# Preliminary insights into artificial intelligence guided dosing in hypertension and diabetes: challenges and lessons learnt in a pilot feasibility study

## TL;DR

This study explores using AI to personalize drug dosing for diabetes and hypertension patients, finding challenges in calibration and data collection.

## Contribution

The study provides preliminary insights into deploying AI-guided dosing in outpatient settings and identifies barriers to effective model calibration.

## Key findings

- Only three out of 18 participants progressed to the active study phase due to limited dose adjustments during calibration.
- Participants showed high satisfaction with monitoring and care, with adherence rates of 76% for visits and 81% for home monitoring.
- Actionable AI dosing recommendations aligned variably with physician decisions depending on data sources.

## Abstract

CURATE.AI is an artificial intelligence platform enabling personalised drug dosing. Aims:

1) Determine the feasibility of using CURATE.AI in the outpatient setting.

2) Compare the consistency of CURATE.AI recommendations derived from different data sources.

3) Assess the alignment of physician and CURATE.AI dosing recommendations.

We conducted a single-arm feasibility study involving type II diabetics and hypertensives recruited from a hospital’s outpatient clinics. Outcomes included recruitment and study completion rates, adherence to study protocols, patient satisfaction, consistency of CURATE.AI recommendations across data sources, and alignment with physicians’ dosing decisions. We calibrated CURATE.AI for each individual using three distinct datapoints that linked drug dose to clinical response. After calibration, participants entered a four-month active phase, receiving monthly CURATE.AI dosing recommendations.

Eighteen participants were recruited, and thirteen completed the study. Only three progressed to the active study phase, primarily due to insufficient dose adjustments required during the calibration phase. Adherence to scheduled visits was 76% and adherence to home monitoring averaged 81%. Barriers to adherence included technical issues and work-related conflicts. Participants expressed high satisfaction with monitoring and care ≥88%. Actionable dosing recommendations were generated for two of the three participants, with varying alignment to physician decisions depending on the data source used.

Calibration challenges emerged when applying AI-guided dosing in a chronic disease population. Limited dose titration opportunities and cautious clinical practice restricted the data generation needed for effective model calibration.

This pilot demonstrates the feasibility of deploying CURATE.AI into outpatient care but underscores the importance of aligning data requirements with patient and clinical characteristics. Future studies should target newly diagnosed patient groups with greater dosing variability to optimise calibration and assess clinical utility.