Study design and population:
We performed a retrospective chart review using data from the Cincinnati Children’s Hospital Medical Center (CCHMC) pediatric cardiology division. The local institutional review board reviewed and approved this study prior to collecting data. Patients were eligible for inclusion in the dataset if they were referred to pediatric cardiology between January 1, 2017, and December 31, 2019. Referrals were placed by a primary care or subspecialist provider. Eligible patients had to be less than 18 years of age at the time of referral. Patients referred to cardiology within the preceding 3 years were excluded. Patients with missing race/ethnicity, gender, language, health insurance type, and address were also excluded. During the study period, there were no significant changes to the referral process or clinic structure.
CCHMC’s pediatric cardiology division operates both general pediatric cardiology and cardiology subspeciality clinics. The primary service area for CCHMC is the Cincinnati metropolitan area. This includes Cincinnati’s urban core as well as suburbs and rural areas in Southwest Ohio, Northern Kentucky, and Southeast Indiana. CCHMC cardiology clinics are located throughout the primary service area. There are approximately 23,000 patients seen each year in CCHMC cardiology clinics. CCHMC outpatient cardiologists are predominantly White. Referrals can be placed electronically or via fax by a primary care provider or subspecialist provider. Whenever a referral is placed, cardiology staff members call the referred family. If contact is not made on the first call, two additional calls are made over a period of several days. Appointments can be scheduled via these phone calls. For all CCHMC primary care and subspecialty clinics, wait time is measured using the number of days required to find the 3rd next available appointment. Wait times for CCHMC’s general cardiology clinics had a mean 3rd next available appointment of 1.1 days during the study period. However, some of the cardiology subspecialty clinics had wait times that were considerably longer. For example, the time to the 3rd next available appointment was 18.5 days for the hypertension clinic and 44.8 days for the preventative cardiology clinic during the study period.
Key outcome and predictor variables:
The primary outcome variable was an incomplete visit following referral to any of the cardiology clinics. Patients were defined as having an incomplete visit if they were referred but never attended a clinic visit, regardless of whether the visit was scheduled. The secondary outcome was an unscheduled appointment. Patients were defined as having an unscheduled appointment if they were referred but never scheduled an appointment. The outcomes were evaluated up until the time of data collection, giving each patient a minimum of two months to schedule and complete their visit before having their visit considered incomplete. Patients were classified as having an appointment scheduled or not scheduled. Of those that were scheduled, they were further classified according to having complete or incomplete visits. We also classified patients as having a complete or incomplete visit regardless of schedule status (Online Resource 1).
We evaluated the association between a variety of factors and unscheduled/incomplete referral visits. Patient level variables included age, race/ethnicity, gender, language, health insurance type, address, and time from referral to appointment (if one was scheduled). These data were all accessible from within the electronic health record. Age was categorized using the distribution among all of those referred, splitting the sample into quantiles. We combined race and ethnicity into a single variable given the ways these variables are assessed across CCHMC and documented within the electronic health record. Race/ethnicity was categorized as Asian, Black, Hispanic/Latino, and White. Gender was classified as male or female. Language was categorized as English, Spanish, or Other. Health insurance was categorized as Private or Public. The patient’s address was geocoded and located to a specific census tract geography and related socioeconomic data. Specifically, we used a widely available, open-source, validated socioeconomic deprivation index (DI). The DI is calculated from census tract-level median household income, fraction of households below the poverty level, fraction of those 25 years and older with at least a high school degree/GED, fraction with insurance, fraction receiving public assistance, and fraction of housing units that are vacant . The DI ranges from 0 to 1 with higher values representing higher levels of deprivation . Census tracts are smaller statistical subdivisions of a county and provide for a more homogeneous population than zip codes allowing for improved study of socioeconomic determinants of health [24,25]. For our analyses, the DI values of comparison were the national mean of 0.38 (slightly higher than our study median of 0.35) and the median of the 5% of patients in our sample living in the most deprived census tracts. We chose to compare the national mean with the most deprived to evaluate the impact of more extreme levels of deprivation on referral completion. For patients in our study living in a census tract with a DI of 0.38, the median income was $45,000, the poverty level was 15%, and 11% of the population did not have health insurance. For patients in our study living in a census tract with a DI of 0.66 (the most deprived 5%), the median income was $23,000, the poverty level was 48%, and 24% of patients did not have health insurance.
Additional variables included reason for referral and season in which the referral was made. The reason for referral was determined via manual review of the chart by a pediatric cardiologist. We grouped reasons for referral into categories capturing the most common reasons for referral. Categories included abnormal echocardiogram/fetal imaging, abnormal electrocardiogram (EKG), evaluations for cardiomyopathy, chest pain, congenital heart disease, cyanosis, dizziness/syncope, preventative cardiology, exercise intolerance/dyspnea, family history, genetic diagnosis, and palpitations (Figure 1). The genetic diagnosis category included those patients with a known or suspected genetic diagnosis that has increased risk of cardiovascular structural abnormalities or disease. Preventative cardiology included referrals for elevated blood pressures, elevated lipids, and obesity or other metabolic conditions that increase risk of heart disease. The family history category included any patient with concerning family cardiac history including sudden cardiac death, arrythmias, ischemic heart disease, and structural abnormalities. Family history of cardiomyopathy, however, was included in the cardiomyopathy category only. If there were multiple referral reasons, the first documented referral reason was used to categorize the referral. Season for referral was grouped into three-month blocks in which the weather conditions are similar for each month in a given block (i.e. December to February in this region is typically cold with occasional snow/ice precipitation).
Medians with interquartile ranges (IQR) for continuous variables (DI and age), and frequencies for categorical variables were calculated for patient demographic and referral information variables. Bivariate analyses were performed to evaluate differences between patients with complete and incomplete referral visits and patients with scheduled and unscheduled appointments. This was done using the Wilcoxon rank-sum test for continuous variables or chi-square test for categorical variables. Separate multivariable logistic regression models were fitted for incomplete referral clinic visits and unscheduled visits using the lrm function in the rms package in R. Model predictors included age, race/ethnicity, gender, language, health insurance type, DI, days from referral to appointment scheduled, referral reason, and season of referral. We allowed for potential non-linear associations for age at referral and DI via the inclusion of restricted cubic spline terms (four knots placed at the 5th, 35th, 65th, and 95th percentiles). The probability of unscheduled visit or incomplete referral according to predictors were obtained from the model estimates. Interactions were assessed between predictors; given the lack of significance of such interactions, we opted to remove them from subsequent models. Model discrimination was further measured by the concordance index (c-index). A c-index of value 1 reflects perfect discrimination, whereas 0.5 reflects random prediction. P values less than 0.05 were considered statistically significant. All statistical analyses were performed using R (version 3.6.1) .