We conducted a retrospective cohort study of pediatric patients enrolled in Kaiser Permanente Southern California (KPSC), an integrated health care organization whose patient population is reflective of the socioeconomic and racial diversity of Southern California.
17 KPSC's electronic health records (EHRs) from 2011 to 2016 were used to identify study-eligible patients.
We focused on children with early onset myopia who were between 4 and 11 years old when they had a refraction measurement between -6 to -1 diopters (Ds). The first measurement where the refractive error was ≤-1 D defined the baseline measurement and all follow-up measurements were included in the analysis. Patients also must have at least one follow-up refraction ≥21 months after the baseline measurement and before the end of 2017. Patients with amblyopia, strabismus, or retinopathy of prematurity were identified through International Classification of Diseases (ICD) codes and excluded from the sample. Patients with strabismus or cataract surgery were identified by Current Procedural Terminology (CPT) codes prior to their first qualifying refraction measurement and were also excluded. Furthermore, patients whose medical records lacked information on gender were excluded from analysis (n = 18).
Patient information on race, ethnicity, and language preferences were abstracted from the KPSC EHR. Patients were surveyed on this information upon enrollment within KPSC and additional details could be added at any time during their care. For children under the age of 12 years old, parents were asked for this information. Patients older than 12 years old were asked to self-report this information. Patients born at KSPC had their maternal race and ethnicity used for identification purposes unless otherwise specified. For race, patients could identify as American Indian or Alaska Native, Asian, Black or African American, Hispanic or Latino, Native Hawaiian or Pacific Islander, white, decline to state, other, or unknown. For ethnicity, patients could select from a list of over 250 groups or select “Decline to State,” “Other,” or “Unknown.” For our study, race/ethnicity categories were collapsed to white, Black, Hispanic, South Asian, East/Southeast Asian, other Asian, and other/unknown. Patients were classified as South Asian if the patient self-identified, or—in the case of children under the age of 12 years—were identified by their parent(s), as Afghan, Asian Indian, Bangladeshi, East Indian, Nepalese, Pakistani, or Sri Lankan, or indicated that their written or spoken language was Bengali, Gujarati, Hindi, Malayalam, Panjabi, Pashto, Punjabi, Sinhalese, Urdu, or Urdu Pakistan. Although other languages are spoken in South Asia, the aforementioned languages were the only ones that patients within this cohort identified as using. Patients were classified as East/Southeast Asian if they were identified as a racial/ethnic group related to or had a primary, spoken, or written language pertaining to East/Southeast Asia. The East/Southeast Asian group included the following racial/ethnic groups: Asian/Pacific Islander, Cambodian, Chinese, Filipino, Indonesian, Japanese, Kinh/Viet, Korean, Laotian, Malaysian, Tagalog, Taiwanese, Thai, and Vietnamese. Languages classifying a patient as East/Southeast Asian were the following: Burmese, Chinese, Dzongkha, Hakka, Japanese, Khmer, Korean, Laotian, Mandarin, Philippine, Tagalog, Thai, Toishanese, and Vietnamese. Patients who were identified as Asian race but were missing more specific race-ethnicity information, specified their language as English only, spoke languages not typically associated with South Asian or East/Southeast Asian regions, or lacked information to further classify the Asian group were categorized as other Asian.
Cycloplegic, manifest, final, and wearing refractions were included for analysis. If a patient had more than one refraction on the same day, the measurement was selected in the same order of priority. The eye with the more negative refractive error at baseline was chosen for analysis. Measurement or recording errors were possible and patients with a biologically implausible average yearly refraction change (calculated using the baseline and final measurements of refractive errors) of ≥ 10 D were excluded from analyses.
Covariates of interest included age, sex, race/ethnicity, body mass index (BMI), year of first examination, screen time, physical activity, and outdoor time. Age at baseline was defined as the patient's age at the time of the first refraction measurement. BMI was calculated using height and weight measurements closest to the date of the initial refraction. Screen time, physical activity, and outdoor time were abstracted from the EHR. At well-child visits, patients were asked whether they had < 2 hours of screen time per day, > 1 hour of physical activity per day, and > 2 hours of outdoor time per day. Responses from the visit closest to baseline were abstracted for analyses. Data on outdoor time were only available in 2017.
A growth curve analysis using linear mixed-effects models was used to trace longitudinal progression of spherical equivalents (SEs) over time by age at baseline. As this longitudinal model relies on person-time, this model traces an average trend across observations among patients of the same age or the same time since onset, rather than tracing each individual's trajectory and then averaging those trajectories.
Analyses adjusted for potential confounders or proxies of confounders including BMI z-score percentiles (< 5%, 5–< 85%, 85–< 95%, and 95–100%), screen time (< 2 vs. ≥ 2 hours per day), and physical activity (≥ 1 vs. < 1 hour per day). We used a conditional growth model with refractive error as the outcome to estimate the fixed and random effects of time since baseline measure. These time effects allowed us to trace the trend of myopia progression by age at baseline and across time, conditional on potential confounders. The intrapatient correlation was specified as an autocorrelation structure of order 1. To understand whether the growth trajectory varied with different baseline ages and race/ethnicities, we included a three-way interaction between the time of refractive error measurement, age at baseline measurement, and race/ethnicity. The post hoc tests of pairwise comparisons of the estimated growth trends between race/ethnicity groups were performed using Tukey's method.
18 Patients missing data on screen time, physical activity, or outdoor time were categorized as unknown for these variables and were included in analyses. Analyses were performed using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA) and R (R version 3.4.3).
Institutional Review Board (IRB) approval was obtained. This research also adhered to the tenets of the Declaration of Helsinki.