Topical losartan for established corneal fibrosis with machine learning-based predictors

Abstract

This study evaluated the effectiveness of topical losartan 1 mg/mL in reducing corneal fibrosis by inhibiting myofibroblast proliferation and improving corneal transparency, offering a potential therapeutic alternative. A prospective, interventional case series enrolled adults with corneal fibrosis. Participants administered topical losartan 1 mg/mL six times daily for 3 months. Visual acuity, slit-lamp examination, corneal tomography, and densitometry were used to assess outcomes. A Random Forest machine learning model identified key predictors of visual improvement. Nineteen eyes from seventeen patients were analyzed. Mean uncorrected distance visual acuity improved from 1.04 ± 0.62 to 0.74 ± 0.46 LogMAR (p = 0.007). Corneal densitometry significantly decreased, particularly in the midperipheral zone (p = 0.0184). Worse baseline visual acuity and longer leukoma duration correlated with greater improvement. The predictive model achieved an AUROC of 0.76, with 86.67% sensitivity, 75.00% specificity, and 84.21% accuracy, confirming its robustness. These findings suggest that topical losartan improves corneal transparency and functional vision in patients with corneal fibrosis. The predictive model provides a clinically useful scoring system to guide treatment selection. Further validation in larger clinical trials is warranted.

Plain language summary

Eye drops for corneal scarring: a new treatment with losartan

Corneal fibrosis, or scarring of the clear front part of the eye, can cause vision loss. Until now, corneal transplantation has been the only option for severe cases. This study tested a new, non-invasive treatment using losartan eye drops to reduce scarring and improve vision. We treated 19 eyes with losartan eye drops six times daily for three months. Results showed that vision improved significantly, and corneal clarity increased. A machine learning model helped identify patients most likely to benefit, with worse starting vision and longer-lasting scars predicting better outcomes. These findings suggest that losartan eye drops could become a simple, affordable alternative for treating corneal scarring, reducing the need for surgery. More research is needed to confirm these results in larger studies

Keywords

corneal fibrosis losartan machine learning visual acuity

Introduction

Corneal opacity is a leading cause of blindness worldwide, affecting over 11 million people.^1,2 Stromal scarring and fibrosis, primarily driven by transforming growth factor-beta (TGF-β), result in irreversible vision loss.^3,4 While corneal transplantation remains the only definitive treatment, accessibility is limited due to donor shortages, high costs, and infrastructure constraints.^5,6

Losartan, an angiotensin II receptor antagonist, has demonstrated anti-fibrotic properties by inhibiting TGF-β activity.⁷ Preclinical studies show that topical losartan effectively reduces fibrosis in rabbit models without adverse effects.^8
–12 Clinical evidence suggests its potential to reverse corneal scarring, as demonstrated in post-refractive surgery fibrosis cases.¹³ Moreover, animal models indicate that losartan may be superior to corticosteroids in certain scenarios, suggesting its role as a promising alternative therapy.^13
–16

This study evaluates the therapeutic potential of topical losartan 1 mg/mL for established corneal fibrosis, representing the first case series in a Latino population. Understanding ethnic differences in disease prevalence and treatment response is essential for optimizing care strategies.^17,18 By targeting the underlying pathophysiology of fibrosis, topical losartan offers an accessible, noninvasive alternative for managing corneal opacity.

Material and methods

Study design and patients

This prospective, interventional, single-center, case series evaluated the efficacy and safety of topical losartan 1 mg/mL in reducing corneal fibrosis. The study was conducted at the Instituto de Oftalmologia Conde de Valenciana in Mexico City between November 2023 and July 2024. Adults 18 years or older with established stromal fibrosis persisting for at least 1 month after complete resolution of the primary corneal pathology, including full re-epithelialization, were eligible for inclusion. Participants were required to have associated visual impairment due to fibrosis, without other ocular conditions affecting vision, such as glaucoma, maculopathy, or clinically significant cataract. Patients requiring additional topical medications other than lubricant, those with active infections, or systemic conditions affecting the cornea were excluded. In two cases where both eyes met the inclusion criteria, each eye was treated as an independent unit.

Topical losartan preparation and administration

Topical losartan was prepared using losartan potassium salt (Sigma-Aldrich, Burlington, VT, USA) following previously published protocols.^13,15 A 1 mg/mL solution was formulated by dissolving 10 mg of losartan potassium in 10 mL of sterile balanced salt solution under laminar flow conditions. The solution was stored at ambient temperature, and participants received a new dropper bottle every 28 days. The treatment regimen consisted of instilling one drop in the affected eye six times daily for 3 months.

Sample size calculation

The sample size was calculated using the standard formula for proportions, assuming a 1.01% corneal fibrosis prevalence, 95% confidence level, and 5% margin of error. The required sample was 16 subjects, increased to 18 to account for a 10% potential loss to follow-up and ensure statistical validity.

Baseline and postoperative follow-up assessment

Subjects were evaluated at baseline and subsequently every 28 days over the three-month study period. Assessments included uncorrected distance visual acuity (UDVA), pinhole visual acuity (PHVA), slit-lamp examination, and anterior segment imaging using anterior segment optical coherence tomography (AS-OCT) (MS-39, Schwind, Kleinostheim, Germany). Corneal densitometry was analyzed to quantify fibrosis severity using the Pentacam HR’s Scheimpflug software (Oculus, Wetzlar, Germany), which acquires 25 images over multiple meridians. Only scans with a “quality specification = OK” were included. A 12-mm-diameter circular zone centered on the corneal apex was automatically subdivided by the software into four concentric zones: 0–2 mm (central), 2–6 mm (paracentral), 6–10 mm (midperipheral), and 10–12 mm (peripheral). For each zone, densitometry values were calculated across three anatomical layers: the anterior 120 μm, the posterior 60 μm, and a central layer defined by subtraction. Densitometry was expressed in grayscale units, ranging from 0 (maximum transparency) to 100 (complete opacity). Total corneal aberrometry was performed using the Pentacam HR, assessing total higher-order aberrations over a 6-mm zone, including coma, trefoil, and spherical aberration. All imaging and clinical evaluations were performed by the principal investigator (JLDH).

Machine learning-based predictive model

Dataset, feature selection, and model development

Patients were classified into two groups based on visual acuity outcomes: the Gain Group (n = 14) for those improving by ⩾1 Snellen line and the No Change/Loss Group (n = 5) for stable or worsened vision. Machine learning analysis using a Random Forest model identified nine key predictors strongly associated with visual improvement. The model was developed in Python 3.9 using Scikit-Learn, with NumPy, SciPy, and Pandas for statistical analysis. A Random Forest classifier with 100 estimators determined feature importance in predicting treatment response. Data was split into training (80%) and testing (20%) subsets. The model analyzed baseline UDVA, PHVA, leukoma duration, etiology (e.g., hydrops and fungal keratitis), and corneal densitometry (anterior, central, posterior, total), with feature importance quantifying each variable’s impact on classification accuracy. Detailed data in Table 1.

Table 1.

Scoring system for predicting visual acuity improvement with topical losartan in corneal fibrosis.

Feature	Importance value (%)	Threshold	Weight assigned	Rationale for threshold and weight
Baseline UDVA (LogMAR)	25.60%	>1.2 LogMAR	+2 points	A worse baseline vision (>1.2 LogMAR) correlates with greater potential improvement post-treatment.
Duration of Leukoma (months)	20.40%	>1.0 month	+2 points	Patients with longer leukoma duration (>1.0 months) showed a higher likelihood of vision gain.
Hydrops	15.30%	Present = 1	+2 points	Hydrops, responded well to losartan therapy.
Fungal Keratitis	12.80%	Present = 1	+2 points	Fungal keratitis cases had higher rates of vision gain.
Bacterial Keratitis	9.70%	Present = 1	+1.5 points	Bacterial infections showed moderate responsiveness to therapy.
Herpetic Keratitis	7.50%	Present = 1	+1.5 points	Chronic viral keratitis showed a slower but notable response.
Interstitial Keratitis	4.20%	Present = 1	+1 point	Interstitial keratitis had a moderate correlation with improvement.
Chemical Burn	3.00%	Present = 1	+2 points	Patients with chemical burns exhibited a strong response.
Pterygium Resection	2.50%	Present = 1	−1 point	Scarring from pterygium resections was associated with lower improvement rates.

UDVA, uncorrected distance visual acuity.

Threshold and weight selection for scoring system

The scoring system, derived from the model’s feature importance values, was validated using multiple statistical techniques, including a confusion matrix, likelihood ratios, and Leave-One-Out Cross-Validation (LOOCV), ensuring predictive accuracy and clinical reliability across different patient cases. The optimal cutoff score for vision gain classification was determined using Youden’s Index to balance sensitivity and specificity.

Statistical analysis

Statistical analyses were performed using GraphPad Prism (version 10.1.1, GraphPad Software, La Jolla, CA, USA). Data were reported as mean ± standard deviation (SD). Normality was assessed using the Shapiro–Wilk and Kolmogorov–Smirnov tests. Paired t-tests compared pre- and post-treatment data, while ANOVA with Tukey’s test analyzed multiple groups. Non-Gaussian distributions were evaluated with the Wilcoxon signed-rank and Friedman tests, followed by Dunn’s multiple comparisons. Statistical significance was set at p < 0.05.

Results

This study included 19 eyes from 17 patients (mean age: 42.56 ± 12.93 years), with most being female (n = 10, 58.82%). The most common leukoma etiologies were herpetic keratitis (n = 5, 26.3%), bacterial keratitis (n = 4, 21.1%), fungal keratitis (n = 3, 15.8%), and hydrops (n = 3, 15.8%). Less frequent causes included interstitial keratitis (n = 2, 10.5%), chemical burns (n = 1, 5.3%), and pterygium resection (n = 1, 5.3%). The mean leukoma duration before treatment was 41.37 ± 78.97 months. An epithelial defect was present in 57.9% (n = 11) during the initial corneal pathology but had fully resolved prior to treatment initiation. No patients had active epithelial defects at baseline. Corneal neovascularization was observed in 31.6% (n = 6) at baseline. Four patients (21.1%) had prior Rose Bengal Photodynamic Therapy (RB-PDT) for infectious keratitis. To minimize confounding, losartan was started after a two-month washout, with an average interval of 3.75 ± 2.06 months between RB-PDT and treatment.

Visual acuity outcomes

During the 3-month follow-up, UDVA improved significantly from 1.04 ± 0.62 LogMAR to 0.74 ± 0.46 LogMAR (Friedman p = 0.007, Dunn p = 0.0331) with a mean improvement of −0.29 ± 0.36 LogMAR. Detailed data in Table 2. Most eyes (n = 9, 47%) gained ⩾2 Snellen lines, while 26% (n = 5) gained 1 line. No change occurred in 21% (n = 4), and only one eye lost ⩾2 lines. In contrast, PHVA remained stable, with 58% (n = 11) unchanged, 32% (n = 6) gaining ⩾2 lines, and 11% (n = 2) gaining 1 line. Notably, no cases lost vision, reinforcing PHVA stability during follow-up.

Table 2.

Patient Ocular Characteristics before and 3 months of treatment.

Characteristic	Baseline	1 Month	2 Months	3 Months	p	Baseline versus 3 months
Characteristic	(n = 19)	(n = 19)	(n = 19)	(n = 19)	p	∆	p
UDVA (LogMAR)	1.04 ± 0.62	0.85 ± 0.43	0.76 ± 0.46	0.74 ± 0.46	0.007	−0.29 ± 0.36	0.0331
Snellen Equivalent (range)	20/25–CF 6’	20/20–20/400	20/20–20/400	20/20–20/400
Pinhole VA (LogMAR)	0.52 ± 0.37	0.44 ± 0.32	0.40 ± 0.35	0.41 ± 0.32	0.0521	−0.10 ± 0.21	0.728
Snellen Equivalent (range)	20/20–20/400	20/20–20/300	20/20–20/200	20/20–20/200
Corneal densitometry (GSU)
Anterior (120 µm)
0–2 mm	75.59 ± 22.05	72.65 ± 26.10	73.57 ± 24.54	74.88 ± 24.41	0.1765	−0.70 ± 10.59	>0.9999
2–6 mm	58.67 ± 17.10	55.34 ± 18.12	56.56 ± 19.33	55.73 ± 17.46	0.1276	−2.94 ± 7.80	0.3809
6–10 mm	38.78 ± 10.31	37.07 ± 10.12	36.74 ± 10.05	34.72 ± 9.49	0.0222	−4.06 ± 6.34	0.0531
10–12 mm	33.65 ± 9.89	32.14 ± 9.77	31.75 ± 8.99	32.46 ± 9.64	0.2702	−1.19 ± 7.35	0.6975
Total	50.97 ± 11.10	48.37 ± 11.70	48.95 ± 12.53	48.36 ± 10.88	0.0823	−2.61 ± 5.52	0.2043
Central
0–2 mm	47.43 ± 18.01	46.57 ± 20.11	47.02 ± 19.43	44.97 ± 17.66	0.4669	−2.45 ± 7.60	0.5103
2–6 mm	37.03 ± 12.15	34.87 ± 11.64	35.22 ± 12.47	33.89 ± 10.27	0.0976	−3.14 ± 7.37	0.2809
6–10 mm	23.91 ± 9.01	22.91 ± 8.16	23.06 ± 8.22	21.44 ± 5.99	0.1048	−2.46 ± 4.91	0.0856
10–12 mm	21.69 ± 5.81	20.94 ± 6.06	20.83 ± 5.97	21.29 ± 4.71	0.5106	−0.40 ± 5.02	>0.9999
Total	32.05 ± 8.79	30.63 ± 8.34	30.43 ± 9.03	29.67 ± 6.46	0.1014	−2.37 ± 5.32	0.2444
Posterior (60 µm)
0–2 mm	21.29 ± 12.17	21.85 ± 11.84	21.39 ± 10.68	18.88 ± 8.996	0.0542	−2.41 ± 5.20	0.0718
2–6 mm	19.42 ± 6.96	18.84 ± 6.60	18.56 ± 6.84	17.84 ± 4.43	0.1027	−1.57 ± 5.20	0.196
6–10 mm	18.33 ± 5.82	17.57 ± 5.31	17.58 ± 5.17	16.32 ± 4.62	0.055	−2.01 ± 4.36	0.0414
10–12 mm	17.63 ± 4.80	17.34 ± 4.60	16.86 ± 3.97	17.27 ± 5.21	0.8059	−0.35 ± 5.21	0.9903
Total	19.07 ± 5.60	18.68 ± 5.16	18.49 ± 5.28	17.25 ± 4.02	0.2188	−1.82 ± 4.05	0.2289
Total
0–2 mm	48.10 ± 13.39	47.02 ± 16.00	47.33 ± 14.92	46.25 ± 13.59	0.5824	−1.85 ± 6.36	0.5928
2–6 mm	38.38 ± 9.72	36.35 ± 9.79	36.78 ± 10.84	35.66 ± 8.92	0.0702	−2.72 ± 5.74	0.2027
6–10 mm	27.02 ± 7.48	25.85 ± 6.86	25.79 ± 6.84	24.15 ± 5.58	0.0184	−2.86 ± 4.36	0.0101
10–12 mm	24.33 ± 6.19	23.47 ± 6.03	23.16 ± 5.59	23.66 ± 5.62	0.8052	−0.66 ± 4.76	>0.9999
Total	34.02 ± 6.80	32.56 ± 6.62	32.80 ± 7.42	31.76 ± 5.17	0.042	−2.25 ± 4.14	0.1187
Total excluding 10–12 mm
Total Anterior (120 µm)	47.16 ± 11.28	44.91 ± 12.10	45.34 ± 12.70	44.57 ± 10.94	0.101	−2.58 ± 5.37	0.1925
Total Central	29.51 ± 8.57	28.28 ± 8.04	28.51 ± 8.79	27.12 ± 6.32	0.0329	−2.38 ± 5.04	0.0231
Total Posterior (60 µm)	16.94 ± 5.31	16.57 ± 4.84	16.40 ± 4.99	15.26 ± 3.40	0.0518	−1.67 ± 3.81	0.0342
Total	31.19 ± 6.68	29.92 ± 6.52	30.08 ± 7.24	28.98 ± 5.18	0.0427	−2.20 ± 4.02	0.115

Data are mean ± SD unless otherwise indicated. ∆ = Difference.

Bold values indicate p < 0.05.

CF, Counting Fingers; GSU, Grayscale Units; SD, standard deviation; UDVA, Uncorrected Distance Visual Acuity.

Corneal densitometry changes

Corneal densitometry significantly decreased, especially in the midperipheral 6–10 mm zone. Anterior densitometry declined from 38.78 ± 10.31 GSU to 34.72 ± 9.49 GSU, with ANOVA showing significance (p = 0.0222), though post-hoc comparison showed only a decreasing trend (p = 0.0531). Posterior densitometry decreased from 18.33 ± 5.82 GSU to 16.32 ± 4.62 GSU, trending toward significance (p = 0.055), but post-hoc analysis confirmed a significant reduction (p = 0.0414). Total densitometry in the 6–10 mm zone significantly decreased from 27.02 ± 7.48 GSU to 24.15 ± 5.58 GSU (Friedman p = 0.0184, Dunn p = 0.0101), confirming reduced corneal opacity. Figure 1.

Figure 1.

Slit-lamp and anterior segment imaging of corneal fibrosis before and after losartan treatment. (a) Fibrosis 2 months after fungal keratitis resolution. (b) Same patient as (a) after 3 months of losartan treatment, showing fibrosis reduction. (c) Fibrosis 1 month after a chemical burn with central opacification. (d) Same patient as (c) after 3 months of losartan treatment, showing improvement. (e) Densitometry map of fibrosis with high central and midperipheral reflectivity. (f) Same patient as (e) after 3 months of losartan treatment, showing reduced reflectivity.

Total corneal densitometry decreased from 34.02 ± 6.80 GSU to 31.76 ± 5.17 GSU, with ANOVA showing significance (p = 0.042), though post-hoc analysis indicated only a decreasing trend (p = 0.1187). Excluding the peripheral 10–12 mm region, total central densitometry decreased from 29.51 ± 8.57 GSU to 27.12 ± 6.32 GSU (Friedman p = 0.0329, Dunn p = 0.0231). Total posterior densitometry decreased from 16.94 ± 5.31 GSU to 15.26 ± 3.40 GSU, with Friedman analysis suggesting a trend (p = 0.0518), while post-hoc analysis confirmed significance (p = 0.0342). Lastly, total corneal densitometry, excluding the 10–12 mm zone, decreased from 31.19 ± 6.68 GSU to 28.98 ± 5.18 GSU, with ANOVA nearing significance (p = 0.0518) and post-hoc analysis confirming significance (p = 0.0342). Detailed data in Table 2.

Subgroup analysis

Patients with baseline UDVA ⩾ 1.2 LogMAR were significantly more likely to experience visual improvement (p = 0.0445, OR = 0.0, 95% CI: 0.000–0.880), suggesting worse initial vision predicted greater gains. Leukoma etiology had no significant impact, though hydrops (n = 3, 15.8%) and interstitial keratitis (n = 2, 10.5%) appeared only in the Gain Group, while pterygium resection (n = 1, 5.3%) was exclusive to the No Change/Loss Group. The Gain Group had a much longer leukoma duration (55.16 ± 88.63 months) than the Loss/No Change Group (2.75 ± 1.85 months), though this difference was not significant (p = 0.2092). Baseline corneal densitometry, history of epithelial defect (resolved prior to treatment), prior RB-PDT, and corneal neovascularization did not significantly affect treatment response. Detailed data in Table 3.

Table 3.

Baseline characteristics between groups based on changes in Visual Acuity.

Characteristic	Loss or No change (n = 5)	Gain 1 or more (n = 14)	p	OR	95% CI
Age (years)	40.20 ± 11.86	45.29 ± 13.49	0.4673
Sex
Male	2 (40.0 )	5 (41.67)	>0.9999	0.9333	0.128–6.09
Female	3 (60.0)	7 (58.33)
UDVA (LogMAR)	0.73 ± 0.38	1.41 ± 0.82	0.109
Cutoff
⩾ 1.2 LogMAR	0 (0)	8 (57.14)	0.0445	0.0	0.000–0.880
< 1.2 LogMAR	5 (100.0)	6 (42.86)
Pinhole VA (LogMAR)	0.36 ± 0.28	0.62 ± 0.39	0.2201
Etiology of leukoma
Bacterial Keratitis	2 (40.0)	2 (14.29)	0.2722	4.0	0.438–31.37
Fungal Keratitis	0 (0)	3 (21.43)	0.5304	0.0	0.00–3.31
Herpetic Keratitis	2 (40.0)	3 (21.43)	0.5696	2.444	0.303–16.52
Interstitial Keratitis	0 (0)	2 (14.29)	>0.9999	0.0	0.00–6.20
Hydrops	0 (0)	3 (21.43)	0.5304	0.0	0.00–3.31
Chemical Burn	0 (0)	1 (7.13)	>0.9999	0.0	0.00–25.20
Pterygium resection	1 (20.0)	0 (0)	0.2632	infinity	0.311–infinity
Duration of leukoma prior to initiating treatment (months)	2.75 ± 1.85	55.16 ± 88.63	0.2092
Etiology with epithelial defect
Yes	3 (60.0)	8 (72.73)	>0.9999	1.125	0.179–7.85
No	2 (40.0)	6 (75.0)
Previous therapy with RB-PDT
Yes	1 (20.0)	3 (21.43)	>0.9999	0.9167	0.057–7.99
No	4 (80.0)	11 (78.57)
Corneal neovascularization
Yes	2 (40.0)	4 (28.57)	>0.9999	1.667	0.222–10.71
No	3 (60.0)	10 (71.43)
Corneal densitometry (GSU)
Total Anterior (120 µm)	51.18 ± 12.59	50.90 ± 11.03	0.963
Total Central	32.48 ± 14.86	31.89 ± 6.23	0.3808
Total Posterior (60 µm)	18.94 ± 7.15	19.12 ± 5.25	0.9644
Total	34.20 ± 11.05	33.96 ± 5.13	0.2976

Data are presented as mean ± SD for continuous variables and as n (%) for categorical variables unless otherwise specified.

Bold values indicate p < 0.05.

GSU, grayscale units; SD, standard deviation; UDVA, uncorrected distance visual acuity.

Correlation analysis

Baseline UDVA strongly correlated with visual improvement (r = 0.6296, p = 0.018), indicating poorer initial vision led to greater treatment response. Longer leukoma duration correlated with better UDVA gains (r = 0.5466, p = 0.0455), and older patients also showed greater improvement (r = 0.5409, p = 0.048). Leukoma duration was significantly associated with reductions in total anterior densitometry (r = 0.6637, p = 0.0115) and total corneal densitometry (r = 0.579, p = 0.0326). Worse baseline UDVA correlated with greater densitometry changes in the total central cornea (r = 0.5699, p = 0.0359). Higher baseline posterior densitometry values correlated with less posterior densitometry change (r = −0.5787, p = 0.0327). Lastly, baseline total central densitometry was significantly correlated with densitometry change in the Loss/No Change Group (r = −1, p = 0.0167), but not in the Gain Group (r = 0.0154, p = 0.9605). Detailed data in Table 4.

Table 4.

Correlations between groups based on changes in visual acuity.

Characteristic	Loss or No change		Gain 1 or more
Characteristic	r	p	r	p
Change in UDVA (LogMAR)
Baseline UDVA (LogMAR)	0.3953	>0.9999	0.6296	0.018
Duration of leukoma prior to initiating treatment (months)	0	>0.9999	0.5466	0.0455
Age (years)	0	>0.9999	0.5409	0.048
Change in Densitometry (GSU)
Change in Total Anterior (120 µm)
Duration of leukoma prior to initiating treatment (months)	0.3	0.6833	0.6637	0.0115
Change in Total Central
Baseline UDVA (LogMAR)	−0.6708	0.3	0.5699	0.0359
Baseline Total Central	−1	0.0167	0.0154	0.9605
Change in Total Posterior (60 µm)
Baseline Total Posterior (60 µm)	−0.9	0.0833	−0.5787	0.0327
Change in Total
Duration of leukoma prior to initiating treatment (months)	0.3	0.6833	0.579	0.0326
Baseline Total Posterior (60 µm)	−1	0.0167	−0.2176	0.4542

Bold values indicate p < 0.05.

GSU, Grayscale Units; UDVA, Uncorrected distance visual acuity.

Corneal topography and higher-order aberrations

At 3 months, keratometric parameters (K1, K2, Kmax), corneal astigmatism, asphericity, and thinnest corneal thickness (TCT) remained stable compared to baseline (p > 0.3). Root mean square (RMS) total corneal higher-order aberrations (HOAs) did not significantly change (Δ = 0.092 ± 2.47 μm; p = 0.9632), and no significant differences were observed in individual total corneal Zernike terms, including vertical trefoil, vertical coma, horizontal coma, oblique trefoil, and spherical aberration (p > 0.4). These findings suggest that while stromal reflectivity decreased, anterior corneal topography and total corneal HOA patterns remained stable throughout the treatment period. Detailed data in Table 5.

Table 5.

Patient anterior corneal surface and aberrations before and 3 months of treatment.

Characteristic	Baseline	Month 1	Month 2	Month 3	p	Baseline vs. 3 months
Anterior corneal surface	(n = 19)	(n = 19)	(n = 19)	(n = 19)	p	∆	p
K1 (D)	40.63 ± 2.92	41.47 ± 2.41	41.58 ± 2.53	41.19 ± 2.90	0.1741	0.563 ± 2.59	0.3558
K2 (D)	45.08 ± 3.55	45.61 ± 3.00	45.27 ± 2.45	45.19 ± 2.40	0.5393	0.105 ± 2.36	0.8483
Kmax (D)	50.94 ± 3.70	51.03 ± 4.31	50.26 ± 4.00	51.67 ± 8.18	0.4065	0.726 ± 8.05	0.4837
Astigmatism (D)	4.468 ± 3.36	4.132 ± 2.85	3.689 ± 2.04	3.979 ± 2.72	0.7414	−0.489 ± 4.11	0.3371
Asphericity (Q)	−0.0594 ± 0.43	−0.1453 ± 0.51	−0.0363 ± 0.43	0.1489 ± 0.84	0.2497	0.208 ± 0.72	0.4122
TCT (µm)	456.4 ± 89.57	460.1 ± 68.79	424.8 ± 109.4	429.6 ± 118.4	0.0625	−26.89 ± 102.3	0.4751
RMS HOA (µm)	3.715 ± 1.19	3.854 ± 2.68	3.617 ± 2.22	3.808 ± 2.12	0.6621	0.092 ± 2.47	0.9632
Vertical Trefoil Z3, −3 (µm)	−0.3800 ± 1.44	−0.7097 ± 1.44	0.09882 ± 2.13	−0.4144 ± 1.26	0.6945	−0.034 ± 1.46	>0.9999
Vertical Coma Z3, −1 (µm)	−0.2685 ± 2.09	0.05041 ± 1.38	0.09071 ± 1.34	−0.3402 ± 1.77	0.9427	−0.071 ± 1.96	0.89
Horizontal Coma Z3, 1 (µm)	0.2516 ± 1.64	0.1043 ± 1.35	0.1884 ± 1.30	−0.3723 ± 1.71	0.0877	−0.623 ± 1.28	0.0617
Oblique Trefoil Z3, 3 (µm)	0.4585 ± 1.44	0.6612 ± 2.67	0.0440 ± 1.81	0.3530 ± 1.98	0.7277	−0.105 ± 2.05	0.4307
Spherical Aberration Z4, 0 (µm)	0.0420 ± 0.62	0.0810 ± 0.95	−0.0013 ± 0.48	0.0995 ± 0.82	0.1133	0.057 ± 0.82	0.8176

Data are mean ± SD unless otherwise indicated. ∆ = difference.

D, diopters; HOA, High Order Aberrations; K, Keratometry; RMS, Root Mean Square; SD, standard deviation; TCT, Thinnest Corneal Thickness; WFE, Wavefront Error.

Tolerability and safety

No cases of corneal staining, punctate epithelial erosions, or epithelial compromise were observed on slit-lamp examination. One patient (5.3%) developed mild conjunctival hyperemia and reported transient ocular discomfort during the first 2 days of treatment; symptoms were tolerable, resolved spontaneously, and did not require discontinuation of therapy. No other patients reported burning, stinging, foreign body sensation, dryness, tearing, or photophobia during the study. All patients completed the full treatment course without the need for discontinuation due to adverse effects.

Feature selection and scoring system

Among model-selected predictors, UDVA was the strongest, contributing 25.6% to treatment response classification, with worse baseline vision (>1.2 LogMAR) assigned +2 points. Leukoma duration (>1.0 month) contributed 20.4% (+2 points), as longer-standing fibrosis correlated with greater improvement. Hydrops, fungal keratitis, and chemical burns (+2 points) were strongly associated with positive outcomes, while bacterial and herpetic keratitis (+1.5 points) showed moderate predictability. Pterygium resection (−1 point) was linked to a lower likelihood of improvement. Densitometry features were excluded due to their negligible predictive value.

The assigned feature weights formed a clinically interpretable scoring system, enabling prediction of treatment response based on key baseline characteristics. The optimal cutoff score for vision gain classification was 2.5, determined using Youden’s Index (0.62) to balance sensitivity (86.67%) and specificity (75.00%). Lowering the threshold to 2.0improved sensitivity to 100% but reduced specificity to 40%, whereas increasing it to 3.0 enhanced specificity to 85.00%but lowered sensitivity to 80.00%, offering minimal additional benefit. Detailed data in Table 1.

Machine learning model performance

The model achieved an AUROC of 0.76, demonstrating strong predictive capability. It correctly identified most responders with high sensitivity (86.67%) and specificity (75.00%), yielding a positive likelihood ratio of 3.47 and a negative likelihood ratio of 0.18. The positive predictive value (PPV) (92.86%) confirmed high accuracy in identifying patients likely to benefit from treatment. The F1-score (0.86) indicated a strong balance between precision and recall, minimizing both false positives and false negatives.

Model validation using LOOCV confirmed an accuracy of 84.21%, reinforcing the model’s robustness across individual patient cases. The AUROC remained stable at 0.76, ensuring consistency across dataset partitions. In addition, the F1-score (0.86) remained stable, further validating the model’s predictive reliability. Overall, the model exhibited strong performance in identifying responders, with a particularly high positive predictive value.

Discussion

This study evaluated the efficacy of topical losartan 1 mg/mL in reducing established corneal fibrosis and improving visual acuity. Significant improvements in UDVA and corneal densitometry were observed, particularly in the midperiphery. A machine learning model identified baseline UDVA as the primary predictor of treatment response, leading to a clinically useful scoring system.

The UDVA improvement from 1.04 ± 0.62 LogMAR to 0.74 ± 0.46 LogMAR, with 47% of eyes gaining at least two Snellen lines, suggests that topical losartan facilitates functional visual recovery in corneal fibrosis. This aligns with Pereira-Souza et al.,¹³ who reported UDVA improvement in a post-LASIK haze case, confirming that losartan enhances corneal clarity rather than altering refraction. Similarly, in this study, PHVA remained stable, indicating that visual gains were due to improved transparency. In contrast, Burgos-Blasco et al.¹⁴ found no significant changes in visual acuity or densitometry after 6 months, suggesting that treatment response may depend on regimen or sample size. One patient’s UDVA declined due to severe dry eye, highlighting the need for tear film stability, though PHVA remained stable. Total corneal topography and HOAs also remained stable, suggesting that visual improvements were primarily driven by enhanced stromal transparency rather than changes in corneal optics.

The reduction in corneal densitometry provides objective evidence of losartan’s anti-fibrotic effects at the tissue level. Changes were most pronounced in the midperipheral cornea (6–10 mm zone), with significant reductions in both anterior and total corneal layers. This suggests regional extracellular matrix (ECM) remodeling differences, possibly influenced by localized TGF-β signaling, keratocyte migration, or ECM turnover.^10,16 As this zone serves as a transition between the organized central stroma and the regenerative limbal periphery, ECM turnover or myofibroblast susceptibility may explain this pattern.^10,19 While losartan reduces myofibroblast density via extracellular signal-regulated kinase inhibition and TGF-β blockade, further research is needed to assess its regional effects on corneal biomechanics and wound healing.^3,4,16

Animal models support topical losartan’s efficacy in treating corneal fibrosis through a biphasic mechanism.¹⁶ This justifies prolonged therapy to prevent fibrosis recurrence and sustain corneal transparency.¹⁰ Pereira-Souza et al.¹³ reported total corneal densitometry reduction with losartan, though without regional analysis. In contrast, Burgos-Blasco et al.¹⁴ found no statistically significant densitometry changes over 6 months, despite a trend toward improvement, possibly due to treatment regimen, sample size, or zonal differences.

While total corneal densitometry decreased significantly, post-hoc analyses showed some reductions, particularly in the anterior stroma, which lacked statistical significance. This aligns with fibrosis regression patterns, where deeper stromal layers remodel at different rates due to collagen density variations.¹⁰ The stronger total densitometry response suggests that losartan improves corneal transparency primarily by inducing apoptosis of opaque myofibroblasts and promoting subsequent removal and reorganization of the disordered extracellular matrix these cells deposit, rather than by modulating epithelial clarity.^3,11,20 This mechanism is consistent with Jester et al., who showed that TGF-β signaling drives myofibroblast differentiation and that fibrosis resolution involves both myofibroblast loss and remodeling of the fibronectin-rich matrix.^3,20

This is supported by the correlation between longer leukoma duration and greater densitometry reduction, indicating chronic fibrosis is more reversible under losartan therapy.⁹ Total posterior densitometry trended toward significance, with post-hoc analysis confirming a reduction, suggesting losartan’s effect on posterior stromal transparency, possibly through posterior keratocyte alterations.¹¹ However, higher baseline posterior densitometry values correlated with smaller changes, indicating severe opacity may limit treatment response. The two-phase mechanism hypothesis suggests that longer treatment durations may yield more uniform densitometry improvements.¹⁰

The exclusion of densitometry features from the predictive model suggests that baseline structural opacity alone does not reliably predict visual improvement, highlighting the importance of functional measures such as UDVA. The correlation between worse baseline UDVA and greater densitometry changes in the total central cornea supports this, indicating that improved corneal transparency leads to functional gains rather than an isolated biophysical change. While corneal densitometry is useful for assessing fibrosis resolution, it may not independently predict visual improvement. Instead, a multimodal approach combining densitometry with functional metrics such as visual acuity provides a more comprehensive evaluation of losartan’s effects.

This study introduced a machine learning-based scoring system to predict treatment response, identifying baseline UDVA as the strongest predictor, with worse initial vision (>1.2 LogMAR) associated with greater visual gains. Statistical analysis confirmed baseline UDVA ⩾1.2 LogMAR as the only significant predictor of vision improvement, highlighting initial visual function over structural parameters. While longer leukoma duration showed a trend toward improved vision, it was not statistically significant, suggesting chronicity alone does not determine treatment response. Other baseline factors, including corneal densitometry, history of epithelial defect (resolved before treatment), neovascularization, and leukoma etiology, had no significant impact on outcomes.

Random Forest analysis reinforced these findings, excluding corneal densitometry features due to negligible predictive value, suggesting structural opacity is less critical than visual function.²¹ Etiology-specific trends emerged, with hydrops, fungal keratitis, and chemical burns responding better to losartan, while pterygium resection scars showed lower improvement, though this observation is based on a single case. These findings suggest fibrosis etiology may influence losartan’s effectiveness, possibly due to differences in stromal remodeling pathways, as seen in animal models.^9
–11

Unlike complex predictive models requiring data balancing (e.g., oversampling, undersampling, and class weighting), a scoring system offers a transparent, practical clinical tool.^22,23 This approach was chosen for three reasons: (1) Interpretability, allowing ophthalmologists to use it without computational expertise; (2) Clinical usability, enabling real-time application without computational resources; and (3) Robustness, minimizing overfitting and ensuring applicability across diverse clinical settings, including small datasets.^22
–24 The scoring system, based on weighted clinical variables, identified patients most likely to benefit. A 2.5 threshold score optimized sensitivity and specificity, improving clinical applicability.²⁵ LOOCV validation confirmed model stability, reinforcing its role in clinical decision-making and patient stratification for topical losartan therapy.²⁶

This is the largest prospective study to date on topical losartan for corneal fibrosis, addressing the need for accessible anti-fibrotic therapies. It analyzed 19 eyes from 17 patients in real-world settings and is the third human study, following a case report¹³ and a case series of eight eyes.¹⁴ Unlike Burgos-Blasco’s study, where patients also used topical cyclosporine and steroids, our patients received only losartan and lubricants, isolating losartan’s anti-fibrotic effects.¹⁴

Unlike prior studies, we used machine learning to identify key treatment predictors and corneal densitometry to assess fibrosis changes, refining patient selection and linking structural and functional outcomes. The scoring system improves clinical applicability, strengthening evidence for losartan in corneal fibrosis management. However, the study has limitations. The sample size was relatively small, and the three-month follow-up may not fully capture long-term effects. While the model demonstrated strong performance, external validation in larger cohorts is needed. In addition, the absence of a control group limits definitive conclusions.

In this study, pinhole visual acuity (PHVA) was used instead of corrected distance visual acuity (CDVA) to assess potential visual improvement. PHVA has been shown to reliably estimate best-corrected vision in eyes with irregular astigmatism and high aberrations. It strongly correlates with contact lens–corrected VA and often exceeds spectacle-corrected vision in corneal disease.²⁷ Similarly, it has been reported that PHVA approximates the best attainable acuity in such cases.²⁸ Given the elevated total corneal HOAs in our cohort (3.715 ± 1.19 μm), PHVA provided meaningful clinical insight. While CDVA with contact lenses would have added value, it was not feasible in our hospital setting. Nonetheless, PHVA served as a clinically valid and practical alternative for evaluating visual potential in patients with corneal fibrosis.

Future studies should explore longer treatment durations, higher dosages, or combination therapies to enhance losartan’s anti-fibrotic effects. At the time this study was designed, early guidance based on preclinical rabbit models and initial clinical experience supported a 3-month treatment duration as sufficient to induce myofibroblast apoptosis and assess clinical response.^10
–12 However, more recent studies by Wilson et al. have highlighted a two-phase mechanism, in which stromal remodeling and epithelial basement membrane regeneration proceed more slowly following the initial reduction in myofibroblast density.^15,16 As such, it is likely that some patients in this series may have experienced additional clinical gains with prolonged treatment, and differences between pre- and post-treatment parameters may have been more pronounced with longer follow-up. Future trials should further evaluate extended treatment durations to optimize transparency recovery and functional outcomes. Given the model’s predictive accuracy, integrating machine learning into clinical workflows could improve personalized treatment selection for corneal fibrosis. Comparative trials with other anti-fibrotic agents, such as corticosteroids, would further assess losartan’s synergistic effects, as shown by Sampaio et al.¹¹

In conclusion, this study demonstrated that topical losartan 1 mg/mL significantly improved visual acuity and reduced corneal fibrosis. The machine learning model identified key predictors of treatment response, refining patient selection. While larger cohort validation is needed, these findings support losartan as a promising, accessible therapy for corneal fibrosis.

Footnotes

Acknowledgements

None.

Declarations

ORCID iDs

Luis Haro-Morlett

Angelica Hernandez-Solis

References

Wang

Kong

Wolle

, et al. Global trends in blindness and vision impairment resulting from corneal opacity 1984-2020: a meta-analysis. Ophthalmology 2023; 130(8): 863–871.

Whitcher

Srinivasan

Upadhyay

MP.

Corneal blindness: a global perspective. Bull World Health Organ 2001; 79(3): 214–221.

Jester

J V.

Huang

Petroll

, et al. TGFβ induced myofibroblast differentiation of rabbit keratocytes requires synergistic TGFβ, PDGF and integrin signaling. Exp Eye Res 2002; 75(6): 645–657.

de Oliveira

Tye

Sampaio

, et al. TGFβ1 and TGFβ2 proteins in corneas with and without stromal fibrosis: Delayed regeneration of apical epithelial growth factor barrier and the epithelial basement membrane in corneas with stromal fibrosis. Exp Eye Res 2021; 202: 108325.

Gain

Jullienne

, et al. Global survey of corneal transplantation and eye banking. JAMA Ophthalmol 2016; 134(2): 167–173.

Singh

Dohlman

Ivanov

, et al. Corneal opacity in the United States an American Academy of Ophthalmology IRIS^® Registry (Intelligent Research in Sight) Study. Ophthalmology 132(1): 52–61.

Lim

Lutucuta

Bachireddy

, et al. Angiotensin II blockade reverses myocardial fibrosis in a transgenic mouse model of human hypertrophic cardiomyopathy. Circulation 2001; 103(6): 789–791.

Sampaio

Hilgert

GSL

Shiju

, et al. Topical losartan inhibits corneal scarring fibrosis and collagen type IV deposition after Descemet’s membrane-endothelial excision in rabbits. Exp Eye Res 2022; 216: 108940.

Sampaio

Villabona-Martinez

Shiju

, et al. Topical Losartan decreases myofibroblast generation but not corneal opacity after surface blast-simulating irregular PTK in rabbits. Transl Vis Sci Technol 2023; 12(9): 20–20.

10.

Martinez

Dutra

BAL

Santhiago

, et al. Effect of topical losartan in the treatment of established corneal fibrosis in rabbits. Transl Vis Sci Technol 2024; 13(8): 22–22. doi:10.1167/TVST.13.8.22

11.

Sampaio

Hilgert

GSL

Shiju

, et al. Topical losartan and corticosteroid additively inhibit corneal stromal myofibroblast generation and scarring fibrosis after Alkali burn injury. Transl Vis Sci Technol 2022; 11(7): 9.

12.

Sampaio

Hilgert

GSL

Shiju

, et al. Losartan inhibition of myofibroblast generation and late haze (scarring fibrosis) after PRK in rabbits. J Refract Surg 2022; 38(12): 820–829.

13.

Pereira-Souza

Ambrósio

Bandeira

, et al. Topical losartan for treating corneal fibrosis (Haze): first clinical experience. J Refract Surg 2022;38(11):741–746.

14.

Burgos-Blasco

Moloney

Topical losartan for corneal fibrosis: a case series with densitometry analysis,www.corneajrnl.com (2024).

15.

Wilson

SE.

Topical losartan: practical guidance for clinical trials in the prevention and treatment of corneal scarring fibrosis and other eye diseases and disorders. J Ocul Pharmacol Ther 2023; 39(3): 191–206.

16.

Wilson

SE.

Two-phase mechanism in the treatment of corneal stromal fibrosis with topical losartan. Exp Eye Res 2024; 242: 109884.

17.

Bains

Osathanugrah

Sanjiv

, et al. Diverse research teams and underrepresented groups in clinical studies. JAMA Ophthalmol. 2023; 141(11): 1037–1044.

18.

Olivier

MMG

Forster

Carter

, et al. Lighting a pathway: The minority ophthalmology mentoring program. Ophthalmology 2020; 127(7): 848–851.

19.

Zhou

Cao

, et al. Role of corneal collagen fibrils in corneal disorders and related pathological conditions. Int J Ophthalmol 2017; 10(5): 803–811.

20.

Jester

Huang

Barry-Lane

, et al. Transforming growth factor(beta)-mediated corneal myofibroblast differentiation requires actin and fibronectin assembly. Invest Ophthalmol Vis Sci 1999; 40(9): 1959–1967.

21.

Balaskas

Glinton

Keenan

TDL

, et al. Prediction of visual function from automatically quantified optical coherence tomography biomarkers in patients with geographic atrophy using machine learning. Sci Rep 2022; 12(1): 15565.

22.

Xie

Chakraborty

Ong

MEH

, et al. AutoScore: a machine learning-based automatic clinical score generator and its application to mortality prediction using electronic health records. JMIR Med Inform 2020; 8(10): e21798.

23.

Moss

Corsar

Shaw

, et al. Demystifying the black box: The importance of interpretability of predictive models in neurocritical care. Neurocrit Care 2022; 37(Suppl. 2):185–191.

24.

Charilaou

Battat

Machine learning models and over-fitting considerations. World J Gastroenterol 2022; 28(5): 605–607.

25.

Habibzadeh

Yadollahie

On determining the most appropriate test cut-off value: the case of tests with continuous results. Biochem Med (Zagreb) 2016;26(3):297–307.

26.

Hassanzad

Hajian-Tilaki

Methods of determining optimal cut-point of diagnostic biomarkers with application of clinical data in ROC analysis: an update review. BMC Med Res Methodol 2024; 24(1): 84.

27.

Kanclerz

Results of the pinhole test correlate with hybrid contact lens visual acuity in patients with visual impairment due to corneal diseases. J Ophthalmol 2022; 2022: 4932856.

28.

Kumar

Narang

Holladay

, et al. Optimum pinhole size determination in pinhole pupilloplasty for higher-order aberrations. J Cataract Refract Surg 2024; 50(3): 264–269.