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 JMIR Public Health and Surveillance, is properly cited. The complete bibliographic information, a link to the original publication on http://publichealth.jmir.org, as well as this copyright and license information must be included.
Existing epidemiological evidence regarding the association between the long-term use of drugs and cancer risk remains controversial.
We aimed to have a comprehensive view of the cancer risk of the long-term use of drugs.
A nationwide population-based, nested, case-control study was conducted within the National Health Insurance Research Database sample cohort of 1999 to 2013 in Taiwan. We identified cases in adults aged 20 years and older who were receiving treatment for at least two months before the index date. We randomly selected control patients from the patients without a cancer diagnosis during the 15 years (1999-2013) of the study period. Case and control patients were matched 1:4 based on age, sex, and visit date. Conditional logistic regression was used to estimate the association between drug exposure and cancer risk by adjusting potential confounders such as drugs and comorbidities.
There were 79,245 cancer cases and 316,980 matched controls included in this study. Of the 45,368 associations, there were 2419, 1302, 662, and 366 associations found statistically significant at a level of
Our web-based system provides an overview of comprehensive quantified data of drug-cancer associations. With all the quantified data visualized, the system is expected to facilitate further research on cancer risk and prevention, potentially serving as a stepping-stone to consulting and exploring associations between the long-term use of drugs and cancer risk.
In recent decades, the prevalence of chronic medical conditions such as arthritis, osteoporosis, diabetes, hypertension, and cardiovascular disease has increased [
There has been a significant rise in the number of published studies in which commonly used medications were found to increase or decrease the risk of cancer [
Big data approaches seem to offer an immense opportunity to generate strong evidence for taking insightful clinical and public health action [
This study is part of a larger project aimed at assessing the effect of most common medications on 20 cancer sites using a population-based nested case-control design. The National Health Insurance Research Database (NHIRD) safeguards the privacy and confidentiality of all beneficiaries and transfers health insurance data to health researchers after ethical approval has been obtained. In this analysis, access to the NHIRD was approved by the Taipei Medical University Joint Institutional Review Board.
This case-control study was carried out using records from the Taiwan NHIRD, which was established in 1995 and has collects all claims of beneficiaries under the National Health Insurance (NHI) program. The program covers more than 99% of the total population (a total of 23,430,000) and has contracted with 97% of the hospitals and clinics in Taiwan. The NHIRD comprised claims data of 2,000,000 individuals randomly selected from all insured enrollees. This sample represents the original medical claims for all residents in Taiwan covered under the NHI program. The database included specific data on medications prescribed, laboratory and diagnostic test data, dates of visits, lengths of hospitalization, and diagnoses. Diagnoses were coded according to the International Classification of Disease, Ninth Revision, Clinical Modification (ICD-9-CM). Drugs were coded based on the World Health Organization Anatomical Therapeutic Chemical (ATC) classifications (WHO Collaborating Centre for Drug Statistics Methodology ATC/DDD Index). The database used in this study can be interlinked by the scrambled, unique, individual personal identification number.
We identified cases in adults who were aged 20 years or older and had received treatment at least two months before the index date. The index date was defined as the date of a cancer diagnosis. We used the ICD-9-CM to identify patients with cancer as cases. Among the NHIRD cases, eligibility criteria for case patients were (1) registration as patients with cancer in the catastrophic illness file, (2) diagnosis of primary cancer in inpatient admission, (3) treatment with any cancer drug from outpatient visits or inpatient admission, (4) a cancer-specific procedure from outpatient visits or inpatient admission, and (5) more than 4 cancer-specific examinations or more than 1 cancer-related procedure (radiotherapy, chemotherapy, or treatment tracking of cancer) from outpatient visits or inpatient admission.
We randomly selected control patients from patients without a cancer diagnosis during the 15 years (1999-2013) of the study period. Case and control patients were matched 1:4 based on age, sex, and visit date (
Workflow of the case-control study design. NHI: National Health Insurance.
We quantified the risks of common cancers in Taiwan, comparing patients treated with (1) antihypertensives; (2) antihyperlipidemics; (3) antidiabetics; (4) antihyperuricemics; (5) anxiolytics, hypnotics, and sedatives; or (6) NSAIDs against those not prescribed any of these medications. We investigated the following cancers using their corresponding ICD-9-CM codes: oral cancer (140-149.xx, excluding 142.xx and 147.xx), esophageal cancer (150.xx), gastric cancer (151.xx), colorectal cancer (153.xx, rectum 154.xx), liver cancer (155.xx), pancreatic cancer (157.xx), lung cancer (162.xx), skin cancer (172-173.xx), female breast cancer (174.xx), cervical cancer (180.xx), endometrial cancer (182.xx), ovarian cancer (183.xx), prostate cancer (185.xx), bladder cancer (188.xx), kidney cancer (189.xx), brain cancer (191.xx), thyroid cancer (193.xx), non-Hodgkin disease (200.xx, 202.xx, 203.xx), leukemia (204-208.xx), and all cancers (140-208.xx).
We defined the index date as the date of a cancer diagnosis. The drug exposure was analyzed only before the index date, and we defined drug users as those who filled prescriptions of at least 60 days during admissions and outpatient visits within the 2 years before the index date (
Drugs exposure.
Comorbidities and medications identified as confounders were adjusted in this study. Comorbidities were defined using the Charlson Comorbidity Index [
The McNamara test and paired
Study variables.
Variables | Type | Descriptive | Statistical model | Definition |
ID | Nominal | N/Aa | No | N/A |
Age | Ordinal | N/A | No | N/A |
Sex | Binomial | 1: Yes, 0: No | No | N/A |
Study drug (exposure) | Binomial | 1: Yes, 0: No | Yes | Independent |
Cancer (outcome) | Binomial | 1: Yes, 0: No | Yes | Dependent |
Myocardial infarction | Binomial | 1: Yes, 0: No | Yes | Confounding |
Congestive heart failure | Binomial | 1: Yes, 0: No | Yes | Confounding |
Peripheral vascular disease | Binomial | 1: Yes, 0: No | Yes | Confounding |
Cerebrovascular disease | Binomial | 1: Yes, 0: No | Yes | Confounding |
Dementia | Binomial | 1: Yes, 0: No | Yes | Confounding |
Chronic pulmonary disease | Binomial | 1: Yes, 0: No | Yes | Confounding |
Rheumatic disease | Binomial | 1: Yes, 0: No | Yes | Confounding |
Peptic ulcer disease | Binomial | 1: Yes, 0: No | Yes | Confounding |
Liver disease (mild, moderate, and severe) | Binomial | 1: Yes, 0: No | Yes | Confounding |
Diabetes (with or without chronic complication) | Binomial | 1: Yes, 0: No | Yes | Confounding |
Hemiplegia or paraplegia | Binomial | 1: Yes, 0: No | Yes | Confounding |
Renal disease | Binomial | 1: Yes, 0: No | Yes | Confounding |
CCIb scores | Ordinal | N/A | Yes | Confounding |
Metformin | Binomial | 1: Yes, 0: No | Yes | Confounding |
Aspirin | Binomial | 1: Yes, 0: No | Yes | Confounding |
Statin | Binomial | 1: Yes, 0: No | Yes | Confounding |
Matching number (case match control) | Nominal | N/A | Yes | Stratified |
aN/A: not applicable.
bCharlson Comorbidity Index.
Data analysis and results were performed using SAS software (version 9.4; SAS Institute) [
After analyzing the associations between the long-term use of drugs and cancer risk, we built a web-based system to include all associations [
We identified 79,245 participants newly diagnosed with cancer between 2002 and 2013 from 2 million people (
Baseline characteristics of case patients and control patients.
Characteristics | Case patients (with cancer) (n=79,245) | Control patients (without cancer) (n=316,980) |
|
||||
|
|||||||
|
|
||||||
Age (years), mean (SD) | 59.20 (15.23) | 59.21 (15.24) | N/Aa |
|
|||
20-39, n (%) | 8,292 (10.5) | 33,168 (10.5) | N/A |
|
|||
40-64, n (%) | 40,504 (51.1) | 162,040 (51.1) | N/A |
|
|||
≥65, n (%) | 30,449 (38.4) | 121,772 (38.4) | N/A |
|
|||
|
|
||||||
Male | 40,259 (50.8) | 161,036 (50.8) | N/A |
|
|||
Female | 38,986 (49.2) | 155,944 (49.2) | N/A |
|
|||
|
|
||||||
Myocardial infarction | 372 (0.47) | 1729 (0.55) | <.0001 |
|
|||
Congestive heart failure | 2557 (3.23) | 9344 (2.95) | <.0001 |
|
|||
Peripheral vascular disease | 1143 (1.44) | 4187 (1.32) | <.0001 |
|
|||
Cerebrovascular disease | 5306 (6.70) | 22,609 (7.13) | <.0001 |
|
|||
Dementia | 1205 (1.52) | 5522 (1.74) | <.0001 |
|
|||
Chronic pulmonary disease | 5951 (7.51) | 20,423 (6.44) | <.0001 |
|
|||
Rheumatic disease | 1138 (1.44) | 3901 (1.23) | <.0001 |
|
|||
Peptic ulcer disease | 12,760 (16.10) | 34,283 (10.82) | <.0001 |
|
|||
Liver disease | 11,671 (14.73) | 20,647 (6.51) | <.0001 |
|
|||
Diabetes | 9143 (11.54) | 32,532 (10.26) | <.0001 |
|
|||
Hemiplegia or paraplegia | 490 (0.62) | 2300 (0.73) | <.0001 |
|
|||
Renal disease | 2793 (3.52) | 7453 (2.35) | <.0001 |
|
|||
|
|
||||||
Metformin | 8236 (10.39) | 33,375 (10.53) | .07 |
|
|||
Aspirin | 9826 (12.40) | 41,726 (13.16) | <.0001 |
|
|||
Statin | 8336 (10.52) | 37,395 (11.80) | <.0001 |
|
aN/A: not applicable.
We successfully developed a web-based system [
Drugs were categorized into 6 groups: (1) antihypertensives; (2) antihyperlipidemics; (3) antidiabetics; (4) antihyperuricemics; (5) NSAIDs; and (6) anxiolytics, hypnotics, and sedatives. As exemplified in
Display of drug and cancer risk.
Associations between different drugs and cancers among different age groups.
Drug (ATCa code), cancer type, and age (years) | Adjusted odd ratio (95% CI) | Case patients, n | Control patients, n | |||||||||
|
|
Exposure | Nonexposure | Exposure | Nonexposure | |||||||
|
|
|
|
|
|
|||||||
|
|
|
|
|
|
|
||||||
|
|
≥65 | 0.954** (0.923-0.985) | 6964 | 23,485 | 28,665 | 93,131 | |||||
40-64 | 0.871*** (0.831-0.913) | 2825 | 37,679 | 12,925 | 149,091 | |||||||
20-39 | 1.000 (0.679-1.472) | 37 | 8255 | 136 | 33,032 | |||||||
≥20 | 0.924*** (0.900-0.949) | 9826 | 69,419 | 41,726 | 275,254 | |||||||
|
|
|
|
|
|
|||||||
|
|
|
|
|
|
|
||||||
|
|
≥65 | 0.881* (0.794-0.979) | 797 | 4532 | 3231 | 18,085 | |||||
40-64 | 0.799*** (0.701-0.912) | 456 | 4757 | 2076 | 18,776 | |||||||
20-39 | 0.448 (0.148-1.358) | 4 | 776 | 28 | 3092 | |||||||
≥20 | 0.845*** (0.779-0.916) | 1257 | 10,065 | 5335 | 39,953 | |||||||
|
|
|
|
|
|
|||||||
|
|
|
|
|
|
|
||||||
|
|
≥65 | 1.901* (1.125-3.213) | 28 | 632 | 50 | 2590 | |||||
40-64 | 2.303* (1.109-4.781) | 16 | 565 | 27 | 2297 | |||||||
20-39 | N/Ab | 0 | 91 | 0 | 364 | |||||||
≥20 | 1.981** (1.298-3.024) | 44 | 1288 | 77 | 5251 | |||||||
|
|
|
|
|
|
|||||||
|
|
|
|
|
|
|
||||||
|
|
≥65 | 1.090 (0.787-1.508) | 78 | 201 | 256 | 860 | |||||
40-64 | 1.456** (1.112-1.905) | 111 | 379 | 287 | 1673 | |||||||
20-39 | 2.409** (1.364-4.257) | 24 | 201 | 46 | 854 | |||||||
≥20 | 1.379** (1.138-1.670) | 213 | 781 | 589 | 3387 |
aATC: Anatomical Therapeutic Chemical.
bN/A: not applicable.
*
**
***
The web-based system provided an overview of associations between cancers and medications sorted by age, gender,
Green colors symbolize a significant association between a drug and a cancer with an AOR less than 1. The darker a green color is, the farther from 1 the AOR is. Red colors symbolize a significant association between a drug and a cancer with an AOR greater than 1. The darker a red color is, the farther from 1 the AOR is. White indicates no significant association between a cancer and a drug.
Display of overview.
In this nationwide longitudinal retrospective study, we evaluated 79,245 patients with cancer and 316,980 control patients matched by variables including age, sex, and visit date at a 1:4 ratio from the NHIRD in Taiwan, including the follow-up data from 2001 to 2013 of 2 million individuals aged 20 years or older. This web-based system aimed to provide information of medication-cancer associations for users (researchers) to choose potentially clinically relevant ones for further studies (eg, a meta-analysis) and offered a filter by
The long-term use of some drugs was associated with increased risk of certain cancers, such as sitagliptin with pancreatic cancer and benzodiazepines (BZDs) with brain cancer. For example, patients aged 40 to 64 years and 65 years or older treated with sitagliptin had a high risk for pancreatic cancer, but there was not sufficient information for us to estimate such risk among patients aged 20 to 39 years. On the contrary, those aged 20 to 39 years receiving BZDs had a higher risk of brain cancer (AOR 2.409, 95% CI 1.364-4.257;
Despite mechanisms between the long-term use of drugs and cancer risk remaining not well understood, our findings were consistent with possible mechanisms proposed in previous studies. Aspirin, metformin, and statins are examples of this. According to previous studies, aspirin reduces prostaglandin generation, which is associated with decreased cellular proliferation, by inhibiting cyclooxygenase isozymes [
Additionally, sitagliptin has been suggested to have an association with elevated risk of pancreatitis and pancreatic cancer [
Despite the immense investment in anticancer therapy, cancer remains the leading cause of death globally [
Aspirin is widely used to treat fever and mild pain, but its long-term use may prevent development of squamous cell carcinoma [
Strengths of this study include the retrospective study design, long-term follow-up, proper identification of case and control patients, and measurement of the magnitude of association between 6 commonly used groups of medications and cancer risks. Furthermore, confounding factors were appropriately adjusted to reduce the study bias.
We also acknowledge that our research has limitations that need to be addressed. First, drug adherence, self-payment, laboratory data, and lifestyles characteristics such as body mass index, smoking, and family history of cancer were unavailable in the NHIRD. Second, other risk factors for cancer, such as phenotype, genotype, and exposure type, might have influenced the results. Although we applied the match method and adjustment for numerous covariates to control confounding factors, it was impossible to eliminate all confounding factors, particularly indications. Third, all data were collected from the Taiwan NHIRD, and hence, the study population limited the generalization of the results to other countries with different ethnic distribution. Fourth, the results showed associations between the long-term use of drugs and cancer risk but not causation.
Moreover, we did not set a threshold for statistical significance at 0.05/45,368 ≈ 1.10 × 10–6 for multiple testing correction, given the large number of statistical tests and the highly selected patients—patients with cancer and long-term users of medications instead of the general population. Had we set the significance level at 1.10 × 10–6, there would not have been enough significant associations to be useful or practical to users. Therefore, we offered in the web-based system a filter by
This comprehensive retrospective study not only provides an overview of associations of cancer risk with 6 commonly prescribed groups of medications but also helps to narrow the gap in the currently insufficient research on the long-term safety of these medications. With all the quantified data visualized, the system is expected to further facilitate research on cancer risk and prevention. Since our findings have proposed only associations between cancers and long-term use of medications, further clinical trials and meta-analyses are required to assess and confirm their causality. This web-based system could potentially serve as a stepping-stone to exploring and consulting associations between long-term use of drugs and cancer risk.
Supplementary table.
angiotensin-converting enzyme inhibitors
adenosine monophosphate–activated protein kinase
adjusted odds ratio
angiotensin II antagonist
Anatomical Therapeutic Chemical
benzodiazepine
3-hydroxy-3-methyl-glutaryl coenzyme A
International Classification of Disease, Ninth Revision, Clinical Modification
National Health Insurance
National Health Insurance Research Database
nonsteroidal anti-inflammatory drug
Hypertext Preprocessor
This research is sponsored in part by the Ministry of Science and Technology (grant number: MOST 109-2222-E-038-002-MY2), the Ministry of Education (grant number: MOE 109-6604-001-400), and Taipei Medical University (grant number: TMU107-AE1-B18).
None declared.