XSci (AI review)

September 28, 2026

Referee Report

Manuscript: "A Scientometrics Analysis of Eye-Drops in China: Research, Patenting, Developing, Licensing, and Innovation"

Field: Scientometrics and research evaluation applied to pharmaceutical innovation and industrial change

Recommendation: Major revision required

Summary and Recommendation

This manuscript revisits a received view in innovation studies and business history — that Japanese pharmaceutical corporations have declined in recent decades, and that the cause is an internal research and development dilemma. The authors pursue the question through a well-chosen vehicle: the eye-drop segment of the mainland China market, a niche historically led by Japanese and Swiss-German firms. The case-selection logic is a genuine strength, since examining a third-party market neutralises many home-market institutional advantages that confound cross-national firm comparisons, while choosing a market that simultaneously hosts a rising indigenous challenger gives leverage on both halves of the question. Organising the evidence around five phases of a drug's life cycle, the study assembles Web of Science records, CNIPA patent filings, ClinicalTrials.gov registrations, licensing transactions from AdisInsight and MarketLine, and approval records from the National Medical Products Administration, tracking Santen, Alcon, Allergan, and Bausch + Lomb against Chinese domestic firms from the late 1970s to the early 2020s. The authors report a sharper contraction among Swiss-German than Japanese firms, and conclude that decisive pressure operates in clinical development rather than early discovery.

A second strength deserves equal emphasis: the insistence on disaggregating the value chain rather than treating pharmaceutical competition as a single contest. Most work on catch-up asks whether latecomers close the gap; this study asks where along the chain it closes first, which is the better question. Several descriptive findings survive independent checking — the absence of any Allergan eye-drop patent filing in China and of any Bausch + Lomb eye-drop trial there are both confirmed by the appendices, and every descriptive statistic in Appendix Table 1 reproduces exactly. Nevertheless the manuscript cannot be published as it stands. Its indicators are not validated against the constructs they represent; its retrieval instruments appear differentially sensitive to the two groups being compared; the analytical methods announced in the Materials and Methods section are not performed; several reported percentages cannot be reproduced from the appendices that supposedly contain them; the organising framework is never defined; and several propositions in the Discussion contradict the Results. Each is remediable, and the research is worth the substantial revision required.

Major Concern Category 1: Measurement Validity and Data Construction

1.1 The core indicators are unvalidated, and retrieval appears differentially sensitive by actor type

The section titled "Early-Stage Research" rests entirely on a Web of Science query matching "Eye Drop" against title and keyword fields, yet the manuscript nowhere argues that indexed publications index early-stage research activity. The problem is acute at firm level, because pharmaceutical companies publish selectively and work through partners whose addresses do not resolve to the sponsor: Appendix Table 3 credits the four focal corporations with roughly one hundred records in total, Santen holding twenty-three with zero in most years before 2009, despite having been founded in 1890 and having entered China in 1988. Reading those zeros as evidence about research activity rather than publication propensity is not defensible without further argument, and at country level the attribution rule is never stated, so the series measures where authors are located rather than where innovation for the Chinese market occurs. A parallel concern attends the patent analysis. Footnote 2 documents a CNIPA search restricted to the title field and a single Chinese dosage-form term, whereas multinationals typically enter via the Patent Cooperation Treaty with titles translated as compositions or formulations, while domestic applicants commonly use the dosage-form noun directly. Such a query would retrieve domestic filings at a higher rate for reasons unrelated to inventive activity, and Appendix Table 4 is consistent with exactly this: fifty-one foreign filings against one thousand one hundred and seventy-four domestic ones, with Alcon appearing once in the entire series despite a Chinese subsidiary since 1995. Because this section supplies the finding that domestic firms captured the middle-upper stream first, a differentially sensitive instrument would render it an artefact. Documented robustness exercises across search fields, lexical variants, and classification-based retrieval, together with triangulation of the publication series against pipeline or priority-date data the authors already license, would address both concerns.

1.2 Source coverage, units of analysis, and scope discipline

Three further construction problems each undercut a specific claim. Table 1 names ClinicalTrials.gov as the sole trial source, but mainland China operates two further platforms — the Chinese Clinical Trial Registry, a World Health Organization primary registry among the world's largest, and the regulator's mandatory registration platform — and published comparisons find domestic Chinese trials splitting roughly seventy to thirty in favour of the domestic registry, while multinational sponsors register on ClinicalTrials.gov as a matter of course. Appendix Table 5 accordingly holds only seventy-seven trials across twenty-two years, with eight empty years. The resulting bias is not straightforwardly convenient for the authors, since undercounting domestic trials would understate the domestic rise; but it cuts both ways, because if domestic registration propensity has risen since the Chinese registry launched in 2007, part of the apparent surge is a reporting artefact. Second, the approval analysis counts regulatory records rather than products, and Chinese registration issues approval numbers per strength, presentation, and site, which is not comparable across groups when domestic generic manufacturers hold many registrations while a multinational markets one branded presentation. Appendix Table 7 also reports seven hundred and eleven domestic approvals in 2015 and seven hundred and forty-seven in 2021 against neighbouring values of fifty, twenty, and fifty-four — order-of-magnitude departures, in years associated with major administrative re-registration exercises, that are neither flagged nor tested for influence though two such observations dominate any share or moving average passing through them. Third, four of the thirteen licensing transactions in Appendix Table 6 fall outside the mainland-China scope the table's title asserts, having worldwide territories and no Chinese counterparty; their removal leaves only the 2006 Pharma Stulln deal in the early period, which cannot support the claim that Japanese and German corporations jointly contributed the early deals. The dataset also selects on disclosure, with six of thirteen values undisclosed, so the observation that all Chinese licensees are listed big pharma may be a fact about who discloses rather than who transacts.

Major Concern Category 2: Analysis, Inference, and Internal Arithmetic

2.1 Reported percentages cannot be reproduced, and the selection documentation does not reconcile

The Results report America at 655 records and 16.71 percent, Japan at 330 and 8.42 percent, Italy at 316 and 8.61 percent, and England at 277 and 7.67 percent. Recomputing implied denominators yields approximately 3,919 for the United States, Japan, mainland China, Germany, and France, but 3,670 for Italy and 3,612 for England, which is impossible on a common corpus. The Italian figure produces a direct conflict: 316 records cannot represent a larger share than Japan's 330, so the two statements cannot both hold, and the description of Italy as ranking third agrees with the counts while contradicting the percentages the authors themselves report. On a denominator of 3,919 Italy would be approximately 8.06 percent and England 7.07 percent, restoring both consistency and the stated ranking. A traceability problem compounds this: summing Appendix Table 2 gives 655, 330, and 250 for the USA, Japan, and mainland China, matching the text exactly, but the appendix totals 4,094 rather than 3,919, on which the American share would be 16.00 percent — while Italy, England, and France have no columns at all, Germany is aggregated with Switzerland though the text reports a Germany-only figure, and the residual "Others" column of 2,537 records, more than half the corpus, is never decomposed despite the paper's thesis concerning emerging economies generally. The PRISMA diagram offered as selection documentation compounds matters. Its identification and deduplication steps reconcile exactly to 5,154 records screened, but 55 human exclusions should leave 5,099 whereas the diagram reports 4,099 sought for retrieval, losing one thousand records at a single transition while every other step reconciles cleanly. The diagram also names Scopus and PubMed, which appear nowhere else in the manuscript, whereas Table 1 and footnote 1 name Web of Science alone; since Appendix Table 2 sums to exactly the diagram's included count of 4,094, Table 1 appears to be the error, and two search strategies require documentation.

2.2 Policy effects are attributed without identification, and one attribution is contradicted by the data

The manuscript attributes share movements to the 2008 financial crisis, to China's Healthcare System Reform of 2009 and its Drugs and Medical Devices Review–Approval Systems Reform of 2015, and to the trade dispute beginning in 2018. Each is a claim about a discrete event's effect on a time-varying outcome, supported only by the proximity of a dashed line to a visual inflection in a smoothed series, with no estimation, counterfactual, or comparison case. More seriously, the 2015 attribution runs against both the manuscript's data and the established characterisation of that reform, which together with accession to the International Council for Harmonisation in 2017, acceptance of foreign clinical data, and the priority review and urgently-needed-overseas-drug pathways is documented as having reduced barriers for imported new drugs and shortened launch delay. Appendix Table 7 agrees with that literature rather than with the manuscript, showing imported eye-drop approvals rising from seventeen in 2016 and seventeen in 2017 to twenty-eight in 2018 — the strongest post-reform year. The reform credited with undermining multinational position is thus followed, in the authors' own table, by multinationals' best post-reform result. The trade-dispute claim encounters the same problem, since the increase occurs in the dispute's first year while the sharp fall occupies 2019 through 2022, coinciding with the approach of the data cut-off and with 2022 being a one-quarter partial year; attributing a terminal decline in a right-censored series to an external event requires ruling out censoring first. The 2008 attribution is similarly unlocalised, as Appendix Table 4 shows foreign filings of zero, zero, one, one, and seventeen across 2008 to 2012, placing the movement four years after the crisis and driving it almost entirely through fifteen Bausch + Lomb filings in one year.

2.3 The announced time-series analysis is absent, and the concentration index is never reported

The Materials and Methods section commits the study to "scientometrics, data visualization, and time-series analysis" and introduces the Herfindahl–Hirschman Index for assessing concentration. Visualisation is delivered; neither of the others is. The manuscript contains no model, coefficient, interval, test statistic, or break test, offering instead moving averages described with directional verbs and no means of distinguishing these readings from noise. The counts make this acute: Santen's four non-consecutive trials, in 2010, 2013, 2019, and 2022, are described as a steadily increasing share, and Alcon's single recorded filing in 2010 as a decrease since the late 2000s, though a series with one non-zero observation has no slope. Smoothing compounds rather than mitigates this, since five-year windows are used in Figures 1 and 4 and nine-year windows in Figures 2 and 3 with no rationale, no statement of whether averages are centred or trailing, and no endpoint rule — decisions that shift apparent inflection dates by several years when the reforms to be dated are only six years apart. The consequences are visible internally: Appendix Table 5 records a raw foreign trial share of one hundred percent in 2022 while Figure 3a places it near five percent, and the claimed 2007 peak is supported by neither the table, which records no foreign trials that year, nor the figure, whose own peak falls near 2005. The concentration index fares worse. No value appears anywhere in the manuscript; judging from Figure 3b it is plotted on a percentage axis though the index is conventionally reported on a zero-to-one or zero-to-ten-thousand scale; and it appears computed over five foreign-sponsor bands summing to one hundred percent by construction, which makes it invariant to the domestic sector the paper argues is displacing those firms. The Methods also assert an extremely high concentration ratio as a premise justifying case selection while introducing the index as the instrument for measuring concentration, and cite Appendix Table 1 for a seventy-five percent share claim though that table contains only founding years, entry dates, and market capitalisations.

Major Concern Category 3: Framework, Sample, and Internal Consistency

3.1 The framework is asserted rather than developed, and national conclusions rest on single-firm samples

The abstract presents framework development as a contribution, yet the Materials and Methods section refers to "these five phases of a drug's life cycle" with no antecedent anywhere prior. The phases must be reconstructed from section headings, and even then licensing is not a temporal phase but a transaction type occurring at any stage — as Appendix Table 6 confirms by recording deals spanning preclinical to marketed assets. Because no framework is stated, no expectations are derived and the analysis has nothing to test: the ordering of domestic advance is read from the data and reported as a finding, leaving readers unable to judge whether it is surprising. The absence also leaves the paper's explanatory mechanism unsupported, since the Conclusions attribute heterogeneous multinational performance to the heterogeneous role of eye drops as simultaneously drug, consumer good, and historically medical device — a genuinely interesting idea, but one for which no series on brand, channel, over-the-counter status, or price is supplied, resting instead on a single citation concerning reputation. Compounding this, conclusions stated at the level of national industries rest on one firm apiece. The abstract's "representative European pharmaceutical corporations" describes Alcon alone, and exactly one of the four focal firms is Japanese, even though the manuscript itself names Santen, Senju, Rohto, and Lion as the relevant Japanese specialists and then selects one. Senju appears in the patent discussion with five Chinese filings, five times Alcon's entire recorded total; Rohto, a major presence in the Chinese consumer market, appears only as having published one article in 2013. The exclusion works against the argument, since the consumer-good mechanism applies most directly to the over-the-counter specialists omitted, and the stated seventy-five percent selection premise holds for patent filings alone — in Appendix Table 3 the "Other MNCs" column exceeds the four focal firms combined in several years.

3.2 Propositions in the Discussion contradict the Results, and firm-level claims conflict with the appendices

Several conflicts occur precisely where the paper states its contribution. The Licensing section reports Chinese royalties of seven to ten and five to twelve percent as remaining "at relatively low levels," while the Discussion states that licensed-in first-in-class eye drops carry "relatively high sales royalties in most cases"; the same royalties cannot be both, and neither claim is well supported, resting on two deals per side with the same licensor appearing on both and no control for territory, stage, or upfront structure. The abstract frames Japanese decline as the phenomenon to be explained while the Conclusions hold that Japanese firms maintain or strengthen their commercialisation position, though Appendix Table 7 gives Santen's approvals as twenty-five in 2015, thirteen in 2018, and then one, zero, one, and zero. The Discussion dates domestic patent dominance to "the 1980s and the 1990s," but Appendix Table 4 begins in 1989 with four consecutive all-zero years and 1990s domestic counts in low single digits. The claim that all Chinese licensees are mature listed big pharma is excepted two sentences later by OcuMension Therapeutics, which the manuscript elsewhere dates to 2017 and lists among recent start-ups. The firm-level record contains parallel conflicts, localised rather than systemic, since the Bausch + Lomb and Allergan patent claims both check out. The text dates Santen's Chinese subsidiary to 2015 whereas Appendix Table 1 gives 2005 — demonstrably the figure used in that table's statistics, since the reported subsidiary mean of 1999 reproduces only from 1995, 2009, 1988, and 2005 — and the inflection the text explains by that date is absent, Santen's filings being zero in every year from 2014 through 2022. Underlying several of these is an unstated rule for corporate identity across mergers: Novartis records from 2011 to 2019 are counted as Alcon though Novartis is a diversified group whose address appears across all therapeutic areas, Allergan's AbbVie period raises the same question more acutely, and Bausch + Lomb's passage through Valeant ownership is unannotated though its filings concentrate precisely in that window. Appendix Table 1 also assigns Allergan a market capitalisation at March 2023 though Allergan ceased separate listing in 2020, and that figure propagates into the reported mean and standard deviation. Finally, the Conclusions' second paragraph asserts findings on safety, quality, access, equity, and regulatory compliance that no indicator in the study measures.

Minor Issues

1. Framework and structure at first use. The reference to "these five phases" has no antecedent, and a "RESULTS" heading is followed immediately by a subheading with no intervening text. A passage defining the framework and its boundaries should precede the first reference, and a brief orienting paragraph under Results would give the authors a natural place to state their analytical strategy.

2. Source documentation. Table 1 presents one database per metric and reads "Patnet" for patent, but more substantively omits AdisInsight and Drug Future, both documented in footnotes as sources actually used. Figure 2a is cited in the text but has no caption, leaving its averaging window and source undocumented while Figure 2b is fully captioned.

3. Mixed data vintage and a partial terminal year. All databases were accessed on 4 April 2022 while Appendix Table 1 reports market capitalisations as of 26 March 2023. The 2022 row consequently captures one quarter — forty-nine publication records against three hundred and ninety-nine in 2021, and zeros throughout Appendix Table 3 — producing an artefactual terminal decline that should be excluded or clearly marked.

4. Inconsistent coverage windows. Appendix Table 2 begins in 1971, Table 3 in 1981, Table 4 in 1989, Table 7 in 1999, and Table 5 in 2001, while the abstract describes a uniform late-1970s-to-early-2020s window; Table 3 also omits 1987 entirely. The staggered coverage should be stated and its implications for cross-stage comparison discussed.

5. Unenumerated residual categories. The "Other MNCs" column carries weight across four appendix tables while its membership is disclosed only piecemeal. Relatedly, the twelve residual foreign patent filings implied by the 76.47 percent figure are attributed only to Senju, Taisho, and Schering-Plough, totalling eight, leaving four unaccounted; Schering-Plough's 2009 acquisition by Merck also goes unremarked though it determines the national column.

6. Unsourced assertions. The explanation that Bausch + Lomb runs no Chinese trials because activities are outsourced to Shenzhen Relin Medicine is uncited, and that sentence contains an ellipsis suggesting a transcription problem. The characterisation of Pharma Stulln's indirect participation is likewise uncited, as is the account attributing the domestic start-up cohort to capital-market access and R&D subsidies.

7. Reporting transparency. Approval retrieval by dosage-form name will miss ophthalmic solutions registered under other nomenclature, with no validation of recall offered, and Drug Future's provenance and deduplication against the NMPA record are undocumented. Shares are reported throughout with denominators suppressed, and how cells with zero numerator and denominator enter the figures is unstated.

8. No robustness analysis, and unconventional axes. Given the discretionary choices involved — window length, four focal firms, keyword filters, single registries, a partial terminal year — the absence of any sensitivity check is conspicuous, and one appendix table showing how principal shares move under alternative specifications would materially increase confidence. Figures 3a and 3b also run right-to-left chronologically, inverting the visual reading of trend direction.

9. Overreaching framing. The claim that almost no prior scholarship addresses eye drops specifically needs documenting or softening. The two decline hypotheses also conflate geographic scope with causal mechanism and are not exhaustive, since a universal decline could have internal causes; restating so that scope and mechanism vary independently would sharpen the paper considerably.

10. Presentation and text-figure mismatches. The Conclusions use author-date form within a numbered system. The passage on Menarini, SIFI, and intraocular lenses is disproportionate to Italy's role and concerns non-eye-drop devices. Appendix Figure 3 includes Qilu Pharmaceutical, never discussed, while omitting Cloudbreak and Affamed, which are; molecule names require correction to International Nonproprietary Names. The recommendations to adopt "fast follow" while focusing more on early-stage research also pull in opposite directions.

Missing and Inadequately Integrated Literature

Cited but Inadequately Integrated

Several works in the reference list are invoked once and left unused where fuller engagement would resolve problems identified above. Rafols et al. (2014) is cited only to establish European contraction, yet it confronts the very difficulty of interpreting corporate publication counts across periods of merger and restructuring; its conventions for handling corporate name changes and address attribution are exactly what this study lacks, and engaging them would let the authors justify their Alcon-Novartis and Allergan-AbbVie treatments by reference to established practice. Hashimoto and Haneda (2008) supplies the premise being reassessed, but its efficiency construct is never brought into contact with the manuscript's share-based indicators, which matters because the two can move independently: Appendix Table 2 shows Japan's absolute output rising from two records in 1978 to twenty-six in 2021, with only its share falling as the global corpus grows faster. The authors should clarify whether their claim concerns capability or relative position. Shimura, Masuda, and Kimura (2014) defines the paper's foil, yet its argument is never described, so readers cannot judge whether the evidence bears on it, and Umemura (2010, 2014) is confined to chronology although its subject is systems of innovation — the stage-by-stage findings could show which components of the Japanese system did and did not travel abroad. Meng et al. (2019) is cited only to date the 2009 reform, a costly omission given that its components operate very differently on imported branded products than on domestic generics; specifying which would produce which movement would replace coincidence with mechanism. Perri, Scalera, and Mudambi (2017) supplies organising vocabulary without shaping the analysis, though its ranking of knowledge conduits is directly testable against the four this study measures. Finally, Wang and Li-Ying (2015) examines when licensed-in technology builds innovative capability, which is precisely what the Conclusions assert it does not; Kobrak (2013) carries the entire theoretical weight of the consumer-good mechanism alone; and Subramanian, Lévesque, and Van De Vrande (2020), Oliver (2001), Kotabe et al. (2007), Huang (2023), and Mathews, Hu, and Wu (2011) each address mechanisms central to the licensing and development stages while appearing only as closing citations.

Truly Missing Papers

Lee and Malerba (2017) is the most consequential absence, supplying what the manuscript lacks: a stated theory of how industrial leadership changes, built on four stages and three windows of opportunity in technology, demand, and institutions. The manuscript's narrative is a catch-up-cycle narrative, its 2009 and 2015 reform lines are institutional-window arguments in all but name, and its finding that incumbents held both ends of the chain while losing the middle maps onto the incumbent-trap mechanism. Lee and Lim (2001) complements this by distinguishing path-following, path-skipping, and path-creating catch-up; the manuscript's own findings describe a recognisable path-following pattern that naming would give theoretical content. Scannell, Blanckley, Boldon, and Warrington (2012) is indispensable, being the canonical account of industry-wide R&D productivity decline and therefore the strongest statement of the generality hypothesis the paper ultimately favours; Pammolli, Magazzini, and Riccaboni (2011) extends it by showing that declining returns concentrate in particular therapeutic areas, which matters because the manuscript generalises from one narrow segment. Hu and Jefferson (2009) bears directly on Appendix Table 4, documenting that China's patent surge was driven substantially by foreign investment intensity, patent-law amendments, and ownership reform rather than R&D intensity — a counter-explanation requiring explicit treatment — while Hu (2010) notes that foreign applications for Chinese patents grew above thirty percent annually, raising a pointed question about whether fifty-one foreign filings across three decades reflects the filing population. Lopez Bernal, Cummins, and Gasparrini (2017) is the single most useful methodological addition, addressing interventions at defined time points with counted outcomes and treating over-dispersion and autocorrelation directly. Donthu, Kumar, Mukherjee, Pandey, and Lim (2021) supplies the complementary reporting standard for query documentation and name disambiguation; Page, McKenzie, Bossuyt, et al. (2021) should be cited given the PRISMA diagram's use; and Kneller (2010) provides the developmental benchmark against which the 2015-to-2019 biotech cohort should be assessed. All have been verified to exist and confirmed absent from the manuscript's twenty-eight numbered references.

Reference List

Donthu, N., S. Kumar, D. Mukherjee, N. Pandey, and W. M. Lim. 2021. J Bus Res 133:285-296.
Hu, A. G. Z. 2010. Res Policy 39(7):985-993.
Hu, A. G. Z., and G. H. Jefferson. 2009. J Dev Econ 90(1):57-68.
Kneller, R. 2010. Nat Rev Drug Discov 9(11):867-882.
Lee, K., and C. Lim. 2001. Res Policy 30(3):459-483.
Lee, K., and F. Malerba. 2017. Res Policy 46(2):338-351.
Lopez Bernal, J., S. Cummins, and A. Gasparrini. 2017. Int J Epidemiol 46(1):348-355.
    [Corrigendum: Int J Epidemiol 50(3):1045, 2021.]
Page, M. J., J. E. McKenzie, P. M. Bossuyt, et al. 2021. BMJ 372:n71.
Pammolli, F., L. Magazzini, and M. Riccaboni. 2011. Nat Rev Drug Discov 10(6):428-438.
Scannell, J. W., A. Blanckley, H. Boldon, and B. Warrington. 2012. Nat Rev Drug Discov
    11(3):191-200.

Conclusion and Path Forward

The intellectual instincts behind this manuscript are sound, and that deserves recording plainly. Selecting a third-party market to neutralise home-market advantages is a genuinely good design idea. Decomposing the value chain to locate where displacement begins, rather than merely asking whether it occurs, is a genuinely good analytical idea that yields the kind of result which generates further research. Assembling five heterogeneous databases around a single therapeutic segment represents substantial labour and produces a description of sectoral change difficult to obtain otherwise. Several descriptive findings survive independent checking, Appendix Table 1's statistics reproduce exactly, and the authors are candid both about the limits of a single case and about the largely generic character of Chinese domestic output. That the firm-level lapses are localised — the Bausch + Lomb and Allergan claims checking out where the Santen and Alcon claims do not — suggests careful work interrupted by specific errors rather than an unreliable record.

What stands between the manuscript and publication is the distance between its conclusions and the warrant offered for them, and much of the required work can be done with data the authors already hold. The percentage inconsistencies appear correctable by recomputation, and the discontinuity in the selection diagram looks like a transcription matter. The patent and approval retrieval concerns call for documented robustness exercises across search fields, lexical variants, and classification-based strategies. The trial analysis needs extending to the Chinese registries, reporting both the combined and single-registry series so readers can see what depends on source choice. The inferential gap can be closed by fitting count models with the 2009 and 2015 interventions entered explicitly and reporting intervals, standardising the smoothing convention, removing the partial 2022 year, and reporting the concentration index on a conventional scale over a partition that includes domestic firms — or by abandoning percentages where event counts are too sparse, narrating individual events instead. The framework must be stated and used to derive at least one testable expectation. The sample should be widened to include the Japanese specialists the manuscript itself names, or the conclusions lowered from national industries to the firms observed. Each contradiction should be resolved in favour of whichever statement the data support, and the paragraph on corporate social responsibility removed. The research question merits this investment, and a manuscript completing this work would make a considerably more durable contribution than the present version can claim.