ABC | Volume 111, Nº4, Octuber 2018

Original Article Silva et al Predictors of family enrollment Arq Bras Cardiol. 2018; 111(4):578-584 in ClinVar, Human Genome Mutation Database (HGMD), British Heart Foundation and Jojo Genetics databases. Functional impact prediction was performed with SIFT, PROVEAN and PolyPhen-2 and mutations without a previous description should be indicated as damaging in at least two algorithms to be considered as potentially pathogenic. Individuals with negative results were also screened for large insertions and deletions via MLPA (MRC-Holland). Point mutations found in ICs were screened in relatives through Sanger sequencing, and large insertions/deletions via MLPA. Statistical analysis The response variable of this study consisted in the number of family members enrolled in the program by each family, starting from a positive IC. The response variable consists of count data, which would suggest the application of a Poisson model. However, as the dependent variable variance was higher than the mean value, the most adequate model in this situation was the negative binomial model, due to data overdispersion (Figure 1). Predictive variables were based on the IC’s clinical and socioeconomic characteristics. We initially performed a distribution analysis on the response variable and the model that appropriately fit this variable was one using a negative binomial distribution. Thus, the estimate for predictor variables for the number of enrolled relatives was derived through a general linear model using a negative binomial regression link function. The following variables were included in the initial model: age, family history of high cholesterol levels, DLNC score, Simon Broome Score, baseline lipid-lowering treatment, employment situation, baseline LDL-C or highest level during treatment, educational level and origin. The mean and standard deviation were calculated for continuous variables. Significance was considered at a p < 0.05. Statistical analyses were performed with SPSS v19.0 (IBM) and R software (Package gamlss, version 3.3.1). Results A total of 183 ICs were analyzed, of which 2316 relatives were contacted and 1605 agreed to enroll in the program (overall enrollment rate of 69.3%). Eighty-seven families were excluded from the study after model adjustment for multiple regression analysis. These were related to 87 ICs that had missing data in at least one of the variables included in the final model. Clinical characteristics of the ICs are shown in Table 1. Regarding the educational level, 30.6% of ICs had college, 25.1% high-school, 22.4% elementary education and 4.9% were illiterate. The greatest percentage of ICs is currently employed (41.0%). Most of ICs were referred by local physicians (81.4%), followed by 7.7% of patients that reached the program via the website. The other 5% were referred from partner centers located at other tertiary care institutions and 3.3% from private physicians. Table 2 shows the univariate negative binomial regression calculated for all the variables in the study. Only family history of altered lipid levels and referral of patients via the website were significantly associated with the number of relatives brought into the program. Results after model adjustment are outlined in Table 3. Family history of high LDL-C levels was an independent predictor associated with a higher number of enrolled relatives, with an increasing estimate of 1.76-fold when comparing ICs with and without family history of dyslipidemia. IC baseline LDL-Cvalues were also associated with a higher number of enrolled relatives. The IC referral origin also significantly influenced the number of relatives in the program. When comparing the origin of ICs, for those enrolled via website the expected number of relatives decreased by 0.42-fold when compared to ICs referred from inside a referral center. Discussion The present study is, to the best of our knowledge, the first to assess the predictors that might influence enrollment of Table 1 – Clinical characteristics of Index cases Variables n Age (Mean ± SD) 183 47 ± 18 Male sex (%) 84 45.9 Tendon xanthomas (%) 26 14.2 Corneal Arcus (%) 49 26.8 Early coronary disease (%) * 54 29.5 Family history of early coronary disease (%) † 72 39.3 Family history of increased LDL-C levels(%) ‡ 98 53.6 Current pharmacological treatment (%) § 145 79.2 DLCN Score (%) Definitive 74 40.4 Probable 48 26.2 Possible 33 18.0 Simon Broome (%) Definitive 29 15.8 Probable 124 67.8 Baseline TC // mg/dL (Mean ± SD) 104 405 ± 112 TC mg/dL (Mean±SD) highest level during treatment 64 305 ± 124 Baseline LDL-C mg/dL (Mean ± SD) 104 326 ± 111 LDL-C mg/dL (Mean ± SD) highest level during treatment 64 238 ± 122 Baseline HDL-C mg/dL (Mean ± SD) 102 47 ± 15 HDL-C mg/dL (Mean±SD) highest level during treatment 64 43 ± 10 Baseline TG ¶ mg/dL (Mean ± SD) 99 144 ± 63 TG mg/dL (Mean ± SD) highest level during treatment 32 132 ± 77 *Coronary disease in men aged < 55 years or women aged < 60 years. †  Family history of coronary disease (e.g.. heart attack) in first or second degree relatives (men aged < 55 years and women < 60 years). ‡ First or second degree relatives with TC > 260 mg/dl or LDL > 160 mg/dL in children (> 16 years old) orTC > 290mg/dLor LDL> 190mg/dL in adults (pre-treatment levels or the highest level during treatment). § Current use of lipid-lowering drugs (e.g. statins). // TC: total cholesterol; TG: triglycerides. SD:standard deviation; DLCN: Dutch Lipid Clinic Network; HDL-C: high-density lipoprotein cholesterol; LDL-C: Low-density lipoprotein cholesterol. 580