SMASH 2026 · Paper #5 · Result card

Visual Acuity Improvement Prediction in Neovascular Age-Related Macular Degeneration

Hadi Moazen, Mélanie Hébert, Alaa Eddine Ayadi, Ali Dirani, Audrey Durand · Université Laval · CHU de Québec – Université Laval · Read the paper

For a simulated patient with BCVA 50 and SRF present, the model gives a 51% chance of gaining 10 or more letters. A misread SRF flag lowers it to 33%.

your patient BCVA 50, CST 380 µm, SRF +, IRF +, PED −, VMT −automation error ×1 paper's error ratesCST error 20 µmcohort 1,203 simulated eyesmodel EBM (interpret)seed 5data SIMULATED, not for clinical use

Your patient's predicted chance

51%

of gaining ≥10 letters at 1 year (BCVA 50, CST 380 µm, SRF +, IRF +, PED −, VMT −); cohort base rate 25%

Biggest single reason

BCVA +0.68

logit contribution in the glass-box model

If the AI misreads one biomarker

51% → 33%

worst case: SRF misread

Glass box vs black box

0.74 vs 0.72

CV AUC, EBM vs gradient boosting on automated biomarkers (measured: 0.75; BCVA alone: 0.72)

From the abstract

We proposed a predictive pipeline that automates the extraction of clinical OCT biomarkers using fine-tuned DL models. The combination of these automated biomarkers with baseline best-corrected visual acuity serves as inputs for an interpretable classifier.

Evidence

Why this prediction? Every term, in log-odds

What if the automated pipeline gets one biomarker wrong?

Glass box vs black box, measured vs automated

Paper vs. your run

Metric	Paper	Your run
EBM (automated biomarkers) AUC ROC	0.77 ± 0.07	0.74 ± 0.04	synthetic cohort
EBM sensitivity / specificity	0.75 ± 0.08 / 0.61 ± 0.03	0.72 ± 0.07 / 0.64 ± 0.05	threshold = prevalence
EBM F1	0.54 ± 0.02	0.52 ± 0.04
XGBoost AUC ROC	0.74 ± 0.06	0.72 ± 0.04	sklearn gradient boosting stand-in
EBM on measured vs automated biomarkers	AUC = 0.77 vs 0.77	0.75 vs 0.74	error scale ×1, CST error 20 µm
Top features	baseline VA, CST, and the interaction between these two features	BCVA, SRF, CST	built into the simulation
Eyes gaining ≥10 letters	299 (25%)	304 (25%)	calibrated
Median letters gained	2 [-3, 10]	3 [-3, 10]	approximate match
BCVA at one year	68 [57, 76]	65 [56, 73]	not matched exactly

Try next

Patient near the ceiling — BCVA 78 is above the 75-letter ceiling where the paper says improvement drops sharply; expect a much lower predicted chance and smaller contributions from BCVA.

{"patient_bcva": 78, "patient_cst": 380, "patient_srf": 1, "patient_irf": 1, "patient_ped": 0, "patient_vmt": 0, "automation_error_scale": 1, "cst_error_um": 20, "n_patients": 1203, "seed": 5}

Much worse automated extraction — Setting automation_error_scale to 3 triples the paper’s reported flip rates; this should widen the gap between EBM-measured and EBM-automated AUCs and make biomarker misread impacts larger.
```
{"patient_bcva": 50, "patient_cst": 380, "patient_srf": 1, "patient_irf": 1, "patient_ped": 0, "patient_vmt": 0, "automation_error_scale": 3, "cst_error_um": 20, "n_patients": 1203, "seed": 5}
```
Perfect automated extraction — Setting automation_error_scale to 0 and cst_error_um to 0 gives the EBM the exact measured biomarkers; the automated and measured EBM AUCs should converge, and misread impact should disappear.
```
{"patient_bcva": 50, "patient_cst": 380, "patient_srf": 1, "patient_irf": 1, "patient_ped": 0, "patient_vmt": 0, "automation_error_scale": 0, "cst_error_um": 0, "n_patients": 1203, "seed": 5}
```

Read this before sharing

The 1,203 eyes are simulated from the paper's reported marginals.
Not for clinical use.