Intention-to-treat and per-protocol analyses provide complementary information about the usefulness of therapies. While intention-to-treat analyses of trials that tested low-protein diets remain debated, per-protocol analyses of the same trials show that low protein intake actually reduces the risk of kidney failure. Per-protocol analyses are notoriously open to bias but intention-to-treat analyses are less immaculate than commonly realized because they critically depend on adherence to the treatment being tested and therefore may not be directly relevant for informing clinical decisions when different degrees of adherence to therapy occur. Over the last 20 years new statistical techniques censoring patients at the time when they become uncompliant and that adjust for confounding attributable to incomplete adherence, i.e. for prognostic factors that predict adherence to treatment, have been developed. These techniques can be usefully applied to reanalyse the Modification Diet in Renal Disease (MDRD) and other trials. Intensive surveillance of patients on a low-protein diet is fundamental for early detection of malnutrition. However, the resources demanded by such surveillance are likely superior to the actual dietitians workforce dedicated to follow-up of the chronic kidney disease (CKD) population. Surveillance efforts may perhaps be preferentially devoted to preselected patients, i.e. patients that maintain good compliance and an adequate metabolic and nutritional status, while patients who are resistant to educational efforts and show persisting uncompliance may be reallocated to a diet with a higher protein content, which poses a lower risk of malnutrition and other adverse health outcomes.