3088 Background: We studied clinical and radiological parameters influencing DIP in patients (pts) participating in phase I clinical trials, aiming to investigate predictive factors affecting DIP, in particular those affecting outcome. Methods: 2439 consecutive stage IV cancer pts on phase I clinical trials from 2007 to 2017 were identified. Pts with respiratory symptoms or abnormal lung imaging were reviewed in detail, with longitudinal analysis of imaging by an experienced radiologist. R-DIP was categorized according to internationally recognized criteria. Results: 60 pts developed R-DIP (overall incidence 2.5%); most frequent in pts receiving drug conjugates (31.1%) followed by targeted therapies (8.3%). Hypersensitivity pneumonitis was most common (33.3%) followed by non-specific interstitial pneumonitis (30%) and cryptogenic organising pneumonitis (26.7%). 45% pts who developed R-DIP were clinically asymptomatic. The number of affected lobes (OR 1.47, 95% CI: 1.19-1.81, p < 0.001) and the pattern of R-DIP (OR 5.83 for ARDS, 95% CI: 0.38-90.26, p = 0.002) were significantly associated with a higher CTCAE pneumonitis grading. 23% pts (14/60) had investigational medicinal product (IMP) temporarily discontinued or had a dose reduction while 42% pts (25/60) had IMP permanently discontinued. 48% pts were treated with steroids. The number of affected lobes, pattern of R-DIP and steroid therapy did not influence an improvement in R-DIP (p = 0.65, 0.27 and 0.23 respectively). Continuation of treatment resulted in worsening of DIP in 42.9% of cases. The only predictive factor for an improvement in DIP was an interruption of treatment (OR 0.05, 95% CI: 0.01-0.35, p = 0.01). 14 pts were retreated with a reoccurrence of R-DIP in 4 pts (28.6%). Conclusions: R-DIP from novel agents in early phase clinical trials presents in varied radiological patterns, with findings often preceding clinical symptoms. Treatment interruption leads to improvement of DIP and should be considered early. Close clinical and radiological surveillance is recommended should IMP be restarted.