Multiple Imputation Approach Research Articles

Improvements Achieved by Multiple Imputation for Single-Cell RNA-Seq Data in Clustering Analysis and Differential Expression Analysis.

In a single-cell RNA-seq (scRNA-seq) data set, a high proportion of missing values (or an excessive number of zeroes) are frequently observed. For the related follow-up tasks, such as clustering analysis and differential expression analysis, a data set without missing values is generally required. Many imputation approaches have been proposed for this purpose. Multiple imputation (MI) is a well-established approach to address possible biases in a follow-up analysis result based on one-time imputed data. There is a lack of investigation on this in the analysis of scRNA-seq data. In this study, we have investigated how to efficiently apply the MI approach to the clustering analysis and the differential expression analysis of scRNA-seq data. We proposed an MI procedure for clustering analysis and an MI procedure for differential expression analysis. To demonstrate the improvements achieved by MI in clustering analysis and differential expression analysis of scRNA-seq data, we analyzed three well-known scRNA-seq data sets. scIGANs, an scRNA-seq imputation method based on the generative adversarial networks (GANs), has been recently proposed for scRNA-seq data imputation. Multiple randomly imputed data sets can be conveniently generated by this method. We implemented our MI procedures based on scIGANs. We demonstrated that MI yielded improved performances on the clustering analysis and differential expression analysis results. Our applications to experimental scRNA-seq data illustrated the advantages of MI over one-time imputation of missing values in scRNA-seq data.

Propensity score adjustment of a treatment effect with missing data in psychiatric health services research


 Background: Missing values are a common problem for data analyses in observational studies, which are frequently applied in health services research. This paper examines the usefulness of different approaches to tackle the problem of incomplete observational data, focusing whether the Multiple Imputation (MI) strategy yields adequate estimates when applied to a complex analysis framework.
 Methods: Based on observational study data originally comparing three forms of psychotherapy, a simulation study with different missing data scenarios was conducted. The considered analysis model comprised a propensity score-adjusted treatment effect estimation. Missing values were handled by complete case analysis, different MI approaches, as well as mean and regression imputation.
 Results: All point estimators of the applied methods lay within the 95% confidence interval of the treatment effect derived from the complete simulation data set. Highest deviation was observed for complete case analysis. A distinct superiority of MI methods could not be demonstrated.
 Conclusion: Since there was no clear benefit of one method to deal with missing values over another, health services researchers faced with incomplete observational data are well-advised to apply different imputation methods and compare the results in order to get an impression of their sensitivity.

The Missing Indicator Approach for Accelerated Failure Time Model with Covariates Subject to Limits of Detection

The limit of detection (LOD) is commonly encountered in observational studies when one or more covariate values fall outside the measuring ranges. Although the complete-case (CC) approach is widely employed in the presence of missing values, it could result in biased estimations or even become inapplicable in small sample studies. On the other hand, approaches such as the missing indicator (MDI) approach are attractive alternatives as they preserve sample sizes. This paper compares the effectiveness of different alternatives to the CC approach under different LOD settings with a survival outcome. These alternatives include substitution methods, multiple imputation (MI) methods, MDI approaches, and MDI-embedded MI approaches. We found that the MDI approach outperformed its competitors regarding bias and mean squared error in small sample sizes through extensive simulation.

Considerations for choosing an imputation method for addressing sparse measurement issues dictated by the study design - An illustration from per-protocol analysis in pragmatic trials

Last Observation Carried Forward (LOCF) is an ad-hoc method, with known limitations. In recent years, several methods publications have used LOCF in estimating the per-protocol effect via inverse probability of adherence weighted (IPAW) model, when a time-varying factor is partially measured by the study design. We compare the statistical performances of LOCF and multiple imputation approaches for estimating the per-protocol effects via the IPAW model in the presence of incomplete treatment adherence. We used a validated pragmatic trial data generating simulation algorithm to generate datasets under 7 different simulation scenarios, where a post-randomization prognostic factor was measured after regular intervals. Unmeasured values of a partially observed factor were imputed using LOCF and multiple imputation approaches, and IPAW model was fitted on the imputed data to obtain the estimates, and statistical performances were assessed. When confounding exists, for higher variability of the time-varying factor, multiple imputation approach shows desirable statistical properties under MCAR assumption; otherwise, LOCF approach can be adequate. Both imputation methods performed well in terms of statistical properties, when there is no confounding or when all necessary confounders are adjusted. A case study from Coronary Primary Prevention Trial data was presented, which included some participants with incomplete treatment adherence.

Proper Use of Multiple Imputation and Dealing with Missing Covariate Data

Missing data is a typical problem in clinical studies, where the value of variables of interest is not measured or collected for some patients. This article aimed to review imputation approaches for missing values and their application in neurosurgery. We reviewed current practices on detecting missingness patterns and applications of multiple imputation approaches under different scenarios. Statistical considerations and importance of sensitivity analysis were explained. Various imputation methods were applied to a retrospective cohort. For illustration purposes, a retrospective cohort of 609 patients harboring both ruptured and unruptured intracranial aneurysms and undergoing microsurgical clip reconstruction at Erasmus MC University Medical Center, Rotterdam, The Netherlands, between 2000 and 2019 was used. modified Rankin Scale score at 6 months was the clinical outcome, and potential predictors were age, sex, size of aneurysm, hypertension, smoking, World Federation of Neurosurgical Societies grade, and aneurysm location. Associations were investigated using different imputation approaches, and the results were compared and discussed. Missing values should be treated carefully. Advantages and disadvantages of multiple imputation methods along with imputation in small and big data should be considered depending on the research question and specifics of the study.

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.

Evaluation of multiple imputation approaches for handling missing covariate information in a case-cohort study with a binary outcome

BackgroundIn case-cohort studies a random subcohort is selected from the inception cohort and acts as the sample of controls for several outcome investigations. Analysis is conducted using only the cases and the subcohort, with inverse probability weighting (IPW) used to account for the unequal sampling probabilities resulting from the study design. Like all epidemiological studies, case-cohort studies are susceptible to missing data. Multiple imputation (MI) has become increasingly popular for addressing missing data in epidemiological studies. It is currently unclear how best to incorporate the weights from a case-cohort analysis in MI procedures used to address missing covariate data.MethodA simulation study was conducted with missingness in two covariates, motivated by a case study within the Barwon Infant Study. MI methods considered were: using the outcome, a proxy for weights in the simple case-cohort design considered, as a predictor in the imputation model, with and without exposure and covariate interactions; imputing separately within each weight category; and using a weighted imputation model. These methods were compared to a complete case analysis (CCA) within the context of a standard IPW analysis model estimating either the risk or odds ratio. The strength of associations, missing data mechanism, proportion of observations with incomplete covariate data, and subcohort selection probability varied across the simulation scenarios. Methods were also applied to the case study.ResultsThere was similar performance in terms of relative bias and precision with all MI methods across the scenarios considered, with expected improvements compared with the CCA. Slight underestimation of the standard error was seen throughout but the nominal level of coverage (95%) was generally achieved. All MI methods showed a similar increase in precision as the subcohort selection probability increased, irrespective of the scenario. A similar pattern of results was seen in the case study.ConclusionsHow weights were incorporated into the imputation model had minimal effect on the performance of MI; this may be due to case-cohort studies only having two weight categories. In this context, inclusion of the outcome in the imputation model was sufficient to account for the unequal sampling probabilities in the analysis model.

A generalized weighted combination test of treatment effect for clinical trials with a sequential parallel comparison design and binary endpoint.

To address the issue of a large placebo effect in certain therapeutic areas, rather than the application of the traditional gold standard parallel group placebo-controlled design, different versions of the sequential parallel comparison design have been advocated. In general, the design consists of two consecutive stages and three treatment groups. Stage 1 placebo nonresponders potentially form a prespecified patient subgroup for formal between-treatment comparison at the final analysis. In this research, a version of the design is considered for a binary endpoint. To fully utilize all available data, a generalized weighted combination test is proposed in case placebo has a relatively small effect for some of the study endpoints. The weighted combination of the test based on stage 1 data and the test based on stage 2 data of stage 1 placebo nonresponders suggested in the literature uses only a part of the study data and is a special case of this generalized weighted combination test. A multiple imputation approach is outlined for handling missing not at random data. Simulation is conducted to evaluate the performances of the methods and a data example is employed to illustrate the applications of the methods.

Impute the missing data using retrieved dropouts

BackgroundIn the past few decades various methods have been proposed to handle missing data of clinical studies, so as to assess the robustness of primary results. Some of the methods are based on the assumption of missing at random (MAR) which assumes subjects who discontinue the treatment will maintain the treatment effect after discontinuation. The agency, however, has expressed concern over methods based on this overly optimistic assumption, because it hardly holds for subjects discontinuing the investigational drug. Although in recent years a good number of sensitivity analyses based on missing not at random (MNAR) assumptions have been proposed, some use very conservative assumption on which it might be hard for sponsors and regulators to reach common ground.MethodsHere we propose a multiple imputation method targeting at “treatment policy” estimand based on the MNAR assumption. This method can be used as the primary analysis, in addition to serving as a sensitivity analysis. It imputes missing data using information from retrieved dropouts defined as subjects who remain in the study despite occurrence of intercurrent events. Then imputed data long with completers and retrieved dropouts are analyzed altogether and finally multiple results are summarized into a single estimate. According to definition in ICH E9 (R1), this proposed approach fully aligns with the treatment policy estimand but its assumption is much more realistic and reasonable.ResultsOur approach has well controlled type I error rate with no loss of power. As expected, the effect size estimates take into account any dilution effect contributed by retrieved dropouts, conforming to the MNAR assumption.ConclusionsAlthough multiple imputation approaches are always used as sensitivity analyses, this multiple imputation approach can be used as primary analysis for trials with sufficient retrieved dropouts or trials designed to collect retrieved dropouts.

Accounting for EGFR Mutations in Epidemiologic Analyses of Non-Small Cell Lung Cancers: Examples Based on the International Lung Cancer Consortium Data.

Somatic EGFR mutations define a subset of non-small cell lung cancers (NSCLC) that have clinical impact on NSCLC risk and outcome. However, EGFR-mutation-status is often missing in epidemiologic datasets. We developed and tested pragmatic approaches to account for EGFR-mutation-status based on variables commonly included in epidemiologic datasets and evaluated the clinical utility of these approaches. Through analysis of the International Lung Cancer Consortium (ILCCO) epidemiologic datasets, we developed a regression model for EGFR-status; we then applied a clinical-restriction approach using the optimal cut-point, and a second epidemiologic, multiple imputation approach to ILCCO survival analyses that did and did not account for EGFR-status. Of 35,356 ILCCO patients with NSCLC, EGFR-mutation-status was available in 4,231 patients. A model regressing known EGFR-mutation-status on clinical and demographic variables achieved a concordance index of 0.75 (95% CI, 0.74-0.77) in the training and 0.77 (95% CI, 0.74-0.79) in the testing dataset. At an optimal cut-point of probability-score = 0.335, sensitivity = 69% and specificity = 72.5% for determining EGFR-wildtype status. In both restriction-based and imputation-based regression analyses of the individual roles of BMI on overall survival of patients with NSCLC, similar results were observed between overall and EGFR-mutation-negative cohort analyses of patients of all ancestries. However, our approach identified some differences: EGFR-mutated Asian patients did not incur a survival benefit from being obese, as observed in EGFR-wildtype Asian patients. We introduce a pragmatic method to evaluate the potential impact of EGFR-status on epidemiological analyses of NSCLC. The proposed method is generalizable in the common occurrence in which EGFR-status data are missing.

Trends in lung cancer incidence by gender, histological type and stage at diagnosis in Japan, 1993 to 2015: A multiple imputation approach.

Continued decrease in smoking prevalence and increasing use of sensitive diagnostic procedures necessitate updated monitoring of trends in lung cancer incidence in Japan. We analyzed histology‐ and stage‐specific trends in 1993 to 2015 using data from 62 870 diagnosed cases from the Monitoring of Cancer Incidence in Japan project. After applying a multiple imputation approach to impute missing/unknown values of stage and histology, we estimated age‐standardized incidence rates and applied joinpoint regression analyses. We observed long‐term growth trends in adenocarcinoma (ADC) and localized cancer among both genders, long‐term declining trends among men and leveling‐off trends among women in small‐cell carcinoma (SMC) and squamous cell carcinoma (SQC). Stratifying by gender, we observed an increase in localized ADC with average annual percentage changes (AAPC) of 4.5 (95% confidence interval: 3.9 to 5.0) and 5.7 (5.0 to 6.4), a decrease in regional ADC with AAPC of −1.5 (−2.5 to −0.6) and −2.3 (−4.6 to 0.0), but an increase in distant ADC with AAPC of 1.5 (1.1 to 1.9) and 1.6 (0.9 to 2.3) among males and females, respectively. Additionally, increasing trends in female‐to‐male incidence rate ratios were observed in localized ADC with significantly above one in the most recent diagnosis period. Our results revealed evidence for a partial shift from advanced to early cancer stage, which may suggest the modest effectiveness of nationwide organized screening programs. The observed increasing localized and distant ADC may be linked to improved diagnostic procedures, especially for metastasis detection. Further investigation is needed for more accurate quantification of these factors.

Improving Penalized Logistic Regression Model with Missing Values in High-Dimensional Data

Analysis without adequate handling of missing values may lead to inconsistent and biased estimates. Despite multiple imputations becoming a widely used approach in handling missing data, manuscript researchers generally encounter missing data in their respective studies. In high-dimensional data, penalized regression is a popular technique for performing feature selection and coefficient estimation simultaneously. However, one of the most vital issues with high-dimensional data is that it often contains large quantities of missing data that common multiple imputation approaches may not work correctly. Therefore, this study uses imputations penalized regression models as an extension of the penalized methods to improve the performance and impute missing values in high-dimensional data. The method was applied to real-life high-dimensional datasets for the different number of features, sample sizes, and missing dataset rates to evaluate its efficiency. The method was also compared with other existing imputation penalized methods for high-dimensional data. The comparative experimental results indicate that the proposed method outperforms its competitors by achieving higher sensitivity, specificity, and classification accuracy values.

Estimates of mortality associated with seasonal influenza for the European Union from the GLaMOR project

BackgroundThe European Centres for Disease Prevention and Control (ECDC) estimates that seasonal influenzacauses 4–50 million symptomatic infections in the EU/EEA each year and 15,000–70,000 European citizens die of causes associated with influenza. We used modelling methods to estimate influenza-associated mortality for the European Union by age group and country. MethodsWe compiled influenza-associated respiratory mortality estimates for 31 countries around the world (11 countries in the EU) during 2002–2011 (excluding the 2009 pandemic). From these we extrapolated the influenza mortality burden for all 193 countries of the world, including the 28 countries of the EU, using a multiple imputation approach. To study the effect of vaccination programs, we obtained data from the EU-funded VENICE project regarding the percentage of persons over 65 who were vaccinated in each country; the data ranged from 2% to 82% between the 21 countries which provided estimates for the 2006/07 reference season. ResultsWe estimated that an average of 27,600 (range 16,200-39,000) respiratory deaths were associated with seasonal influenza in the 28 EU countries per winter; 88% were among people 65 years and older, and the rates of mortality in this age group were roughly 35 times higher compared with those < 65 years. Estimates varied considerably across the EU; for example, rates in the elderly ranged from 21.6 (12.5–35.1) per 100,000 in Portugal to 36.5 (16.4–62.5) in Luxembourg, a difference of nearly 70%. We were unable to find a negative correlation between vaccination coverage rates and influenza-associated mortality estimates in the elderly. ConclusionOur EU estimate of influenza-associated respiratory mortality is broadly consistent with the ECDC estimate. More research is needed to explain the observed variation in mortality across the EU, and on possible bias that could explain the unexpected lack of mortality benefits associated with European elderly influenza vaccination programs.

Is young-onset esophageal adenocarcinoma increasing in Japan? An analysis of population-based cancer registries.

BackgroundWhile esophageal squamous cell carcinoma (ESCC) is the predominant histological type in Japan, concern has been expressed over an increase in the proportion of esophageal adenocarcinoma (EAC), especially in middle‐aged populations. This study aimed to assess long‐term trends in esophageal cancer incidence by histological type.MethodsWe used data from three population‐based cancer registries in Japan with 10,642 esophageal cancer cases diagnosed between 1993 and 2014. The multiple imputation approach was used to impute a specific histological type (ESCC, EAC, and others) for cases with “Unknown” or missing status. We calculated the age‐standardized incidence rates by histological type from 1993 to 2014 and fitted age‐period‐cohort models to estimate the annual percent changes (APCs) and adjusted incidence rate ratios (IRRs).ResultsAfter imputation of missing data, the largest mean APC increase was seen in the incidence of EAC in men aged 40–49 years (7.1%) followed by those aged 50–59 years (5.5%). The age‐period‐cohort analysis showed that men who were born in the 1960s and later were more likely to develop EAC relative to men who were born in 1950–1959 (1960–1969 cohort, IRR: 1.42; 1970–1974 cohort, IRR: 2.23), with a period effect indicating a constant increase after 2003. For women, no significant trend in EAC incidence was observed.ConclusionsThe incidence of EAC has increased more prominently compared with that of ESCC, especially in men aged 40–59 years, suggesting the impact of increasing obesity in men and a reduction in H. pylori prevalence in Japan.

Uncertain Data Analysis with Regularized XGBoost

Uncertainty is a ubiquitous element in available knowledge about the real world. Data sampling error, obsolete sources, network latency, and transmission error are all factors that contribute to the uncertainty. These kinds of uncertainty have to be handled cautiously, or else the classification results could be unreliable or even erroneous. There are numerous methodologies developed to comprehend and control uncertainty in data. There are many faces for uncertainty i.e., inconsistency, imprecision, ambiguity, incompleteness, vagueness, unpredictability, noise, and unreliability. Missing information is inevitable in real-world data sets. While some conventional multiple imputation approaches are well studied and have shown empirical validity, they entail limitations in processing large datasets with complex data structures. In addition, these standard approaches tend to be computationally inefficient for medium and large datasets. In this paper, we propose a scalable multiple imputation frameworks based on XGBoost, bootstrapping and regularized method. XGBoost, one of the fastest implementations of gradient boosted trees, is able to automatically retain interactions and non-linear relations in a dataset while achieving high computational efficiency with the aid of bootstrapping and regularized methods. In the context of high-dimensional data, this methodology provides fewer biased estimates and reflects acceptable imputation variability than previous regression approaches. We validate our adaptive imputation approaches with standard methods on numerical and real data sets and shown promising results.

A general approach for imputation of non-normal continuous data based on copula transformation

Dealing with missing data problems for skewed data is a difficult task especially since most of the imputation and data augmentation methodologies assume multivariate normality. The performance of imputation and hence the accuracy of inference on parameters become questionable when the violation of the above assumption occurs. One approach to solve the normality violation is to apply normalizing transformation prior to the imputation phase. However, this approach may introduce new problems such as altering the dependence structure among random variables. We present a general purpose multiple imputation approach based on Copula transformation. The approach is used to effectively transform any continuous multivariate non-normal data to multivariate normal, thereby allowing the imputation using standard normality-based techniques. The method then allows to conveniently back transform the data into original space. Real data sets are used to illustrate the techniques. We then compare the performance of our Copula-based method with other traditional normality-based multiple imputation approaches through extensive simulated and real non-normal multivariate datasets. We demonstrate that this method significantly mitigates the problem and hence the practice of making the blind assumption of multivariate normality for non-normal multivariate data under the assumption that data are missing at different mechanisms.

Communication, Coordination, and Security for People with Multiple Sclerosis (COCOS-MS): a randomised phase II clinical trial protocol

IntroductionPatients with multiple sclerosis (MS) have complex needs that range from organising one’s everyday life to measures of disease-specific therapy monitoring to palliative care. Patients with MS are likely to...

ЗАСТОСУВАННЯ МОДЕЛЕЙ, ЩО ВРАХОВУЮТЬ МЕХАНІЗМ ВИБУВАННЯ ТА МЕТОД ПЕРЕЛОМНОЇ ТОЧКИ В СОЦІАЛЬНИХ ДОСЛІДЖЕННЯХ

Nowadays, social economics focuses on many critical issues; among them, public health and morbidity are among the highest priorities since they directly impact human capital formation, which is an important component in the development of the economy. Within public health issues, one of the crucial directions is the analysis of the effectiveness of drugs, which is typically performed on micro-level involving patients in hospitals. The data collected usually is not complete, and it causes problems during the analysis as if a significant part of the critical data is missed, that invalidates finding. The multiple imputation method is one of the most common approaches in dealing with this problem. Both primary and sensitivity analysis were performed involving multiple imputation approaches. While the preliminary analysis was performed assuming that the missing-data values are overlooked at random, the sensitivity analysis was conducted on the two approaches of missing not-at-random algorithm – the pattern mixture models and the tipping point method. In the paper, the methodological aspects of the usage of these methods were highlighted. Also, the practical implementation of these methods was given in the example of imputing the missing values of the laboratory parameter at different time points with subsequent calculation of AUC and testing the hypothesis of drug efficacy using the analysis of the covariance model. The primary analysis showed the effectiveness of the new drug compared to the placebo. A sensitivity analysis proved the results of the primary analysis. The tipping point method showed that if the assumption that the mean value of dropout is more significant than observed values for more than 196, then the result of the primary analysis is questionable.

A comparative study of multiple imputation and subset correspondence analysis in dealing with missing data

Methods: Multiple imputation and subset correspondence analysis are applied to a set of child asthma data that is mainly categorical and suffers from non-response. Differences in the methods and in the outcomes they produce are studied. In addition, the inclusion of interactions in a subset correspondence analysis is illustrated. Results: Despite the vast differences in the two approaches, they yielded similar results in the identification of genetic, environmental and socio-economic factors that affect childhood asthma. A number of exposure related variables were found to be associated with the greater severity of asthma. It was also found that a finer distinction between the asthma severity levels and their associations with factors was possible with a subset correspondence analysis, compared to the multiple imputation approach. Conclusions: Both multiple imputation and subset correspondence analysis were able to identify several factors associated with childhood asthma while at the same time successfully managing the missing data. This offers the researcher a choice to select the method that best suits his/her study. Keywords: Missing data, Multiple imputation, Subset correspondence analysis

A multiple imputation approach to evaluate the accuracy of diagnostic tests in presence of missing values

Diagnostic tests are used to determine the presence or absence of a disease. Diagnostic accuracy is the main tool to evaluate a test. Four accuracy measures are used to evaluate how well the results of the test under evaluation (index test) agree with the outcome of the reference test (gold standard). These measures are sensitivity, specificity, positive predictive value and negative predictive value. Some subjects are only measured by a subset of tests which result in missing values. This leads to biased results. The mechanism of missing data could be missing completely at random (MCA), missing at random (MAR), or missing not at random (MNAR). Various methods such as the complete-case analysis (CCA) and the maximum likelihood (ML) method are used to handle missing data. Also, imputation methods could be used. The article aims to use a multiple imputation approach to evaluate binary diagnostic tests with missing data under the MCAR mechanism. The proposed approach is applied to a real data set. Also, a simulation study is conducted to evaluate the performance of the proposed approach.