The objective of this case-control study was to identify any association of metformin intake with the survival of patients with ovarian cancer.
The objective of this case-control study was to identify any association of metformin intake with the survival of patients with ovarian cancer.
In this retrospective case-control study, women with ovarian cancer who received metformin (cases) were compared with women with ovarian cancer who did not receive metformin (controls). A 2-layered analysis was conducted. In preliminary analysis, all cases (the OC cohort) were compared with controls at a 1:2 ratio. Subsequently, in definitive analysis, only patients who had epithelial ovarian cancer (the EOC cohort) were compared with controls at a 1:3 ratio. In the EOC cohort, cases were matched with controls for age (±5 years), International Federation of Gynecology and Obstetrics stage, and residual disease. Prognostic variables and disease specific survival were compared using chi-square tests, the Kaplan-Meier (log-rank) method, and Cox proportional hazards analysis.
In a preliminary analysis of the OC cohort (72 cases and 143 controls), cases had better survival (5-year disease-specific survival for cases vs controls, 73% vs 44%; P = .0002). In the definitive analysis of the EOC cohort (61 cases and 178 controls), the distribution of age, disease stage, optimal cytoreduction, serous histology, and platinum chemotherapy remained similar between cases and controls (P > .05). Despite these similarities, cases had significantly better survival (5-year disease-specific survival for cases vs controls, 67% vs 47%; P = .007). On multivariate analysis, metformin remained an independent predictor of survival (hazard ratio, 2.2; 95% confidence interval, 1.2-3.8; P = .007) after controlling for disease stage, grade, histology, chemotherapy, body mass index, and surgical cytoreduction.
The results of this study indicated an association of metformin intake with survival in patients with ovarian cancer. The receipt of metformin was associated with better survival, and the authors concluded that metformin is worthy of clinical trials in ovarian cancer. Cancer 2013. © 2012 American Cancer Society.
The case fatality ratio of ovarian cancer is extremely high.1 This is because 75% of patients have extensive stage disease (stage >III) at diagnosis, and the currently available therapies in up-front and recurrent ovarian cancer have limited efficacy.2 Consequently, there is a great need to develop improved up-front and salvage therapies for ovarian cancer.2
Novel therapies for ovarian cancer can originate by designing and discovering new drugs de novo. Although promising, this approach carries the limitation of being tremendously expensive and very time consuming. In fact, it is estimated that de novo drug discovery may take as long as 10 to 17 years and $800 million (US) per agent.3 Therefore, alternative means of finding novel therapies need to be explored. One such avenue is drug repositioning, which is an approach in which already existing drugs are used to treat a different and new disease. This approach carries the promise of accelerating the translation of laboratory research to the clinical practice in a shorter time frame.4 Numerous examples of successful drug repositioning exist, one of which is the US food and Drug Administration approval of a morning sickness drug, lenalidomide, for multiple myeloma.5
Originating from the French Lilac plant (Galega officinalis), metformin is one of the most commonly used medications for diabetes worldwide. We previously demonstrated that metformin has anticancer effects in ovarian cancer both in vitro6 and in vivo.7 Similar observations from others have suggested the anticancer effects of metformin in neoplasms of the prostate,8 colon,9 pancreas,10 and brain.11 In breast cancer, a large, phase 3, randomized clinical trial is already underway to test metformin in >200 oncology centers (National Clinical Trial identifier NCT01101438). However, to date, only a small study has assessed the association of metformin with the prognosis of actual patients with ovarian cancer.12 The objective of the current study was to investigate whether there is an association between metformin and prognosis in patients with ovarian cancer.
This study was approved by the Institutional Review Board of the Mayo Clinic (Rochester, Minn). Consecutive patients with a diagnosis of ovarian cancer who received metformin were identified by inpatient hospital records (metformin given) as well as outpatient pharmacy prescription records (prescription filled) of the Mayo Clinic in Rochester between 1995 and 2010. Diagnosis codes according to the International Classification of Diseases for Oncology, third revision, were used for patient identification. The mean duration of metformin intake was 2.3 years (range, 1-11 years), and the minimum length of metformin use included in the study was 1 year. The receipt of metformin, as captured by our hospital records, was at or after the cancer diagnosis. Metformin was prescribed at doses from 500 mg twice daily to 1 g twice daily. The duration of metformin use before cancer diagnosis was unknown because patient records before the cancer diagnosis were not available. The reason patients received metformin was the presence of diabetes in the cases.
In this report, patients with ovarian cancer who received metformin are referred to as cases, and patients with ovarian cancer without diabetes who did not receive metformin are referred to as controls (nondiabetic controls). Patients with ovarian cancer with diabetes who received insulin or other antidiabetic medications (but not metformin) are referred to as diabetic controls.
Body mass index (BMI) was calculated as the weight in kilograms/height in meters squared (kg/m2). Stage of disease was assigned in accordance with International Federation of Gynecology and Obstetrics (FIGO) surgical staging criteria. For the purposes of comparison, early or low-stage disease was classified as FIGO stage I and II, and late or advanced-stage disease was classified as FIGO stage III and IV. For statistical brevity without losing biologic significance; the histologic grade was categorized as high grade or low grade. Surgical cytoreduction was classified as optimal if the greatest dimension of residual disease was ≤1 cm or suboptimal if it was >1 cm. Patients who had no evidence of gross residual disease were classified as completely cytoreduced. Comparison of histology was completed using each distinct histologic group as well as nonserous versus serous histologies. The type of chemotherapy, the number of cycles, and the number of agents received were recorded for each patient. In addition, platinum chemotherapy was compared with nonplatinum chemotherapy. Disease-specific survival (DSS) also was recorded. None of the patients had breast cancer susceptibility gene (BRCA) mutations, and none received neoadjuvant chemotherapy. Patients who had a platinum-free interval of ≥12 months were deemed to have platinum-sensitive disease.
In total, 72 women were identified who had ovarian cancer and had received metformin. A 2-layered analysis was conducted. In the initial, preliminary analysis, all 72 cases were compared with 143 randomly selected controls. Because this analysis involved all ovarian cancer patients irrespective of histology, we labeled this the all OC cohort. In OC cohort cases, all but 1 patient had 2 matched controls. Consequently, the OC cohort consisted of 215 patients in total. The purpose of this preliminary analysis was to describe the full spectrum of ovarian cancer observed in association with metformin intake de novo and to compare them with controls.
In the subsequent, definitive analysis, epithelial ovarian cancer cases (n = 61) were identified from the parent cohort of 72 cases after excluding patients who had nonepithelial ovarian cancer (n = 5), tumors with low malignant potential (n = 2), and atypical, rare histology types (transitional and small cell cancers of the ovary; n = 4). These 61 cases were matched with controls at a 1:3 ratio based on 3 variables: age at diagnosis (±5 years), FIGO stage, and surgical cytoreduction status (optimal vs suboptimal). Of those matched cases, 57 had 3 controls each, 3 had 2 controls each, and 1 had only 1 control. Therefore, the definitive analysis cohort of cases matched with controls consisted of 239 patients (61 cases and 178 controls). Because the definitive cohort had patients with epithelial ovarian cancer only, we labeled this the EOC cohort. The purpose of the definitive analysis was to identify the association of metformin with survival in epithelial ovarian cancer, because the later accounts for the vast majority of the mortality ascribed to the disease in the adult population, and the distribution of histology and FIGO stage was disparate in our initial analysis of the OC cohort (Table 1) between cases and controls.
|No. of Patients (%)|
|Age: Mean±SD, y||60.6±11||60.3±11||.8|
|BMI: Mean±SD, kg/m2||33±7||29±7||< .001|
|Median year of diagnosis||2005||2004|
|Yes||57 (89.1)||114 (89.1)|
|No||7 (10.9)||14 (10.9)|
|I||22 (30.6)||20 (14)||.01|
|II||5 (6.9)||4 (2.8)|
|III||38 (52.8)||97 (67.8)|
|IV||7 (9.7)||22 (15.4)|
|Early: I and II||27 (37.5)||24 (16.8)||.0008|
|Late: III and IV||45 (62.5)||119 (83.2)|
|High||48 (70.6)||126 (88.1)||.002|
|Low||20 (29.4)||17 (11.9)|
|Serous||39 (54.2)||106 (74.1)||0.03|
|Clear cell||5 (6.9)||9 (6.3)|
|Endometrioid||14 (19.4)||18 (12.6)|
|Mucinous||3 (4.2)||4 (2.8)|
|LMP||2 (2.8)||0 (0)|
|Sarcoma||1 (1.4)||5 (3.5)|
|Granulosa cell tumor||4 (5.6)||0 (0)|
|Others||4 (5.6)||1 (0.7)|
|Serous||39 (54.2)||106 (74.1)||.003|
|Nonserous||33 (45.8)||37 (25.9)|
|Platinum||52 (77.6)||116 (83.5)||.2|
|Nonplatinum||1 (1.5)||9 (6.7)|
|None||14 (20.9)||14 (10.1)|
Data were collected on the 103 patients who had both diabetes and epithelial ovarian cancer and had received insulin or other antidiabetic medications rather than metformin. This cohort was referred to as the diabetic control group. Because of the small sample size (n = 103), matching was not feasible between the cases in the EOC cohort and the diabetic control group.
The Student t test and chi-square/Fisher exact tests were used to compare groups as appropriate for parametric and nonparametric comparisons. Kaplan-Meier estimates were used to compare survival using the log-rank test. Univariate and multivariate Cox proportional hazards model were used to access the effect of risk factors on survival outcomes. In the later analysis, biologically as well as clinically important variables that had well known association with prognosis (disease stage, histology, grade, chemotherapy, etc) in patients with ovarian cancer were included. This model was constructed to assess whether metformin exhibits an independent association with survival in ovarian cancer. All tests were 2-sided, and P values ≤ .05 were considered statistically significant. Statistical analyses were done using the SAS statistical software package (version 9.1.3; SAS Institute, Cary, NC).
Initial analyses (the OC cohort) were based on 72 cases and 143 randomly selected controls. One case had only 1 control (thus, in total, there were 143 controls rather than 144). Table 1 lists the key variables for the cases and controls. The distribution of age, year of diagnosis, optimal surgical cytoreduction, and platinum-based chemotherapy was similar in cases and controls (Table 1). The BMI of cases was higher than that of controls, as expected. Early stage, lower grade, and nonserous histology tumors were significantly more prevalent in cases than in controls, as indicated in Table 1. The entire cohort had a median DSS of 5.5 years (95% confidence interval [CI], 4.1-7.5 years). The median survival for controls was 4.1 years, and the median survival for cases was not reached (log-rank P = .002). The 5-year survival rate for controls was 44% versus 73% for cases (log-rank P = .002) (Fig. 1). Results from the multivariable analysis for the OC cohort are presented in Table 2. In this analysis, after adjusting for age, year of diagnosis, BMI, disease stage, histology, and chemotherapy, only grade (hazard ratio, 8.6; 95% CI, 1.2-63; P = .03) and receipt of metformin (hazard ratio, 2.7; 95% CI, 1.4-5.4; P = .004) remained independent predictors of survival.
|Variable||HR (95% CI)||P|
|Late: III and IV||2.4 (0.8-7.0)||.10|
|Early: I and II||1.0|
|High: 3 and 4||8.6 (1.2-63.0)||.03|
|Low: 1 and 2||1.0|
In the definitive analysis of the EOC cohort, 61 cases of epithelial ovarian cancer were matched with 178 controls. The distribution of the key variables among cases and controls in the definitive analysis of the EOC cohort is depicted in Table 3. The mean duration of follow-up for cases was 3.7 years (median 3.4 years) vs 4.3 years (median 3.3 years) for controls. The distribution of matching variables (age, FIGO stage, and surgical cytoreductive effort measured by postoperative residual disease) remained similar among cases and controls (Table 3). Furthermore, the distribution of grade, histology (serous vs nonserous), and chemotherapy remained similar between cases and controls (P > .05) (Table 3). Recurrence after chemotherapy was reported in 194 patients (47 cases and 147 controls). Platinum-sensitive disease was present in 29 of 47 cases (62%) versus 82 of 147 controls (56%; P = .5). The median recurrence-free survival was 32 months for cases and 22 months for controls (P = .04).
|No. of Patients (%)|
|Variable||Cases||Nondiabetic Controls||Diabetic Controlsa||P|
|All patients||61 (100)||178 (100)||103 (100)|
|Age: Mean±SD, yb||62±11||61±11||65±13||.06|
|BMI: Mean±SD, kg/m2||34±7||28±6||30±8||< .001|
|I||15 (25)b||43 (24)b||16 (17)||.7|
|II||5 (8)b||13 (7)b||5 (5)|
|III||35 (57)b||105 (59)b||59 (63)|
|IV||6 (10)b||17 (10)b||13 (14)|
|High||45 (74)||141 (79)||81 (80)||.6|
|Low||16 (26)||37 (21)||20 (20)|
|Serous||39 (64)||126 (71)||78 (76)||.2|
|Nonserous||22 (36)||52 (29)||25 (24)|
|Yes||49 (80)b||146 (82)b||80 (78)||.2|
|No||11 (18)b||30 (17)b||21 (20)|
|Unknown||1 (2)||2 (1)||2 (2)|
|Platinum based||49 (86)||152 (88)||74 (83)||.3|
|Nonplatinum||0 (0)||5 (3)||2 (2)|
|None||8 (14)||15 (9)||13 (15)|
The 5-year DSS rate for the entire cohort was 52%. The 5-year DSS rate in cases (67%) was significantly better compared with the DSS rate in controls (47%; log-rank P = .006) (Fig. 2). Results from the multivariable analysis are presented in Table 4. After adjusting for BMI, grade, histology, and chemotherapy, metformin remained an independent predictor of survival in ovarian cancer.
|Variable||HR (95% CI)||P|
|High: 3 and 4||1.3 (0.6-2.8)||.5|
|Low: 1 and 2||1.0|
The definitive study cohort of 239 patients with epithelial ovarian cancer was divided further into 2 subgroups based on the absence (residual disease measuring 0 cm) or presence (residual disease measuring >0 cm) of gross residual disease to assess the association of metformin with prognosis in these 2 subgroups. Of 239 patients in the EOC cohort, 130 (55%) underwent complete cytoreduction (residual disease measuring 0 cm), and 106 (45%) underwent incomplete cytoreduction (residual disease measuring >0 cm); whereas, in 3 patients, the status of residual disease remained unknown. In patients with epithelial ovarian cancer who underwent complete cytoreduction, the 5-year survival of cases was 90% versus 69% in controls (P = .04) (Fig. 3, top left). Conversely, for patients with epithelial ovarian cancer who underwent incomplete cytoreduction, the 5-year survival rate was 41% for cases was versus 21% for controls (P = .047) (Fig. 3, top right). Finally, restricting the survival analysis in the EOC cohort to patients who had advanced-stage (FIGO stage III and IV) disease (5-year DSS for cases vs controls: 55% vs 33%, respectively; HR, 2.0; 95% CI, 1.2-3.5; P = .008) (Fig. 3, bottom left) or serous histology only (5-year DSS for cases vs controls: 60% vs 38%, respectively; HR, 2.2; 95%CI, 1.3-4; P = .007) (Fig. 3, bottom right), the survival among cases still was better compared with the survival among controls.
A comparison of data between the diabetic control group and the EOC cohort is presented in the Table 3. Compared with cases in the EOC cohort, controls in the diabetic control group were slightly older and had marginally lower BMI. However, the distribution of disease stage, histology, grade, extent of surgical cytoreduction, and platinum-based chemotherapy remained similar between cases in the EOC cohort and diabetic controls. Despite these similarities, the 5-year DSS rate for diabetic controls was the lowest (40%) compared with cases in the EOC cohort (67%) and nondiabetic controls (47%; P = .003) (Fig. 4). Survival also was assessed based on metformin intake versus insulin or other antidiabetic medication intake in patients with ovarian cancer. Receipt of metformin was associated with the best 5-year DSS rate (67%) compared with insulin (43%) and other antidiabetic medications (34%; P = .004), as illustrated in Figure 5.
The current study is the largest case-control study to our knowledge that evaluates the association between metformin and survival in women with ovarian cancer. Although causation could not be assessed; we observed that receipt of metformin was associated with significantly better survival in ovarian cancer. The interest in metformin as an anticancer compound initially was ignited by 2 key epidemiologic reports. Evans et al reported a reduced risk of cancer in diabetic patients who were receiving metformin,13 whereas Bowker et al demonstrated reduced cancer-related mortality in association with metformin.14 These epidemiologic observations were complimented by in vitro and in vivo data on the anticancer activity of metformin in several human cancers.8-11
The current study placed heavy emphasis on the survival comparisons between cases and nondiabetic controls (referred to herein as controls) rather than the diabetic control group. There are 2 reasons for this approach: 1) It has been demonstrated that diabetes has a negative impact on ovarian cancer survival15 and, thus, a comparison of survival between the cases and diabetic control groups was likely to overestimate the magnitude of survival benefit associated with metformin; and 2) the vast majority of patients with ovarian cancer are not diabetic and will be candidates for enrolment into future clinical trials using metformin in ovarian cancer. Therefore, in terms of clinical applicability, the study results using nondiabetic controls are more relevant. However, we could argue that, in an ideal study, the cases and controls should differ only in terms of exposure (metformin) and not in terms of other confounders (eg, diabetes). To overcome this limitation, we have also provided clinical and survival data for the diabetic control group (Table 3, Fig. 4).
Before metformin can be received by patients for its anticancer effect, prospective trials using metformin in ovarian cancer are warranted. Our data provide support for such trials. To that end, the most important question is to decipher whether our current observations are merely by chance or are the results of a true clinical effect. It is noteworthy that, in our initial analysis of the OC cohort that incorporated all metformin-associated ovarian cancers, metformin-associated tumors were more likely to be lower grade, earlier stage, and had a more favorable histology profile. One could attribute the association of better prognosis in the OC cohort cases with that of metformin on the constellation of these lower risk attributes. Part of our data in the OC cohort lends some support to this hypothesis. For example, 4 cases (5.6%) in the OC cohort had granulosa cell tumors (Table 1), and patients who have granulosa cell tumors have a much better prognosis compared with patients who have epithelial ovarian cancer. In addition, diabetes and obesity are commonly the result of hormone effects of the tumor itself,16 leading to the prescription of metformin. Along similar lines, tumors with low malignant potential were observed more frequently in cases than in controls (Table 1) and have a better prognosis compared with frankly invasive epithelial ovarian cancers. Despite these limitations, the primary value of the preliminary analysis of the OC cohort was to describe the full spectrum of ovarian tumors associated with metformin intake. Furthermore, the emergence of metformin as an independent predictor of survival in the OC cohort multivariable analysis (Table 2), in which disease stage, grade, and histology were controlled, tends to favor a true association between metformin and a clinically beneficial impact in ovarian cancer.
Next, we conducted our definitive analysis of the EOC cohort to overcome the above-mentioned limitations posed by the nonuniform distribution of FIGO stage and tumor types in the OC cohort. The EOC cohort only included women with epithelial ovarian cancers, and the controls were matched for age, FIGO stage, and residual disease after cytoreductive surgery (Table 3). All 3 of these matching criteria are the primary determinants affecting survival in ovarian cancer.17, 18 The distribution of all variables analyzed was statistically similar between cases and controls in the EOC cohort with the exception of BMI (Table 3). Cases had higher BMI, because metformin is usually prescribed to obese diabetic individuals.19 It is unlikely that this unequal distribution of BMI affected our survival analysis, because: 1) BMI is not considered to have prognostic value in ovarian cancer,20 and 2) BMI was controlled in the multivariable analysis. In our analysis of the EOC cohort, the association of metformin with better survival persisted in the overall analysis as well as in the subgroup analysis of the complete and incomplete cytoreduction subgroups. Second, when the analysis was restricted to serous histology only or advanced-stage tumors only, the survival advantage associated with metformin persisted. Finally, the multivariable analysis, after controlling for BMI, histology, tumor grade, and type of chemotherapy, indicated that metformin is an independent predictor of survival in patients with ovarian cancer. These observations suggest that confounding is unlikely to be the sole factor accounting for the remarkable difference in prognosis between cases and controls. Moreover, previous data from our laboratory indicated a potent inhibition of cell division, cell survival, and antitumor activity of metformin in a mouse model of ovarian cancer.6, 7 Taken together, these data suggest a possible causative role of metformin in exhibiting antitumor activity in ovarian cancer.
The limitations of this study are its retrospective nature, small sample size, and lack of information on the receipt of metformin before cancer diagnosis because of a lack of medical records before patient care was established. However, to our knowledge, this is the largest clinical study to date evaluating the association between metformin and prognosis in ovarian cancer. One previous study evaluated the association between metformin and survival in patients with ovarian cancer12 and demonstrated better recurrence-free survival, similar to our report. However, the current report overcomes several limitations that were posed by the previous study. These include a much larger sample size of 72 cases (vs 16 cases in the previous study), the ability to match cases in the EOC cohort with nondiabetic controls for important clinical variables (vs nonmatched controls in the previous study), and the ability to demonstrate differences in overall survival (vs only recurrence-free survival advantage in cases in the previous study in multivariable analysis). Moreover, although a sample size of 72 cases may be perceived as relatively small, the receipt of metformin because of diabetes in cancer patients is rather uncommon. This notion is supported by a study of metformin in patients with breast cancer from The University of Texas M. D Anderson Cancer Center by Jiralerspong et al, in which only 68 such patients could be identified.21 Conversely, our ability to match cases and controls for 3 variables (age, FIGO stage, and surgical cytoreduction) is a noteworthy strength of the current study. The later is especially true, because surgical effort is largely surgeon-dependent (rather than depending on tumor biology), and diabetic patients have higher operative morbidity,22 which may limit the fraction of patients who have comprehensive surgical staging at a given institution.15 Thus, this factor was eliminated by matching for surgical effort.
In conclusion, the results from our current case-control study demonstrate that metformin intake independently predicts better survival in patients with ovarian cancer. Although causation could not be assessed by our retrospective study, metformin nevertheless is a strong contender for further clinical studies in ovarian cancer. These may be retrospective, confirmatory studies by other institutions and/or prospective phase 2/3 studies in women with ovarian cancer.
This work was supported in part by the grants from the Fred C. and Katherine B. Andersen Foundation (VS) and National Institutes of Health Grants P50 CA136393 (VS) and CA123249 (VS).
CONFLICT OF INTEREST DISCLOSURES
The authors made no disclosures.