Data Resource Profile: Prostate cancer data from Clinical Practice Research Datalink linked hospital records, mortality data and cancer registry standardized to the Observational Medical Outcomes Partnership common data model (CPRD-PCa-OMOP)

CPRD-PCa-OMOP comprises men from England with incident prostate cancer (PCa) selected from the Clinical Practice Research Datalink (CPRD) GOLD and Aurum primary care databases, linked to hospital admissions, Office for National Statistics mortality data, and national cancer registry records. Each dataset was standardized to the Observational Medical Outcomes Partnership Common Data Model to facilitate data integration, collaborative research, external validation, and network studies. All the standardized datasets were merged to create CPRD-PCa-OMOP. Men aged ≥18 years, diagnosed with their first PCa in primary care between 2010 and 2022, and confirmed in the cancer registry, with ≥1 year of general practitioner registration, were included. There are 69 109 patients from 1611 English practices: 5566 from CPRD GOLD and 63 543 from CPRD Aurum. CPRD-PCa-OMOP includes information on demographics, diagnoses, medications, procedures, referrals, test results, hospitalizations, episodes of care, cancer diagnoses, tumour characteristics such as Tumour, Node, Metastasis staging and Gleason score, date of death, and primary cause of death. This comprehensive data resource will enable the study of risk factors, treatment patterns, and longitudinal outcomes in PCa. CPRD data access can be requested at https://www.cprd.com/access-data. The digital collection of UK patient care data was started in the late 1980s to manage general practitioner (GP) practices and patients, and was expanded over the years to include hospital events and registry records, generating large data sources that are now used for observational research [1]. Observational data for cancer patients are complex, distributed across different care settings. This requires the integration of various pieces of information, such as tumour histology, biomarkers, and treatment regimens, as well as longitudinal comorbidities and comedications. In non-cancer observational research, cohorts can be defined by the absence or presence of a health condition. The additional challenges of conducting observational research in cancer arise from diagnostic attributes (cancer staging, histology, grade, and biomarkers) that can determine the treatment options and prognoses [2]. One of the UK data providers that allow cancer data integration through a linked data service is the Clinical Practice Research Datalink (CPRD). CPRD is a not-for-profit, cost-recovery UK government research service delivered by the Medicines and Healthcare products Regulatory Agency (MHRA) with support from the National Institute for Health and Care Research, as part of the Department of Health and Social Care [3]. CPRD hosts two primary care observational databases: CPRD GOLD [4] and CPRD Aurum [5], hereinafter called GOLD and Aurum, respectively. GOLD contains records collected by using the Vision® software from historical contributing practices in England, and from historical and current contributing practices in Wales, Scotland, and Northern Ireland. Aurum contains records collected by using the EMIS Web® software from historical and current contributing practices in England only. CPRD can provide a variety of linked patient-level data from England for both GOLD and Aurum, including the Hospital Episode Statistics (HES) Admitted Patient Care (APC), Office for National Statistics (ONS) Mortality, and National Cancer Registration and Analysis Service (NCRAS) datasets [6]. These datasets have been utilized individually or as part of a linkage in numerous peer-reviewed publications [7]. CPRD-PCa-OMOP is a new data source created to characterize and assess long-term outcomes in patients with prostate cancer (PCa) in the context of the public–private partnership OPTIMA (Optimal Treatment for Patients with Solid Tumours in Europe Through Artificial Intelligence). The OPTIMA consortium aims to achieve an interoperable real-world data and evidence generation platform to improve care for patients with prostate, breast, and lung cancer. To create this data source, we used the GOLD [4, 8] and Aurum [5, 9] December 2023 releases, both linked to the latest available HES APC [10–12], NCRAS cancer registration (NCRASCR) [13–15], and ONS Mortality datasets [16–18]. To avoid possible patient duplications, we have excluded from GOLD those practices that migrated from the Vision® software to the EMIS Web® software within CPRD. Because of the different data structures of these datasets and to facilitate the interoperability and execution of distributed network studies, we transferred each of these datasets to the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM) version 5.4 [19] and merged the resulting databases. The OMOP CDM is a well-established standardization method, which has been adopted to facilitate interoperability and federated analytics in various frameworks such as the European Health Data and Evidence Network (EHDEN) consortium [20], the Data Analysis and Real World Interrogation Network (DARWIN) EU® initiative [21], and the OPTIMA consortium [22]. CPRD-PCa-OMOP is the first data source in OMOP CDM format that includes CPRD primary care data linked to HES APC, ONS Mortality, and NCRASCR datasets. CPRD-PCa-OMOP includes 69 109 men from England aged ≥18 years diagnosed with their first PCa in GOLD or Aurum between 2010 and 2022 and confirmed in NCRASCR, with ≥1 year of registration at a GP practice [23]. In CPRD, only consenting GP practices from England are eligible for linkage and we linked primary care records to HES APC, ONS Mortality, and NCRASCR data (Table 1). Patients who opt out of having their data utilized in clinical research were excluded from CPRD database releases. We excluded ONS Mortality- and HES-linked data that were provided with a weak linkage likelihood (1 to 5 = the most likely, the least likely), following a previously published algorithm [24]. We only accepted records with a linkage likelihood of between 1 and 2 (i.e. hes_patient.match_rank <=2). To prevent data duplication, we retained only the most recently registered patient identifier when multiple identifiers were linked to the same individual via a general identifier provided in the HES data (i.e. hes_patient.hes_gen_id). Key details about CPRD-PCa-OMOP. IQR, interquartile range. Table 2 reports the patient demographics of CPRD-PCa-OMOP. There were 5566 patients in GOLD and 63 543 in Aurum. The age distribution was similar between both data sources, with most patients (38.40%) first diagnosed between 70 and 79 years of age. The median number of follow-up years from date of diagnosis until the end of observation was 6.13 [interquartile range (IQR) 3.71–8.68]. CPRD-PCa-OMOP patient demographics. IQR, interquartile range. In Table 3, we describe the patient characteristics of CPRD-PCa-OMOP in terms of cancer stage and grade, and the prevalence of comorbidities and comedications. Most patients had Stage 1 cancer (26.16%). A higher proportion of patients had Stage 3 and 4 cancers in Aurum as compared with GOLD. Prevalent comorbidities before PCa diagnosis included hypertension (52.30%), hyperlipidaemia (24.77%), and type 2 diabetes (14.92%). Common medications prescribed in the year before PCa diagnosis included lipid-modifying agents (45.62%), drugs for acid-related disorders (33.68%), and drugs for obstructive airway diseases (20.34%). CPRD-PCa-OMOP patient clinical characteristics. Invalid values were any values outside of 2–10. We standardized all the datasets described above to the OMOP CDM v. 5.4, the latest version recommended at the time of writing, to enable research collaboration through standardized analytical packages across database partners using a distributed federated network strategy. Although other CDMs exist, the OMOP CDM has been chosen by health regulatory agencies to support policy-making [21] and is widely used nationally [3, 25] and internationally [26]. The harmonization of the dictionaries used in the datasets was produced by the EHDEN and Observational Health Data Sciences and Informatics (OHDSI) initiatives: standardized vocabularies are available for downloading at https://athena.ohdsi.org. For this study, we used the Athena vocabulary bundle v5.0 29-FEB-24 and the sources for standard mapped values are documented in Supplementary Tables S1–S11. The data standardization to the OMOP CDM for primary care, secondary care, and mortality data was performed by following previously used and established techniques (https://github.com/oxford-pharmacoepi/etl_ndorms). For the cancer registration data, we developed new coding techniques by following the approach recommended by the OHDSI Oncology Working Group [2]. This approach, which provided part of the supporting rationale for the development of the OMOP CDM version 5.4, recognizes that cancer observational data are more challenging than those for most other conditions because they contain an individual’s sequence of cancer states and treatments that define the journeys of patients through the disease. Because CPRD-PCa-OMOP includes ONS Mortality data, which are based on death certificates and are the primary source of information for NCRAS mortality, we did not utilize the NCRAS information on death. For all the datasets, source data field names and values were stored in the corresponding source reference variables in the CDM tables to maintain data provenance. Table 4 summarizes the information available in the linked datasets and the corresponding source and target vocabularies. Key information and vocabularies of CPRD-PCa-OMOP. Read SNOMED SNOMED, OMOP Extension Read SNOMED SNOMED, CPT4, OMOP Extension dm+d gemscript RxNorm, RxNorm Extension CVX dm+d gemscript dm+d, SNOMED Read SNOMED SNOMED, LOINC, OMOP Extension Read SNOMED SNOMED, ICD-O-3 Cancer Modifier, SNOMED SNOMED, OPCS-4 Bespoken RxNorm Extension RxNorm, RxNorm Extension CPT4, Current Procedural Terminology, Fourth Edition; CVX, vaccine administered code; dm+d, NHS dictionary of medicines and devices; ICD-10, International Statistical Classification of Diseases and Related Health Problems, 10th Revision; ICD-O-3, International Classification of Diseases for Oncology, 3rd Edition; LOINC, Logical Observation Identifiers Names and Codes; NCRASCR, National Cancer Registration and Analysis Service cancer registration data; OPCS-4, Classification of Interventions and Procedures, version 4; SNOMED, Systemized Nomenclature of Medicine. To map NCRAS cancer diagnoses of the ‘tumour’ dataset, we employed the World Health Organization International Classification of Diseases for Oncology, 3rd Edition (ICD-O-3) international standard for tumour registries, which uses a concatenation of cancer’s attributes: [Histology]/[behaviour]-[Topography] (Supplementary Table S1). We mapped 69 106 PCa diagnoses to the OMOP CDM CONDITION_OCCURRENCE table by using 26 standard concept identifiers, of which 24 belonged to the ICD-O-3 vocabulary. One of the two standard concept identifiers that belonged to the Systemized Nomenclature of Medicine (SNOMED) vocabulary (4161028 = ‘8140/3-C61.9’ = ‘Adenocarcinoma of prostate’) covered 92.1% of the diagnoses. Six Aurum patients had diagnoses not present in the OHDSI vocabularies: the latter were stored in the OMOP CDM OBSERVATION table. Whenever possible, we mapped cancer modifiers of the ‘tumour’ dataset (i.e. stage, grade, size, number of local and regional lymph nodes tested and those classified as metastatic, and Charlson comorbidities indexes over 2 and 6 years) into the OMOP CDM MEASUREMENT table by using the OHDSI ‘Cancer Modifier’ and SNOMED standardized vocabularies, for a total of 510 359 records. When cancer modifiers were not adequately represented in the OHDSI vocabularies, we stored them in the OMOP CDM OBSERVATION table, following the OHDSI best practice recommendations [27]. Using the ‘Cancer Modifier’ vocabulary, we mapped several cancer stage attributes reported in the NCRAS dataset and classified them by using the American Joint Committee on Cancer (AJCC)/Union for International Cancer Control (UICC) Tumour, Node, Metastasis (TNM) system [28]. AJCC/TNM is a globally recognized standard that categorizes malignant tumours based on the primary tumour (T), the regional lymph node involvement (N), and distant metastasis (M). NCRAS derived stage attributes by combining related variables: ‘stage_best’ (‘t_best’, ‘n_best’, ‘m_best’), ‘stage_img’ (‘t_img’, ‘n_img’, ‘m_img’), and ‘stage_path’ (‘t_path’, ‘n_path’, ‘m_path’). We used 43 standard concept identifiers to map stages 1[A–E] to 4[A–E] associated with different versions of the AJCC/UICC TNM staging system for different patients and diagnoses (i.e. 6th, 7th, 8th, and no version). We mapped to standard concept identifiers several cancer grade attributes: the ‘Grade of tumour’ (G1 to G4 = more abnormal cells, greater likelihood of aggressive growth and spread) (Supplementary Table S2) and the scoring system for malignant neoplasm of the prostate ‘Gleason Primary/Secondary/Tertiary Pattern Grade’ (1 to 5 = the more aggressive, the least differentiated) by using the ‘Cancer Modifier’ vocabulary (Supplementary Tables S3–S5), and the ‘Gleason Combined Pattern Grade’ (Supplementary Table S6) by using the SNOMED vocabulary. There were 28 088 ‘Grade of tumour’ instances, of which 51.08% were classified as G3, 32.94% as G2, 12.49% as G4, and 3.50% as G1. The 57 852 ‘Gleason Combined’ instances were associated with 21 distinct values, of which we accepted values from 2 to 10. The most frequent value was 7 (39.98%), followed by 6 (19.94%), 9 (13,36%), 8 (9.07%), and 10 (1.13%). There were 440 instances of ‘Size of tumour’ reported as the diameter of a tumour in millimetres (if more than one tumour was present, the size of the largest tumour was reported). These instances were associated with 66 distinct values, of which the most frequent 5 (20, 15, 25, 10, and 12 mm) represented 33.18% of the total. There were 4482 instances of ‘Number of local and regional metastatic lymph nodes’. Of these, 17.0% reported no positive lymph nodes (value = 0) and 50.4% reported values from 1 (8.41%) to 9 (3.10%). The remaining 32.62% covered values from 11 (3.10%) to 89 (0.02%). There were 4375 instances of ‘Number of local and regional lymph nodes examined’ in excised specimens, with 89.74% reporting a value of 0. There were 68 912 instances of both Charlson comorbidity scores on diagnosis date: one covering 2 years, looking back from 27 to 3 months before diagnosis; and the other covering 6 years, looking back from 78 to 6 months before diagnosis. In both cases, the most frequent score was 0 (83.50% and 74.83%, respectively), followed by 1 (8.91%, 13.49%), 2 (4.09%, 6.30%), 3 (1.80%, 2.80%), 4 (0.86%, 1.28%), and 5 (0.36%, 0.59%), with higher scores having minimum representation. Cancer procedures from the NCRAS ‘Treatment’ dataset were mapped to the OMOP CDM PROCEDURE_OCCURRENCE table by using the OHDSI SNOMED and OPCS-4 standardized vocabularies (Supplementary Tables S7 and S8), for a total of 159 673 records. The five most frequent procedures accounted for 81.64% of the records, with ‘Imaging’ covering 40.07%, ‘Hormone therapy’ covering 20.27%, ‘External beam radiation therapy procedure’ covering 9.66%, ‘Transrectal needle biopsy of prostate’ covering 6.14%, and ‘Radical prostatectomy’ covering 5.50%. Cancer treatments from the NCRAS ‘Treatment’ dataset were mapped to the OMOP CDM DRUG_EXPOSURE table by using OHDSI ‘RxNorm’ and ‘RxNorm Extension’ standardized vocabularies (Supplementary Table S9). A total of 2913 records were mapped using 39 standard concept identifiers and the five most frequent concepts covered 85.65% of the records. Bicalutamide accounted for 38.76%, goserelin for 26.16%, cyproterone for 14.07%, gonadorelin for 3.50%, and docetaxel for 3.16%. All cancer diagnoses, modifiers, and treatments (Supplementary Table S10) were additionally mapped to the OMOP CDM EPISODE and EPISODE_EVENT tables to identify cancer patients’ trajectories through the disease (Supplementary Table S11). A total of 243 095 episodes were created associated with the domains of Procedure (42.61%), Regimen (28.96%), and Condition (28.43%). The number of episode events was 1 548 516, linking the events to the OBSERVATION (46.40%), MEASUREMENT (38.64%), PROCEDURE_OCCURRENCE (10.31%), CONDITION_OCCURRENCE (4.46%), and DRUG_EXPOSURE (0.19%) tables. All primary causes of death were mapped to the OMOP CDM DEATH table, together with the date of death. A total of 25 230 deaths were reported, with 40.29% (39.96% in Aurum, 44.16% in GOLD) associated with ‘Primary malignant neoplasm of prostate’ (concept_id = 200962). We assessed the quality of the data mapped to the OMOP CDM by using Kahn’s framework [29], implemented by the OHDSI software DataQualityDashboard (DQD, version: 2.6.3), available at https://github.com/OHDSI/DataQualityDashboard. DQD performs validation and verification checks according to conformance, completeness, and plausibility at the concept, field, and table levels by using check-specific thresholds. A total of 4684 data-quality checks were performed on the data, of which 98% with both and of the quality and plausibility The data-quality checks that were and of them was a cause for for study used in the UK were not recognized by DQD and a checks software all of which we reported to the DQD The linked datasets in their source format have been used in other to the quality and of cancer diagnoses. CPRD-PCa-OMOP is a dataset mapped to the CDM to allow in network studies. The data resource was created to data to the OPTIMA consortium and will be used to by related to the of patients with risk treatment patterns, and We are the characteristics and outcomes of patients who in a study The of CPRD-PCa-OMOP is data sources, which are comprehensive and of the from England, with longitudinal follow-up information and linkage to hospital admissions, cancer registry data, and mortality data, the latter for the of cancer linked data sources to between data and define the variables according to the study For the first date of cancer diagnosis between the different data that in diagnosis was across different cancer and higher between HES and For the of this data source, we used the date in NCRAS as the reference A proportion of patients, had in primary care and HES records. by in primary care the coding of cancers before cancer diagnoses for a of patients in HES records have from in NCRAS NCRAS records on which is a HES records the first such as one based on of CPRD-PCa-OMOP is that is which data integration, collaborative research, external validation, and network studies. of cancer records has been performed by England we have additionally mapped episodes to study patient trajectories and followed the OMOP CDM best practice The of CPRD-PCa-OMOP is the of information about and regimens, which we requested to a in the of the datasets by a This has the of the study to or and cancer we to the data and the corresponding and to the research as as is information cancer stage and Gleason grade which are in cancer research, was that the of cancer stage and grade has over data standardization interoperability between when different standard concepts are used to the same To this we have provided of the standardization with the corresponding In concept harmonization is based on the Athena vocabularies, their include and that the data are available at and The and is documented at and can for CPRD data access to and For about the mapped dataset, to the which information on the standardized variables and on the harmonization The data used in this study were provided the of CPRD and be the 3 and the are available on and be used to or the mapped database or as a for other cancer data CPRD has from the UK Health Research to support research using patient data and an NHS Data and to that the standard for data CPRD data are only via a CPRD a CPRD and a via the Research Data CPRD data be with not in the and they to be when the research has been to CPRD data and to the used to create CPRD-PCa-OMOP was by number We from the of for to the into the OMOP This study is based in part on data from the CPRD from the UK The data are provided by patients and collected by the NHS as part of their care and The and in this study are those of the The OPTIMA consortium in European of the of of and of the of for Research and Treatment of Cancer A Sciences and for Medicine 39 the OMOP data and the and the performed the data-quality All the the will as for the Supplementary is available at The Health Data Sciences research has research from the European Medicines from the Medicines from and from as well as or from and from outside the All other no additional of OPTIMA is through the Joint and is support from the European research and and the European of and collaborative research and of and to in The within are those of the the European any are for any that be of the information The and in this study are those of the The data this were provided by the CPRD the by the of These data can be requested at by CPRD was not used in the data, or or in this