Corneal Specialists’ Confidence in Identifying Causal Organisms of Microbial Keratitis

Abstract

Purpose:

Microbial keratitis (MK) is a potentially blinding corneal disease caused by an array of microbial etiologies. However, the lack of early organism identification is a barrier to optimal care. We investigated clinician confidence in their diagnosis of organism type on initial presentation and the relationship between confidence and presenting features.

Methods:

This research presents secondary data analysis of 72 patients from the Automated Quantitative Ulcer Analysis (AQUA) study. Cornea specialists reported their confidence in organism identification. Presenting sample characteristics were recorded including patient demographics, health history, infection morphology, symptoms, and circumstances of infection. The association between confidence and presenting characteristics was investigated with 2-sample t-tests, Wilcoxon tests, and Chi-square or Fisher’s exact tests.

Results:

Clinicians reported being “confident or very confident” in their diagnosis of the causal organism in MK infections for 39 patients (54%) and “not confident” for 33 patients (46%). Confidence was not significantly associated with patient demographics, morphologic features, or symptoms related to MK. MK cases where clinicians reported they were confident, versus not confident in their diagnosis, showed significantly smaller percentages of previous corneal disease (0% versus 15%, p=0.017), were not seen by an outside provider first (69% versus 94%, p=0.015), or had no prior labs drawn (8% versus 33%, p=0.046), and a significantly larger percentage of cases wore contact lenses (54% versus 28%, p=0.029).

Conclusion:

In almost half of MK cases, cornea specialists reported lack of confidence in identifying the infection type. Confidence was related to ocular history and circumstances of infection but not by observable signs and symptoms or patient demographics. Tools are needed to assist clinicians with early diagnosis of MK infection type to expedite care and healing.

Introduction:

Microbial Keratitis (MK) is an infection of the cornea that causes pain, visual impairment, and is a major cause of blindness. Annually, there are an estimated 2 million cases of MK worldwide.¹ MK requires immediate medical intervention and close monitoring to prevent progression. If not effectively treated, patients may lose vision or have permanent ocular damage. Common predisposing factors include contact lens use, trauma, ocular surgery, ocular surface disease, and systemic disease (including diabetes mellitus and herpetic keratitis).^2–4

Various species of bacteria, viruses, fungi, and parasites may cause MK.⁵ Diagnosis of MK typically involves a patient history of microbial exposure and features of MK on examination.⁶ These components, however, are typically insufficient to accurately predict the causal organism which requires culture, staining or sending the organism out for genetic testing.^6,7 Additionally, laboratory studies have lower yield rates and results can be delayed limiting clinicians ability to diagnose the organism at the initial clinical encounter.⁸ Culture results have been reported to be negative in up to 60–80% of cases.^8–10 As a consequence of clinicians’ diagnostic uncertainty, initial treatment often involves broad microbial coverage.¹¹ Use of non-specific or multiple antimicrobial agents may lead to ocular surface toxicity, competition of medication on the eye, and undue costs to the patient.^12–14 Overuse of antibiotics is a serious concern when considering multidrug resistant bacteria and also can change the individual patient’s ocular surface microbiome.^15–18

Cornea specialists rely on clinical judgment to diagnose the infectious organism for treatment decisions. Clinical judgment can be guided by several factors, including knowledge, experience, and geographic area. However, it may not result in an accurate diagnosis. Although certain features may be more likely in certain infection types,^19,20 no single MK feature is considered truly pathognomonic for any specific microbe.²¹ A study by Redd et al. evaluated the performance of 66 cornea specialists from 16 countries in distinguishing between images of bacterial and fungal corneal ulcers. These corneal specialists were informed that 50 images would include ulcers from culture-proven bacterial infections and 50 images would include ulcers from culture-proven fungal infections. Overall, the average area under the receiver operating characteristic curve was 0.61, which suggests that visual diagnosis of MK etiologies is not particularly accurate.²²

In this study, we aimed to investigate the degree of confidence cornea specialists have in their diagnosis of the class of organism causing MK infection on initial presentation and to evaluate if there are factors of the patient’s history or morphologic features of the infection related to their confidence. The clinicians involved in this study were cornea specialists at a tertiary referral center who see high volumes of MK cases. Thus, their rating of confidence is likely a reflection of the best possible confidence rating among United States clinicians.

Materials and Methods:

This research presents secondary data analysis of a sample of patient recruited for the Automated Quantitative Ulcer Analysis (AQUA) study. Briefly, the primary AQUA study is designed to create decision-aid tools for ophthalmologists to manage MK. AQUA includes patients from two sites, the Kellogg Eye Center of the University of Michigan (UM) and Aravind Eye Care Systems in Madurai, India. Patients were enrolled in the study if they were age 15 or older and had clinically significant MK (stromal infiltrate > 2mm²) and provided written informed consent. Patients were excluded if they had prior incisional cornea surgery, no light perception vision or if they were members of vulnerable populations. The AQUA study was reviewed and approved by the University of Michigan Institutional Review Board (HUM00174923) and adheres to the Tenets of the Declaration of Helsinki.

Only participants enrolled at the UM site between August 2020 and June 2022 were included in the current research to investigate clinician confidence in diagnosis the causal organism in cases of MK. The process of collecting specimens and determining organism types adheres to a different workflow between the AQUA sites. Aravind has on-site laboratory testing where specimens are stained and interpreted within hours of collection. Therefore, confidence in determination of organism type at Aravind is influenced by laboratory test results. At UM, clinicians follow a more standard workflow for eye clinics in the United States with off-site laboratory diagnostic testing where results are not available the same day. Therefore, confidence in determination of organism type at UM is not biased by laboratory results and instead is reliant on cornea specialist clinical judgment.

Eight board-certified cornea specialist faculty and four cornea fellows participated in the AQUA study at UM. Cornea specialists examined MK patients in-person at presentation for presence/size of epithelial defects, stromal infiltrate length and width, greatest posterior ulcer depth (as a percentage of total cornea thickness), corneal thinning (percentage), hypopyon height, ulcer location (central or peripheral), presence of satellite lesions, conjunctival injection, and other features of corneal ulcers. Examinations were performed by one cornea specialist per patient. After conclusion of the patient examination, the treating cornea specialist provided their diagnosis of the class of organism causing MK and estimated the confidence level for that diagnosis. Confidence ratings were reported as “not very confident”, “confident”, and “very confident”. Clinicians used all data available at presentation to determine their confidence in diagnosis including examination data of infection morphology, patient questionnaire responses, and patient-reported history (including previous therapy for MK). Participant data were collected via electronic health records and patient questionnaires and managed in a study database (REDCap electronic data capture tool).^23,24 Data collected (Table 1) include demographic information, including reported gender, race, ethnicity, level of education, income, and whether the patient had medical insurance. We also recorded ocular history, including previous corneal disease, whether the patient initially visited a provider outside of the UM system, date of symptom onset, laboratory studies and medications ordered by an outside provider, reported symptoms, and circumstances related to infection (trauma, swimming, contact lens use). Outside provider was defined as any physician outside of the cornea department at UM. In the current analysis, only data collected on a patient’s initial visit to UM were used.

Table 1.

Summary of data collected.

Demographics	History	Slit-Lamp Exam
Age (years)	Previous Corneal Disease (Yes, No)	Presenting Logmar VA
Sex (Male, Female)	Outside provider seen (Yes, No)	Epi Defect Size (Height, Width, Longest Axis)
Race (White, Black, Asian, Other/Multiracial)	Outside provider prescribed medications (Yes, No)	Stromal Infiltrate Size (Height, Width, Longest Axis)
Ethnicity (Hispanic, Non-Hispanic)	Outside provider labs performed (Yes, No)	Hypopyon Height (if present)
Level of Education (< High School, High School Diploma, Some College, College Degree, Other)	Reported Symptoms (Pain, Redness, Glare, Blurry Vision)	Posterior Depth of Ulcer
Income (<$25k, $25k-$50k, $51k-$100k, >$100k)	Circumstances of Infection (Trauma, Swimming, Contact Lens Use)	Percent Corneal Thinning
Medical Insurance (Yes, No)		Epi Defect Presence
		Stromal Infiltrate Area
		Location (Central, Not Central)
		Satellite Lesions (Yes, No)

Statistical Methods

Characteristics of the participant sample were summarized descriptively with means and standard deviations (SD) for continuous measures and frequencies and percentages for categorical measures. Participant demographics, circumstances of infection, and characteristics of the infection were compared between cases where cornea specialists were very confident or confident in their diagnosis of MK infection type and those where cornea specialists were not very confident using 2-sample t-tests, 2-sample Wilcoxon tests, Chi-square tests, and Fisher’s exact tests. Corneal specialists’ confidence in diagnosis was compared between the specific classes of organisms diagnosed as causing MK with Fisher’s exact test, including post-hoc pairwise comparisons with Holm’s adjustment for multiple comparisons. The association between corneal specialist experience (faculty versus fellow) and confidence in diagnosis was investigated with a Chi-square test, and differences between the 12 corneal specialists for confidence in diagnosis was investigated with a Fisher’s exact test. All statistical analyses were performed with SAS version 9.4 (SAS Institute, Cary, NC).

Results:

Seventy-two participants with MK were enrolled during the study period with complete data at the UM site. Cornea specialists reported they were very confident with their diagnosis in 3% of cases (n=2), confident in 51% (n=37), and not very confident in 46% (n=33). Patient participants were on average 51.5 years old (SD=19.7) at presentation, 42% Male, 86% White, 10% Black, and 5% Hispanic. Education was reported as less than high school for 11%, high school graduate for 19%, some college for 28%, college graduate for 29%, and graduate education for 13%. Income was distributed with 24% of participants reporting total household yearly income <$25,000, 25% reporting $25,000-$50,000, 24% reporting $51,000-$100,000, and 27% reporting >$100,000.

Cornea specialists diagnosed MK infections as bacterial in 58% of cases (n=42), fungal in 14% (n=10), viral in 7% (n=5), inflammatory/non-infectious in 8% (n=6), acanthamoeba in 7% (n=5), and polymicrobial for 6% (n=4). Confidence in diagnosis was associated with type of MK infection type diagnosed (p=0.0001, Table 2). Cornea specialists were confident or very confident for 73.8% of infections they diagnosed as bacterial, 40.0% as fungal, 50.0% as inflammatory, 0% as acanthamoeba, 20.0% as viral, and 0% as polymicrobial. After adjustment for multiple comparisons, the only significant difference in confidence was between those diagnosed as bacterial versus acanthamoeba (73.8% versus 0%, Holm-adjusted p=0.0420).

Table 2.

Association between treating corneal specialist diagnosis and confidence in diagnosis

		Very Confident/Confident (n=39)	Not Very Confident (n=33)
Diagnosis	n	# (Row %)	# (Row %)	p-value*
Bacterial	42	31 (73.8)	11 (26.2)	0.0002a
Fungal	10	4 (40.0)	6 (60.0)
Inflammatory	6	3 (50.0)	3 (50.0)
Acanthamoeba	5	0 (0.0)	5 (100.0)
Viral	5	1 (20.0)	4 (80.0)
Polymicrobial	4	0 (0.0)	4 (100.0)

^*Fisher’s exact test

^apost-hoc pairwise comparisons with Holm-adjustment showed that clinicians had more confidence in their fungal diagnoses than acanthamoeba diagnoses (p=0.0420)

The distribution of participant demographic characteristics was similar between those cases where cornea specialists were very confident/confident in their MK infection type diagnosis and those where cornea specialists were not very confident (Table 3). No significant differences were found for mean age (50.0 years versus 53.3 years, respectively; p=0.476), reported gender (46% male versus 36%; p=0.401), race (80% White versus 94%; p=0.217), ethnicity (9% Hispanic versus 0%; p=0.243), education (28% at least high school education versus 33%; p=0.996), or income (41% ≤$50,000 versus 59%; p=0.338). A similar percentage of participants reported having medication insurance among cases where cornea specialists were very confident/confident in their MK infection type diagnosis and not very confident (97% versus 87%; p=0.163).

Table 3.

Comparison of participant demographic characteristics between cases where cornea specialists were very confident/confident in their diagnosis and those where cornea specialists were not very confident.

	Very Confident/Confident (n=39)	Not Very Confident (n=33)
Continuous Variable	Mean (SD), Median	Mean (SD), Median	p-value*
Age (years)	50.0 (16.7), 53.2	53.3 (22.9), 51.7	0.476
Categorical Variable	# (Column %)	# (Column %)	p-value**
Sex
Male	18 (46.2)	12 (36.4)	0.401
Female	21 (53.9)	21 (63.6)
Race
White	31 (79.5)	31 (93.9)	0.217
Black	5 (12.8)	2 (6.1)
Asian	0 (0.0)	0 (0.0)
Other/Multiracial	3 (7.7)	0 (0.0)
Ethnicity
Hispanic	3 (8.6)	0 (0.0)	0.243
Non-Hispanic	32 (91.4)	30 (100.0)
Education
<High School	4 (10.3)	4 (12.1)	0.996
High School Diploma	7 (18.0)	7 (21.2)
Some College	11 (28.2)	9 (27.3)
Graduate Degree	12 (30.8)	9 (27.2)
Other	5 (12.8)	4 (12.1)
Income	2 (5.1)	0 (0.0)
<$25k	6 (17.7)	9 (31.0)
$25k-$50k	8 (23.5)	8 (27.6)	0.338
$51k-$100k	11 (32.4)	4 (13.8)
>$100k	9 (26.5)	8 (27.6)
Medical Insurance	38 (97.4)	27 (87.1)	0.163

SD, standard deviation; k, $1000

^*2-sample t-test;

^**Chi-square or Fisher's exact test (if cell counts <5) for categorical variables

note: percentages are reported on the non-missing sample

Some participant characteristics related to ocular history and circumstances of infection showed significant differences between participants with different diagnostic confidence (Table 3). Specifically, participants where cornea specialists were very confident/confident in their diagnosis, versus not very confident, showed significantly smaller percentages of previous corneal disease (0% versus 15%; p=0.017), being seen by an outside provider before presenting to UM (69% versus 94%; p=0.017), and having laboratory testing by an outside provider before presenting to UM (8% versus 33%; p=0.046). A significantly larger percentage of participants with very confident/confident diagnoses had used contact lenses in the seven days prior to infection than those with a not very confident diagnosis (54% versus 28%; p=0.029). No significant differences were found between cases with different diagnostic confidence with respect to participants being given medications to treat their MK by an outside provider (p=0.099), reported symptoms (pain, p=0.659; redness, p=0.966; glare, p=0.686; blurry vision, p=0.782), trauma to the infected eye (p=0.327), or swimming/water exposure within seven days of infection (p=0.327).

Confidence did not have statistically significant differences in relation to MK morphology (Table 4). Epithelial defect was present in 95% of participants with very confident/confident diagnoses and in 79% with not very confident diagnoses (p=0.070); hypopyon was present in 36% of participants with very confident/confident diagnoses and in 21% with not very confident diagnoses (p=0.172). Infiltrate sizes were not significantly different between participants with very confident/confident diagnoses versus not very confident diagnoses, including the longest axis length of the epithelial defect (mean ± SD, 4.1±2.7mm versus 3.6±1.9mm; p=0.706), longest axis length of the stromal infiltrate (4.1±2.2mm versus3.9±1.6mm; p=0.821), stromal infiltrate area (>5mm² in 67% versus 67%; p=1.000), and hypopyon height (1.4±0.9mm versus 1.4±1.1mm; p=1.000). Percentage depth of infection (50±29% versus 38±23%; p=0.092) and thinning (20±22% versus 18±26%; p=0.346) were not significantly different for cases where cornea specialists reported very confident/confident diagnoses versus not very confident diagnoses.

Table 4.

Comparison of participant ocular history and circumstance of infection between cases where cornea specialists were very confident/confident in their diagnosis and those where cornea specialists were not very confident.

	Very Confident/Confident (n=39)	Not Very Confident (n=33)
Categorical Variable	# (Column %)	# (Column %)	p-value*
Previous Corneal Disease	0 (0.0)	5 (15.1)	0.017
Outside provider seen	27 (69.2)	31 (93.9)	0.015
Outside provider meds	21 (53.9)	24 (72.7)	0.099
Outside provider labs	2 (8.3)	10 (33.3)	0.046
Reported Symptoms
Pain	29 (74.4)	26 (78.8)	0.659
Redness	27 (69.2)	23 (69.7)	0.966
Glare	16 (41.0)	12 (36.4)	0.686
Blurry Vision	19 (48.7)	15 (45.5)	0.782
Circumstances of Infection
Trauma	4 (10.5)	6 (19.4)	0.327
Swimming	1 (2.3)	3 (9.1)	0.327
CTL use	21 (53.9)	9 (28.1)	0.029

CTL, contact lens

^*Chi-square or Fisher's exact test (if cell counts <5)

note: percentages are reported on the non-missing sample

The 12 corneal specialists participating in this research included 8 board-certified cornea specialist faculty (66.6%) and 4 fellows (33.3%). Faculty had experience ranging from 3 to 48 years (mean=20, median=15.5). No association was observed between corneal specialist type and confidence in diagnosis such that 55% of MK cases seen by faculty had confident or very confident diagnoses compared to 50% of cases seen by fellows (p=0.7274). Further, no significant differences in confidence in diagnosis were observed between the 12 corneal specialists (p=0.3665).

Discussion:

Cornea specialists were not confident in identifying the MK organism type for almost half (46%) of MK patients at presentation. Although it is important for confidence to ultimately be linked to a final accurate diagnosis, the focus of this study was to investigate confidence in diagnosis at presentation. Laboratory results are often not known at presentation, yet clinicians still initiate medications or therapeutic procedures. This is a barrier to optimal care.

In this cohort, there was no relationship between cornea specialists’ confidence and patient demographics and ulcer morphologic features or clinical symptoms. Organisms that cause MK may have distinct morphological features.²⁵ These morphological features, however, do not always exist and may overlap between organisms. Therefore, the determination of causal organisms in MK can be difficult to determine based on physical features alone and is reflected in clinician confidence in organism diagnosis.

MK cases where cornea specialists were not confident in their diagnosis of MK organism type were more likely to have reported previous corneal disease. One possible explanation for this result is that previous corneal disease may increase the variability of exam findings or circumstances of infection. Corneal disease may alter innate immunity, including corneal nerves, epithelium, and keratocytes.²⁶ Prior corneal scarring may also alter appearance on slit lamp examination.

MK cases where cornea specialists were not confident in their diagnosis of Mk organism type were more likely to have received laboratory testing from an outside provider. One possible explanation for this finding is that patients with laboratory results from outside providers were referred to UM for more complex cases. Patients who present to UM, a tertiary care center, are often first seen by outside providers. Cornea specialists may be less confident after patients had seen an outside provider because more complex cases are often referred to tertiary care hospitals. Upon closer review of the twelve patients with laboratory results from outside providers, 4 had negative cultures, 5 had pending results, and only 3 had positive cultures. This further demonstrates the limitations of current laboratory testing in determining causal organisms of MK.

MK cases where cornea specialists were confident in diagnosing the MK infection type were more likely to have used contact lenses. Bacterial organisms causing MK, such as pseudomonas aeruginosa, have been associated with contact lens wear.^27,28 Acanthamoeba and fungal infections can be associated with MK in contact lens wearers, but are less prevalent in UM’s northern United States climate.^29,30 This local knowledge of organism prevalence and risk factors likely led to increased confidence in determining MK infection type in patients who use contact lenses.³¹

There are some limitations to this study. By using data collected from one institution, the confidence ratings were from a select number of cornea specialists. This may limit the generalizability of our findings. The cornea specialists reported confidence on a user-friendly Likert scale, which may be inferior to a continuous variable for statistical analysis. Patients included in this study presented to UM at variable time points after their initial symptoms. The quality of a patient’s history of ocular infection was also a variable that could not be controlled. We did not distinguish between prior cultures obtained at UM (including emergency department) and outside facilities. The power of this study is limited by the small sample size of MK patients and thus limited our analyses to univariate associations instead of more rigorous investigations of models and interactions. It is possible that not all observable features of corneal ulcers were recorded by the corneal specialists, which may have limited our analysis. Our study only includes data from patients’ initial visit to UM and does not examine confidence levels at any other time point during the course of treatment, which could show improvement over time with response to treatment and laboratory results. Lastly, confidence in diagnosis alone means nothing without accuracy and therefore future studies should examine the relationship between confidence and accuracy.

Despite its limitations, our study highlights the lack of confidence in diagnosis the causal organisms of MK on initial presentation. Innovative diagnostic aids, including decision-aid software, may help to address this problem. Decision-aid tools have been studied for several ophthalmological applications, including MK, keratoconus, dry eye syndrome, and Fuchs endothelial dystrophy.^32–35 Specifically for MK, deep learning algorithms have been developed, some outperforming the diagnostic skills of ophthalmologists.^36–38 In addition to potential improved diagnostic accuracy, deep learning algorithms may be used as point-of-care testing, allowing for faster initial treatment decisions prior to laboratory results. Accessible deep learning tools may also improve delivery of care to MK patients in remote locations, areas without cornea specialists, and locales with limited resources.

Our findings suggest that cornea specialists generally lack confidence in clinically determining the causal organism in MK infections. The development of highly effective diagnostic tools will help to increase the confidence and accuracy of diagnoses and, ultimately, improve patient outcomes.

Table 5.

Comparison of infection morphology between cases where cornea specialists were very confident/confident in their diagnosis and those where cornea specialists were not very confident.

	Very Confident/Confident (n=39)		Not Very Confident (n=33)
Continuous Variable	n	Mean (SD), Median	n	Mean (SD), Median	p-value*
Epithelial Defect
Height (mm)	37	3.80 (2.57), 3.40	26	3.09 (2.02), 3.00	0.328
Width (mm)	37	3.55 (2.70), 3.00	26	3.12 (1.85), 2.93	0.769
Longest Axis (mm)	37	4.10 (2.73), 3.40	26	3.61 (1.91), 3.85	0.706
Stromal Infiltrate
Height (mm)	39	3.64 (1.98), 3.40	33	3.47 (1.67), 3.00	0.799
Width (mm)	39	3.57 (2.22), 3.00	33	3.33 (1.70), 3.70	0.991
Longest Axis (mm)	39	4.11 (2.21), 3.50	33	3.92 (1.57), 3.90	0.821
Hypopyon Height (mm)	13	1.38 (0.89), 1.00	6	1.43 (1.12), 1.05	1.000
Depth (%)	37	50.3 (28.6), 50.0	32	38.3 (23.2), 30.0	0.092
Thinning (%)	38	20.0 (21.7), 10.0	33	17.7 (26.2), 10.0	0.346
Categorical Variable	# (%)		# (%)		p-value**
Epithelial Defect
Present	37 (94.9)		26 (78.8)		0.070
Absent	2 (5.1)		7 (21.2)
Approx. Stromal Infiltrate Area
≤5mm2	13 (33.3)		11 (33.3)		1.000
>5mm2	26 (66.7)		22 (66.7)
Central
Yes	33 (86.8)		30 (90.9)		0.716
No	5 (13.2)		3 (9.1)
Hypopyon
Present	14 (35.9)		7 (21.2)		0.172
Absent	25 (64.1)		26 (78.8)
Satellite Lesions
Yes	5 (12.8)		4 (12.1)		1.000
No	34 (87.2)		29 (87.9)

SD, standard deviation; VA, visual acuity; mm, millimeter; approx., approximate

^*2-sample Wilcoxon test;

^**Chi-square or Fisher's exact test (if cell counts <5) for categorical variables

note: percentages are reported on the non-missing sample

Data Availability Statement:

The participants of this study did not give written consent for their data to be shared publicly. Supporting data is, therefore, not publicly available.