Fig 1: (A) Illustration of the interconnected challenges inherent to the successful development of a model-based approach for leveraging longitudinal ctDNA to support early clinical decision-making in oncology, namely, the access to highly longitudinal samples and the understanding of disease biology and bioinformatic nature of the data. (B) Database description: Weber et al. (2021) clinical study published in 2021. The dataset was shared with the authors without further obligations according to the policy of the Journal of Clinical Oncology Precision Oncology and according to the local patient-level sharing privacy rules. The dataset is composed of 167 patients with non-small cell lung cancer. The ctDNA sample collection was performed between 2015 and 2018. The Roche AVENIO assay was used for ctDNA measurements and reported as (average) mutant molecules per milliliter measured consistently at two time points (baseline, i.e., cycle 1 1 and 28 days in median, SD = 10 days). For 15 patients, the ctDNA value at both baseline and on treatment was recorded 0, making the calculation of change from baseline impossible. These patients were removed from the analysis. Patients suffered from stage IIIB (13.2%) or stage IV (86.8%) NSCLC; all were treated with checkpoint inhibitors (nivolumab for more than 80%). The overall response rate was 26.3% (n = 44), and the mean treatment duration was 13 cycles. Anonymized patient-level survival data were available. IMspire170: Roche sponsored study on patients with melanoma (Gogas et al., 2021) Data consisted of 411 patients, of which 209 patients were treated with the checkpoint inhibitor atezolizumab with the tyrosine kinase (MEK-) inhibitor cobimetinib and 202 patients with pembrolizumab. ctDNA measurements were reported as a ctDNA tumor fraction (cTF) based on aneuploidy and variant allele fraction (VAF) and were available for 241 patients (122 in the group treated with atezolizumab and cobimetinib and 119 in the group treated with pembrolizumab). Two time points were available: baseline (cycle 1 day 1) and cycle 2 day 1 (corresponding to 21 days for the pembrolizumab cohort and 28 days for the atezolizumab/cobimetinib cohort). Tumor size (sum of longest diameters) was available longitudinally for 409 patients out of the 411 patients. Patient-level survival data were available. OAK: Roche sponsored the study on non-small cell lung cancer patients treated with the checkpoint inhibitor atezolizumab or chemotherapy docetaxel (Rittmeyer et al., 2017). The study included 1,225 patients. ctDNA data were available for a subset only (n = 94). Our analysis included only the data from the atezolizumab arm (n = 613 and 46 with ctDNA data). The data from the chemotherapy arm were only used for developing the ctDNA time course model with the treatment arm regarded as a covariate. The Roche AVENIO assay was performed for ctDNA measurements and reported as mutant molecules per milliliter, and data at four time points were available: cycle 1 (baseline), cycle 2 (around 21 days), cycle 3 (around 42 days), and cycle 4 (around 63 days). These data were used for longitudinal modeling using the Stein et al. model. OAK ctDNA data analysis has been already published (Zou et al., 2021).