On the use of multiple imputation to address data missing by design as well as unintended missing data in case-cohort studies with a binary endpoint

BACKGROUND

Case-cohort studies are conducted within cohort studies, with the defining feature that collection of exposure data is limited to a subset of the cohort, leading to a large proportion of missing data by design. Standard analysis uses inverse probability weighting (IPW) to address this intended missing data, but little research has been conducted into how best to perform analysis when there is also unintended missingness. Multiple imputation (MI) has become a default standard for handling unintended missingness and is typically used in combination with IPW to handle the intended missingness due to the case-control sampling. Alternatively, MI could be used to handle both the intended and unintended missingness. While the performance of an MI-only approach has been investigated in the context of a case-cohort study with a time-to-event outcome, it is unclear how this approach performs with a binary outcome.

METHODS

We conducted a simulation study to assess and compare the performance of approaches using only MI, only IPW, and a combination of MI and IPW, for handling intended and unintended missingness in the case-cohort setting. We also applied the approaches to a case study.

RESULTS

Our results show that the combined approach is approximately unbiased for estimation of the exposure effect when the sample size is large, and was the least biased with small sample sizes, while MI-only and IPW-only exhibited larger biases in both sample size settings.

CONCLUSIONS

These findings suggest that a combined MI/IPW approach should be preferred to handle intended and unintended missing data in case-cohort studies with binary outcomes.

Risk factors for suicide one year after discharge from hospitalization for physical illness in Denmark

While suicide risk following psychiatric hospitalization has been studied extensively, risk following hospitalization for physical illness is less well understood. We used random forests to examine risk factors for suicide in the year following physical illness hospitalization in Denmark. In this case-cohort study, suicide cases were all individuals who died by suicide within one year of a hospitalization for a physical illness (n = 4563) and the comparison subcohort was a 5% random sample of individuals living in Denmark on January 1, 1995 who had a hospitalization for a physical illness between January 1, 1995 and December 31, 2015 (n = 177,664). We used random forests to examine identify the most important predictors of suicide stratified by sex. For women, the top 10 most important variables for random forest prediction were all related to psychiatric diagnoses. For men, many physical health conditions also appeared important to suicide prediction. Among the top 10 variables in the variable importance plot for men were influenza, injuries to the head, nervous system surgeries, and cerebrovascular diseases. Suicide prediction after a physical illness hospitalization requires comprehensive consideration of different and multiple factors for each sex.

Physical activity attenuates but does not eliminate coronary heart disease risk amongst adults with risk factors: EPIC-CVD case-cohort study

AIMS

This study aimed to evaluate the association between physical activity and the incidence of coronary heart disease (CHD) in individuals with and without CHD risk factors.

METHODS AND RESULTS

EPIC-CVD is a case-cohort study of 29 333 participants that included 13 582 incident CHD cases and a randomly selected sub-cohort nested within the European Prospective Investigation into Cancer and Nutrition (EPIC) cohort. Self-reported physical activity was summarized using the Cambridge physical activity index (inactive, moderately inactive, moderately active, and active). Participants were categorized into sub-groups based on the presence or the absence of the following risk factors: obesity (body mass index ≥30 kg/m2), hypercholesterolaemia (total cholesterol ≥6.2 mmol/L), history of diabetes, hypertension (self-reported or ≥140/90 mmHg), and current smoking. Prentice-weighted Cox regression was used to assess the association between physical activity and incident CHD events (non-fatal and fatal).Compared to inactive participants without the respective CHD risk factor (referent), excess CHD risk was highest in physically inactive and lowest in moderately active participants with CHD risk factors. Corresponding excess CHD risk estimates amongst those with obesity were 47% [95% confidence interval (CI) 32-64%] and 21% (95%CI 2-44%), with hypercholesterolaemia were 80% (95%CI 55-108%) and 48% (95%CI 22-81%), with hypertension were 80% (95%CI 65-96%) and 49% (95%CI 28-74%), with diabetes were 142% (95%CI 63-260%), and 100% (95%CI 32-204%), and amongst smokers were 152% (95%CI 122-186%) and 109% (95%CI 74-150%).

CONCLUSIONS

In people with CHD risk factors, moderate physical activity, equivalent to 40 mins of walking per day, attenuates but does not completely offset CHD risk.

How many crashes does cellphone use contribute to? Population attributable risk of cellphone use while driving

BACKGROUND

Cellphone distraction is a major contributing factor for traffic crashes, a leading cause of death worldwide. The novel naturalistic driving study (NDS) study with continuously collected in situ driving videos provides an opportunity to accurately estimate the safety impact of cellphone distraction.

METHODS

We apply a case-cohort study design to the Second Strategic Highway Research Program NDS, the largest NDS up-to-date with more than 3400 participants. The data include with 842 level 1-3 crashes and 19,338 randomly selected control driving segments. We propose a partial Population Attributable Risk (PAR) estimator that provides consistent and stable estimation over time and across different driving behaviors.

RESULTS

The US population-adjusted PAR show that 8% of crashes (PAR = 0.08, 95 %CI: [0.06, 0.19]) can be reduced if cellphone distraction were switched to sober, alert, and attentive driving behavior. Young adults (age 20-29 years) and middle-aged drivers (age 30-64 years) each contribute 39% of the population level PAR. Within each age group, the PARs vary substantially from 18% for young adult drivers to 5% for middle-aged drivers. The contribution of cellphone visual-manual tasks to crashes is more than 4 times larger than cellphone talking and accounts for 87.5% of cellphone-related crashes (PAR = 0.07).

CONCLUSIONS

Cellphone distraction contributes to a considerable part of crashes. Young drivers are more susceptible to the influence of cellphone distraction and visual-manual distraction accounts for the majority of cellphone-related crashes.

Plasma perfluoroalkyl substance exposure and incidence risk of breast cancer: A case-cohort study in the Dongfeng-Tongji cohort

Experimental studies have suggested perfluoroalkyl substances (PFASs) as mammary toxicants, but few studies evaluated the prospective associations of PFASs with breast cancer risk. We performed a case-cohort study within the Dongfeng-Tongji cohort, including incident breast cancer cases (n = 226) and a random sub-cohort (n = 990). Baseline plasma concentrations of four perfluorinated carboxylic acids (PFCAs) [perfluorooctanoic acid (PFOA), perfluorononanoic acid (PFNA), perfluorodecanoic acid (PFDA), and perfluoroheptanoic acid (PFHpA)] and two perfluorinated sulfonic acids (PFSAs) [perfluorooctane sulfonic acid (PFOS) and perfluorohexane sulfonic acid (PFHxS)] were measured. Barlow-weighted Cox regression models revealed that each 1-unit increase in ln-transformed PFOA and PFHpA was associated with a separate 35% and 20% elevated incident risk of breast cancer [HR(95%CI) = 1.35(1.03, 1.78) and 1.20(1.02, 1.40), respectively], which were also significant among postmenopausal females [HR(95%CI) = 1.34(1.01, 1.77) and 1.23 (1.02, 1.48), respectively]. Quantile g-computation analysis observed a 19% increased incident risk of breast cancer along with each simultaneous quartile increase in all ln-transformed PFCA concentrations [HR(95%CI) = 1.19(1.01, 1.41)], with PFOA accounting for 56% of the positive effect. Our findings firstly revealed the impact of short-chain PFHpA on increased incident risk of breast cancer, suggested exposure to PFASs as a risk factor for breast cancer, and shed light on breast cancer prevention by regulating PFASs as a chemical class.

Predictors of nonfatal suicide attempts within 30 days of discharge from psychiatric hospitalization: Sex-specific models developed using population-based registries

BACKGROUND

Risk for nonfatal suicide attempts is heightened in the month after psychiatric hospitalization discharge. Investigations of factors associated with such attempts are limited.

METHODS

We conducted a case-subcohort study using data from Danish medical, administrative, and social registries to develop sex-specific risk models using two machine learning methods: classification trees and random forests. Cases included individuals who received a diagnostic code for a nonfatal suicide attempt within 30 days of discharge following a psychiatric hospitalization between January 1, 1995 and December 31, 2015 (n = 3166, 56.5% female). The comparison subcohort consisted of a 5% random sample of individuals living in Denmark (n = 24,559, 51.3% female) on January 1, 1995 who had a psychiatric hospitalization during the study period.

RESULTS

Histories of self-poisoning, substance-related disorders, and eating disorders were important predictors of nonfatal suicide attempt among women, with notable interactions observed between age, self-poisoning history, and other characteristics (e.g., medication use). Self-poisoning, substance-related disorders, and severe stress reactions were among the most important variables for men, with key interactions noted between self-poisoning history, age, major depressive disorder diagnosis, and prescription classes.

LIMITATIONS

Findings are based on Danish administrative data, which may be subject to inaccuracies, missingness, etc. It is unclear whether results would generalize to other populations.

CONCLUSIONS

Markers of behavioral dysregulation were important predictors of nonfatal suicide attempts in the 30 days after psychiatric hospitalization discharge for both sexes. Examining risk markers for nonfatal suicide attempt following discharge is important to enhance support for this vulnerable population.

Urinary cadmium and stroke - a case-cohort study in Danish never-smokers

BACKGROUND

and Purpose: Cadmium has been associated with risk of cardiovascular events, including stroke. Human cadmium exposure occurs primarily through diet and tobacco smoke. Recent cohort studies have found an association with stroke, but residual confounding from smoking, could not be ruled out. We therefore conducted a case-cohort study to evaluate whether cadmium is associated with stroke in never-smokers.

METHODS

The Danish Diet Cancer and Health cohort consists of Danes 50-64 years old, recruited in 1993-1997. From never-smoking cohort members without previous cancer or stroke we sampled a sub-cohort of 1200 persons. We also identified all (n = 534) cases in the cohort with a validated stroke diagnosis between baseline and 2009. We quantified cadmium and creatinine concentrations from baseline urine samples and used cadmium per creatinine as our main exposure metric. We used Cox proportional hazards models to estimate hazard ratios (HRs) with age as time scale and adjusting for BMI, education and urinary cotinine with and without stratification by sex.

RESULTS

The median urinary cadmium concentration was 0.21 μg cadmium/g creatinine in cases and 0.19 μg/g in the sub-cohort. The majority (83%) of stroke cases were diagnosed with ischemic stroke. The HR for stroke in the highest quartile of exposure (median 0.44 μg/g creatinine) was 1.11 (95% CI: 0.79-1.54) compared with the lowest quartile (median 0.10 μg/g creatinine). The HR per inter quartile range (IQR, 0.19 μg/g creatinine) was 1.02 (95% CI: 0.92-1.12). Among men, the HR per IQR higher levels of cadmium (0.16 μg/g creatinine) was 1.18 (95% CI: 0.92-1.52), and 1.00 (95% CI: 0.89-1.12) among women. Adjusting for creatinine or using osmolality instead of creatinine standardization generally attenuated observed relationships.

CONCLUSIONS

Our results do not support that low levels of cadmium exposure among never-smokers are strongly associated with risk of stroke, although results varied somewhat by sex and method of accounting for urinary dilution.

Urine cadmium and acute myocardial infarction among never smokers in the Danish Diet, Cancer and Health cohort

Cadmium exposure has been associated with cardiovascular disease. Cigarette smoking is a key source of cadmium exposure and thus a potential confounder in observational studies of environmental cadmium and cardiovascular disease that include tobacco smokers. We leveraged up to 20 years of follow-up in the Danish Diet, Cancer and Health cohort to test the hypothesis that cadmium exposure is associated with acute myocardial infarction (AMI) among people who never smoked. Between 1993 and 1997, 19,394 never-smoking participants (ages 50-64 years) were enrolled and provided a urine sample. From this sample, we randomly selected a subcohort of 600 males and 600 females. We identified 809 AMI cases occurring between baseline and the end of 2015 using the Danish National Patient Registry. We quantified cadmium, creatinine, and osmolality in baseline urine samples. Using an unweighted case-cohort approach, we estimated adjusted hazard ratios (aHR) for AMI in Cox proportional hazards models with age as the time axis. Participants had relatively low concentrations of urinary cadmium, as expected for never smokers (median = 0.20; 25th, 75th = 0.13, 0.32 μg cadmium/g creatinine). We did not find strong evidence to support an association between higher urinary cadmium and AMI when comparing the highest versus lowest quartile (aHR = 1.16; 95% CI: 0.86 - 1.56) and per IQR increment in cadmium concentration (aHR = 1.02; 95% CI: 0.93 - 1.12). Results were not materially different across strata defined by sex. Results were generally similar using creatinine or osmolality to account for differences in urine dilution. While cadmium exposure has been identified as a risk factor for cardiovascular disease, we did not find strong evidence that urinary cadmium at relatively low-levels is associated with AMI among people who have never smoked.

Assessing risk model calibration with missing covariates

When validating a risk model in an independent cohort, some predictors may be missing for some subjects. Missingness can be unplanned or by design, as in case-cohort or nested case-control studies, in which some covariates are measured only in subsampled subjects. Weighting methods and imputation are used to handle missing data. We propose methods to increase the efficiency of weighting to assess calibration of a risk model (i.e. bias in model predictions), which is quantified by the ratio of the number of observed events, $\mathcal{O}$, to expected events, $\mathcal{E}$, computed from the model. We adjust known inverse probability weights by incorporating auxiliary information available for all cohort members. We use survey calibration that requires the weighted sum of the auxiliary statistics in the complete data subset to equal their sum in the full cohort. We show that a pseudo-risk estimate that approximates the actual risk value but uses only variables available for the entire cohort is an excellent auxiliary statistic to estimate $\mathcal{E}$. We derive analytic variance formulas for $\mathcal{O}/\mathcal{E}$ with adjusted weights. In simulations, weight adjustment with pseudo-risk was much more efficient than inverse probability weighting and yielded consistent estimates even when the pseudo-risk was a poor approximation. Multiple imputation was often efficient but yielded biased estimates when the imputation model was misspecified. Using these methods, we assessed calibration of an absolute risk model for second primary thyroid cancer in an independent cohort.

Brand-specific rates of pertussis disease among Wisconsin children given 1-4 doses of pertussis Vaccine, 2010-2014

BACKGROUND

Acellular pertussis vaccines were initially licensed based on placebo-controlled efficacy trials, but such trials are no longer ethical. The effectiveness of current pertussis vaccines among properly vaccinated children <5 years is so high that a randomized trial is infeasible. Fluctuations in pertussis incidence and characteristics of the US vaccine marketplace make selection of suitable controls for a case-control study problematic. To satisfy an FDA requirement to evaluate rates of pertussis following licensure of Pentacel® vaccine, we used a case-cohort study design with a novel method for characterizing the cohort population.

METHODS

This prospective, observational study was conducted in Wisconsin from 2010 to 2014 among Wisconsin residents <60 months of age who received ≤four doses of pertussis vaccine (surveillance population). Cases were identified by the Wisconsin Division of Public Health. Characteristics and pertussis vaccinations of the surveillance population were estimated by ongoing random telephonic survey. The primary objective was to determine rates of pertussis disease among those who received only Pentacel vaccine (Group 1) vs those who received a single brand of vaccine other than Pentacel vaccine (Group 2).

RESULTS

1195 pertussis cases were identified. It was estimated that the surveillance population accrued a total of 1,133,403 person-years (Group 1, 39%; Group 2, 41%; Group 3 [those not in Group 1 or Group 2], 20%). Pertussis rates were similar in Group 1 (98.9/100,000) and Group 2 (96.2/100,000); rate ratios were 1.03 (unadjusted; 90% CI, 0.92-1.15) and 0.99 (adjusted; 90% CI, 0.89-1.12). Persons with one or more delayed vaccinations had a 66% higher risk of pertussis (90% CI, 39-96%).

DISCUSSION

Pertussis protection was not found to differ for recipients of the newly licensed vs other available pertussis vaccines. Delayed vaccination substantially increased risk of pertussis. Sample survey methodology was able to characterize the study cohort and enable an otherwise-infeasible study. Clinical Trial Registry number: ClinicalTrials.gov, NCT01129362.

Biomarker-defined pathways for incident type 2 diabetes and coronary heart disease-a comparison in the MONICA/KORA study

Background

Biomarkers may contribute to our understanding of the pathophysiology of various diseases. Type 2 diabetes (T2D) and coronary heart disease (CHD) share many clinical and lifestyle risk factors and several biomarkers are associated with both diseases. The current analysis aims to assess the relevance of biomarkers combined to pathway groups for the development of T2D and CHD in the same cohort.

Methods

Forty-seven serum biomarkers were measured in the MONICA/KORA case-cohort study using clinical chemistry assays and ultrasensitive molecular counting technology. The T2D (CHD) analyses included 689 (568) incident cases and 1850 (2004) non-cases from three population-based surveys. At baseline, the study participants were 35–74 years old. The median follow-up was 14 years. We computed Cox regression models for each biomarker, adjusted for age, sex, and survey. Additionally, we assigned the biomarkers to 19 etiological pathways based on information from literature. One age-, sex-, and survey-controlled average variable was built for each pathway. We used the R²_PM coefficient of determination to assess the explained disease risk.

Results

The associations of many biomarkers, such as several cytokines or the iron marker soluble transferrin receptor (sTfR), were similar in strength for T2D and CHD, but we also observed important differences. Lipoprotein (a) (Lp(a)) and N-terminal pro B-type natriuretic peptide (NT-proBNP) even demonstrated opposite effect directions. All pathway variables together explained 49% of the T2D risk and 21% of the CHD risk. The insulin-like growth factor binding protein 2 (IGFBP-2, IGF/IGFBP system pathway) best explained the T2D risk (about 9% explained risk, independent of all other pathway variables). For CHD, the myocardial-injury- and lipid-related-pathways were most important and both explained about 4% of the CHD risk.

Conclusions

The biomarker-derived pathway variables explained a higher proportion of the T2D risk compared to CHD. The ranking of the pathways differed between the two diseases, with the IGF/IGFBP-system-pathway being most strongly associated with T2D and the myocardial-injury- and lipid-related-pathways with CHD. Our results help to better understand the pathophysiology of the two diseases, with the ultimate goal of pointing out targets for lifestyle intervention and drug development to ideally prevent both T2D and CHD development.

Adipose tissue content of alpha-linolenic acid and development of peripheral artery disease: a Danish case-cohort study

PURPOSE

The aim of this study was to investigate the association between adipose tissue content of the plant-derived n-3 fatty acid, alpha-linolenic acid, and the rate of incident peripheral artery disease (PAD).

METHODS

We conducted a case-cohort study nested within the Danish Diet, Cancer and Health cohort (n = 57,053), which was established between 1993 and 1997. Potential PAD cases were identified using linkage with The Danish National Patient Register and all potential cases were validated. Adipose tissue samples from the buttock were collected at baseline and fatty acid composition was determined in cases and in a random sample (n = 3500) from the cohort by gas chromatography. Statistical analyses were performed using weighted Cox regression allowing for different baseline hazards among sexes.

RESULTS

During a median of 13.5 years of follow-up, we identified 863 PAD cases with complete information. The median adipose tissue content of ALA in the sub-cohort (n = 3197) was 0.84% (interquartile range 0.73-0.94%) of total fatty acids. In multivariate analyses including adjustment for established risk factors, we observed a U-shaped association between ALA in adipose tissue and rate of PAD, but the association was not statistically significant (P = 0.131). Similar pattern of associations were observed between ALA content in adipose tissue and the rate of PAD among men and women.

CONCLUSIONS

We found indications of a U-shaped association between adipose tissue content of ALA and the rate of PAD, but the association was not statistically significant.

Racial disparities in end-stage renal disease in a high-risk population: the Southern Community Cohort Study

Introduction

The Southern Community Cohort Study is a prospective study of low socioeconomic status (SES) blacks and whites from the southeastern US, where the burden of end-stage renal disease (ESRD) and its risk factors are high. We tested whether the 2.4-fold elevated risk of ESRD we previously observed in blacks compared to whites was explained by differences in baseline kidney function.

Methods

We conducted a case-cohort study of incident ESRD cases (n = 737) with stored blood and a probability sampled subcohort (n = 4238) and calculated estimated glomerular filtration rate (eGFR) from serum creatinine. 86% of participants were enrolled from community health centers in medically underserved areas and 14% from the general population in 12 states in the southeastern United States. Incident ESRD after entry into the cohort was ascertained by linkage of the cohort with the US Renal Data System (USRDS).

Results

Median (25th, 75th percentile) eGFR at baseline was 63.3 (36.0, 98.2) ml/min/1.73m² for ESRD cases and 103.2 (86.0, 117.9) for subcohort. Black ESRD cases had higher median (25th, 75th) eGFR [63.3 (35.9, 95.9)] compared to whites [59.1 (39.4, 99.2)]. In multivariable Cox models accounting for sampling weights, baseline eGFR was a strong predictor of ESRD risk, and an interaction with race was detected (P = 0.029). The higher ESRD risk among blacks relative to whites persisted (hazard ratio: 2.58; 95% confidence interval: 1.65, 4.03) after adjustment for eGFR.

Conclusion

In this predominantly lower SES cohort, the racial disparity in ESRD risk is not explained by differences in baseline kidney function.

Electronic supplementary material

The online version of this article (10.1186/s12882-019-1502-z) contains supplementary material, which is available to authorized users.

Association between plasma CD36 levels and incident risk of coronary heart disease among Danish men and women

BACKGROUND AND AIMS

CD36 is a cholesterol receptor involved in the uptake of oxidized low-density lipoprotein cholesterol and development of atherosclerotic plaques. Cross-sectional studies have shown correlations between plasma CD36 and atherosclerosis but no prospective study has examined the association yet. We prospectively examined the association between plasma CD36 levels and risk of incident coronary heart disease (CHD) in a Danish population.

METHODS

Plasma CD36 levels were measured in a case-cohort study nested within the Danish population-based cohort, the Diet, Cancer and Health Study. A total of 1963 incident CHD events occurred between baseline (1993-1997) and 2008, and a sub-cohort of 1759 participants were randomly selected as reference. Cox proportional hazard regression models were used to compute the hazard ratio (HR) and corresponding 95% confidence interval (CI).

RESULTS

After adjusting for CHD risk factors, including history of hypercholesterolemia and diabetes, elevated plasma CD36 levels were not associated with higher CHD risk in the total population, and the HR comparing the highest versus lowest tertile of CD36 levels was 1.02 (95% CI: 0.84-1.23). High CD36 levels were only found to be associated with risk of CHD in combination with prevalent diabetes (HR = 2.83, 95% CI: 1.08-7.45) vs. the joint reference group of lowest CD36 tertile and no diabetes.

CONCLUSIONS

Plasma CD36 levels were not predictive of CHD risk in the general population.

Serum mercury concentration and the risk of ischemic stroke: The REasons for Geographic and Racial Differences in Stroke Trace Element Study

BACKGROUND

Although biologically plausible, epidemiological evidence linking exposure to methylmercury with increased risk of ischemic stroke is limited. The effects of methylmercury may be modified by selenium, which is an anti-oxidant that often co-exists with mercury in fish.

OBJECTIVES

To examine the association between serum mercury levels with the incidence of ischemic stroke and to explore the possible effect modifications by serum selenium levels and demographic and geographic factors.

METHODS

A case-cohort study was designed nested in the REasons for Geographic and Racial Differences in Stroke cohort, including 662 adjudicated incident cases of ischemic stroke and 2494 participants in a randomly selected sub-cohort. Serum mercury was measured using samples collected at recruitment. Multivariable-adjusted hazard ratios (HRs) and the corresponding 95% confidence intervals (CIs) were estimated using the Barlow-weighting method for the Cox proportional hazards regression model.

RESULTS

No statistically significant association was observed between serum mercury concentration and the incidence of ischemic stroke (the highest vs. lowest quintile of mercury levels: HR = 0.82; 95% CI = 0.55-1.22; P for linear trend = 0.42). Sex (P for interaction = 0.06), but not serum selenium levels, modified the association; a more evident trend toward lower incidence of ischemic stroke with higher mercury levels was observed among women.

CONCLUSION

This study does not support an association between mercury and the incidence of ischemic stroke within a population with low-to-moderate level of exposure. Further studies are needed to explore the possibility of mercury-induced ischemic stroke toxicity in other populations at higher exposure levels.

Analysis of multiple survival events in generalized case-cohort designs

Generalized case-cohort design has been proposed to assess the effects of exposures on survival outcomes when measuring exposures is expensive and events are not rare in the cohort. In such design, expensive exposure information is collected from both a (stratified) randomly selected subcohort and a subset of individuals with events. In this article, we consider extension of such design to study multiple types of survival events by selecting a proportion of cases for each type of event. We propose a general weighting scheme to analyze data. Furthermore, we examine the optimal choice of weights and show that this optimal weighting yields much improved efficiency gain both asymptotically and in simulation studies. Finally, we apply our proposed methods to data from the Atherosclerosis Risk in Communities study.

Multiple imputation of missing data in nested case-control and case-cohort studies

The nested case-control and case-cohort designs are two main approaches for carrying out a substudy within a prospective cohort. This article adapts multiple imputation (MI) methods for handling missing covariates in full-cohort studies for nested case-control and case-cohort studies. We consider data missing by design and data missing by chance. MI analyses that make use of full-cohort data and MI analyses based on substudy data only are described, alongside an intermediate approach in which the imputation uses full-cohort data but the analysis uses only the substudy. We describe adaptations to two imputation methods: the approximate method (MI-approx) of White and Royston (2009) and the “substantive model compatible” (MI-SMC) method of Bartlett et al. (2015). We also apply the “MI matched set” approach of Seaman and Keogh (2015) to nested case-control studies, which does not require any full-cohort information. The methods are investigated using simulation studies and all perform well when their assumptions hold. Substantial gains in efficiency can be made by imputing data missing by design using the full-cohort approach or by imputing data missing by chance in analyses using the substudy only. The intermediate approach brings greater gains in efficiency relative to the substudy approach and is more robust to imputation model misspecification than the full-cohort approach. The methods are illustrated using the ARIC Study cohort. Supplementary Materials provide R and Stata code.

Associations of serum 25-hydroxyvitamin D level with incidence of lung cancer and histologic types in Norwegian adults: a case-cohort analysis of the HUNT study

Previous prospective studies have shown inconsistent associations between serum 25-hydroxyvitamin D [25(OH)D] level and lung cancer incidence. The aim of the present study was to explore the associations of serum 25(OH)D levels with incidence of lung cancer overall and different histologic types. We performed a population-based prospective case-cohort study including 696 incident lung cancer cases and 5804 individuals in a subcohort who participated in the second survey of the Nord-Trøndelag Health Study in Norway. Cox proportional hazards regression models counting for the case-cohort design were used to estimate hazard ratios (HRs) with 95% confidence interval (CIs) for lung cancer overall or histologic types in relation to serum 25(OH)D levels. Compared with the fourth season-specific quartile of 25(OH)D (median 68.0 nmol/L), lower 25(OH)D levels were not associated with the incidence of overall, small or squamous cell lung cancer. However, the risk of adenocarcinoma was lower in the second and third quartiles (median 39.9 and 51.5 nmol/L) compared with the fourth quartile, with HRs of 0.63 (95% CI 0.41-0.98) and 0.58 (0.38-0.88), respectively. The associations of lower levels of 25(OH)D with a reduced risk of adenocarcinoma were only observed in the overweight/obese subjects [HRs for second and third quartiles: 0.40 (0.22-0.72) and 0.50 (0.27-0.92)] but not in the normal weight subjects [HRs: 0.95 (0.52-1.75) and 0.60 (0.32-1.10)]. Serum 25(OH)D levels were not associated with the risk of lung cancer in general. The observation that lower 25(OH)D levels were associated with a lower risk of adenocarcinoma should be interpreted with caution.

Estrogen receptor quantitative measures and breast cancer survival

PURPOSE

While the estrogen receptor (ER) is the single most widely used biomarker to evaluate breast cancer outcomes, aspects of ER marker biology remain poorly understood. We sought to determine whether quantitative measures of ER, such as protein expression and intensity, were associated with survival, or with survival disparities experienced by Hispanic women.

METHODS

A case-cohort study included a 15% random sample of invasive breast cancer cases diagnosed from 1997 to 2009 in six New Mexico counties and all deaths due to breast cancer-related causes. Pathology reports and tissue microarrays served as sources of ER information. Analyses were restricted to women with ≥1% ER immunohistochemical staining. Hazard ratios (HR) and 95% confidence intervals (CI) for breast cancer death were estimated using Cox proportional hazards models.

RESULTS

Included women represented 4336 ER+ breast cancer cases and 448 deaths. Median follow-up was 93 months. ER percent expression was not associated with breast cancer survival after adjustment for standard prognostic factors (p trend = 0.76). ER intensity remained a strong and independent risk factor for breast cancer survival in multivariate analyses: Women whose tumors expressed ER at intensity = 2 (HR 0.6; 95% CI 0.4-1.0) or 3 (HR 0.5; 95% CI 0.2-0.9) had a reduced risk of breast cancer mortality, compared to ER intensity = 1 (p trend = 0.02). Neither ER protein expression nor intensity influenced Hispanic survival disparities.

CONCLUSIONS

Estrogen receptor percent positive staining is not independently related to breast cancer survival after adjustment for other survival-related factors. ER intensity, in contrast, demonstrates promise for prognostic utility.

Neonatal Vitamin D Levels in Relation to Risk of Overweight at 7 Years in the Danish D-Tect Case-Cohort Study

BACKGROUND

Vitamin D level in pregnancy may be associated with risk of overweight in the offspring later in life.

METHODS

In a case-cohort study based on Danish biobanks and registers we examined the association between 25-hydroxy-vitamin D (25(OH)D) level at birth and overweight at 7 years. Cases of overweight (n = 871) were randomly selected among 7-year-old children from the Copenhagen School Health Records Register (CSHRR) with a BMI above the 90th percentile. The cohort (n = 1,311) was a random sample selected among all Danish children born during the same period. Neonatal 25(OH)D was measured in dried blood spots.

RESULTS

25(OH)D3 exhibited the expected seasonal variation. Median level of 25(OH)D3 was 20.6 (11.9-33.3) nmol/l in the overweight group and 23.4 (13.5-34.3) nmol/l in the cohort. We found no association between neonatal 25(OH)D3 level and risk of overweight at age 7 years, neither in the crude model (OR (CI) 1.00 (0.99; 1.00)) nor in a model adjusted for maternal ethnicity, educational level, civil status, parity, season and year of birth, and offspring ponderal index (OR (CI) 1.00 (0.99; 1.01)).

CONCLUSION

Risk of overweight at 7 years of age was not associated with vitamin D level at birth.

More efficient estimators for case-cohort studies

The case-cohort study design, used to reduce costs in large cohort studies, is a random sample of the entire cohort, named the subcohort, augmented with subjects having the disease of interest but not in the subcohort sample. When several diseases are of interest, several case-cohort studies may be conducted using the same subcohort, with each disease analyzed separately, ignoring the additional exposure measurements collected on subjects with the other diseases. This is not an efficient use of the data, and in this paper, we propose more efficient estimators. We consider both joint and separate analyses for the multiple diseases. We propose an estimating equation approach with a new weight function, and we establish the consistency and asymptotic normality of the resulting estimator. Simulation studies show that the proposed methods using all available information gain efficiency. We apply our proposed method to the data from the Busselton Health Study.