BACKGROUND: Value-Based Healthcare (VBHC) focuses on Patient-Centred Outcomes (PCOs) to maximize healthcare value. Benchmarking across providers requires comparing outcomes among clinically and therapeutically comparable patients. However, traditional classification is based on biological features without accounting for the therapeutic trajectory. Establishing patient archetypes based on therapeutic pathways is therefore essential to prevent case-mix bias in VBHC benchmarking. The aim of this paper is to develop a model-based framework using Latent Class Analysis (LCA) to (i) identify patient archetypes in breast cancer and (ii) establish a probabilistic tool to classify new patients. METHODS: A cohort of 2485 patients diagnosed with breast cancer between 2018 and 2020 was followed-up. Demographic and treatment variables were collected at baseline and at 6-months follow-up. LCA was applied to derive archetypes in a training set (80%). Fit and classification quality were assessed using aBIC, entropy, and posterior probabilities. Tool classification robustness was evaluated using Random Forest (RF) in a test set (20%), and generalizability was tested in an external multicenter cohort (N = 352). RESULTS: A six-class solution (Arch-1-Arch-6) was selected, achieving optimal fit, parsimony, high entropy (0.932), and clinical interpretability. Evaluated against RF, the tool demonstrated high classification concordance (99.68% accuracy). In the external cohort, high individual classification certainty was maintained (mean posterior probability = 0.965), supporting consistent assignment across care settings. CONCLUSIONS: LCA provides a transparent, efficient and robust tool to identify patient archetypes and assign new patients based on therapeutic pathways, enabling fair benchmarking in VBHC. Prospective validation in broader cohorts is recommended to confirm transferability to other clinical conditions.