Cost-Effectiveness Analysis of Artificial Intelligence-Driven Risk Stratification in Patients With Diabetic Kidney Disease in the US Veterans Population

Abstract

Rationale & Objective

Efficient risk stratification is essential to optimize care and allocate resources for treatment of diabetic kidney disease (DKD). This study evaluates the cost-effectiveness of an artificial intelligence-driven in vitro kidney disease risk assay (AIKD).

Study Design

Cost-effectiveness analysis using a hybrid model, combining a decision tree followed by a Markov model.

Setting & Population

Patients with early-stage DKD receiving care within the US Veterans Health Administration health care system.

Intervention(s)

Risk stratification using AIKD versus Kidney Disease: Improving Global Outcomes (KDIGO) standard of care (SoC).

Outcomes

Five-year health care costs, quality-adjusted life-years (QALYs), and incremental cost-effectiveness ratio (ICER).

Model, Perspective, & Timeframe

The decision tree delineated clinical pathways based on the prevalence of progressive decline in kidney function and risk stratification performance of AIKD versus KDIGO. The subsequent Markov model simulated DKD stage transitions across the underlying risk–treatment pathways. Model inputs included test performance characteristics, risk prevalence, transition probabilities, costs, and utilities. One-way and probabilistic sensitivity analyses assessed uncertainty. The analysis was conducted from the perspective of the Veterans Health Administration health care system over a 5-year time horizon.

Results

AIKD-guided care resulted in a total cost of $146,437 and 2.8277 QALYs, compared with $145,120 and 2.8164 QALYs for the SoC arm. The ICER for AIKD relative to SoC was $116,349 per QALY gained. One-way sensitivity analysis showed that the sensitivity and specificity of AIKD and SoC, as well as the prevalence of underlying risk of progressive decline in kidney function, were the most influential inputs affecting the ICER. From the probabilistic sensitivity analysis, AIKD has 69% likelihood of being accepted at the conventional willingness-to-pay threshold of $150,000 per QALY gained.

Limitations

Model assumptions regarding risk stratification performance and long-term treatment effects may limit generalizability.

Conclusions

AIKD is cost-effective compared to KDIGO for patients with early-stage DKD. Its adoption could improve health outcomes and support efficient health care resource utilization management.

Plain-Language Summary

We studied whether a new artificial intelligence tool, artificial intelligence-driven in vitro kidney disease risk assay (AIKD), is a worthwhile investment for early identification of diabetic kidney disease compared with the current Kidney Disease: Improving Global Outcomes (KDIGO) guideline-recommended risk stratification method. This tool assists clinicians in identifying patients at high risk of kidney function decline, allowing for earlier and more targeted interventions. Using data from the US Veterans Health Administration, we found that AIKD modestly increases care costs but helps slow disease progression and improves quality of life. Over 5 years, AIKD was found to be cost-effective, and its overall impact on the health care budget was modest. Our findings suggest that AIKD-assisted personalized care decision making has the potential to result in cost-effective health care resource utilization in patients with diabetic kidney disease.