This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on http://www.jmir.org/, as well as this copyright and license information must be included.
Studies using Taiwan’s National Health Insurance (NHI) claims data have expanded rapidly both in quantity and quality during the first decade following the first study published in 2000. However, some of these studies were criticized for being merely data-dredging studies rather than hypothesis-driven. In addition, the use of claims data without the explicit authorization from individual patients has incurred litigation.
This study aimed to investigate whether the research output during the second decade after the release of the NHI claims database continues growing, to explore how the emergence of open access mega journals (OAMJs) and lawsuit against the use of this database affect the research topics and publication volume and to discuss the underlying reasons.
PubMed was used to locate publications based on NHI claims data between 1996 and 2017. Concept extraction using MetaMap was employed to mine research topics from article titles. Research trends were analyzed from various aspects, including publication amount, journals, research topics and types, and cooperation between authors.
A total of 4473 articles were identified. A rapid growth in publications was witnessed from 2000 to 2015, followed by a plateau. Diabetes, stroke, and dementia were the top 3 most popular research topics whereas statin therapy, metformin, and Chinese herbal medicine were the most investigated interventions. Approximately one-third of the articles were published in open access journals. Studies with two or more medical conditions, but without any intervention, were the most common study type. Studies of this type tended to be contributed by prolific authors and published in OAMJs.
The growth in publication volume during the second decade after the release of the NHI claims database was different from that during the first decade. OAMJs appeared to provide fertile soil for the rapid growth of research based on NHI claims data, in particular for those studies with two or medical conditions in the article title. A halt in the growth of publication volume was observed after the use of NHI claims data for research purposes had been restricted in response to legal controversy. More efforts are needed to improve the impact of knowledge gained from NHI claims data on medical decisions and policy making.
Health care administrative data, also known as administrative claims data, [
Taiwan’s National Health Insurance Research Database (NHIRD), one of the largest health care administrative databases in the world, has provided a great opportunity for researchers to perform population-based studies [
While more and more authors have successfully published studies using NHI claims data as the data source, some of these studies were criticized for being merely data-dredging studies rather than hypothesis-driven [
On the other hand, with the widespread use of the NHIRD in research, several human rights groups protested the use of claims data without the explicit authorization from individual patients and launched a lawsuit against the NHI Administration in 2012 [
Studies based on the NHIRD have expanded substantially in both quantity and quality during the period of 2000 to 2009 [
This study used PubMed to locate publications that may have used NHI claims data as the primary data source because PubMed is the most widely used database for searching medical literature. Articles published in English that entered PubMed between Jan 1, 1996 and Dec 31, 2017 with “journal article” as their publication type were included. Following the work by Chen et al [
Because journals may change their titles or even merge, this study always adopted the last title of a journal. JCR Science Edition and Social Sciences Edition (Clarivate Analytics, 2018) was used to retrieve 2017 Journal Impact Factors and journal categories. Journals were classified from Q1 to Q4 according to the impact factor quartiles in the specific journal category, where Q1 journals stand for journals with higher impact factors. Journals not indexed by the JCR were classified as other. Open access journals were identified through the Directory of Open Access Journals. OAMJs were defined as described in a previous study [
All the articles were downloaded from PubMed and preprocessed using the “easyPubMed” package in R. Because the list of potential articles was quite lengthy, several heuristic rules were applied to determine whether an article used NHI claims data as the primary data source. Basically, regular expression pattern matching was used to detect the mentioning of using NHI claims data in article abstracts and adjusted the matching patterns by trial and error. This study finally found two inclusion rules and one exclusion rule. The rules are shown in
The remaining 2046 articles were reviewed by the first and second authors. Each author independently classified an article as “using NHI claims data,” “not using NHI claims data,” or “using data from undetermined source” by examining its abstract or full text, when necessary. This process achieved an agreement of 99.1% (kappa=0.978), and discrepancies (19 articles) were resolved by consensus. Among them, 632 articles were considered not using NHI claims data and thus excluded. In the end, a total of 4473 articles were included in this study (
1. (from|data|study|using|cohort|used|based|patients|identified|population|obtained|
claim|conducted|retrieved|collected|selected|analyz)[[:print:]]{1,20}(National Health Insurance|NHI|Longitudinal Health Insurance|insurance claims|Registry for Catastrophic Illness Patient)[[:print:]]{1,20}(claim|data|file)”)
2. (National Health Insurance|NHI|Longitudinal Health Insurance|insurance claims|Registry for Catastrophic Illness Patients)[[:print;]]{1,20}(claim|data|file)[[:print:]]{1,20}(from|
used|patients|identified)
1. (Korea)[[:print:]]{1,20}(National Health Insurance|NHI)
This study used MetaMap as the tool to mine medical entities, such as symptoms, clinical findings, diseases, and medications, from article titles. MetaMap is a natural language processing tool developed by the National Library of Medicine. It analyzes input text through tokenization, sentence boundary determination, part-of-speech tagging, and parsing and generates variants of resulting phrases or words [
Because an article title typically indicates what the article is about, this study attempted to mine knowledge from article titles. Although MeSH terms are also used in PubMed to describe what an article is about, we analyzed MetaMap-derived concepts instead of MeSH concepts for two reasons. First, we focused on the article title, but the MeSH concepts were determined by examining the whole article; second, the UMLS Metathesaurus contains far more medical concepts than the MeSH vocabulary. This study focused on two categories of medical entities: (1) medical conditions including diseases, symptoms and signs, and findings and (2) interventions, including medications, procedures, and surgery. Specifically, medical entities were categorized based on their UMLS semantic types (see
Researchers may have different preferences for research topics when using administrative databases as the primary data source [
In order to offer background statistics, PubMed was queried to identify the number of articles published in English that entered PubMed between 2000 and 2017 with “journal article” as their publication type. The number of articles published between 2000 and 2017 in the top 20 journals that have published the most studies using NHI claims data were also obtained. Furthermore, the titles of the articles in PubMed and the top 20 journals were screened for presence of the top 10 medical conditions and top 10 interventions that were the most prevalent among studies using NHI claims data. The percentage of each medical condition or intervention among published articles was calculated.
Categorical variables are reported as counts (percentages). Comparisons between groups used chi-square tests. Trends in continuous variables were assessed using the Cuzick test. Trends in categorical outcomes were evaluated using the Cochran-Armitage trend test for binomial proportions and the multinomial Cochran-Armitage Trend Test implemented in the R package “multiCA” [
Statistical analyses and visualizations were performed using Stata 15.1 (StataCorp, College Station, Texas) and R version 3.5.0 (R Foundation for Statistical Computing, Vienna, Austria). Two-tailed
Since the first article appeared in 2000, the number of publications grew tremendously until 2015, when the publication output seemed to reach a plateau (
Annual number of publications between 2000 and 2017 (A) in PubMed, (B) in the top 20 journals that have published the most studies using National Health Insurance (NHI) claims data, and (C) based on NHI claims data, separated by JCR (2017 edition) ranking (Q1, Q2, Q3, Q4). The first year of each major OAMJ is indicated. The release of the National Health Insurance Research Databases (NHIRD) began in 2000 and ended in 2015. The Health and Welfare Database (HWD) was created in 2009 and is still available for research use.
Characteristics of and trends for articles and journals.
Articles and journals | Period | Total number | ||||
2000–2005 | 2006–2011 | 2012–2017 | ||||
|
|
|
|
|
|
|
|
Indexed in PubMed, n | 86 | 636 | 3751 | 4473 | <.001 |
|
Indexed in JCRa 2017, n (%) | 78 |
625 |
3635 (96.9) | 4338 |
.219 |
|
First author from hospitals, n (%) | 33 |
303 |
2479 |
2815 |
<.001 |
|
First author from abroad | 3 (3.5) | 20 (3.1) | 71 (1.9) | 94 |
.019 |
|
Published in OAJsb, n (%) | 16 |
103 |
1447 (38.6) | 1566 |
<.001 |
|
Published in OAMJsc, n (%) | 0 (0) | 8 (1.3) | 898 |
906 |
<.001 |
|
|
|
|
|
<.001 | |
|
With ≥1 intervention, n (%) | 29 |
202 |
1315 (35.1) | 1546 |
.002 |
|
With ≥2 conditions, n (%) | 8 |
132 |
1435 (38.3) | 1575 |
<.001 |
|
With only 1 condition, n (%) | 30 |
210 |
797 |
1037 |
<.001 |
|
Others, n (%) | 19 (22.1) | 92 (14.5) | 204 (5.4) | 315 (7.0) | <.001 |
|
|
|
|
|
|
|
|
Indexed in PubMed, n | 59 | 297 | 766 | 841 | <.001 |
|
Indexed in JCR 2017, n | 53 | 287 | 727 | 791 | <.001 |
|
Journal categories, n | 32 | 56 | 68 | 68 | <.001 |
aJCR: Journal Citation Reports.
bOAJ: open access journal.
cOAMJ, open access mega journal.
A total of 1763 medical entities were retrieved from article titles using MetaMap. Among these, 1132 entities belonged to the category of medical conditions, and 631 belonged to the category of interventions. The most commonly investigated medical conditions were diabetes (n=263), stroke (n=189), and dementia (n=139), whereas the most commonly studied interventions were statin (n=100), metformin (n=40), and Chinese herbal medicine (n=38). The top 50 medical conditions and interventions are listed in
Relative frequency of the top 50 medical conditions and interventions mentioned in article titles of studies based on National Health Insurance (NHI) claims data.
Medical conditions and interventions | Number of times mentioned | |
|
|
|
|
Diabetes | 263 |
|
Stroke | 189 |
|
Dementia | 139 |
|
Type 2 diabetes | 132 |
|
Cancer | 124 |
|
End stage renal disease | 102 |
|
Atrial fibrillation | 88 |
|
Chronic obstructive pulmonary disease | 88 |
|
Chronic kidney disease | 87 |
|
Schizophrenia | 82 |
|
Ischemic stroke | 80 |
|
Asthma | 79 |
|
Depression | 74 |
|
Rheumatoid arthritis | 71 |
|
Breast cancer | 68 |
|
Tuberculosis | 65 |
|
Parkinson disease | 59 |
|
Acute myocardial infarction | 55 |
|
Bipolar disorder | 55 |
|
Hepatocellular carcinoma | 55 |
|
Hypertension | 55 |
|
Osteoporosis | 55 |
|
Hip fracture | 54 |
|
Lupus erythematosus, systemic | 54 |
|
Pneumonia | 53 |
|
Fracture | 51 |
|
Attention deficit-hyperactivity disorder | 49 |
|
Acute pancreatitis | 46 |
|
Cardiovascular disease | 45 |
|
Erectile dysfunction | 41 |
|
Lung cancer | 41 |
|
Acute coronary syndrome | 40 |
|
Coronary artery disease | 38 |
|
Depressive disorder | 38 |
|
Infection | 37 |
|
Peripheral arterial disease | 37 |
|
Prostate cancer | 37 |
|
Epilepsy | 36 |
|
Traumatic brain injury | 36 |
|
Sleep disorder | 35 |
|
Colorectal cancer | 34 |
|
Migraine | 34 |
|
Psoriasis | 34 |
|
Hearing loss, sudden | 31 |
|
Gout | 27 |
|
Liver abscess, pyogenic | 27 |
|
Obstructive sleep apnea | 27 |
|
Sleep apnea | 27 |
|
Alzheimer's disease | 26 |
|
Psychiatric disorder | 26 |
|
|
|
|
Statin | 100 |
|
Metformin | 40 |
|
Chinese herbal medicine | 38 |
|
Hemodialysis | 38 |
|
Antidepressant | 36 |
|
Antipsychotic | 35 |
|
Proton pump inhibitor | 29 |
|
Dialysis | 27 |
|
Nonsteroidal anti-inflammatory drugs | 27 |
|
Corticosteroid | 24 |
|
Influenza vaccination | 24 |
|
Angiotensin-converting enzyme inhibitor | 22 |
|
Benzodiazepine | 21 |
|
Zolpidem | 19 |
|
Antihypertensive agents | 18 |
|
Dialysis, peritoneal | 17 |
|
Thiazolidinedione | 17 |
|
Antidiabetic | 16 |
|
Angiotensin receptor blockers | 14 |
|
Reduction | 14 |
|
Tamoxifen | 14 |
|
Antibiotic | 13 |
|
Cholecystectomy | 13 |
|
Sitagliptin | 13 |
|
Angiotensin 2 receptor blockers | 12 |
|
Antiepileptic drug | 12 |
|
Chemotherapy | 12 |
|
Selective serotonin reuptake inhibitors | 12 |
|
Acupuncture | 11 |
|
Intervention, percutaneous coronary | 11 |
|
Mechanical ventilation | 11 |
|
Pioglitazone | 11 |
|
Appendectomy | 10 |
|
Aspirin | 10 |
|
Clopidogrel | 10 |
|
Hormone therapy | 10 |
|
Interferon | 10 |
|
Splenectomy | 10 |
|
Total knee arthroplasty | 10 |
|
Antiplatelet agents | 9 |
|
Caesarian section | 9 |
|
Coronary artery bypass grafting | 9 |
|
Drug eluting stent | 9 |
|
Liver transplantation | 9 |
|
Radiotherapy | 9 |
|
Resection | 9 |
|
Alendronate | 8 |
|
Antiviral | 8 |
|
Digoxin | 8 |
|
Hypnotic | 8 |
Number and percentage of articles with the corresponding condition or intervention in the article title between 2000 and 2007.
Articles | Studies using NHIa claims data |
Articles in top 20 journals |
Articles in PubMed |
|
|
||||
|
Diabetes | 263 (5.9) | 4195 (1.2) | 112,484 (0.9) |
|
Stroke | 189 (4.2) | 1944 (0.5) | 56,781 (0.5) |
|
Dementia | 139 (3.1) | 664 (0.2) | 24,217 (0.2) |
|
Type 2 diabetes | 132 (3.0) | 1724 (0.5) | 38,757 (0.3) |
|
Cancer | 124 (2.8) | 22,030 (6.1) | 503,283 (4.1) |
|
End stage renal disease | 102 (2.3) | 169 (0.0) | 4387 (0.0) |
|
Atrial fibrillation | 88 (2.0) | 1189 (0.3) | 21,867 (0.2) |
|
Chronic obstructive pulmonary disease | 88 (2.0) | 450 (0.1) | 10,356 (0.1) |
|
Chronic kidney disease | 87 (1.9) | 671 (0.2) | 12,617 (0.1) |
|
Schizophrenia | 82 (1.8) | 1215 (0.3) | 33,536 (0.3) |
|
|
|||
|
Statin | 100 (2.2) | 319 (0.1) | 5130 (0.0) |
|
Metformin | 40 (0.9) | 372 (0.1) | 6408 (0.1) |
|
Chinese herbal medicine | 38 (0.8) | 158 (0.0) | 674 (0.0) |
|
Hemodialysis | 38 (0.8) | 58 (0.0) | 3175 (0.0) |
|
Antidepressant | 36 (0.8) | 696 (0.2) | 7327 (0.1) |
|
Antipsychotic | 35 (0.8) | 337 (0.1) | 5540 (0.0) |
|
Proton pump inhibitor | 29 (0.6) | 38 (0.0) | 1105 (0.0) |
|
Dialysis | 27 (0.6) | 416 (0.1) | 16,301 (0.1) |
|
Nonsteroidal anti-inflammatory drugs | 27 (0.6) | 2 (0.0) | 260 (0.0) |
|
Corticosteroid | 24 (0.5) | 109 (0.0) | 4475 (0.0) |
aNHI: National Health Insurance.
Distribution of study types across the years.
Network graphs displaying the most common (A) condition-condition pairs and (B) condition-intervention pairs.
Until the end of 2017, 4473 articles were published in 841 journals, with an average of 5.3 articles per journal. Among these journals, 791 were indexed in the JCR Science Edition or Social Sciences Edition and were spread across 68 disciplines. The journals PLOS ONE and Medicine ranked first and second, respectively, in the number of publications and yielded 18.0% (804/4473) of the articles, whereas 333 journals published only one article each. This study applied Bradford's law and divided journals into three groups by the rank of journals, with each group of journals publishing approximately the same number of articles (
Scattering of articles in journals.
Group | Journals (n=841), n | Articles (n=4473), n | Cumulative, n (%) | Description |
Top third | 20 | 1504 | 1504 (33.6) | Publishing 26-419 articles |
Middle third | 113 | 1315 | 2819 (63.0) | Publishing 8-25 articles |
Bottom third | 708 | 1654 | 4473 (100.0) | Publishing 1-7 articles |
Open access journals published 35% of the studies, whereas OAMJs published around one-fifth of the studies (
Top 20 journals ranked by published articles between 2000 and 2017.
Journal name | IFa | Rank in JCRb in 2017 | Articles, n (%) | OAJc | OAMJd |
PLOS ONE | 2.766 | Multidisciplinary sciences (15/64) | 419 (9.4) | Y | Y |
Medicine | 2.028 | Medicine, general & internal (56/154) | 385 (8.6) | Y | Ye |
International Journal of Cardiology | 4.034 | Cardiac & cardiovascular systems (41/128) | 72 (1.6) | N | N |
Journal of the Formosan Medical Association | 2.452 | Medicine, general & internal (42/154) | 52 (1.2) | Y | N |
Oncotarget | N/Af | N/A | 51 (1.1) | Yg | N |
Journal of Affective Disorders | 3.786 | Clinical neurology (46/197), psychiatry (37/142), psychiatryh (27/142) | 51 (1.1) | N | N |
Scientific Reports | 4.122 | Multidisciplinary sciences (12/64) | 49 (1.1) | Y | Y |
Pharmacoepidemiology and Drug Safety | 2.314 | Public, environmental & occupational health (66/180), |
47 (1.1) | N | N |
BMJ Open | 2.413 | Medicine, general & internal (43/154) | 47 (1.1) | Y | Y |
Journal of the Chinese Medical Association | 1.660 | Medicine, general & internal (72/154) | 39 (0.9) | Y | N |
Journal of Ethnopharmacology | 3.115 | Plant sciences (38/222); chemistry, medicinal (20/59); |
36 (0.8) | N | N |
BMC Health Services Research | 1.843 | Health care sciences & services (53/94) | 34 (0.8) | Y | N |
European Journal of Internal Medicine | 3.282 | Medicine, general & internal (27/154) | 30 (0.7) | N | N |
Research in Developmental Disabilities | 1.820 | Education, specialh (8/40), rehabilitationh (19/69) | 30 (0.7) | N | N |
Health Policy | 2.293 | Health care sciences & services (40/94), health policy & servicesh (22/79) | 29 (0.6) | N | N |
Osteoporosis International | 3.856 | Endocrinology & metabolism (40/143) | 28 (0.6) | N | N |
Evidence-based Complementary and Alternative Medicine | 2.064 | Integrative & complementary medicine (10/27) | 27 (0.6) | Y | N |
QJM | 3.204 | Medicine, general & internal (30/154) | 27 (0.6) | N | N |
Journal of Clinical Psychiatry | 4.247 | Psychiatry (26/142), psychiatryh (19/142), psychology, clinicalh (11/127) | 26 (0.6) | N | N |
International Journal of Environmental Research and Public Health | 2.145 | Environmental sciences (116/241); public, environmental & occupational health (73/180); public, environmental & occupational healthh (44/156) | 25 (0.6) | Y | N |
aIF: impact factor.
bJCR: Journal Citation Reports.
cOAJ: open access journal.
dOAMJ: open access mega journal.
eConverted to an OAMJ in 2014.
fN/A: not available.
gNot listed in the Directory of Open Access Journals.
hSocial Sciences Edition.
Distribution of study types per article title across journal types.
Study type | Open access journal, n (%) | Open access mega journal, n (%) | ||
Yes (n=1566) | No (n=2907) | Yes (n=906) | No (n=276) | |
With ≥1 intervention | 552 (35.2) | 994 (34.2) | 308 (34.0) | 1238 (34.7) |
With ≥2 conditions | 537 (34.3) | 1038 (35.7) | 400 (44.2) | 1175 (32.9) |
With only 1 condition | 353 (22.5) | 684 (23.5) | 159 (17.5) | 878 (24.6) |
Others | 124 (7.9) | 191 (6.6) | 39 (4.3) | 276 (7.7) |
The visualization in
When a prolific author was defined as one who had at least 100 articles published between 2000 and 2017, a total of 8 authors were qualified as prolific.
Co-authorship networks during (A) 2000-2005, (B) 2006-2011, and (C) 2012-2017.
Distribution of study types per article title between studies authored by prolific authors and those authored by others.
Study type | Prolific author (≥100 articles), n (%) | |
Yes (n=1399) | No (n=3074) | |
With ≥1 intervention | 385 (27.5) | 1161 (37.8) |
With ≥2 conditions | 707 (50.5) | 868 (28.2) |
With only 1 condition | 250 (17.9) | 787 (25.6) |
Others | 57 (4.1) | 258 (8.4) |
This study yielded several interesting findings. First, a rapid growth in publications was observed from 2009 to 2015, just as it was between 2000 and 2009. However, the growth dramatically ceased after 2015. Second, certain medical conditions, such as diabetes, stroke, and dementia, and certain interventions such as statin therapy, metformin, and Chinese herbal medicine, received more attention from researchers using NHI claims data as the study material. Third, almost all the studies were published in JCR-indexed journals, most ranking as Q1 or Q2 in their corresponding JCR categories. OAMJs appeared to provide fertile soil for the rapid growth of research based on NHI claims data, particularly studies with ≥2 medical conditions in the article title. Fourth, while the top 8 most prolific authors contributed nearly one-third of all studies, they published more studies with ≥2 medical conditions in the article title than nonprolific authors. These studies mainly investigated the association between two medical conditions and might be easier to conduct than studies examining the effect of an intervention on a medical condition.
As described by Chen et al [
Diabetes, stroke, and dementia represented the most commonly investigated medical conditions. All these conditions are highly prevalent diseases that naturally attract more attention from researchers. Furthermore, their high prevalence enabled researchers to study these diseases using merely the Longitudinal Health Insurance Database, which is the 1 million–person subset of the NHIRD that entails a lower cost than the whole dataset of the NHIRD. In particular, because the diagnostic codes for diabetes and stroke have been validated within NHI claims data [
As for interventions, statins and metformin are commonly prescribed to patients with stroke and diabetes, respectively. Naturally, they were among the most frequently investigated interventions. In addition, the pleiotropic effects of statins and metformin might also intrigue researchers to test their effects on other diseases using large health care databases like the NHIRD [
Writing for publication is essential for academics. Currently, not only are academics evaluated against how well they publish but universities are also ranked according to their academic publication rates. The long-existing “publish or perish” culture of academia has now prevailed in Taiwan’s hospitals. Taiwan’s hospital accreditation system, in addition to assessing the quality of health care, also aims to determine the teaching status of a hospital [
In addition to these internal factors, the external environment is just suitable for catalyzing the growth of publications. The increasing availability of open access journals, in particular OAMJs, provides unprecedented capacity to accommodate a large volume of publications. Furthermore, several OAMJs (eg, PLOS ONE, Medicine, Scientific Reports, and BMJ Open) have decent impact factors and above-average JCR ranking (Q1 or Q2). All these factors have driven researchers to utilize secondary data analysis to augment their research output. Although the current system might have misdirected some hospital practitioners to “shallow research,” it has also encouraged positive involvement of practitioners in academic research.
Based on the text mining analysis, prolific authors tended to produce articles with ≥2 medical conditions in the title, while such articles were more likely to be published in OAMJs. From the pragmatic point of view, it is easier to investigate the association between two medical conditions than to study the effect of an intervention on a medical condition, in particular when the intervention, such as a medication, is time-dependent. Testing multiple hypotheses at the same time definitely increases the likelihood of finding an association [
Despite the negative impression, the following strategies were proposed to increase the impact of research based on NHI claims data. First, the percentage of first authors who are not Taiwanese citizens was very low (
This study has the following limitations. First, this study included only articles written in English and indexed in the PubMed database to make the results comparable with the study by Chen et al [
As Taiwan has recently become an aged society and is expected to become a super-aged society by 2025 [
Flowchart of included articles.
Unified Medical Language System (UMLS) semantic types for categorizing medical entities.
Journal Citation Reports.
medical subject heading.
National Health Insurance.
National Health Insurance Research Database.
open access journal.
open access mega journal.
Unified Medical Language System.
We would like to thank Ms. Li-Ying Sung for English language editing. This research was supported in part by the Ministry of Science and Technology (grant number MOST 107-2314-B-705-001) and the Center for Innovative Research on Aging Society from The Featured Areas Research Center Program within the framework of the Higher Education Sprout Project by the Ministry of Education in Taiwan.
None declared.