Association Between Sociodemographic Risk Factors and No-Show Propensity in a Glaucoma Population Before and During Covid-19 Pandemic

Abstract

Purpose:

To identify sociodemographic risk factors for higher likelihood to no-show among glaucoma subjects before and during the Covid-19 pandemic using the no-show propensity factor (NSPF), a novel attendance metric, which improves upon no-show percentages by adjusting for number of visits.

Methods:

We analyzed de-identified demographic, visit attendance, and social risk factor data (social vulnerability index (SVI) and area deprivation index (ADI) scores) of de-identified glaucoma subjects from the Bascom Palmer Glaucoma Repository, computed NSPF, and categorized scores as low, intermediate, or high by the 75th and 90th percentiles for the pre-pandemic and pandemic periods. We identified predictors of NSPF scores using univariable, multivariable, and logistic regression analyses.

Results:

Of 15,342 subjects, 11,474, 2,238, and 1,630 subjects had low, intermediate, and high NSPF scores respectively with no-show rates of 9.5%, 39.2%, and 57.8% respectively. Age (β=−0.039 per decade, p<0.001), Black race (β=0.152, p<0.001), Hispanic ethnicity (β=0.115, p<0.001), Medicaid (β=0.073, p<0.001), Spanish primary language (β=0.076, p<0.001), SVI scores (β=0.047 per 25% increase, p<0.001), ADI ranking (β=0.057 for highest quartile, p<0.001), and baseline moderate (β=−0.046, p<0.001) or severe (β=−0.077, p<0.001) glaucomatous disease were significant predictors of NSPF. Older age (odds ratio (OR) 1.15 per decade, p<0.001), higher overall SVI (OR 1.09 per 25% increase, p<0.001), Medicare/Tricare insurance (OR 1.13, p=0.044), and non-English/Spanish primary language (OR 1.43, p=0.020) were associated with worsening NSPF during the pandemic.

Conclusion:

Younger age, non-white race, Hispanic ethnicity, non-English primary language, Medicaid, milder glaucoma, and residence in vulnerable areas are risk factors for greater propensity to no-show.

Précis

Greater social vulnerability, younger age, non-white race, Hispanic ethnicity, non-English speaking, Medicaid insurance, and milder glaucoma were associated with higher no-show propensity, which worsened during the Covid-19 pandemic among those subjects who were older and more socially vulnerable.

Introduction

Glaucoma is an irreversible optic neuropathy, which can lead to severe visual impairment or blindness.¹ Globally, an estimated 64.3 million people were affected by glaucoma in 2013, with a projected increase to 111.8 million people by 2040.² Likewise, in the United States, the number of people suffering from glaucoma was 2.71 million in 2011 and is projected to increase to 7.32 million by 2050.³ Since most cases of glaucoma are asymptomatic and initially restricted to the peripheral visual field, effective disease management requires frequent monitoring longitudinally to prevent visual impairment.^4–6

Poor adherence to recommended follow-up visits is a major challenge in glaucoma management, given that 41% of patients fail to undergo a follow-up eye examination despite having screened positive for glaucomatous disease.⁷ Further, over 50% of patients with a confirmed glaucoma diagnosis often do not subsequently utilize ophthalmic services.⁷ Previous studies among smaller study cohorts have identified demographic risk factors such as younger age, Hispanic ethnicity, and severe glaucomatous disease to be associated with poor follow-up adherence and higher no-show rates among patients with glaucoma, diabetic retinopathy, and age-related macular degeneration.^8,9

Prior studies evaluating no-show rates have used raw no-show percentages to quantify visit non-compliance.^9–11 However, given the complex nature of visit non-compliance, accounting for the number of scheduled visits and the overall no-show rate in the study population may better characterize no-show propensity. Ignoring the number of scheduled visits could generate misleading comparisons (e.g., a patient with 2 no-shows out of 4 scheduled visits ought not to be weighed equally as a patient with 10 no-shows out of 20 scheduled visits although both have a 50% no-show rate). In this regard, the no-show propensity factor (NSPF) was developed by Hwang et al. and transforms patient visit data into a score representing the patient’s likelihood to no-show.¹² This measure accounts for the total number of visits when calculating the score while considering individual rates in the context of the overall no-show rate. While two patients may have the same no-show percentage, the patient with more scheduled visits would receive a more extreme NSPF score than the patient with fewer visits (e.g., the patient with 10 no-shows out of 20 scheduled visits would have a worse NSPF score than the patient with 2 no-shows out of 4 scheduled visits). Such adjustments make NSPF a more robust metric compared to raw no-show percentages. Previous work using NSPF in the context of primary care visits categorized NSPF scores into low, intermediate, and high no-show propensity using the 75^th and 90^th percentiles.¹²

The purpose of this study was to determine key risk factors for greater no-show propensity among patients with or suspected of glaucoma using NSPF before the Covid-19 pandemic and key risk factors for worsening in NSPF during the Covid-19 pandemic, with an emphasis on geocoded social risk factor data. We hypothesized that patients from minority communities residing in areas of greater social vulnerability would have a higher propensity to no-show, and that they were at greater risk for a worsening in NSPF during the pandemic.

Methods

The University of Miami Institutional Review Board approved the study and waived the requirement for informed consent because of the retrospective nature of the study. All procedures and protocols adhered to the tenets of the Declaration of Helsinki and complied with the Health Insurance Portability and Accountability Act (HIPAA) to maintain patient confidentiality and integrity.

Study Design and Data Source

A retrospective cohort study was conducted using the Bascom Palmer Glaucoma Repository (BPGR). Sourced from Epic systems electronic health records (EHR; Epic Systems, Verona, WI) and Zeiss Forum (Carl Zeiss Meditech, Dublin, CA), the BPGR is a compilation of demographic and clinical data from over 70,000 patients evaluated for glaucoma at the Bascom Palmer Eye Institute (BPEI), thus containing data from patients with or suspected of glaucoma.^13,14 The repository contains data regarding demographics, ophthalmic examinations, procedures, standard automated perimetry (SAP), and optical coherence tomography (OCT). These data represent the racially and ethnically diverse population of South Florida, with large Black and Hispanic communities.^13,15 The SAP and OCT data, procedural data, and EHR ophthalmic examination data were available starting November 1996, January 2002, and April 2014 respectively. The repository contains these data through February 1, 2022.^13,15 Previous studies outline the inclusion and exclusion criteria for the BPGR.^14,16

Patient Selection and Data Collection

The study included adult subjects (age ≥18 years at the time of the baseline visit) that were diagnosed with or suspected of glaucoma using ICD-10 codes (H40.X). Codes used for inclusion in the study are listed in Table 1.^17,18 The first visit reported in the EHR was treated as the baseline visit. Patient demographic data included age, self-reported sex, self-reported race, self-reported ethnicity, and self-reported primary language. The patient’s primary insurance coverage at the time of the baseline visit was used to categorize insurance status into four groups: commercial, Medicaid/ Public Health Trust (PHT), Medicare/ Tricare, and self-pay (i.e., uninsured). The Public Health Trust is a government-funded organization in Miami-Dade county that provides safety net insurance coverage to underserved patients.¹⁵

Table 1.

List of International Classification of Diseases codes used to identify patients with or suspected of glaucoma.

Inclusion ICD Codes
	ICD-9	ICD-10
Glaucoma suspect	365.01, 365.02, 365.03, 365.04, 365.05, 365.06 377.14a V19.11b	H40.00X, H40.01X, H40.02X, H40.03X, H40.04X, H40.05X, H40.06X H47.23Xa Z83.511b
Open angle glaucoma	365.10 365.11 365.12 365.13 365.14 365.15 365.52 365.7X	H40.1X Q15.0
Chronic angle closure glaucoma (excluding acute angle closure)	365.20, 365.21, 365.23, 365.24	H40.22X, H40.23X, H40.24X
Traumatic glaucoma	365.65	H40.3X
Uveitic glaucoma	365.62	H40.4X
Glaucoma due to other disorders	365.63	H40.5X
Drug-induced glaucoma	365.31, 365.32	H40.6X
Episcleral venous pressure-induced glaucoma, hypersecretion glaucoma, malignant glaucoma	365.81, 365.82, 365.83	H40.8X

ICD: International Classification of Diseases

X replaces any combination of numbers

^aGlaucomatous optic disc atrophy; in the absence of other ICD codes for glaucoma.

^bFamily history of glaucoma; in the absence of other ICD codes for glaucoma.

Social Vulnerability Index and Area Deprivation Index Data

Social vulnerability index (SVI) is a consolidated metric that indicates the relative vulnerability of US Census tracts by computing scores for 15 individual social factors, with values ranging from zero (least vulnerable) to one (most vulnerable).^13,19 These scores are broadly classified into 4 themes, namely socioeconomic status (Theme 1), household characteristics (Theme 2), minority status and language (Theme 3), housing type and transportation (Theme 4).^19,20 The area deprivation index (ADI) is an index based on a measure created by the Health Resources and Services Administration, which was further refined and validated by researchers at the University of Wisconsin-Madison.²¹ It ranks neighborhoods by socio-economic disadvantage at a national (1–100) and state level (1–10) based on factors such as income, education, employment, and housing quality.²¹ We used the residential addresses of the subjects, while excluding addresses containing PO boxes, to calculate SVI and ADI scores, as has been completed in previous studies.^13,15 We obtained SVI scores at a census tract level and ADI data at a block group level using geocoded data from the 2014–2018 American Community Survey. As in previous studies, we converted SVI values into percentages, categorized ADI national rankings into quartiles (1–25, 26–50, 51–75, 76–100), and categorized ADI state rankings into tertiles (1–3, 4–7, 8–10).^13,15,22,23 We also grouped SVI percentages into quartiles for subsequent analyses.

Standard Automated Perimetry Data

We only utilized reliable SAP tests, defined as those that had a false positive rate of <15% without any “abnormally high sensitivity” result on the Glaucoma Hemifield Test.^13,24 We defined glaucomatous field loss as an “outside normal limits” result on the Glaucoma Hemifield Test or a pattern standard deviation with a probability less than 5%; all others were categorized as suspects. We used the first reliable visual field test to determine the baseline severity of glaucomatous disease according to the Hodapp-Anderson-Parrish (HAP) criteria.²⁵ For patients with differing disease severities between their two eyes, we used the worse severity to classify the patient.

Visit Attendance Data

We extracted appointment data from 2014 to 2021 from the EHR. The 2014–2019 period was defined as pre-pandemic period. Patients were required to have visit attendance data for at least 2 years to be included. We included visit data from glaucoma as well as comprehensive ophthalmology departments at BPEI sites, as some subjects were followed by comprehensive ophthalmologists or optometrists. No-shows were defined as visits in the EHR that were marked as no-show, late cancelations (cancelations within 24 hours of the scheduled appointment), and visits when the patient left without being evaluated by the clinician. The total number of visits was defined as the total number of scheduled appointments during the study period. We excluded appointments that were canceled over 24 hours prior to scheduled appointment time. These data were then used to calculate NSPF using the following formula¹²:

$$\mathit{\text{NSPF}}=ln\left[\frac{\left(\frac{\left(N_S\times N_R\right)+H_N}{N_S+H_C+H_N}\right)}{N_R}\right]$$

• N_S = NSPF smoothing factor (set to 10, per Hwang et al¹²)

• N_R = Average no-show rate of the study cohort

• H_N = Total number of no-shows

• H_C = Total number of completed visits

Based on NSPF percentiles, patients were classified into three groups: low (<75^th percentile), intermediate (75^th-90^th percentile), and high (≥90^th percentile).¹² Of note, prior sensitivity analyses by Hwang et al demonstrated the robustness of NSPF with different cutoffs.¹² For the pandemic period, NSPF scores were computed and categorized separately using visit data from 2020 and 2021 in an analogous fashion.

Statistical Analysis

We completed both univariable and multivariable linear regression models, in which NSPF score was the dependent variable and was treated as a continuous variable. Stepwise regression using the Akaike information criterion (AIC) was utilized for the multivariable model for variable selection. Continuous variables were centered on their respective means. Positive coefficients indicated an association with a greater propensity to no-show, while negative coefficients indicated an association with a reduced likelihood to no-show. We introduced an interaction term between baseline severity and SVI. We completed separate multivariable models for national and state ADI, as well as each SVI theme, with the inclusion of other predictors. We ensured the absence of collinearity using variance inflation factors given the use of both demographic and geocoded social risk factor data. We assessed residuals to ensure the lack of heteroskedasticity and to confirm normality in residual distribution. We used the Kruskal-Wallis test to compare NSPF categorization among SVI and ADI groups.

Given that the NSPF values were computed independently for the pre-pandemic and pandemic study periods, we compared the change in NSPF categorization from the pre-pandemic to the pandemic period instead of the raw NSPF scores. We assigned values of 0, 1, and 2 to low, intermediate, and high categories, respectively, and we calculated the difference in this value between the two periods: Δ_NSPF = NSPF_pandemic – NSPF_pre-pandemic. For example, a change from “low” during the pre-pandemic period to “high” during the pandemic period for a patient would lead to a value of 2 (Δ_NSPF = NSPF_pandemic – NSPF_pre-pandemic = 2–0 = 2). Δ_NSPF was treated as an ordered factor. We then completed ordinal logistic regression to identify predictors of worsening in NSPF categorization. We completed separate multivariable models for each social risk factor index (i.e., overall SVI, each SVI theme, national ADI, and state ADI).

Results

A total of 15,342 subjects were evaluated, of which 11,474 (74.8%) had low NSPF scores, 2,238 (14.6%) had intermediate NSPF scores, and 1,630 (10.6%) had high NSPF scores during the pre-pandemic period. The average overall absolute no-show rate in the cohort was 19.9%, with mean no-show rates of 9.5%, 39.2%, and 57.8% in the low, intermediate, and high NSPF categories, respectively. Figure 1 shows the distribution of NSPF scores for subjects. NSPF score cutoffs for the 75^th and 90^th percentiles were 0.185 and 0.433, respectively. The mean age of the cohort was 65.2 ± 13.3 years (Table 2). Most subjects were insured by Medicare or Tricare (51.5%), followed by commercial insurance (38.0%). While most subjects (72.3%) resided in areas with an ADI rank higher than the national median, the average overall SVI score of the cohort was 54.9 ± 28.2%. A statistically significant difference was noted in the distribution of self-reported race among the NSPF categories (p<0.001), with Black subjects constituting 18.2% of the low NSPF group but 31.0% in the high NSPF group. Subjects with high NSPF scores tended to be younger than the other groups, with a mean age of 61.7 ± 14.9 years and had a greater proportion of subjects with Hispanic ethnicity (47.7%), a primary language other than English (37.3%), and Medicaid insurance (13.5%). We observed statistically significant differences in overall SVI (p<0.001) and ADI national rank (p<0.001) among the three NSPF categories. The high NSPF category had the greatest mean overall SVI (65.6 ± 26.3%) and the highest proportion of patients residing in areas with an ADI national rank in the fourth quartile (10.8%). Notably, no statistically significant difference was noted among the three groups regarding glaucomatous disease severity (p = 0.094). The most common disease severity at baseline was severe stage (34.7%).

Figure 1.

Distribution of no-show propensity factor (NSPF) scores of subjects. The black, yellow, and red lines represent the 50^th, 75^th, and 90^th percentile thresholds respectively. A higher NSPF score implies a greater propensity to no-show.

Table 2.

Cohort characteristics by no-show propensity factor categorization

Characteristic	Overall, N = 15,342	Low, N = 11,474	Intermediate, N = 2,238	High, N = 1,630	p-value
Age (years)	65.2 ± 13.3 66.5 (57.2, 74.5)	66.1 ± 12.8 67.5 (58.7, 74.9)	62.6 ± 14.2 63.2 (53.7, 73.0)	61.7 ± 14.9 61.8 (52.6, 72.5)	<0.001 1
Self-reported Sex					0.047 2
Female	8,880 (57.9%)	6,576 (57.3%)	1,338 (59.8%)	966 (59.3%)
Male	6,462 (42.1%)	4,898 (42.7%)	900 (40.2%)	664 (40.7%)
Self-reported Race					<0.001 2
White	10,986 (71.6%)	8,549 (74.5%)	1,448 (64.7%)	989 (60.7%)
Black	3,224 (21.0%)	2,089 (18.2%)	629 (28.1%)	506 (31.0%)
Other	1,132 (7.4%)	836 (7.3%)	161 (7.2%)	135 (8.3%)
Self-reported Ethnicity					<0.001 2
Non-Hispanic	9,437 (61.5%)	7,396 (64.5%)	1,189 (53.1%)	852 (52.3%)
Hispanic	5,905 (38.5%)	4,078 (35.5%)	1,049 (46.9%)	778 (47.7%)
Self-reported Primary Language					<0.001 2
English	10,974 (71.5%)	8,538 (74.4%)	1,413 (63.1%)	1,023 (62.8%)
Spanish	4,008 (26.1%)	2,714 (23.7%)	737 (32.9%)	557 (34.2%)
Other	360 (2.3%)	222 (1.9%)	88 (3.9%)	50 (3.1%)
Insurance Status					<0.001 2
Commercial	5,837 (38.0%)	4,302 (37.5%)	907 (40.5%)	628 (38.5%)
Medicaid/PHT	1,194 (7.8%)	711 (6.2%)	263 (11.8%)	220 (13.5%)
Medicare/Tricare	7,898 (51.5%)	6,172 (53.8%)	998 (44.6%)	728 (44.7%)
Self-Pay	413 (2.7%)	289 (2.5%)	70 (3.1%)	54 (3.3%)
SVI: Overall (%)	54.9 ± 28.2 54.9 (31.5, 80.5)	51.8 ± 28.2 50.6 (28.5, 76.3)	63.2 ± 26.3 67.1 (42.5, 86.9)	65.6 ± 26.3 71.4 (45.2, 88.9)	<0.001 1
Socioeconomic Status (%)	49.0 ± 29.0 47.5 (23.6, 74.9)	45.8 ± 28.6 42.8 (21.3, 70.8)	57.0 ± 28.1 60.4 (32.1, 82.1)	59.9 ± 27.9 64.5 (35.7, 84.8)	<0.001 1
Household Composition / Disability (%)	40.4 ± 21.7 38.6 (23.5, 55.4)	39.7 ± 21.3 37.5 (23.2, 54.1)	41.7 ± 22.4 40.7 (23.9, 58.8)	43.8 ± 22.7 43.0 (25.4, 60.8)	<0.001 1
Minority Status / Language (%)	77.7 ± 23.1 85.6 (68.1, 95.3)	74.9 ± 24.3 82.7 (62.7, 94.3)	85.2 ± 17.2 91.6 (80.5, 96.8)	86.6 ± 15.8 92.4 (82.2, 96.9)	<0.001 1
Housing Type / Transportation (%)	52.5 ± 27.5 52.7 (30.6, 75.5)	50.1 ± 27.5 50.0 (27.5, 72.4)	59.2 ± 26.1 62.8 (39.9, 81.1)	60.2 ± 26.0 64.0 (40.5, 81.8)	<0.001 1
State ADI Rank					<0.001 2
1–3	8,210 (53.5%)	6,532 (56.9%)	998 (44.6%)	680 (41.7%)
4–7	5,467 (35.6%)	3,800 (33.1%)	962 (43.0%)	705 (43.3%)
8–10	1,665 (10.9%)	1,142 (10.0%)	278 (12.4%)	245 (15.0%)
National ADI Rank					<0.001 2
1–25	5,716 (37.3%)	4,635 (40.4%)	651 (29.1%)	430 (26.4%)
26–50	5,377 (35.0%)	3,935 (34.3%)	848 (37.9%)	594 (36.4%)
51–75	3,027 (19.7%)	2,064 (18.0%)	533 (23.8%)	430 (26.4%)
76–100	1,222 (8.0%)	840 (7.3%)	206 (9.2%)	176 (10.8%)
Baseline HAP Severity					0.0942
Suspect	4,193 (27.3%)	3,155 (27.5%)	633 (28.3%)	405 (24.8%)
Mild	3,183 (20.7%)	2,367 (20.6%)	486 (21.7%)	330 (20.2%)
Moderate	2,645 (17.2%)	1,966 (17.1%)	383 (17.1%)	296 (18.2%)
Severe	5,321 (34.7%)	3,986 (34.7%)	736 (32.9%)	599 (36.7%)

ADI = Area Deprivation Index; HAP = Hodapp-Anderson-Parrish; PHT = Public Health Trust; SVI = Social Vulnerability Index

¹Kruskal-Wallis rank sum test;

²Pearson’s Chi-squared test

Univariable regression analyses identified several associations between NSPF scores and predictors (Table 3). In multivariable analyses, higher overall SVI scores were associated with higher NSPF scores (i.e., greater no-show propensity; β = 0.047 per 25% increase; p<0.001; Table 4). Among individual SVI themes, socioeconomic status (theme 1; β = 0.048 per 25% increase, p<0.001), minority status and language (theme 3; (β = 0.098 per 25% increase, p<0.001), and housing type and transportation (theme 4; β = 0.048 per 25% increase, p<0.001) were significant predictors of higher NSPF scores. Figure 2 shows the categorization of subjects based on NSPF scores across SVI quartiles. Other predictors that were associated with higher NSPF scores were Black race (β = 0.152; p<0.001), Hispanic ethnicity (β = 0.115; p<0.001), Medicaid insurance (β = 0.073; p<0.001), and Spanish as a primary language (β = 0.076; p<0.001). Older age (β = −0.039 per decade older, p<0.001), and moderate (β = −0.046; p < 0.001) or severe (β = −0.077; p<0.001) glaucomatous disease at baseline were associated with lower NSPF scores (i.e., reduced likelihood to no-show).

Table 3.

Univariable analysis of impact of individual predictors on no-show propensity factor scores

Characteristic	Beta	95% CI	p-value
Age (per decade older)	−0.061	−0.067, −0.055	<0.001
Self-reported Sex
Female	—	—
Male	−0.023	−0.038, −0.007	0.004
Self-reported Race
White	—	—
Black	0.164	0.145, 0.183	<0.001
Other	0.084	0.054, 0.113	<0.001
Self-reported Ethnicity
Non-Hispanic	—	—
Hispanic	0.160	0.144, 0.175	<0.001
Self-reported Primary Language
English	—	—
Spanish	0.165	0.148, 0.182	<0.001
Other	0.214	0.164, 0.264	<0.001
Insurance Status
Commercial	—	—
Medicaid/PHT	0.182	0.152, 0.212	<0.001
Medicare/Tricare	−0.084	−0.100, −0.068	<0.001
Self-Pay	0.024	−0.024, 0.072	0.3
Overall SVI (per 25% higher)	0.103	0.097, 0.110	<0.001
SVI Theme 1 (per 25% higher)	0.095	0.089, 0.102	<0.001
SVI Theme 2 (per 25% higher)	0.032	0.023, 0.040	<0.001
SVI Theme 3 (per 25% higher)	0.152	0.144, 0.160	<0.001
SVI Theme 4 (per 25% higher)	0.082	0.075, 0.089	<0.001
ADI State Ranking Tertile
1–3	—	—
4–7	0.125	0.108, 0.141	<0.001
8–10	0.136	0.111, 0.161	<0.001
ADI National Ranking Quartile
1–25	—	—
26–50	0.086	0.068, 0.104	<0.001
51–75	0.168	0.147, 0.189	<0.001
76–100	0.150	0.121, 0.180	<0.001
Baseline HAP Severity
Suspect	—	—
Mild	−0.013	−0.035, 0.010	0.3
Moderate	−0.046	−0.070, −0.023	<0.001
Severe	−0.062	−0.082, −0.043	<0.001

ADI = Area Deprivation Index; CI = Confidence Interval; HAP = Hodapp-Anderson-Parrish; PHT = Public Health Trust; SVI = Social Vulnerability Index

Table 4.

Impact of overall social vulnerability index and demographic factors on no-show propensity factor scores using a multivariable regression model

Characteristic	Beta	95% CI	p-value
Age (per decade older)	−0.039	−0.046, −0.032	<0.001
Self-reported Race
White	—	—
Black	0.152	0.129, 0.174	<0.001
Other	0.105	0.077, 0.133	<0.001
Self-reported Ethnicity
Non-Hispanic	—	—
Hispanic	0.115	0.094, 0.137	<0.001
Insurance Status
Commercial	—	—
Medicaid/PHT	0.073	0.043, 0.103	<0.001
Medicare/Tricare	0.002	−0.019, 0.022	0.9
Self-Pay	0.037	−0.010, 0.083	0.12
Self-reported Primary Language
English	—	—
Spanish	0.076	0.053, 0.100	<0.001
Other	0.098	0.048, 0.147	<0.001
Overall SVI (per 25% higher)	0.047	0.034, 0.060	<0.001
Baseline HAP Severity
Suspect	—	—
Mild	−0.012	−0.033, 0.010	0.3
Moderate	−0.046	−0.069, −0.023	<0.001
Severe	−0.077	−0.096, −0.057	<0.001
Overall SVI (per 25% higher) * Baseline HAP Severity
Overall SVI (per 25% higher) * Mild	0.002	−0.017, 0.020	0.9
Overall SVI (per 25% higher) * Moderate	0.013	−0.007, 0.033	0.2
Overall SVI (per 25% higher) * Severe	0.025	0.008, 0.042	0.003

ADI = Area Deprivation Index; CI = Confidence Interval; HAP = Hodapp-Anderson-Parrish; PHT = Public Health Trust; SVI = Social Vulnerability Index

Figure 2.

Categorization of subjects based on no-show propensity factor (NSPF) scores across overall social vulnerability index (SVI) quartiles.

Of note, we observed a significant interaction between overall SVI and severe baseline disease (β = 0.025; p = 0.003; Table 4). Figure 3 demonstrates the effect of SVI quartile on the relationship between baseline glaucomatous disease and NSPF scores among glaucoma suspects and subjects with severe glaucoma. Although NSPF scores tended to be lower among subjects with severe disease (i.e., lower no-show propensity), the noted difference between subjects with severe disease and glaucoma suspects became progressively smaller with higher SVI values. In separate multivariable models with individual SVI themes, this interaction term was statistically significant only in minority status/language (theme 3) and housing type/transportation (theme 4) models. A similar multivariable model demonstrated that ADI national ranking in the third quartile (β = 0.051; p<0.001) or fourth quartile (β = 0.057; p<0.001) was associated with higher NSPF scores (Table 5). An interaction between ADI quartile and baseline disease severity was not statistically significant, and thus was not included in the final model. Of note, sensitivity analyses using the 70^th and 85^th percentiles as cutoffs (as previously completed by Hwang et al) did not alter our findings.¹²

Figure 3.

Split violin plots demonstrating the distribution of no-show propensity factor (NSPF) scores across social vulnerability index (SVI) quartiles among glaucoma suspects and subjects with severe glaucoma based on Hodapp-Anderson-Parrish (HAP) criteria at baseline evaluation. The yellow and red lines represent the 75^th and 90^th percentile thresholds respectively, which were used to categorize NSPF scores as low, intermediate, and high.

Table 5.

Impact of area deprivation index national ranking and demographic factors on no-show propensity factor scores using a multivariable regression model

Characteristic	Beta	95% CI	p-value
Age (per decade older)	−0.043	−0.051, −0.036	<0.001
Self-reported Race
White	—	—
Black	0.207	0.185, 0.229	<0.001
Other	0.109	0.080, 0.137	<0.001
Self-reported Ethnicity
Non-Hispanic	—	—
Hispanic	0.141	0.120, 0.163	<0.001
Insurance Status
Commercial	—	—
Medicaid/PHT	0.093	0.063, 0.123	<0.001
Medicare/Tricare	0.009	−0.011, 0.030	0.4
Self-Pay	0.031	−0.016, 0.077	0.2
Self-reported Primary Language
English	—	—
Spanish	0.105	0.082, 0.128	<0.001
Other	0.115	0.065, 0.164	<0.001
ADI National Quartile
1–25	—	—
26–50	0.011	−0.007, 0.028	0.2
51–75	0.051	0.029, 0.074	<0.001
76–100	0.057	0.027, 0.087	<0.001
Baseline HAP Severity
Suspect	—	—
Mild	−0.012	−0.034, 0.009	0.3
Moderate	−0.045	−0.068, −0.023	<0.001
Severe	−0.071	−0.091, −0.052	<0.001

ADI = Area Deprivation Index; CI = Confidence Interval; HAP = Hodapp-Anderson-Parrish; PHT = Public Health Trust; SVI = Social Vulnerability Index

When evaluating the pandemic period (data available from 16,295 subjects), the average overall absolute no-show rate was 17.9%, with mean no-show rates of 3.4%, 40.2%, and 76.6% in the low, intermediate, and high NSPF categories, respectively. NSPF score cutoffs for the 75^th and 90^th percentiles during the pandemic period were 0.179 and 0.348 respectively. When comparing the change in NSPF categorization between the pre-pandemic and pandemic periods, 9,150 subjects had data during both periods. We observed that older age (OR 1.15 per decade, p <0.001), higher overall SVI (OR 1.09 per 25% increase, p<0.001), Medicare/Tricare insurance (OR 1.13, p = 0.044), and primary language other than English or Spanish (OR 1.43, p = 0.020) were all associated with a worsening in NSPF categorization (Table 6). Similar statistically significant patterns were observed for individual SVI themes, namely socioeconomic status (theme 1; OR 1.10 per 25% increase, p<0.001), minority status and language (theme 3; OR 1.06 per 25% increase, p = 0.042), and housing type and transportation (theme 4; OR 1.05 per 25% increase, p = 0.013), as well as ADI national quartile (OR 1.21 for highest quartile, p = 0.036) and ADI state tertile (OR 1.18 for highest tertile, p = 0.037).

Table 6.

Impact of social and demographic predictors on changes in no-show propensity factor categorization before and during the Covid-19 pandemic

Characteristic	OR	95% CI	p-value
Age (per decade older)	1.15	1.09, 1.20	<0.001
Self-reported Sex
Female	—	—
Male	1.02	0.93, 1.11	0.743
Self-reported Race
White	—	—
Black	1.03	0.90, 1.18	0.654
Other	1.15	0.95, 1.40	0.148
Self-reported Ethnicity
Non-Hispanic	—	—
Hispanic	1.05	0.92, 1.19	0.519
Insurance Status
Commercial	—	—
Medicaid/PHT	0.94	0.77, 1.15	0.549
Medicare/Tricare	1.13	1.00, 1.27	0.044
Self-Pay	1.34	1.00, 1.79	0.052
Self-reported Primary Language
English	—	—
Spanish	1.09	0.94, 1.27	0.235
Other	1.43	1.06, 1.93	0.020
Overall SVI (per 25% higher)	1.09	1.04, 1.14	<0.001
Baseline HAP Severity
Suspect	—	—
Mild	1.00	0.88, 1.14	0.975
Moderate	0.92	0.80, 1.05	0.199
Severe	0.98	0.87, 1.10	0.713

ADI = Area Deprivation Index; CI = Confidence Interval; HAP = Hodapp-Anderson-Parrish; OR = Odds Ratio; PHT = Public Health Trust; SVI = Social Vulnerability Index

Discussion

This study of over 15,000 diverse glaucoma subjects demonstrates the association between key demographic, social, and ocular risk factors obtained from the EHR and the propensity to no-show to glaucoma appointments, highlighting the value of geocoded data in identifying patients at high risk to no-show. Younger age, Black race, Hispanic ethnicity, Medicaid insurance, a primary language other than English, milder disease, higher overall SVI scores, and worse ADI national ranking were all associated with a greater likelihood to no-show. The study also identified risk factors for worsened no-show propensity during the Covid-19 pandemic, providing insight into which patient groups may be at greatest risk for no-shows during future public health emergencies. Older age, Medicare or Tricare insurance, higher overall SVI scores, higher ADI rankings, and a primary language other than English and Spanish were associated with a worsening in NSPF categorization during the pandemic period. This study is novel in its use of the NSPF metric to quantify patient no-show propensity for eye care visits and in its comparison of changes in no-show propensity before and during the Covid-19 pandemic. The use of both SVI and ADI allows for a holistic analysis of social vulnerability and deprivation; SVI accounts for socio-economic and socio-demographic themes while ADI focuses on socio-economic deprivation. These findings highlight health inequities among vulnerable patient groups, identified by both demographic and social risk factors collated from EHR data.

Among SVI themes, higher values for themes 1 (socioeconomic status), 3 (minority status), and 4 (housing type/transportation) were predictors of higher NSPF scores. Subjects with high theme 1 scores may have limited financial resources for healthcare.¹⁹ Individuals with higher SVI scores are also likely to have more systemic co-morbidities, which may take precedence over an ophthalmic condition.^26,27 They may also avoid multiple healthcare visits to avoid losing wages.²⁸ Racial and ethnic minorities often report suboptimal patient-physician communication, which may contribute to poor health literacy.²⁹ Scanzera et al. studied the barriers and facilitators to visit adherence to ophthalmology appointments, and found lack of transportation to be the most significant factor when interviewing patients with SVI scores >0.61.³⁰ Restricted vehicular access, possibly exacerbated by limitations in Florida’s public transit as well as changes in residence, may explain greater no-show propensity among subjects with high SVI theme 4 scores.¹⁹ Likewise, higher NSPF scores among younger subjects might be attributed to workplace and parental commitments in addition to likely milder disease. Among Medicaid patients, socioeconomic challenges may explain higher NSPF scores.

As a tertiary care eye center, BPEI cares for patients with advanced disease; over 50% of subjects had moderate or severe disease at baseline (Table 2). Interestingly, subjects with moderate and severe baseline glaucomatous disease had lower propensities to no-show. This reduced likelihood may be attributed to a higher perceived need to attend visits due to significant functional impairment, compared to often asymptomatic subjects with mild disease. Notably, the interaction between overall SVI scores and glaucomatous disease suggested that while subjects with more severe disease tend to no-show less frequently, those subjects with advanced disease from highly socially vulnerable areas remained more likely to no-show. Figure 3 highlights the impact of SVI quartile when comparing glaucoma suspects with subjects with severe glaucoma. These findings have implications for public health strategies such as glaucoma screening; while subjects with severe glaucoma were less likely to no-show overall, those advanced disease subjects residing in areas with higher SVI still had a greater propensity to no-show. Community screening should not only target higher SVI communities, but should also emphasize the importance of follow-up for patients with advanced disease in such high-risk communities. These findings also highlight that it is essential to emphasize the importance of follow-up not only to patients with advanced glaucoma but also to those with mild disease.

Our findings regarding disease severity differed from those of prior studies, which observed an association between more severe disease and worse follow-up adherence.^8,31,32 Ung et al. studied the impact of disease status on follow-up adherence among approximately 200 patients cared for at a safety-net hospital.⁸ The authors measured follow-up adherence in a binary fashion (adherent/non-adherent) using American Academy of Ophthalmology (AAO) Preferred Practice Pattern guidelines as well as personalized recommendations. The limited study cohort was also focused on indigent and uninsured communities per the authors. Recent work by Wasser et al. and Williams et al. evaluated factors associated with loss to follow-up (LTFU) in the AAO Intelligent Research in Sight (IRIS) database among primary open-angle glaucoma patients, and identified that patients with severe disease were more likely to experience LTFU.^31,32 Disparate findings between our work and these studies are likely due to differences in the measured outcome and experimental design. Wasser et al. and Williams et al. measured the prevalence of LTFU, defined as the percentage of POAG patients who went a full calendar year without a documented encounter in the IRIS registry between their index visit in 2014 and 2019.^31,32 Conversely, in our study, NSPF was a continuous outcome variable, which indicated the likelihood to no-show to scheduled appointments without commenting on adherence to a recommended interval. Thus, a patient could be adherent to the recommended follow-up schedule yet have a high propensity to no-show. NSPF provides greater granularity compared to other common metrics like the prevalence of LTFU, while accounting for longitudinal patient attendance. Although sample size is one of the key strengths of the IRIS database, the authors had to use ICD-9 coding to classify disease severity; notably, almost 90% of eyes were categorized as “unspecified” stage and only 3% were classified as severe. In contrast, our study utilized perimetry data to stage disease severity per a well-established classification system. Advanced disease patients in the IRIS analyses may have been referred by physician offices to academic medical centers, of which only 33% are represented in the IRIS database.³² These patients would have been misclassified as LTFU if they had actually transitioned their care to these academic medical centers for further management. In contrast, subjects with severe glaucoma constituted the largest percentage of our study cohort, approximately 35% (Table 2). Given these differences in study design and cohort, we believe that our findings regarding disease severity are complementary to those of prior works.

A comparison of NSPF categorization before and during the pandemic showed that older age, Medicare or Tricare insurance, and higher SVI were statistically significant risk factors for worsening NSPF during the pandemic. Older subjects, who are typically insured by Medicare or Tricare, had higher rates of morbidity and mortality due to Covid-19, likely explaining the greater no-show propensity.³³ The exacerbation of social vulnerabilities, particularly among disadvantaged groups, and higher infection rates among socially vulnerable populations during the pandemic might have increased their no-show propensity.^34–36 This finding is consistent with the strong association between SVI socioeconomic status (theme 1) and a worsening in NSPF categorization during the pandemic; these subjects may have struggled with employment during the pandemic, and may have had to prioritize other needs over healthcare due to limited resources. These findings highlight how social factors were key contributors to a worsening in no-show propensity, independent of insurance status and race or ethnic status. Communication regarding appointments during the Covid-19 pandemic may have been insufficient for non-English and non-Spanish speaking patients (e.g., Haitian Creole patients), leading to the observed worsening of NSPF categorization among subjects with a primary language other than English or Spanish.

Limitations of this study include its retrospective design and dependence on EHR data, which may contain inconsistencies or missing information. Race, ethnicity, and primary language were self-reported, making these data vulnerable to inconsistencies while also ignoring differences among subgroups.³⁷ While it would have been interesting to evaluate differences in NSPF categorization by glaucoma type, we were concerned about data accuracy given potential ICD code misclassification in the EHR data and thus did not include this variable as a predictor in our models. As previously noted, our study evaluated no-show propensity; we did not evaluate how closely a patient adhered to individual follow-up recommendations or national recommendations. This nuance poses a challenge in interpreting EHR data for strategic interventions to reduce no-shows. Future studies may evaluate visit adherence by comparing the provider’s recommended timeframe for follow-up with the patient’s actual time to the next encounter, as completed for diabetic retinopathy patients.³⁸ Incorporating an algorithm that detects such deviation from the provider’s recommendations would likely improve long-term care of patients with glaucoma and other chronic eye diseases. Another limitation is that insurance status was determined at baseline and did not account for subsequent change, which could have influenced no-show propensity. It is well-established that insurance is a key determinant of healthcare utilization.³⁹ We did not evaluate if a patient rescheduled a visit after a late cancelation, making it difficult to comment on true visit adherence. The study analyzed visit data from the sites of a single academic center, which may limit the generalizability of these findings. Future work may focus on analyzing data from multiple ophthalmic centers. In addition, given the relative ease in accessing these data in the EHR, decision support systems could identify high-risk patients in the future. Simple interventions such as sending EHR-generated text messages to patients after a missed appointment may be used to encourage re-engagement. Targeted interventions, such as social work consults, could reduce no-show propensity among these individuals.^40–42

Overall, this study demonstrates the association between no-show propensity and demographic factors, greater social vulnerability, and greater area deprivation among glaucoma subjects. Geocoded information from the EHR can easily provide composite social risk factor data to potentially risk-stratify patients. Notably, younger age, Black race, Hispanic ethnicity, non-English as a primary language, and Medicaid/ PHT insurance increased a patient’s likelihood to no-show while more advanced disease decreased the likelihood to no-show. During the pandemic, older age, primary language other than English or Spanish, Medicare/Tricare insurance, and higher social vulnerability were associated with a worsening of no-show propensity. An awareness of these risk factors may help personalize strategies to minimize the likelihood of no-shows, a key component of longitudinal care in this population.