The Effect of Partly Missing Covariates on Statistical Power in Randomized Controlled Trials With Discrete-Time Survival Endpoints

Abstract. In randomized controlled trials (RCTs), a common strategy to increase power to detect a treatment effect is adjustment for baseline covariates. However, adjustment with partly missing covariates, where complete cases are only used, is inefficient. We consider different alternatives in trials with discrete-time survival data, where subjects are measured in discrete-time intervals while they may experience an event at any point in time. The results of a Monte Carlo simulation study, as well as a case study of randomized trials in smokers with attention deficit hyperactivity disorder (ADHD), indicated that single and multiple imputation methods outperform the other methods and increase precision in estimating the treatment effect. Missing indicator method, which uses a dummy variable in the statistical model to indicate whether the value for that variable is missing and sets the same value to all missing values, is comparable to imputation methods. Nevertheless, the power level to detect the treatment effect based on missing indicator method is marginally lower than the imputation methods, particularly when the missingness depends on the outcome. In conclusion, it appears that imputation of partly missing (baseline) covariates should be preferred in the analysis of discrete-time survival data.

Optimal designs in longitudinal trials with varying treatment effects and discrete-time survival endpoints

It is plausible to assume that the treatment effect in a longitudinal study will vary over time. It can become either stronger or weaker as time goes on. Here, we extend previous work on optimal designs for discrete-time survival analysis to trials with the treatment effect varying over time. In discrete-time survival analysis, subjects are measured in discrete time intervals, while they may experience the event at any point in time. We focus on studies where the width of time intervals is fixed beforehand, meaning that subjects are measured more often when the study duration increases. The optimal design is defined as the optimal combination of the number of subjects, the number of measurements for each subject, and the optimal proportion of subjects assigned to the experimental condition. We study optimal designs for different optimality criteria and linear cost functions. We illustrate the methodology of finding optimal designs using a clinical trial that studies the effect of an outpatient mental health program on reducing substance abuse among patients with severe mental illness. We observe that optimal designs depend to some extent on the rate at which group differences vary across time intervals and the direction of these changes over time. We conclude that an optimal design based on the assumption of a constant treatment effect is not likely to be efficient if the treatment effect varies across time.

Mother-Daughter Correlation of Central Obesity and Other Noncommunicable Disease Risk Factors

This study aimed to investigate the mother-daughter correlation for central obesity and other noncommunicable disease risk factors. The authors used metabolic and anthropometric data from the Tehran Lipid and Glucose Study, enrolling 1041 mother-daughter pairs for the current study. Three age strata were defined: 3 to 9 years for childhood (146 mother-daughter pairs), 10 to 17 years for adolescence (395 mother-daughter pairs), and 18 to 25 years for early adulthood (500 mother-daughter pairs). Familial associations for central obesity and other noncommunicable disease risk factors were assessed. The prevalence of central obesity was 44.7% in mothers and 11.2% in daughters (6.2% in the 3-9, 19.2% in the 10-17, and 6.4% in the 18-25 years groups). Mothers with central obesity were more likely than nonobese mothers to have daughters with central obesity (10.5% and 1.7%, respectively; P = .0001). Central obesity indices among daughters were positively correlated with those of their mothers in all 3 age strata. Correlations for other noncommunicable disease risk factors were analyzed before and after adjusting the risk factor levels for mothers’ and daughters’ waist circumferences (WCs) within each group to determine whether risk factor correlations were, in part, a result of the central obesity correlations. After the non-communicable disease risk factor levels of participants were adjusted for their WCs, the mother-daughter correlations remained significant. The consistent association of central obesity between mothers and daughters may indicate the key role that could be played by the mother in the primary prevention of central obesity, particularly in high-risk families.

"Predictability of body mass index for diabetes: affected by the presence of metabolic syndrome?"

Background

Metabolic syndrome (MetS) and body mass index (BMI, kg.m^-2) are established independent risk factors in the development of diabetes; we prospectively examined their relative contributions and joint relationship with incident diabetes in a Middle Eastern cohort.

Method

participants of the ongoing Tehran lipid and glucose study are followed on a triennial basis. Among non-diabetic participants aged≥ 20 years at baseline (8,121) those with at least one follow-up examination (5,250) were included for the current study. Multivariate logistic regression models were used to estimate sex-specific adjusted odd ratios (ORs) and 95% confidence intervals (CIs) of baseline BMI-MetS categories (normal weight without MetS as reference group) for incident diabetes among 2186 men and 3064 women, aged ≥ 20 years, free of diabetes at baseline.

Result

During follow up (median 6.5 years); there were 369 incident diabetes (147 in men). In women without MetS, the multivariate adjusted ORs (95% CIs) for overweight (BMI 25-30 kg/m2) and obese (BMI≥30) participants were 2.3 (1.2-4.3) and 2.2 (1.0-4.7), respectively. The corresponding ORs for men without MetS were 1.6 (0.9-2.9) and 3.6 (1.5-8.4) respectively. As compared to the normal-weight/without MetS, normal-weight women and men with MetS, had a multivariate-adjusted ORs for incident diabetes of 8.8 (3.7-21.2) and 3.1 (1.3-7.0), respectively. The corresponding ORs for overweight and obese women with MetS reached to 7.7 (4.0-14.9) and 12.6 (6.9-23.2) and for men reached to 3.4(2.0-5.8) and 5.7(3.9-9.9), respectively.

Conclusion

This study highlights the importance of screening for MetS in normal weight individuals. Obesity increases diabetes risk in the absence of MetS, underscores the need for more stringent criteria to define healthy metabolic state among obese individuals. Weight reduction measures, thus, should be encouraged in conjunction with achieving metabolic targets not addressed by current definition of MetS, both in every day encounter and public health setting.

A new approach to compare the predictive power of metabolic syndrome defined by a joint interim statement versus its components for incident cardiovascular disease in Middle East Caucasian residents in Tehran

BACKGROUND

Metabolic syndrome (MS) has an increased risk for developing cardiovascular disease (CVD); however, whether the concept of MS, applying the new joint interim statement definition, has a greater impact on incident CVD than its individual components is debated.

METHODS

The authors related MS and its components to CVD incidence in 1856 men and 2392 women, Iranian adults aged ≥40 years, free of CVD at baseline, using Cox proportional hazards models. To examine whether MS would improve prediction of CVD beyond that achieved by its components; model fitness, discrimination and integrated discrimination improvement (IDI) statistics were used.

RESULTS

During a median follow-up of 8.6 years, there were 244 CVD events in men and 189 in women. MS resulted in HRs (95% CIs) of 1.97 (1.50 to 2.57) in men and 2.25 (1.57 to 3.21) in women after adjusting for CVD risk factors; in another model including all the five MS components, high blood pressure in both genders [men: 1.99 (1.48 to 2.67), women: 1.62 (1.14 to 2.30)), high waist circumference (≥94.5 cm) in men (1.47 (1.12 to 1.93)) and high fasting plasma glucose (FPG) in women (1.88 (1.39 to 2.55)) remained as independent predictors of CVD after adjusting for CVD risk factors. Adding the MS variable to this model did not yield any improvement in model fitness, C-statistic or significant IDI value.

CONCLUSION

In the Middle East population, MS did not provide CVD predictive risk information beyond its individual components; clinical focus should remain on hypertension in both sexes, high FPG in women and central adiposity in men rather than MS.

Predictive power of the components of metabolic syndrome in its development: a 6.5-year follow-up in the Tehran Lipid and Glucose Study (TLGS)

BACKGROUND/OBJECTIVES

To determine which component of the metabolic syndrome (MetS) is the best predictor of its development.

SUBJECTS/METHODS

In this cohort study, 2279 subjects aged 20-87 years without MetS selected from among the participants of the cross-sectional phase of the Tehran Lipid and Glucose Study (TLGS) were followed up for development of MetS.

RESULTS

After a mean interval of 6.5 years, 462 and 602 new cases of MetS were diagnosed on the basis of the modified Adult Treatment Panel III (ATP III) and International Diabetes Federation (IDF) criteria, respectively. The adjusted odds ratio for development of MetS by ATP III criteria was highest for central obesity in men, 2.8 (2.2-3.7), and for triglycerides (TGs) in women, 2.8 (2.0-4.1). The adjusted odds ratio for the development of MetS by IDF criteria was highest for TGs in both men and women: odds ratio 2.8 (2.2-3.7) and 2.9 (1.9-4.3), respectively. A model that included waist circumference (WC) and TGs or WC and high-density lipoprotein (HDL) predicted MetS similar to a model that included all five MetS components.

CONCLUSION

Screening for the timely prediction of the development of MetS should include measurement of WC, TGs and plasma HDL.

Lipid measures for prediction of incident cardiovascular disease in diabetic and non-diabetic adults: results of the 8.6 years follow-up of a population based cohort study

BACKGROUND

Diabetes is a strong risk factor for cardiovascular disease (CVD).The relative role of various lipid measures in determining CVD risk in diabetic patients is still a subject of debate. We aimed to compare performance of different lipid measures as predictors of CVD using discrimination and fitting characteristics in individuals with and without diabetes mellitus from a Middle East Caucasian population.

METHODS

The study population consisted of 1021 diabetic (men = 413, women = 608) and 5310 non-diabetic (men = 2317, women = 2993) subjects, aged > or = 30 years, free of CVD at baseline. The adjusted hazard ratios (HRs) for CVD were calculated for a 1 standard deviation (SD) change in total cholesterol (TC), log-transformed triglyceride (TG), high density lipoprotein cholesterol (HDL-C), low density lipoprotein cholesterol (LDL-C), non-HDL-C, TC/HDL-C and log-transformed TG/HDL-C using Cox proportional regression analysis. Incident CVD was ascertained over a median of 8.6 years of follow-up.

RESULTS

A total of 189 (men = 91, women = 98) and 263(men = 169, women = 94) CVD events occurred, in diabetic and non-diabetic population, respectively. The risk factor adjusted HRs to predict CVD, except for HDL-C, TG and TG/HDL-C, were significant for all lipid measures in diabetic males and were 1.39, 1.45, 1.36 and 1.16 for TC, LDL-C, non- HDL-C and TC/HDL-C respectively. In diabetic women, using multivariate analysis, only TC/HDL-C had significant risk [adjusted HR1.31(1.10-1.57)].Among non-diabetic men, all lipid measures, except for TG, were independent predictors for CVD however; a 1 SD increase in HDL-C significantly decreased the risk of CVD [adjusted HR 0.83(0.70-0.97)].In non-diabetic women, TC, LDL-C, non-HDL-C and TG were independent predictors.There was no difference in the discriminatory power of different lipid measures to predict incident CVD in the risk factor adjusted models, in either sex of diabetic and non-diabetic population.

CONCLUSION

Our data according to important test performance characteristics provided evidence based support for WHO recommendation that along with other CVD risk factors serum TC vs. LDL-C, non-HDL-C and TC/HDL-C is a reasonable lipid measure to predict incident CVD among diabetic men. Importantly, HDL-C did not have a protective effect for incident CVD among diabetic population; given that the HDL-C had a protective effect only among non- diabetic men.