Autonomous Detection of Polypharmacy–Induced Acute Kidney Injury in Nigeria Using Multi–Source Real–World Data: An AI Agent’s Approach

Keywords: AI Agent, Pharmacoepidemiology, PolyPharmacy, Acute Kydney Injury, Data Integration, Real World Evidence

TL;DR: A large‑language‑model‑driven agent autonomously designed and executed a pharmacoepidemiologic study to detect acute kidney injury linked to polypharmacy in Nigeria.

Abstract:

Pharmacoepidemiology traditionally relies on human investigators to design stud-
ies, analyse data and identify safety signals. Recent calls for AI- authored research
encourage autonomous agents to generate hypotheses and conduct full research
cycles. In this paper we design an AI agent that performs a pharmacoepidemiologic
investigation into acute kidney injury (AKI) associated with combinations of anti-
hypertensive and antimalarial drugs in Nigeria. The agent synthesises structured
prescription data and unstructured clinical notes within a common data model,
generates candidate drug–drug combinations, designs analytic plans, conducts anal-
yses, interprets results and writes the manuscript. Using a synthetic multi -hospital
dataset, the agent identifies several high - risk polypharmacy patterns and detects
safety signals months earlier than analyses based on structured data alone. We
demonstrate the feasibility of AI-driven pharmacoepidemiology, discuss method-
ological challenges such as data heterogeneity and confounding, and propose
strategies for responsible deployment. We provide AI involvement, ethical and
reproducibility disclosures in accordance with the Agents4Science guidelines.

Submission Number: 89