Because of the no

Abstract

Counting or not counting nutrients in HCL systems: mental load as a clinical variable

Before every meal, a user of hybrid closed-loop (HCL) systems must observe, estimate, calculate, and decide. Five or six times a day, every day, with no possibility of suspending it. This silent arithmetic rarely appears in our metrics, yet it constitutes a clinical variable: the mental load that sustains treatment.

Sustained cognitive effort becomes depleted, and with it the quality of decisions: this is decision fatigue. It appears as skipped calculations, unannounced meals, and abandonment of an unsustainable task. It occurs in the child who eats lunch at school with no adult to validate the count³, and in the adult managing midday between meetings and caregiving responsibilities. Attention is a finite resource, and counting does not wait until it is available, it overlaps with work, study, and the care of others.

Simplified meal announcement offers a concrete alternative: instead of quantifying grams, the user declares meal size through personalized categories — small, medium, or large — calibrated with the care team, while the algorithm takes over postprandial correction. Developments such as Medtrum's AutoMeal extend this logic to the mere declaration of meal type. This does not mean abandoning nutritional education, but redistributing the load toward initial calibration and follow-up.

The SMASH trial, a randomized, crossover trail in youth aged 12 to 20, showed the noninferiority of simplified announcement compared with precise counting¹. In real-world conditions, an analysis of 195,529 users across four countries confirmed it for time in range, time below range, and mean glucose².

The argument extends to fat and protein. Their postprandial effect is real, but delayed, prolonged, and variable between individuals and between days⁴. Anticipating it requires predicting a response that is not constant even within the same person. That slow profile is precisely what algorithms correct best reactively, counting them transfers to the user a prediction the system resolves by observation.

Algorithms were designed to tolerate imprecision. Demanding constant precision turns counting into a barrier to access, it excludes from automation those who need it most².

Keeping the calculation on the patient's side, when the system can handle it, has ceased to be a technical decision and become a clinical one. The discussion is no longer whether counting works, but how much of that load we continue to delegate.