OBJECTIVES : A novel binocular digital therapeutic for the treatment of amblyopia has the potential to improve binocular function, such as visual acuity (VA) and stereopsis. This research aims to develop a cost-effectiveness model that captures the natural history of amblyopia and the treatment effects on binocular vision.

METHODS : A literature review revealed that most previously developed models were one-eye models with decision-tree structures, or Markov models for screening. Studies have shown that only considering VA of the amblyopic eye cannot capture the impact of the fellow eye, leading to inaccurate estimates of Quality-of-Life (QoL). Furthermore, these models did not consider and compliance.

RESULTS : A de-novo Markov model was developed to compare binocular treatment with traditional monocular therapies (patching of the fellow eye). Patients enter the model in one of eight amblyopic states defined by VA (normal/mild/moderate/severe) of both eyes. After 6-months of treatment, patients either move to another VA state or remain in their current state. Transition probabilities were estimated by assuming that treatment effect follows a normal distribution. Non-compliant patients were associated with a reduced treatment effect. Some patients may regress, following the natural history of amblyopia. Patients may develop bilateral visual impairment from age-specific, non-amblyopia-related causes identified in the literature with any associated mortality and costs assumed. Utility was estimated based on VA of both eyes using published regression coefficients. Patients with improved stereoacuity were assigned utility increments estimated by change in percentage of patients having nil stereoacuity from baseline. The utility of nil stereoacuity was estimated from published SF36 scores, mapped onto EQ5D using a published algorithm.

CONCLUSIONS : The proposed model structure describes important clinical features of amblyopia, including natural disease history, visual acuity of both eyes, stereopsis, long-term risks, and impact on QoL. This model can be useful for health technology assessment and decision-making.