A warning concerning the estimation of multinomial logistic models with correlated responses in SAS

Rooij, Mark de; Worku, Hailemichael M.

doi:10.1016/j.cmpb.2012.01.008

Cited by 11 publications

(9 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To measure the generational increase in alcohol use, we used multinomial logistic regression for corre- lated responses. Confidence intervals for the parameter estimates were obtained from 1000 bootstrapped samples as demonstrated by de Rooij and Worku 16 using SAS software (SAS Institute Inc). This marginal model treats mothers' and daughters' drinking in addition to the social and economic factors as repeated measures of the same unit of observation.…”

Section: Discussionmentioning

confidence: 99%

Generational Increase in Young Women’s Drinking

et al. 2014

View full text Add to dashboard Cite

urveys in Western countries suggest that members of recent birth cohorts are consuming more alcohol at younger ages than preceding cohorts. 1 The patterns of drinking also appear to have changed, with recent cohorts in the United States reporting more frequent heavy drinking at younger ages but reduced alcohol use at later ages. 2,3 In Australia, less sophisticated analyses of cross-sectional surveys over time suggest that initiation of drinking has occurred at increasingly earlier ages 4 and that risky alcohol consumption among youth has increased. 5,6 Especially noteworthy in the increasing alcohol use among young cohorts is increased drinking among young females. 1 There was a large increase in high-risk drinking among girls 14 to 17 years of age (1%-9%) in Australia between 1998 and 2001, while levels of drinking in men 18 to 24 years of age decreased. 7 In the United Kingdom, the prevalence of alcohol misuse increased in those aged 18 to 24 years during the 1990s, particularly among women. 8 Likewise, data from the Netherlands twin registry show that alcohol consumption increased among adolescents between 1993 and 2005 to 2008, with the sharpest increase in girls aged 12 to 15 years. 9 Quantifying generational trends using repeated crosssectional survey data is challenging. The use of different survey questions at different periods means that the rates of al-IMPORTANCE Increases in alcohol use in young women over recent decades are shown by national survey data but have yet to be replicated using prospective data. OBJECTIVE To compare change in alcohol use over a generation of young women born in Australia from 1981 to 1983 at 21 years with that of their mothers at the same age. DESIGN, SETTING, AND PARTICIPANTS Data came from the Mater University Study of Pregnancy, a prospective prebirth cohort study that recruited all pregnant mothers attending a hospital in Brisbane, Australia, from 1981 to 1983. The analyses were restricted to 1053 mothers who were aged 18 to 25 years of age at the baseline measurement and their daughters who were between the same ages when assessed 21 years later. MAIN OUTCOME AND MEASURE Assessing the increase in the same prospective measures of 4 levels of alcohol use between mothers and daughters. RESULTS Multinomial logistic regression for clustered data indicated daughters were at greater odds of consuming high (odds ratio [OR] = 5.68 [95% CI, 4.24-7.57]) and moderate (OR = 2.81 [95% CI, 2.18-3.63]) levels of alcohol than their mothers. Not having a dependent child roughly doubled the odds of all levels of drinking in both cohorts. Undertaking or completing tertiary education had no effect on consumption. There was an interaction between mothers' or daughters' drinking and partner status (χ 2 3 = 12.56; P = .007); having a partner doubled the daughters' odds of consuming high levels of alcohol (no partner: OR = 0.51 [95% CI, 0.31-0.80]) while the odds of drinking at the highest level were more than 5 times for mothers who were single (OR = 5.65 [95% CI, 2.99-12.35]). CONCLUSIONS A...

show abstract

Section: Discussionmentioning

confidence: 99%

Generational Increase in Young Women’s Drinking

et al. 2014

View full text Add to dashboard Cite

show abstract

“…PROC NLMIXED in SAS (Statistical Analysis System) was applied to determine the predictive relationship between subtypes of patients and siblings. An adaptive Gaussian quadrature with ten quadrature points was specified to integrate out the random effect of the likelihood function and to estimate the parameters (i.e., subtypes of patients) and their standard errors. The intra‐cluster correlation coefficient (ICC) was calculated to estimate the familial correlation between pairs of unaffected siblings and probands in the same family.…”

Section: Methodsmentioning

confidence: 99%

Long‐term cognitive trajectories and heterogeneity in patients with schizophrenia and their unaffected siblings

Islam

Habtewold

et al. 2018

Acta Psychiatr Scand

View full text Add to dashboard Cite

ObjectiveThis study aimed to assess the heterogeneity and stability of cognition in patients with a non‐affective psychotic disorder and their unaffected siblings. In addition, we aimed to predict the cognitive subtypes of siblings by their probands.MethodAssessments were conducted at baseline, 3 and 6 years in 1119 patients, 1059 siblings and 586 controls from the Genetic Risk and Outcome of Psychosis (GROUP) study. Group‐based trajectory modeling was applied to identify trajectories and clustered multinomial logistic regression analysis was used for prediction modeling. A composite score of eight neurocognitive tests was used to measure cognitive performance.ResultsFive stable cognitive trajectories ranging from severely altered to high cognitive performance were identified in patients. Likewise, four stable trajectories ranging from moderately altered to high performance were found in siblings. Siblings had a higher risk of cognitive alteration when patients’ alteration was mild (OR = 2.21), moderate (OR = 5.70), and severe (OR = 10.07) compared with patients with intact cognitive function. The familial correlation coefficient between pairs of index patients and their siblings was 0.27 (P = 0.003).ConclusionsThe cognitive profiles identified in the current study might be suitable as endophenotypes and could be used in future genetic studies and predicting functional and clinical outcomes.

show abstract

“…We opted to use multinomial rather than ordinal regression because we considered non‐smoking qualitatively different to any level of smoking, and as it is not constrained by the proportional odds assumption, it allows greater flexibility in exploring relationships with the covariables at different levels of the response variable. Confidence intervals for the parameter estimates were obtained from 1000 bootstrapped samples as demonstrated by De Rooij [23] using SAS software (SAS Institute Inc., Cary, NC, USA). This marginal model treats mothers' and daughters' smoking, as well as education and depressive symptoms, as repeated measures of the same unit of observation (i.e.…”

Section: Methodsmentioning

confidence: 99%