Validation of Diagnostic Code Definitions for Identifying Patient in Korean Health Insurance Claims Data: A Scoping Review Authors: Kyoung-Hoon Kim Published: 2025 ABSTRACT BackgroundHealth administrative data are widely used in health services research, but diagnostic codes may be subject to misclassification. This scoping review examined the validity and reporting practices of operational definitions using the Korean National Health Insurance Claim Database (KHICD), focusing on compliance with the REporting of studies Conducted using Observational Routinely collected health Data (RECORD) statement.MethodsPubMed was searched for studies validating operational definitions or using the KHICD for population identification, published between 2020 and 2024. After screening 29 validation studies and 239 KHICD-based studies, 12 validation studies and 157 KHICD-based studies were included. Data on operational definitions, validation methods, and adherence to RECORD guidelines were extracted.ResultsAmong 12 validation studies, most focused on cancer, and algorithms combining diagnostic codes with rare intractable disease program codes, prescription, or procedure codes demonstrated higher positive predictive values than diagnostic codes alone. Among the 157 KHICD-based studies, 131 (83.4%) used diagnosis codes to identify patients, of which 71 (45.2%) combined them with supplementary codes. However, 60 (38.2%) studies relied solely on diagnosis codes, often without specifying diagnosis scope or acknowledging misclassification risks. Only three studies conducted validation, and overall compliance with the RECORD statement was limited.ConclusionWhile operational definitions using combined codes improve patient identification in the KHICD, methodological rigor and reporting transparency remain suboptimal. Systematic validation studies and strict adherence to reporting guidelines are needed to enhance reproducibility and comparability of KHICD-based research. INTRODUCTION Health administrative data are increasingly utilized in health services research because they are less costly than randomized clinical trials and are less prone to selection bias.1 These advantages enable population-based studies of disease epidemiology, treatment effectiveness, and health outcomes. However, administrative data also have inherent limitations, most notably the risk of misclassification of diagnoses, which may compromise study validity.2 To improve transparency and reproducibility in research using routinely collected health data, the REporting of studies Conducted using Observational Routinely collected health Data (RECORD) statement provides reporting guidelines.3 Specifically, the RECORD statement recommends that codes or algorithms used to identify study populations be described in sufficient detail, with explicit references to validation studies. When validation is conducted, both methodology and results should be clearly reported. The Korean National Health Insurance Claim Database (KHICD) is a representative source of real-world data, containing information on diagnoses, procedures, tests, and prescriptions.4 Because procedures and prescriptions are directly tied to reimbursement under the fee-for-service system, their validity is relatively high. By contrast, diagnostic codes often do not represent confirmed clinical diagnoses, but instead reflect conditions most relevant to resource utilization. This increases the likelihood of misclassification bias and threatens validity. Consequently, operational definitions that combine diagnostic codes with additional information are essential to improve diagnostic accuracy. Several studies have developed operational definitions by linking International Classification of Diseases, 10th revision (ICD-10) codes with medication prescriptions, requiring repeated healthcare encounters, or incorporating rare and intractable disease (RID) program codes, which are directly tied to reimbursement benefits. For example, Lee et al.5 defined chronic obstructive pulmonary disease (COPD) cases as patients with ICD-10 codes J43–J44 plus at least two outpatient visits with a COPD prescription, while Chang et al.6 defined cancer patients using ICD-10 codes C00–C96 in combination with RID codes. Such approaches enhance reproducibility and generalizability of study results. Despite the growing use of the KHICD, however, no systematic review has comprehensively evaluated the validity of these operational definitions. This review therefore aims to summarize validation efforts and examine reporting practices in relation to the RECORD statement. METHODS Search for studies We conducted a scoping review of studies that either validated operational definitions or used the KHICD to identify study populations. A search of PubMed was performed in April 2025 using predefined search strategies: · Validation studies: (Korea OR Korean) AND (“health insurance claim* data*” OR “health insurance claim*” OR HIRA OR NHIS) AND (“Validation Study” OR validation OR “operational definition*”) AND (“2020/01/01”: “2024/12/31”) · Validation studies: (Korea OR Korean) AND (“health insurance claim* data*” OR “health insurance claim*” OR HIRA OR NHIS) AND (“Validation Study” OR validation OR “operational definition*”) AND (“2020/01/01”: “2024/12/31”) · KHICD-based studies: (Korea OR Korean) AND (“health insurance claim*” OR claims data OR NHIS OR HIRA) AND (“2024”) · KHICD-based studies: (Korea OR Korean) AND (“health insurance claim*” OR claims data OR NHIS OR HIRA) AND (“2024”) The search period was determined according to the purpose of this review. To capture the most recent reporting practices and current trends in the use of administrative data in Korea, KHICD-based studies were restricted to those published in 2024. In contrast, because only a limited number of validation studies were published in 2024, the search window for validation studies was extended to include the previous five years (2020–2024). This broader time frame ensured adequate coverage and allowed inclusion of a representative and contemporary set of validation studies. Selection of study All retrieved records underwent title, abstract, and full-text screening. We excluded duplicate records, review articles, studies using non-Korean data, and studies relying on alternative data sources such as electronic medical records, immunization registries, or health screening databases. Among the 29 validation studies identified, 12 were retained after abstract and full-text review. Of the 239 KHICD-based studies initially retrieved, 215 remained after excluding review articles (n = 9), studies using non-Korean data (n = 11), and those unrelated to KHICD use (n = 4) (Fig. 1). We further excluded studies relying on alternative data sources such as electronic medical record and those conducted at the hospital-level rather than the patient-level (n = 58). Ultimately, 157 studies met the eligibility criteria and were included in the review. Flowchart for study selection for review. KHICD = Korean Health Insurance Claim Database. Evaluation framework for included studies To ensure a rigorous and transparent assessment, we set an evaluation framework comprising three complementary domains: 1) review of validation methodologies, 2) review of operational definitions, and 3) evaluation of adherence to the RECORD statement. All included studies were independently assessed by two reviewers (Kim KH and Lee SH). Any discrepancies in evaluation were resolved through discussion and consensus. For validation studies, we reviewed the methodological rigor by examining the study population, the reference data sources used for comparison with the KHICD, and the diagnostic code agreement metrics reported. These metrics included positive predictive value (PPV), accuracy, and other relevant indicators of validity. In addition, we reviewed the final operational definitions proposed by the authors for case identification. For KHICD-based studies, operational definitions were examined with particular attention to the specification of diagnostic, procedure, and drug codes used to identify cases, as well as the use of supplementary codes. Adherence to the RECORD statement was evaluated using a modified checklist tailored to administrative data research. Specifically, we assessed whether each study adequately reported its operational definition algorithms, including code lists and case-finding rules; whether justification for the selected definitions was provided through references to prior research or explicit reporting of validation procedures; and whether potential risks of misclassification—such as coding inaccuracies, inherent limitations of claims data, or the absence of clinical confirmation—were appropriately acknowledged. Each study was ultimately classified according to whether these key elements were explicitly described in the manuscript. RESULTS Validation studies for algorithm or diagnostic codes to identify study population Twelve studies evaluated the validity of diagnostic codes in the KHICD (Table 1). Nine employed chart review, while three compared against external data sources such as the National Cancer Registry (NCR) or the Korea National Health and Nutrition Examination Survey (KNHNES). Most validation studies focused on cancer. Yang et al.8 assessed algorithms for multiple cancer types (breast, cervical, colorectal, liver, lung, prostate, stomach) against the NCR, and found that algorithms incorporating RID codes were more accurate than those using primary diagnoses alone. Similar findings were reported for bile duct cancer.7 Park et al.9 and Kim et al.10 conducted literature reviews to find operational definitions for colorectal and liver cancer, respectively, and reported that primary diagnosis codes produced age-standardized incidence rates comparable to national cancer statistics from the NCR. Table 1Reviews of validation studies for operational definitionAuthor, yearCondition validatedReference standardRecommended operational definitionHwang et al. 20207 Bile duct cancerChart reviewDiagnostic code and codes for rare and intractable diseases registration program (PPV: 98.3%)Yang et al., 20218 Cancer (stomach, liver, colorectal, lung, breast, cervical, prostate)National Cancer Registry DatabaseCodes for rare and intractable diseases registration programPark et al., 20239 Colorectal cancerCancer statisticsC18–C20 as the main diagnosisKim et al., 202310 Liver cancerCancer statisticsC22.0 or C22.9 for men and C22.0 for women as the main diagnosisPark et al., 201911 AMI, gastrointestinal bleeding, stroke, intracranial hemorrhageChart reviewAMI, gastrointestinal bleeding: hospitalization with primary diagnostic code (PPV: 92.0%, 92%, respectively)Stroke, intracranial hemorrhage: hospitalization with primary diagnostic code (over 3 days) and prescription (PPV: 94.2%, 83.1%, respectively)Ham et al., 202012 PsoriasisChart review≥ 1 visits and prescription of vitamin D derivatives (PPV: 96.5%)Lee et al., 202013 Inflammatory bowel diseaseChart review≥ 1 claim and ≥ 1 prescription claims (PPV: 97.5%)Lee et al., 202114 Distal radius fracturesChart reviewEpidemiology study: any diagnosis codeComparative effectiveness study: any diagnosis code and procedure codeYu et al., 202215 Osteoporotic vertebral fracturesChart reviewDiagnostic code as primary diagnosis and procedure code and imaging procedure code (PPV: 59.7%)Lee et al., 202216 Osteoporotic hip fracturesChart reviewPrimary, secondary, or tertiary diagnosis codes and procedure codes (PPV: 95.7%)Kim et al., 202217 Diabetic microvascular complicationsChart reviewDiagnostic code and blood test and urinary proteinuria (PPV: 81.4%)Baek et al., 202318 Type 2 diabetes mellitusKorea National Health and Nutrition Examination Survey Diagnostic codes and prescription codes for anti-diabetic drugs (accuracy 93%, specificity 0.96%)PPV = predictive positive value, AMI = acute myocardial infarction. Reviews of validation studies for operational definition PPV = predictive positive value, AMI = acute myocardial infarction. Non-cancer conditions were also studied. Kim et al.17 validated diabetic microvascular complications via chart review, demonstrating higher PPVs of 81.4% when combining diagnostic codes with laboratory tests. Baek et al.18 found that adding prescription data improved identification of type 2 diabetes when compared with KNHNES. For orthopedic conditions such as distal radius fractures, osteoporotic vertebral fractures, and hip fractures, algorithms combining diagnosis codes with procedure codes yielded higher PPVs than diagnosis codes alone.141516 Validation studies on stroke, psoriasis, and inflammatory bowel disease have demonstrated that algorithms incorporating at least one diagnostic code in combination with a prescription claim yielded higher positive predictive values.111213 Operational definitions to identify study population Among the 157 included studies, patients were identified using diagnosis codes, health insurance procedure codes, drug codes, RID codes, or disease related group codes, depending on the study population (Fig. 2). Twenty-six studies relied primarily on procedure codes and drug codes to identify surgical or drug-based cohorts. Excluding these, 131 studies used diagnosis codes as one of the criteria for patient selection (Table 2). Of these, 38 combined diagnosis codes with procedure codes and/or drug codes. Twenty-nine studies operationally defined patients using diagnosis codes plus prescriptions, mostly for chronic diseases such as COPD, asthma, dyslipidemia, hypertension, and diabetes, or psychiatric disorders. Thirty-one studies combined diagnosis codes with RID codes, typically focusing on cancer or rare diseases such as inflammatory bowel disease, ankylosing spondylitis, and Wilson’s disease. Number of studies by combination of codes for operational definition (n = 157). RID = rare and intractable diseases registration program. Table 2Compliance with the RECORD statement for studies using diagnostic codeCriteria for defining a study population sTotalDescribing diagnosis scopeDescribing validity of algorithmsPrimary diagnosisPrimary or sub-diagnosisUnreportedAdopted from previous researchValidated in studyMentioned only in limitationsUnreportedTotal131 (100.0)18 (13.7)25 (19.1)88 (67.2)31 (23.7)3 (2.3)30 (22.9)67 (51.1)Diagnostic code60 (45.8)11 (18.3)10 (16.7)39 (65.0)5 (8.3)2 (3.3)20 (33.3)33 (55.0)Diagnostic code + drug code29 (22.1)3 (10.3)10 (34.5)16 (55.2)9 (31.0)0 (0.0)6 (20.7)14 (48.3)Diagnostic code + procedure code7 (5.3)1 (14.3)1 (14.3)5 (71.4)1 (14.3)0 (0.0)0 (0.0)6 (85.7)Diagnostic code + procedure code + drug code2 (1.5)0 (0.0)1 (50.0)1 (50.0)0 (0.0)0 (0.0)1 (50.0)1 (50.0)Diagnostic code + RID23 (17.6)2 (8.7)3 (13.0)18 (78.3)14 (60.9)1 (4.3)2 (8.7)6 (26.1)Diagnostic code + RID + procedure code and/or drug code8 (6.1)0 (0.0)0 (0.0)8 (100.0)2 (25.0)0 (0.0)1 (12.5)5 (62.5)Othersa 2 (1.5)1 (50.0)0 (0.0)1 (50.0)0 (0.0)0 (0.0)0 (0.0)2 (100.0)Values are presented as number (%).RECORD = REporting of studies Conducted using Observational Routinely collected health Data, RID = rare and intractable diseases registration program.aOthers: combination of diagnostic code and disease related group, combination of diagnostic code and specific code that makes a person eligible for financial assistance during national disasters. Compliance with the RECORD statement for studies using diagnostic code Values are presented as number (%). RECORD = REporting of studies Conducted using Observational Routinely collected health Data, RID = rare and intractable diseases registration program. aOthers: combination of diagnostic code and disease related group, combination of diagnostic code and specific code that makes a person eligible for financial assistance during national disasters. Despite the inherent limitations of diagnosis codes, 60 studies identified their populations solely on diagnosis codes. Among these, seven studies refined case definitions based on the number of encounters. Woo et al.19 and Kim et al.20 defined kidney disease as ≥ 1 inpatient or ≥ 2 outpatient diagnoses, and ≥ 2 inpatient or outpatient diagnoses, respectively. Cancer patients were defined as those with ≥ 3 outpatient visits and ≥ 1 inpatient admission or those with ≥ 2 claims for lung cancer.2122 Studies of menopause, obstructive sleep apnea, and diabetes defined patients as those with ≥ 3 healthcare encounters for the respective diagnosis.232425 Compliance with the RECORD statement Among the 131 diagnosis-based studies described their algorithms or codes (Table 2). However, only 43 explicitly specified the scope of diagnosis codes (e.g., primary, primary/sub-diagnosis), while 88 did not. Thirty-one studies adopted algorithms previously reported in the literature. Kim and Lee26 classified dementia patients as those with ≥ 1 medical claim with relevant diagnosis codes (F00, F01, F02, F03, G30, G31.82) or ≥ 1 prescription for dementia-related medications (donepezil, galantamine, rivastigmine, or memantine). Similarly, Park et al.27 defined COPD as a primary or secondary diagnosis plus ≥ 2 prescriptions for COPD medications per year. Validation was performed in only three studies. Two employed chart review, and one used an external data source. Kim et al.28 validated primary hyperparathyroidism in 431 patients (29.4%) through chart review, using a definition based on repeated ICD-10 code E21.0 and parathyroid hormone testing. They demonstrated that the combined approach yielded the highest PPV (92%), with high specificity (97%) but moderate sensitivity (71%) among algorithms. Kim et al.29 also validated Crohn’s disease and ulcerative colitis using ICD-10 + RID codes versus ICD-10 alone, reporting sensitivities of 94.5% and 96.4%, respectively. Park et al.30 compared the prevalence of non-alcoholic fatty liver disease identified using their algorithm with that reported in previous studies and the KNHNES, and found it to be lower, attributing the discrepancy to differences in study populations and diagnostic criteria. Thirty studies (22.9%) briefly mentioned misclassification risk due to the inherent limitations of the KHICD, whereas 67 (51.1%) made no mention at all. Notably, among the 60 studies using only diagnosis codes, 33 (55.0%) did not report limitations related to diagnosis accuracy. DISCUSSION This scoping review examined the use of operational definitions and reporting practices in studies employing the KHICD of Korea, with particular attention to compliance with the RECORD statement. We found that while many studies enhanced case definitions by combining diagnostic codes with procedure, drug, or RID codes, a substantial proportion still relied solely on diagnosis codes, often without clearly specifying diagnosis scope or acknowledging misclassification risk. Compliance with the RECORD statement was limited, particularly regarding transparent reporting of operational definitions and validation efforts. Of the 131 studies that used diagnostic codes to identify study population, 71 studies used combinations of diagnostic code and supplementary codes including procedure code, drug code, and RID code, but 60 studies used only diagnostic codes. Furthermore, while 30 out of 131 studies (22.9%) briefly acknowledged the risk of misclassification inherent to claims data, more than half (51.5%) failed to address this issue. Among studies relying exclusively on diagnostic code, 55.0% did not mention misclassification and 65.0% studies did not specify the diagnostic scope, reflecting a lack of transparency that contradicts the principles of the RECORD statement, which emphasizes detailed reporting to enhance reproducibility and comparability across studies. Validation of health administrative data for patient identification remains a research priority.2 The efforts generally assessed diagnostic accuracy against external gold standards such as chart review or independent data sources, using measures including sensitivity, specificity, PPV, and negative predictive value.3132 To improve methodological rigor, reporting guidelines have been developed. The Standards for Reporting of Diagnostic Accuracy steering committee produced a checklist and a generic flow diagram through literature reviews and expert consensus.33 Building on this, Benchimol et al.2 developed a 40-point data collection tool and applied the tool to the 271 articles published from 1980 to 2009, reporting considerable variation in the quality of studies validating health states using administrative data and substantial deficiencies in the reporting of diagnostic accuracy measures. Most validation studies using the KHICD employed chart review or external registries as the gold standard, and consistently showing that algorithms incorporating supplementary information such as RID codes, procedures, or prescriptions yielded higher validity than diagnostic codes alone. However, validation was conducted in only a small fraction of studies, and restricted to a limited set of diseases. Systematic validation studies encompassing a wide range of disease populations, grounded in the aforementioned guidelines, are essential to prevent the misclassification and improve the reproducibility of study utilizing the KHICD. Of particular note, RID codes—unique to the Korean health system—link diagnoses of rare or severe diseases directly to reimbursement benefits, providing a robust complement to ICD-10 codes. This approach has been shown to substantially enhance the accuracy of case identification, especially in studies of cancer and rare diseases. This review has several limitations. First, although this review included a substantial number of recently published studies, the search was restricted to English-language publications indexed in PubMed. As a result, relevant Korean-language studies may have been inadvertently excluded. Future research incorporating domestic databases such as KoreaMed would provide a more comprehensive and balanced assessment of reporting practices in Korean administrative data research. Second, compliance was assessed based on reporting within published articles, which may underestimate actual validation efforts undertaken but not documented. In conclusion, although the use of the KHICD in health service research is expanding rapidly, significant gaps remain in the methodological rigor and transparency of reporting. Strengthening adherence to established reporting guidelines and promoting validation research systematically will enhance the reliability, comparability, and generalizability of the KHICD–based studies in Korea.