Factors Associated with Vision Outcomes in Microbial Keratitis – a Multisite Prospective Cohort Study

Abstract

Objective:

To investigate factors associated with 90-day visual acuity in microbial keratitis (MK) patients.

Design:

This multicenter prospective cohort study recruited patients with MK from the United States (US) and India from July 23, 2020 through May 1, 2024, and followed them for up to 90 days.

Participants:

Individuals 15 years and older with MK greater than 2mm² in stromal infiltrate area without study exclusion criteria of prior corneal surgery or gluing, impending corneal perforation or keratoplasty, no light perception vision, or current pregnancy or incarceration.

Methods:

Data on sociodemographic features, health and ocular history, symptoms, and clinical measures were gathered. Best-corrected visual acuity (BCVA) was measured in the affected eye at initial and 90-day visits with carried forward BCVA for those healed prior to 90 days. Features were summarized overall and by site. Site-stratified multivariable linear regression models were used to investigate associations with 90-day logMAR BCVA.

Main Outcome Measures:

90-day logMAR BCVA.

Results:

There were 479 participants analyzed after exclusions (n=31) and participants without a final BCVA (n=52). Participants had an average 90-day BCVA of 1.36±1.40 logMAR units (Snellen approximate 20/460 ± 14.0 lines) in the US and 0.70±0.99 (Snellen approximate 20/100 ± 9.9 lines) in India (P<0.0001). For the US cohort, worse 90-day logMAR BCVA was associated with worse presenting BCVA (β=0.05 logMAR unit increase in 90-day logMAR BCVA per 0.1 logMAR unit increase in presenting BCVA, P<0.0001), longer time until presentation (β=0.01 per day, P<0.0001), no contact lens use (β=0.46, P=0.0131), and larger stromal infiltrate area (β=0.02 per 1mm² increase in area, P=0.0082 for bacterial ulcers; β=0.10 per 1 mm² increase in area, P=0.0002 for fungal; interaction p=0.0017). For the India cohort, worse 90-day logMAR BCVA was associated with worse presenting BCVA (β=0.04 logMAR unit increase in 90-day logMAR BCVA per 0.1 logMAR unit increase in presenting BCVA, P<0.0001), longer delays to presentation (β=0.03 per day, P=0.0004), diabetes mellitus (β=0.41, P=0.0019), hypopyon (β=0.27, P=0.0083), no recent ocular trauma (β=0.21, P=0.0370), and larger stromal infiltrate area (β=0.03 per 1mm² increase in area, P<0.0001 for fungal; interaction p=0.0001; nonsignificant β=−0.02 per 1mm² increase in area, P=0.07 for bacterial ulcers;).

Conclusions:

Initial vision, longer time until presentation, and larger stromal infiltrate size conferred risk for worse 90-day BCVA in both cohorts, while other factors were unique. Systems to mitigate care delays and support communities to access care are needed so that clinicians can intervene and improve vision outcomes.

Introduction

Despite microbial keratitis (MK) being a leading preventable cause of blindness,^1–4 clinicians lack knowledge on how patients will respond to treatment. MK is a complex infection to manage given diversity in causal pathogens, patient-related health comorbidities, and presenting disease severity.^1,4–6 An advanced-appearing infection may be highly responsive to treatment, while a moderate, persistent infection can cause permanent vision loss.

Many patients with MK have lasting visual impairment (VI) after resolution.^3,5–9 VI may be due to corneal thinning, surface irregularity, corneal perforation, endophthalmitis, or corneal scarring.^8,10,11 Prior studies and clinical trials in specific MK populations have identified risk factors for poor vision outcomes including worse presenting visual acuity,^8,12–18 delays in time to presentation^{12,13,16,19,20}, longer symptom duration,^18,21,22 and existing ocular surface disease.^18,23 However, these studies and trials have largely been retrospective, restricted to one geographic area, or restricted to one MK organism class. Thus, we are limited in our understanding of organisms, patient’s health, and circumstantial factors associated with MK-related VI.²⁴

Broadening our understanding of these factors would help us to capture phenotypes of MK. This could help us to improve our assessments of disease severity and differentiate MK in nuanced ways to identify potential targets for interventions.³ Studies are also needed to improve clinicians’ ability to assess risk of poor outcomes. Early interventions to prevent MK have been successful, such as providing topical antimicrobial ointment for corneal abrasions;²⁵ however, more work is needed to identify other interventions to address MK-related VI. The purpose of this study was to evaluate the effect of features at patients’ initial clinical presentation on 90-day best-corrected visual acuity (BCVA) in a multisite prospective cohort of MK patients in the United States (US) and India, with the goal of identifying risk factors that may be amenable to novel interventions.

Methods

Study Design & Setting

The Automated Quantitative Ulcer Analysis (AQUA) study is a multisite prospective cohort study of patients with MK recruited at the University of Michigan (UM) Kellogg Eye Center in the US and Aravind Eye Care Systems (AECS) hospital in India from July 23, 2020 to May 1, 2024. The study was sponsored by the National Eye Institute. Participants were enrolled after written informed consent and followed for up to 90 days. Data on patient sociodemographic features, history, infection circumstances and characteristics, and vision were gathered and managed in a Research Electronic Data Capture database hosted at UM.^26,27 The UM Institutional Review Board, AECS Ethics Committee, and the Government of India approved this study and Clinicaltrials.gov registration is available (NCT04420962). This study adhered to the tenets of the Declaration of Helsinki. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines were used for data reporting.²⁸

Participants

Participants included individuals 15 years and older presenting with new MK ≥2mm² in area. Participants were excluded if they had prior incisional cornea surgery (cataract surgery not excluded), corneal gluing, current bandage contact lens placement, perforation or impending perforation that would need keratoplasty, acanthamoeba infection (due to low enrollment numbers), no light perception vision, current pregnancy or incarceration, or inability to provide written informed consent. The study aimed to measure vision over time in MK, which is much more limited in patients with no light perception or perforation requiring keratoplasty or gluing. Additionally, since the study aimed to assess primarily bacterial and fungal keratitis, we opted to exclude patients who had undergone refractive surgery and may be more likely to have other types of infections or those with acanthamoeba. There was no cut-off for maximum length or width of epithelial defect (ED) or stromal infiltrate (SI) area, so long as there was no perforation. Participants enrolled in the study who subsequently had keratoplasty within 7 days, had a positive acanthamoeba PCR, or were found to have had prior corneal incisional surgery, corneal gluing, or MK <2mm² were considered screen failures and withdrawn from the study sample. Given the observational nature of the study, follow-up and treatment was determined by the supervising cornea specialist managing each participant.

Study Variables

Participants’ sociodemographic characteristics, initial symptoms, health history, and infection circumstances, including time between symptom onset and presentation to the study site, were collected. Income classification for the India cohort was based on the modified Kuppusuwamy scale.^29,30 Board-certified ophthalmologists measured MK morphologic features, including ED and SI height, width, and longest length; hypopyon presence and height; and degree of corneal thinning. ED and SI area were approximated by the product of measured ED and SI heights and widths, respectively, as is standard practice. Certified ophthalmic technicians measured the affected eye’s BCVA at 20 feet (US) and 6 meters (India), using Snellen charts. When a participant’s last BCVA was before 90 days but 20/40 or better or their last documented exam reflected a healed infection, their last BCVA was carried forward as their 90-day BCVA to reduce missing data. If these data were missing, the participant was excluded from analysis. A healed infection was defined as no active infiltrate or epithelial defect on their documented exam. For participants who underwent a keratoplasty before 90 days, their last pre-keratoplasty BCVA was carried forward to avoid capturing keratoplasty-related vision changes. Microbe organism classification was determined by cornea specialists on systematic review of laboratory stains and cultures, pathology reports, clinical visit notes, and treatment response over 90 days. Culture-negative cases were included given that many MK cases are culture negative.^2,31–36

Statistical Analysis

Baseline participant features and microbe classification were summarized overall and by site with descriptive statistics (mean, standard deviation [SD], median, interquartile range [IQR], frequency, percentage). No standard Snellen equivalent for BCVA measured as count fingers, hand motion, light perception, or no light perception exists. Therefore, these vision measures were approximated by Snellen values of 20/4000 for count fingers (estimate between Holladay et. al. and Grover et. al.), 20/40000 for hand motion (10 times worse than count fingers), and 20/400000 for light perception (10 times worse than hand motion).^37,38 Vision measured as no light perception was not observed in our sample at baseline (due to exclusion criteria) nor for the 90-day outcome measure, thus a Snellen conversion was not necessary. As a sensitivity analysis, we also used the conversions proposed by Grover et. al. of 20/8000 for count fingers, 20/10000 for hand motion, and 20/12500 for light perception.³⁸ BCVA was then converted to logMAR for analysis.³⁷ Site differences were investigated using two-sample t-, two-sample Wilcoxon, Chi-square, and Fisher’s exact statistical tests, with multiple comparison adjustment using the Holm procedure. Given significant site differences, further analyses were stratified by site. Univariate associations of baseline features and microbe classification with 90-day logMAR BCVA were analyzed using Spearman correlations (r_s) for continuous measures and two-sample Wilcoxon or Kruskal-Wallis (KW) tests for categorical measures. Significant KW tests were followed by post-hoc pairwise comparisons using the Dwass, Steel, Critchlow-Fligner (DSCF) multiple comparison procedure. Site-stratified multivariable linear regression models identified measures independently associated with 90-day logMAR BCVA, using the method of best subsets with the adjusted R square criterion. Interactions with microbe classification were investigated. Model results are presented with estimates and 95% confidence intervals (CI). Given the nature of the logMAR scale, positive model estimates indicate worse 90-day vision. For context, every 0.02 unit or 0.1-unit difference in logMAR equals one letter or one line difference on the Snellen chart, respectively. All statistical analysis used SAS version 9.4 (SAS Institute, Cary, NC).

Results

There were 562 participants enrolled in the AQUA study from 7/23/2020 to 5/1/2024 (Figure 1). There were 479 participants (n=138 US, n=341 India) analyzed after excluding screen failures (n=31) and participants without a 90-day BCVA measure or one to carry forward (n=52). Most participants had a 90-day BCVA versus those who had a prior BCVA carried forward (US: 70.3% vs 29.7%; India: 83.9% vs 16.1%). US, versus India, participants reported a smaller percentage of male sex (46.7% versus 65.7%, P=0.0018) and Asian race (0% versus 100%, P<0.0001) (Table 1, Table S2). More US participants used contact lenses (44.9% versus 0.6%, P<0.001), had self-reported rheumatological conditions (9.4% versus 0.3%, P<0.0001), and prior limited vision (30.4% versus 0.9%, P<0.0001), while fewer had trauma to the infected eye (16.7% versus 71.6%, P<0.0001). US, versus India, participants had significantly worse presenting logMAR BCVA (mean±SD=2.07±1.38 [approximate Snellen equivalent 20/2350 ± 13.8 lines] versus 1.06±0.88 [approximate Snellen equivalent 20/230 ± 8.8 lines], P<0.001) and significantly worse 90-day logMAR BCVA (1.36±1.40 [approximate Snellen equivalent 20/460 ± 14.0 lines] versus 0.70±0.99 [approximate Snellen equivalent 20/100 ± 9.9 lines], P<0.001; Figure 2). US, versus India, participants were less likely to report initial symptoms of pain (73.9% versus 97.4%, P<0.0001) and redness (65.2% versus 94.4%, P<0.0001). There were no significant differences between the cohorts’ initial symptoms of glare from bright lights (38.4% versus 32.6%, P=0.8840) or blurry vision (43.5% versus 40.5%, P=1.0000). US, versus India, participants had significantly larger EDs (19.0±25.1 mm² versus 9.1±10.8 mm², P=0.0172) and deeper infections (48.0±30.9% versus 33.4±16.6%, P=0.0026) with more thinning (14.1±20.9% versus 3.5±9.9%, P<0.0001; percentage showing 0% thinning: 43.8% versus 84.8%, P<0.0001), but no difference in presence of hypopyon was observed (41.2% versus 39.9%, P=1.0000). US participants also had longer time from initial reported symptoms to presentation at the study site (19.8±29.7 days versus 6.3±5.5 days, P<0.0001). In the US, versus India, expert microbe classification revealed greater frequency of bacterial keratitis (BK) (62.3% versus 11.1%, P<0.0001) and lesser frequency of fungal keratitis (FK) (19.6% versus 80.1%, P<0.0001).

Figure 1.

Flow chart of AQUA participant sample. Key: AQUA, Automated Quantitative Ulcer Analysis; US, United States; BCVA, Best-Corrected Visual Acuity; PKP, Penetrating Keratoplasty; CF, Carried Forward

Table 1.

Presenting characteristics of the AQUA study participant sample

	US (N=138)			India (N=341)			Holm-Adj
Continuous Variable	N	Mean (SD)	Median (IQR)	N	Mean (SD)	Median (IQR)	P-value*
Age (years)	138	54.7 (18.0)	55.4 (43.7–68.1)	341	50.3 (13.0)	52.4 (40.6–60.1)	0.0792
Epithelial Defect - Height (mm)	138	3.5 (2.7)	3.0 (1.2–5.0)	341	2.6 (1.8)	2.5 (1.5–3.2)	0.0530
Epithelial Defect - Width (mm)	138	3.6 (2.7)	3.1 (1.3–5.1)	341	2.5 (1.7)	2.4 (1.3–3.5)	0.0027
Epithelial Defect - Longest Length (mm)	138	4.0 (2.8)	3.5 (1.8–6.0)	341	2.8 (1.8)	3.0 (1.5–4.0)	0.0006
Approx Epithelial Defect Area (mm2)	138	19.0 (25.1)	8.8 (2.5–24.8)	341	9.1 (10.8)	6.0 (2.1–11.0)	0.0187
Stromal Infiltrate - Height (mm)	138	3.6 (2.0)	3.2 (2.0–4.8)	341	3.3 (1.4)	3.0 (2.2–4.0)	1.0000
Stromal Infiltrate - Width (mm)	138	3.8 (2.0)	3.5 (2.0–5.0)	341	3.2 (1.4)	3.0 (2.0–4.0)	0.1104
Stromal Infiltrate - Longest Length (mm)	138	4.2 (2.1)	3.7 (2.5–5.6)	341	3.6 (1.4)	3.3 (2.5–4.5)	0.1104
Approx Stromal Infiltrate Area (mm2)	138	16.8 (17.7)	10.1 (4.4–23.0)	341	12.1 (10.4)	9.0 (5.0–15.4)	1.0000
Greatest Posterior Depth (%)	134	48.0 (30.9)	40.0 (20.0–75.0)	341	33.4 (16.6)	30.0 (20.0–40.0)	0.0028
Degree of Thinning (%)	137	14.1 (20.9)	10.0 (0.0–20.0)	341	3.5 (9.9)	0.0 (0.0–0.0)	<0.0001
Presenting Visual Acuity	138	2.07 (1.38)	1.91 (0.80–3.30)	341	1.06 (0.88)	1.00 (0.48–1.48)	<0.0001
Days to presentation	138	19.8 (29.7)	7.0 (4.0–26.0)	341	6.3 (5.5)	4.0 (3.0–7.0)	<0.0001
Categorical Variable	n/Nnon-missing (%)			n/Nnon-missing (%)			Holm-Adj P-value
Sex - Male	64/137 (46.7)			224/341 (65.7)			0.0019
Race
White	123/137 (89.8)			0/341 (0.0)
Black	11/137 (8.0)			0/341 (0.0)			<0.0001
Asian	0/137 (0.0)			341/341 (100.0)
Other/Multiracial	3/137 (2.2)			0/341 (0.0)
Ethnicity - Hispanic	4/130 (3.1)			0/341 (0.0)			0.0830
Diabetes	28/137 (20.4)			43/341 (12.6)			0.1776
Rheumatologic Condition	13/138 (9.4)			1/341 (0.3)			<0.0001
Prior Limited Vision	42/138 (30.4)			3/341 (0.9)			<0.0001
Contact Lens wear	62/138 (44.9)			2/333 (0.6)			<0.0001
Trauma	23/138 (16.7)			244/341 (71.6)			<0.0001
Outdoor Occupation	17/135 (12.6)			--			NA
Healthcare Occupation	15/137 (11.0)			--			NA
Medical Insurance	131/135 (97.0)			--			NA
Ulcer within central 4mm	118/135 (87.4)			228/341 (66.9)			0.0001
Hypopyon	56/136 (41.2)			136/341 (39.9)			1.0000
First Symptoms
Pain	102/138 (73.9)			332/341 (97.4)			<0.0001
Redness	90/138 (65.2)			322/341 (94.4)			<0.0001
Glare from bright lights	53/138 (38.4)			111/341 (32.6)			1.0000
Blurry vision	60/138 (43.5)			138/341 (40.5)			1.0000
Baseline Symptoms
Pain
None	6/138 (4.4)			1/323 (0.3)
Mild	15/138 (10.9)			16/323 (5.0)
Moderate	33/138 (23.9)			143/323 (44.3)			<0.0001a
Severe	53/138 (38.4)			128/323 (39.6)
Very Severe	31/138 (22.5)			35/323 (10.8)
Redness
No problem at all	4/138 (2.9)			0/323 (0.0)
A little bit of a problem	11/138 (8.0)			36/323 (11.2)			<0.0001b
Somewhat of a problem	17/138 (12.3)			111/323 (34.4)
Very much a problem	106/138 (76.8)			176/323 (54.5)
Glare/Light Sensitivity
No problem at all	8/138 (5.8)			0/324 (0.0)
A little bit of a problem	9/138 (6.5)			20/324 (6.2)			0.0005c
Somewhat of a problem	27/138 (19.6)			102/324 (31.5)
Very much a problem	94/138 (68.1)			202/324 (62.4)
Blurry vision
No problem at all	8/138 (5.8)			0/324 (0.0)
A little bit of a problem	4/138 (2.9)			23/324 (7.1)			<0.0001d
Somewhat of a problem	17/138 (12.3)			107/324 (33.0)
Very much a problem	109/138 (79.0)			194/324 (59.9)
Expert Diagnosis
Bacterial	86/138 (62.3)			38/341 (11.1)			<0.0001e
Fungal	27/138 (19.6)			273/341 (80.1)
Viral	6/138 (4.4)			0/341 (0.0)
Inflammatory	15/138 (10.9)			3/341 (0.9)
Polymicrobial	3/138 (2.2)			18/341 (5.3)
Unsure	1/138 (0.7)			9/341 (2.6)

AQUA, Automated Quantitative Ulcer Analysis; US, United States; mm, millimeter; Approx, Approximate; HS, High School; USD, United States Dollar; k, $1000; NA, Not Applicable

^*2-sample t-test, 2-sample Wilcoxon test, Chi-square test, or Fisher's exact test

^aPost-hoc pairwise comparisons revealed a significantly larger percentage of US participants reported no eye pain (p=0.0035), mild eye pain (p=0.0202), or very severe eye pain (p=0.0011), while a significantly smaller percentage reported moderate eye pain (p<0.0001) than India participants

^bPost-hoc pairwise comparisons revealed a significantly larger percentage of US participants reported eye redness was no problem at all (p=0.0078) or very much a problem (p<0.0001), and a significantly smaller percentage reported eye redness was somewhat a problem (p<0.0001) than India participants

^cPost-hoc pairwise comparisons revealed a significantly larger percentage of US participants reported glare/light sensitivity was no problem at all (p<0.0001) and a significantly smaller percentage reported that glare/light sensitivity was somewhat a problem (p= 0.0090) than India participants

^dPost-hoc pairwise comparisons revealed a significantly larger percentage of US participants reported blurry vision was no problem at all (p<0.0001) or very much a problem (p<0.0001), and a significantly smaller percentage reported blurry vision was somewhat a problem (p<0.0001) than India participants

^ePost-hoc pairwise comparisons revealed a significantly larger percentage of US participants were classified as bacterial (p<0.0001), viral (p=0.0005), and inflammatory (p<0.0001), and a significantly smaller percentage were classified as fungal (p<0.0001) than India participants

Figure 2.

Boxplots displaying the distribution of 90-day logMAR best-corrected visual acuity in the AQUA participant sample. Boxplots display the mean (black dot), median (line within the box), 25^th and 75^th percentiles (bottom and top of the box, respectively), and outliers (observations greater than 1.5 times the interquartile range; asterisks) Key: AQUA, Automated Quantitative Ulcer Analysis; US, United States

In the US, univariate analysis showed associations of worse 90-day BCVA with older age (r_s=0.20; P=0.0336), worse presenting BCVA (r_s=0.65; P<0.0001), larger ED area (r_s=0.51, P<0.0001), larger SI area (r_s=0.50, P<0.0001), greater posterior depth of the infection (r_s=0.40, P<0.0001), greater degree of thinning (r_s=0.20; P=0.0336), and longer time until presentation (r_s=0.23; P=0.0196) (Table 3). Additionally, those participants with central ulcers had worse 90-day logMAR BCVA than those without central ulcers (mean±SD=1.47±1.39 [approximate Snellen equivalent 20/600 ± 13.9 lines] versus 0.47±1.05 [approximate Snellen equivalent 20/60 ± 10.5 lines], P=0.0034; Table S4) and those with hypopyon had worse 90-day logMAR BCVA than those without hypopyon (1.86±1.50 logMAR units [20/1500±15 lines] versus 0.98±1.20 logMAR units [20/200±12 lines], P=0.0042, Table S4). There were no significant univariate associations of 90-day BCVA with other demographic characteristics, health history, or infection circumstances; however, participants who reported that blurry vision was somewhat of a problem on presentation had significantly better 90-day vision than those reporting that blurry vision was very much a problem on presentation (Table S2). Multivariable linear regression modeling identified independent associations of worse 90-day logMAR BCVA with main effects for worse presenting logMAR BCVA (β=0.05 per 0.1 logMAR unit, 95% CI 0.03 to 0.06; P<0.0001), and longer time until presentation (β=0.01 per day, 95% CI 0.007 to 0.02; P<0.0001) (Table 5). Conversely, contact lens use (β=−0.46, 95% CI −0.83 to −0.10; P=0.0131) was associated with better 90-day logMAR BCVA (Table 5). A significant interaction was observed between presenting SI area and expert microbe classification such that worse 90-day logMAR BCVA was observed with increasing SI area for both bacterial (β=0.02 per 1mm² increase in SI area, Snellen equivalent of 1 letter, 95% CI 0.004 to 0.03, P=0.0082) and fungal infections (β=0.10 per 1mm² increase in SI area, Snellen equivalent of 5 letters, 95% CI 0.05 to 0.15, P=0.0002), but the effect for fungal infections was significantly larger than that of bacterial infections (interaction P=0.0017; Figure 3A). Multivariable models using ED area, instead of SI area, showed similar association with 90-day vision (data not shown). Additionally, multivariable models using the Grover et al. vision conversions showed similar results (Table S6).³⁸

Table 3.

Univariate association of 90-day best-corrected visual acuity with characteristics of the AQUA participant sample

	US (n=138)		India (n=341)
Continuous Variable	Spearman Correlation with 90-day BCVA	Holm P-value	Spearman Correlation with 90-day BCVA	Holm P-value
Age (years)	0.20	0.0336	0.41	<0.0001
Epithelial Defect - Height (mm)	0.51	<0.0001	0.23	0.0001
Epithelial Defect - Width (mm)	0.47	<0.0001	0.26	<0.0001
Epithelial Defect - Longest Length (mm)	0.49	<0.0001	0.25	<0.0001
Approx Epithelial Defect Area (mm2)	0.51	<0.0001	0.26	<0.0001
Stromal Infiltrate - Height (mm)	0.45	<0.0001	0.36	<0.0001
Stromal Infiltrate - Width (mm)	0.44	<0.0001	0.35	<0.0001
Stromal Infiltrate - Longest Length (mm)	0.42	<0.0001	0.35	<0.0001
Approx Stromal Infiltrate Area (mm2)	0.50	<0.0001	0.39	<0.0001
Greatest Posterior Depth (%)	0.40	<0.0001	0.22	0.0003
Degree of Thinning (%)	0.20	0.0336	0.17	0.0062
Presenting Visual Acuity	0.65	<0.0001	0.63	<0.0001
Days to presentation	0.23	0.0196	0.25	<0.0001

AQUA, Automated Quantitative Ulcer Analysis; US, United States; BCVA, Best Corrected Visual Acuity; Approx, Approximate; mm, millimeter

Table 5.

Multivariable linear regression model results for independent associations of participant and infection characteristics with 90-day logMAR best-corrected visual acuity in the AQUA participant sample, stratified by site

Variable	Estimate	95% CI	P-value*
US
Main Effects
Contact Lens Use	−0.46	(−0.83, −0.10)	0.0131
Presenting logMAR VA (per 0.1)	0.05	(0.03, 0.06)	<0.0001
Days to Presentation (per 1 day)	0.01	(0.007, 0.02)	<0.0001
Interaction: Expert classification × Presenting SI Area (Effect of 1mm2 increase in presenting SI area by expert classification)
Bacteriala	0.02	(0.004, 0.03)	0.0082
Fungalb	0.10	(0.05, 0.15)	0.0002
Inflammatory	−0.03	(−0.08, 0.03)	0.3720
Polymicrobial	0.03	(−0.06, 0.13)	0.4509
Viral	0.12	(−0.12, 0.37)	0.3114
India
Main Effects
Diabetes	0.41	(0.15, 0.67)	0.0019
Trauma	−0.21	(−0.40, −0.01)	0.0370
Presenting logMAR VA (per 0.1)	0.04	(0.03, 0.05)	<0.0001
Days to Presentation (per 1 day)	0.03	(0.01, 0.04)	0.0004
Hypopyon	0.27	(0.07, 0.47)	0.0083
Interaction: Expert classification × Presenting SI Area (Effect of 1mm2 increase in presenting Stromal Infiltrate area by expert classification)
Bacteriala	−0.02	(−0.04, 0.002)	0.0719
Fungal	0.03	(0.02, 0.04)	<0.0001
Inflammatory	0.15	(−1.77, 2.07)	0.8775
Polymicrobial	−0.05	(−0.16, 0.05)	0.3230
Unsure	0.04	(−0.02, 0.10)	0.1649

AQUA, Automated Quantitative Ulcer Analysis; CI, Confidence Interval; US, United States; BCVA, Best- Corrected Visual Acuity; Approx, Approximate; mm, millimeter

^aInteraction: Bacterial slope is significantly different than fungal slope (US: p=0.0017, India: p=0.0001)

^bInteraction: Fungal slope is significantly different than inflammatory slope (p=0.0014)

Figure 3.

Plot displaying the interaction effect between presenting stromal infiltrate area and microbial keratitis type on 90-day logMAR best-corrected visual acuity, for A) US AQUA participants, and B) India AQUA participants. US, United States

In India, univariate analysis showed associations of worse 90-day logMAR BCVA with older age (r_s=0.41; P<0.0001), worse presenting BCVA (r_s=0.63; P<0.0001), larger ED area (r_s=0.26, P=0.0002), larger SI area (r_s=0.39, P<0.0001), greater posterior depth (r_s=0.22, P=0.0003), greater degree of thinning (r_s=0.17, P=0.0062), longer time until presentation (r_s=0.25, P<0.0001), and lower educational attainment (P<0.0001) (Table 3; Table S2). Additionally, those participants with central ulcers had worse 90-day logMAR BCVA than those without central ulcers (mean±SD=0.84±1.09 [approximate Snellen equivalent 20/140 ± 10.9 lines] versus 0.40±0.65 [approximate Snellen equivalent 20/50 ± 6.5 lines], P=0.0034; Table S4) and those with hypopyon had worse 90-day logMAR BCVA than those without hypopyon (1.12±1.20 logMAR units [20/250±12 lines] versus 0.42±0.71 logMAR units [20/50±7 lines], P<0.0001, Table S4). Multivariable linear regression modeling identified independent associations of worse 90-day logMAR BCVA with main effects for worse presenting logMAR BCVA (β=0.04 per 0.1 logMAR unit, 95% CI 0.03 to 0.05; P<0.0001), longer time until presentation (β=0.03 per day, 95% CI 0.01 to 0.45; P=0.0004), self-reported diabetes mellitus (β=0.41, 95% CI 0.15 to 0.67; P=0.0019), and hypopyon (β=0.27, 95% CI 0.07 to 0.47; P=0.0083) (Table 5). Conversely, better 90-day logMAR BCVA was associated with recent trauma to the affected eye (β=−0.21, 95% CI −0.40 to −0.01; P=0.0370). A significant interaction was also observed between presenting SI area and organism classification such that worse 90-day logMAR BCVA was observed with increasing SI area for fungal infections (β=0.03 per 1mm² increase in SI area, Snellen equivalent 1.5 letters, 95% CI 0.02 to 0.04, P<0.0001) but not for bacterial infections (β=−0.02 per 1mm² increase in SI area, 95% CI −0.04 to 0.002, P=0.0719; interaction P=0.0001; Figure 3B). Multivariable models using the Grover et. al. vision conversions showed similar results (Table S6).³⁸

Discussion

This study analyzed a large multisite prospective cohort of MK patients. US participants had more advanced, vision-threatening infections with worse presenting BCVA, larger average ED area, and greater average infection depth and corneal thinning than India participants, possibly due to differences in site-specific referral patterns, causative organisms, and recruitment to the study. It is more common practice in India to perform urgent keratoplasty for patients with large infections, a population that would then not be recruited for prospective observation. US participants had worse 90-day logMAR BCVA than India participants. There was a distinct distribution of MK microbes between sites: most US participants had BK, while most India participants had FK, consistent with geographic organismal differences.^1,39 There were also notable site differences in ocular history factors, with significantly greater contact lens use in the US versus India, and frequent antecedent ocular trauma in India versus minimal in the US. As a result of differences, site-specific analysis was executed. Despite differences, when site-stratified multivariable models were performed, there were overlapping associations with 90-day BCVA in both cohorts. Other factors were unique to each site. These findings offer important prognostic and management implications.

In both cohorts, worse presenting BCVA and longer time until presentation were significantly associated with worse 90-day BCVA. Every one-line worsening in presenting BCVA was associated with a two (US) or two and a half (India) letter worsening in 90-day BCVA, aligned with previous studies.^{8,12,15,17,40–43} The associations of worse presenting BCVA with worse 90-day BCVA could be explained by organismal virulence, causing more severe initial VI due to toxin release and, consequently, poorer recovery.^44,45,46 Longer time from initial symptoms to presentation was independently associated with worse 90-day BCVA in both cohorts. Delays in care have been associated with worse outcomes.^{12,13,16,19,20} Care delays represent a key modifiable factor, so future work should focus on minimizing delays, including identifying system-level barriers to improve timely MK care, improving referral pathways, and identifying patients’ social risk factors that keep them from care.^42,47

Larger SI area was also associated with worse 90-day vision in both cohorts; however, this effect was modified by organism classification in different ways for each cohort. For example, US participants with either BK or FK had a stepwise decrease in their 90-day vision for every 1mm² increase in their SI area, which was independent of presenting BCVA or time until presentation. Additionally, the effect size of this stepwise decrease in 90-day BCVA was significantly larger for US participants with FK (5 letters per 1mm²) than those with BK (1 letter per 1mm², interaction P=0.0017). In contrast, only India participants with FK were observed to have a stepwise decrease (1.5 letters per 1mm²) in 90-day BCVA for every 1mm² increase in SI area; there was no significant association of 90-day BCVA and SI area for India participants with any non-fungal keratitis, which was significantly different from India participants with FK (interaction P=0.0001). However, the lack of association between SI area and 90-day BCVA in BK for India participants may in part be due to low power as the prevalence of BK in our India sample was low (n=38, 11.1%). Signs of more severe morphology, such as larger ED and SI area, presence of hypopyon, and severe corneal thinning or perforation, have correlated with worse outcomes.^{12,14,17,18,48,49} In our analysis, each of these morphologic features had significant univariate associations; however, in a multivariable model only SI area remained independently associated, implying that these features overlap in their effect on outcomes. Size is likely a marker for organism burden and the effectiveness (or lack thereof) of the host response, so patients with larger infections will take longer to respond to treatment and, consequently, have more potential for scarring and correspondingly worse vision. Interestingly, at both sites, we identified a novel interaction between SI area and organism classification, which underscores the poor prognosis of FK relative to non-fungal keratitis.^8,12,17,50 Fungal infections may be more difficult to treat due to poor ocular penetration of antifungal medications, which could explain this finding.⁵¹ New agents to treat fungal or more effectively phenotype fungal disease are warranted, as well as close follow-up of these patients at greater risk for permanent VI.

In the US, contact lens use was independently associated with better 90-day vision. This finding has been attributed to less virulent strains and less severe presentations of Pseudomonas BK – the most common MK etiology in contact lens wearers.^52–54 Contact lens wearers likely are more aware of infection risks, have better access to eye care, or have higher socioeconomic status than non-wearers. Additionally, clinicians who treat contact lens patients may have a higher index of suspicion and treat with more tailored medications prior to receiving the culture results. There may also be latent, unmeasured factors that account for the protective relationship between contact lens use and 90-day vision, which is an important consideration. The apparently “protective” effect of contact lens use should be tempered by the fact that contact lens wear is a known risk factor to develop MK, so contact lens hygiene remains a pillar to emphasize with patients.^55,56

In India, patients with hypopyon and diabetes mellitus had worse 90-day vision. Hypopyon presence indicates greater MK inflammation, previously associated with greater risk for corneal perforation and poor treatment response.^57–59 While there were similar rates of hypopyon between the two cohorts, hypopyon only conferred a worse visual prognosis in India. It is unclear the cause of this difference between cohorts. Previous research has shown that BK patients with diabetes mellitus in Saudi Arabia were 3.5 times more likely to have a final visual acuity <20/200 or need surgery.^59,60 Diabetes mellitus patients have increased MK risk due to greater ocular surface fragility from compromised tear film production.^60,61 Additionally, neural dysregulation in diabetes mellitus confers delayed corneal healing,^60,62–64 which may explain worse or delayed vision recovery and greater need for surgical intervention.^59,60 Identifying diabetes mellitus as a risk factor for poor vision outcomes may encourage clinicians to take a more cautious, proactive approach to managing these patients.

In India, trauma-induced keratitis was a vision-protective factor. Ocular trauma is a frequently cited MK risk factor in regions with farming and with high-risk machinery labor;^9,18 antecedent ocular trauma occurred in 70.4% of India participants versus 15.5% of US participants. However, there is conflicting evidence on ocular trauma’s impact on vision outcomes.^9,13,65–68 Perhaps patients with trauma-induced MK present sooner than non-trauma patients; however, trauma patients had better vision even after controlling for time until presentation.¹⁹ Organisms associated with trauma-related MK may also be more amenable to available treatments, which could explain improved outcomes. Alternatively, clinicians in India may be more familiar with targeted therapies for trauma-induced keratitis, like how US clinicians are used to treating contact lens wear-related ulcers. Unfortunately, the sample size of this study was not amendable to a deep investigation of the various medication treatments provided to participants.

Our study’s strengths are its sample size, clinical diversity, and multicenter design. Prior prospective studies and trials have been restricted to one geographic region or organism subtype.^{12–14,17,19,40,41,43,67–69} We identified overlapping and cohort-specific features associated with MK vision outcomes in the US and India, and compared prognostic features in western versus eastern geographic regions and tropical versus temperate climates. Given the prospective nature of this study, a comprehensive assessment of baseline morphology and history factors was possible. Yet, this study has limitations. Our study exclusion criteria limited enrollment of patients with severe MK who may have had perforation or gluing necessitating early intervention, or severe pain limiting their ability to participate. As such, our findings are not generalizable all MK types or severity and may be more applicable to those with mild-to-moderate MK. It was not feasible to screen all patients with MK or follow all screened patients given study team constraints and, thus, it is impossible to know how representative this study population was. While most participants were healed by the study’s conclusion, results are only applicable to the 90-day study window, so long-term outcomes are unknown. Several participants were lost to follow up with a last BCVA <20/40, potentially skewing results. Carrying forward the last measured BCVA in these cases was deemed inappropriate, as these patients could have continued to heal and, thus, chose not to attend appointments. The goal was to reduce bias in the analysis. Finally, associations with vision do not imply causality, so clinical trials are a necessary next step to codify the benefits of risk factor modification.

Multiple MK prognostic indicators offer avenues for novel interventions to address MK-related VI, including rapid triage systems and more aggressive management of patients with large ulcers, fungal ulcers, or those with worse presenting vision. In the US, non-contact lens wearers and those with FK are at higher risk for worse VI after MK, so caution should be taken in their care. In India, patients with diabetes mellitus, corneal thinning, and non-traumatic causes of keratitis should be monitored closely. Policy changes to improve care gaps should be evaluated to determine if outcomes can be improved. We suspect that institutional changes to support timely, accessible treatment will be critical in improving MK-related VI, especially in rural settings or with vulnerable patients. Evaluation of referral and follow-up pathways in tandem with social work services will be essential to ensure timely, equitable access to follow up visits with trained clinicians. Further research is necessary to establish how treatment responsiveness affects vision outcomes as patients improve, remain unchanged, or deteriorate. Latent factors that could not be explored in the present study, including the profound effect of medication adherence and the home environment, should be evaluated in future work. Finally, the expansion of novel, targeted interventions to address modifiable factors will be critical as we strive to reduce the global burden of MK-related VI.