Introduction Clinical trial eligibility criteria can unfairly exclude patients or expose patients to risk if they are not justified by drug safety. Our prior analysis of acute myeloid leukemia (AML) trials identified unjustified variability in the use of criteria and in the limits placed when criteria were used (ASCO 2023 Abstract #6525). As additional safety data are available in later phase studies, we sought to assess if criteria were more justified in phase III versus phase II trials, and we extended our analysis to include acute lymphoblastic leukemia (ALL). Methods Trials and criteria were identified using ClinicalTrials.gov. Trials were included if they treated patients in the United States, were interventional, therapeutic, phase II or III, treated previously untreated adults with AML and/or ALL, were not cellular therapy trials, and had study start dates from 1/1/2010-12/31/2019. Drug safety data from labels, trial protocols, and previous study results available as of study start dates, as identified in the ClinicalTrials.gov record, were compared to systematically coded eligibility criteria following FDA publications.Primary endpoints were differences in discordant criteria use (presence of criteria without a known, relevant drug safety signal or absence of criteria with a signal) and limits (numerical values of drug safety- or criteria-based limits) between phase II and III studies assessed via Fisher's exact, rank sum, and median absolute deviation (MAD) testing. Regression modeling assessed differences in infectious disease (ID) exclusions by phase adjusted for disease type and trial sponsor, year, size, and drug safety profiles (drug-drug interaction and association with viral reactivation). Results There were 250 eligible trials, of which 203 were phase II and 47 phase III; 190 trials were in AML and 60 ALL. Similar proportions of phase II and III trials had exclusion criteria (with specific limits when applicable) for low left ventricular ejection fraction (LVEF; 36 vs 25%), high corrected QT interval (QTc; 21 vs 27%), presence of human immunodeficiency virus (HIV; 57 vs 47%), and hepatitis B (43 vs 38%) and C (42 vs 38%). More phase II than phase III trials had renal function (70 vs 53%; p=0.04), bilirubin (73 vs 49%; p<0.01), and alanine transaminase/aspartate transaminase (AST/ALT; 26 vs 64%; p<0.01) based exclusions. Table 1 shows proportions and odds ratios (OR) of discordant criteria by trial phase. OR of discordance for renal function (1.90), AST/ALT (2.30), and bilirubin (2.45) were significantly higher for phase III studies (all p<0.05); no criteria were less discordant for phase III studies. Drug safety- and trial-based limits for renal and liver function are shown in Figure 1. Median differences between limits numerically improved from phase II to phase III (renal: 0.62 to 0.20; AST/ALT: 0.50 to 0; bilirubin: 0.20 to 0); only the change in renal function was significant (rank sum p=0.03). The dispersion of these differences was numerically smaller for renal function (phase III to III MAD ratio: 0.65) and bilirubin (0.91) and higher for AST/ALT (1.17); no changes in dispersion were significant. In multivariable regressions adjusted for other trial and drug safety factors, phase III trials were equally likely to exclude persons diagnosed with HIV and hepatitis B and C as phase II (adjusted OR 1.48, 1.37, 1.41; all p>0.05). Conclusions Despite increasing understanding of drug safety profiles in later phase trials, the use of eligibility criteria for phase III acute leukemia clinical trials were at least as discordant with known drug safety as earlier phase studies. When drug safety- and trial-based limits were present, phase III criteria were not consistently more aligned with drug safety and were equally likely to exclude persons diagnosed with HIV and hepatitis. Future phase III trials in acute leukemia should standardize and rigorously justify eligibility criteria to enhance patient representativeness and welfare.