Accurately predicting patient survival and progression-free survival remains challenging due to substantial inter- and intra-tumor heterogeneity, which drives variable drug responses and eventual relapse. A modeling framework is developed to address this by integrating heterogeneity inferred from single-cell data. Variability in drug sensitivity is represented as lognormal distributions derived from transcriptomic similarity between single cells and cell lines using Spearman correlation. These distributions parameterize growth rate–based tumor dynamics under continuous drug exposure, enabling simulation of treatment response over time. The framework captures both inter- and intra-patient variability and extends to combination therapies through multivariate modeling. Validation in preclinical mouse models and exploratory clinical datasets shows promising agreement with observed outcomes, supporting the potential of single-cell-informed approaches for predicting therapeutic response.