"AI" in this study means aromatase inhibitor, not artificial intelligence: sticking to treatment raises cardiometabolic risk but lowers mortality

By Medscape · October 12, 2026.

A note up front: Medscape's original headline refers to "AI adherence," and here AI stands for *aromatase inhibitors*, hormone therapy drugs used against breast cancer. It has nothing to do with artificial intelligence. We point this out because the acronym is misleading, and because the article itself states that it was produced with editorial tools that include AI, with review by human editors.

The key finding: in postmenopausal women with early-stage HR-positive (hormone-sensitive) breast cancer, five years of adherence to aromatase inhibitors was associated with a higher 10-year risk of hypertension, dyslipidemia (abnormal blood fat levels) and diabetes. Even so, adherence to either of the two endocrine therapies, aromatase inhibitors or tamoxifen, was associated with lower all-cause mortality.

The study, led by Jorge R. Ledesma (Kaiser Permanente Northern California) and published in JNCI Cancer Spectrum, is a prospective cohort of 8,365 women with stage I–III cancer (median age, 64) who began treatment between 2005 and 2013. Patients were considered adherent if more than 80% of days were covered by medication, based on pharmacy dispensing data. The authors estimated counterfactual scenarios ("what would have happened with a different duration of adherence?") using doubly robust estimation with Super Learner algorithms.

The figures, as reported by Medscape: 42.7% were adherent for five years or more; 12.3% for 3 to 4 years; 12.9% for 1 to 2 years; and 12.8% were non-adherent. Compared with non-adherent patients, five years on aromatase inhibitors was associated with more hypertension (9.37%; 95% CI: 4.54-14.2), dyslipidemia (6.84%; 1.76-11.9) and diabetes (3.29%; 0.58-6.00). Tamoxifen was associated only with more hypertension (9.38%; 2.71-16.1). In a direct comparison among adherent patients, aromatase inhibitors showed a 9.86% higher incidence of dyslipidemia than tamoxifen (5.09-14.6).

On the other side of the scale, five-year adherence was associated with lower 10-year mortality: all-cause (10.7%; 7.74-13.7), breast cancer (5.37%; 3.36-7.38) and cardiovascular disease (2.89%; 1.24-4.53).

Methodological caveats. This is an observational study: it shows association, not causation. According to the authors themselves, there may be residual confounding (diet, exercise, recurrence, access to care), adherence was measured by dispensing rather than actual intake, and the outcomes are binary variables defined by clinical algorithms, not biomarkers such as cholesterol or glucose. In addition, the confidence intervals are wide, especially for diabetes. The authors speculate that adherent patients are more engaged with the healthcare system, which allows their risk factors to be managed.

Our take: the useful message is not "beware of the treatment," but the opposite. The net balance favors staying on therapy, and the cardiometabolic cost is a manageable risk through screening, diet and exercise, as the authors recommend. More broadly, with more than 4 million breast cancer survivors in the U.S., oncology is starting to measure not only whether patients survive, but how they live in the years that follow.

This is where our long-term thesis fits in. When cancer stops being a death sentence, medicine shifts toward managing the side effects of curing it. That is good news disguised as a problem. And it is an area where large-scale data analysis, including real AI, can help: screening those at higher risk earlier and personalizing therapy. Today that is a reasonable aspiration, not something demonstrated by this study, which does not use AI as a clinical tool.