Dataset Information

Much ado about nothing: A comparison of missing data methods and software to fit incomplete data regression models.

ABSTRACT: Missing data are a recurring problem that can cause bias or lead to inefficient analyses. Development of statistical methods to address missingness have been actively pursued in recent years, including imputation, likelihood and weighting approaches. Each approach is more complicated when there are many patterns of missing values, or when both categorical and continuous random variables are involved. Implementations of routines to incorporate observations with incomplete variables in regression models are now widely available. We review these routines in the context of a motivating example from a large health services research dataset. While there are still limitations to the current implementations, and additional efforts are required of the analyst, it is feasible to incorporate partially observed values, and these methods should be utilized in practice.

SUBMITTER: Horton NJ

PROVIDER: S-EPMC1839993 | biostudies-literature | 2007 Feb

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Much ado about nothing: A comparison of missing data methods and software to fit incomplete data regression models.

Horton Nicholas J NJ Kleinman Ken P KP

The American statistician 20070201 1

Missing data are a recurring problem that can cause bias or lead to inefficient analyses. Development of statistical methods to address missingness have been actively pursued in recent years, including imputation, likelihood and weighting approaches. Each approach is more complicated when there are many patterns of missing values, or when both categorical and continuous random variables are involved. Implementations of routines to incorporate observations with incomplete variables in regression ...[more]

PMID: 17401454

Similar Datasets

Project description:IntroductionType-2 diabetes mellitus (T2DM) is a progressive disease, and many patients eventually require insulin therapy. This study examined real-world outcomes of switching basal insulin analogs among patients with T2DM.MethodsUsing two large United States administrative claims databases (IMPACT(®) and Humana(®)), this longitudinal retrospective study examined two cohorts of adult patients with T2DM. Previously on insulin glargine, Cohort 1 either continued insulin glargine (GLA-C) or switched to insulin detemir (DET-S), while Cohort 2 was previously on insulin detemir, and either continued insulin detemir (DET-C) or switched to insulin glargine (GLA-S). One-year follow-up treatment persistence and adherence, glycated hemoglobin (HbA1c), hypoglycemia events, healthcare utilization and costs were assessed. Selection bias was minimized by propensity score matching between treatment groups within each cohort.ResultsA total of 5,921 patients (mean age 60 years, female 50.0%, HbA1c 8.6%) were included in the analysis (Cohort 1: IMPACT(®): n = 536 DET-S matched to n = 2,668 GLA-C; Humana(®): n = 256 DET-S matched to n = 1,262 GLA-C; Cohort 2: n = 419 GLA-S matched to n = 780 DET-C), with similar baseline characteristics between treatment groups in each cohort. During 1-year follow-up, in Cohort 1, DET-S patients, when compared with GLA-C patients, had lower treatment persistence/adherence with 33-40% restarting insulin glargine, higher rapid-acting insulin use, worse HbA1c outcomes, significantly higher diabetes drug costs, and similar hypoglycemia rates, health care utilization and total costs. However, in Cohort 2 overall opposite outcomes were observed and only 19.8% GLA-S patients restarted insulin detemir.ConclusionsThis study showed contrasting clinical and economic outcomes when patients with T2DM switched basal insulin analogs, with worse outcomes observed for patients switching from insulin glargine to insulin detemir and improved outcomes when switching from insulin detemir to insulin glargine. Further investigation into the therapeutic interchangeability of insulin glargine and insulin detemir in the real-world setting is needed.

Project description:BackgroundRecent studies suggested the existence of (poly-)microbial infections in human brains. These have been described either as putative pathogens linked to the neuro-inflammatory changes seen in Parkinson's disease (PD) and Alzheimer's disease (AD) or as a "brain microbiome" in the context of healthy patients' brain samples.MethodsUsing 16S rRNA gene sequencing, we tested the hypothesis that there is a bacterial brain microbiome. We evaluated brain samples from healthy human subjects and individuals suffering from PD (olfactory bulb and pre-frontal cortex), as well as murine brains. In line with state-of-the-art recommendations, we included several negative and positive controls in our analysis and estimated total bacterial biomass by 16S rRNA gene qPCR.ResultsAmplicon sequencing did detect bacterial signals in both human and murine samples, but estimated bacterial biomass was extremely low in all samples. Stringent reanalyses implied bacterial signals being explained by a combination of exogenous DNA contamination (54.8%) and false positive amplification of host DNA (34.2%, off-target amplicons). Several seemingly brain-enriched microbes in our dataset turned out to be false-positive signals upon closer examination. We identified off-target amplification as a major confounding factor in low-bacterial/high-host-DNA scenarios. These amplified human or mouse DNA sequences were clustered and falsely assigned to bacterial taxa in the majority of tested amplicon sequencing pipelines. Off-target amplicons seemed to be related to the tissue's sterility and could also be found in independent brain 16S rRNA gene sequences.ConclusionsTaxonomic signals obtained from (extremely) low biomass samples by 16S rRNA gene sequencing must be scrutinized closely to exclude the possibility of off-target amplifications, amplicons that can only appear enriched in biological samples, but are sometimes assigned to bacterial taxa. Sequences must be explicitly matched against any possible background genomes present in large quantities (i.e., the host genome). Using close scrutiny in our approach, we find no evidence supporting the hypothetical presence of either a brain microbiome or a bacterial infection in PD brains. Video abstract.

Dataset Information

Much ado about nothing: A comparison of missing data methods and software to fit incomplete data regression models.

Publications

Much ado about nothing: A comparison of missing data methods and software to fit incomplete data regression models.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure