Metformin failure in people with type 2 diabetes is very common, particularly among those with high hemoglobin A1c levels at the time of diagnosis, new findings suggest.
An analysis of electronic health record data for more than 22,000 patients starting metformin at three U.S. clinical sites found that over 40% experienced metformin failure.
This was defined as either failure to achieve or maintain A1c less than 7% within 18 months or the use of additional glucose-lowering medications.
Other predictors that metformin use wouldn’t be successful included increasing age, male sex, and race/ethnicity. However, the latter ceased to be linked after adjustment for other clinical risk factors.
“Our study results suggest increased monitoring with potentially earlier treatment intensification to achieve glycemic control may be appropriate in patients with clinical parameters described in this paper,” Suzette J. Bielinski, PhD, and colleagues wrote.
“Further, these results call into question the ubiquitous use of metformin as the first-line therapy and suggest a more individualized approach may be needed to optimize therapy,” they added in their article, published online in the Journal of Clinical Endocrinology and Metabolism.
The study is also noteworthy in that it demonstrated the feasibility of using EHR data with a machine-learning approach to discover risk biomarkers, Dr. Bielinski, professor of epidemiology at the Mayo Clinic, Rochester, Minn., said in an interview.
“We wanted to repurpose clinical data to answer questions ... I think more studies using these types of techniques repurposing data meant for one thing could potentially impact care in other domains. ... If we can get the bang for the buck from all these data that we generate on people I just think it will improve health care and maybe save health care dollars.”
Baseline A1c strongest predictor of metformin failure
The investigators identified a total of 22,047 metformin initiators from three clinical primary care sites: the University of Mississippi’s Jackson centers, which serves a mostly African American population, the Mountain Park Health Center in Arizona, a seven-clinic federally qualified community health center in Phoenix that serves a mostly Latino population, and the Rochester Epidemiology Project, which includes the Mayo Clinic and serves a primarily White population.
Overall, a total of 43% (9,407) of patients met one of two criteria for metformin failure by 18 months. Among those, median time to failure on metformin was 3.9 months.
Unadjusted failure rates were higher among African Americans, Hispanics, and other racial groups, compared with non-Hispanic White patients.
However, the racial groups also differed by baseline characteristics. Mean A1c was 7.7% overall, 8.1% for the African American group, 7.9% for Asians, and 8.2% for Hispanics, compared with 7.6% for non-Hispanic Whites.
Of 150 clinical factors examined, higher A1c was the strongest predictor of metformin failure, with a rapid increase in risk appearing between 7.5% and 8.0%.
“The slope is steep. It gives us some clinical guidance,” Dr. Bielinski said.
Other variables positively correlated with metformin failure included “diabetes with complications,” increased age, and higher levels of potassium, triglycerides, heart rate, and mean cell hemoglobin.
Factors inversely correlated with metformin failure were having received screening for other suspected conditions and medical examination/evaluation, and lower levels of sodium, albumin, and HDL cholesterol.
Three variables – body mass index, LDL cholesterol, and creatinine – had a U-shaped relationship with metformin failure, so that both high and low values were associated with increased risk.
“The racial/ethnic differences disappeared once other clinical factors were considered suggesting that the biological response to metformin is similar regardless of race/ethnicity,” Dr. Bielinski and colleagues wrote.
They also noted that the abnormal lab results which correlated with metformin failure “likely represent biomarkers for chronic illnesses. However, the effect size for lab abnormalities was small compared with that of baseline A1c.”
Dr. Bielinski urged caution in interpreting the findings. “Electronic health records data have limitations. We have evidence that these people were prescribed metformin. We have no idea if they took it. ... I would really be hesitant to be too strong in making clinical recommendations.”
However, she said that the data are “suggestive to say maybe we need to have some kind of threshold where if someone comes in with an A1c of X that they go on dual therapy right away. I think this is opening the door to that.”
The authors reported no relevant financial relationships.
A version of this article first appeared on.