Results of this study have been presented previously as an abstract at the annual conference of the Association for Research in Vision and Ophthalmology, Seattle, Washington, USA May 2013.
To estimate the prevalence of, and factors associated with, dilated eye examination guideline compliance among patients with diabetes mellitus (DM), but without diabetic retinopathy.
Utilizing the computerized billing records database, we identified patients with International Classification of Diseases (ICD)-9 diagnoses of DM, but without any ocular diagnoses. The available medical records of patients in 2007–2008 were reviewed for demographic and ocular information, including visits through 2010 (n=200). Patients were considered guideline compliant if they returned at least every 15 months for screening. Participant street addresses were assigned latitude and longitude coordinates to assess their neighborhood socioeconomic status (using the 2000 US census data), distance to the screening facility, and public transportation access. Patients not compliant, based on the medical record review, were contacted by phone or mail and asked to complete a follow-up survey to determine if screening took place at other locations.
The overall screening compliance rate was 31%. Patient sociodemographic characteristics, insurance status, and neighborhood socioeconomic measures were not significantly associated with compliance. However, in separate multivariable logistic regression models, those living eight or more miles from the screening facility were significantly less likely to be compliant relative to those living within eight miles (OR=0.36 (95% CI 0.14 to 0.86)), while public transit access quality was positively associated with screening compliance (1.34 (1.07 to 1.68)).
Less than one-third of patients returned for diabetic retinopathy screening at least every 15 months, with transportation challenges associated with noncompliance. Our results suggest that reducing transportation barriers or utilizing community-based screening strategies may improve compliance.
Dilated eye examination guideline compliance among patients with diabetes mellitus is low.
Quality of access to public transportation was positively correlated with screening compliance.
Transportation assistance interventions and more comprehensive community-based screening models need to be developed and tested.
The number of American adults 20 years of age and older living with diabetes mellitus (DM) has increased by approximately 75% over the past two decades, with the largest absolute increases in prevalence occurring in those 65 years of age and older.
Early detection of diabetic retinopathy is critical, given that prompt treatment increases the likelihood of preserving vision.
Unfortunately, not everyone who should do so receives annual dilated eye examinations. Factors which influence screening compliance include lack of insurance and healthcare access, knowledge of diabetes-specific ocular risk, and health literacy, cultural, and language barriers.
While it is important to evaluate adherence for all persons with diabetes, regardless of whether retinopathy has already been diagnosed, there is scant literature about screening adherence in persons with diabetes who do not have a retinopathy diagnosis. We identified factors associated with compliance for annual dilated eye examination guidelines among patients with diagnosed DM, but without diabetic retinopathy. Given that geographic access to healthcare and screening facilities is critically important for compliance, poor access (such as longer traveling time, high transportation cost, and burdensome public transportation options) can be a potent barrier.
Utilizing computerized billing records, we selected the Bascom Palmer Eye Institute patients initially seen with International Classification of Disease (ICD)-9 diagnoses of DM (types 1 and 2) without diabetic complications and without diabetic retinopathy or any other eye disease. Since we were interested in examining transit-related compliance factors, we further restricted our sample to patients who resided within the same county as the screening facility (ie, Miami-Dade County). The sample of available and eligible patient records first seen in 2007–2008 (n=203) was reviewed for demographic information (eg, age, sex, and race/ethnicity) at the screening visit, and all clinic visits through 2010 were ascertained by chart review. Insurance status was obtained from medical billing records based on current information available as of June 2, 2011. On review, three patients had diabetic retinopathy diagnoses recorded in their medical record and were eliminated, resulting in a final sample size of 200 patients.
Although compliance guidelines for dilated eye examinations specify annual visits, we defined compliance in our sample as returning at least every 15 months (coded as 0=noncompliant, 1=complaint). We chose 15 months to account for annual scheduling variations, such as patient/provider-requested scheduling issues or being away from the area as a consequence of travel or other reasons.
To determine if they received care at a location other than the Bascom Palmer Eye Institute, we attempted to call those patients whose medical records indicated that they were not guideline compliant. A minimum of five attempts were made. Patients who were unreachable by telephone were mailed a questionnaire sent to their last known address. A total of 45 and 8 patients, respectively, completed the telephone interviews or returned the mailed questionnaires.
Each participant's last known address was geocoded to abstract sociodemographic neighborhood characteristics using publicly available block and tract-level data from the 2000 US Census. While demographic data (such as race/ethnicity and household information) were available at the block level, economic data were only available at the census tract level. Generally, a block is small in area; for example, a block in a city is bounded on all sides by streets. Blocks are nested within block groups and block groups within census tracts. The population size of a census tract ranges from 1200 to 4000 people.
Bascom Palmer Eye Institute, the eye care facility, is located near downtown Miami, which is connected to most public transport facilities (metro rail and bus services). We first calculated the Euclidean distance (in miles) between the eye care facility and residential locations of participants. This point-to-point calculation is a crude indicator of the level of geographic access to the eye care facility since it does not take into account roadway patterns and the availability of public transportation to the site. We therefore extracted data from the Walk Score website using web-mining techniques (
Comparisons of patient-level factors and US census-derived estimates of neighborhood sociodemographic status as a function of compliance status were made at the univariate level using the IBM/SPSS statistical package (IBM SPSS Statistics V.21). The means of age and socioeconomic variables in compliant versus noncompliant participants were compared using analysis of variance. The statistical significance of associations between sex, race/ethnicity, and health insurance status was assessed with the χ2 test.
Logistic regression modeling was employed to examine the association between transit scores and select individual-based and area-based measures with adjustment for the spatial trend using the logit function with a robust SE option in Stata (STATA/SE V.10.1). We first undertook multivariable analyses to determine if the Euclidean distance between the eye care facility and the patients’ home was associated with compliance. In this analysis, we partitioned distance using the average distance into two categories: less than the average distance coded as 0, and the rest coded as 1 (<8 vs ≥8 miles). In separate logistic regression modeling, we examined the association between transit score and compliance. Given that there was evidence of spatial trends in these data (ie, presence of similarities in geographic distribution of compliance and noncompliance), a distance-weighted autocovariate was generated in R using the spdep library which was included as a covariate in the regression models.
For both sets of analyses, we employed a stepwise approach to demonstrate how compliance varies with transit scores with and without the control for individual-based and area-based measures and spatial trend. We first modeled the association of transit score with compliance independent of any other variables (model 1), followed by a model which controlled for spatial autocorrelation (model 2). The next model included spatial autocorrelation and patient-level factors including age, sex, and insurance status (model 3). The final model included these variables along with two representative census-based neighborhood sociodemographic indicators, namely the percentage of neighborhood white race designation and median household income. Results were considered significant if p<0.05.
Characteristics of the study population are provided in
Sociodemographic characteristics of the study sample (N=200)
| Characteristic | n | Mean/ per cent | SE |
|---|---|---|---|
| Individual level | |||
| Screening age | 198 | 50.9 | 1.2 |
| Female | 107 | 53.5% | |
| Race-ethnicity | |||
| Black, non-Hispanic | 45 | 22.5% | |
| White, non-Hispanic | 35 | 17.5% | |
| Hispanic | 86 | 43.0% | |
| Other/unknown | 34 | 17.0% | |
| Insurance status | |||
| Uninsured (self-pay) | 18 | 9.8% | |
| Public insurance | 85 | 46.4% | |
| Private insurance | 80 | 43.7% | |
| Area-based characteristics | |||
| Average household age (years) | 195 | 38.5 | 9.4 |
| Percentage of white population | 195 | 62.1 | 2.1 |
| Percentage of married couples | 195 | 64.1 | 0.8 |
| Median household income ($) | 195 | 34 984 | 780 |
| Percentage of families receiving public assistance | 195 | 7.4 | 0.3 |
| Median real estate taxes ($) | 195 | 1828 | 62 |
| Percentage of owner occupied houses | 195 | 56.3 | 1.2 |
Incorporating results from the telephone and mailing outreach efforts yielded an estimated compliance rate of 31% (62/200). Of the 200 patients, 12.5% (n=25) were judged to be guideline compliant based on a medical record review. Of the 175 patients whose medical records indicated that they were not guideline compliant, 30% (n=53) completed a telephone interview (n=45) or returned a mailed questionnaire (n=8). Seventy percent (37/53) of these patients indicated that they had received follow-up care every 15 months from other providers and were reclassified as compliant. Since we were unable to reach 122 of the potentially noncompliant patients, we compared the patient-level and area-based census characteristics of those who did and did not complete either a telephone or mail survey and found no statistically significant differences between groups (results not shown).
Neither patient characteristics nor neighborhood socioeconomic measures were significantly associated with compliance (
Sociodemographic characteristics among study participants compliant and noncompliant with screening guidelines
| Characteristic | Not fully compliant (n=138) | Fully compliant (n=62) | p Value | ||
|---|---|---|---|---|---|
| Mean/per cent | SE | Mean/per cent | SE | ||
| Individual level | |||||
| Age | 52.0 | 1.4 | 48.4 | 2.4 | 0.17 |
| Sex | 1.00 | ||||
| Male (n=92) | 68.5% | 31.5% | |||
| Female (n=107) | 69.2% | 30.8% | |||
| Race-ethnicity | 0.47 | ||||
| Black, non-Hispanic (n=45) | 75.6% | 24.4% | |||
| White, non-Hispanic (n=35) | 68.6% | 31.4% | |||
| Hispanic (n=86) | 65.1% | 34.9% | |||
| Insurance status | 0.08 | ||||
| Uninsured (self-pay) (n=18) | 77.8% | 22.2% | |||
| Public insurance (n=85) | 60.0% | 40.0% | |||
| Private insurance (n=80) | 75.0% | 25.0% | |||
| Area-based characteristics | |||||
| Average household age (years) | 38.3 | 0.8 | 39.2 | 1.2 | 0.54 |
| Percentage of white population | 60.9 | 1.4 | 55.5 | 2.3 | 0.40 |
| Percentage of married couples | 63.9 | 1.0 | 64.4 | 1.5 | 0.78 |
| Median household income ($) | 34 867 | 926 | 35 247 | 1455 | 0.82 |
| Percentage of families receiving public assistance | 7.5 | 0.3 | 7.4 | 0.5 | 0.94 |
| Median real estate taxes ($) | 1803 | 77 | 1885 | 106 | 0.55 |
| Percentage of owner occupied houses | 56.7 | 1.4 | 55.5 | 2.3 | 0.64 |
Distance to screening facility and odds of compliance following adjustment for spatial, individual-level, and neighborhood-level characteristics
| Model 1 | Model 2 | Model 3 | Model 4 | |||||
|---|---|---|---|---|---|---|---|---|
| OR | 95% CI | OR | 95% CI | OR | 95% CI | OR | 95% CI | |
| Distance to facility (<8 miles=reference vs ≥8 miles) | 0.64 | 0.35 to 1.18 | 0.55 | 0.29 to 1.06 | 0.52 | 0.27 to 1.01 | ||
| Age (<45 years (reference) vs 45+) | 0.63 | 0.32 to 1.26 | 0.67 | 0.33 to 1.37 | ||||
| Sex (male=reference vs female) | 0.82 | 0.43 to 1.53 | 0.86 | 0.45 to 1.62 | ||||
| Insurance (yes=reference vs no) | 0.61 | 0.26 to 1.42 | 0.70 | 0.30 to 1.66 | ||||
| Percentage of white population in the neighborhood | 1.00 | 0.99 to 1.02 | ||||||
| Median household income in the neighborhood | 1.00 | 1.00 to 1.00 | ||||||
*p<0.05.
Quality of access to public transportation and odds of compliance following adjustment for spatial, individual-level, and neighborhood-level characteristics
| Model 1 | Model 2 | Model 3 | Model 4 | |||||
|---|---|---|---|---|---|---|---|---|
| OR | 95% CI | OR | 95% CI | OR | 95% CI | OR | 95% CI | |
| Transit score | 1.19 | 1.00 to 1.43 | ||||||
| Age (<45 years (reference) vs 45+) | 0.68 | 0.34 to 1.37 | 0.71 | 0.34 to 1.50 | ||||
| Sex (male=reference) | 0.90 | 0.48 to 1.70 | 0.99 | 0.52 to 1.88 | ||||
| Insurance (yes=reference vs no) | 0.62 | 0.25 to 1.47 | 0.72 | 0.29 to 1.76 | ||||
| Percentage of white population in the neighborhood | 1.01 | 0.99 to 1.02 | ||||||
| Median household income in the neighborhood | 1.00 | 0.99 to 1.00 | ||||||
*p<0.05.
We found that all examined patient-level and neighborhood-level census measures failed to distinguish between patients with diabetes diagnosis who did and did not have dilated eye examinations at least every 15 months during the 24–36-month study period. However, after adjusting for demographics, those living more than eight miles from the eye care facility were less likely to be compliant. We also found that the quality of access to public transportation was strongly associated with compliance. We do not believe that an association between the quality of public transportation access and dilated eye examination compliance has ever been reported in the scientific literature. Both distance to the eye care facility and transit score are highly correlated (r=0.72, p<0.001), which precluded modeling both simultaneously. However, of the two measures, we believe that the findings for transit score, which captures the ease with which a person can return for repeat eye examinations, especially individuals who do not have motor vehicle access, have the most implications for improving compliance in urban settings. This is especially relevant for interpretation of results since the eye care facility is contracted to provide ophthalmic care for county residents lacking healthcare access.
Our findings suggest that transportation access is one barrier that could be addressed through interventions designed to lower such barriers via travel vouchers and arranging for transportation to eye care facilities. However, even the provision of free transportation may not mitigate these low compliance rates, especially in economically distressed communities. For example, in one comprehensive community-based eye disease screening program, those who tested positive and needed follow-up care were offered free transportation to the clinic site. Despite this offer, only about 50% of those who agreed to the follow-up examination completed the visit.
Thus, unless they are paired with interventions designed to increase compliance with eye care guidelines, interventions focused solely on lowering transportation barriers may not substantially improve compliance rates. A randomized trial testing an educational intervention targeting African-Americans with diabetes but with no dilated eye examination in the previous 14 months yielded a significantly higher subsequent examination rate in those in the intervention versus usual care arms of the trial (55% vs 27%).
Alternately, results of the present analysis suggest that dispersed eye care access throughout communities could be more effective in increasing dilated eye examination compliance rates.
Our study was limited by the incomplete follow-up of those judged to be noncompliant based on a medical record review. We had only limited success in reaching these patients by phone or mail. Therefore, our compliance rate of 31% is most likely an underestimate of the true rate. This misclassification may have also influenced our ability to accurately identify predictors of compliance. Our relatively small sample size, combined with lack of information in the medical record, prevented us from examining the role of language barriers in compliance rates. Finally, this study took place in an urban setting, so results may not generalize to those living in suburban and rural settings.
Less than one-third of patients in our study were compliant with dilated eye examination guidelines. Compliance was associated only with living within eight miles from the eye care facility and more strongly with quality of access to public transportation in the urban setting in which this study took place. Study results reinforce the notion that the current environment for routinely meeting dilated eye examination guidelines is far from adequate, and new models which seamlessly embed eye care and educational opportunities within communities, with robust mechanisms for follow-up care for those who test positive, need to be developed and tested.