Impact of Metformin on Clinical Outcomes in Prostate Cancer
Impact of Metformin on Clinical Outcomes in Prostate Cancer
We retrieved a total of 230 citations through electronic databases and gray literature. We excluded the following studies: studies assessing outcomes following metformin use in animal models; in vitro studies; reviews; RCTs on interventions other than metformin; assessing risk of prostate cancer with the use of metformin; and assessing risk of prostate cancer with presence of diabetes. In addition, we excluded an RCT on metformin use along with ADT among men with prostate cancer as this study did not assess biochemical recurrence or any other prostate cancer-related outcomes of our interest. We also excluded one observation study as we were not able to distinguish between metformin users as mono- or combination therapy with thiazolidinedione. After all these exclusions, a total nine studies were available for our review and meta-analysis. See Figure 1 for details of study selection.
(Enlarge Image)
Figure 1.
Preferred reporting items for systematic reviews and meta-analyses flow chart for study selection. Pca, prostate cancer; RCT, randomized clinical trials.
Characteristics of all the nine studies are presented in Table 1, Table 2 and Table 3. These were published between 2010 and 2014. Five studies were conducted in the United States, two in the United Kingdom and one in Canada, while, the last study utilized data from both UK and US. Five studies restricted the study sample to men with diabetes and prostate cancer, while four studies included sample of men with prostate cancer with presence or absence of diabetes. The metformin users ranged from 32.6 to 63.5% among men with prostate cancer. The sample size of the studies varied from 233 to 3837. Two studies explicitly mentioned exclusion of men with type 1 diabetes mellitus. Four studies restricted the cohort of men with prostate cancer to those treated with RP, while one study included men with prostate cancer treated with external-beam-radiation therapy (EBRT).
The STROBE checklist was used to assess quality of the included studies as shown in Table 4. Except one nested case-control study, the other eight studies utilized a retrospective cohort design. The one nested case-control study utilized Cancer Registry linked Medical records. Out of eight retrospective studies, four studies utilized single institutional based electronic medical records; two studies utilized multicenter electronic medical data from primary care practice, one study utilized data from the Veterans Affairs Medical Centers, and the last study utilized amalgamation of population-based administrative claims data with other clinical databases.
Only four studies explicitly mentioned the ascertainment of diabetes via diagnostic codes or self-report. Three studies mentioned the ascertainment of prostate cancer via diagnostic codes and biopsy-proven prostate cancer diagnosis by a trained pathologist. Metformin exposure was measured prior to RP in two studies, as cumulative exposure of metformin during the entire study period in two studies, anytime during study period without providing timing, index-date, or duration of prostate cancer exposure in two studies, at the time of cancer diagnosis in one study, 90 days before and after diagnosis of cancer in one study, and prior to EBRT or anytime post-EBRT in the final study.
The included studies controlled for multiple potential confounders such as demographic (age, race), socio-economic status, access to primary care, prostate cancer severity (PSA, Gleason score, stage), prostate cancer primary treatment (RP, RT, ADT), personal healthcare practices (body mass index (BMI)), anti-diabetes medications (insulin, sulfonylurea, thiazolidinedione), other medications (statins and COX inhibitors) and comorbidity status in multivariable modeling frameworks. Three studies controlled for the comorbidity status and three studies controlled for African-American race as a potential confounder. Except one study, none of the other studies controlled for diabetes severity or duration.
Seven studies had median age of population in the range of 61–70 years,, while two studies had the median age above 70 years. Among three studies that reported findings by race, African-Americans constituted 20–47% among the metformin users. Among two studies that reported diabetes duration, the median duration of diabetes ranged from 3.08 to 12.80 years. Among three studies that reported comorbidities, two studies utilized the Charlson comorbidity index score while one study reported pre-existing cardiac disease as a measure of comorbidity. Five studies reported BMI and most of the study participants were in the overweight or obese categories. There were no significant differences in BMI levels and metformin use. Only one study reported 15–17% smoking rates in the study population. (See Table 2 for details.)
The baseline characteristics related to prostate cancer stage and severity, and treatments in the included studies are presented in Table 3. Seven studies reported the status of pre-operative PSA levels. Six of the seven studies did not find significant differences in the baseline pre-operative PSA levels between metformin users and nonusers. The seventh study reported higher median PSA levels among the metformin users as compared to non-metformin users. Eight out of nine studies reported the Gleason score among metformin users and nonusers. A greater proportion of men had Gleason scores of 7 or 8 with no significant difference between metformin users and non-metformin users in these studies. Six studies reported the American Joint Cancer Commission (AJCC) guideline based tumor stages. Three studies had higher percentages of men with T1 stage (84.8–56.8%) and the rates of T1, T2, T3 stages were significantly different between the metformin users and nonusers. One study reported a significantly higher proportion of men with stage three or advanced stage among the non-metformin users as compared to the metformin users (57.4 versus 51.6%). There were no significant differences in the rates of positive surgical margin, use of ADT, and radiation therapy between metformin users and nonusers.
All the nine studies conducted survival analysis of outcomes using Kaplan–Meir Plots, or univariate and multivariate Cox-proportional hazard regressions.
The hazard of biochemical recurrence was reported in five studies. Metformin use was marginally associated with an 18% reduction in the risk of biochemical recurrence in random-effects model (pHR: 0.82, 95% CI: 0.67, 1.01, P-value=0.06) without any evidence of heterogeneity (See Figure 2 for details). With respect to publication bias, we did not detect a statistically significant publication bias based on the Egger's test (P=0.41). In addition, each study fell under white area of the contour-enhanced funnel plot (see Figure 4 for details). In the contour enhance plot, as all studies were on the left side of the null effect line, we suspected publication bias and hence we conducted a trim-fill analysis but did not find any missing studies.
(Enlarge Image)
Figure 2.
Forest plots on the association of metformin with clinical outcomes: (1) biochemical recurrence; (2) metastases; (3) all-cause mortality; and (4) prostate-specific mortality using a random-effect model.
(Enlarge Image)
Figure 4.
Funnel plot (contour enhanced); for publication bias on association of metformin with all-cause mortality (a) statistical significance contour (b) I contour.
Two studies reported CRPC as an outcome. Allott et al. found that the rate of CRPC was 3.8% (N=14) among the included study population, and reported no significant association between metformin use and CRPC (HR, 2.98; 95% CI, 0.98–9.05). However, Spratt et al. reported that metformin use was associated with lower odds of development of CRPC (aOR, 0.067; 95% CI, 0.01–0.55) with a median follow-up period of 8.7 years. We did not conduct a meta-analysis for this outcome because difference in the reported statistics parameters for outcomes such as unadjusted hazard ratio and adjusted odds ratio.
Four studies reported metastases or lymph node metastases as an outcome. Except Allott et al., other three studies reported adjusted hazards ratio for the development of metastases. With incidence of 14 events of metastases in the study population, Allott et al. found no significant association between metformin use and metastases (HR: 2.53, 95% CI: 0.70, 9.22). Similar to this finding, we found no significant association between metformin use and risk of development of metastases among individuals with prostate cancer (pHR: 0.59, 95% CI: 0.30–1.18, P-value=0.14) with the presence of significant heterogeneity (I=74%). We did not observe any publication bias for metastases based on the Egger's test and contour-enhanced funnel plot (see Figure 3 for details).
(Enlarge Image)
Figure 3.
Funnel plot (contour enhanced) for publication bias on association of metformin with (a) biochemical recurrence and (b) metastases.
Out of the nine included studies, six studies assessed all-cause mortality as the primary outcome. Overall, the rates of all-cause mortality ranged from 2.2–35% (35% Margel et al.; 33% Currie et al.; 27.5% Bensimon et al.; 14% in Spratt et al., 4% in Kaushik et al., and 2.2% in Allott et al). We found that use of metformin was not significantly associated with all-cause mortality (pHR: 0.86; 95% CI, 0.67, 1.10, P-value=0.23), however, there was heterogeneity among the studies (I=73%). We did not detect a statistically significant publication bias based on the Egger's test (P=0.47) or by using the contour-enhanced funnel plot (see Figure 4). All of the studies fell below 0.1 significance level contour and on both sides of the null effect line indicating publications with even nonsignificant results. After the trim-fill analysis, we found one study missing on the bottom right side of the funnel plot using the random-effect model (see Figure 5, left). After adjusting for this study in the forest plot, the pooled HR was (pHR: 0.95, 95% CI: 0.76, 1.20) (see Figure 5, right). However, the heterogeneity was still evident in the simulated meta-analysis (I=75%, P=0.001). Additionally, due to considerable heterogeneity (I=78%) in the pooled effect for all-cause mortality, we developed a contour-enhanced funnel plot based on I contours (see Figure 4 right) which suggest that a significant heterogeneity will persist even there was publication of a new study with large or small sample size in the future.
(Enlarge Image)
Figure 5.
(a) Funnel plot (trim-fill) for publication bias and (b) adjusted forest plot with missing study on association of metformin with all-cause mortality.
Five studies reported estimates of prostate cancer-specific mortality rates by metformin use. Allott et al. found that only 8 men out of 371 men died due to prostate cancer (2.2%). This study did not find a statistically significantly association between metformin use and prostate cancer-specific mortality (HR: 2.89, 95% CI: 0.68, 12.3) in unadjusted analysis. Due to the small sample size, even after controlling for other factors in a conditional manner, Allott et al. did not find any significant association between metformin use and prostate cancer-specific mortality. In the pooled results of the other four studies, metformin use was not associated with prostate cancer-specific mortality in a random-effects model (pHR: 0.76, 95% CI: 0.43, 1.33, P-value=0.33) without significant heterogeneity. We did not detect publication bias statistically based on the Egger's test (P=0.11) (see Figure 6). However, visualization of funnel plot, and after doing trim-fill analysis, we found two missing studies on the bottom left side of the funnel plot using the random-effect model (see Figure 6, left). After adjusting the forest plot with these studies, the pooled HR was (pHR: 0.83, 95% CI: 0.60, 1.16) without any heterogeneity in the simulated meta-analysis (I=0%, P-value=0.32).
(Enlarge Image)
Figure 6.
(a) Funnel plot (contour enhanced) for publication bias and (b) adjusted forest plot with two missing studies on association of metformin with prostate cancer-specific mortality.
Sensitivity analysis by status of diabetes and other prostate cancer therapy for each clinical outcome among the included studies are presented in Supplementary Appendix 2 http://www.nature.com/pcan/journal/v18/n2/suppinfo/pcan201452s1.html.
a. Men With Prostate Cancer and Diabetes. Among the studies with included sample of men with prostate cancer and diabetes, metformin use was significantly associated with reduction in the all-cause mortality (pHR: 0.95, 95% CI: 0.91–0.99, P-value=0.02) and prostate cancer-specific mortality (pHR: 0.81, 95% CI: 0.75,0.87), while metformin use was not significantly associated with all-cause and prostate cancer-specific mortality among the studies with men with prostate cancer with and without diabetes.
b. Types of Prostate Cancer Therapy (Radiation Therapy vs RP). Studies with men treated with RP, we found no significant difference in the any of clinical outcomes with the use of metformin. However, one study with a sample of men treated with EBRT found a significant improvement in all the clinical outcomes with the use of metformin.
Results
We retrieved a total of 230 citations through electronic databases and gray literature. We excluded the following studies: studies assessing outcomes following metformin use in animal models; in vitro studies; reviews; RCTs on interventions other than metformin; assessing risk of prostate cancer with the use of metformin; and assessing risk of prostate cancer with presence of diabetes. In addition, we excluded an RCT on metformin use along with ADT among men with prostate cancer as this study did not assess biochemical recurrence or any other prostate cancer-related outcomes of our interest. We also excluded one observation study as we were not able to distinguish between metformin users as mono- or combination therapy with thiazolidinedione. After all these exclusions, a total nine studies were available for our review and meta-analysis. See Figure 1 for details of study selection.
(Enlarge Image)
Figure 1.
Preferred reporting items for systematic reviews and meta-analyses flow chart for study selection. Pca, prostate cancer; RCT, randomized clinical trials.
Characteristics of Included Studies
Characteristics of all the nine studies are presented in Table 1, Table 2 and Table 3. These were published between 2010 and 2014. Five studies were conducted in the United States, two in the United Kingdom and one in Canada, while, the last study utilized data from both UK and US. Five studies restricted the study sample to men with diabetes and prostate cancer, while four studies included sample of men with prostate cancer with presence or absence of diabetes. The metformin users ranged from 32.6 to 63.5% among men with prostate cancer. The sample size of the studies varied from 233 to 3837. Two studies explicitly mentioned exclusion of men with type 1 diabetes mellitus. Four studies restricted the cohort of men with prostate cancer to those treated with RP, while one study included men with prostate cancer treated with external-beam-radiation therapy (EBRT).
Quality Assessment of Included Studies
The STROBE checklist was used to assess quality of the included studies as shown in Table 4. Except one nested case-control study, the other eight studies utilized a retrospective cohort design. The one nested case-control study utilized Cancer Registry linked Medical records. Out of eight retrospective studies, four studies utilized single institutional based electronic medical records; two studies utilized multicenter electronic medical data from primary care practice, one study utilized data from the Veterans Affairs Medical Centers, and the last study utilized amalgamation of population-based administrative claims data with other clinical databases.
Only four studies explicitly mentioned the ascertainment of diabetes via diagnostic codes or self-report. Three studies mentioned the ascertainment of prostate cancer via diagnostic codes and biopsy-proven prostate cancer diagnosis by a trained pathologist. Metformin exposure was measured prior to RP in two studies, as cumulative exposure of metformin during the entire study period in two studies, anytime during study period without providing timing, index-date, or duration of prostate cancer exposure in two studies, at the time of cancer diagnosis in one study, 90 days before and after diagnosis of cancer in one study, and prior to EBRT or anytime post-EBRT in the final study.
The included studies controlled for multiple potential confounders such as demographic (age, race), socio-economic status, access to primary care, prostate cancer severity (PSA, Gleason score, stage), prostate cancer primary treatment (RP, RT, ADT), personal healthcare practices (body mass index (BMI)), anti-diabetes medications (insulin, sulfonylurea, thiazolidinedione), other medications (statins and COX inhibitors) and comorbidity status in multivariable modeling frameworks. Three studies controlled for the comorbidity status and three studies controlled for African-American race as a potential confounder. Except one study, none of the other studies controlled for diabetes severity or duration.
Characteristics of Men With Prostate Cancer
Seven studies had median age of population in the range of 61–70 years,, while two studies had the median age above 70 years. Among three studies that reported findings by race, African-Americans constituted 20–47% among the metformin users. Among two studies that reported diabetes duration, the median duration of diabetes ranged from 3.08 to 12.80 years. Among three studies that reported comorbidities, two studies utilized the Charlson comorbidity index score while one study reported pre-existing cardiac disease as a measure of comorbidity. Five studies reported BMI and most of the study participants were in the overweight or obese categories. There were no significant differences in BMI levels and metformin use. Only one study reported 15–17% smoking rates in the study population. (See Table 2 for details.)
Prostate Cancer-specific Characteristics by Metformin use
The baseline characteristics related to prostate cancer stage and severity, and treatments in the included studies are presented in Table 3. Seven studies reported the status of pre-operative PSA levels. Six of the seven studies did not find significant differences in the baseline pre-operative PSA levels between metformin users and nonusers. The seventh study reported higher median PSA levels among the metformin users as compared to non-metformin users. Eight out of nine studies reported the Gleason score among metformin users and nonusers. A greater proportion of men had Gleason scores of 7 or 8 with no significant difference between metformin users and non-metformin users in these studies. Six studies reported the American Joint Cancer Commission (AJCC) guideline based tumor stages. Three studies had higher percentages of men with T1 stage (84.8–56.8%) and the rates of T1, T2, T3 stages were significantly different between the metformin users and nonusers. One study reported a significantly higher proportion of men with stage three or advanced stage among the non-metformin users as compared to the metformin users (57.4 versus 51.6%). There were no significant differences in the rates of positive surgical margin, use of ADT, and radiation therapy between metformin users and nonusers.
Metformin use and Clinical Outcomes
All the nine studies conducted survival analysis of outcomes using Kaplan–Meir Plots, or univariate and multivariate Cox-proportional hazard regressions.
Metformin and Biochemical Recurrence
The hazard of biochemical recurrence was reported in five studies. Metformin use was marginally associated with an 18% reduction in the risk of biochemical recurrence in random-effects model (pHR: 0.82, 95% CI: 0.67, 1.01, P-value=0.06) without any evidence of heterogeneity (See Figure 2 for details). With respect to publication bias, we did not detect a statistically significant publication bias based on the Egger's test (P=0.41). In addition, each study fell under white area of the contour-enhanced funnel plot (see Figure 4 for details). In the contour enhance plot, as all studies were on the left side of the null effect line, we suspected publication bias and hence we conducted a trim-fill analysis but did not find any missing studies.
(Enlarge Image)
Figure 2.
Forest plots on the association of metformin with clinical outcomes: (1) biochemical recurrence; (2) metastases; (3) all-cause mortality; and (4) prostate-specific mortality using a random-effect model.
(Enlarge Image)
Figure 4.
Funnel plot (contour enhanced); for publication bias on association of metformin with all-cause mortality (a) statistical significance contour (b) I contour.
Metformin and Development of CRPC
Two studies reported CRPC as an outcome. Allott et al. found that the rate of CRPC was 3.8% (N=14) among the included study population, and reported no significant association between metformin use and CRPC (HR, 2.98; 95% CI, 0.98–9.05). However, Spratt et al. reported that metformin use was associated with lower odds of development of CRPC (aOR, 0.067; 95% CI, 0.01–0.55) with a median follow-up period of 8.7 years. We did not conduct a meta-analysis for this outcome because difference in the reported statistics parameters for outcomes such as unadjusted hazard ratio and adjusted odds ratio.
Metformin and Metastases
Four studies reported metastases or lymph node metastases as an outcome. Except Allott et al., other three studies reported adjusted hazards ratio for the development of metastases. With incidence of 14 events of metastases in the study population, Allott et al. found no significant association between metformin use and metastases (HR: 2.53, 95% CI: 0.70, 9.22). Similar to this finding, we found no significant association between metformin use and risk of development of metastases among individuals with prostate cancer (pHR: 0.59, 95% CI: 0.30–1.18, P-value=0.14) with the presence of significant heterogeneity (I=74%). We did not observe any publication bias for metastases based on the Egger's test and contour-enhanced funnel plot (see Figure 3 for details).
(Enlarge Image)
Figure 3.
Funnel plot (contour enhanced) for publication bias on association of metformin with (a) biochemical recurrence and (b) metastases.
Metformin and All-cause Mortality
Out of the nine included studies, six studies assessed all-cause mortality as the primary outcome. Overall, the rates of all-cause mortality ranged from 2.2–35% (35% Margel et al.; 33% Currie et al.; 27.5% Bensimon et al.; 14% in Spratt et al., 4% in Kaushik et al., and 2.2% in Allott et al). We found that use of metformin was not significantly associated with all-cause mortality (pHR: 0.86; 95% CI, 0.67, 1.10, P-value=0.23), however, there was heterogeneity among the studies (I=73%). We did not detect a statistically significant publication bias based on the Egger's test (P=0.47) or by using the contour-enhanced funnel plot (see Figure 4). All of the studies fell below 0.1 significance level contour and on both sides of the null effect line indicating publications with even nonsignificant results. After the trim-fill analysis, we found one study missing on the bottom right side of the funnel plot using the random-effect model (see Figure 5, left). After adjusting for this study in the forest plot, the pooled HR was (pHR: 0.95, 95% CI: 0.76, 1.20) (see Figure 5, right). However, the heterogeneity was still evident in the simulated meta-analysis (I=75%, P=0.001). Additionally, due to considerable heterogeneity (I=78%) in the pooled effect for all-cause mortality, we developed a contour-enhanced funnel plot based on I contours (see Figure 4 right) which suggest that a significant heterogeneity will persist even there was publication of a new study with large or small sample size in the future.
(Enlarge Image)
Figure 5.
(a) Funnel plot (trim-fill) for publication bias and (b) adjusted forest plot with missing study on association of metformin with all-cause mortality.
Metformin and Prostate Cancer-specific Mortality
Five studies reported estimates of prostate cancer-specific mortality rates by metformin use. Allott et al. found that only 8 men out of 371 men died due to prostate cancer (2.2%). This study did not find a statistically significantly association between metformin use and prostate cancer-specific mortality (HR: 2.89, 95% CI: 0.68, 12.3) in unadjusted analysis. Due to the small sample size, even after controlling for other factors in a conditional manner, Allott et al. did not find any significant association between metformin use and prostate cancer-specific mortality. In the pooled results of the other four studies, metformin use was not associated with prostate cancer-specific mortality in a random-effects model (pHR: 0.76, 95% CI: 0.43, 1.33, P-value=0.33) without significant heterogeneity. We did not detect publication bias statistically based on the Egger's test (P=0.11) (see Figure 6). However, visualization of funnel plot, and after doing trim-fill analysis, we found two missing studies on the bottom left side of the funnel plot using the random-effect model (see Figure 6, left). After adjusting the forest plot with these studies, the pooled HR was (pHR: 0.83, 95% CI: 0.60, 1.16) without any heterogeneity in the simulated meta-analysis (I=0%, P-value=0.32).
(Enlarge Image)
Figure 6.
(a) Funnel plot (contour enhanced) for publication bias and (b) adjusted forest plot with two missing studies on association of metformin with prostate cancer-specific mortality.
Sensitivity Analyses
Sensitivity analysis by status of diabetes and other prostate cancer therapy for each clinical outcome among the included studies are presented in Supplementary Appendix 2 http://www.nature.com/pcan/journal/v18/n2/suppinfo/pcan201452s1.html.
a. Men With Prostate Cancer and Diabetes. Among the studies with included sample of men with prostate cancer and diabetes, metformin use was significantly associated with reduction in the all-cause mortality (pHR: 0.95, 95% CI: 0.91–0.99, P-value=0.02) and prostate cancer-specific mortality (pHR: 0.81, 95% CI: 0.75,0.87), while metformin use was not significantly associated with all-cause and prostate cancer-specific mortality among the studies with men with prostate cancer with and without diabetes.
b. Types of Prostate Cancer Therapy (Radiation Therapy vs RP). Studies with men treated with RP, we found no significant difference in the any of clinical outcomes with the use of metformin. However, one study with a sample of men treated with EBRT found a significant improvement in all the clinical outcomes with the use of metformin.
Source...