Background: PD-L1 tumor proportion score (TPS) is often used to determine eligibility for first line therapy with immune checkpoint inhibitors (ICIs) in advanced non-small cell lung cancer (NSCLC). However, PD-L1 expression can vary over time and between tumor sites, potentially leading to inaccurate patient stratification. Therefore, it is critical to understand the clinicopathologic and genomic factors that are associated with PD-L1 changes in NSCLC. Methods: Clinicopathologic and genomic data were collected from patients with NSCLC and quantitative PD-L1 immunohistochemistry (IHC) on at least two different biopsies. NSCLC biopsies were categorized as PD-L1 negative, low, and high if they had a PD-L1 TPS < 1%, 1-49%, and ≥50%, respectively. Intrapatient changes in PD-L1 TPS between samples (DPD-L1) were defined as follows: major decrease (decrease in PD-L1 TPS from ≥50% to < 50% or from ≥1% to < 1%), major increase (increase in PD-L1 TPS from < 1% to ≥1% or < 50% to ≥50%), and non-major change (all other cases). Next-generation sequencing (NGS) was used to evaluate copy number (CN) variations at the CD274 locus, which encodes PD-L1. Wilcoxon and Kruskal-Wallis rank sum tests were used to analyze continuous variables and Fisher’s exact test was used to analyze categorical variables. Results: Among 250 patients with NSCLC with PD-L1 IHC assays performed on at least two distinct tissue samples, PD-L1 TPS of the first biopsy was < 1% in 104 (41.6%), 1-49% in 80 (32.0%), and ≥50% in 66 (26.4%) samples, for a median PD-L1 TPS of 2% (range: 0% to 100%). When intrapatient DPD-L1 was examined, there were major decreases and major increases in PD-L1 TPS in 49 (19.6%) and 65 (26.0%) cases, respectively, and non-major changes were observed in the remaining 136 samples (54.4%), with a median DPD-L1 of 0% (range: -90% to +90%). Baseline PD-L1 TPS and DPD-L1 were not significantly affected by histology, smoking status, sex, or treatment. Among 219 NSCLC samples that underwent tissue NGS and had full CN data available, the median PD-L1 TPS differed significantly based on CD274 CN: PD-L1 TPS 1% with single copy deletion vs. 5% with copy neutral vs. 42.5% with low amplification vs. 97.5% with high amplification (p < 0.01). Among 56 patients with paired PD-L1 TPS and NGS on both samples, there was a significant difference in median DPD-L1 according to CD274 CN change: DPD-L1 TPS -49% with acquired CD274 CN loss vs. 0% with no major change in CD274 CN vs. +1.75% with acquired CD274 CN gain (p = 0.01). Conclusions: PD-L1 TPS varies within the same patient in almost half of NSCLC cases, with few clinicopathologic correlates of change in expression. Variation in PD-L1 TPS correlates with changes in CD274 CN across biopsies. These findings suggest a genomic correlate to predict PD-L1 TPS across samples, as well as a potential complementary method in determining in ICI initiation.