BACKGROUND: Idiopathic Pulmonary Fibrosis (IPF) is challenging to diagnose. Patients are diagnosed through a combination of multidisciplinary discussion (MDD), high-resolution computed tomography, and surgical lung biopsy. Surgical lung biopsy is considered a core part of the gold standard. The costly invasive procedure carries a significant risk in both morbidity and mortality. Novel non-invasive classifier algorithms may improve the accuracy of IPF diagnosis and reduce the need for biopsy.

OBJECTIVES: We conducted a cost-effectiveness analysis of a machine learning algorithm for the diagnosis of IPF.

METHODS: We developed a decision-analytic model to determine the cost-effectiveness of a machine learning algorithm for diagnosing IPF in patients with chronic interstitial lung disease following an initial inconclusive assessment from an MDD. From the health system perspective using a lifetime horizon, we compare (1) the algorithm, (2) biopsy all patients, and (3) treating all patients. Input parameters were from the published literature and retrospective analyses of IPF patients. The primary outcome measures were costs, quality-adjusted life-years (QALYs), and biopsies averted.

RESULTS: The algorithm reduced the number of biopsies by 41%. Treat all had an ICER of $3,190,788 per QALY compared to the algorithm. Compared to biopsy all, the algorithm had an ICER of $382,991 per QALY. The cost-effectiveness of the algorithm was driven primarily by treatment costs. In addition, results were sensitive to the risk of mortality from surgical lung biopsy and assumptions regarding follow-up diagnostics.

CONCLUSIONS: We found that a novel classifier algorithm may provide additional benefits for diagnosing IPF as a clinical middle-ground. While at a high cost-effectiveness ratio, this is driven by high drug costs. The first generic treatments with reduced costs of ~20% greatly improve the ICER. As additional generic IPF treatments enter the market, cost-effectiveness will likely improve further. Sensitivity analyses illustrate screening technologies' complicated role in identifying diseases with limited, high-cost treatment options.