Our study has identified an association between the use of metformin and/or a TZD and a lower likelihood of developing lung cancer in patients with DM. In addition, those who developed lung cancer while receiving metformin were more likely to present with metastatic disease, had a lower survival rate (after correction for stage and age), and had a different histology distribution than those who were not receiving either of these drugs.
For an association to be considered causal, several criteria need to be considered. Ideally, there should be a strong association that increases as the exposure increases; the exposure should occur before the disease appears; the association should not conflict significantly with what is already known of the disease; the association should be biologically plausible; and the association cannot be due to any source of error.
The association between the development of lung cancer and the use of metformin and/or a TZD reported here was strong (control group was 1.5 times more likely to have used these medications, OR 0.47) and increased with greater exposure duration (control group was 2.3 times more likely to have used one of these medications for longer than 24 months). All exposures in the cancer group occurred prior to the diagnosis of cancer.
The associations identified in this study are consistent with prior reports. A population based study of 11876 diabetics included 923 subjects with a cancer. Those with cancer were less likely to be receiving metformin than those without (36.4% vs. 39.7%) [14
]. In a study of 10309 diabetics, cancer related mortality was reported in 3.5% of metformin users, 5.9% of sufonylurea users, and 5.8% of insulin users [15
]. A report of 22,621 females with DM using oral glucose lowering drugs included 305 women with breast cancer. Only 17 of the women with breast cancer had used metformin
5 years, as compared to 120 matched controls (OR 0.44) [16
]. In a neoadjuvant chemotherapy trial for breast cancer 24% of 68 subjects with DM receiving metformin had a complete response to therapy, whereas only 8% of 87 who were not taking metformin, and 16% on non-diabetics did [17
]. Metformin use in a mouse model of lung cancer development was shown to reduce the lung tumor burden by 53-72% [18
]. Finally, in a cohort trial of 87,678 newly diagnosed diabetics over age 40, 1,371 lung cancers developed over a 7 year timeframe. A 33% reduction in lung cancer risk was seen among TZD users after adjustment for covariates [19
The associations reported in this study are biologically plausible. Metformin impairs mitochondrial ATP production inducing an energy stress at the cellular level. AMP-activated protein kinase (AMPK) is a key cellular energy sensor. During energy stresses, when the ratio of AMP:ATP increases, AMP binds to AMPK, allowing its activation through phosphorylation by the constituitively active upstream kinase LKB1. Activation of AMPK leads to a shift from energy depleting synthetic metabolic pathways to energy conserving metabolic processes; to an increase in P53 activity, resulting in cell cycle arrest at the G1/S checkpoint, apoptosis, and activation of autophagy pathways; and to a decrease in mTOR activity resulting in a decrease in protein synthesis, cell growth, and a reduction in survivin levels [20
]. AMPK has a direct link with cell proliferation during the M-phase of the cell cycle, represses mitosis-gene families, and cytokinesis genes [27
]. In LKB1 and P53 mutant tumors there may be an inability to compensate for metformin induced energy stresses [25
]. Additional potentially protective actions of metformin include decreases in circulating insulin levels, as well as inhibition of phosphorylation of IGF-1R/IR, Akt, ERK, and mTOR [12
]. It is possible that lung cancers that develop in the face of metformin use are less sensitive to energy stresses or bypass the pathways influenced by metformin use, resulting in a more aggressive cancer phenotype and the observation of increased metastatic disease, and shorter survival. Thiazolidinediones are ligands for PPAR-gamma, a transcription factor highly expressed in cancer cell lines [28
]. Exposure to TZDs in vitro leads to cell cycle arrest, apoptosis, and/or redifferentiation through PPAR-gamma dependent and independent actions [29
The potential for the identified associations to be related to sources of error can never be discounted in retrospective epidemiologic studies. Attempts to minimize the potential influence of random errors and biases were made by including all of the identified lung cancer subjects, having a relatively large sample size, by directly reviewing each of the study subjects’ medical records, and by matching controls based on known and identifiable risk factors. An effect-cause error was avoided by insuring medication use prior to lung cancer diagnosis was documented. The greatest remaining potential for error is the presence of a confounding variable. The level of glucose control and the BMI are potential confounders. The level of glucose control, as estimated by the HbA1c values, did not differ between cancer and control subjects, or within the groups of lung cancer subjects who were and were not receiving metformin and/or a TZD. The BMI of the control group was slightly higher than the cancer group. This is in keeping with other reports suggesting a potential protective effect of obesity on the risk of developing lung cancer [4
]. It is difficult to know if the lower BMI in the cancer group represents the protective effect of obesity or if it is related to weight loss from the cancer itself. There was no difference in the BMI of the cancer subjects who were and those who were not receiving metformin and/or a TZD. These factors were not found to be associated with lung cancer risk or survival when assessed with multivariate statistics. There has been a secular trend in lung cancer histology in the United States. An influence of this trend, combined with an increased use of the studied DM medications over time, on the association of histology distribution with medication use cannot be discounted. When only subjects who developed their lung cancer after 1995 were included in the analysis (480 of 507) the results were identical to those presented in the results section (results not shown). Other unknown confounders, such as the potential selection of DM medication based on symptoms or co-morbidities that are associated with lung cancer, cannot be controlled for with this study design. Overall, the strength and consistency of the associations identified support the findings despite the design.