The investigation and correction of recall bias for an ordinal response in a case-control study

Consider a case-control study designed to investigate the possible association between development of a particular disease and the value of a putative risk factor measured on an ordinal scale. Let E denote a subject's true risk factor value and let E* denote a subject's recorded risk factor value. Misclassification bias occurs if conclusions reached regarding the relationship between disease status and E* do not also apply to the relationship between disease status and E. We propose a model for the conditional probability distribution of E* given E. We show how the model may be used to investigate misclassification bias in a validation study where measurements of E* and E are available for both cases and controls and apply the methods developed to data from a test-retest study of recall bias in the context of screening for hypertension. We also consider a situation where the validation study is carried out on a subset of the subjects within a larger case-control study. In that case, values for E* are available for all subjects but values for E are available only for those subjects included in the validation study. We show how correct likelihood-based inference concerning association between disease status and risk factor value may be carried out using all of the available data. A Monte Carlo study shows how the inclusion of a validation study leads to a correction of recall bias problems at the cost of an increased standard error for the estimated association parameter.     

Bias in association studies resulting from gene-environment interactions and competing risks

The etiology, particularly the genetic basis, of multifactorial late-onset diseases is the subject of many genetic epidemiologic studies. The authors' aim in this paper was to investigate the circumstances under which competing risks can lead to bias in studies of genetic susceptibility to late-onset diseases. The authors used a model built in an epidemiologic framework to show that when a genetic risk factor and an environmental risk factor interact to cause a frequent competing risk of death, the measure of the association between the disease under investigation and the genetic risk factor will be biased if the environmental risk factor is also associated with the latter disease and is omitted from the analysis. This is an example of confounding bias, and it is the consequence of an association between the genetic risk factor and the environmental risk factor that appears over time. Numerical examples show that under certain conditions this bias can be substantial. The authors present several patterns of association in favor of such a bias. Because competing risks of death are likely to be present in older subjects, researchers studying the etiology of late-onset diseases should be aware of the possibility of this bias.     

Vitamin D and cardiovascular disease: Systematic review and meta-analysis of prospective studies

BackgroundLow serum 25-hydroxyvitamin D (25-OH-D) has recently been linked to cardiovascular diseases. This review summarizes evidence from prospective studies evaluating the prognostic value of 25-OH-D for cardiovascular disease incidence and mortality.MethodA systematic literature search in EMBASE and Pubmed-Medline databases was performed until November 2009. Prospective studies published in English were selected reporting estimates for the association of 25-OH-D with primary or secondary cardiovascular event incidence or mortality in the general population or subjects with prevalent cardiovascular disease. Pooled risk estimators were derived by meta-analysis using a random effects model approach.ResultsFour incidence and five independent mortality studies were included. Two incidence and three mortality studies reported a two- to five-fold risk increase for both outcomes in subjects with lower 25-OH-D, while the others did not detect a significant association. Meta-analysis supported the existence of an inverse association.ConclusionData from prospective investigations suggest an inverse association between 25-OH-D and cardiovascular risk. However, given the heterogeneity and small number of longitudinal studies, more research is needed to corroborate a potential prognostic value of 25-OH-D for cardiovascular disease incidence and mortality.     

Estimating age at onset distributions: The bias from prevalent cases and its impact on risk estimation

Since many disorders have a variable age at onset, knowing the age at onset distribution of a disease facilitates epidemiologic analyses in several ways. The age at onset distribution is commonly used to estimate morbidity risks or the recurrence risks in genetic counseling. Unfortunately, estimation of a disease's age at onset distribution is not straightforward. The observed age at onset distribution obtained from prevalent cases is usually used in these epidemiologic analyses. Through simulation studies, we show that, in certain situations, the observed age at onset distribution has a non-negligible downward bias. This bias can lead to a substantial underestimation of the morbidity risk or the recurrence risk. The simulations also demonstrate that a non-parametric approach for correcting the age at onset distribution works well even when mortality increases after onset. The results have implications for diseases that have adult onset and/or increased mortality after onset. We suggest that researchers should use corrected age at onset distributions, rather than relying on observed distributions, in the calculation of either morbidity risks or recurrence risks. © 1993 Wiley-Liss. Inc.     

Nondifferential disease misclassification may bias incidence risk ratios away from the null

BACKGROUND AND OBJECTIVE: When estimating incidence risk ratios in follow-up studies, subjects testing positive for the disease at baseline are excluded. Although the effect of disease misclassification on estimated incidence risk ratios has otherwise been extensively explored, the effect of disease misclassification at baseline has not previously been analyzed. STUDY DESIGN AND SETTING: The design was theoretical calculations assuming dichotomous disease and a follow-up study with a baseline and a follow-up examination, analyzed using cumulative incidence. Calculations consider nondifferential misclassification of disease mainly at baseline, but no misclassification of exposure. RESULTS: Nondifferential misclassification of disease at baseline can lead to bias either away or toward null in estimated cumulative incidence risk ratios. This bias is mainly a function of sensitivity at baseline, because imperfect sensitivity leads to failure to exclude all diseased subjects from the follow-up. Imperfect specificity at baseline has less effect. Bias is increased with high true prevalence of disease and low true incidence. Bias is also increased with large differences in true risk ratios at baseline and at follow-up, because observed incidence risk ratios in the presence of misclassification reflect both the true association at baseline and at follow-up. CONCLUSION: Nondifferential disease misclassification at baseline examination of a follow-up study can lead to over- or underestimation of the cumulative incidence risk ratios. The bias can be substantial for disease with low incidence and high prevalence, such as asthma or myocardial infarction. The results underscore the need to select a highly sensitive test for disease at baseline to exclude all diseased subjects from the follow-up.     

The influence of prevalent cohort bias in the association between periodontal disease progression and incident coronary heart disease

PurposeIn longitudinal studies, the onset of the index condition (e.g. exposure) does not always coincide with the start of a study's observation period, leading to the possibility of bias in estimation that derives from studying prevalent exposure rather than new exposure. We investigate the possible role of this bias in the relationship between periodontitis progression and coronary heart disease (CHD) among a cohort of men participating in the Veterans Administration Dental Longitudinal Study.MethodsAt baseline, there were 298 men with existing (i.e., prevalent) periodontitis. During follow-up, routine dental inspection identified 163 new (i.e., incident) cases of periodontitis. Change in mean alveolar bone loss score (MBLS) served as the measure of disease progression. Tabular analyses were performed to obtain crude, stratified, and adjusted measures of the association for periodontitis cases overall and separately for prevalent and incident cases. Potential bias was evaluated by comparing estimates across these subcohorts.ResultsAmong all periodontitis cases, increasing MBLS was associated with increasing risk of CHD event. Subdividing periodontal cases into new and prevalent cases revealed that the relationship was most pronounced among incident periodontitis cases (incident rate ratio for MBLS change >0.5 = 5.4), compared with prevalent cases (incident rate ratio for MBLS change >0.5 = 2.5).ConclusionsStudying prevalent cases of periodontitis underestimates the association between incidence periodontitis and CHD.     

Problems in the use of dead controls in case-control studies. II. Effect of excluding certain causes of death

A recent comparison of 493 dead and 697 living controls from a case-control study of cancer in the Minneapolis-St. Paul area showed that the dead controls of both sexes were reported to have been significantly heavier consumers of cigarettes, hard liquor, beer, and certain drugs, and to have had more adulthood diseases than living controls. The present analysis examines the effect of excluding causes of death associated with those exposures found in excess in the dead controls. Exclusion of individuals with smoking-related causes of death reduced but did not eliminate the excess of cigarette smokers among the dead controls. Deletion of individuals with alcohol-related causes of death only slightly reduced the excess among dead controls. Adjustment for cigarette smoking, however, nearly eliminated the association with alcohol consumption, particularly among males. For certain adulthood diseases and medications, the exclusion of individuals with exposure-associated causes of death also virtually eliminated the excesses found in the dead controls when compared with the living controls. Thus, it appears that even after extensive exclusion of smoking-related causes of death, the association between dead controls and cigarette smoking still remains, and the use of dead controls in case-control studies where cigarette smoking is the risk factor being evaluated may lead to a biased underestimated of risk. For the other exposures found in significant excess among the dead controls, the exclusion of exposure-related causes of death and proper adjustment for confounders may eliminate much or all of the excess.     

Genome‐wide association scans for secondary traits using case‐control samples

Genome‐wide association studies (GWAS) require considerable investment, so researchers often study multiple traits collected on the same set of subjects to maximize return. However, many GWAS have adopted a case‐control design; improperly accounting for case‐control ascertainment can lead to biased estimates of association between markers and secondary traits. We show that under the null hypothesis of no marker‐secondary trait association, naïve analyses that ignore ascertainment or stratify on case‐control status have proper Type I error rates except when both the marker and secondary trait are independently associated with disease risk. Under the alternative hypothesis, these methods are unbiased when the secondary trait is not associated with disease risk. We also show that inverse‐probability‐of‐sampling‐weighted (IPW) regression provides unbiased estimates of marker‐secondary trait association. We use simulation to quantify the Type I error, power and bias of naïve and IPW methods. IPW regression has appropriate Type I error in all situations we consider, but has lower power than naïve analyses. The bias for naïve analyses is small provided the marker is independent of disease risk. Considering the majority of tested markers in a GWAS are not associated with disease risk, naïve analyses provide valid tests of and nearly unbiased estimates of marker‐secondary trait association. Care must be taken when there is evidence that both the secondary trait and tested marker are associated with the primary disease, a situation we illustrate using an analysis of the relationship between a marker in FGFR2 and mammographic density in a breast cancer case‐control sample. Genet. Epidemiol. 33:717–728, 2009. © 2009 Wiley‐Liss, Inc.     

Relation of serum ascorbic acid to mortality among US adults.

PURPOSE: To examine the relation between serum ascorbic acid (SAA), a marker of dietary intake (including supplements), and cause-specific mortality. SUBJECTS AND METHODS: We analyzed data from a probability sample of 8,453 Americans age > or = 30 years at baseline enrolled in the Second National Health and Nutrition Examination Survey (NHANES II), who were followed for mortality endpoints. We calculated relative hazard ratios as measures of disease association comparing the mortality rates in three biologically relevant SAA categories. RESULTS: Participants with normal to high SAA levels had a marginally significant 21% to 25% decreased risk of fatal cardiovascular disease (CVD) (p for trend = 0.09) and a 25% to 29% decreased risk of all-cause mortality (p for trend <0.001) compared to participants with low levels. Because we determined that gender modified the association between SAA levels and cancer death, we analyzed these associations stratified by gender. Among men, normal to high SAA levels were associated with an approximately 30% decreased risk of cancer deaths, whereas such SAA levels were associated with an approximately two-fold increased risk of cancer deaths among women. This association among women persisted even after adjustment for baseline prevalent cancer and exclusion for early cancer death or exclusion for prevalent cancer. CONCLUSIONS: Low SAA levels were marginally associated with an increased risk of fatal CVD and significantly associated with an increased risk for all-cause mortality. Low SAA levels were also a risk factor for cancer death in men, but unexpectedly were associated with a decreased risk of cancer death in women. If the association between low SAA levels and all-cause mortality is causal, increasing the consumption of ascorbic acid, and thereby SAA levels, could decrease the risk of death among Americans with low ascorbic acid intakes.     

A method to visualize and adjust for selection bias in prevalent cohort studies

Selection bias and confounding are concerns in cohort studies where the reason for inclusion of subjects in the cohort may be related to the outcome of interest. Selection bias in prevalent cohorts is often corrected by excluding observation time and events during the first time period after inclusion in the cohort. This time period must be chosen carefully-long enough to minimize selection bias but not too long so as to unnecessarily discard observation time and events. A novel method visualizing and estimating selection bias is described and exemplified by using 2 real cohort study examples: a study of hepatitis C virus infection and a study of monoclonal gammopathy of undetermined significance. The method is based on modeling the hazard for the outcome of interest as a function of time since inclusion in the cohort. The events studied were "hospitalizations for kidney-related disease" in the hepatitis C virus cohort and "death" in the monoclonal gammopathy of undetermined significance cohort. Both cohorts show signs of considerable selection bias as evidenced by increased hazard in the time period after inclusion in the cohort. The method was very useful in visualizing selection bias and in determining the initial time period to be excluded from the analyses.     

Chronic obstructive pulmonary disease and the risk of cardiovascular diseases

Previous large epidemiological studies reporting on the association between chronic obstructive pulmonary disease (COPD) and cardiovascular diseases mainly focussed on prevalent diseases rather than on the incidence of newly diagnosed cardiovascular outcomes. We used the UK-based General Practice Research Database (GPRD) to assess the prevalence and incidence of cardiovascular diseases in COPD patients aged 40–79 between 1995 and 2005, and we randomly matched COPD-free comparison patients to COPD patients. In nested-case control analyses, we compared the risks of developing an incident diagnosis of cardiac arrhythmias, venous thromboembolism, myocardial infarction, or stroke between patients with and without COPD, stratifying the analyses by COPD-severity, using COPD-treatment as proxy for disease severity. We identified 35,772 patients with COPD and the same number of COPD-free patients. Most cardiovascular diseases were more prevalent among COPD patients than among the comparison group of COPD-free patients. The relative risk estimates of developing an incident diagnosis of cardiac arrhythmia (OR 1.19, 95% CI 0.98–1.43), deep vein thrombosis (OR 1.35, 95% CI 0.97–1.89), pulmonary embolism (OR 2.51, 95% CI 1.62–3.87), myocardial infarction (OR 1.40, 95% CI 1.13–1.73), or stroke (OR 1.13, 95% CI 0.92–1.38), tended to be increased for patients with COPD as compared to COPD-free controls. The findings of this large observational study provide further evidence that patients with COPD are at increased risk for most cardiovascular diseases.     

成人睡眠时间与2型糖尿病相关性队列研究的meta分析

目的 系统回顾睡眠时间长短与2型糖尿病（type 2 diabetes mellitus, T2DM）发生相关性队列研究，通过meta分析研究睡眠时间长短与T2DM发生相关性，为预防T2DM提供理论基础。方法 检索PubMed、Web of science、Scopus、Embase、Cochrane Library、ProQuest、CNKI、万方、维普、SinoMed建库至2020年5月，由2位研究者按照纳入与排除标准独立检索、筛选文献、质量评价与提取资料。利用RevMan 5.3进行meta分析、发表偏倚分析等。结果 共纳入17项队列研究，共计737 002名成人。T2DM发生率短睡眠时间组(short sleep duration, SSD)（t≤6 h）4.73%，正常睡眠时间组(normal sleep duration, NSD)（6 h＜t＜9 h）4.39%，长睡眠时间组(long sleep duration, LSD)（t≥9 h）4.99%。meta分析显示SSD与NSD相比患T2DM风险增加（RR = 1.22，95% CI:1.15～1.29，P＜0.001），LSD与NSD相比患T2DM风险增加（RR = 1.26，95% CI:1.15～1.39，P＜0.001）。各研究间敏感性稳定，发表偏倚小。结论 SSD或LSD均增加T2DM发病风险。     

Disease based spectrum bias in referred samples and the relationship between type a behavior and coronary artery disease

Using referred samples to study predictors of disease produces statistical problems that reduce the likelihood of obtaining statistically significant results even when a substantial relationship is present between a risk factor and a disease. The present paper refers to these problems as disease based spectrum (DBS) bias. DBS bias is present when subjects are directed into or excluded from the study sample according to their disease status. For example, healthy individuals are excluded from and diseased individuals are directed into referred samples. Therefore, DBS bias is present in referred samples. Examples from the literature on Type A behavior and coronary artery disease (CAD) are presented to illustrate how DBS bias reduces statistical associations. The results of the current research indicate DBS bias has reduced the association between Type A behavior and CAD in a number of studies reported in recent years. In addition, the present article discusses techniques for assessing and controlling for DBS bias.     

Respiratory morbidity in smokers of low- and high-yield cigarettes

To study the association between smoking cigarettes with a low yield of tar and nicotine (tar less than 15.0 mg per cigarette and nicotine less than 1.0 mg) and respiratory disease, we reviewed the medical records of 4,610 current, regular cigarette smokers and 2,035 persons who had never used any form of tobacco and who were enrolled in a smoking study. In the year after recruitment to the study, the percentage of subjects with pneumonia or influenza was lower in female but not in male smokers of low-yield cigarettes. The percentage of subjects with any disease of the respiratory tract was lower in both male and female smokers of low-yield cigarettes. In multiple logistic regression analyses in which tar was included as a continuous variable and in which we also controlled for age, sex, race, and number of cigarettes smoked per day, smoking lower tar cigarettes was associated with lower risk for pneumonia or influenza, but not with the risk for other acute respiratory infections, other diseases of the upper respiratory tract, chronic obstructive pulmonary disease and allied conditions, or all respiratory diseases considered as a group. In other multiple logistic regression analyses, in which we controlled for age, race, and sex, smokers of low-yield cigarettes had a higher risk for pneumonia or influenza and chronic obstructive pulmonary disease when compared with subjects who had never used tobacco. We conclude that, with regard to pneumonia and influenza seen in an outpatient setting, smoking low-yield cigarettes is probably less hazardous than smoking high-yield cigarettes, but it still represents a considerable hazard compared with not smoking cigarettes at all.     

The prevalent cohort study and the acquired immunodeficiency syndrome

The acquired immunodeficiency syndrome (AIDS) is caused by a retrovirus, the human immunodeficiency virus (HIV). A rapid and convenient method to identify additional cofactors or risk modifiers and markers of disease progression is to study a cohort prevalent with HIV antibody. However, because the time of viral infection is usually unknown in the cohort, there are several potential sources of bias. Three sources of bias in a prevalent cohort study are identified assuming a proportional hazards model: onset confounding, differential length-biased sampling, and frailty selection. A number of problems in the interpretation of results on markers from a prevalent cohort also are considered. It is concluded that risk estimates derived from a prevalent cohort are not directly comparable to risk estimates derived from an incident cohort.     

Risk factor studies of age-at-onset in a sample ascertained for Parkinson disease affected sibling pairs: a cautionary tale

An association between exposure to a risk factor and age-at-onset of disease may reflect an effect on the rate of disease occurrence or an acceleration of the disease process. The difference in age-at-onset arising from case-only studies, however, may also reflect secular trends in the prevalence of exposure to the risk factor. Comparisons of age-at-onset associated with risk factors are commonly performed in case series enrolled for genetic linkage analysis of late onset diseases. We describe how the results of age-at-onset studies of environmental risk factors reflect the underlying structure of the source population, rather than an association with age-at-onset, by contrasting the effects of coffee drinking and cigarette smoking on Parkinson disease age-at-onset with the effects on age-at-enrollment in a population based study sample. Despite earlier evidence to suggest a protective association of coffee drinking and cigarette smoking with Parkinson disease risk, the age-at-onset results are comparable to the patterns observed in the population sample, and thus a causal inference from the age-at-onset effect may not be justified. Protective effects of multivitamin use on PD age-at-onset are also shown to be subject to a bias from the relationship between age and multivitamin initiation. Case-only studies of age-at-onset must be performed with an appreciation for the association between risk factors and age and ageing in the source population.     

Roaming through methodology. XXXV. Bias in prevalent cohort studies of the natural course of disease]

The natural history of infectious diseases with a long asymptomatic incubation period has mainly been studied in cohorts of individuals already infected at study entry: the so-called prevalent cohort study. Because the time of infection is usually unknown in the prevalent cohort, in standard survival analysis it is common to use the time since entry into the cohort instead of the time since infection to study risk factors for disease progression. However, the use of the time since study entry may bias results. The two most important sources of bias are onset confounding and differential length-bias sampling. Because bias may occur, results derived from a prevalent cohort are not directly comparable to results derived from an incident cohort where the moment of infection is known.     

Choice of time-scale in Cox's model analysis of epidemiologic cohort data: a simulation study

Cox's regression model is widely used for assessing associations between potential risk factors and disease occurrence in epidemiologic cohort studies. Although age is often a strong determinant of disease risk, authors have frequently used time-on-study instead of age as the time-scale, as for clinical trials. Unless the baseline hazard is an exponential function of age, this approach can yield different estimates of relative hazards than using age as the time-scale, even when age is adjusted for. We performed a simulation study in order to investigate the existence and magnitude of bias for different degrees of association between age and the covariate of interest. Age to disease onset was generated from exponential, Weibull or piecewise Weibull distributions, and both fixed and time-dependent dichotomous covariates were considered. We observed no bias upon using age as the time-scale. Upon using time-on-study, we verified the absence of bias for exponentially distributed age to disease onset. For non-exponential distributions, we found that bias could occur even when the covariate of interest was independent from age. It could be severe in case of substantial association with age, especially with time-dependent covariates. These findings were illustrated on data from a cohort of 84,329 French women followed prospectively for breast cancer occurrence. In view of our results, we strongly recommend not using time-on-study as the time-scale for analysing epidemiologic cohort data.     

Meta-analysis of the effect of excluding early deaths on the estimated relationship between body mass index and mortality.

OBJECTIVES: Prospective cohort studies typically observe U- or J-shaped relationships between body mass index (BMI) (kg/m2) and mortality. However, some studies suggest that the elevated mortality at lower BMIs is due to confounding by pre-existing occult disease and recommend eliminating subjects who die during the first several (k) years of follow-up. This meta-analysis tests the effects of such early death exclusion on the BMI-mortality association. RESEARCH METHODS AND PROCEDURES: Studies identified from MEDLINE, review articles, ancestry analyses, and the "invisible college." Included studies: 1) measured relative body weight at baseline; 2) included at least 1000 subjects; 3) reported results with and without early-death exclusion, or relevant data; and 4) did not study exclusively diseased populations. Blank tables were mailed to 131 investigators covering 59 databases. Completed tables (n = 16 databases), electronic raw data (n = 7 databases), and original articles (n = 6 databases) provided final data. Meta-analytic regressions compared the BMI-mortality association with and without early death exclusion. The sample included 29 studies and 1,954,345 subjects. RESULTS: The effect of eliminating early deaths was statistically significant but minuscule in magnitude. Implementation of early death exclusion was estimated to shift the BMI associated with minimum mortality only 0.4 units for men and 0.6 units for women at age 50. Even at a BMI 16, the estimated relative risk (compared to BMI 25) decreased only 0.008 units for men and 0.076 units for women at age 50. DISCUSSION: Results indicate that either pre-existing disease does not confound the BMI-mortality association or eliminating early deaths is inefficient for reducing that confounding.