Background

Current AAO 2016 guidelines recommend annual screening for all HCQ users after 5 years, regardless of individualized risk trajectory. This uniform approach generates substantial ophthalmology burden: with an estimated 5+ million HCQ users globally, annual screening produces millions of visits, the vast majority of which are negative in low-risk patients. Meanwhile, high-risk patients may benefit from more frequent monitoring than annually.

Hypothesis

We hypothesize that a Bayesian adaptive screening model incorporating serial spectral-domain OCT (SD-OCT) measurements — specifically parafoveal ellipsoid zone thickness, outer nuclear layer thinning, and retinal pigment epithelium reflectivity changes — can dynamically adjust screening intervals to reduce visit frequency by ≥40% in patients with consistently stable biomarkers, while maintaining ≥95% sensitivity for early toxicity detection.

Testable Predictions

1. Visit reduction: Patients classified as "stable trajectory" by the Bayesian model after 3 consecutive normal annual OCTs will safely extend to biennial screening, reducing visits by 40-50% over a 10-year horizon.
2. Sensitivity maintenance: The adaptive model will detect ≥95% of cases that progress to definite toxicity (defined as parafoveal photoreceptor loss on SD-OCT confirmed by mfERG), with median lead time of ≥6 months before functional visual field loss.
3. Posterior convergence: After 3 annual observations, the posterior probability of toxicity in stable patients will converge to <0.5%, justifying interval extension by the Bayesian decision rule.
4. Subgroup specificity: The model will correctly identify the 10-15% of patients whose biomarker trajectories warrant accelerated screening (every 6 months) despite low baseline risk scores.

Proposed Validation

• Retrospective analysis of serial OCT data from HCQ cohorts (target n≥500 patients, ≥3 OCTs each)
• Prior elicitation from AAO prevalence data (Melles 2020) and cumulative dose curves
• Simulation of adaptive vs. fixed screening schedules over 10-year horizons
• Sensitivity analysis varying prior assumptions by ±50%

Clinical Significance

Optimizing screening intervals addresses both patient burden (unnecessary visits, anxiety) and healthcare system capacity. The Bayesian framework naturally integrates new data at each visit, creating a personalized surveillance trajectory rather than one-size-fits-all scheduling.

References

• Marmor MF et al. Ophthalmology 2016;123:1386-94
• Melles RB, Jorge AM. JAMA Ophthalmol 2020;138(4):e200370
• Browning DJ. Am J Ophthalmol 2014;158(6):1207-12
• Yusuf IH et al. Eye 2017;31:828-845
• Lee DH et al. Retinal layer segmentation in hydroxychloroquine toxicity. Br J Ophthalmol 2022;106(8):1132-1138