Participants of this study were enrolled in an integrated prepaid health plan of Kaiser Permanente Southern California (KPSC) which provided comprehensive health care to over 3.3 million patients as of July 2010. Members received their care in medical offices and hospitals owned by KPSC throughout the seven-county region. All health care providers have access to the KP health record system in which all medical information is stored.
For this cross-sectional study, we used data on children enrolled in the KPSC Children’s Health Study which is described in detail elsewhere [9
]. Briefly, we identified 816,204 children 2 through 19 years of age who were members of KPSC in 2007–2008 and had at least one medical visit during this study period. To eliminate pregnancy-related weight gain, we then excluded pregnant adolescents. After the exclusion of pregnant members (n=6,475), 809,729 patients were eligible for participation in the present cross-sectional study. Of these patients, 710,949 (87.8 % of eligible patients) had at least one valid body weight and height available in the electronic health record in 2007–2008 and were included in the final study cohort. The study protocol was reviewed and approved by the Institutional Review Board of KPSC.
Patients with prevalent psoriasis (n=2,119) were identified by extracting International Classification of Disease (ICD-9) code 696.1 from electronic health records from all inpatient and outpatient encounters since enrollment into the health plan. Details of the chart review process are described elsewhere [17
]. To control for undercoding of psoriasis, additional searches were performed to identify potentially un-coded patients with psoriasis using a comprehensive search for prescription of medications consistent with psoriasis (n=154) and patients with a diagnosis of psoriatic arthritis (ICD-9 code 696.0, n=23). A board-certified dermatologist and trained research staff validated all psoriasis diagnoses by confirming diagnosis codes for psoriasis from physician’s notes in the electronic health record as follows: (1) Cases in which the diagnosis was made by a primary care provider, who were younger than 5 years of age, had a diagnosis of psoriatic arthritis, or were prescribed psoriasis-specific medications, were reviewed by a dermatologist (n=1,082); (2) Potential psoriasis cases identified by ICD-9 code older than 5 years of age and with a diagnosis made at least once by a dermatologist and (n=1,037) were reviewed by trained research staff. Terms used for confirmation included “silvery, flaky, papulosquamous, red, papules, plaques” in the following body locations: “knees, elbows”. Additionally, the description of the location of the rash, family history, and physical findings were used to confirm or reject the diagnosis of psoriasis. Any unclear or questionable records reviewed by research staff were additionally referred to a dermatologist for further review. The diagnosis of 1,350 children was confirmed as prevalent psoriasis. Out of these 1,350 confirmed cases, 90% (n=1,234) were diagnosed by a dermatologist in the electronic medical record.
As an indicator of severe or widespread psoriasis, we identified patients (n = 53) who received any treatment with traditional systemic medications (acitretin, cyclosporine, methotrexate), other less commonly used systemic medications (azathioprine, mycophenolate mofetil, hydroxyurea), biologic medications (TNF-α inhibitors, T-cell inhibitors, or interleukin-12/23 inhibitors) or had received phototherapy (UVB or PUVA) [18
Information from electronic health records was utilized to extract body weight and height data. Only encounters with weight and height measurement from the same day were selected. BMI was calculated as weight (kilograms) divided by the square of the height (meters). For each year 2007 and 2008, the median BMI-for-age of all encounters for a patient from the most recent available year was used for this analysis. Based on a validation study including 15,000 patients with 45,980 medical encounters, the estimated error rate in body weight and height data was <0.4% [19
Definitions for overweight and obesity in children and adolescents are based on the sex-specific BMI-for-age growth charts developed by the CDC and World Health Organization definitions for overweight and obesity in adults [20
]. Children were categorized as underweight (BMI-forage <5th
percentile), normal weight (BMI-for-age ≥ 5th
and < 85th
percentile), overweight (BMI-for-age ≥85th percentile or a BMI ≥25 kg/m2
and BMI-for-age <95th percentile or a BMI <30 kg/m2
), moderately obese (BMI-for age ≥95th percentile or a BMI ≥30 kg/m2
and BMI-for age <1.2 × 95th percentile or a BMI <35 kg/m2
), and extremely obese (BMI-for age ≥1.2 × 95th percentile or a BMI ≥35 kg/m2
Race and ethnicity information was obtained from health plan administrative records and birth certificates. We categorized race/ethnicity as non-Hispanic White, Hispanic White, Black (regardless of ethnicity), Asian or Pacific Islander, other or multiple race/ethnicity, and unknown due to missing information (52.5%). A validation study compared health plan administrative records and birth certificate records of 325,810 children. The positive predictive value (PPV) for Hispanic ethnicity was 95.6%. The PPV for White, Black, Asian/Pacific Islander, American Indian/Alaskan Native, multiple and other was 89.3%, 86.6%, 73.8%, 18.2%, 51.8% and 1.2%, respectively.
For unknown race and ethnicity information, administrative records were supplemented by an imputation algorithm based on surname lists and address information derived from the U.S. Census Bureau [23
]. Hispanic ethnicity and Asian race were assigned based on surnames. For Blacks and non-Hispanic Whites, the child’s home address was used to link racial/ethnic information from the U.S. Census Bureau. Race/ethnicity was hierarchically assigned using probability cut-offs of >50% for Asian surname, >50% for Hispanic surname, >75% for Black race from geocoding if probability for Asian surname was <50% (Hispanic Blacks are assigned to Black race for this study), and White race >45% from geocoding if no other assignment could be made before. The specificity and PPV were >98% for all races/ethnicities [9
We used Medi-Cal status as an indicator for low socioeconomic status. Medi-Cal is the California state-subsidized program providing health care coverage for more than six million low-income children and families as well as elderly, blind, or disabled individuals.
According to KPSC preventive screening guidelines, all children with a BMI-for-age ≥95th percentile or ≥85th percentile with additional potential risk factors such as familial hypertension should undergo screening for dyslipidemia and abnormal liver enzymes. Typically, blood screening is recommended for children pubertal age or older. For the present study, information on total cholesterol, low-density lipoprotein (LDL) cholesterol, high-density lipoprotein (HDL) cholesterol, triglycerides and alanine aminotransferase (ALT) was extracted from electronic health records of children ≥12 years with a BMI-for-age ≥85th percentile (n=133,270).
Each child was assigned to a specific age group (2–5, 6–11, or 12–19 years) according to the child’s mid-year age. Differences in the distribution of basic demographics across groups defined by weight class were assessed with the chi-square test.
Multiple logistic regression models were generated to estimate odds ratios (OR) and their 95% confidence intervals (CI) for psoriasis vs. weight class (underweight, normal weight (reference), overweight, moderate obesity, extreme obesity), sex, age and race/ethnicity. Logistic regression models, stratified by sex, race/ethnicity, and age group (2–5 years, 6–11 years, and 12–19 years) were used to test the association between psoriasis and weight class, adjusted for medical center (low vs. high number of patients with psoriasis) and Medi-Cal status (yes/no), as well as sex, age, and race/ethnicity where appropriate. Tests of linear trend across weight categories were conducted by modeling the exposure as a single continuous variable in the multivariate model, the coefficient for which was evaluated using a Wald test. The association between psoriasis and continuous cardiovascular risk factors was assessed with multiple linear regression models adjusted for age, sex, race/ethnicity, medical center, Medi-Cal status and BMI.
Among overweight and obese adolescents who met the criteria for preventive screening for cardiovascular risk factors (n=133,270), a substantial proportion of patients did not undergo screening and, thus, had missing data for total cholesterol (n= 73,884; 55.4%), LDL cholesterol (n= 89,519; 67.1%), HDL cholesterol (n= 84,685; 63.5%), triglycerides (n= 97,091; 72.9%) and ALT (n= 92,455; 69.4%). We compared adolescents with missing values for any of the above metabolic traits with those adolescents who did have data for differences in the distribution of demographic factors and/or proportion affected with psoriasis, compared with those with complete data with chi-square tests. Adolescents missing one or more lipid or ALT values were more likely to be male, white, or unknown race/ethnicity; they were also less likely to receive Medi-Cal covered services or to have psoriasis.
Because the analysis of only complete data may introduce bias in the estimation of parameters and reduce power to detect significant effects, we considered multiple imputation as an alternative method that allows individuals with incomplete data to be included in analyses [24
]. Briefly, multiple imputations involve the replacement of missing values with values that have been sampled from their predictive distribution based on the observed data. Several values per observation are sampled, resulting in several simulated data sets. Standard regression analyses are then performed on each data set and parameter estimates combined to yield one final estimate, with standard errors calculated according to Rubin’s method, which takes into account the variability between data sets [24
]. Thus, we performed multiple imputation of missing values using the Markov Chain Monte Carlo method, with posterior mode computed by Expectation-Maximization (EM) algorithm [25
]. The means and standard deviations from available observations were used as the initial estimates for the EM algorithm. The association between psoriasis and continuous clinical measures was then assessed with multivariable linear regression models on existing and imputed data, adjusted for age, sex, race/ethnicity, Medi-Cal status and BMI. All analyses were conducted using SAS version 9.1 (Proc MI and PROC MIANALYZE, SAS Institute, Inc, Cary, NC) and PASW release 17.0 (SPSS Inc., Chicago, IL).