Correcting for ascertainment bias of relative-risk estimates obtained using affected-sib-pair linkage data

Locus-specific sibling relative risk is often estimated using affected-sib-pair lod score analysis of affected sibships and may be used to decide whether to continue or discontinue the search for additional susceptibility genes. We showed that relative-risk estimates obtained using affected-sib-pair data are asymptotically unbiased when each pair is given a weight inversely proportional to the sibship ascertainment probability. Here we show by simulation that the extent of the bias of relative risks estimated using the incorrect ascertainment weights is small for dominant models, but large for single-locus recessive models and some two-locus heterogeneity models. Since in practice the ascertainment scheme is often unknown, we investigate methods for jointly estimating ascertainment and relative risks from affected-sibship data. Given a sufficient sample size, a reasonable estimate of relative risk may always be obtained using only affected pairs from sibships with two affected and no unaffected siblings. This estimate, which has a large variance, may then be used in a three-stage procedure (which we call the alpha method) to estimate consistently both the ascertainment probabilities and the relative risks with greater precision. We additionally propose correction factors to eliminate small-sample bias of relative risks and investigate the bias due to error in the estimate of disease locus location.     

Correcting for the bias caused by drop-outs in hypertension trials

A major source of bias in hypertension trials can arise from patients who are withdrawn during the course of the trial because of inadequate blood pressure control. We develop a mechanism which allows for such withdrawals while preserving the potential to make an unbiased comparison of the treatment effects. The approach is illustrated using data from a large multicentre trial of two anti-hypertensive agents in patients with mild to moderate essential hypertension.     

Linkage mapping methods applied to the COGA data set: presentation Group 4 of Genetic Analysis Workshop 14

Presentation Group 4 participants analyzed the Collaborative Study on the Genetics of Alcoholism data provided for Genetic Analysis Workshop 14. This group examined various aspects of linkage analysis and related issues. Seven papers included linkage analyses, while the eighth calculated identity-by-descent (IBD) probabilities. Six papers analyzed linkage to an alcoholism phenotype: ALDX1 (four papers), ALDX2 (one paper), or a combination both (one paper). Methods used included Bayesian variable selection coupled with Haseman-Elston regression, recursive partitioning to identify phenotype and covariate groupings that interact with evidence for linkage, nonparametric linkage regression modeling, affected sib-pair linkage analysis with discordant sib-pair controls, simulation-based homozygosity mapping in a single pedigree, and application of a propensity score to collapse covariates in a general conditional logistic model. Alcoholism linkage was found with > or =2 of these approaches on chromosomes 2, 4, 6, 7, 9, 14, and 21. The remaining linkage paper compared the utility of several single-nucleotide polymorphism (SNP) and microsatellite marker maps for Monte Carlo Markov chain combined oligogenic segregation and linkage analysis, and analyzed one of the electrophysiological endophenotypes, ttth1, on chromosome 7. Linkage was found with all marker sets. The last paper compared the multipoint IBD information content of several SNP sets and the microsatellite set, and found that while all SNP sets examined contained more information than the microsatellite set, most of the information contained in the SNP sets was captured by a subset of the SNP markers with approximately 1-cM marker spacing. From these papers, we highlight three points: a 1-cM SNP map seems to capture most of the linkage information, so denser maps do not appear necessary; careful and appropriate use of covariates can aid linkage analysis; and sources of increased gene-sharing between relatives should be accounted for in analyses.     

Correcting for selection bias in estimation of within-individual variance

Consider a variable whose expected value distributes among individuals in a population, and which also has an important component of within-individual variance. In a screening study that involves repeated observations only for those individuals whose initial observation exceeds an arbitrary cutoff point, the usual estimator of within-individual variance is biased. Assuming normality and independence, this note gives the derivation of the expected value of the estimator and uses it to obtain an unbiased estimator. The results generalize to the bivariate case that involves selection on only one variable of the pair. A companion paper provides an example with use of blood pressure.     

Ghost-time bias from imperfect mortality ascertainment in aging cohorts

Purpose

Many cohort studies in the United States link with the National Death Index to detect deaths. Although linkage with National Death Index is relatively sensitive, some participant deaths will be missed. These participants continue to contribute person-time to the data set after their death, resulting in bias, which we refer to as ghost-time bias. We sought to evaluate the influence of ghost-time bias on mortality relative risk (RR) estimates.

Correcting for bias in the selection and validation of informative diagnostic tests

When developing a new diagnostic test for a disease, there are often multiple candidate classifiers to choose from, and it is unclear if any will offer an improvement in performance compared with current technology. A two‐stage design can be used to select a promising classifier (if one exists) in stage one for definitive validation in stage two. However, estimating the true properties of the chosen classifier is complicated by the first stage selection rules. In particular, the usual maximum likelihood estimator (MLE) that combines data from both stages will be biased high. Consequently, confidence intervals and p ‐values flowing from the MLE will also be incorrect. Building on the results of Pepe et al. (SIM 28 :762–779), we derive the most efficient conditionally unbiased estimator and exact confidence intervals for a classifier's sensitivity in a two‐stage design with arbitrary selection rules; the condition being that the trial proceeds to the validation stage. We apply our estimation strategy to data from a recent family history screening tool validation study by Walter et al. (BJGP 63 :393–400) and are able to identify and successfully adjust for bias in the tool's estimated sensitivity to detect those at higher risk of breast cancer. © 2015 The Authors. Statistics in Medicine published by John Wiley & Sons Ltd.     

Comparison of GENEHUNTER and MFLINK for analysis of COGA linkage data

We compared the NPLALL statistic from GENEHUNTER with two-point and three-point MALODs and MFLODs from MFLINK for all autosomal markers in the Collaborative Study on the Genetics of Alcoholism (COGA) data set. In general MFLINK produced more significant results than GENEHUNTER and implicated two regions containing candidate genes (ADH3 and DRD2). Many regions of interest identified in other studies reported at this workshop produced MALODs significant at p < or = 0.05, but these would not have been picked up by GENEHUNTER unless a less significant threshold were used.     

Correcting for nonrandom ascertainment in generalized linear mixed models (GLMMs), fitted using Gibbs sampling

Gibbs sampling-based generalized linear mixed models (GLMMs) provide a convenient and flexible way to extend variance components models for multivariate normally distributed continuous traits to other classes of phenotype. This includes binary traits and right-censored failure times such as age-at-onset data. The approach has applications in many areas of genetic epidemiology. However, the required GLMMs are sensitive to nonrandom ascertainment. In the absence of an appropriate correction for ascertainment, they can exhibit marked positive bias in the estimated grand mean and serious shrinkage in the estimated magnitude of variance components. To compound practical difficulties, it is currently difficult to implement a conventional adjustment for ascertainment because of the need to undertake repeated integration across the distribution of random effects. This is prohibitively slow when it must be repeated at every iteration of the Markov chain Monte Carlo (MCMC) procedure. This paper motivates a correction for ascertainment that is based on sampling random effects rather than integrating across them and can therefore be implemented in a general-purpose Gibbs sampling environment such as WinBUGS. The approach has the characteristic that it returns ascertainment-adjusted parameter estimates that pertain to the true distribution of determinants in the ascertained sample rather than in the general population. The implications of this characteristic are investigated and discussed. This paper extends the utility of Gibbs sampling-based GLMMs to a variety of settings in which family data are ascertained nonrandomly.     

Asymptotic power of likelihood-ratio tests for detecting quantitative trait loci using the COGA data

We investigated the asymptotic power of the likelihood-ratio test for detecting linkage to a quantitative trait locus (QTL) using the data set from the Collaborative Study on the Genetics of Alcoholism (COGA). Assuming a total trait heritability of 50% as determined for the COGA Cz P300 phenotype, we determined the minimum QTL heritability required at each point in the genome to achieve 80% power to detect linkage with a lod of 3.0. We find that there are regions of the genome where it is not possible to detect a QTL of any effect with 80% power, and that the overall minimum detectable QTL heritability for the COGA data is 0.35-0.40.     

The effect of phenotype variation on detection of linkage in the COGA data

Error in phenotypic measurement can significantly compromise ability to detect linkage. We assessed the impact of introducing phenotypic measurement error on our ability to detect a quantitative trait locus in the Collaborative Study on the Genetics of Alcoholism (COGA) data. The impact of introducing three different types of errors was evaluated: 1) errors generated by sampling from a normal distribution; 2) errors generated by permuting phenotype values between subjects; and 3) errors generated by sampling from a uniform error distribution.     

The effect of provider-level ascertainment bias on profiling nursing homes

Profiling health care providers for the purpose of public reporting and quality improvement has become commonplace. Recently, the Centers for Medicare and Medicaid Services (CMS) began publishing measures of quality for every Medicare/Medicaid-certified nursing home in the country. The facility-specific quality indicators (QIs) reported by CMS are based on quarterly measures from the minimum data set (MDS). However, some QIs from the MDS are potentially subject to ascertainment bias. Ascertainment bias would occur if there was variation in the way items that make up QIs are measured by nurses from each facility. This is potentially a problem for difficult-to-measure items such as pain and pressure ulcers. To assess the impact of ascertainment bias on profiling, we utilize data from a reliability study of nursing homes from six states. We develop methods for profiling providers in situations where the data consist of a response variable for each subject based on assessments from an internal rater, and, for a subset of subjects in each facility, a response variable based on assessments from an independent (external) rater. The internal assessments are potentially subject to provider-level ascertainment bias, whereas the independent assessments are considered the 'gold standard'. Our methods extend popular Bayesian approaches for profiling by using the paired observations from the subset of subjects with error-prone and error-free assessments to adjust for ascertainment bias. We apply the methods to MDS merged with the reliability data, and compare the bias-corrected profiles with those of standard approaches.     

Family density of alcoholism and linkage information in the analysis of the COGA data

In this study of GAW11 Problem 1, we analyzed the genome scan data in families weighted according to the density of alcoholism among the probands' siblings. We hypothesized that certain disease-predisposing alleles may be common in the general population, rendering high-density sibships less informative for linkage. Three types of families were found in the data, with the prevalence of alcoholism of 1.0, 0.78, and 0.24 in the probands' sibships. The linkage results showed several peak lod scores on chromosomes 1, 2, 4, 8, 11, 19, and 21, the majority of which originated in only one or two types of families. However, for almost all markers, the maximum lod scores observed without the weights were equal to or exceeded the values obtained for any single type of family. These results indicate that although the stratification of families may be theoretically justified, in practice the best strategy is to use all available information.     

On the validity of the TDT test in the presence of comorbidity and ascertainment bias.

Comorbidity, the association of two disorders, occurs commonly with complex diseases. In this paper, we investigate the effects of both true (within-family) comorbidity and spurious comorbidity due to ascertainment bias on the validity of both the parental and sibling control transmission/disequilibrium test. Specifically, we consider settings in which a candidate gene is unlinked to the target phenotype but is in linkage disequilibrium with a comorbid phenotype. We derive conditions under which the presence of true and/or spurious comorbidity will result in an artificial correlation between the target phenotype and the candidate gene.     

A novel form of ascertainment bias in case-control studies of cancer screening.

In case-control studies of cancer screening, some have generally admonished investigators against case definitions based on diagnosis dates because of lead-time bias. However, perhaps partly due to vagueness, the admonitions have been frequently ignored. A recurrence-time model simulates case ascertainment when diagnosis must occur within a specific calendar period. The model depends on screening test sensitivity and rate, age-specific preclinical incidence rates, and preclinical duration time and survival time distributions. For one study of sigmoidoscopic screening for colorectal cancer, when the true odds ratio is 1, its estimate is 0.50 to 0.75 under plausible assumptions. This bias can affect any observational study wherein case definition depends on diagnosis times (e.g., health-plan enrollment data). To avoid bias in observational investigations of cancer screening wherein the case definition depends on the diagnosis date, one must ensure that both screening and preclinical incidence do not occur before the case definition period.     

The increasing incidence of the hemolytic-uremic syndrome in King County, Washington: lack of evidence for ascertainment bias

The annual incidence of the hemolytic-uremic syndrome was determined for the well-defined population of King County, Washington, between 1971 and 1986, inclusive, to ascertain temporal trends in the epidemiology of this disease. The average annual incidence rose from 0.69 cases per 100,000 children under age 15 years between 1971 and 1975 to 1.77 cases between 1976 and 1980 and 1.74 cases between 1981 and 1986. The mean hematocrits, platelet counts, and blood urea nitrogen and creatinine concentrations on admission were similar in all periods, as were the mean length of hospital stay and the proportions of patients requiring erythrocyte and/or platelet transfusions and dialysis. These results indicate that the increased incidence of hemolytic-uremic syndrome in childhood has been sustained in King County, Washington, and that this increase is not due to ascertainment bias caused by the diagnosis of less severely ill cases. Further investigations are needed to determine whether this increased incidence is being experienced in other populations and to assess strategies for the prevention of microangiopathic sequelae to hemorrhagic colitis.     

Correcting for lead time and length bias in estimating the effect of screen detection on cancer survival

Determination of survival time among persons with screen-detected cancer is subject to lead time and length biases. The authors propose a simple correction for lead time, assuming an exponential distribution of the preclinical screen-detectable period. Assuming two latent categories of tumors, one of which is more prone to screen detection and correspondingly less prone to death from the cancer in question, the authors have developed a strategy of sensitivity analysis for various magnitudes of length bias. Here they demonstrate these methods using a series of 25,962 breast cancer cases (1988-2004) from the West Midlands, United Kingdom.     

Investigating the numerical effects of ascertainment bias in linkage analysis: Development of methods and preliminary results

It is general practice to have nonsingle ascertainment of pedigrees for linkage studies, along with intrafamilial sampling that is dependent on who among the related individuals was initially ascertained (Proband dependent or PD sampling). Vieland and Hodge [1995; 1996] have shown that under these conditions, the likelihood used in calculating the lod score is not strictly correct and can produce asymptotically biased estimates of the recombination fraction, θ. However they speculated that this bias would be small in most applications. This paper presents preliminary work aimed at quantifying the numerical magnitude of the bias introduced by PD sampling and nonsingle ascertainment in linkage analysis. We considered five generating models where we varied the ascertainment procedure, intrafamilial sampling scheme, and the sample size, for each model. In this limited initial set of simulations, asymptotic bias θˇ in appears to be trivial, while PD sampling procedures can increase the efficiency of θˇ. These preliminary results support the view that the advantages of unsystematic ascertainment may offset any small estimation bias that may arise. © 1997 Wiley-Liss, Inc.