Developing novel outcomes for clinical trials in Stargardt disease using structure/function relationship and deep learning

Stargardt disease is an inherited retinal disease that typically starts to cause visual symptoms during childhood or adolescence. The by far most prevalent form of Stargardt disease (Stargardt disease type 1, STGD1) is inherited in an autosomal recessive manner and caused by mutations in the ABCA4 gene. With the genetic discovery of the disease, novel treatment options are on the horizon, but currently there is no approved treatment for the disease available. One of the challenges for the approval of a drug is to find a sufficient main outcome parameter. In STGD1 disease visual acuity is not a good outcome, because the disease is characterized by a progressive loss visual fields. Functional measures such as microperimetry (MP) show large variability, structural measures such as optical coherence tomography (OCT) are not patient-relevant. We hypotheisze that an optimized estimator of visual field deterioration over time can be developed and that structural data can be used to improve prediction of visual function in STGD1 disease. To test this hypothesis, the data of the ProgStar study, which enrolled 345 patients with STGD1, will be used. Baseline OCT and MP data will be used for the characterization of the structure/function relationship. To refine the structure/function relationship we will apply a deep learning-based model. We will use the structure/function relationship as obtained from the baseline data to predict the future development of visual field decline in STGD1. At the end of this grant we expect to have an improved combined structure/function that can be used in clinical studies involving STGD1 patients. This will facilitate pivotal clinical trials in this field and may accelerate the approval of novel treatments for STGD1 disease.