ORIGINAL ARTICLE
Association between metformin use and cervical cancer risk in women with diabetes in Japan: a retrospective cohort study
2026 年 8 巻 2 号 p. 46-53
詳細
2026 年 8 巻 2 号 p. 46-53
BACKGROUND
This study aimed to evaluate the association between metformin use and the risk of cervical cancer in women with diabetes compared to dipeptidyl peptidase-4 inhibitors (DPP-4is), using a new-user active-comparator design.
METHODS
We analyzed data from the JMDC claims database for new users of metformin or DPP-4i between 2010–2022 who were women. Propensity score overlap-weighting was applied to adjust for differences in age, complications, comorbidities, and other anti-diabetic medications. Kaplan–Meier curves and Cox proportional hazards models were used to compare cervical cancer incidence between the metformin and DPP-4i groups.
RESULTS
A total of 31,269 eligible individuals were identified, comprising 11,466 metformin users and 19,803 DPP-4i users. Cervical cancer occurred in 11 women (34.0/100,000 person-years) from the metformin group and 34 (59.0/100,000 person-years) in the DPP-4i group. Metformin use was associated with a lower risk of cervical cancer in our unadjusted Kaplan–Meier analysis (P = 0.058), and multivariable-adjusted Cox model (hazard ratio, 0.49; 95% CI, 0.22–1.09; P = 0.082), although neither difference was statistically significant.
CONCLUSIONS
Our findings did not show a statistically significant association between metformin use and cervical cancer incidence. However, this study’s limited sample size precluded definitive conclusions.
Cervical cancer is one of the most prevalent malignant tumors in women. According to the World Health Organization (WHO), an estimated 660,000 new cases and ~350,000 deaths related to this malignancy were reported globally in 20221). In Japan, the Ministry of Health, Labour and Welfare reports ~11,000 new cases and 2,900 deaths related to cervical cancer annually2). Persistent human papillomavirus (HPV) infection is the primary cause of cervical cancer. Effective prevention strategies include HPV vaccination, regular screening, and timely treatment of precancerous lesions.
Numerous studies have shown that diabetes is associated with an increased risk of various cancers3),4). Several studies have reported that type 2 diabetes is associated with an increased risk of gynecological cancers such as ovarian, cervical, vulvar, and endometrial5). Reducing cancer risks among individuals with diabetes is crucial, as malignancies remain the leading cause of death in individuals with type 2 diabetes6).
Among the medications available to treat type 2 diabetes, metformin is widely recommended as a first-line therapy in many major clinical guidelines across Europe and North America. By contrast, Japan does not specify any particular medication as a first-line treatment for diabetes7). Although several meta-analyses of randomized controlled trials have reported that metformin use does not improve overall survival8),9), a meta-analysis of observational studies suggested that metformin may reduce cancer risk. This discrepancy may be attributed to the immortal time bias inherent to observational studies. Another meta-analysis of studies examining the association between metformin use and mortality in patients with pancreatic cancer found no association when the immortal time bias was accounted for—whereas nine studies with this bias misleadingly indicated a protective effect10). Observational studies that use the active-comparator new-user design with appropriate adjustments for confounders are required to clarify whether metformin indeed has any anti-cancer effect11). However, to our knowledge, no published observational studies, thus far, have used this design to investigate the association between metformin use and cervical cancer.
In this context, the present study aimed to investigate the association between metformin use and the risk of cervical cancer among Japanese women with diabetes, when compared with the use of dipeptidyl peptidase-4 inhibitors (DPP-4is), which represent the most commonly prescribed treatments for type 2 diabetes in the country12).
This retrospective cohort study used the JMDC Claims Database (JMDC Inc., Tokyo, Japan). The database contains data from >60 insurers that compiles de-identified information on insured individuals and their claims—including hospitalizations, outpatient treatments, medication prescriptions, and dental care. It also contains health check-up data for ~30% of the Japanese population. Medication records are classified using the WHO’s Anatomical Therapeutic Chemical (ATC) codes, with all diagnoses coded according to the International Classification of Diseases, 10th Revision (ICD-10) system13). We used only definitive diagnoses and excluded suspected or rule-out diagnoses when defining the diseases.
The Institutional Review Board of the Graduate School of Medicine of The University of Tokyo (10862-(1)) approved the study’s protocol. Owing to the anonymous nature of the data, the requirement for informed consent was waived.
STUDY POPULATIONThis study included new users of either metformin (WHO-ATC code: A10BA02) or DPP-4is (WHO-ATC code: A10BH) among Japanese women with diabetes, since January 2010. Individuals were required to have ≥6 months of continuous enrollment prior to the first prescription, which was used as a lookback period. Those who were prescribed medications containing metformin were classified as metformin users, while those who were prescribed any of sitagliptin, vildagliptin, alogliptin, linagliptin, teneligliptin, anagliptin, saxagliptin, trelagliptin, or omarigliptin were categorized as DPP-4i users. Each patient’s first prescription date served as their index date.
We applied the following exclusion criteria: (i) individuals who were prescribed both metformin and DPP-4is in their index month; (ii) those with a history of any cancer or who received kidney replacement therapy during their lookback period; (iii) those who were prescribed buformin or imeglimin during their lookback period; and (iv) those with follow-up periods of <6 months.
VARIABLESIn addition to each participant’s age (in years), their year of birth was also considered, to account for the probability of vaccination against HPV. According to the Statistics Bureau of Japan’s Ministry of Health, Labour and Welfare, HPV vaccination rates among women born between 1994–1999 are significantly higher compared to other birth cohorts14), potentially affecting the incidence of cervical cancer. Considering this, the participants were classified into two groups: those born before 1994 or after 1999 (the unvaccinated generation), and those born between 1994–1999 (the vaccinated generation).
We collected data concerning diabetes-related complications, major comorbidities, and other anti-diabetic medications (other than metformin and DPP-4i) for each patient over their 6-month lookback period. Each participant’s Charlson comorbidity index was also included as a covariable.
Diabetes-related complications and major comorbidities were defined using ICD-10 codes, including diabetic nephropathy (E102, E112, E122, E132, and E142), diabetic retinopathy (E103, E113, E123, E133, and E143), and diabetic neuropathy (E104, E114, E124, E134, and E144)15), hypertension (I10–15)16), stroke (I60–63)16), ischemic heart disease (I20–25)16), chronic obstructive pulmonary disease (J41, J42, J43, J440, J441, and J449)17), and obesity (E66)18). Dyslipidemia was defined as at least one prescription for any drugs with WHO-ATC codes beginning with C1018).
Anti-diabetic medications were identified using ingredient names or WHO-ATC codes. These included thiazolidinedione, sulfonylurea, glinides, alpha-glucosidase inhibitors, sodium-glucose co-transporter-2 inhibitors, glucagon-like peptide-1 receptor agonists, and insulin. The ICD-10 or WHO-ATC codes used in this study are summarized in Supplementary Table 1.
OUTCOMEThe outcome of this study was an initial diagnosis of cervical cancer that occurred ≥6 months after the first prescription date, as defined by the ICD-10 code C5319). The observation period for each individual began on the first prescription date and ended either when cervical cancer was diagnosed, the individual died, or the observation period ended.
STATISTICAL ANALYSISDescriptive statistics are reported as means and numbers (percentages).
To balance the baseline characteristics between the two groups, we employed overlap weighting using propensity scores, which has distinct advantages over propensity score matching and inverse probability of treatment weighting20),21). This method achieves improved covariate balance by reducing standardized mean differences to zero and retains all individuals, unlike propensity score matching, which excludes some participants20). In terms of bias reduction and estimation efficiency, overlap weighting has been shown to outperform inverse probability of treatment weighting20),21), while also avoiding extreme weights (e.g., >10 or <0.10), which often require trimming21). The propensity score was calculated using logistic regression analysis with metformin prescription as the dependent variable. The independent variables included age, whether the participant belonged to the HPV-vaccinated generation, diabetes-related complications, major comorbidities, anti-diabetic medications, and Charlson comorbidity index. Overlap weights were defined as one minus the propensity score for those who had been prescribed metformin, and as the propensity score values for those who were prescribed DPP-4is. We examined whether the baseline characteristics of the two groups were balanced after weighting by calculating standardized mean differences, with absolute standardized mean differences of <10% indicating negligible imbalance22).
Survival analyses were conducted using Kaplan–Meier curves and Cox regression to assess whether metformin use was associated with a risk of cervical cancer. The possible association was evaluated using the log-rank test in the Kaplan–Meier curves, and via hazard ratios and their 95% confidence intervals (CIs) in the Cox regression analyses. These analyses were performed with and without overlap-weighting. For the weighted analysis, we applied robust variance estimators to compute 95% CIs, consistent with the methods used for weighted analyses23).
This study adopted a similar design to the intention-to-treat approach used in randomized controlled trials. We employed the active comparator new-user design because this method is well-suited for evaluating the effects of drugs in observational studies, as it reduces confounding by indication and improves comparability between groups11). After categorizing each individual based on their initial drug prescription, subsequent changes in medication were not considered.
We also conducted a sensitivity analysis of a per-protocol analysis as an alternative to the intention-to-treat approach. In the per-protocol analysis, a switch to the other drug (e.g., a DPP-4i prescription in the metformin group) or termination of the drug (for gaps >6 months) was treated as censoring, in addition to the censoring events considered in the intention-to-treat analysis.
Statistical significance was set at P < 0.05. All statistical analyses were performed using Stata version 18 (StataCorp, College Station, TX, USA).
After applying our inclusion and exclusion criteria, we used data from 42,544 individuals for further screening (Fig. 1). A total of 31,269 individuals from this initial cohort were considered eligible for the final analysis, comprising 11,466 (36.7%) in the metformin group and 19,803 (63.3%) in the DPP-4i group.
Table 1 presents the baseline characteristics of the study participants. Before the adjustment, those in the metformin group were younger on average (mean age, 49.1 vs 55.1 years); more likely to have hypertension, dyslipidemia, ischemic heart disease, and higher Charlson comorbidity indexes; more likely to have prescriptions for sulfonylureas and alpha-glucosidase inhibitors; and less likely to have prescriptions for glucagon-like peptide-1 receptor agonists and insulin.
| Unadjusted | Propensity-score overlap weighted | |||||
|---|---|---|---|---|---|---|
| Metformin (n = 11,466) |
DPP-4i (n = 19,803) |
SMD | Metformin (n = 6654) |
DPP-4i (n = 6654) |
SMD | |
| Age, years, mean | 49.1 | 55.1 | –54.7% | 51.9 | 51.9 | 0.0% |
| HPV-vaccinated generation, n (%) | 144 (1.3) | 79 (0.4) | 9.5% | 0.8 | 0.8 | 0.0% |
| Diabetes complications, n (%) | ||||||
| Nephropathy | 634 (5.5) | 886 (4.5) | 4.8% | 5.1 | 5.1 | 0.0% |
| Retinopathy | 987 (8.6) | 1496 (7.6) | 3.9% | 8.1 | 8.1 | 0.0% |
| Neuropathy | 180 (1.6) | 276 (1.4) | 1.5% | 1.4 | 1.4 | 0.0% |
| Major comorbidities, n (%) | ||||||
| Hypertension | 4203 (36.7) | 8895 (44.9) | –16.9% | 40.7 | 40.7 | 0.0% |
| Dyslipidemia | 3178 (27.7) | 7254 (36.6) | –19.2% | 31.5 | 31.5 | 0.0% |
| Stroke | 224 (2.0) | 614 (3.1) | –7.3% | 2.3 | 2.3 | 0.0% |
| Ischemic heart disease | 476 (4.2) | 1294 (6.5) | –10.6% | 5.0 | 5.0 | 0.0% |
| Chronic obstructive pulmonary disease | 263 (2.3) | 510 (2.6) | –1.8% | 2.4 | 2.4 | 0.0% |
| Obesity | 445 (3.9) | 413 (2.1) | 10.6% | 2.9 | 2.9 | 0.0% |
| Anti-diabetic medications, n (%) | ||||||
| TZD | 278 (2.4) | 775 (3.9) | –8.5% | 2.9 | 2.9 | 0.0% |
| SU | 543 (4.7) | 1556 (7.9) | –12.9% | 5.7 | 5.7 | 0.0% |
| Glinides | 145 (1.3) | 366 (1.8) | –4.7% | 1.5 | 1.5 | 0.0% |
| α-GI | 524 (4.6) | 1398 (7.1) | –10.7% | 5.4 | 5.4 | 0.0% |
| SGLT2i | 901 (7.9) | 1173 (5.9) | 7.6% | 7.1 | 7.1 | 0.0% |
| GLP-1RA | 226 (2.0) | 78 (0.4) | 14.6% | 0.8 | 0.8 | 0.0% |
| Insulin | 968 (8.4) | 1044 (5.3) | 12.6% | 6.7 | 6.7 | 0.0% |
| Charlson comorbidity index, mean | 2.2 | 2.4 | –10.0% | 2.3 | 2.3 | 0.0% |
Note: The proportions of patients with comorbid conditions and types of medications in the propensity-score overlap-weighted population are not shown because the results were weighted in fractional counts.
DPP-4i, dipeptidyl peptidase-4 inhibitor; TZD, thiazolidinedione; SU, sulfonylurea; α-GI, alpha-glucosidase inhibitor; SGLT2i, sodium-glucose co-transporter-2 inhibitor; GLP-1RA, glucagon-like peptide-1 receptor agonist; SMD, standardized mean difference.
We obtained data for 970 (interquartile range, 512–1639) and 1000 (interquartile range, 543–1700) follow-up days in the metformin and DPP-4i groups, respectively. In the metformin and DPP-4i groups, over follow-up periods of 32,341 and 57,668 person-years, respectively, we noted that 11 (34.0/100,000 person-years) and 34 (59.0/100,000 person-years) individuals developed cervical cancer, respectively. Over the follow-up period, we observed death before cancer development in 0.26% of the patients in the metformin group, and 0.60% in the DPP-4i one. In our propensity-score overlap-weighted model, cervical cancer developed at a rate of 31.2/100,000 person-years in the metformin group, and 61.8/100,000 person-years in the DPP-4i group. The patients who developed cervical cancer over the follow-up period were younger and had a higher median Charlson comorbidity index score compared with those who did not (Supplementary Table 2).
Fig. 2 shows the Kaplan–Meier curves for the incidence of cervical cancer among individuals in metformin and DPP-4i groups, in the multivariable-adjusted models. The log-rank test did not reveal a statistically significant result (P = 0.058).
Table 2 shows the incidence of cervical cancer in the Cox regression analyses. In both unadjusted and adjusted analyses, the metformin group did not show a statistically significant reduction in risk compared with the DPP-4i group. The hazard ratio for the metformin group vs the DPP-4i group was 0.58 (95% CI: 0.27–1.24; P = 0.163) in the unadjusted analysis, and 0.49 (95% CI: 0.22–1.09; P = 0.082) in the adjusted analysis.
| HR (95% CI) | P value | |
|---|---|---|
| Unadjusted | 0.58 (0.27–1.24) | 0.163 |
| Adjusted | 0.49 (0.22–1.09) | 0.082 |
| Unadjusted (Sensitivity analysis) | 0.55 (0.18–1.65) | 0.285 |
| Adjusted (Sensitivity analysis) | 0.43 (0.13–1.39) | 0.160 |
Hazard ratios of metformin with reference to DPP 4i
HR, hazard ratio; CI, confidence interval
The results of the sensitivity analysis were similar to those of the main analysis (Fig. 3 and Table 2). In both unadjusted and adjusted Cox regression analyses, the metformin group did not show a statistically significant reduction in risk compared with the DPP-4i group. The hazard ratio for the metformin group vs the DPP-4i group was 0.55 (95% CI: 0.18–1.65; P = 0.285) in the unadjusted analysis, and 0.43 (95% CI: 0.13–1.39; P = 0.160) in the adjusted analysis.
In this analysis of data concerning women with diabetes from a Japanese administrative claims database spanning 2010–2022, using a new-user active-comparator design, we found a statistically insignificant association between metformin use and cervical cancer incidence. Further studies with larger sample sizes are warranted to clarify this association.
Current evidence regarding the anti-cancer effects of metformin in clinical settings is scarce. For example, most anti-cancer effects related to metformin have been highlighted by observational studies, which may have been influenced by the immortal time bias11). Randomized controlled trials have failed to confirm the efficacy of the anti-cancer effects of metformin in terms of suppressing cancer development24),25) or improving overall survival8),26). Although a recent observational study suggested a possible anti-cancer effect of metformin on cervical cancer in women with incident diabetes, that study did not adopt a new-user or active-comparator design19).
In animal experiments, metformin has demonstrated anti-cancer effects both in vitro and in vivo27). Research indicates that metformin may reduce insulin-mediated tumor growth by lowering systemic blood glucose and insulin levels. Metformin also activates adenosine monophosphate-activated protein kinase and inhibits the mechanistic/mammalian target of rapamycin complex 1 signaling pathway, thereby suppressing cell growth and proliferation. It may exert its anti-cancer effects through multiple mechanisms, such as activating hypoxia-inducible factor-1α and inhibiting nuclear factor kappa B. These experimental results suggest that metformin may suppress the development of certain cancer types.
The main strengths of this study related to how its active-comparator design preserved the comparability between the two participant groups, and the removal of the potential effects of the immortal time bias through the integration of the new-user design. The integration of the intention-to-treat principle in addition to the new-user active-comparator design represents one of the most effective strategies for addressing the issue of immortal time bias in observational studies28). Second, we adjusted for confounders using propensity-score overlap-weighting, which reduced differences in the distributions of covariables between the two groups to almost zero.
However, this study was also subject to several limitations worth noting. First, owing to the limited sample size, we were only able to analyze 45 cases of cervical cancer, which may have led to insignificant results. Second, this study did not use biomarkers to assess blood glucose control and body mass index—both factors that may also have affected the choice of prescribed drugs and cervical cancer risk. We also did not consider information related to metformin dosages, durations of treatment regimens, or cancer stages. Consequently, the potential impact of different medication doses or cancer stages could not be assessed. Finally, HPV infection represents the main causative factor of cervical cancer; however, we could only classify the participant population by birth year based on the time points when the HPV vaccination rates in the overall Japanese population increased or decreased. Unfortunately, information concerning specific vaccination statuses on the individual level could not be obtained from the database, which may also have affected the results.
This study used a new-user, active comparator design to examine the association between metformin use and cervical cancer risk among Japanese women with diabetes. Our findings did not reveal a statistically significant association. Further research is warranted to confirm these results in larger population-based studies.
Akira Okada is a member of the Department of Prevention of Diabetes and Lifestyle-Related Diseases, which is a cooperative program between the University of Tokyo and the Asahi Mutual Life Insurance Company.
This work was supported by a grant from the Ministry of Health, Labour and Welfare of Japan (grant no.: 23AA2003).
None.
Hui Yuan and Akira conceived and designed the study, performed the statistical analysis, and wrote, edited, and reviewed the manuscript. Hideo Yasunaga contributed to the discussion and interpretation of the data, and edited and reviewed the manuscript. All authors have approved the final manuscript for publication.
Akira Okada and Hideo Yasunaga are the Editorial Board members of Annals of Clinical Epidemiology. They were not involved in the peer-review or decision-making process for this paper.
