Source-linked AI summary
Quantifying the impact of clinical-academic collaborations
Mohamad Zeina, Nick McNally, Karl S. Peggs, Parashkev Nachev
TL;DR
Clinical-academic collaboration had not been robustly quantified, motivating a framework for analysing academic networks with open data. The paper applies that framework to NIHR BRC partnerships and reports that UCLH/UCL’s most underserved partners gain the most. The framework supports comparison between networks and identification of attributes associated with bibliometric impact, while its adjustment is not a causal estimator.
Problem
The impact of clinical-academic collaborations had not been robustly quantified, creating a need for quantitative representation.
Method
The paper introduces an open-data framework for quantitative analysis of academic networks and applies it to NIHR BRC clinical-academic partnerships.
Results
UCLH/UCL’s most underserved partners gain the most.
Takeaways & Limitations
The framework enables faithful comparison between networks and identification of attributes that predict bibliometric impact.
Takeaways & Limitations
The adjustment is a flexible covariate correction, not a fixed-effects or doubly robust causal estimator.
Abstract
from arXiv · showhide
Academic collaboration is of self-evident value but requires a quantitative representation to be optimally guided by policy. No established methodological approach to such representation exists. Here we introduce a general framework of graphical and bibliometric analysis of open data for the task of quantifying the impact of academic networks, with NIHR Biomedical Research Centre (BRC) clinical-academic partnerships in England as the prototype. We define publication-level identities for the 20 English BRCs based on the conjunction of authors from each BRC's partner institutions. Drawing on bibliometric and administrative records, we characterise the graphical properties of each network, estimate what the university adds to the hospital's papers, what the partnership adds to the papers of relatively infrastructure-poor collaborating institutions, and how that gain depends on existing infrastructure. We apply our framework to the NIHR UCLH/UCL BRC as an exemplar. UCLH/UCL authored 20,985 network papers from April 2007, in collaboration with 9,868 distinct external partners over the whole record, forming the most central node of the graph of networks across England. University co-authored papers exhibited 1.6 times the field-weighted citation impact (FWCI) of hospital-only papers, and were 2.1 times as likely to be cited by a patent. Across 60 of the exemplar's most partnered with UK healthcare organisations, the benefit rose from 1.8 times where local NIHR infrastructure activity was densest to 3.4 times where it was sparsest, while impact without the exemplar varied little. Academic networks can be robustly identified from open data, enabling comparative analysis of collaborative impact. Applied to NIHR BRCs, the approach enables quantification of the impact across networks and reveals that benefit is most pronounced where infrastructure is least developed.
Introduction
The paper addresses the lack of robust quantitative measures for clinical-academic collaboration by defining networks from publication metadata and comparing their outputs.
- Clinical-academic collaboration has not been robustly quantified, despite evidence that research-active hospitals and collaborative papers achieve stronger outcomes.
- The authors identify two obstacles: network relationships are poorly represented in bibliographic metadata, and faithful cross-network comparators are publicly limited.
- The framework assigns a paper to a network when its authors include both the partner hospital trust and university, creating a common public criterion.
- This identity enables identical analysis across the 20 English BRCs and comparisons between network attributes and published outputs.
- The study examines university value for hospitals, benefits to infrastructure-poor healthcare collaborators, and whether benefits vary with existing infrastructure.
Methods
The study combines open bibliographic, administrative, and infrastructure records with explicit rules for identifying BRC networks and defining analysis windows.
- OpenAlex supplies publication, author, institution, citation, and field data, while Altmetric and NIHR datasets provide external-impact and infrastructure information.
- Analysis windows: The study analyzes papers from 1 April 2007 onward for output and citation accrual, uses the whole record for network reach, and uses 2015–2024 for partner gains.
- Establishing BRC bibliographic identity: Affiliation strings were manually structured to distinguish university from hospital authors because OpenAlex may conflate UCLH and UCL.
- Establishing BRC bibliographic identity: Network papers require at least one author affiliated with the host NHS trust and at least one affiliated with its corresponding university.
- Cohort definition: The analysis uses the current 2022–2027 BRC cohort, although Guy’s and St Thomas’ exited the cohort and Exeter joined only in December 2022.
Network structure
The authors represent BRC networks as nodes in an England-wide collaboration graph and quantify their external institutional reach.
- For each network, the analysis counts external institutions and measures the share of its external partners also reached by the UCLH/UCL exemplar.
- The England-wide graph contains 22 network nodes, comprising 20 BRCs plus King’s and Edinburgh, with edges weighted by shared papers.
- Eigenvector centrality is calculated on the weighted graph, whereas betweenness centrality is calculated on the unweighted graph.
The value of clinical-academic partnership
The analysis compares hospital-only and university-involved papers, then evaluates benefits to healthcare partners across differing infrastructure environments.
- Hospital-only and university-involved papers are compared using citation, FWCI, patent, policy, and news outcomes with field, year, and author-count adjustment.
- Binary outcomes use linear probability models, and results are supplemented with unadjusted ratios of means and incidence.
- Collaborative benefit to underserved organisations: Partner organizations are classified by location and healthcare status, with established partnerships requiring at least five joint papers and a founding paper.
- Infrastructure classification: Early-phase translational infrastructure includes BRCs, CRFs, HealthTech Research Centres, ECMCs, and BioResource centres, while applied and delivery schemes are excluded.
- Partnership persistence: The study tracks recent partnership activity, distance, and persistence among collaborations whose first joint paper appeared from 2007 to 2015.
Collaborative benefit to underserved organisations
The analysis compares papers produced with and without exemplar or other clinical-academic networks across healthcare organisations, using outcome ratios and covariate-adjusted estimates. It also tests whether collaboration benefits differ across organisations and infrastructure contexts.
- Comparison design: The study compares healthcare organisations’ papers written with the exemplar, with another university-hospital pair, or with neither.The primary sample covers organisations linked to English university-hospital pairs and assigns papers to three collaboration groups.
- Pooled comparison: The pooled exemplar-to-without-exemplar FWCI ratio is computed on work-deduplicated papers using organisation-clustered bootstrap resampling.The primary comparison includes 40 organisations and 86,983 papers.
- Robustness analysis: Top-decile FWCI rates and rate ratios distinguish the exemplar’s association from broader effects of elite collaboration.Comparisons include exemplar papers against other-pair and neither arms, with organisation-clustered bootstrap intervals.
- Outcome measures: Impact is assessed using FWCI, raw citations, patent citation, and policy citation across a sample of organisations selected for frequent exemplar collaboration.The four-endpoint analysis uses an earlier sample of 14 organisations selected by at least 50 co-authorships with the exemplar.
- Adjustment: The analysis reports unadjusted with-versus-without ratios alongside estimates adjusted for publication year, field, author count, and organisation.Adjustment uses cross-fitted gradient-boosted regression and organisation-clustered standard errors.
- Interpretation: The adjustment is a flexible covariate correction rather than a fixed-effects or doubly robust causal estimator.A comparison group of 24 non-BRC NHS trusts in served cities is also analysed.
Dependence of benefit on local infrastructure
Across 60 healthcare organisations collaborating substantially with the exemplar, its FWCI benefit increased as local NIHR-supported research infrastructure became scarcer. Impact without the exemplar varied little across infrastructure tertiles, while the gradient was also observed for regional economic activity.
- Analysis scope: The analysis covered 60 exemplar partners spanning BRC host trusts, served-city trusts and organisations in unserved places.Four axes were computed: nearby NIHR-supported activity, distance to the nearest BRC, regional gross value added per head, and catchment-weighted deprivation.
Statistical reporting and reproducibility
The analyses were applied consistently across all 20 networks, with uncertainty estimates and reproducibility materials reported alongside the results.
- Reporting: All analyses were applied identically to the 20 networks, with descriptive metrics and within-hospital contrasts reported in supplementary tables.Supplementary Tables S2 and S3 provide the network-wide descriptive metrics and within-hospital comparisons.
- Uncertainty: 95% confidence intervals used organisation-clustered bootstrap methods for ratios and HC3 or cluster-robust standard errors for regression coefficients.The reporting specifies separate uncertainty procedures for ratios and regression estimates.
- Reproducibility: Network definitions, derived tables and production scripts are publicly available, with manifest entries naming each table’s script, window and checksum.Analyses were performed in Python using pandas, statsmodels, scikit-learn and NetworkX.
Results
Across English BRC networks, UCLH/UCL was the most central network and showed higher impact for university-coauthored papers than hospital-only papers. Its partnerships reached many organisations beyond established NIHR infrastructure, and the collaboration benefit was greatest where local infrastructure was scarcest.
- Network-wide results: UCLH/UCL had the highest weighted eigenvector centrality among 20 BRC networks, at 0.599, ahead of GOSH at 0.569 and Oxford at 0.237.The centrality measure weights a node by the importance of its neighbours; output and citation totals also varied substantially across networks.
- University contribution: University-coauthored UCLH/UCL papers had 1.57 times the FWCI of hospital-only papers, 3.54 versus 2.25.After adjustment for publication year, field and team size, they received 68% more citations and had higher patent and news inclusion, but lower policy inclusion.
- University contribution: The adjusted citation premium across all 20 networks ranged from 20% at Royal Marsden to 131% at Maudsley, with a median of 58%.Adjusted FWCI benefit was significant in 12 networks; patent and news effects were significant in 18 and 20, respectively.
- Reach beyond infrastructure: UCLH/UCL collaborated with 688 UK institutions in sites without NIHR early-phase infrastructure, including 418 NHS or healthcare organisations.These partners spanned 237 towns and cities, and the exemplar led or tied for first in unserved-place reach every year since 2020.
- Reach beyond infrastructure: Among exemplar partnerships established by five or more joint papers, 92% of those beginning in 2007–2015 remained active in 2021–2024.The exemplar had 289 established unserved-site partnerships, 286 active in 2021–2024, and 155 more than 200 km from infrastructure.
- Benefit to partner organisations: Across 14 unserved-place organisations, exemplar coauthorship produced 4.04 times the mean FWCI and 2.60 times the policy-citation incidence.The study also found 3.19 times the mean citation count and 2.59 times patent-citation incidence before adjustment.
- Benefit to partner organisations: The FWCI gain increased from 1.83 where local NIHR activity was greatest to 3.42 where it was lowest, while impact without the exemplar varied little.The same direction appeared across regional economic tertiles, but not for catchment deprivation.
Discussion
The framework uses transparent open-data publication identities and graph-based bibliometrics to compare clinical-academic networks and quantify collaboration benefits. Applied to NIHR BRCs, it finds benefits across networks and especially strong gains for infrastructure-poor partners, while causal interpretation remains limited.
- Cross-network findings: Across NIHR BRC networks, collaboration benefits over hospital-only research vary substantially, and the benefit does not appear to be merely a function of network size.The framework quantifies the benefit of involving academic partners across disparate contexts, while observed outputs, citations, and benefits differ across networks.
- Partner infrastructure: Among underserved partners, collaboration benefits are substantial across endpoints and increase as partner NIHR support decreases.The exemplar’s most underserved partners gain the most, contrary to an expectation that collaborative increments would remain constant or decline with lower resources.
- Implications: The findings suggest that collaborative contribution interacts with partner need, with effective network operation potentially as important as network scale and span.The authors suggest strengthening effective hubs or replicating their collaborative policies may benefit underserved nodes, while presenting these as policy possibilities rather than established causal effects.
- Limitations: The study cannot infer causal effects because co-authorship is non-random, although within-organisation comparisons and symmetric sampling address some selection concerns.More sophisticated author fixed-effects and event-study designs are proposed as possible mitigations.
- Limitations: Interpretation is also bounded by binary city-level infrastructure classification, a changing BRC cohort, imperfect affiliation coverage, and differing analysis windows.The gradient estimates concern 60 UK healthcare co-author organisations, most near early-phase infrastructure of their own.
- Framework: The framework identifies networks from publication co-authorship and institutional identities, enabling transparent, equitable comparison across networks using open data.The introductory implementation uses co-authorship involving at least one representative of constituent institutions, while more complex weighting schemes remain possible.
Declarations
The study reports funding, data availability, authorship, competing-interest, and ethics declarations. Person-level data were excluded, while bibliographic and infrastructure data were openly available or appropriately licensed.
- The work was supported by the NIHR UCLH Biomedical Research Centre, and all authors were employed by UCL and/or UCLH and affiliated with the NIHR UCLH BRC.
- The authors state that the views expressed are not necessarily those of the NIHR or the Department of Health and Social Care, and report no other competing interests.
- OpenAlex and NIHR Infrastructure Supported Projects data were openly available, while Altmetric data were used under licence.
- Network definitions, derived tables, production manifests, checksums, and analysis scripts are available in the project repository, but person-level data are not included.
- The study used publicly available institutional bibliographic metadata, involved no human participants, and therefore required no ethical approval.
Supplementary material
The supplementary material defines the 20 English NIHR BRC networks and documents descriptive and within-hospital comparison tables. These materials specify network scope, metric definitions, data windows, and adjustment procedures.
- Supplementary Table S1: Supplementary Table S1 lists the 20 English NIHR BRC networks for the 2022–2027 funding round and identifies the OpenAlex site families counted for each trust.Names follow OpenAlex records, with noted trust mergers and renamings.
- Supplementary Table S2: Supplementary Table S2 reports papers, citations, partners, centrality, served places, partnership activity, retention, and overlap using specified record windows and thresholds.The table is sorted by all-years paper counts.
- Supplementary Table S3: Supplementary Table S3 compares with-university and hospital-only papers from April 2007 to 2025 using text-based arm definitions.Adjusted estimates use OLS with publication-year and primary-field fixed effects, log(1 + author count), and HC3 standard errors; incidence ratios are unadjusted.