Literature Review Refutes Claim of Cancer Risk with Insulin Glargine

^th

A controversy broke out last year after a German study by Hemkens and colleagues claimed that insulin glargine was associated with increased cancer risk. But in a presentation at the American Diabetes Association’s 70 Scientific Sessions, John M. Lachin, ScD, professor of biostatistics and epidemiology, and statistics, at The Biostatistics Center at the George Washington University, said that the data wasn’t sufficient to make that claim and there is, in fact, no evidence that insulin glargine is associated with increased cancer risk.

The Hemkens study consisted of 127,031 patients who were exposed for an average of 1.63 years and who were identified retrospectively; 95,804 were on human insulin, 23,855 were on insulin glargine. All patients were 18 or older who had no history of a malignancy for 3 years prior. Those who used insulin combinations and/or changed their insulin regimens were excluded.

But according to Lachin, who spoke during the session “Insulin Glargine and Cancer—Facts and Fallacies” on Sunday, June 27, there were several omitted covariates; this adjustment for observed factors cannot eliminate bias due to imbalances in unmeasured covariates. The study compared type 1 with type 2, the patients had different durations of diabetes, degrees of glucose control, and varying body mass indexes, and all types of malignancies weren’t reported. After the study in Germany was published, three more followed—one each from Scotland, England, and Sweden. But all 4 of these studies had issues with dose adjustment.

According to Pocock and Smeeth (2009), the Cox PH model is valid if covariable values for all subjects are obtained prior to the time of the event. But the German study wasn’t analyzed how a Cox PH model is supposed to be conducted. The German study computed the average dose for each subject over the entire period of follow-up, including doses after a diagnosis of cancer had been delivered. But despite this issue, Lachin explored the dose adjustments.

Lachin explained that the dose adjustment estimates the difference between glargine and human insulin if there was no difference in insulin dose. Statistical models essentially estimate the risk in the complete sample under the assumption that the average dose or distribution in both groups was the same. This brings up the pivotal question of whether the adjustment for insulin dose was statistically appropriate. Lachin asked, “What are the possible reasons for the imbalance in dose and does the statistically adjusted estimate have a meaningful population model interpretation?” Lachin explained that if the answer is no to the latter, than adjusting for dose could introduce more bias.

So despite these fallacies, Lachin investigated the how the possible dose adjustment had such a profound effect. He wanted to know ‘how does one go from a 14% reduction to a 14% increase in risk for glargine versus human insulin,” which is what the German study showed. When comparing glargine versus the NPH doses, negligible differences in the doses are needed to achieve comparable levels of glycemic control in randomized trials, which are the ideal type to test for something like this. This would mean that with all things being equal there should be no difference in dose.

According to Lachin, there could be several reasons for the dose imbalance. They could have confounded by indication, or possibly allocation bias. High or low doses may have been determined by unmeasured patient factors that were differently distributed within the groups, for instance a high glargine dose was only administered to severely ill patients. But whatever the reason, it is impossible to statistically adjust for these confounding factors because the adjusted analysis would be biased.

Citing a study by Ehninger and Schmidt (2009), Lachin said the reason for the dose imbalance was the difference in groups. For instance, those receiving human insulin had either type 1 or type 2 diabetes compared to the glargine group only having type 2; without type 1 patients in the glargine group its impossible to appropriately adjust the doses.

This intensive statistical evaluation of the Hemkens study led Lachin to several conclusions. First, the distribution of insulin doses in the study likely reflected the actual dose distribution in the population meeting the selection criteria. Therefore, the analysis that didn’t include the dose adjustment better reflects the relative risk of glargine versus human insulin in this population. Also, when analyzed for dose, there is no sign of increased risk with glargine, which is confirmed by all of the other studies, whether they were randomized or not.

Lachin also concluded that since the dose-adjusted increase in the risk with using glargine was due to a 42% higher risk in the 14% of glargine subjects who received more than 40 IU that it is possible, and probably even likely, that this risk is due to confounding by indication or even imbalances in other factors, such as the omitted covariates. And, since none of the other studies investigated the dose-effect on risk, there was no replication.

Ultimately, Lachin’s final conclusion is that without replicated evidence that insulin glargine at any dose is a cause of, or is associated with, an increase in the risk of malignancies, the previous claims to the contrary have no basis.