPWT¹². A 2014 Cochrane review found insufficient evidence of clear benefits for surgical wounds healing by primary intention and noted the potential influence of commercial funding¹³. A subsequent Cochrane update reported uncertainty for several clinically important outcomes, although NPWT may reduce surgical-site infection in selected settings¹⁴. Adequately powered, independently funded trials are therefore needed, particularly in high-risk surgical populations. Optimal pressure settings, dressing selection, site-specific protocols, and long-term prognostic factors also remain incompletely defined. Research activity is geographically uneven. Foundational experimental and clinical studies of NPWT were conducted by U.S. research groups, and subsequent global analyses indicate that U.S. and European institutions became prominent contributors to the field’s research and collaboration networks^7,8,15. Chinese research groups have contributed a growing body of clinical evidence involving diabetic foot ulcers¹, burns¹⁶, and orthopedic and open-fracture wounds¹⁷. In the present dataset, however, Chinese publications were more frequently produced within single-country research networks, and participation in multinational collaborations was less extensive than that observed among several established U.S. and European groups. These findings highlight opportunities to strengthen multinational collaboration and mechanistically oriented research on NPWT.

Bibliometric analysis provides a quantitative framework for mapping publication trends, collaboration networks, influential contributors, core journals, highly cited studies, and emerging topics. Although a bibliometric analysis of NPWT has previously been reported¹⁵, a comprehensive evaluation of publication trends, collaboration networks, influential sources, and thematic evolution remains limited. This study, therefore, mapped publication trends, collaboration networks, influential sources, and thematic evolution in NPWT research.

Protocol

The complete bibliometric workflow consists of literature retrieval, eligibility assessment, duplicate and version checking, independent relevance screening, data standardization, descriptive analysis, network analysis, and visualization. The raw records, screening decisions, and reproducible analysis package are provided in Supplemental File 1, Supplemental File 2, and Supplemental File 3, respectively.

Data source and search strategy

The bibliometric analysis was conducted using the Web of Science Core Collection (WoSCC), with the Science Citation Index Expanded (SCI-EXPANDED/SCIE) selected as the data source. The publication period was restricted to 1 January 1991 through 30 June 2026; therefore, the 2026 publication cohort represents a partial year because the search window ended on 30 June 2026.

The following search strategy was applied: TS=("negative pressure wound therapy" OR "vacuum assisted closure" OR "vacuum sealing drainage") AND DOP=(1991-01-01/2026-06-30). The Topic (TS) field was used to search titles, abstracts, author keywords, and Keywords Plus, whereas the Date of Publication (DOP) field was used to restrict the publication period.

No language or document-type restrictions were applied during record retrieval. Eligibility assessment was performed after data export to preserve a complete and auditable record-selection process. The final WoSCC search was conducted on 23 July 2026, yielding 5,957 records, which were exported on 24 July 2026. The complete search strategy, database and index information, date range, search and export dates, and eligibility workflow are summarized in  Figure 1 and further documented in the README Reproducible Workflow.md file provided in Supplemental File 3.

Record export and initial data verification

The retrieved records were exported from the Web of Science Core Collection with Full Record and Cited References selected as the record content and Plain Text File as the export format. Because the total dataset exceeded the maximum number of records permitted per export, the 5,957 records were downloaded in 12 consecutive batches comprising records 1–500, 501–1,000, 1,001–1,500, 1,501–2,000, 2,001–2,500, 2,501–3,000, 3,001–3,500, 3,501–4,000, 4,001–4,500, 4,501–5,000, 5,001–5,500, and 5,501–5,957.

Following export, the records were verified to ensure that the principal bibliographic fields required for subsequent analyses were present, including the accession number (UT), DOI, title, authors, affiliations, source journal, publication year (PY), publication date (PD), document type (DT), language, abstract, author keywords, Keywords Plus, total citations (TC), usage counts (U1 and U2), and cited references (CR). The original plain-text export files were preserved without modification and are provided as Supplemental File 1.

Because the WoSCC plain-text export does not retain the search-history page or a dedicated exported Date of Publication (DOP) field, the original search string and search date were documented in Figure 1, the manuscript, and the README Reproducible Workflow.md file included in Supplemental File 3. The original DOP restriction applied during database retrieval was retained and was not replaced by post-export filtering based on publication year (PY), publication date (PD), or Early Access (EA) date.

Document-type screening and retraction assessment

The 5,957 retrieved records were imported into the screening dataset for eligibility assessment. Original research articles and review articles were retained without applying any language restrictions. Eligible Web of Science document-type classifications included Article, Article; Proceedings Paper, Article; Early Access, Review, and Review; Early Access. Early Access was treated as an indexing status rather than as an independent publication type to avoid duplicate counting of the same publication.

During screening, 1,078 records with ineligible document types were excluded, followed by the removal of 59 retraction-related records. After document-type screening and retraction assessment, 4,820 Article- and Review-type records were retained for duplicate and version assessment. The excluded records and the corresponding record-specific reasons are provided in the DT_Exclusions_1078 and Retraction_Exclusions_59 worksheets of Supplemental File 2.

Duplicate and publication-version assessment

Duplicate and publication-version assessment was performed using the Web of Science accession number (UT) as the primary record identifier. When duplicate or related records had different accession numbers, they were compared using the DOI, normalized title, authors, source journal, publication year, volume, issue, page numbers, abstract, document type, and publication-version information. For each set of related records, the final or most appropriate publication version was retained, and the relationship between retained and excluded records was documented.

A total of 23 duplicate or superseded-version records were removed, including nine superseded Cochrane review versions and 14 verified duplicate, reprint, duplicate-indexing, language-version, or other related-version records. After duplicate and publication-version assessment, 4,797 unique records were retained for topical relevance screening. The accession numbers of retained and excluded records, duplicate-verification criteria, and record-specific rationales are provided in the Duplicate_Exclusions_23 worksheet of Supplemental File 2.

Independent topical relevance screening

Topical relevance screening was independently performed by two trained researchers on the 4,797 unique records. Each reviewer evaluated the title, abstract, author keywords, and Keywords Plus for every record and recorded an independent inclusion or exclusion decision together with a record-specific rationale before reconciliation.

Records were considered eligible when negative pressure wound therapy (NPWT), vacuum-assisted closure (VAC), or vacuum sealing drainage (VSD) constituted the principal intervention or comparator, the technique under evaluation, a primary clinical outcome, the subject of a mechanistic investigation, a guideline or consensus topic, or a principal component of the wound-management pathway. Records were excluded when NPWT, VAC, or VSD was mentioned only as background information, included only in a list of treatment options, reported solely as a previously failed treatment or non-use, used only as an incidental rescue measure, described only as a complication or covariate, fell outside the principal subject of the study, or appeared only as a false-positive Keywords Plus term.

The independent screening decisions were compared, with initial agreement achieved for 4,782 records and disagreement identified for 15 records. All discrepancies were resolved through discussion, and, when necessary, adjudication by a third researcher. The final decision and supporting rationale were documented for each disputed record. In the audited dataset, all 15 disputed records were ultimately excluded, and no disagreements remained unresolved.

Following topical relevance screening, 734 records were excluded and 4,063 records were retained for the bibliometric analysis. Supplemental File 2 contains the decisions of both reviewers, reviewer-specific rationales, disagreement indicators, final inclusion decisions, supporting evidence statements, and the final dataset of 4,063 records.

The final record-selection process was internally verified, confirming that the 5,957 retrieved records comprised 1,137 records excluded because of document type or retraction status, 23 duplicate or superseded records, 734 records excluded during topical relevance screening, and 4,063 records included in the final bibliometric analysis (5,957 = 1,137 + 23 + 734 + 4,063). Figure 1 presents a PRISMA-style flow diagram to illustrate the bibliographic record-selection process. Because this study is a bibliometric analysis rather than a systematic review, a full-text risk-of-bias assessment was not performed.

Final dataset construction and cited-reference integration

The final dataset of 4,063 records was finalized and frozen on 24 July 2026 and was used for all subsequent descriptive, citation, network, and visualization analyses. All included records were verified to contain unique Web of Science accession numbers (UTs).

To enable cited-reference analyses, the cited-reference (CR) fields were retrieved from the complete 5,957-record Full Record and Cited References export and integrated into the final dataset by exact matching of UT identifiers. This integration process was performed without altering dataset membership or the citation and usage-count snapshot associated with the original export.

Following integration, 4,048 included records were confirmed to contain at least one exported cited-reference (CR) field, whereas 15 records contained no cited-reference list in the Web of Science Core Collection export. The cited-reference integration script and the corresponding audit report are provided in Supplemental File 3.

Data cleaning and descriptive analysis

Author names, institutional names, journal titles, cited references, countries/regions, and keywords were standardized using unit-specific thesauri. Separate thesauri were applied to each analytical unit, and author, organization, country, keyword, journal, and cited-reference thesauri were not combined during the standardization process.

Annual publication output was grouped according to the publication year (PY) recorded in the Web of Science Core Collection. Years with no publications between 1991 and 2026 were retained in the analysis and displayed as zero to ensure a continuous time series. Document types were classified into five mutually exclusive categories, with Early Access treated as an indexing status rather than an independent publication type to avoid duplicate counting.

Total citations (TC), usage counts during the preceding 180 days (U1), and usage counts since 2013 (U2) were aggregated by publication year. Because TC, U1, and U2 are dynamic indicators in the Web of Science Core Collection, these metrics represent the database snapshot exported on 24 July 2026.

Data processing, screening-log management, descriptive statistical analyses, and generation of Figure 2A,B and Figure 3A–C were performed using Python 3.12.10 with the pandas 3.0.3, openpyxl 3.1.5, and matplotlib 3.11.0 libraries, together with Microsoft Excel.

VOSviewer network analyses

Bibliometric network analyses were performed using VOSviewer version 1.6.21. The final Web of Science Core Collection plain-text dataset was imported as bibliographic database files. Full counting was applied for all analyses, and association-strength normalization was used to construct and visualize the bibliometric networks.

Unit-specific thesauri were applied according to the analytical unit under investigation. Keyword co-occurrence analysis was performed using All keywords with a minimum occurrence threshold of 30, resulting in a network comprising 158 nodes and 6,468 edges (Figure 4A). Author collaboration analysis was conducted using the Authors unit with a minimum threshold of six documents per author, producing a network of 51 nodes and 134 edges (Figure 5A). Institutional collaboration analysis was performed using the Organizations unit with a minimum threshold of 10 documents per organization, yielding 109 nodes and 351 edges (Figure 5B). Country collaboration analysis used the Countries unit with a minimum threshold of seven documents per country, generating a supplemental network containing 50 nodes and 438 edges. Source-journal bibliographic coupling analysis was conducted using the Sources unit with a minimum threshold of 20 documents per source, resulting in 39 nodes and 741 edges (Figure 6A). Cited-reference co-citation analysis was performed using the Cited references unit with a minimum threshold of 68 citations per reference, producing a network of 91 nodes and 3,626 edges (Figure 6C).

The VOSviewer map files and corresponding network files generated for each analysis were retained to ensure reproducibility. The complete mapping parameters, threshold values, thesauri, mapping audits, and saved VOSviewer map and network files are provided in Supplemental File 3.

CiteSpace keyword burst and journal dual-map analyses

Keyword burst detection and journal dual-map analyses were performed using CiteSpace version 7.0.R0 (64-bit, Advanced) with the final frozen Web of Science Core Collection dataset. Keyword burst analysis was conducted using a time span of 1991–2026 with 1-year time slices. Terms were extracted from the title, abstract, author keywords, and Keywords Plus fields. Keyword was selected as the node type, the selection criterion was set to Top N = 50 per time slice, and no pruning was applied.

Citation burst analysis was performed using the Burstness function, and the 25 keywords with the strongest citation bursts were retained for presentation in Figure 4B. Because the literature search covered publications only through 30 June 2026, citation bursts terminating in 2026 were interpreted as extending to the endpoint of the partial search window rather than representing activity across a complete calendar year.

A journal dual-map overlay was generated from the same frozen Web of Science dataset to visualize citation pathways between citing and cited journal disciplines (Figure 7). The complete CiteSpace parameter settings and the saved Top-25 keyword burst results are provided in Supplemental File 3.

Institutional, journal, and country-level visualizations

Institutional, journal, and country-level bibliometric visualizations were generated using bibliometrix and complementary visualization software. Institutional names were harmonized using the standardized affiliation data described above. The collaboration network was then constructed based on co-authorship relationships among institutions meeting the predefined publication threshold. Nodes represent institutions, links denote collaborative relationships, and thicker links indicate higher collaboration intensity.

To examine citation relationships among the most productive journals, the 15 journals with the highest publication output were identified. Directed citation links were constructed from citing journals to cited journals after excluding journal self-citations. The resulting network comprised 15 nodes and 200 directed non-self citation edges. For visualization, only the 56 citation links with weights of at least 45 were displayed in  Figure 6B. This network represents direct citation relationships among the 15 most productive journals and does not constitute a Bradford core analysis or a bibliographic-coupling map.

Country and region analyses were performed using full counting across all author affiliations, with each country contributing no more than once to an individual publication. Publications were classified as single-country publications (SCP) or multiple-country publications (MCP) according to the corresponding author's country and whether the complete set of author affiliations represented one or multiple countries. For country-level analyses only, Taiwan, Hong Kong, and Macao were aggregated under China, while the original institutional names and address information were preserved.

Country/region analyses and SCP/MCP visualizations presented in Figure 8A,B were generated using R version 4.6.1 with the bibliometrix (v5.4.1) and ggplot2 (v4.0.3) packages, together with SCImago Graphica version 1.0.55.

Reproducibility and data verification

Reproducibility and technical validation were supported through three complementary supplemental files. Supplemental File 1 contains the unmodified raw Web of Science Core Collection (WoSCC) export used as the primary data source. Supplemental File 2 documents all document-type, retraction, duplicate/publication-version, and topical-relevance screening decisions, enabling complete reproduction of the study selection process.

Supplemental File 3 provides the materials required to reproduce the data-cleaning procedures, cited-reference integration, descriptive analyses, VOSviewer network analyses, CiteSpace keyword burst analyses, institutional collaboration network, journal direct-citation network, and technical validation procedures. Reproduced outputs were compared with the archived node counts, edge counts, screening totals, cited-reference coverage, and figure-source audit files to verify analytical consistency.

All analyses were confirmed to use the same frozen dataset comprising 4,063 records and the same citation and usage-count snapshot obtained on 24 July 2026. Consolidated software versions and key analytical parameters are provided in the Table of Materials.

Results

Basic publication output and annual trend analysis

A total of 4,063 NPWT-related publications were included: 3,563 Articles (87.7%) and 500 Reviews (12.3%). These totals classify Early Access as an indexing status rather than a separate document type. Annual publication output is shown in Figure 2A, and the plotting categories used to display document types and Early Access status are shown in Figure 2B.

Discussion

The analysis identified a sustained expansion of NPWT research and a clear thematic transition from foundational mechanisms to clinical optimization¹⁸. Early studies emphasized sub-atmospheric pressure, tissue perfusion, and biological responses. For example, Morykwas et al. (1999) found that applying negative pressure to partial-thickness burns in pig models prevented wound deepening, revealing a 12-hour treatment window post-injury, with as little as 6 hours of application sufficient to stop inj...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
`CiteSpace 7.0.R0 (64-bit) Advanced	Chaomei Chen, Drexel University, USA	https://citespace.podia.com/	Used for keyword burst analysis and the dual-map overlay. Settings: 1991-2026, 1-year slices, Keyword node type, Top N = 50 per slice, no pruning, and the 25 strongest burst terms.
Python 3.12.10; pandas 3.0.3; openpyxl 3.1.5; matplotlib 3.11.0; NetworkX 3.6.1; Microsoft Excel 16.0.20131.20154	Python Software Foundation; pandas, openpyxl, Matplotlib, and NetworkX development teams; Microsoft Corporation	https://www.python.org/; https://pandas.pydata.org/; https://openpyxl.readthedocs.io/; https://matplotlib.org/; https://networkx.org/; https://www.microsoft.com/microsoft-365/excel	Used for screening, deduplication, data auditing, descriptive analyses, and figure generation. The final dataset contained 4,063 records. NetworkX was used for the Top-15 journal direct-citation network (15 nodes; 56 displayed edges with weights >=45).
R 4.6.1; bibliometrix 5.4.1; ggplot2 4.0.3	R Foundation for Statistical Computing; Massimo Aria and Corrado Cuccurullo; ggplot2 development team	https://www.r-project.org/; https://www.bibliometrix.org/; https://ggplot2.tidyverse.org/	Used for country/region productivity and collaboration analysis. Full counting was applied to all addresses; SCP/MCP used the corresponding author's country. Taiwan, Hong Kong, and Macao were aggregated under China at the country level only.
SCImago Graphica 1.0.55	SCImago Lab	https://graphica.app/	Used for final layout and visualization of the journal direct-citation network and country/region maps.
VOSviewer 1.6.21	Centre for Science and Technology Studies (CWTS), Leiden University, the Netherlands	https://www.vosviewer.com/	Full counting with association-strength normalization. Thresholds and final networks: keywords >=30 (158 nodes; 6,468 edges); authors >=6 documents (51; 134); institutions >=10 documents (109; 351); source journals >=20 documents (39; 741); cited references >=68 citations (91; 3,626). Unit-specific thesauri were applied.