J-LSMS 2014 | Annual Archive

Journal of the Louisiana State Medical Society

Pediatric developmental screening The PDS survey question is: “Sometimes a child’s doctor or other healthcare provider will ask a parent to fill out a questionnaire at home or during their child’s visit. During the past 12months, did a doctor or other healthcare provider have you fill out a questionnaire about specific concerns or observations you may have about [child]’s development, communication, or social behaviors?” 10 If the response was affirmative and based on the child’s age (10-23, 24-71 months), two questions were asked to de- lineate false positives. If both the initial and two follow-up questions were answered “yes,” then the child is defined to have received PDS. If any of the three questions was answered “no,” the child is defined to have not received PDS ( n =22,270). Medical home The MH is a composite of 21 questions that operation- alizes five model domains: 1) personal doctor or nurse; 2) usual source of care; 3) access to needed referrals; 4) fam- ily-centered care; and 5) care coordination. 10 The MH was dichotomized into a response of “yes” or “no” ( n = 26,694). Covariates CSHCN encompass children with a broad spectrum of health needs defined as meeting one or more of the follow- ing chronic conditions for 12 or more months: (1) currently need prescription medication; (2) need more medical care, mental health, or educational services than a typical child; (3) are limited in the ability to do things; (4) need physical, occupational, or speech therapy; and (5) have any kind of emotional, developmental, or behavioral problem for which he/she needs treatment or counseling. 10 CSHCN status was modeled as “yes” or “no” ( n =27,566). Ethnicity was dichotomized, Hispanic or non-Hispanic (NH) ( n =27,260). Race was grouped as: White, Black, Other, and Multiple races ( n =25,814). Other merged Asian, Native American, Alaskan Native, Native Hawaiian, or Pacific Islander into one group. Household federal poverty level (FPL) is binary: less than or equal to 200% or greater than 200% based on the average Medicaid household income eligibility criteria ( n =25,125). The highest educational level of any household member is binary: < high school or > high school ( n =27,290). Primary health insurance was grouped: public (Medicaid), private (employer or self-insured), or currently uninsured ( n =27,271). Family structure is: two-parent (biologic or adoptive), two-parent (stepparent), single-mother, and other ( n =27,414). The 50 states and DC were categorized as: South, Northwest, Midwest, and West ( n =27,566). Lastly, the child’s age at the time of the survey (10 months to 5 years) ( n =27,566), and sex (male or female) ( n =27,547) were included. Missing values and responses of either “don’t know” or “refused” were coded as missing ( n-missing range: 19 - 2,441).

Statistical analyses SAS-Callable SUDAAN v. 9.1 was used to account for the complex sampling design of NSCH data in order to estimate weighted percentages, odds ratios (OR), and 95% confidence intervals (CI) with proper standard errors so that parameters reflect population estimates. Preliminary descriptive analyses assessing the PDS rate and theMH-PDS association in covariate-adjustedmodels for each state were done to determine if multilevel logistic regression models (MLM) were appropriate for this study. Clusters for this study are states ( n =51). Two-level MLMs were conducted to determine the overall MH-PDS association controlling for covariates. The MLM framework (Figure 1) outlines how level-2 influences the level-1 MH-PDS association. Mplus v.6.0 was used to build three MLMs sequentially tested for goodness-of-fit. The reduced null model did not includeMH and covariates. The random intercept model included MH with covariates and assumed that PDS prevalence varied but the MH-PDS association did not differ by state. The final random intercept and random slope model expanded on the second but assumed the MH-PDS association varied by state. All covariates were assumed to have fixed effects, meaning confounding on the MH-PDS association did not vary by state. Akaike’s and Bayesian’s information criteria and likelihood-ratio tests were used to evaluate model fit. The intra-class correlation coefficient (ICC) measures the percent of variability of the PDS between states. To improve parameter estimate accuracy, the raw sampling weight was scaled by applying scaling method A as suggested by Aspa- rouhov and yields new weights that add up to the cluster sample size. This method is particularly recommended when the cluster size is greater than 20. 12 Because of study inclusion criteria, a zero-weight approachwas applied to the new weight so ineligible children ( n =5,296) were excluded from analysis with a designated zero weight prior to MLM building. As subpopulations are nested above the cluster level, the zero-weight approach is deemed an appropriate method for MLM subpopulation analyses. Alpha was set at 0.05 for statistical significance. RESULTS Our study captured 81% ( n =22,270) of the five years and younger child population. Of this, a fraction (19.5%) received a PDS (Table 1), whereas, almost two-thirds (63%) report a MH. A majority (58.3%) is privately insured, and 13.2% are CSHCN. About one-fifth are Hispanic (22%) and 72.4% white. Educational attainment was equally distrib- uted, as was sex and age (given the inclusion criteria only children less than one year represented a comparatively smaller fraction of the study sample). More than 75% reside in two-parent (biologic/adopted) families, 61.2% live at or above 200% FPL, and nearly 40% are from the South. A comparable proportion of children with or without a MH received a PDS (20.4% vs. 18.4%, respectively) (Table 2). Only significant confounders were included in the adjusted models (age, insurance, race, CSHCN, family structure, and

112 J La State Med Soc VOL 166 May/June 2014

Made with FlippingBook - Online catalogs