Accounting for missing data in statistical analyses: multiple imputation is not always the answer

Hughes, Rachael A.; Heron, Jon; Sterne, Jonathan A C; Tilling, Kate

doi:10.1093/ije/dyz032

Cited by 519 publications

(457 citation statements)

References 59 publications

Supporting

Mentioning

448

Contrasting

Unclassified

Order By: Relevance

“…These results are important for informing analysis strategies and the likely direction and magnitude of bias due to conditioning only on those who participate. For the participator-only analysis for a given model to be unbiased, it is necessary for the outcome variable to be independent of missingness, given the variables in the analysis model 7 . Thus when examining the factors affecting physical activity, all the factors that we have shown here to be related to participation in the physical activity monitoring (BMI, height, education, intelligence, ADHD, age at menarche, should be either included in the analysis model or used in other strategies such as inverse probability weighting (IPW) or multiple imputation (MI).…”

Section: Discussionmentioning

confidence: 99%

“…Thus when examining the factors affecting physical activity, all the factors that we have shown here to be related to participation in the physical activity monitoring (BMI, height, education, intelligence, ADHD, age at menarche, should be either included in the analysis model or used in other strategies such as inverse probability weighting (IPW) or multiple imputation (MI). Where selection is related to the underlying concept(s) measured by the optional component, then this concept will be missing not at random and analyses where it is the outcome will likely be biased 7 . On the other hand, a participator-only analysis of a model that involves only characteristics that are unrelated to participation will not be biased by conditioning on participation.…”

Section: Discussionmentioning

confidence: 99%

“…In general, an analysis will give biased estimates if the exposure and the outcome variable (or causes of them) are associated with participation (conditional on the other variables in the analysis model 7 ). Selection bias can also occur under other circumstances when only the outcome is related to selection, for example if exposure does cause the outcome, and the outcome causes selection 7 . Thus, to understand the impact of selective participation for a particular analysis, we need to identify factors that influence participation.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Genetic predictors of participation in optional components of UK Biobank

Tyrrell

Zheng

Beaumont

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

Large studies (e.g. UK Biobank) are increasingly used for GWAS and Mendelian randomization (MR) studies. Selection into and dropout from studies may bias genetic and phenotypic associations. We examine genetic factors affecting participation in four optional components in up to 451,306 UK Biobank participants.We used GWAS to identify genetic variants associated with participation, MR to estimate effects of phenotypes on participation, and genetic correlations to compare participation bias across different studies. 32 variants were associated with participation in one of the optional components (P<6x10 -9 ), including loci with known links to intelligence and Alzheimer's disease. Genetic correlations demonstrated that participation bias was common across studies. MR showed that longer educational duration, older menarche and taller stature increased participation, whilst higher levels of adiposity, dyslipidaemia, neuroticism, Alzheimer's and schizophrenia reduced participation. Our effect estimates can be used for sensitivity analysis to account for selective participation biases in genetic or non-genetic analyses.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Genetic predictors of participation in optional components of UK Biobank

Tyrrell

Zheng

Beaumont

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…non‐responses were removed). This decision was made to ensure an accurate representation of respondents’ perceptions of the quality and benefits of the care received (Hughes, Heron, Sterne, & Tilling, ).…”

Section: Resultsmentioning

confidence: 99%

An Australian study on the benefits of pastoral care to aged care residents in Christian affiliated homes

Gordon¹,

Jones²,

Taylor³

et al. 2019

Health Soc Care Community

View full text Add to dashboard Cite

This study aimed to understand the experience of pastoral care (PC), that is, the provision of support, comfort and spiritual counselling, from the perspective of Australian aged care residents. A survey research design captured feedback on participants' PC experience. Outcomes were reported by 575 residents (aged 53-102) across 41 sites.The majority perceived that they received a high quality of care (92%) and benefited from their meeting with the pastoral practitioner (80%), 'often' or 'all of the time'.A few significant differences were found based on participants' gender, spirituality (i.e. connection and meaning), religiosity (i.e. faith beliefs and religious practices) and well-being. Females and participants who identified as both religious and spiritual were more likely to feel that their faiths/beliefs were valued. Those with greater psychological well-being, as defined by the World Health Organisation (1998), were more likely to report receiving a high quality of care and greater benefits from receiving PC than those with poorer well-being. Three overarching themes and eight subthemes were identified from the open-ended responses: 1) personal qualities of the pastoral practitioner; caring, supportive, understanding and empathetic; 2) pastoral practitioner met specific needs; spiritual and religious, friendship and company and assistance, advice and help; and 3) positive impact on the participant; feeling listened to, peaceful and valued, accepted and respected. The qualitative findings resonate with Maslow's Hierarchy of Needs, to feel safe, belong and have self-esteem.There was a synergy between what participants desire in the care they receive, asexpressed in the open-ended questions, and what the pastoral practitioners provide, as indicated in the quantitative findings. A study strength was its mixed-method, multi-site and cross-organisational context, enabling PC to be explored across a diverse sample. Future research should consider a pre-and post-test survey to more comprehensively capture the impact and benefits of PC. K E Y W O R D Saged care, chaplaincy, pastoral care, spiritual care | 367 GORDON et al.

show abstract

“…Additionally, participants missing from the multivariable model had more complex clinical profiles than those included. However, missingness was not associated with the outcome variable reducing the potential for missing at random bias .…”

Section: Discussionmentioning

confidence: 97%

Correlates of indicators of potential extra‐medical opioid use in people prescribed opioids for chronic non‐cancer pain

Santo

Larance

Bruno

et al. 2019

Drug and Alcohol Review

View full text Add to dashboard Cite

Introduction and Aims. The opioid-related behaviours in treatment (ORBIT) scale are a measure of recent indicators of potential extra-medical opioid use. Indicators of potential extra-medical opioid use are divergent practices among people prescribed opioids that may place them at risk of harm. This study aimed to examine the correlates of indicators of potential extra-medical opioid use in people prescribed opioids for chronic non-cancer pain (CNCP). Design and Methods. The Pain and Opioids IN Treatment (POINT) study is a prospective cohort study of people prescribed opioids for CNCP in Australia. People prescribed opioids solely for opioid dependence were excluded. This cross-sectional study utilised logistic regression to determine the characteristics associated with reporting any indicators of potential extra-medical opioid use. Results. Of the 1505 participants, 38% reported at least one indicator of potential extra-medical opioid use, most commonly asking for an increase in prescribed opioid dose (21%) and early prescription renewal (12%). Indicators of potential extra-medical opioid use were associated with younger age (adjusted odds ratio [AOR] = 0.98; 95% confidence interval [CI] = 0.92, 0.99), male sex (AOR = 1.53; 95% CI = 1.15, 2.04), lifetime pharmaceutical opioid use disorder (AOR = 1.87; 95% CI = 1.31, 2.66) and lifetime illicit drug use disorder (AOR = 1.72; 95% CI = 1.18, 2.52). Discussion and Conclusions. Over one-third of the POINT cohort reported one or more indicators of potential extra-medical opioid use. Lifetime substance use disorders were associated these divergent practices, highlighting the importance of clinical monitoring and patient education for this patient group. Longitudinal studies are needed to investigate whether indicators of potential extra-medical opioid use predict opioid use disorders in this population. [Santo T Jr, Larance B, Bruno R, Gisev N, Nielsen S, Degenhardt L, Campbell G. Correlates of indicators of potential extra-medical opioid use in people prescribed opioids for chronic non-cancer pain. Drug Alcohol Rev 2020;39: [128][129][130][131][132][133][134]

show abstract

Accounting for missing data in statistical analyses: multiple imputation is not always the answer

Cited by 519 publications

References 59 publications

Genetic predictors of participation in optional components of UK Biobank

Genetic predictors of participation in optional components of UK Biobank

An Australian study on the benefits of pastoral care to aged care residents in Christian affiliated homes

Correlates of indicators of potential extra‐medical opioid use in people prescribed opioids for chronic non‐cancer pain

Contact Info

Product

Resources

About