Estimating Model-Adjusted Risks, Risk Differences, and Risk Ratios From Complex Survey Data

Statistics and Epidemiology Unit, RTI International, Research Triangle Park, North Carolina 27709-2194, USA.
American journal of epidemiology (Impact Factor: 5.23). 03/2010; 171(5):618-23. DOI: 10.1093/aje/kwp440
Source: PubMed


There is increasing interest in estimating and drawing inferences about risk or prevalence ratios and differences instead
of odds ratios in the regression setting. Recent publications have shown how the GENMOD procedure in SAS (SAS Institute Inc.,
Cary, North Carolina) can be used to estimate these parameters in non-population-based studies. In this paper, the authors
show how model-adjusted risks, risk differences, and risk ratio estimates can be obtained directly from logistic regression
models in the complex sample survey setting to yield population-based inferences. Complex sample survey designs typically
involve some combination of weighting, stratification, multistage sampling, clustering, and perhaps finite population adjustments.
Point estimates of model-adjusted risks, risk differences, and risk ratios are obtained from average marginal predictions
in the fitted logistic regression model. The model can contain both continuous and categorical covariates, as well as interaction
terms. The authors use the SUDAAN software package (Research Triangle Institute, Research Triangle Park, North Carolina) to
obtain point estimates, standard errors (via linearization or a replication method), confidence intervals, and P values for the parameters and contrasts of interest. Data from the 2006 National Health Interview Survey are used to illustrate
these concepts.

Full-text preview

Available from:
  • Source
    • "Missing values for education and number of chronic conditions were treated as not missing completely at random in the weighted frequency calculation. To evaluate the association between survey year and receipt of preventive services, adjusted prevalence of receipt and marginal prevalence ratio (PR) and the 95% confidence intervals (CI) (Bieler et al., 2010) were calculated for each insurance type using multivariable logistic regression controlling for age, gender, race/ethnicity , education, marital status, region, residence, and number of chronic conditions . In order to assess if the association varied by socioeconomic status (Damiani et al., 2011; Hoeck et al., 2014; Sambamoorthi and McAlpine, 2003) and health status, we further conducted stratified analyses by family income and the number of chronic conditions, particularly for adults aged 18–64 years with any private insurance. "
    [Show abstract] [Hide abstract]
    ABSTRACT: An early provision of the Affordable Care Act (ACA) eliminated cost-sharing for a range of recommended preventive services. This provision took effect in September 2010, but little is known about its effect on preventive service use. We evaluated changes in the use of recommended preventive services from the 2009 (before the implementation of ACA cost sharing provision) and 2011/2012 (after the implementation) in the Medical Expenditure Panel Survey, a nationally representative household interview survey in the US. Specifically, we examined: blood pressure check, cholesterol check, flu vaccination, and cervical, breast, and colorectal cancer screening, controlling for demographic characteristics and stratifying by insurance type. There were 64,280 (21,310 before and 42,970 after the implementation of ACA cost-sharing provision) adults included in the analyses. Receipt of recent blood pressure check, cholesterol check and flu vaccination increased significantly from 2009 to 2011/2012, primarily in the privately insured population ages 18-64 years, with adjusted prevalence ratios (95% confidence intervals) 1.03 (1.01-1.05) for blood pressure check, 1.13 (1.09-1.18) for cholesterol check and 1.04 (1.00-1.08) for flu vaccination (all p-values <0.05). However, few changes were observed for cancer screening. We observed little change in the uninsured population. These early observations suggest positive benefits from the ACA policy of eliminating cost sharing for some preventive services. Future research is warranted to monitor and evaluate longer term effects of the ACA on access to care and health outcomes. Copyright © 2015. Published by Elsevier Inc.
    Full-text · Article · Jul 2015 · Preventive Medicine
  • Source
    • "Tests for trend are reported as Wald F values with DF=(1, >100); DF (denominator) = number of PSUs with relevant observations (1268) minus number of survey strata (2). Predicted marginal risk ratios (model-adjusted risk ratios) were reported from logistic regression instead of odds ratios [35,36]. Risk ratios and odds ratios are close for uncommon outcomes but for more common outcomes odds ratios are more extreme than risk ratios. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Abstract Background: Criteria for the diagnosis of alcohol abuse or dependence in DSM-IV or ICD-10 do not include measures of alcohol consumption. However the Alcohol Use Disorders Identification Test (AUDIT) contains three consumption questions (AUDIT-C) plus seven problem questions. The AUDIT-C has often been used as a short screening questionnaire. Here drinking patterns in the past year are analysed, and the AUDIT-C and other combinations of those three questions are related to alcohol problems or diagnoses in the same period. Methods: The 2003-2004 New Zealand Mental Health Survey (N=12,992), a nationally representative survey, included the AUDIT and the Composite International Diagnostic Interview (CIDI) 3.0. Latent class analyses were used to discover patterns of alcohol consumption (AUDIT-C) and patterns of alcohol problems. Cross-tabulations, Receiver Operating Characteristic Curves and logistic regression were used to relate consumption to problems and diagnoses. Results: Analyses indicated that drinking frequency (Q1) was an ineffective screening question. Amount consumed per drinking day (Q2) plus frequency of per-occasion heavy drinking (Q3) was as good as or better than the AUDIT-C, with Q3 alone nearly as good. For a given consumption score, males were only slightly more likely than females to experience negative consequences from their drinking whereas age differences were more substantial. For both sexes and all age groups, a reasonable sensitivity of around 80% was achieved with often rather low specificity for detection of any drinking problems (specificities 57- 81%). However there was higher specificity for detection of multiple problems, or diagnosis (specificities 72-85%). Conclusion: Usual drinking frequency is a poor screening indicator of past year alcohol problems and alcohol diagnoses, and does not improve on frequency of heavy per-occasion drinking, or that question plus usual quantity consumed. Retention of the usual drinking frequency question in the AUDIT-C must be based on considerations apart from its value in screening.
    Full-text · Article · Jan 2015
  • Source
    • "prevalence of needs improvement – health risk CRF combined with unhealthy MSF). To investigate potential for internal selection bias in CRF, MSF, or urban vs. rural, we calculated the relative prevalence of a missing value by other demographic and fitness variables using logistic regression [49]. Missings in BMI and WC were not evaluated due to very few missing values in the working sample (BMI: n = 0; WC: n = 9). "
    [Show abstract] [Hide abstract]
    ABSTRACT: Purpose In addition to excess adiposity, low cardiorespiratory fitness (CRF) and low musculoskeletal fitness (MSF) are important independent risk factors for future cardio-metabolic disease in adolescents, yet global fitness surveillance in adolescents is poor. The objective of this study was to describe and investigate geographical variation in levels of health-related physical fitness, including CRF, MSF, body mass index (BMI), and waist circumference (WC) in Chilean 8th graders. Methods This cross-sectional study was based on a population-based, representative sample of 19,929 8th graders (median age = 14 years) in the 2011 National Physical Education Survey from Chile. CRF was assessed with the 20-meter shuttle run test, MSF with standing broad jump, and body composition with BMI and WC. Data were classified according to health-related standards. Prevalence of levels of health-related physical fitness was mapped for each of the four variables, and geographical variation was explored at the country level by region and in the Santiago Metropolitan Area by municipality. Results Girls had significantly higher prevalence of unhealthy CRF, MSF, and BMI than boys (p<0.05). Overall, 26% of boys and 55% of girls had unhealthy CRF, 29% of boys and 35% of girls had unhealthy MSF, 29% of boys and 44% of girls had unhealthy BMI, and 31% of adolescents had unhealthy WC. High prevalence of unhealthy fitness levels concentrates in the northern and middle regions of the country and in the North and Southwest sectors for the Santiago Metropolitan Area. Conclusion Prevalence of unhealthy CRF, MSF, and BMI is relatively high among Chilean 8th graders, especially in girls, when compared with global estimates. Identification of geographical regions and municipalities with high prevalence of unhealthy physical fitness presents opportunity for targeted intervention.
    Full-text · Article · Sep 2014 · PLoS ONE
Show more