CohortFlow: A Steerable Clinical Cohort Curation Agent with Localized Semantic Correction

Language model agents for clinical cohort curation should accommodate evolving eligibility criteria without repeatedly processing unchanged requirements. We introduce a steerable agent that represents natural language corrections as localized updates to a structured cohort specification and reconciles them with a persistent execution graph. Explicitly recorded semantic dependencies enable affected computations to be selectively invalidated while preserving valid prior results. On 100 scenarios derived from clinical-trial eligibility criteria, comprising 126 correction turns, localized steering achieves 75.5\% requirement micro-F1 versus 76.2\% for full regeneration using the same 8B model, with no statistically significant difference in requirement-level accuracy ($p=0.510$). It reduces mean correction latency by 64.3\% after completion and 44.9\% during execution, with 35.9\% fewer model calls. In a separate proof-of-concept evaluation over MIMIC, all 13 cohort states execute successfully, while correction trajectories demonstrate selective preservation of unaffected computations and recomputation of changed or dependent nodes. Together, these results show that localized semantic steering can substantially reduce correction cost while preserving requirement-level accuracy.