2.1 Search strategy and selection criteria

We queried records related to PD by specifying database name terms and common variants in the Topic field. This functioned similarly to MeSH terms within PubMed, grouping categories relative to specific terms for easier filtering. The search yielded 981 results in total, which contained all articles published since database inception. Filtering the search results via “journal articles” resulted in 943 articles. Articles were not excluded by topic within the specialties of medicine or surgery. Original articles and registry data were included, whereas meta-analyses, systematic reviews, guidelines, and other review articles were excluded. These results were then ranked within Web of Science by total citation count and exported to Covidence systematic review software (Veritas Health Innovation, Melbourne, Australia) for screening.

Two reviewers (RC and DDO) independently screened all titles and abstracts for relevance and then reviewed abstracts and full texts according to the inclusion and exclusion criteria. Where any discrepancies were found, these were resolved through consensus discussion. The top 100 most cited articles which fulfilled our criteria were then included in the final dataset.

Inclusion: (1) original research articles (clinical studies, basic science investigations, registry data reports, prospective and retrospective case series, randomized controlled trials, cohort studies, case-control studies, and biomechanical studies); (2) published in English; (3) utilized PD.

Exclusion: (1) reviews (systematic reviews, meta-analyses, narrative reviews, scoping reviews), guidelines, consensus statements, position papers, letters to the editor, commentaries, editorials, conference abstracts or proceedings, book chapters, technical notes without original data, or case reports with fewer than 5 patients; (2) articles which did not directly utilize PD and articles that were retrieved due to a reference to PD somewhere in the text of the article.

2.2 Data extraction

The following information was listed for all articles: title, first author's name, journal name, year of publication, impact factor of the journal in 2024, total number of citations of the article, geographic origin, institutions, research theme, level of evidence, and keywords.

The altmetric bookmarklet (https://www.altmetric.com/solutions/free-tools/bookmarklet/) was used to identify the AAS and number of X (formerly Twitter) posts. For each article, the AAS was derived using its Digital Object Identifier (DOI), based on a weighted algorithm that evaluates the level of attention from various non-traditional sources.

2.3 Statistical analysis

Descriptive statistics were calculated for all article-level metrics. Articles were ranked by total citations in the Web of Science Core Collection and by citations per year (CPY). Each article was assigned to an orthopaedic surgery subspecialty based on the domain reported in the manuscript; if not specified, subspecialty was inferred from the senior author's departmental section. Comparison between means of the article metrics were made using one-way analysis of variance (ANOVA). Correlation between variables was determined using Spearman rank or Pearson product-moment tests, and P < 0.05 was considered statistically significant. Analysis was performed using SPSS 30.0 (IBM Corp).

For keyword analysis, we identified all keywords associated with each article. Articles without keywords were excluded from this portion of the analysis. Network analysis was performed using Gephi v0.10.1 (Gephi Consortium, Paris, France) to identify the relationships between keywords and their centrality in the literature. Visual representations of keyword networks were created to identify central themes and evolving research focuses within the field.

Citation density, defined as the CPY since publication, was calculated for each article to adjust for the advantage of older publications in accumulating more citations. The publication months were used to account for fractions of a year. Citation density was plotted against publication year to identify temporal trends in citation patterns.

4.1 Utilization and focus

Since their inception, PD and other insurance claims databases, have become widely used to supplement orthopaedic research,^1–3 whilst focusing on epidemiology,^11–14 high-volume surgical procedures,^15,16 medical treatments,¹⁷ health policy topics,^18,19 and cost analyses.²⁰ Some examples include Montgomery et al. who reported a 365 % increase in hip arthroscopy incidence between 2004 and 2009,¹¹ while Abrams et al. documented sharp increases in meniscus repair, both isolated and combined with ACL reconstruction, between 2005 and 2011.¹³ Sing et al. later demonstrated that older age groups undergoing hip arthroscopy faced higher conversion rates to arthroplasty.¹⁴

The predominance of sports, spine, and arthroplasty studies amongst the most-cited articles reflects the widespread use of claims data in procedures with high national incidence and resource implications, such as ACL reconstruction, hip arthroscopy, and total joint replacement. PD has provided large-scale, cost-efficient insights into population-level trends. Importantly, our keyword network analysis revealed clustering around utilization and complication themes, suggesting that PD is most impactful when applied to questions of national burden and outcomes variation rather than highly granular clinical variables.

4.2 Methodological challenges and inconsistencies

PD contains millions of patient records across multiple payer datasets, enabling correlations between demographic variables and clinical outcomes at relatively low cost compared to prospective multicenter trials.²¹ Despite these contributions, PD studies share inherent limitations. Exposure and outcome definitions within PD are dependent on CPT and ICD coding, the validity of which varies across diagnoses, procedures, and coding eras. The ICD codes are limited in their ability to stratify diagnoses. The ICD10 code changes in 2018 substantially improved this, however the PD Mariner dataset contains records from 2010 which means a large portion of the data is not optimally categorized. Furthermore, even the ICD-10 codes are limited in their ability to categorize injury and fracture subtypes.²² For example, McCormick et al. demonstrated increasing use of cartilage restoration techniques,²³ yet without radiographic or intraoperative data, the clinical nuance is limited. Azad et al. showed evolving fixation preferences for distal radius fractures, but fracture type, injury severity, quality of reduction, and surgeon rationale cannot be ascertained. These examples highlight how coding-driven studies risk over-simplification. The lack of granular data within PD, limits clinical integration and has resulted in the production of more descriptive reporting styles.

The difficulty of aggregating sufficiently granular data is shared by all databases and not limited to insurance claims databases.²⁴ In trauma care, databases poorly stratify fracture subtypes, verify the quality of fracture reduction and fixation, and are limited by adequate follow-up for long term complications. These limitations may have contributed to the restructuring of the AAOS Fracture and Trauma Registry that has shifted the highest value datasets, hip fractures and proximal humerus fractures into other AAOS registry datasets and archived the remaining fracture datasets.²⁵

PD continues to evolve, from its initial database to its alliance with Humana in 2014,²⁶ to its most recent dataset within the Mariner version,²¹ which introduces uncertainty around data maintenance, population capture and reporting of quantitative outcomes. Earlier studies employed Humana or Medicare subsets,¹¹ while newer studies employed the Mariner database,^12,18 potentially complicating longitudinal analyses and raising questions about representation by race or other characteristics in the dataset. Moreover, with samples in the millions, very small differences can achieve statistical significance, risking overstatement of clinical relevance.

4.3 Citation and altmetric performance

Our results show that highly cited PD articles frequently appeared in higher impact subspecialty journals such as Arthroscopy, American Journal of Sports Medicine, and Journal of Arthroplasty. Citation density was highest among recent sports medicine and hip arthroscopy related studies,¹² reflecting the rapid expansion of these domains. In contrast, arthroplasty related studies such as Debbi et al. demonstrated greater altimetric scores, potentially due to the relatability of these procedures for the general public as well as the association with health policy and apparent cost impact.²⁷

The moderate correlation between traditional and alternative metrics suggests that online visibility (e.g., AAS, XP) does not uniformly translate into long-term citation accrual. Nevertheless, the elevated AAS of recent PD studies reflects growing interest from clinicians, policymakers, and the public in “big data” orthopaedics.

4.4 Clinical impact

PD has facilitated highly cited, widely visible research that maps national utilization and complications in orthopaedics. Montgomery quantified the explosive rise of hip arthroscopy,¹¹ Abrams mapped national meniscus trends,¹³ and Buckland identified a critical high-risk cohort following THA.¹⁰ These papers and others provide critical data for policymakers and guideline committees and represent cost-efficient, scalable research but without some of the deeper level information that can inform clinical decision making. For these reasons, PD appears to have primary utility as a hypothesis generating tool, identifying system-level trends and high-risk cohorts that can then be validated in registries or prospective studies. Integration with patient reported outcomes, further imaging, and intraoperative data remains essential to achieve full clinical utility. Without such an evolution of the database, PD remains most valuable at the population level, guiding policy and economics rather than individual patient decision-making.

Future directions may include comparisons of PD with other orthopaedic databases to clarify relative strengths. Furthermore, validation studies assessing coding accuracy across PD iterations and linkage with patient-reported outcomes could help with both population and patient-level decision making.