BACKGROUND: We examined how crossover therapy might affect the association between progression-free survival (PFS) and overall survival (OS) in non-small cell lung cancer (NSCLC). METHODS: We extracted PFS- and OS-hazard ratios (HRs) in phase III trials of molecular-targeted agents for advanced NSCLC. Their relationship was modeled in a linear function with the coefficient of determination (R-squared) to assess the correlation between PFS and OS. RESULTS: Thirty-four trials with 35 pairs for the investigational and reference arms were identified (24,158 patients). Overall, there was little correlation between PFS- and OS-HRs (R-squared = 0.14), suggesting PFS-HR could account only for 14% of variation in OS-HR. The median proportion of crossover therapy per trial was 20%. If patients seldom crossed over (none or <1%), the association between PFS- and OS-HRs was strong (R-squared = 0.69). When the proportion of crossover was ≥1%, however, R-squared declined considerably (≥1% to <20% crossover, R-squared = 0.27; ≥20% to <40%, R-squared = 0.06; and ≥40%, R-squared = 0.27). CONCLUSIONS: A PFS advantage seldom is associated with an OS advantage any longer. Our analysis suggests this is due to a high level of crossover now that an increasing number of active agents are available for NSCLC.