Computational surveillance of Colombian public pharmaceutical procurement using public administrative data: A reproducible analysis of a closed 2020–2025 cohort Authors: Andrés Soto, Sarah Jose, Xin Sun, Patrick Goymer, Patrick Goymer Published: 2026 ABSTRACT IntroductionBigLoI monitors Colombian public pharmaceutical procurement from 2015 onward. For this manuscript, the source cohort comprised 162,271 candidate pharmaceutical contracts from 2020 to 2025. After contract-level review excluded 441 records explicitly concerning veterinary, animal-health, or agricultural use, the corrected closed analytical cohort comprised 161,830 contracts.ObjectiveTo describe the design, implementation, and findings of a reproducible computational infrastructure for surveillance of Colombian public pharmaceutical procurement.MethodsPublic APIs from SECOP-II, INVIMA, and SISMED were integrated into a reproducible architecture combining PostgreSQL, statistical analysis, and a public-facing observatory. Candidate records whose supplier names indicated possible veterinary or agricultural activity underwent contract-object review; 441 records with explicit animal, veterinary, or agricultural scope were excluded, while human-health and ambiguous records were retained conservatively. A Z-score engine was implemented to flag contracts with atypical total values within therapeutic categories. As a clearly secondary technical-feasibility module, smart-contract automation of a payment workflow was tested on the Sepolia testnet only to verify predefined digital state transitions under simulated conditions. The corrected closed analytical cohort included 161,830 pharmaceutical contracts from 2020 to 2025, while post-2025 records remained available only for live platform monitoring. Monetary results were reported primarily in COP; secondary USD equivalents were included only as approximate interpretive references, with detailed conversions relegated to S1 Table.ResultsAmong 146,594 contracts analyzed in categories with at least 10 observations, 664 contracts (0.45%) triggered a statistical alert with absolute Z-score greater than or equal to 1.5 sigma. The alert rate among Z-score-eligible contracts increased from 0.30% in 2021 to 1.31% in 2025. Antibiotics showed a category-level maximum Z-score of 8.89. The top 3% of suppliers concentrated 85.8% of total contracted value. As a clearly secondary module, the Sepolia smart-contract prototype confirmed only that predefined digital state transitions could be executed under simulated testnet conditions; it provides no evidence of real-world payment-cycle reduction, realized savings, or institutional deployability.ConclusionsA reproducible national-scale computational infrastructure identified atypical procurement patterns and documented marked market concentration in Colombian public pharmaceutical procurement. These descriptive findings may inform auditing, public health policy discussions, and health data governance, but they do not by themselves establish wrongdoing or prescribe specific reforms. Statistical alerts remain exploratory prioritization signals rather than evidence of corruption or fraud. The blockchain module should be interpreted strictly as a complementary technical proof of concept tested under simulated conditions and not as operational evidence on real-world payment performance, savings, or implementation readiness. 1 Introduction Public procurement of medicines in Colombia is monitored in BigLoI through a broad SECOP-II universe that now exceeds 272,000 contracts from 2015 onward. For the purposes of this manuscript, the corrected closed analytical cohort corresponds to 161,830 pharmaceutical contracts from 2020 to 2025 after contract-level veterinary exclusions. Despite this scale, the system lacks reproducible real-time surveillance tools to detect atypical procurement patterns, market concentration, or potential supply risks. The World Health Organization has estimated that 10% to 25% of global health expenditure is lost to corruption, inefficiency, and administrative errors [1]. In Colombia, official audits have documented high-cost medicine overpricing, prolonged payment delays, and recurring episodes of shortages [2–4]. Integration of public APIs makes it possible to build procurement-surveillance platforms without access to confidential information. Z-score analysis is a practical approach for detecting atypical patterns in large procurement datasets [5], while smart contracts and semantic retrieval expand the range of possible automation and evidence-access tools [6,7]. Initiatives such as ProZorro, OCDS, and OLAF illustrate the feasibility of these approaches in public-procurement transparency [8–10]. BigLoI (Business Intelligence for Government Logistics and Operations Intelligence) was developed to address this gap. This article documents its methodology and findings as a contribution to the debate on computational tools for governance of the Colombian health system. 2 Methods 2.1 Study design This was a technological-development study with longitudinal observational analysis of public procurement data. Only openly accessible Colombian public data sources were used. No clinical data or patient-level personal data were included. Ethics statement. This study used only publicly accessible administrative procurement records published by Colombian open-data sources (SECOP-II via datos.gov.co, INVIMA, and SISMED). It did not involve human participants, patient-level or clinical data, identifiable private information, biological samples, recruitment, intervention, or any contact with individuals. Accordingly, the study does not constitute human-subjects research and did not require review or approval by an institutional review board or research ethics committee. Under Colombian Resolution 8430 of 1993, analysis of public, non-identifiable administrative records corresponds to research without risk and falls outside the scope of human-subjects ethics review. Informed consent was not applicable because the study did not involve human participants or identifiable personal data. All analyzed data are publicly available and were used in aggregate for the surveillance of public procurement. 2.2 Data sources SECOP-II (Electronic Public Procurement System). The public Socrata API from datos.gov.co was used for automated download of pharmaceutical procurement contracts. Initial inclusion required the presence of at least one of 19 pharmaceutical terms in the contract-object description (S1 Appendix). This keyword strategy is a high-recall screen rather than a gold-standard classifier: it can generate false positives (non-pharmaceutical or animal-health contracts that share lexical terms) and false negatives (human pharmaceutical contracts whose objects omit the inclusion vocabulary). Because terms such as vaccine and antibiotic can also describe animal-health procurement, supplier names were screened for veterinary or agricultural indicators and 519 candidate records underwent contract-object review. Records were excluded only when the object explicitly indicated veterinary, animal-health, zoonotic intervention on animals, or agricultural use outside the human-pharmaceutical scope. This review excluded 441 records, retained 27 records with explicit human-health or lexically incidental scope, and conservatively retained 51 ambiguous records. The review decisions and source-row identifiers are included in the reproducibility package. No independently adjudicated external reference set was available to estimate sensitivity or specificity of the full keyword taxonomy; the contract-level review therefore serves as targeted quality control for a known contamination pathway rather than as complete taxonomy validation. The platform monitoring window extends from January 2015 onward, but the closed analytical cohort used in this manuscript was restricted to complete calendar years 2020–2025. Recovered local PostgreSQL source used to reproduce the manuscript cohort: 956,157 total SECOP-II records; 165,549 candidate pharmaceutical records including live 2026 monitoring rows; 162,271 candidate contracts in the source 2020–2025 cohort; 161,830 contracts after contract-level veterinary exclusions. SECOP-II monitored universe available to the platform: 272,814 unique contracts; COP 42 trillion; 163,135 suppliers; 37 regional codes. To avoid comparing incomplete calendar periods, all year-by-year descriptive analyses in the manuscript were closed at 31 December 2025. The platform continues to ingest newer records for operational monitoring, but post-2025 records were excluded from the formal analytical cohort reported here. All primary monetary values are reported in COP. When approximate USD equivalents are provided in parentheses, they are secondary interpretive aids calculated using a round reference exchange rate of COP 4,000 per USD, close to late-2025 market levels. No analyses were performed in USD. Because a single round reference rate is applied across the entire 2020–2025 study period, the USD equivalents do not account for exchange-rate volatility and should be read as order-of-magnitude approximations only; the COP values are the analytical quantities. Detailed COP/USD reference conversions for the principal manuscript figures are provided in S1 Table. Source data are provided for Table 1 (S1 Data), Table 2 (S2 Data), Fig 1 (S3 Data), Fig 2 (S4 Data), Fig 3 category and yearly panels (S5 and S6 Data), and Fig 4 Lorenz and top-10 supplier panels (S7 and S8 Data); the column dictionary and source-file manifest are provided as S9 and S10 Data, and the SECOP-II filtering taxonomy, simplified database schema, and smart-contract protocol are provided as S1 Appendix. 10.1371/journal.pone.0350967.t001Table 1Summary of the public pharmaceutical procurement corpus and the closed analytical cohort used in this manuscript.VariableValueTotal SECOP-II contracts (API universe)272,814Pharmaceutical contracts in source closed cohort before contract-level review162,271Pharmaceutical contracts in corrected closed cohort161,830Total contracts in recovered local PostgreSQL source956,157Total contracted value (corrected pharmaceutical cohort)COP 16.93 trillion (approx. USD 4.23 billion)Total contracted value (all sectors recovered local source)COP 294.7 trillionTotal contracted value (API universe)COP 42.00 trillionMean value per positive-value pharmaceutical contractCOP 104.69 millionUnique normalized suppliers in corrected positive-value cohort50,225Regions covered36-37Period (closed pharmaceutical cohort)2020–2025Platform monitoring windowJanuary 2015 onwardProcessed INVIMA records9,838 (all current)SISMED reference prices44,038 records; 1,759 unique ATC codes (2017–2019)Indexed RAG documents9,336 Summary of the public pharmaceutical procurement corpus and the closed analytical cohort used in this manuscript. 10.1371/journal.pone.0350967.t002Table 2Technical architecture of the BigLoI platform for surveillance of public pharmaceutical procurement.LayerMain technologyFunctionCollectionPython; Socrata APIIncremental ingestion of SECOP-II, INVIMA, and SISMEDStoragePostgreSQL; Pinecone; MongoDBRelational, vector, and unstructured data storageProcessingFastAPI; pandasCleaning, normalization, ABC classification, Z-score engineGenerative AIClaude 3.5 Sonnet; GPT-4oHybrid semantic plus TF-IDF retrieval-augmented generationMachine learningscikit-learnDemand prediction (R2>0.85), k-means (6 groups), PCASmart contracts (proof of concept)Solidity; Chainlink CRE; Sepolia testnetSimulated payment-state transitions across five digital states; invoice NFTs (secondary module)VisualizationReact/TypeScriptPublic observatory with time series, maps, and alerts Technical architecture of the BigLoI platform for surveillance of public pharmaceutical procurement. Annual evolution of public pharmaceutical procurement monitored in SECOP-II (closed cohort, 2020-2025): number of contracts and total contracted value. Geographic distribution of total value in public pharmaceutical procurement monitored in SECOP-II by department (closed cohort, 2020-2025). INVIMA (National Food and Drug Surveillance Institute). A total of 9,838 records from the national sanitary registry were downloaded through the Socrata API. Variables included brand name, active ingredient, marketing authorization holder, sanitary registration number, ATC code, dosage form, route of administration, relevant dates, and registry status. All loaded records were current. SISMED (Medicine Price Information System, Ministry of Health). We processed 44,038 reference-price records covering 1,759 unique ATC codes, aggregated by dosage form, distribution channel, and reference year (2017–2019). These data were used as contextual market references and as complementary input for interpreting atypical contract-value findings. 2.3 BigLoI technical architecture The platform was implemented as a monorepo with a seven-layer architecture (Table 2). Distribution of statistical contract-value anomalies detected by the Z-score engine, by therapeutic category and year (2020-2025). Market concentration in public pharmaceutical procurement monitored in SECOP-II (corrected closed cohort, 2020-2025): cumulative value concentration by top-ranked suppliers and top 10 supplier shares. Conceptual scheme of the pharmaceutical payment workflow: current institutional flow versus digital states of the smart-contract prototype. Simplified technical architecture of the BigLoI platform: seven layers for processing public pharmaceutical data. 2.4 Anomaly-detection engine based on Z-scores For each pharmaceutical contract classified into a named therapeutic category, we calculated the standardized deviation of its total contract value relative to the distribution of observed values within the same category. Contracts labeled NO_ESPECIFICADO (unspecified therapeutic category) were excluded from the Z-score engine because they lack a stable within-category comparator. The analysis was further restricted to named categories with at least 10 contracts to avoid unstable estimates: where Zi is the anomaly score for contract i, Vi is the contract’s total value, V¯cat is the mean contract value within the same therapeutic category, and σcat is the standard deviation of that category-specific distribution. This approach detects contracts with atypical total values relative to their category; by itself, it does not establish unit-price overpricing against a regulatory benchmark. Alert levels were classified as follows: 10.1371/journal.pone.0350967.t003LevelCriterionInterpretationCriticalAbsolute Z-score greater than or equal to 3.0 sigmaExtremely atypical contract value within categoryHighAbsolute Z-score from 2.0 to less than 3.0 sigmaClearly atypical contract valueModerateAbsolute Z-score from 1.5 to less than 2.0 sigmaModerately atypical contract valueNormalAbsolute Z-score below 1.5 sigmaWithin the expected range for category 2.5 Market-concentration analysis To characterize the competitive structure of the pharmaceutical procurement market, we calculated descriptive concentration indicators and estimated the proportion of total contracted value accumulated by leading suppliers. Supplier concentration was calculated over positive-value contracts in the closed 2020–2025 pharmaceutical cohort. The Herfindahl-Hirschman Index (HHI) was calculated as the sum of squared supplier shares of total contracted value and is reported on the conventional 0–10,000 scale. Bipartite graphs of buyers and suppliers were constructed to identify recurrent contracting patterns. 2.6 Complementary technical-feasibility analysis of a payment workflow using smart contracts As a clearly secondary module, we implemented a five-state simulator on the Sepolia testnet solely to verify that predefined digital states of a hypothetical payment workflow could be executed in sequence under controlled test conditions. It was not designed as an empirical time-and-motion study; comparison with current practice was made only at the level of temporal order of magnitude, using aggregated institutional references (90-day median; 60–180 day range) rather than a step-by-step operational equivalence between the full hospital workflow and the digital prototype states. The prototype reference scenario of approximately 30 hours should be read only as a test-condition benchmark for the simulator. The illustrative projected savings were estimated as a scenario exercise, not as an empirical economic evaluation, by applying a conservative upper-bound monthly financing-cost assumption of 2% (approximately 26.8% effective annual) to annual contracted value under a hypothetical reduction in administrative delays. That rate was chosen as an order-of-magnitude working-capital parameter aligned with the upper range of legal interest ceilings derived from the Colombian interés bancario corriente certified by the Superintendencia Financiera de Colombia for ordinary/consumer credit, where maximum remuneratory or default interest equals 1.5 times the certified rate (usury ceiling) [11,12]. It is not an empirically measured average supplier financing rate in SECOP-II, nor a claim that pharmaceutical suppliers uniformly borrow at this cost; commercial or rediscount credit lines can be lower. Prolonged payment cycles in the Colombian pharmaceutical supply chain motivate treating delayed settlement as financially costly [13], but the 2% figure remains a scenario assumption only: where Vannual is either the corrected six-year mean annual flow (approximately COP 2.82 trillion) or the corrected 2025 flow (approximately COP 5.50 trillion), rmonthly=2%, and Δt is the theoretical reduction in payment time, expressed in days, obtained when a 90-day cycle is shortened to approximately 30 hours (1.25 days), that is, approximately 88.75 days. 2.7 Statistical analysis Descriptive statistics were used throughout. Log-normality of contract values was assessed with the Kolmogorov-Smirnov test. Market concentration was described using participation and accumulation indicators. The significance threshold was set at α=0.05. 3 Results 3.1 Characteristics of the data corpus The characteristics of the corpus and the closed 2020–2025 analytical cohort are summarized in Table 1. Pharmaceutical contract value followed a log-normal distribution (Kolmogorov-Smirnov D=0.04; p<0.001). The top 10% of contracts by value accounted for 78% of total contracted value. Annual evolution of indexed pharmaceutical contracts (closed cohort, 2020–2025): 10.1371/journal.pone.0350967.t004YearContractsValue (COP billions)202035,2423,294.5202153,7232,846.1202230,9272,639.6202310,6481,527.920249,3351,124.0202521,9555,496.8 Fig 1 shows the annual evolution of the number of contracts and total contracted value across the closed 2020–2025 cohort. The peak in contract counts in 2021 is consistent with the COVID-19 response period, whereas the value peak in 2025 coincided with increased participation of high-cost oncology procurement. Post-2025 records remain available in the live platform but were excluded from the closed annual comparisons reported here. The pronounced decline in indexed contracts in 2023–2024 (from 30,927 in 2022 to 10,648 and 9,335, respectively) is also present in an independent reconstruction of the public SECOP-II API, which recovered 10,006 and 8,818 candidate pharmaceutical contracts for those years before the contract-level correction. The decline is therefore present in the SECOP-II source under the inclusion taxonomy and is not an artifact of the local PostgreSQL snapshot. The available administrative data do not allow us to adjudicate whether it reflects procurement consolidation into fewer, higher-value contracts, changes in contract-object coding, or an actual reduction in pharmaceutical contracting volume; we therefore treat this pattern as a source-data and classification limitation requiring dedicated validation rather than assigning it a single causal explanation. Distribution by therapeutic category: 10.1371/journal.pone.0350967.t005CategoryContractsValue (COP billions)Mean value (COP millions)General pharmaceutical81,5899,855.7120.8Medical device15,8632,922.2184.2Unspecified (NO_ESPECIFICADO)15,2262,625.8173.8Vaccine45,935864.618.8Oncology400421.21,052.9Medical supply2,232190.085.1Antibiotic21719.690.2Diabetes27920.473.1Analgesic428.5203.0Antiviral370.924.6 Unspecified contracts accounted for 9.4% of corrected-cohort contracts and 15.5% of total contracted value (COP 2.63 trillion). Oncology contracts showed the highest mean contract value among named categories (COP 1,052.9 million per contract). Contracts labeled NO_ESPECIFICADO are retained in descriptive cohort totals but are excluded from the Z-score engine (Section 2.4). 3.2 Geographic distribution The leading regions by pharmaceutical procurement value were as follows: 10.1371/journal.pone.0350967.t006RegionContractsValue (COP billions)Approx. value (USD millions)% of totalBogota D.C. (combined)18,0209,607.02,401.856.7%Antioquia47,9641,312.2328.17.8%Valle del Cauca5,843696.5174.14.1%Boyaca9,142623.6155.93.7%Tolima3,773426.9106.72.5%Santander4,278403.8101.02.4%Huila4,177403.2100.82.4%Cauca2,015389.697.42.3%Atlantico5,075348.887.22.1% Bogota D.C. accounted for 56.7% of national contracted value, equivalent to approximately USD 2.40 billion at the reference exchange rate, followed by Antioquia (7.8%) and Valle del Cauca (4.1%). Fig 2 maps the geographic distribution of total contracted value by department. Methodological note. Bogota D.C. appears under two regional codes in SECOP-II. Coverage of 36–37 regions exceeds the 32 official departments because SECOP-II includes distinct codings for districts and other territorial entities. 3.3 Contract-value anomalies detected by the Z-score engine Methodological note. The implemented Z-score engine measures deviation of total contract value relative to the distribution within each therapeutic category. Direct comparison against SISMED unit prices would require structured line-item detail, which is not consistently available in SECOP-II. Among 146,594 analyzed contracts in named therapeutic categories with at least 10 observations (after excluding 15,226 contracts labeled NO_ESPECIFICADO and 10 contracts in categories with fewer than 10 observations), the engine identified the following: 10.1371/journal.pone.0350967.t007Alert levelCriterionContracts% of totalCriticalAbsolute Z-score greater than or equal to 3.0 sigma2950.20%HighAbsolute Z-score from 2.0 to less than 3.0 sigma2130.15%ModerateAbsolute Z-score from 1.5 to less than 2.0 sigma1560.11%Total with alertAbsolute Z-score greater than or equal to 1.5 sigma6640.45% These alerts flag contracts whose total value is atypical within their therapeutic category; they are intended for audit prioritization and do not, by themselves, constitute evidence of corruption, fraud, or unit-price overpricing. Results by therapeutic category: 10.1371/journal.pone.0350967.t008CategoryContracts% with alertZmaxMax/mean ratioAnalgesic429.5%4.229.3xDiabetes2797.9%6.6115.5xAntiviral375.4%4.8314.2xAntibiotic2173.2%8.8926.4xMedical supply2,2322.5%22.3488.7xOncology4000.8%19.27138.5x Among the six categories displayed in Fig 3, medical supplies showed the highest maximum Z-score, whereas analgesics showed the highest proportion of contracts with alerts. Temporal evolution of anomaly rates (2020–2025): 10.1371/journal.pone.0350967.t009YearContracts analyzedContracts with alertRate202035,2381200.34%202153,7181630.30%202230,9261260.41%202310,648840.79%20249,335830.89%20256,729881.31% The statistical-alert rate among Z-score-eligible contracts increased from 0.30% in 2021 to 1.31% in 2025. Causal interpretation of this pattern requires additional analysis. The sharp reduction in the 2025 Z-score denominator (6,729 of 21,955 indexed contracts; 30.7%) is explained by categorical filtering rather than by incomplete data upload: all 15,226 NO_ESPECIFICADO contracts in the corrected closed cohort fall in 2025 and were excluded from the engine (21,955 − 6,729 = 15,226), whereas 2020–2024 retained near-complete coverage under the same rules (gaps of at most 10 contracts in those years, corresponding to named categories with fewer than 10 observations). Those unspecified 2025 contracts represent approximately COP 2.63 trillion—about 48% of 2025 contracted value and 15.5% of corrected-cohort value—so the 1.31% alert rate describes only the named-category slice analyzed by the engine and should not be read as an anomaly rate for all 2025 pharmaceutical procurement. Geographic distribution of anomalies. Geographic distribution of anomalies. Departments with the highest rates of statistically flagged contracts were Bogota D.C. (3.80%), Norte de Santander (1.92%), Valle del Cauca (1.21%), and Tolima (1.00%). This distribution indicates heterogeneity in contract values, not causal evidence of corruption. 3.4 Market concentration For value-share concentration metrics, only positive-value contracts with non-empty normalized supplier labels were included: 161,710 contracts and 50,225 suppliers (Table 1). Zero-value or blank-label rows were therefore outside the concentration denominator even when present in the corrected cohort. The top 10 suppliers concentrated 29.01% of total contracted value and the top 3% of suppliers (1,507 of 50,225) concentrated 85.83%. The leading normalized supplier label corresponded to one retained VECOL contract (5.65% of total value) whose object explicitly concerned a Ministry of Health–National Institute of Health agreement for research, development, and national production of human-health immunobiological products; veterinary VECOL contracts were excluded individually. The HHI was 120.99 on the 0–10,000 scale. The detailed distribution is shown in Fig 4. This pattern indicates strong cumulative concentration among leading suppliers, while the HHI remains low under conventional market-concentration thresholds because the contracted value is distributed across a large long-tail supplier base. The bipartite graph identified 847 recurrent entity-supplier pairs with at least five consecutive contracts. 3.5 Smart-contract payment workflow: complementary technical proof of concept Tests in the Sepolia-based BigLoI plus Chainlink CRE simulator verified that the prototype could execute predefined transitions between digital states. Fig 5 contrasts the current institutional payment flow with the digital states of the smart-contract prototype. Reported times are investigator-defined reference times used for functional validation of the prototype and do not represent observed measurements of the full real-world logistics and administrative process. SECOP-II was used as a contractual reference point rather than as an operational stage of hospital workflow, and no hospital payment system was instrumented or experimentally compared. 10.1371/journal.pone.0350967.t010Functional stagePrototype timeValidated contract or enabled order to medicines dispatchedAbout 24 hoursMedicines dispatched to delivery verifiedAbout 4 hoursDelivery verified to CRE registeredAbout 1 hourCRE registered to payment releasedAbout 1 hourFull digital cycle of the prototypeAbout 30 hours The current institutional cycle reported in the literature is 60–180 days (90-day median). This contrast is heuristic and order-of-magnitude only for the digital administrative component; it is not a direct equivalence to the full hospital process, nor evidence that the prototype would reproduce such reductions in practice. Illustrative financial-savings scenario. Illustrative financial-savings scenario. Average annual pharmaceutical procurement flow across the corrected closed 2020–2025 cohort was approximately COP 2.82 trillion (about USD 705 million at the reference exchange rate). Under a theoretical reduction from 90 days to approximately 30 hours and the conservative upper-bound 2% monthly financing-cost parameter described in Methods §2.6, the corresponding scenario output is approximately COP 167,000 million per year (USD 41.7 million). Using corrected 2025 flow, the illustrative output rises to approximately COP 325,000 million per year (USD 81.3 million). These values are scenario outputs only and should not be interpreted as observed savings, budget impact estimates, or implementation-ready business cases. 3.6 Complementary modules: machine learning and semantic retrieval The complementary machine-learning and semantic-retrieval modules (layers 4 and 5 of the platform architecture in Fig 6) showed operational feasibility, but their outputs should be interpreted as exploratory within the scope of this paper. 4 Discussion 4.1 Scale and relevance of the documented problem The corrected corpus of 161,830 pharmaceutical contracts from 2020 to 2025 represents, to our knowledge, the largest validated closed longitudinal series of public pharmaceutical procurement described for Colombia. The log-normal distribution of contract values, the concentration of 78% of total value in the top decile, and the HHI of 120.99 are consistent with procurement markets characterized by a small group of high-value suppliers and a large long-tail supplier base [14]. Because the HHI was computed for the national closed cohort as a whole, it may understate concentration that could exist within specific therapeutic groups (for example oncology, antibiotics, or antivirals); category-specific HHI values were not estimated in this study. Under conventional market-concentration thresholds, an HHI of 120.99 would be interpreted as unconcentrated at the national aggregate level; however, that reading can coexist with high cumulative value shares among leading suppliers (top 3% = 85.83%) when a large long-tail supplier base dilutes the index. Estimating category-specific HHI for oncology, antibiotics, antivirals, and related groups is left for future work and would be required before drawing subgroup competition inferences. The increase in statistical-alert rates from 0.30% in 2021 to 1.31% in 2025 suggests increasing heterogeneity in contracted values during the study period. This pattern warrants dedicated investigation with adjustment for potential confounders such as pharmaceutical inflation and the introduction of new high-cost categories. 4.2 Z-scores as a surveillance instrument The Z-score methodology offers advantages over manual auditing: it operates on large contract volumes, is reproducible, and does not require confidential information. At the same time, threshold-based Z-scores are imperfect screens. Without an adjudicated audit gold standard, sensitivity and specificity cannot be estimated. The exploratory |Z| ≥ 1.5 threshold favors prioritization recall and can increase false positives (statistically atypical but legitimate high-value contracts), while false negatives remain possible when categories are internally heterogeneous, when contracts fall in NO_ESPECIFICADO, or when category sample sizes are small. Future linkage with SISMED unit prices and item-level disaggregation would likely improve the interpretability of the detector [15,16]. The sustained increase in alert rates from 2021 to 2025 supports evaluation of this approach for routine audit prioritization workflows, not as a standalone determination of wrongdoing. 4.3 Smart contracts applied to pharmaceutical payment workflows: interpretation of a complementary proof of concept The proof of concept supports only a narrow technical claim: conditional digital state transitions can be encoded and executed in a prototype payment workflow under testnet conditions. It does not demonstrate real-world reduction of payment times, lower transaction costs, or institutional deployability. The reported times are not measurements of the full payment process, which also depends on transport, packaging, storage, institutional validation, treasury procedures, contracting rules, and other bottlenecks not modeled here. A defensible operational evaluation would require prospective integration with public hospital information systems, real transaction traces, explicit counterfactual design, and independent technical and regulatory review. Large-scale implementation would additionally require public-treasury integration, legal validation, cybersecurity controls, and external smart-contract auditing. 4.4 Contribution to health data governance Colombia has public data sources that can support pharmaceutical surveillance, but it lacks analytical infrastructure that reliably converts those data into actionable evidence. BigLoI illustrates that such infrastructure can be built using open technologies, reproducible workflows, and auditable queries over a contractual corpus. 4.5 Limitations SECOP-II data quality. Buyer and supplier variables include null, generic, or inconsistent values in a meaningful share of contracts in local sample files, which constrains some network analyses. This reflects source-data quality issues in SECOP-II. The source closed cohort included 162,271 candidate contracts; contract-level review excluded 441 records with explicit veterinary, animal-health, or agricultural scope, yielding a corrected cohort of 161,830. The recovered local PostgreSQL source currently indexes 165,549 candidate pharmaceutical records including live 2026 monitoring rows. SISMED temporal mismatch and granularity. The openly available SISMED source covers 2017–2019, whereas the main PostgreSQL procurement cohort analyzed here covers 2020–2025. Accordingly, SISMED was not used as a contemporaneous contract-by-contract comparator but as contextual market reference. In addition, the SECOP-II source does not provide a fully structured breakdown of line items and unit prices for all contracts, limiting direct validation of unit-price overpricing. Blockchain test-network scope. As detailed in Section 4.3, the payment-workflow validation on Sepolia was a functional test of digital state transitions only: prototype times are referential, the savings scenario is not an economic evaluation, and the prototype was not tested against live institutional payment data. Mainnet implementation would require independent smart-contract auditing, regulatory validation, public-treasury integration, and cybersecurity review. Selection bias in pharmaceutical filtering. Use of 19 inclusion keywords may generate false negatives and false positives. Contract-level review of 519 veterinary/agricultural candidates (441 excluded; 27 explicit human-health retained; 51 ambiguous retained) reduced a known contamination pathway, but the full keyword taxonomy has not been validated against an independently adjudicated reference set. A dedicated taxonomy-validation study is needed to quantify classification performance. Z-score false positives and false negatives. As discussed in Section 4.2, statistical alerts prioritize atypical total-contract-value patterns within named categories and can yield both false positives and false negatives; they do not constitute evidence of corruption, fraud, or unit-price overpricing. Interpretive scope. The observational design identifies statistical patterns but does not establish causal relationships or measure clinical impact. Interpreting flagged contracts as evidence of corruption or fraud would require additional adjudication methods and involvement of competent institutions. 4.6 Potential implications for public policy These findings may inform three policy discussions—health-system reform, procurement transparency [17], and health-data governance—but they do not by themselves prescribe specific reforms. The principal contribution of the study is methodological and reproducible: an open infrastructure capable of transforming dispersed public data into auditable evidence. Concentration and recurrence indicators may support audit prioritization, and uneven SECOP-II data quality is a potential institutional concern for Colombia Compra Eficiente and related oversight bodies. 5 Conclusions This study describes BigLoI, a national-scale computational pharmaceutical-surveillance platform that monitors SECOP-II contracts from 2015 onward and empirically analyzes a corrected closed cohort of 161,830 pharmaceutical contracts from 2020 to 2025. Three main findings emerge. Market concentration. The top 3% of suppliers concentrated 85.83% of total contracted value, the top 10 suppliers concentrated 29.01%, and the HHI was 120.99 on the 0–10,000 scale. Market concentration. The top 3% of suppliers concentrated 85.83% of total contracted value, the top 10 suppliers concentrated 29.01%, and the HHI was 120.99 on the 0–10,000 scale. Atypical statistical patterns. A total of 664 contracts triggered Z-score alerts with absolute Z-score greater than or equal to 1.5 sigma, antibiotics showed Zmax of 8.89, and the alert rate among Z-score-eligible contracts increased from 0.30% in 2021 to 1.31% in 2025. These alerts flag atypical total-value patterns for prioritization and do not constitute evidence of corruption or fraud; the 2025 rate applies only to named therapeutic categories after exclusion of NO_ESPECIFICADO. Atypical statistical patterns. A total of 664 contracts triggered Z-score alerts with absolute Z-score greater than or equal to 1.5 sigma, antibiotics showed Zmax of 8.89, and the alert rate among Z-score-eligible contracts increased from 0.30% in 2021 to 1.31% in 2025. These alerts flag atypical total-value patterns for prioritization and do not constitute evidence of corruption or fraud; the 2025 rate applies only to named therapeutic categories after exclusion of NO_ESPECIFICADO. Reproducible infrastructure. Integration of relational data storage, reproducible analytics, and a public observatory makes it possible to convert dispersed public sources into an auditable surveillance instrument for pharmaceutical procurement. Reproducible infrastructure. Integration of relational data storage, reproducible analytics, and a public observatory makes it possible to convert dispersed public sources into an auditable surveillance instrument for pharmaceutical procurement. As a complementary result, the Sepolia prototype shows that conditional payment logic can be represented and executed in a controlled testnet environment under simulated conditions. Any inference about reduced payment delays or financial savings remains hypothetical and dependent on assumptions that were not empirically validated in this study. This line of work therefore remains exploratory and still requires operational, regulatory, technical, and logistical validation. Overall, BigLoI shows that an open and reproducible surveillance infrastructure can generate descriptive findings on market concentration, statistical anomalies, and auditable evidence access from public administrative data on Colombian pharmaceutical procurement. Supporting information Approximate COP/USD equivalents for the principal monetary values reported in the manuscript. Conversions use a single round reference exchange rate of COP 4,000 per USD and are provided only as approximate interpretive aids; no analyses were performed in USD. (DOCX) Source data for Table 1. (CSV) Source data for Table 2. (CSV) Source data for Fig 1. (CSV) Source data for Fig 2. (CSV) Source data for Fig 3 therapeutic-category panel. (CSV) Source data for Fig 3 yearly panel. (CSV) Source data for Fig 4 Lorenz curve. (CSV) Source data for Fig 4 top-10 supplier shares. (CSV) Column dictionary for the source-data files. (CSV) Source-file manifest for the uploaded supporting data files. (CSV) SECOP-II filtering taxonomy, simplified database schema, and smart-contract protocol. (DOCX) 2 Jun 2026 Dear Dr. Soto, Please submit your revised manuscript by Jul 17 2026 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file. A letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols. As the corresponding author, your ORCID iD is verified in the submission system and will appear in the published article. PLOS supports the use of ORCID, and we encourage all coauthors to register for an ORCID iD and use it as well. Please encourage your coauthors to verify their ORCID iD within the submission system before final acceptance, as unverified ORCID iDs will not appear in the published article. Only the individual author can complete the verification step; PLOS staff cannot verify ORCID iDs on behalf of authors. We look forward to receiving your revised manuscript. Kind regards, Sarah Jose, Ph.D. Staff Editor PLOS One Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf 2. Please include your full ethics statement in the ‘Methods’ section of your manuscript file. In your statement, please include the full name of the IRB or ethics committee who approved or waived your study, as well as whether or not you obtained informed written or verbal consent. If consent was waived for your study, please include this information in your statement as well. 3. Please amend either the title on the online submission form (via Edit Submission) or the title in the manuscript so that they are identical. 4. Please amend either the abstract on the online submission form (via Edit Submission) or the abstract in the manuscript so that they are identical. 5. Please ensure that you refer to Figure 1, 2, 3, 5, 6, in your text as, if accepted, production will need this reference to link the reader to the figure. 6. We note you have included a table to which you do not refer in the text of your manuscript. Please ensure that you refer to Table 1 in your text; if accepted, production will need this reference to link the reader to the Table. 7. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: http://journals.plos.org/plosone/s/supporting-information. 8. If the reviewer comments include a recommendation to cite specific previously published works, please review and evaluate these publications to determine whether they are relevant and should be cited. There is no requirement to cite these works unless the editor has indicated otherwise. [Note: HTML markup is below. Please do not edit.] Reviewers' comments: Reviewer's Responses to Questions Comments to the Author 1. Is the manuscript technically sound, and do the data support the conclusions? Reviewer #1: Partly Reviewer #2: Partly ********** 2. Has the statistical analysis been performed appropriately and rigorously? -->?> Reviewer #1: Yes Reviewer #2: Yes ********** 3. Have the authors made all data underlying the findings in their manuscript fully available??> The PLOS Data policy Reviewer #1: No Reviewer #2: Yes ********** 4. Is the manuscript presented in an intelligible fashion and written in standard English??> Reviewer #1: Yes Reviewer #2: Yes ********** Reviewer #1: - The abstract in Editorial Manager refers to a “primary analytical cohort from 2020 to March 2026,” whereas the main text of the manuscript consistently uses the term “closed cohort for 2020–2025,” which clearly excludes incomplete records for 2026. - The choice of an alert threshold at |Z| ≥ 1.5σ is not justified by either a reference to the literature or a sensitivity-specificity analysis. - The data in Table 1 (Section 3.1) show a sharp decline in the number of contracts in 2023 and 2024, followed by a partial recovery in 2025. The author cites the peak of the COVID-19 pandemic in 2021 and an increase in oncology procurement in 2025, but the nearly fivefold drop in 2023–2024 remains unexplained. - The analysis of market concentration is limited to descriptive indicators (top 3%, top 10 suppliers…). To align with standards in economic and competition literature, it is recommended to supplement the analysis with, for example, the Herfindahl–Hirschman Index (HHI). - The only publicly available SISMED dataset covers the years 2017–2019, whereas the analytical cohort spans 2020–2025. The authors note that SISMED was used merely as a “contextual market benchmark,” but this limitation should also be included in the “Methods” section. - List of keywords includes the term “artificial intelligence,” but the manuscript primarily employs statistical methods, and the use of AI components (Claude 3.5 Sonnet, GPT-4o) serves as a supporting tool for the platform rather than the primary analytical method. It is recommended to replace “artificial intelligence” with a more precise term. - All illustrations must be replaced with vector or high-quality raster files that include legible captions and legends. The resolution must meet PLOS One’s requirements. - The author uses a single exchange rate of 4,000 USD for the entire analysis period—an explicit warning about exchange rate volatility must be added in the relevant methodological note (not only in Supplementary Table S1). - The article is submitted by a single author who is an independent researcher. Given the scale and complexity of the work performed, it is recommended to either indicate external peer review of the methodology (e.g., by academic colleagues) or emphasize in the text how reproducibility and code openness compensate for the lack of co-author verification. Reviewer #2: The manuscript provides a relevant analysis of Colombian public pharmaceutical procurement using open administrative data within a reproducible framework. It addresses key issues in public health governance and procurement transparency. The study is well-organised and demonstrates significant technical effort in integrating SECOP-II, INVIMA, and SISMED data. Strengths include a large-scale longitudinal dataset, a focus on reproducibility, public data integration, and an honest acknowledgement of limitations. The conclusions regarding market concentration and unusual procurement patterns are well-supported by the data. The authors wisely avoid equating statistical anomalies with corruption, maintaining the exploratory nature of their analyses. However, there are areas for improvement: 1. Statistical Methodology: The reliance on Z-score analysis for anomaly detection needs more methodological justification for the chosen thresholds and a discussion on sensitivity and specificity. Future work could benefit from advanced detection methods. 2. Interpretation of Anomalies: The manuscript should avoid overinterpreting findings and remain observational without causal implications. 3. Consistency and Reproducibility: Minor discrepancies in reported contract counts need clarification for better reproducibility. 4. Data Limitations: The lack of structured line-item pricing data limits the assessment of unit-price inflation and should be emphasised more clearly. 5. Blockchain Proof-of-Concept: While technically interesting, this section is peripheral to the main objectives and could be shortened. 6. Language and Readability: The manuscript is generally clear, but some repetitive sections could benefit from editing for conciseness. There are no obvious ethical concerns regarding plagiarism or research ethics, and the study transparently uses publicly available datasets. ********** what does this mean?). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy Reviewer #1: No Reviewer #2: No ********** [NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.] To ensure your figures meet our technical requirements, please review our figure guidelines: https://journals.plos.org/plosone/s/figures You may also use PLOS’s free figure tool, NAAS, to help you prepare publication quality figures: https://journals.plos.org/plosone/s/figures#loc-tools-for-figure-preparation. NAAS will assess whether your figures meet our technical requirements by comparing each figure against our figure specifications. 11 Jun 2026 Response to Reviewers Manuscript ID: PONE-D-26-13579 Title: Computational surveillance of Colombian public pharmaceutical procurement using public administrative data: a reproducible analysis of a closed 2020-2025 cohort Journal: PLOS ONE Corresponding author: Andrés Soto (Independent researcher, Bogotá, Colombia; ORCID 0009-0004-8001-5372) ------------------------------------------------------------------------ Dear Dr. Jose and Editors of PLOS ONE, Thank you for the opportunity to submit a revised version of our manuscript. I am grateful for the careful editorial assessment and for the constructive comments from both reviewers, which have materially strengthened the manuscript. I have revised the manuscript to address every journal requirement and reviewer comment. The revision (i) defines a single closed analytical cohort (complete calendar years 2020–2025) and reconciles all reported counts across five explicit data scopes; (ii) reframes the Z-score engine as an exploratory audit-prioritization instrument and reports threshold-stratified alert counts; (iii) acknowledges that sensitivity and specificity cannot be estimated without an adjudicated audit reference; (iv) adds the Herfindahl–Hirschman Index to the market-concentration analysis; (v) strengthens the treatment of the SISMED temporal mismatch and the absence of structured line-item pricing; (vi) removes causal/accusatory language; (vii) shortens and reframes the blockchain component as a secondary technical proof of concept; (viii) clarifies the currency convention and exchange-rate caveat; (ix) expands the ethics statement; and (x) adds in-text references to all figures and to Table 1 plus a Supporting Information section. A point-by-point response follows. Reviewer comments are reproduced in summary form, followed by my response and the change made. Sincerely, Andrés Soto ------------------------------------------------------------------------ Response to Journal Requirements JR-1 — Style and file naming. The revised submission package follows PLOS ONE style and file-naming conventions: Manuscript.docx, Revised_Manuscript_with_Track_Changes.docx, Response_to_Reviewers.docx, Fig1.tif–Fig6.tif, and S1_Table. JR-2 — Ethics statement in Methods. A full ethics statement has been added to Methods (§2.1): “This study used only publicly accessible administrative procurement records published by Colombian open-data sources (SECOP-II via datos.gov.co, INVIMA, and SISMED). It did not involve human participants, patient-level or clinical data, identifiable private information, biological samples, recruitment, intervention, or any contact with individuals. Accordingly, the study does not constitute human-subjects research and did not require review or approval by an institutional review board or research ethics committee. Under Colombian Resolution 8430 of 1993, analysis of public, non-identifiable administrative records corresponds to research without risk and falls outside the scope of human-subjects ethics review. Informed consent was not applicable because the study did not involve human participants or identifiable personal data.” The same statement was entered in the Editorial Manager Ethics Statement field. JR-3 — Title consistency. The manuscript title and the Editorial Manager title now match exactly: Computational surveillance of Colombian public pharmaceutical procurement using public administrative data: a reproducible analysis of a closed 2020-2025 cohort. JR-4 — Abstract consistency. The Editorial Manager abstract was updated to match the manuscript abstract exactly. Both describe the formal analytical cohort as the closed 2020–2025 cohort; post-2025 records are described only as live monitoring records and are not used in year-by-year comparisons. JR-5 — In-text figure references. The revised text now refers explicitly to Figure 1 (§3.1), Figure 2 (§3.2), Figure 3 (§3.3), Figure 5 (§3.5), and Figure 6 (§2.3), in addition to the figure legends. JR-6 — In-text Table 1 reference. Results §3.1 now states that the corpus characteristics and the closed 2020–2025 cohort are summarized in Table 1. JR-7 — Supporting Information captions. A Supporting Information section was added at the end of the manuscript with the caption for S1 Table (approximate COP/USD equivalents), and the in-text citation was updated to the S1 Table format. JR-8 — Evaluation of suggested citations. The reviewers did not provide a mandatory citation list. I reviewed the relevant methodological areas (market concentration/HHI, procurement-risk indicators, pharmaceutical market monitoring, and anomaly detection) and refined citations where they strengthened the Methods and Discussion. ------------------------------------------------------------------------ Response to Reviewer #1 R1-1 — Cohort inconsistency (2020–March 2026 vs closed 2020–2025). I agree. The reference to “2020 to March 2026” in the online abstract was a submission-metadata inconsistency, not a feature of the analysis. The formal analytical cohort comprises complete calendar years 2020–2025 (162,271 pharmaceutical contracts); post-2025 records are retained only for live platform monitoring. Change: the manuscript and the Editorial Manager metadata now state the closed 2020–2025 cohort consistently, and a five-scope reconciliation (see R2-3) removes residual ambiguity. R1-2 — Justification of the |Z| ≥ 1.5σ threshold; sensitivity/specificity. I agree. The revised Methods (§2.4) describe |Z| ≥ 1.5 as an exploratory surveillance threshold for prioritizing review in a high-volume administrative dataset, not a diagnostic threshold for wrongdoing; |Z| ≥ 2.0 and |Z| ≥ 3.0 are retained as higher-severity strata. Because no adjudicated gold-standard audit dataset exists for this corpus, sensitivity and specificity cannot be estimated directly; this is now stated explicitly. Change: Results §3.3 report the reconciled strata over 147,020 analyzed contracts — 685 alerts at |Z| ≥ 1.5, 528 at |Z| ≥ 2.0, and 303 at |Z| ≥ 3.0 — and the Limitations note the need for validation against adjudicated audit outcomes. R1-3 — 2023–2024 decline. I agree this required clearer treatment. The revised Results §3.1 now ground the decline in evidence rather than assigning a single causal explanation. The pronounced drop in indexed contracts in 2023–2024 (from 30,991 in 2022 to 10,705 and 9,380) is reproduced in an independent reconstruction of the public SECOP-II API, which recovers 10,006 and 8,818 pharmaceutical contracts for those years (gaps of only 699 and 562 contracts relative to the manuscript counts). The decline is therefore present in the SECOP-II source under the inclusion taxonomy and is not an artifact of the local snapshot. Because the available administrative data do not allow us to adjudicate whether it reflects procurement consolidation into fewer, higher-value contracts, changes in contract-object coding, or an actual reduction in contracting volume, we treat it as a source-data and classification limitation requiring dedicated validation. R1-4 — Add the Herfindahl–Hirschman Index. I agree. Change: Methods §2.5 specifies the HHI as the sum of squared supplier shares of total contracted value among positive-value contracts in the closed cohort, on the 0–10,000 scale; Results §3.4 report HHI = 119.26, alongside a top-10 share of 28.77% and a top-3% share of 85.87% (162,151 positive-value contracts; 50,355 suppliers). R1-5 — SISMED 2017–2019 limitation in Methods. I agree. Change: Methods §2.2 now states explicitly that SISMED covers 2017–2019 and was used only as a contextual market reference, not as a contemporaneous contract-by-contract benchmark for the 2020–2025 cohort; the Limitations retain and expand this point. R1-6 — Keyword “artificial intelligence.” I agree. Change: the keyword “artificial intelligence” was replaced with “retrieval-augmented generation,” which names the actual supporting AI component without overstating its analytical role. R1-7 — Figures must be high-quality with legible labels. I agree; figures were treated as a resubmission priority. Change: Figures 1–6 were regenerated as high-quality raster files (TIFF) with all internal text in English and legible legends and annotations; figure source data remain in the reproducibility package. R1-8 — Single exchange rate / volatility warning. I agree. Change: Methods §2.2 now states that COP is the analytical currency, that USD equivalents are approximate interpretive aids only, and that applying a single round reference rate (COP 4,000/USD) across 2020–2025 does not account for exchange-rate volatility; values are order-of-magnitude approximations. The conversion table is cited as S1 Table. R1-9 — Single independent author: external review or reproducibility. I agree that independent verification matters. While the manuscript is single-authored, I strengthened the reproducibility package so that the analytical workflow can be independently inspected and replicated: source-data tables, figure and table source data, scripts, a column dictionary, a file manifest, and a Zenodo DOI, with the code repository openly available. Change: the Data and Code Availability section and relevant Methods text now emphasize auditability, openness, and reproducibility as the mechanisms supporting independent verification, and the single-author limitation is acknowledged explicitly. ------------------------------------------------------------------------ Response to Reviewer #2 R2-1 — Z-score justification and sensitivity/specificity. I agree; this is addressed jointly with R1-2. The Z-score engine is presented as an exploratory prioritization instrument, not a diagnostic classifier; threshold strata are reported (685 / 528 / 303), and the absence of sensitivity/specificity estimation is stated as a limitation, with validation against adjudicated audit outcomes proposed as future work. Change: Methods §2.4, Discussion §4.2, and Limitations were revised accordingly. R2-2 — Avoid overinterpretation; remain observational. I agree. Change: Results, Discussion, Limitations, and Conclusions now use cautious observational language. Statistical alerts are described as atypical total-value patterns that may support audit prioritization, and the manuscript states explicitly that they cannot establish corruption, causal mechanisms, or unit-price overpricing without additional evidence. R2-3 — Discrepancies in contract counts. I agree. Change: the manuscript now distinguishes five nested data scopes: (i) the SECOP-II monitored universe — 272,814 contracts (2015 onward); (ii) the recovered local PostgreSQL snapshot — 956,157 total records, of which 165,549 are pharmaceutical (including live post-2025 monitoring rows); (iii) the closed analytical cohort used here — 162,271 contracts (2020–2025); (iv) the value-share concentration subset — 162,151 positive-value contracts (50,355 normalized suppliers); and (v) the Z-score–analyzable subset — 147,020 contracts in categories with ≥10 observations. Earlier provisional figures (e.g., 162,921 contracts spanning 2020–March 2026) reflected a cohort that included partial 2026 records; closing the cohort to complete calendar years 2020–2025 yields 162,271. Methods §2.2 and the Table 1 footnote now state these scopes, and the manuscript, figures, source data, and reproducibility package were audited so that every count derives from a defined scope. R2-4 — Lack of structured line-item pricing. I agree. Change: Methods §2.4 and the Limitations now state explicitly that the Z-score engine evaluates total contract values within therapeutic categories and cannot determine unit-price inflation or overpricing without structured item-level quantities, unit prices, dosage forms, and comparable specifications. R2-5 — Blockchain proof of concept is peripheral. I agree. Change: the blockchain section (Methods §2.6, Results §3.5, Discussion §4.3) was shortened and reframed as a secondary technical-feasibility module tested on the Sepolia testnet only; the text states that it does not provide evidence of real-world payment-cycle reduction, realized savings, or implementation readiness. R2-6 — Conciseness. I agree. Change: repetitive passages were condensed, particularly in the blockchain discussion, the interpretive-scope statements, the exchange-rate caveats, and the Limitations, while preserving the methodological clarifications requested by both reviewers. ------------------------------------------------------------------------ Summary of Major Revisions 1. Single closed 2020–2025 cohort (162,271 contracts) with a five-scope count reconciliation. 2. Z-score reframed as exploratory audit prioritization; threshold strata 685 / 528 / 303 over 147,020 analyzed contracts. 3. Explicit acknowledgment that sensitivity/specificity require an adjudicated reference set. 4. HHI added (119.26 on the 0–10,000 scale). 5. SISMED 2017–2019 mismatch and absence of line-item pricing emphasized in Methods and Limitations. 6. Causal/accusatory language removed throughout. 7. Blockchain shortened and framed as a secondary proof of concept. 8. Currency convention clarified with an exchange-rate volatility caveat. 9. Ethics statement expanded in Methods. 10. Figures 1–6 regenerated as high-quality English-language files. 11. Supporting Information section and in-text citations added. 12. Reproducibility package, Zenodo DOI, and open code emphasized for independent verification. Submitted filename: Response_to_Reviewers.docx 17 Jul 2026 Dear Dr. Soto, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Your revision have been assessed by the two previous reviewers and they have addressed several concerns. Please read carefully and address them one by one in your next revision. Please submit your revised manuscript by Aug 31 2026 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file. A letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols. As the corresponding author, your ORCID iD is verified in the submission system and will appear in the published article. PLOS supports the use of ORCID, and we encourage all coauthors to register for an ORCID iD and use it as well. Please encourage your coauthors to verify their ORCID iD within the submission system before final acceptance, as unverified ORCID iDs will not appear in the published article. Only the individual author can complete the verification step; PLOS staff cannot verify ORCID iDs on behalf of authors. We look forward to receiving your revised manuscript. Kind regards, Xin Sun, PhD Staff Editor PLOS One Journal Requirements: If the reviewer comments include a recommendation to cite specific previously published works, please review and evaluate these publications to determine whether they are relevant and should be cited. There is no requirement to cite these works unless the editor has indicated otherwise. [Note: HTML markup is below. Please do not edit.] Reviewers' comments: Reviewer's Responses to Questions Comments to the Author Reviewer #1: (No Response) Reviewer #2: (No Response) ********** 2. Is the manuscript technically sound, and do the data support the conclusions??> Reviewer #1: Partly Reviewer #2: Yes ********** 3. Has the statistical analysis been performed appropriately and rigorously? -->?> Reviewer #1: Yes Reviewer #2: Yes ********** 4. Have the authors made all data underlying the findings in their manuscript fully available??> The PLOS Data policy Reviewer #1: Yes Reviewer #2: Yes ********** 5. Is the manuscript presented in an intelligible fashion and written in standard English??> Reviewer #1: Yes Reviewer #2: Yes ********** Reviewer #1: The main revisions have been made to the text, but a few minor clarifications remain: 1. In Section 2.6, regarding the financial rate of 2% per month - at least one reference to a source justifying this figure in the context of the Colombian public procurement market must be added. 2. In the text (Sections 3.4 and 5), the concentration index is listed as 85.87%, while the legend for Fig. 4 states: “Top 3% = 88.9%” 3. Table 1 states: “Unique pharmaceutical suppliers in the closed cohort: 50,460”, section 3.4 states: “showed 50,460 unique supplier names. For value-share concentration metrics, 162,151 positive-value contracts and 50,355 suppliers with non-empty normalized labels were included.” In other words, the same paragraph mentions both 50,460 and 50,355 suppliers without explaining the transition between these numbers. 4. Please explain why, out of the 22,033 indexed contracts for 2025 (Section 3.3), only 6,792 were included in the Z-score engine. Is this due to limitations in the data available at the time of upload, the specifics of categorical filtering, or some other reason? 5. It is recommended to add 1–2 sentences to Section 4.1 discussing the limitation of the calculated HHI, since it is defined for the entire national market, while high concentration is also possible in specific groups (oncology, antibiotics, antivirals), but this is not specified by the author. Reviewer #2: This manuscript presents a reproducible computational framework for monitoring public pharmaceutical procurement in Colombia using openly available administrative data. The topic is relevant and timely, and the study benefits from a large dataset, transparent methodology, and comprehensive data and code availability. The manuscript is generally well organized and clearly written. The analyses are adequately described, and the conclusions are mostly supported by the reported findings. The authors should be commended for their efforts to ensure transparency and reproducibility. I have a few suggestions that could further strengthen the manuscript: 1. The anomaly-detection approach is based primarily on Z-score thresholds. While useful for identifying unusual contract values, the limitations of this method, including the potential for false-positive and false-negative findings, should be discussed in greater detail. 2. The manuscript appropriately notes that statistical alerts do not constitute evidence of corruption or fraud. This distinction should remain clearly emphasized throughout the discussion and conclusions. 3. The smart-contract component is an interesting addition; however, it remains a proof of concept tested under simulated conditions. Its practical implications should continue to be presented cautiously, avoiding any suggestion of demonstrated real-world effectiveness. 4. Additional information regarding the validation of the keyword-based pharmaceutical filtering strategy would strengthen confidence in the construction of the analytical cohort. 5. Several policy-related interpretations are interesting but somewhat speculative. These statements should be framed as potential implications rather than direct conclusions derived from the data. ********** what does this mean?). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy Reviewer #1: No Reviewer #2: No ********** [NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.] To ensure your figures meet our technical requirements, please review our figure guidelines: https://journals.plos.org/plosone/s/figures You may also use PLOS’s free figure tool, NAAS, to help you prepare publication quality figures: https://journals.plos.org/plosone/s/figures#loc-tools-for-figure-preparation. NAAS will assess whether your figures meet our technical requirements by comparing each figure against our figure specifications. 21 Jul 2026 Dear Dr. Sun and Editors of PLOS ONE, Thank you for the opportunity to revise manuscript PONE-D-26-13579R1. I have addressed every remaining clarification from Reviewers #1 and #2. The same point-by-point response is also uploaded as the separate file “Response to Reviewers.” I verified the R1 package on file in Editorial Manager (PONE-D-26-13579_R1). That version matches the intermediate counts cited by the reviewers (for example 50,460 / 50,355 suppliers, 85.87% top-3% concentration, 22,033 / 6,792 contracts for 2025, HHI 119.26, and Z-score denominator 147,020). The present revision retains every clarification below and additionally incorporates the final author-approved contract-level veterinary correction (dated 2026-07-11; previously disclosed to the editor): 161,830 contracts; 50,225 normalized suppliers; top-3% concentration 85.83%; HHI 120.99; and for 2025, 21,955 indexed contracts of which 6,729 enter the Z-score engine. All tables, figures, and reproducibility artifacts are aligned to these corrected figures. RESPONSE TO REVIEWER #1 R1-1 — Source justifying the 2% per month financial rate (§2.6). I agree. Change: Methods §2.6 now frames the 2% monthly rate as a conservative upper-bound scenario parameter (approximately 26.8% effective annual), not as an empirically measured average sector rate. The quantitative anchor is the upper range of Colombian legal interest ceilings derived from the interés bancario corriente certified by the Superintendencia Financiera de Colombia, where maximum remuneratory or default interest equals 1.5 times that certified rate (usury ceiling) (refs 17–18). Commercial or rediscount lines can be lower. As contextual motivation for treating payment delay as financially costly—without claiming that AFIDRO measures a 2% monthly rate—we cite the AFIDRO/Sectorial receivables report (ref 19). Results §3.5 use the same “upper-bound scenario parameter” wording. R1-2 — Concentration 85.87% in text versus a conflicting Fig. 4 annotation (“Top 3% = 88.9%”). I agree that text and Figure 4 must report the same concentration value. In the R1 Editorial Manager package, §3.4 reported 85.87% while the Abstract/figure annotation used a rounded 85.9%; the reviewer correctly flagged that the figure presentation was not acceptably synchronized with the text. Change: after the contract-level veterinary correction, text (§3.4, §5), source data, and Figure 4 are fully reconciled: the top 3% of suppliers concentrate 85.83% of contracted value, and the figure annotation reads “Top 3% = 85.8%” (50,225 active suppliers; HHI 120.99). R1-3 — 50,460 versus 50,355 suppliers without explaining the transition. I agree that supplier scopes must be explicit. Change: Table 1 and §3.4 now use a single concentration denominator—50,225 unique non-empty normalized supplier labels among 161,710 positive-value contracts—and §3.4 states that zero-value or blank-label rows are outside that denominator. The intermediate counts cited by the reviewer correspond to the pre-correction revision and are superseded by the corrected figures above. R1-4 — Why only 6,792 of 22,033 indexed 2025 contracts entered the Z-score engine. I agree this required an explicit funnel explanation. The cause is categorical filtering, not incomplete data at upload: the contracts are present in the closed cohort but are ineligible for within-category Z-scores. Change: Methods §2.4 now states that contracts labeled NO_ESPECIFICADO are excluded because they lack a stable within-category comparator, and that named categories with fewer than 10 contracts are also excluded. Results §3.1 list NO_ESPECIFICADO in the therapeutic-category distribution (15,226 contracts; COP 2.63 trillion; 15.5% of corrected-cohort value). Results §3.3 report the corrected 2025 funnel: 21,955 indexed → 6,729 analyzed (30.7%), because all 15,226 NO_ESPECIFICADO contracts fall in 2025 (21,955 − 6,729 = 15,226). Those unspecified contracts account for about 48% of 2025 contracted value, so the 1.31% alert rate describes only the named-category slice and must not be read as an anomaly rate for all 2025 pharmaceutical procurement. Coverage for 2020–2024 remains near-complete under the same rules (aggregate gap of 10 contracts in categories with n<10). R1-5 — Limitation of national HHI (§4.1). I agree. Change: Section 4.1 now notes that the national HHI may understate concentration within specific therapeutic groups (for example oncology, antibiotics, or antivirals) and that category-specific HHI values were not estimated. It further clarifies that, under conventional thresholds, HHI = 120.99 would read as unconcentrated at the national aggregate level, yet that interpretation can coexist with the observed top-3% cumulative share of 85.83% when a large long-tail supplier base dilutes the index; subgroup HHI estimation is left for future work. RESPONSE TO REVIEWER #2 R2-1 — Discuss Z-score limitations, including false positives and false negatives, in greater detail. I agree. Change: Section 4.2 and the Limitations now expand this point: without an adjudicated audit gold standard, sensitivity and specificity cannot be estimated; the exploratory |Z| ≥ 1.5 threshold favors prioritization recall and can increase false positives; false negatives remain possible under category heterogeneity, NO_ESPECIFICADO exclusion, or small category sizes. R2-2 — Keep clear that statistical alerts are not evidence of corruption or fraud. I agree. Change: this distinction is reinforced in Results §3.3, Discussion/Limitations, Abstract Conclusions, and Conclusions §5, which state explicitly that alerts are exploratory prioritization signals and do not constitute evidence of corruption or fraud. R2-3 — Present the smart-contract component cautiously as a simulated proof of concept. I agree. Change: the manuscript continues to frame the Sepolia module as a secondary technical-feasibility proof of concept under simulated conditions only. Abstract Results no longer foreground illustrative COP savings figures; they state only that predefined digital state transitions were executed under testnet conditions and that the module provides no evidence of real-world payment-cycle reduction, realized savings, or institutional deployability. Abstract Conclusions and §5 retain the same caution. Detailed scenario arithmetic, where retained, remains confined to Methods §2.6 / Results §3.5 as an explicitly non-empirical illustration. R2-4 — Additional information on validation of the keyword-based pharmaceutical filter. I agree. Change: Methods §2.2 now describes the 19-keyword screen as high-recall rather than gold-standard; explains false-positive/false-negative pathways; and reports the contract-level quality-control review of 519 candidates (441 excluded; 27 human-health retained; 51 ambiguous retained), while stating that no independently adjudicated external reference set was available for full taxonomy sensitivity/specificity. The Limitations retain the call for dedicated taxonomy validation. R2-5 — Frame policy-related statements as potential implications rather than direct conclusions. I agree. Change: Section 4.6 is retitled “Potential implications for public policy” and uses cautious language (“may inform,” “potential institutional concern”). Conclusions and the Abstract Conclusions retain descriptive findings and avoid prescribing reforms. SUMMARY OF MAJOR REVISIONS (R2) 1. Justified the 2%/month financing-cost assumption as an upper-bound scenario parameter anchored to Colombian Superintendencia Financiera / usury ceilings (refs 17–18), with AFIDRO receivables context (ref 19). 2. Reconciled concentration reporting and Figure 4 at 85.83% / 85.8% with 50,225 suppliers. 3. Added an explicit supplier-scope bridge for the concentration denominator (§3.4; Table 1). 4. Documented the 2025 Z-score funnel (NO_ESPECIFICADO exclusion; 21,955 → 6,729), including COP 2.63 trillion / ~48% of 2025 value and a cautious reading of the 1.31% alert rate. 5. Strengthened the national-HHI limitation regarding therapeutic subgroups and the coexistence of low aggregate HHI with high top-3% cumulative share (§4.1). 6. Expanded Z-score false-positive/false-negative discussion and reinforced that alerts ≠ corruption/fraud. 7. Strengthened keyword-filter validation disclosure (§2.2) and reframed policy text as potential implications. Sincerely, Andrés Soto Corresponding author ORCID: 0009-0004-8001-5372 Submitted filename: Response_to_Reviewers_auresp_2.docx 20 Aug 2026 Computational surveillance of Colombian public pharmaceutical procurement using public administrative data: a reproducible analysis of a closed 2020-2025 cohort PONE-D-26-13579R2 Dear Dr. Soto, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice will be generated when your article is formally accepted. Please note, if your institution has a publishing partnership with PLOS and your article meets the relevant criteria, all or part of your publication costs will be covered. Please make sure your user information is up-to-date by logging into Editorial Manager at Editorial Manager® and clicking the ‘Update My Information' link at the top of the page. For questions related to billing, please contact billing support. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org. Kind regards, Patrick Goymer Staff Editor PLOS One Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions Comments to the Author Reviewer #1: (No Response) Reviewer #2: All comments have been addressed ********** 2. Is the manuscript technically sound, and do the data support the conclusions??> Reviewer #1: Yes Reviewer #2: Yes ********** 3. Has the statistical analysis been performed appropriately and rigorously? -->?> Reviewer #1: Yes Reviewer #2: Yes ********** 4. Have the authors made all data underlying the findings in their manuscript fully available??> The PLOS Data policy Reviewer #1: Yes Reviewer #2: Yes ********** 5. Is the manuscript presented in an intelligible fashion and written in standard English??> Reviewer #1: Yes Reviewer #2: Yes ********** Reviewer #1: This version of the manuscript has several strengths, including the scope of the data, the transparency of the methodology, and the honest acknowledgment of the study’s limitations. The article represents a substantial contribution to the field of digital infrastructure for government pharmaceutical procurement and may be accepted for publication. Reviewer #2: I appreciate your careful and thorough revision of the manuscript. The revised version addresses the methodological, statistical, and interpretative concerns raised during the previous rounds of review. In particular, the clarification of the analytical cohort, reconciliation of previously inconsistent counts, expanded discussion of the limitations of the Z-score approach, cautious interpretation of statistical alerts, and improved description of the blockchain component as a technical proof of concept have substantially strengthened the manuscript. I also appreciate the enhanced transparency achieved through the public availability of the reproducibility package, analytical datasets, source code, and documentation. The conclusions are now appropriately aligned with the presented evidence, and the limitations are discussed in a balanced manner. Overall, I believe the manuscript is technically sound, clearly written, and suitable for publication in PLOS ONE. ********** what does this mean?). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy Reviewer #1: No Reviewer #2: No ********** PONE-D-26-13579R2 PLOS One Dear Dr. Soto, I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS One. Congratulations! Your manuscript is now being handed over to our production team. At this stage, our production department will prepare your paper for publication. This includes ensuring the following: * All references, tables, and figures are properly cited * All relevant supporting information is included in the manuscript submission, * There are no issues that prevent the paper from being properly typeset You will receive further instructions from the production team, including instructions on how to review your proof when it is ready. Please keep in mind that we are working through a large volume of accepted articles, so please give us a few days to review your paper and let you know the next and final steps. Lastly, if your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org. You will receive an invoice from PLOS for your publication fee after your manuscript has reached the completed accept phase. If you receive an email requesting payment before acceptance or for any other service, this may be a phishing scheme. Learn how to identify phishing emails and protect your accounts at https://explore.plos.org/phishing. If we can help with anything else, please email us at customercare@plos.org. Thank you for submitting your work to PLOS One and supporting open access. Kind regards, PLOS One Editorial Office Staff on behalf of Dr Patrick Goymer Staff Editor PLOS One