Identification of patient characteristics associated with survival benefit from metformin treatment in patients with stage I non-small cell lung cancer

Abstract

Background:

Non-small cell lung cancer (NSCLC) continues to be a major cause of cancer deaths. Previous investigation has suggested that metformin use can contribute to improved outcomes in NSCLC patients. However, this association is not uniform in all analyzed cohorts, implying that patient characteristics might lead to disparate results. Identification of patient characteristics that affect the association of metformin use with clinical benefit might clarify the drug’s effect on lung cancer outcomes and lead to more rational design of clinical trials of metformin’s utility as an intervention. In this study, we examined the association of metformin use with long-term mortality benefit in patients with NSCLC and the possible modulation of this benefit by body mass index (BMI) and smoking status, controlling for other clinical covariates.

Methods:

This was a retrospective cohort study in which we analyzed data from the Veterans Affairs (VA) Tumor Registry in the United States. Data from all patients with stage I NSCLC from 2000 to 2016 were extracted from a national database, the Corporate Data Warehouse that captures data from all patients, primarily male, who underwent treatment through the VA health system in the United States. Metformin use was measured according to metformin prescriptions dispensed to patients in the VA health system. The association of metformin use with overall survival (OS) after diagnosis of stage I NSCLC was examined. Patients were further stratified according to BMI and smoking status (previous vs current) to examine the association of metformin use with OS across these strata.

Results:

Metformin use was associated with improved survival in patients with stage I NSCLC (average hazard ratio, 0.82; P< .001). Significant interaction between the effect of metformin use and BMI on OS was observed (χ² = 3268.417; P < .001), with a greater benefit of metformin use observed in patients as BMI increased. Similarly, an interaction between smoking status and metformin use on OS was also observed (χ² = 2997.048; P < .001), with a greater benefit of metformin use observed in previous smokers compared with current smokers.

Conclusions:

In this large retrospective study, we showed that a survival benefit is enjoyed by users of metformin in a robust stage I NSCLC patient population treated in the VA health system. Metformin use was associated with an 18% improved OS. This association was stronger in patients with a higher BMI and in previous smokers. These observations deserve further mechanistic study and can help rational design of clinical trials with metformin in patients with lung cancer.

Graphical Abstract

Metformin and its interaction with BMI and smoking is associated with improved survival in patients with stage I NSCLC.

Lung cancer is the leading cause of cancer death in the United States with non-small cell lung cancer (NSCLC) subtype representing approximately 80% of all lung cancer patients. Although the clinical outlook for patients with stage I NSCLC treated primarily with surgery is markedly better than the prospects of patients with more advanced disease, early stage NSCLC is still associated with a poor 5-year survival rate.¹ Furthermore, although adjuvant therapies (eg, chemotherapy, immunotherapy) have improved the outcomes of some lung cancers,² these are not indicated for stage I disease.³ It is anticipated that, with the increased adoption of lung cancer screening, there will be an eventual stage migration to stage I disease.⁴ Therefore, new therapeutic modalities for improving survival for this subgroup of patients are needed.

Although its molecular mechanism is poorly understood, metformin (N,N-dimethylbiguanide) is widely used to treat type 2 diabetes and has a long history of safety and minimal side effects. It is also known to possess anticancer activity,⁵ for which a myriad of molecular and cellular mechanisms have been proposed.^6,7 In lung cancer, metformin reduces tobacco carcinogen-induced lung carcinogenesis in animal experiments, and epidemiologic studies suggest reduced cancer risk among metformin users.^8–10 Small retrospective studies by us and others have shown that metformin use is associated with improved survival in patients with lung cancer.^11,12 However, this benefit is seen almost exclusively in obese patients and is a result of immune reprogramming in the tumor microenvironment. This finding is very much in line with metformin’s reported ability to enhance immune mobilization against tumors.¹³ This benefit is disproportionately enjoyed by a specific patient subpopulation (ie, obese patients) and associated with indications of favorable immune reprogramming in the tumor microenvironment—a finding very much in line with metformin s reported ability to enhance immune mobilization against tumors.¹³ Previous investigators have also shown that the effect of metformin in a mouse tumor model depends on diet as well as the metabolic state of the mice.^14–17 Complementary studies show that the anticancer effect of metformin is present only in patients in a hyperinsulinemic state or with a metabolic syndrome, consistent with observations of a prospective clinical trial in breast cancer.¹⁸ These findings clearly suggest that distinct metabolic conditions might determine the effect of this drug on lung cancer. Therefore, the associations of the influence of body mass index (BMI) on overall survival (OS) and the influence of metformin use on outcome are confounded by the association of BMI and metformin use.¹⁹ In addition, metformin use is associated with improved cardiovascular disease outcomes in smokers and ex-smokers, consistent with the hypothesis that metformin might mitigate the proinflammatory effects of smoking.^20,21

On the basis of this previous knowledge, we hypothesized that metformin would provide a survival benefit in a robust cohort of patients with stage I NSCLC. We also hypothesized that this effect is influenced by BMI, an anthropometric measure of the metabolic syndrome, and smoking status.

METHODS

We conducted a retrospective study of US veterans diagnosed with clinical stage I NSCLC between 2000 and 2016 using patient electronic health care record data contained in the Department of Veterans Affairs (VA) Cen-tral Cancer Registry and the corporate data warehouse (HHS approval number 2296, dated August 23, 2018). We established a cohort of 24,048 patients by extracting data from the Department of VA tumor registry. Pa-tients who survived <90 days after diagnosis were removed from the anal-ysis (4.73%). Our final study cohort was 19,940 patients.

Outcome, Treatment, Adherence, and Additional Predictors

We chose 5-year survival as our outcome. A patient was entered into the study at their date of diagnosis and continued in the study until death occurred or they were censored. A patient was censored if they exceeded 5 years of survival with a truncated value of 1825 days of survival.

Metformin prescriptions were queried using the VA National Drug Name with 2141 patients having ever used metformin. Total days of metformin use after diagnosis was calculated from dispense date information and included up to 90 days before diagnosis. Metformin duration was defined as the total number of days of metformin supply within the 5 years after diagnosis date. Metformin adherence was defined as duration divided by the number of days that patient survived after their date of diagnosis. BMI was added as a continuous variable. Because of implausible coding (eg, a height of 2 inches; a weight >2000 pounds), 36 (0.18%) of patient’s BMI were set to missing.In addition, to understand different clinically relevant levels of BMI, we categorized BMI into the following categories: <18.5, 18.5 to 24, 25 to 29, 30 to 34, 35 to 40, and >40 (https://www.cdc.gov/obesity/adult/defining.html). Smoking status was categorized into “never smoker,” “previous smoker,” and “current smoker.” Smoking status for 9.5% of patients was missing and was coded in a separate level as “missing.”

Demographic variables, comorbid conditions, and cancer-specific variables were considered potential covariates. Elixhauser comorbidity category mappings from the International Classification of Diseases Clinical Modification, 9th and 10th editions^22,23 ‘ were used. From the potential set of comorbidities, the likelihood ratio test was used to determine covariate inclusion in the model²⁴. Tumor histologic types were classified into squamous cell, adenocarcinoma, and “other” histologies. Indicators for whether a patient had surgery, radiation, or none/other treatment were included.

Statistical Analysis

Analysis was performed using R 3.5.3 and packages survival, KMsurv, survminer, and Coxphw (R Foundation for Statistical Computing). Summary statistics for patient baseline characteristics and comorbidities for NSCLC patients taking metformin versus patients not taking metformin, conditioned on BMI, described with mean, standard deviation, median, and 25th and 75th percentile ranges, and standardized mean differences for quantitative variables and count, percentage, and Cramer V for qualitative variables are reported. The Kruskal-Wallis test was used to compare the groups for quantitative variables and the χ² test for categorical variables. The main analysis used a Cox proportional hazards regression model. The length of follow-up for OS was defined as the time from NSCLC diagnosis to death or the end of the study in days. An unadjusted Kaplan-Meier analysis stratified according to metformin use was performed, followed by 2 subanalyses involving the interactions between our variable of interest, “metformin use,” and BMI or smoking status. For interpretation and clinical value, we used the discretized BMI variable in the interaction analysis. The continuous BMI interaction with metformin is included in Table E1.

So that the interaction terms could be easily interpretable, new factors derived as the interaction of BMI and metformin use and the interaction of smoking status and metformin use were created. BMI ≥40 and metformin use was used as the reference category for the ordinal discretized BMI and metformin use interaction. For the smoking interaction, continuous BMI was included as a confounding factor. Mirroring the main analysis, covariate adjusted Cox proportional hazards models and unadjusted Kaplan-Meier analyses for our subanalyses were also performed.

Hazard ratios (HRs) and 95% confidence intervals are shown. Because the proportional hazard²⁵ assumption is often violated, causing the relative risk of a predictor to be under- or overestimated, weighted Cox regression as proposed by Schemper, Wakounig, and Heinze was performed, and these HR estimates are reported. Weighted Cox regression can account for time-dependent effects by weighting the contributions of each event to the partial likelihood according to the product of the survivor function and the inverse cumulative probability of follow-up at that time. Therefore, the weighting function is proportional to the expected number of subjects at risk if censoring had not occurred and yields estimates of average HRs (AHR).

RESULTS

Of the 19,940 patients included in our analysis, the median age was 69 (25th to 75th percentile, 63–69). Our cohort was primarily male (97.5%) and 30.2% had type 2 diabetes. Of the 19,940 patients 9996 had undergone surgery and 2173 had undergone radiation as their primary treatment.

Metformin use and nonuse specific information is summarized in Table 1. The patients receiving metformin tended to have a higher BMI (29.87 [SD, 5.97] vs 26.10 [SD, 5.58]; P < .001), and they contained a slightly higher percentage of patients who underwent surgery as opposed to radiation as a primary treatment.

TABLE 1.

Descriptive statistics for metformin users compared with non-metformin users in a stage 1 non-small cell lung cancer cohort

	Mean (SD)/median (25th-75th percentile)		Effect size,* standardized mean difference
	No metformin use (n = 17,779)	Metformin use (n = 2141)
Age, y	69.21 (9.29)/69 (63–76)	68.64 (7.91)/68 (63–74)	0.063 (P <.001)
OS, d	1147.05 (639.03)/1221 (514–1825)	1271.08 (599.45)/1533 (719–1825)	−0.195 (P <.001)
BMI	26.10 (5.58)/29.545 (22.336–29.193)	29.87 (5.97)/29.49 (25.846–33.270)	−0.671 (P <.001)
	n (%)	n (%)	Cramer V
Male sex	17,343 (97.5)	2090 (96.9)	0.003 (P = .67)
Histology-ACA	6699 (33.6)	805 (37.2)	-
Histology-squamous	6337 (35.6)	840 (49.7)	-
Histology-other	4751 (26.7)	495 (23.1)	0.029 (P = .003)
Smoker status (current)	9444 (53.1)	970 (45.3)
Smoker status (never)	448 (2.5)	65 (3)
Smoker status (previous)	6224 (35)	928 (43.3)	0.06 (P < .001)
Diabetes mellitus, type 2	3883 (21.8)	2141 (100)	N/A
Surgery	8794 (49.5%)	1202 (56.1%)	-
Radiation	1973 (11.1%)	200 (9.3%)	0.0418 (P < .001)

OS, Overall survival; BMI, body mass index; ACA, adenocarcinoma; N/A, not applicable.

Primary Analysis

To determine the effect of metformin use on unadjusted OS, Kaplan-Meier analysis was performed. This showed a survival advantage for patients receiving metformin (P<.001; Figure 1, A). For the Cox regression analysis, adjusting the model by taking demographic characteristics, comorbid conditions, and cancer-associated variables into account, metformin users still showed improved survival (Figure 1, B; HR, 0.84; 95% CI, 0.779–0.905; P<.001).

FIGURE 1.

Unadjusted and adjusted overall survival for metformin (MET) use with 95% confidence intervals. A, Kaplan-Meier unadjusted curve analysis with number at risk table. Number at risk is unadjusted. B, Marginal Cox proportional hazards regression adjusted analysis with curves produced balance with respect to the confounding variables (using the mean of quantitative variables and the reference category for qualitative variables).

To identify potential combined effects of covariates on the relative hazard risks with and without metformin use, the global Shoefeld test was applied to the Cox proportional hazards assumption made in the primary analysis model. Significant results (X² = 146; P < .001) were found for the following variables: pulmonary, histology, BMI, diabetes, surgery or radiation, and metformin use. Therefore, weighted Cox proportional hazards regression analyses were performed, and AHRs and their respective 95% CIs are reported (Figure 2). Our main treatment, metformin (AHR, 0.817; 95% CI, 0.756–0.882); BMI (AHR, 0.976; 95% CI, 0.972–0.981); never smoker versus current smoker status (AHR, 0.717; 95% CI, 0.628–0.819); and previous smoker versus current smoker status (AHR, 0.837; 95% CI, 0.799–0.876) remained significantly associated with improved OS.

FIGURE 2.

Cox regression hazard ratios (HRs) and 95% confidence intervals (95% CIs) and weighted Cox regression averaged HRs (AHRs) and 95% CIs for the primary analysis of assessing overall survival and metformin use for stage 1 non-small cell lung cancer patients. AHR and HR of 1.00 indicates lack of association, >1.00 implies an increased risk of death, and <1.00 implies a decreased risk of death in the forestplot. Error bars indicate 95% CI. The size of the markers in the plot correspond to the total number of cases within each category/level. BMI, Body mass index; CHF.

Interaction Between Metformin Use and BMI

In a subsequent Kaplan-Meier analysis of the role of BMI in metformin’s prosurvival effect, patients using metformin with an ordinal discretized BMI showed a higher survival rate than other groups (P < .001; Figure E1). For the adjusted analysis using Cox regression, the derived interaction term had a statistically significant Wald value (X² = 3268.417; P < .001), showing an interaction for BMI and metformin use. In addition, as shown in Figure E2, high BMI (ie, BMI ≥25) metformin users showed improved survival compared with all other categories with BMI 30 to <35 and metformin use having the highest improved survival rate. Compared with a BMI ≥40 and metformin use, the highest category of BMI with metformin use, was statistically different than a BMI <18.5, BMI ≥40 and no metformin use, and BMI 18.5 to <25 and no metformin use (BMI <18.5 and metformin: HR, 2.00 [1.24–3.23]; AHR, 1.83 [1.14–2.95]; [P = .012], BMI <18.5 and no metformin: HR, 2.05 [1.55–2.72]; AHR, 2.086 [1.55–2.79]; [P < .001], BMI 18.5 to <25 and no metformin: HR, 1.37 [1.05–1.80]; AHR, 1.4 [1.05–1.85]; [P = .021], and BMI >40 and no metformin: HR, 1.37 [1.02–1.85]; AHR, 1.37 [1.00–1.86]; [P = .049]). The global test for Cox proportional hazards assumption for the interaction BMI and metformin use was significant (X² = 147; P < .001) with the variables histology, the interaction, diabetes, insu-lin, surgery or radiation, and pulmonary significant and therefore, AHRs are provided as well. In addition, Tukey pairwise multiple comparisons were run (results are sum-marized in Table E1 and Figure 3). In Figure 3, we show that in the BMI category 25 to <30, patients not taking metformin had worse survival compared with those taking metformin (difference = 0.203; P = .025). In addition, those not taking metformin in the ≥40 category had worse survival than those taking metformin in the BMI cate-gories 30 to <35 and 35 to <40 (30 to <35: 0.411 [P < .001]; 35 to <40: 0.386 [P = .04]). Interaction term Cox regression estimates are also provided (Table E2) to show that the pairwise comparison estimates are similar to running a main effect and interaction model.

FIGURE 3.

Tukey pairwise comparisons for Cox regression hazard ratio estimates and 95% confidence intervals for metformin and body mass index (BMI). Blue color indicates P<.05. Red color indicates P<.1 level. Yes/No refers to metformin use; BMI categories are<18.5, 18.5 to<25, 25 to ^Q34 <30, 30 to<35, 35 to<40, and ≥40.

Interaction for Metformin Use and Smoking Status

In the analysis of patient smoker status at diagnosis and the survival benefit linked to metformin, never smokers were excluded because they formed a very small proportion of the entire population. Kaplan-Meier curves revealed an interaction effect for current or previous smoker status and metformin use (Figure 4, A). When the model was adjusted for covariates using Cox regression, previous smokers using metformin had an improved survival compared with all other categories (Figure 4, B; Table E2; Wald test [X²] = 2997.048; P < .001). Pairwise multiple comparisons (Table E1) showed that the curves for current smokers using metformin and previous smokers without metformin use and the curves for using metformin and being a current or previous smoker were not significantly different (P = .95; P = .28). In addition, using metformin for current smokers improved survival (difference, 0.22; P < .001). Further analysis is needed to determine whether metformin use will differentially attenuate mortality risk in current versus past smokers as postulated by others.²¹ The global test for Cox proportional hazards assumption for the interaction for smoking status and metformin use was significant (X² = 135; P .001) with the variables, histology, the interaction variable, insulin, BMI, diabetes, surgery or radiation, and pulmonary disease being identified as significant covariates. Similar to the interaction for BMI and metformin use, the main effect and interaction model as well as pairwise comparisons yielded results in line with the model presented (Table E2).

FIGURE 4.

Unadjusted and adjusted overall survival curves for the interaction of metformin (Met) use and smoking status with 95% confidence bands. A, Kaplan-Meier unadjusted curve analysis with number at risk table. Number at risk is unadjusted. B, Marginal Cox proportional hazards regression adjusted analysis with curves produced balance with respect to the confounding variables (using the mean of quantitative variables and the reference category for qualitative variables).

DISCUSSION

This large retrospective study showed an association between metformin use and improved OS of patients with stage I NSCLC. In these patients, metformin use was associated with a 5-year all-cause mortality advantage of 18%, and this benefit was more robust in overweight and obese patients and greater in previous smokers compared with current smokers.

The ability of metformin to extend patient survival and enhance response rates in multiple cancer forms has been previously described.^26,27 However, these anticancer effects are not completely understood. Because type 2 diabetes mellitus itself is associated with heightened cancer risk,²⁶ it stands to reason that metformin users might experience better survival outcomes because of amelioration of the disease. However, studies have yielded mixed results concerning the effects of diabetes on cancer-specific survival outcomes,²⁸ and treating diabetes with insulin analogs has been tied to increased cancer risk in some studies^9,29–32 and a lack of a beneficial effects on cancer development in others.^33,34 These observations suggest that metformin’s diabetes-correcting effects are not likely to account for the marked survival benefits this drug imparts in NSCLC patients. Future studies including the previously untapped nondiabetic (ie, recreational, off-label) metformin user populations will no doubt shed light on this matter.

A number of potential mechanisms for metformin’s antitumor effects (context-dependent and otherwise) are suggested by the literature. For instance, metformin is a known activator of the energy stress sensor known as adenosine-mono-phosphate-activated protein kinase (AMPK). This key metabolic regulator governs a host of cellular processes including protein synthesis, the uptake of important metabolic and biosynthetic fuels, and energy-generating processes. It is also a known negative regulator of mammalian target of rapamycin (mTOR), another central element of cellular metabolism. The ability to trigger AMPK activity in cancer cells can, in turn, reduce mTOR activity—a turn of events with decidedly negative implications for the proliferation and survival of malignant cells.

In addition to playing critical roles governing tumor cell biology, these hubs of metabolic regulation are also well known to direct the function and fitness of leukocyte subsets critical to the antitumor immune response. By activating AMPK while presumably suppressing the mTOR cascade, metformin can be expected to sway processes such as T-helper cell differentiation³⁵ and CD8⁺ T -cells memory/ effector state determination³⁶ which are highly sensitive to regulation by mTOR. In line with metformin’s known effects on immune-relevant processes, the drug’s potential as an agent for reprogramming the antitumor immune response has been the focus of several recent studies. Particularly, the effects of metformin on the CD8⁺ T-cell component of the antitumor immune response was highlighted by work of Eikawa and colleagues.¹³ In this study, the drug’s ability to reinvigorate exhausted T cells in the cancer setting was shown. Specifically, treatment of mice with metformin resulted in better infiltration of solid tumors by T cells that were more capable of producing inflammatory cytokines and resistant to apoptosis compared with vehicle control mice. Accordingly, metformin treatment suppressed tumor growth in this study.¹³ Another report showed that metformin enhances intratumoral T-cell activity indirectly by counteracting suppressive metabolic elements of the tumor microenvironment.³⁷ Thus, it is likely that metformin improves NSCLC survival outcomes by enhancing antitumor immunity.

Although the precise mechanism responsible for metformin’s prosurvival effects in NSCLC patients remains unknown, there is mounting evidence that obesity is linked to specific changes in the immune defenses that are likely to undercut effective tumor eradication and control. It is known that obesity induces a state of chronic metainflammation typified by chronic cytokine production, widespread dysfunction of innate and adaptive immune cells, and premature immune aging that yields an abundance of activated yet exhausted and dysfunctional T cells.^38,39 Notably, obesity has been linked to the upregulation of immune checkpoint molecules, which hinder effective antitumor immune responses. Preclinical studies show that obese mice support more robust tumor growth than normal control mice while harboring CD8⁺ T cells with high levels of surface markers (PD-1, LAG3, and TIM3) and gene expression profiles associated with exhaustion and immune suppression.⁴⁰ In line with a role for metformin in the reprogramming of the immune tumor microenvironment, we previously reported that use of the drug by advanced NSCLC patients was associated with the downregulation of several immune checkpoint molecules with potent immunosuppressive function (eg, PD-1, CTLA-4).¹⁹ Metformin might preferentially reprogram antitumor immunity in high-BMI individuals by modulating responsiveness to the hormone, leptin,⁴¹ a known regulator of the hunger response and energy balance that is also markedly upregulated in obesity. The immunomodulatory capacity of this hormone has recently been reported, and the ability of AMPK to dampen leptin responsiveness has as well. Interestingly, leptin levels are correlated with PD-1 levels on human and murine T cells. Disrupting leptin signaling in T cells undermined the exhausted phenotype typical in obese mice and improved the efficacy of anti-PD-1 immunotherapy.⁴⁰

Another possible mechanism of action would be “sugar steal syndrome” in which reducing insulin resistance moves sugar from the blood stream into the muscles and away from the tumor. This premise is supported by the observations that tumors are highly dependent on glucose availability and that ketogenic diets might be helpful in prolonging survival in cancer.³⁹

Several limitations of the study are acknowledged. This was a retrospective study with the attendant concerns for bias that such studies entail. In addition, the cohort was primarily male (97.5%), and therefore, a sex effect could not be measured. Further analysis should be done for an equally balanced sex cohort. In addition, 38.9% of patients did not have a recorded surgery or radiation code in the cohort and therefore, these treatments might have been given outside of the VA hospital. Although information on several comorbidities was available, more granular information was not. For example, pulmonary function measures, a critical determinant of long-term survival in patients with chronic obstructive pulmonary disease, a common comorbidity in these patients, were unavailable. In the absence of this end point, the possibility of other explanations for our observation cannot be excluded. Another inferential limitation of this study is that the mechanism of metformin’s benefit cannot be delineated. It is entirely possible that the benefit of metformin might be partly or largely because of its nononcologic health, especially because our end point was OS and not recurrence-free survival. For example, metformin has been shown to have beneficial effects on aging and longevity in preclinical models⁴² and decreasing risk factors of cardiovascular disease in humans.⁴³ It is also unclear if the benefit associated with metformin can be extrapolated to administration of this drug to nondiabetic patients, because all patients receiving metformin in this study were diabetic. This question can only be answered by a prospective clinical trial.

It is important to note that the 5-year survival of even stage I NSCLC is low. It is expected that several of these patients will end up with a higher pathologic stage, and staging methods over this time period (2000–2016) have dramatically improved, both factors leading to an apparent improved survival in pathologic stage I disease. However, these patients do not have any proven adjuvant treatment options to move the needle on this relatively disappointing outcome. In this context, metformin therapy represents a potential important adjunct to surgery in the care of patients with stage I NSCLC. Advantages of the use of metformin include its long safety record, favorable side effect profile, and low cost. The effect size of 5% 5-year survival advantage is impressive considering the landmark International Adjuvant Therapy Trial showed a 5-year survival advantage of only 4.1%² with administration of adjuvant chemotherapy, a treatment associated with greater morbidity.

CONCLUSIONS

Further clinical trials of metformin in nondiabetic patients with early stage lung cancer would have merit and should be urgently conducted to verify and extend our observations. Our observations also suggest that overweight patients and previous smokers might be more likely to benefit from metformin, a finding that can help design more efficient trials of this agent.

FIGURE 5.

Non–small cell lung cancer (NSCLC) is a major cause of cancer deaths and whereas metformin use has been shown to improve outcomes in NSCLC patients, disparate results across cohorts imply confounding with patient-specific characteristics. In a retrospective cohort of 19,940 US veterans with stage I NSCLC from 2000 to 2016, metformin use was associated with improved survival (average hazard ratio, 0.82; P<.001), especially in high body mass index (BMI) patients and former smokers. The right side of the illustration shows the unadjusted Kaplan–Meier curves stratified according to metformin use and the interactions between metformin use and BMI and smoking.

PERSPECTIVE.

Metformin use has been associated with improved survival in lung cancer patients. However, our analysis identifies specific populations that have a higher association between metformin use and survival. These results identify patient populations that need to be targeted with the design of rational trials for the use of this affordable and safe therapeutic agent.

Abbreviations and Acronyms

AHR

average hazard ratio

AMPK

adenosine-mono-phosphate-activated protein kinase

BMI

body mass index

HR

hazard ratio

mTOR

mammalian target of rapamycin

NSCLC

non-small cell lung cancer

OS

overall survival

VA

Veterans Affairs