Imputation Of Covariates Research Articles

The association between tobacco use and COVID-19 diagnoses in three Nordic countries: a pooled analysis.

Previous research has suggested an unexpected negative association between smoking and susceptibility to COVID-19. This study, drawing on population-based data from three Nordic countries-Sweden, Norway, and Finland-aims to investigate this association further, capitalizing on diversity introduced by different containment measures. The objective of this research was to examine the association between cigarette smoking and snus (smokeless tobacco) use and the risk of confirmed COVID-19 infection. A pooled analysis integrating original data from 547,685 participants across three countries. We used a multiple imputation approach based on conditional probabilities to impute the systematically missing covariates. The associations between tobacco use and COVID-19 infection were assessed, controlling for potential confounding factors. Current cigarette smokers had a lower risk of a confirmed COVID-19 case, whereas there was an increased risk among snus users. Our sensitivity analysis confirmed that the associations between tobacco use and COVID-19 infection risk are robust, remaining consistent regardless of whether covariate imputation was applied. Findings support a negative association between smoking and SARS-CoV-2 infection, but not the hypothesis that nicotine may be protective against the risk of contracting SARS-CoV-2 infection.

Dirichlet Process Log Skew-Normal Mixture with a Missing-at-Random-Covariate in Insurance Claim Analysis

In actuarial practice, the modeling of total losses tied to a certain policy is a nontrivial task due to complex distributional features. In the recent literature, the application of the Dirichlet process mixture for insurance loss has been proposed to eliminate the risk of model misspecification biases. However, the effect of covariates as well as missing covariates in the modeling framework is rarely studied. In this article, we propose novel connections among a covariate-dependent Dirichlet process mixture, log-normal convolution, and missing covariate imputation. As a generative approach, our framework models the joint of outcome and covariates, which allows us to impute missing covariates under the assumption of missingness at random. The performance is assessed by applying our model to several insurance datasets of varying size and data missingness from the literature, and the empirical results demonstrate the benefit of our model compared with the existing actuarial models, such as the Tweedie-based generalized linear model, generalized additive model, or multivariate adaptive regression spline.

Abstract 5364: Language modeling of peptide-HLA interactions achieves state-of-the-art performance on prediction of peptide presentation by HLA Class II

Abstract Precise and sensitive prediction of neoantigen presentation to the immune system via human leukocyte antigen (HLA) class II molecules remains a challenge despite the early success of neural networks applied to HLA class I. However, it is necessary to address this modeling challenge because presentation of a neoantigen epitope by both classes of HLA molecules may be valuable to induce a sustained immune response with therapeutic cancer vaccines. Previously we have developed a machine-learning based platform, EDGETM, that provides a state-of-the-art model to predict presentation of peptides by HLA Class I. Here we propose a new addition to our EDGETM platform: a model that leverages structural information of putative epitopes and HLA class II alleles from their in-situ context to predict presentation of peptides by HLA class II. Our model achieves this by leveraging the Evolutionary Scale Model pre-trained protein language model (LM), which has been demonstrated to embed protein sequences with rich structural information. The input to the model is a linear peptide consisting of an epitope and its flanking amino acids, concatenated with structurally relevant amino acids from each HLA allele. This allows our model to treat the modeling problem entirely as a natural language processing task, which minimizes imputation of covariates found in prior approaches when performing inference in the context of vaccine design, while maximizing the richness of the LM embeddings on longer linear peptides. Crucially, this also allows our model to generalize to any allele that has a known sequence. Additionally, DR-, DP-, and DQ-specific immunoaffinity purified mass spectrometry multi-allelic (MA) presentation data were generated per tumor or cell line sample, spanning 89 alleles in aggregate. We demonstrate that incrementally decreasing HLA class II allele MA resolution during training results in substantially improved predictions for situations where MA presentation data has completely ambiguous epitope presentation across DR/DP/DQ alleles. Our model achieves an Average Precision (AP) of 0.92 and ROC-AUC of 0.98 on the same benchmark validation data as the current state-of-the-art model BERTMHC, which achieved an AP of 0.81 and ROC-AUC of 0.95. These are the best AP and ROC-AUC for an HLA Class II presentation model on this benchmark dataset to the best of our knowledge. Our model is a significant advancement in HLA class II epitope prediction that allows our EDGETM platform to bring neoantigen vaccine design optimized for both class I and class II presentation within reach. Citation Format: Daniel Sprague, Joshua Klein, Italo Faria do Valle, Olivia Petrillo, Melissa Rotunno, Matthew Davis, Monica Lane, Karin Jooss, Ankur Dhanik. Language modeling of peptide-HLA interactions achieves state-of-the-art performance on prediction of peptide presentation by HLA Class II. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5364.

Selective Imputation of Covariates in High Dimensional Censored Data

Efficient modeling of censored data, that is, data which are restricted by some detection limit or truncation, is important for many applications. Ignoring the censoring can be problematic as valuable information may be missing and restoration of these censored values may significantly improve the quality of models. There are many scenarios where one may encounter censored data: survival data, interval-censored data or data with a lower limit of detection. Strategies to handle censored data are plenty, however, little effort has been made to handle censored data of high dimension. In this article, we present a selective multiple imputation approach for predictive modeling when a larger number of covariates are subject to censoring. Our method allows for iterative, subject-wise selection of covariates to impute in order to achieve a fast and accurate predictive model. The algorithm furthermore selects values for imputation which are likely to provide important information if imputed. In contrast to previously proposed methods, our approach is fully nonparametric and therefore, very flexible. We demonstrate that, in comparison to previous work, our model achieves faster execution and often comparable accuracy in a simulated example as well as predicting signal strength in radio network data. Supplementary materials for this article are available online.

Alternative expectation approaches for expectation-maximization missing data imputations in cox regression

Missing data is common in survival analysis. It is either removed or imputed using various methods. Expectation-maximization (EM) imputation is a popular method in Cox regression studies. This paper investigated the effect of different regression methods on Cox regression modeling within the framework of EM. A stratified Cox regression model was derived from a dataset of categorical and numerical variables. Missing data were imputed using the EM framework with five machine learning algorithms and then were compared to the full model. The results show that the recursive partition and regression tree (RPART) method performed better than others. However, all regression methods performed poorly in categorical covariate imputation. R code is available online.

A stacked approach for chained equations multiple imputation incorporating the substantive model.

Multiple imputation by chained equations (MICE) has emerged as a popular approach for handling missing data. A central challenge for applying MICE is determining how to incorporate outcome information into covariate imputation models, particularly for complicated outcomes. Often, we have a particular analysis model in mind, and we would like to ensure congeniality between the imputation and analysis models. We propose a novel strategy for directly incorporating the analysis model into the handling of missing data. In our proposed approach, multiple imputations of missing covariates are obtained without using outcome information. We then utilize the strategy of imputation stacking, where multiple imputations are stacked on top of each other to create a large data set. The analysis model is then incorporated through weights. Instead of applying Rubin's combining rules, we obtain parameter estimates by fitting a weighted version of the analysis model on the stacked data set. We propose a novel estimator for obtaining standard errors for this stacked and weighted analysis. Our estimator is based on the observed data information principle in Louis' work and can be applied for analyzing stacked multiple imputations more generally. Our approach for analyzing stacked multiple imputations is the first method that can be easily applied (using R package StackImpute) for a wide variety of standard analysis models and missing data settings.

Handling Missing Values in Interrupted Time Series Analysis of Longitudinal Individual-Level Data.

BackgroundIn the interrupted time series (ITS) approach, it is common to average the outcome of interest at each time point and then perform a segmented regression (SR) analysis. In this study, we illustrate that such ‘aggregate-level’ analysis is biased when data are missing at random (MAR) and provide alternative analysis methods.MethodsUsing electronic health records from the UK, we evaluated weight change over time induced by the initiation of antipsychotic treatment. We contrasted estimates from aggregate-level SR analysis against estimates from mixed models with and without multiple imputation of missing covariates, using individual-level data. Then, we conducted a simulation study for insight about the different results in a controlled environment.ResultsAggregate-level SR analysis suggested a substantial weight gain after initiation of treatment (average short-term weight change: 0.799kg/week) compared to mixed models (0.412kg/week). Simulation studies confirmed that aggregate-level SR analysis was biased when data were MAR. In simulations, mixed models gave less biased estimates than SR analysis and, in combination with multilevel multiple imputation, provided unbiased estimates. Mixed models with multiple imputation can be used with other types of ITS outcomes (eg, proportions). Other standard methods applied in ITS do not help to correct this bias problem.ConclusionAggregate-level SR analysis can bias the ITS estimates when individual-level data are MAR, because taking averages of individual-level data before SR means that data at the cluster level are missing not at random. Avoiding the averaging-step and using mixed models with or without multilevel multiple imputation of covariates is recommended.

Imputed spatial data: Cautions arising from response and covariate imputation measurement error

When data for observations are missing, scientists often remove those observations from an analysis, or replace them with imputations, with few other options available to these analysts. Confining an analysis to those observations with complete data can waste resources expended for, especially, those observations with near-complete data. Selectively retaining variables with complete data for observations with incomplete data can compromise mathematical properties of data analyses (e.g., covariance matrices constructed with pairwise deletion having negative eigenvalues). Unless missing data observations are a random subsample of a given sample, entirely removing these observations can result in biased statistical results. Imputations, such as those furnished by kriging, are best linear unbiased predictors (BLUPs), which can be conditional expectations. As such, they are smoothed values that may be viewed as representing attribute values with measurement error, and hence estimated attribute variance based upon them tends to be biased downward. Because, for example, regression coefficients use variance estimates in their calculations, this downward bias can propagate through a data analysis and its statistical inferences. This paper compares spatial regression analyses between complete datasets and the same datasets in which data values are suppressed and then these missing data are imputed, investigating the presence of such imputations in a response variable as well as in covariates. This study employs the following three imputation methods: kriging, spatial autoregression, and Moran eigenvector spatial filtering. Its emphasis is on predictive modeling as well as spatial data quality and uncertainty.

Multiple imputation in Cox regression when there are time-varying effects of covariates.

In Cox regression, it is important to test the proportional hazards assumption and sometimes of interest in itself to study time‐varying effects (TVEs) of covariates. TVEs can be investigated with log hazard ratios modelled as a function of time. Missing data on covariates are common and multiple imputation is a popular approach to handling this to avoid the potential bias and efficiency loss resulting from a “complete‐case” analysis. Two multiple imputation methods have been proposed for when the substantive model is a Cox proportional hazards regression: an approximate method (Imputing missing covariate values for the Cox model in Statistics in Medicine (2009) by White and Royston) and a substantive‐model‐compatible method (Multiple imputation of covariates by fully conditional specification: accommodating the substantive model in Statistical Methods in Medical Research (2015) by Bartlett et al). At present, neither accommodates TVEs of covariates. We extend them to do so for a general form for the TVEs and give specific details for TVEs modelled using restricted cubic splines. Simulation studies assess the performance of the methods under several underlying shapes for TVEs. Our proposed methods give approximately unbiased TVE estimates for binary covariates with missing data, but for continuous covariates, the substantive‐model‐compatible method performs better. The methods also give approximately correct type I errors in the test for proportional hazards when there is no TVE and gain power to detect TVEs relative to complete‐case analysis. Ignoring TVEs at the imputation stage results in biased TVE estimates, incorrect type I errors, and substantial loss of power in detecting TVEs. We also propose a multivariable TVE model selection algorithm. The methods are illustrated using data from the Rotterdam Breast Cancer Study. R code is provided.

Conditional expectation estimation through attributable components

Abstract A general methodology is proposed for the explanation of variability in a quantity of interest x in terms of covariates z = (z1, …, zL). It provides the conditional mean $\bar{x}(z)$ as a sum of components, where each component is represented as a product of non-parametric one-dimensional functions of each covariate zl that are computed through an alternating projection procedure. Both x and the zl can be real or categorical variables; in addition, some or all values of each zl can be unknown, providing a general framework for multi-clustering, classification and covariate imputation in the presence of confounding factors. The procedure can be considered as a preconditioning step for the more general determination of the full conditional distribution $\boldsymbol{\rho}(x|z) $ through a data-driven optimal-transport barycenter problem. In particular, just iterating the procedure once yields the second order structure (i.e. the covariance) of $\boldsymbol{\rho}(x|z) $. The methodology is illustrated through examples that include the explanation of variability of ground temperature across the continental United States and the prediction of book preference among potential readers.

Multiple imputation of missing covariates for the Cox proportional hazards cure model.

We explore several approaches for imputing partially observed covariates when the outcome of interest is a censored event time and when there is an underlying subset of the population that will never experience the event of interest. We call these subjects 'cured', and we consider the case where the data are modeled using a Cox proportional hazards (CPH) mixture cure model. We study covariate imputation approaches using fully conditional specification. We derive the exact conditional distribution and suggest a sampling scheme for imputing partially observed covariates in the CPH cure model setting. We also propose several approximations to the exact distribution that are simpler and more convenient to use for imputation. A simulation study demonstrates that the proposed imputation approaches outperform existing imputation approaches for survival data without a cure fraction in terms of bias in estimating CPH cure model parameters. We apply our multiple imputation techniques to a study of patients with head and neck cancer. Copyright © 2016 John Wiley & Sons, Ltd.

Multiple Imputation of Covariates by Substantive-model Compatible Fully Conditional Specification

Multiple imputation is a practical, principled approach to handling missing data. When used to impute missing values in covariates of regression models, imputation models may be misspecified if they are not compatible with the substantive model of interest for the outcome. In this article, we introduce the smcfcs command, which imputes covariates by substantive-model compatible fully conditional specification. This modifies the popular fully conditional specification or chained-equations approach to multiple imputation by imputing each covariate compatibly with a user-specified substantive model. We compare the smcfcs command with standard fully conditional specification imputation using mi impute chained in a simulation study and illustrative analysis of data from a study investigating time to tumor recurrence in breast cancer.

Multiple imputation of covariates by fully conditional specification: Accommodating the substantive model

Missing covariate data commonly occur in epidemiological and clinical research, and are often dealt with using multiple imputation. Imputation of partially observed covariates is complicated if the substantive model is non-linear (e.g. Cox proportional hazards model), or contains non-linear (e.g. squared) or interaction terms, and standard software implementations of multiple imputation may impute covariates from models that are incompatible with such substantive models. We show how imputation by fully conditional specification, a popular approach for performing multiple imputation, can be modified so that covariates are imputed from models which are compatible with the substantive model. We investigate through simulation the performance of this proposal, and compare it with existing approaches. Simulation results suggest our proposal gives consistent estimates for a range of common substantive models, including models which contain non-linear covariate effects or interactions, provided data are missing at random and the assumed imputation models are correctly specified and mutually compatible. Stata software implementing the approach is freely available.

Analysis of Melanoma Onset: Assessing Familial Aggregation by Using Estimating Equations and Fitting Variance Components via Bayesian Random Effects Models

We investigate whether relative contributions of genetic and shared environmental factors are associated with an increased risk in melanoma. Data from the Queensland Familial Melanoma Project comprising 15,907 subjects arising from 1912 families were analyzed to estimate the additive genetic, common and unique environmental contributions to variation in the age at onset of melanoma. Two complementary approaches for analyzing correlated time-to-onset family data were considered: the generalized estimating equations (GEE) method in which one can estimate relationship-specific dependence simultaneously with regression coefficients that describe the average population response to changing covariates; and a subject-specific Bayesian mixed model in which heterogeneity in regression parameters is explicitly modeled and the different components of variation may be estimated directly. The proportional hazards and Weibull models were utilized, as both produce natural frameworks for estimating relative risks while adjusting for simultaneous effects of other covariates. A simple Markov Chain Monte Carlo method for covariate imputation of missing data was used and the actual implementation of the Bayesian model was based on Gibbs sampling using the free ware package BUGS. In addition, we also used a Bayesian model to investigate the relative contribution of genetic and environmental effects on the expression of naevi and freckles, which are known risk factors for melanoma.

Analysis of melanoma onset: assessing familial aggregation by using estimating equations and fitting variance components via Bayesian random effects models.

We investigate whether relative contributions of genetic and shared environmental factors are associated with an increased risk in melanoma. Data from the Queensland Familial Melanoma Project comprising 15,907 subjects arising from 1912 families were analyzed to estimate the additive genetic, common and unique environmental contributions to variation in the age at onset of melanoma. Two complementary approaches for analyzing correlated time-to-onset family data were considered: the generalized estimating equations (GEE) method in which one can estimate relationship-specific dependence simultaneously with regression coefficients that describe the average population response to changing covariates; and a subject-specific Bayesian mixed model in which heterogeneity in regression parameters is explicitly modeled and the different components of variation may be estimated directly. The proportional hazards and Weibull models were utilized, as both produce natural frameworks for estimating relative risks while adjusting for simultaneous effects of other covariates. A simple Markov Chain Monte Carlo method for covariate imputation of missing data was used and the actual implementation of the Bayesian model was based on Gibbs sampling using the free ware package BUGS. In addition, we also used a Bayesian model to investigate the relative contribution of genetic and environmental effects on the expression of naevi and freckles, which are known risk factors for melanoma.

Genetic analysis of the age at menopause by using estimating equations and Bayesian random effects models.

Multi-wave self-report data on age at menopause in 2182 female twin pairs (1355 monozygotic and 827 dizygotic pairs), were analysed to estimate the genetic, common and unique environmental contribution to variation in age at menopause. Two complementary approaches for analysing correlated time-to-onset twin data are considered: the generalized estimating equations (GEE) method in which one can estimate zygosity-specific dependence simultaneously with regression coefficients that describe the average population response to changing covariates; and a subject-specific Bayesian mixed model in which heterogeneity in regression parameters is explicitly modelled and the different components of variation may be estimated directly. The proportional hazards and Weibull models were utilized, as both produce natural frameworks for estimating relative risks while adjusting for simultaneous effects of other covariates. A simple Markov chain Monte Carlo method for covariate imputation of missing data was used and the actual implementation of the Bayesian model was based on Gibbs sampling using the freeware package BUGS.

Analysis of gene-smoking interaction in lung cancer.

The Louisiana Lung Cancer Dataset, consisting of 337 extended pedigrees, is analyzed to determine whether a major Mendelian gene interacts with cumulative tobacco smoking (pack-years). The proportional hazards model is utilized, as it is a natural framework for estimating relative risks while adjusting for variability in age of disease onset. Segregation analyses show evidence that a Mendelian gene is segregating in these families, with the most parsimonious model, including sex, pack-years, pack-years squared, and a dominant major gene. The estimated frequency of the high-risk allele is 2% and carriers are estimated to have relative risk of 17.3 for developing lung cancer, compared to noncarriers. The addition of a gene x pack-years interaction does not significantly improve the fit of the model, indicating that on a multiplicative scale, these two factors independently influence lung cancer risk. Smoking history is missing for 23% of the study subjects and degree of "missingness" depends on disease status, age, and birth-year. To account for the nonrandomness of the missing data, a Markov chain Monte Carlo method for covariate imputation is proposed and implemented. Results from this analysis also support a nonsignificant gene-smoking interaction and an allele frequency of 2%, but a lower genetic relative risk (9.0) compared to the "complete case" analysis.