Risk-set Sampling Research Articles

Serum levels of C‐peptide, IGFBP‐1 and IGFBP‐2 and endometrial cancer risk; Results from the European prospective investigation into cancer and nutrition

We conducted a case-control study nested within the European Prospective Investigation into Cancer and Nutrition, to examine the associations between prediagnostic serum concentrations of C-peptide, insulin-like growth factor binding protein (IGFBP)-1 and IGFBP-2, and endometrial cancer risk. Among pre- and post-menopausal women, who were not currently using exogenous hormones, 286 women developed incident endometrial cancer during an average 5.1 years follow-up. Using risk set sampling, 555 matched control subjects were selected. In conditional logistic regression models adjusted for matching factors only, endometrial cancer risk increased with increasing serum levels of C-peptide (relative risks (RR) for the top vs. bottom quartile = 2.13 [95% confidence interval (CI) 1.33-3.41], p(trend) = 0.001, and decreasing serum levels of IGFBP-2 (RR for the top vs. bottom quartile = 0.56 [95% CI 0.35-0.90], p(trend) = 0.03, but was not significantly associated with IGFBP-1 levels (RR for the top vs. bottom quartile = 0.76 [95% CI 0.47-1.21], p(trend) = 0.25). In BMI-adjusted models, only the C-peptide association remained marginally statistically significant (RR for the top vs. bottom quartile = 1.56 [95% CI 0.94-2.57], p(trend) = 0.05 for C-peptide; 0.84 [95% CI 0.50-1.40], p(trend) = 0.74 for IGFBP-2; and 1.08 [95% CI 0.65-1.78], p(trend) = 0.86 for IGFBP-1 levels). These associations were stronger among nonfasting women (< or =< or =6 hr since last meal; 63% of subjects) but were not evident among fasting women, although the interactions were not statistically significant. The C-peptide-risk association was substantially attenuated after adjustment for free estradiol in postmenopausal women (RR for the top vs. bottom quartile = 1.28 [95% CI 0.67-2.45], p(trend) = 0.42. Our results provide modest support to the hypothesis that hyperinsulinaemia is a risk factor for endometrial cancer.

Postmenopausal Hormone Replacement Therapy and Risk of Acute Pancreatitis: A Population-Based Case-Control Study

To examine whether acute pancreatitis is associated with the use of postmenopausal hormonal replacement therapy in Danish women over 45 yr of age. We based this population-based case-control study on data from three Danish counties for the years 1991-2003. We identified all women (>45 yr of age) with a first hospital discharge diagnosis of acute pancreatitis in the county hospital discharge registries (N = 1,054). Using the Danish Civil Registration System, we selected 10 age-matched population controls for each case, using risk set sampling (N = 10,540). Data on all prescriptions for estrogens or combined estrogen/progestins redeemed within 90 days before the hospitalization (current users) and 91-365 days before (former users) were collected from the prescription databases. Conditional logistic regression was used to estimate the relative risk of acute pancreatitis after exposure to estrogen or combined estrogen/progestin, adjusted for other risk factors for acute pancreatitis. The adjusted relative risk for acute pancreatitis in current users of menopausal estrogens was 1.1 (95% confidence interval [CI] 0.8-1.4), and 1.1 (95% CI 0.8-1.5) in former users. For current users of combined estrogen/progestins, the adjusted relative risk was 1.2 (95% CI 0.9-1.6), and for former users, 1.6 (95% CI 1.0-2.5). Our data did not support a substantial association between acute pancreatitis and the use of postmenopausal hormone therapy.

Lung Cancer Mortality and Carbon Black Exposure: A Nested Case???Control Study at a German Carbon Black Production Plant

The objective of this study was to conduct a case-control study of lung cancer nested within a cohort of 1528 German carbon black workers, 1976-1998. The authors conducted risk-set sampling of two controls matched on year of birth and conditional logistic regression modeling of cumulative carbon black exposure, duration of work in different departments, feedstock contact, asbestos exposure, smoking, age at hire, exposures before the carbon black job, and serving as a soldier in World War II or being a prisoner of war. Analyses were performed with both the full cohort and members of an inception cohort subset. Exposures were lagged by 10 years. Analysis of 50 lung cancer deaths showed no association to carbon black exposure. Carbon black exposure was not linked to lung cancer risk. Suggestions of positive associations with asbestos exposure, feedstock contact, and work in specific departments are inconclusive due to small numbers.

Plasma Adiponectin Levels and Endometrial Cancer Risk in Pre- and Postmenopausal Women

Adiponectin, an adipocytokine secreted by adipose tissue, is decreased in obesity, insulin resistance, type 2 diabetes, and polycystic ovary syndrome, all of which are well-established risk factors for endometrial cancer. We conducted a case-control study nested within the European Prospective Investigation into Cancer and Nutrition to examine the relation between prediagnostic plasma adiponectin levels and endometrial cancer risk. Among pre- and postmenopausal women who were not currently using exogenous hormones, 284 women developed incident endometrial cancer during an average of 5.1 yr of follow-up. Using risk set sampling, 548 control subjects were selected, matched on center, age, menopausal status, phase of menstrual cycle, time of blood draw, and fasting status. Conditional logistic regression models were used to estimate relative risks and 95% confidence intervals. Adiponectin levels were inversely associated with endometrial cancer risk [body mass index-adjusted relative risk for the top vs. bottom quartile = 0.56 (95% confidence interval 0.36-0.86), P(trend) = 0.006]. There was evidence of a stronger inverse association among obese women than among nonobese women (P(heterogeneity) = 0.03). The inverse association also appeared stronger for women who were postmenopausal or perimenopausal than premenopausal at baseline, but this was not statistically significantly heterogeneous (P(heterogeneity) = 0.51). The association remained statistically significant after separate adjustment for other obesity-related physiological risk factors such as C-peptide, IGF binding protein-1, IGF binding protein-2, SHBG, estrone, or free testosterone but only marginally statistically significant after simultaneous adjustment for these factors. High circulating adiponectin levels are associated with reduced endometrial cancer risk, largely independent of other obesity-related risk factors.

Beryllium and Lung Cancer-Some Methodological Issues Involving Risk-Set Sampling and Exposure Lagging in a Nested Case-Control Study

Beryllium and Lung Cancer-Some Methodological Issues Involving Risk-Set Sampling and Exposure Lagging in a Nested Case-Control Study

Valproic Acid and Risk of Acute Pancreatitis: A Population-Based Case-Control Study

To examine whether acute pancreatitis is associated with use of valproic acid. Through the population-based hospital discharge registries we identified all patients with an incident hospitalization of acute pancreatitis in the counties of North Jutland (data 1991 to 2003), Aarhus (data 1996 to 2003), and Viborg (data 1998 to 2003), Denmark. From the Danish Civil Registration System, we selected 10 sex-matched and age-matched population controls per case on the basis of risk set sampling. All prescriptions of valproic acid and other antiepileptic drugs within 90 days (present users) or 91 to 365 days (past users) before hospital admission with acute pancreatitis, or index date among controls, were collected from the prescription databases in the counties. We performed conditional logistic regression to estimate the relative risk of acute pancreatitis after exposure to valproic acid or other antiepileptic drugs, adjusting for gallstone diseases, alcohol-related diseases, hyperlipidemia, and hypercalcemia. We included 3083 cases of acute pancreatitis and 30,830 population controls. The adjusted odds ratio (OR) for acute pancreatitis in present users of valproic acid was 1.9 [95% confidence interval (CI), 1.1-3.3); for past users, the adjusted OR was 2.6 (95% CI, 0.8-8.7). For users of other antiepileptic drugs, the corresponding adjusted ORs were 1.6 (95% CI, 1.2-2.2) and 1.8 (95% CI, 1.1-3.0). Use of valproic acid is associated with an elevated relative risk estimate for acute pancreatitis, but it was not materially different from past use or use of other antiepileptic drugs. Therefore, our data challenge the hypothesis that valproic acid is an independent risk factor for acute pancreatitis.

Understanding the effect of deployment on the risk of fatal motor vehicle crashes: A nested case–control study of fatalities in Gulf War era veterans, 1991–1995

Motor vehicle crashes (MVCs) are an important cause of morbidity and premature loss of life among military personnel during peacetime and particularly following combat. A nested case–control study of fatal MVC occurring between 1991 and 1995 was conducted in a cohort of Gulf War era veterans. Cases were validated MVC deaths in the Fatality Analysis Reporting System. Controls were selected using risk set sampling by gender and year of case ascertainment in a 10:1 ratio. Preliminary results, consistent with previous reports of increased fatal MVC risk among returning combat veterans, showed a crude odds ratio of 1.45 (95% confidence interval 1.27–1.65). Multivariable logistic regression modeling was used to identify important independent predictors, as well as to quantify the influence of deployment on a risk profile for fatal MVC. Because of significant interaction between deployment and inpatient diagnosis of substance abuse, the final model was stratified by deployment status. Results suggest that demographic, military, and behavioral characteristics of deployed healthy warriors are similar to the risk profile for fatal MVC. In addition to young, single, high school-educated, enlisted male personnel, those who served during times of ground combat, particularly in infantry, gun crews, or seamanship occupations, should be targeted for preventive interventions.

Reduced Mortality Rate Associated with Annual Mammograms after Breast Cancer Therapy

Guidelines have been developed for appropriate posttherapy surveillance for breast cancer recurrence. One purpose of posttherapy surveillance is to detect potentially curable local recurrences and new cancers in the opposite breast. The objective of this investigation was to assess the impact of annual mammography on all-cause mortality in breast cancer survivors. We conducted a case-control analysis nested in a cohort of 865 stage I or II breast cancer patients diagnosed from 1996 to 1999. The exposure variable was the number of mammograms received after completing primary therapy. Cases were decedents and we used risk-set sampling to match eight controls to each case on follow-up time. The mortality rate declined with an increasing number of mammograms (p for trend=0.007). The age- and therapy-adjusted odds ratio associating receipt of an additional mammogram, compared with receipt of no mammogram, equaled 0.77 (95% confidence interval [CI] 0.53-1.1). These results are consistent with a protective effect of regular surveillance mammography after completing therapy for early stage breast cancer.

Non–High-Density Lipoprotein Cholesterol and Apolipoprotein B in the Prediction of Coronary Heart Disease in Men

Apolipoprotein B (apoB) plasma levels reflect the concentration of proatherogenic lipoproteins very low-density lipoprotein and low-density lipoprotein (LDL), whereas non-high-density lipoprotein cholesterol (non-HDL-C) levels reflect the concentration of cholesterol transported by these particles. The aim of our study was to compare apoB, non-HDL-C, LDL cholesterol (LDL-C), and other lipid markers as predictors of coronary heart disease (CHD) in a nested case-control study among 18 225 participants in the Health Professionals Follow-up Study. Among men who were free of diagnosed cardiovascular disease at the time of blood collection, 266 had nonfatal myocardial infarction or fatal CHD during 6 years of follow-up. Through the use of risk set sampling, control subjects were selected at a 2:1 ratio and matched with regard to age, date of blood collection, and smoking status. After adjustment for matching factors, the relative risk of CHD in the highest quintile compared with the lowest quintile was 2.76 (95% confidence interval [CI], 1.66 to 4.58) for non-HDL-C, 3.01 (95% CI, 1.81 to 5.00) for apoB, 1.81 (95% CI, 1.12 to 2.93) for LDL-C, 0.31 (95% CI, 0.18 to 0.52) for HDL-C, 2.41 (95% CI, 1.43 to 4.07) for triglycerides (all P trend <0.001), and 1.42 (95% CI, 0.86 to 2.32, P trend =0.19) for lipoprotein(a). When non-HDL-C and LDL-C were mutually adjusted, only non-HDL-C was predictive of CHD. When non-HDL-C and apoB were mutually adjusted, only apoB was predictive; the relative risk was 4.18 (95% CI, 1.30 to 13.49; P trend =0.02) for apoB compared with 0.70 (95% CI, 0.21 to 2.27; P trend =0.72) for non-HDL-C. Triglycerides added significant information to non-HDL-C but not to apoB for CHD risk prediction. Although non-HDL-C and apoB were both strong predictors of CHD in this male cohort, more so than LDL-C, the findings support the concept that the plasma concentration of atherogenic lipoprotein particles measured by apoB is more predictive in development of CHD than the cholesterol carried by these particles, measured by non-HDL-C.

Nonalcoholic Fatty Liver Disease and Risk of Future Cardiovascular Events Among Type 2 Diabetic Patients

Nonalcoholic fatty liver disease (NAFLD) is closely correlated to several metabolic syndrome features. We assessed prospectively whether NAFLD predicts future cardiovascular disease (CVD) events among type 2 diabetic individuals, independent of metabolic syndrome features and other classical risk factors. We carried out a prospective nested case-control study in 2,103 type 2 diabetic patients who were free of diagnosed CVD at baseline. During 5 years of follow-up, 248 participants (case subjects) subsequently developed nonfatal coronary heart disease (myocardial infarction and coronary revascularization procedures), ischemic stroke, or cardiovascular death. Using risk-set sampling, 496 patients (control subjects) among those who remained free of diagnosed CVD during follow-up were randomly selected in a 2:1 ratio, matched for age and sex to the case subjects. After adjustment for age, sex, smoking history, diabetes duration, HbA1c, LDL cholesterol, liver enzymes, and use of medications, the presence of NAFLD was significantly associated with an increased CVD risk (odds ratio 1.84, 95% CI 1.4-2.1, P < 0.001). Additional adjustment for the metabolic syndrome (as defined by National Cholesterol Education Program Adult Treatment Panel III criteria) appreciably attenuated, but did not abolish, this association (1.53, 1.1-1.7, P = 0.02). In conclusion, NAFLD is significantly associated with a moderately increased CVD risk among type 2 diabetic individuals. This relationship is independent of classical risk factors and is only partly explained by occurrence of metabolic syndrome.

Polymorphisms in Cytoplasmic Serine Hydroxymethyltransferase and Methylenetetrahydrofolate Reductase Affect the Risk of Cardiovascular Disease in Men

Genetic variation in folate-regulating enzymes contributes to the risk of cardiovascular disease (CVD). The cytoplasmic serine hydroxymethyltransferase (cSHMT) enzyme is proposed to regulate a key metabolic intersection in folate metabolism. We hypothesized that a variant in cSHMT (cSHMT 1420C-->T) affects CVD risk, and that the effect depends on a linked step in the metabolic pathway catalyzed by methylenetetrahydrofolate reductase (MTHFR). A nested case-control study of incident CVD was conducted within the all-male Normative Aging Study cohort. Of the incident CVD cases, 507 had DNA samples; 2 controls/case were selected by risk set sampling (matched on age and birth year). A significant gene-gene interaction (P-values 0.0013, 0.0064) was found between MTHFR and cSHMT, and there was little or no change in the coefficients in covariate-adjusted models. The effect of MTHFR 677C-->T genotype on CVD risk varied by cSHMT 1420C-->T genotype. Among men with cSHMT 1420C-->T TT genotype, the odds ratios (OR) for CVD risk for MTHFR 677C-->T CT and TT genotypes compared with the MTHFR 677C-->T CC genotype were 3.6 (95% CI, 1.7-7.8) and 10.6 (95% CI, 2.5-46.0), respectively. Among men with the cSHMT 1420C-->T CC/CT genotype, the corresponding ORs were 1.0 (95% CI, 0.8-1.2) and 1.3 (95% CI, 0.9-1.8). Plasma total homocysteine concentrations were highest in the subgroup of men with both polymorphisms, MTHFR 677C-->T TT and cSHMT 1420C-->T TT, consistent with a higher risk of CVD in this subgroup. A more complete understanding of the molecular mechanism awaits identification of the functional effect of the polymorphism.

Peroxisome Proliferator-Activated Receptor-γ2 P12A Polymorphism and Risk of Coronary Heart Disease in US Men and Women

Activation of the peroxisome proliferator-activated receptor-gamma (PPARgamma) improves insulin sensitivity and exerts antiatherogenic effects. A common alanine for proline substitution at codon 12 in the PPARG2 gene is related to lower receptor activity. Studies suggest that the A12 allele is associated with reduced risk of type 2 diabetes; however, data on the risk of coronary heart disease (CHD) are scarce and controversial. We examined the relationship between PPARG2 P12A and CHD risk in women (Nurses' Health Study) and men (Health Professionals Follow-Up Study) in nested case control settings. Among participants free of cardiovascular disease at baseline, 249 women and 266 men developed nonfatal myocardial infarction (MI) or fatal CHD during 8 and 6 years of follow-up, respectively. Using risk-set sampling, controls were selected 2:1 matched on age, smoking, and date of blood draw. The relative risk (RR) of nonfatal MI or fatal CHD of carriers compared with noncarriers of the A12 allele was 1.17 (95% CI, 0.82 to 1.68) among women and 1.44 (95% CI, 1.00 to 2.07) among men (pooled RR, 1.30 [95% CI, 1.00 to 1.67]). We found a significantly increased risk associated with the A12 allele among individuals with a body mass index > or =25 kg/m2 (women: RR, 1.88; 95% CI, 1.01 to 3.50; men: RR, 1.55; 95% CI, 0.92 to 2.60; pooled: RR, 1.68; 95% CI, 1.13 to 2.50) but not among those <25 kg/m2 (pooled RR, 0.86; 95% CI, 0.37 to 1.97; P heterogeneity overweight versus nonoverweight 0.16). These data do not support the hypothesis that the A12 allele is associated with a decreased risk of CHD. The potential interaction between PPARG2 P12A, overweight, and increased CHD risk needs further evaluation.

Leveraging existing databases to study vehicle crashes in a combat occupational cohort: Epidemiologic methods

The US military is a large, well-defined occupational cohort offering tremendous opportunities to study risk factors for important health outcomes. This article describes our nested case-control methods to evaluate risk factors for fatal motor vehicle crashes (MVC) within all Service branches in a 1991 Gulf War era cohort. We identified 1,343 cases of fatal MVC between 1991 and 1995 that were also included in the Department of Transportation's Fatality Analysis Reporting System database and, using risk set sampling, selected 13,430 controls. Our final analytic dataset consisted of 980 male driver cases and 12,807 controls linked to multiple databases. Cases were disproportionately younger, less educated, not married, enlisted, and deployed to the Gulf War, compared to controls. The ability to leverage multiple databases to study risk factors for fatal MVC is clearly advantageous and could eventually lead to the reduction of fatalities in similar occupational cohorts.

Reduced rates of cancer-related worries and mortality associated with guideline surveillance after breast cancer therapy

Guidelines have been developed for appropriate post-therapy surveillance for breast cancer recurrence. Two objectives of post-therapy surveillance are to support and counsel patients and to detect potentially curable local recurrences and new cancers in the opposite breast. The objective of this investigation was to assess the impact of guideline surveillance (history, physical examination, and annual mammography) on cancer-related worries and all-cause mortality. We collected data on a cohort of 303 Massachusetts women with stages I or II breast cancer diagnosed between 1992 and 1994. Cases were women with increasing cancer-related worries or decedents. We used risk-set sampling to match five controls to each case on follow-up time. Cases and members of their matched risk set were characterized with respect to receipt of guideline surveillance and covariates preceding the date of their outcomes. The adjusted odds ratio associating guideline surveillance in the preceding year with an increase in cancer-related worries equaled 0.37 (95% CI = 0.14-0.99). The adjusted odds ratio associating continuous guideline surveillance with all-cause mortality equaled 0.66 (95% CI = 0.51-0.86). The results are consistent with the stated objectives of surveillance follow-up of breast cancer patients after the completion of their primary therapy.

Inflammatory Markers and the Risk of Coronary Heart Disease in Men and Women

Few studies have simultaneously investigated the role of soluble tumor necrosis factor alpha (TNF-alpha) receptors types 1 and 2 (sTNF-R1 and sTNF-R2), C-reactive protein, and interleukin-6 as predictors of cardiovascular events. The value of these inflammatory markers as independent predictors remains controversial. We examined plasma levels of sTNF-R1, sTNF-R2, interleukin-6, and C-reactive protein as markers of risk for coronary heart disease among women participating in the Nurses' Health Study and men participating in the Health Professionals Follow-up Study in nested case-control analyses. Among participants who provided a blood sample and who were free of cardiovascular disease at baseline, 239 women and 265 men had a nonfatal myocardial infarction or fatal coronary heart disease during eight years and six years of follow-up, respectively. Using risk-set sampling, we selected controls in a 2:1 ratio with matching for age, smoking status, and date of blood sampling. After adjustment for matching factors, high levels of interleukin-6 and C-reactive protein were significantly related to an increased risk of coronary heart disease in both sexes, whereas high levels of soluble TNF-alpha receptors were significant only among women. Further adjustment for lipid and nonlipid factors attenuated all associations; only C-reactive protein levels remained significant. The relative risk among all participants was 1.79 for those with C-reactive protein levels of at least 3.0 mg per liter, as compared with those with levels of less than 1.0 mg per liter (95 percent confidence interval, 1.27 to 2.51; P for trend <0.001). Additional adjustment for the presence or absence of diabetes and hypertension moderately attenuated the relative risk to 1.68 (95 percent confidence interval, 1.18 to 2.38; P for trend = 0.008). Elevated levels of inflammatory markers, particularly C-reactive protein, indicate an increased risk of coronary heart disease. Although plasma lipid levels were more strongly associated with an increased risk than were inflammatory markers, the level of C-reactive protein remained a significant contributor to the prediction of coronary heart disease.

Statistical estimation in the proportional hazards model with risk set sampling

Thomas’ partial likelihood estimator of regression parameters is widely used in the analysis of nested case-control data with Cox’s model. This paper proposes a new estimator of the regression parameters, which is consistent and asymptotically normal. Its asymptotic variance is smaller than that of Thomas’ estimator away from the null. Unlike some other existing estimators, the proposed estimator does not rely on any more data than strictly necessary for Thomas’ estimator and is easily computable from a closed form estimating equation with a unique solution. The variance estimation is obtained as minus the inverse of the derivative of the estimating function and therefore the inference is easily available. A numerical example is provided in support of the theory.

Plasma adiponectin levels and risk of myocardial infarction in men.

Adiponectin, a recently discovered adipocyte-derived peptide, is involved in the regulation of insulin sensitivity and lipid oxidation and, purportedly, in the development of atherosclerosis and coronary heart disease in humans. To assess prospectively whether plasma adiponectin concentrations are associated with risk of myocardial infarction (MI). Nested case-control study among 18 225 male participants of the Health Professionals Follow-up Study aged 40 to 75 years who were free of diagnosed cardiovascular disease at the time of blood draw (1993-1995). During 6 years of follow-up through January 31, 2000, 266 men subsequently developed nonfatal MI or fatal coronary heart disease. Using risk set sampling, controls were selected in a 2:1 ratio matched for age, date of blood draw, and smoking status (n = 532). Incidence of nonfatal MI and fatal coronary heart disease by adiponectin level. After adjustment for matched variables, participants in the highest compared with the lowest quintile of adiponectin levels had a significantly decreased risk of MI (relative risk [RR], 0.39; 95% confidence interval [CI], 0.23-0.64; P for trend <.001). Additional adjustment for family history of MI, body mass index, alcohol consumption, physical activity, and history of diabetes and hypertension did not substantively affect this relationship (RR, 0.41; 95% CI, 0.24-0.70; P for trend <.001). Further adjustment for hemoglobin A1c or C-reactive protein levels also had little impact, but additional adjustment for low- and high-density lipoprotein cholesterol levels modestly attenuated this association (RR, 0.56; 95% CI, 0.32-0.99; P for trend =.02). High plasma adiponectin concentrations are associated with lower risk of MI in men. This relationship can be only partly explained by differences in blood lipids and is independent of inflammation and glycemic status.

Relationship between dietary antioxidants and childhood asthma: more epidemiological studies are needed

Approximately 15 million Americans including over 5 million children suffer from asthma, the most common chronic disease in childhood. The prevalence of pediatric asthma has risen sharply over the past four decades, with the sharpest increases occurring in children younger than 6 years and in urban, predominantly minority, populations. The reasons for this dramatic increase are not yet clear. Recent epidemiological studies indicate a higher prevalence of dietary antioxidant deficiency among asthmatics patients. However, the results of these studies are inconsistent or even contradictory. Epidemiological studies with robust design and use of novel epidemiological tools are urgently needed to examine the impact of dietary antioxidants on the incidence of asthma in preschool children. An incidence density case-control study which includes non-atopic controls, who are matched for age, gender, race, study center and sampling time to each asthmatic case will offer a robust study design. A validated food frequency questionnaire and an asthma and atopy severity score can be used to interview the parents of the recruited children. Risk set sampling may enable us to explore possible associations between the type and quantity of dietary antioxidants and the development and severity of asthma in such an epidemiological study.

Marginal Analysis For Clustered Failure Time Data

Clustered failure time data are commonly encountered in biomedical research where the study subjects from the same cluster (e.g., family) share the common genetic and/or environmental factors such that the failure times within the same cluster are correlated. Two approaches that are commonly used to account for the intra-cluster association are frailty models and marginal models. In this paper, we study the marginal proportional hazards model, where the structure of dependence between individuals within a cluster is unspecified. An estimation procedure is developed based on a pseudo-likelihood approach, and a risk set sampling method is proposed for the formulation of the pseudo-likelihood. The asymptotic properties of the proposed estimators are studied, and the related issues regarding the statistical efficiencies are discussed. The performances of the proposed estimator are demonstrated by the simulation studies. A data example from a child vitamin A supplementation trial in Nepal (Nepal Nutrition Intervention Project-Sarlahi, or NNIPS) is used to illustrate this methodology.

Cohort case-control design and analysis for clustered failure-time data.

Cohort case-control design is an efficient and economical design to study risk factors for disease incidence or mortality in a large cohort. In the last few decades, a variety of cohort case-control designs have been developed and theoretically justified. These designs have been exclusively applied to the analysis of univariate failure-time data. In this work, a cohort case-control design adapted to multivariate failure-time data is developed. A risk set sampling method is proposed to sample controls from nonfailures in a large cohort for each case matched by failure time. This method leads to a pseudolikelihood approach for the estimation of regression parameters in the marginal proportional hazards model (Cox, 1972, Journal of the Royal Statistical Society, Series B 34, 187-220), where the correlation structure between individuals within a cluster is left unspecified. The performance of the proposed estimator is demonstrated by simulation studies. A bootstrap method is proposed for inferential purposes. This methodology is illustrated by a data example from a child vitamin A supplementation trial in Nepal (Nepal Nutrition Intervention Project-Sarlahi, or NNIPS).