Missing data and multiple imputation in clinical epidemiological research

Pedersen, Alma B; Mikkelsen, Ellen M.; Cronin-Fenton, Deirdre; Kristensen, Nickolaj Risbo; Pham, Tra My; Pedersen, Lars; Petersen, Irene

doi:10.2147/clep.s129785

Cited by 681 publications

(579 citation statements)

References 28 publications

Supporting

Mentioning

555

Contrasting

Unclassified

Order By: Relevance

“…Allowed responses to survey questions included ‘don’t know’ but occasionally residents refused to answer particular questions. To avoid a sizeable loss of respondents due to missing data as a result of not knowing the answer in several survey responses, notably with 54 (21%) of individuals not knowing the household income, we used a combination of simple and multiple imputation 16. Household income was imputed using the most common response of others in the same household, if possible.…”

Section: Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Social determinants of health among residential areas with a high tuberculosis incidence in a remote Inuit community

Kilabuk

Momoli

Mallick

et al. 2019

J Epidemiol Community Health

View full text Add to dashboard Cite

BackgroundTuberculosis (TB) remains a significant health burden among Inuit in Canada. Social determinants of health (SDH) play a key role in TB infection, disease and ongoing transmission in this population. The objective of this research was to estimate the prevalence of social determinants of Inuit health as they relate to latent TB infection (LTBI) among people living in residential areas at high risk for TB in Iqaluit, Nunavut.MethodsInperson home surveys were conducted among those who lived in predetermined residential areas at high risk for TB identified in a door-to-door TB prevention campaign in Iqaluit, Nunavut in 2011. Risk ratios for SDH and LTBI were estimated, and multiple imputation was used to address missing data.Results261 participants completed the questionnaire. Most participants identified as Inuit (82%). Unadjusted risk ratios demonstrated that age, education, smoking tobacco, crowded housing conditions and Inuit ethnicity were associated with LTBI. After adjusting for other SDH, multivariable analysis showed an association between LTBI with increasing age (relative risk, RR 1.07, 95% CI 1.04 to 1.11), crowded housing (RR 1.48, 95% CI 1.10 to 2.00) and ethnicity (RR 2.76, 95% CI 1.33 to 5.73) after imputing missing data.ConclusionAmong high-risk residential areas for TB in a remote Arctic region of Canada, crowded housing and Inuit ethnicity were associated with LTBI after adjusting for other SDH. In addition to strong screening and treatment programmes, alleviating the chronic housing shortage will be a key element in the elimination of TB in the Canadian Inuit Nunangat.

show abstract

Section: Methodsmentioning

confidence: 99%

“…Otherwise, we used multiple imputation to create and analyse imputed data sets (m=20). This is thought to be a scientifically sound approach to analyses with missing data16 if the mechanisms and assumptions are approximately true. Our imputation model included all variables described in the analysis model.…”

Section: Methodsmentioning

confidence: 99%

Social determinants of health among residential areas with a high tuberculosis incidence in a remote Inuit community

Kilabuk

Momoli

Mallick

et al. 2019

J Epidemiol Community Health

View full text Add to dashboard Cite

show abstract

“…We addressed missing data by the missing indicator method, which involves grouping missing values into a ‘missing’ category 34. In particular, data on education were incomplete, as it was not systematically registered before 1973.…”

Section: Methodsmentioning

confidence: 99%

Socioeconomic inequality in drug reimbursement during end-of-life care: a nationwide study

Daugaard

Neergaard

Vestergaard

et al. 2019

J Epidemiol Community Health

View full text Add to dashboard Cite

BackgroundIn Denmark, patients who are terminally ill have the right to drug reimbursement due to terminal illness (DRTI). DRTI, a proxy marker of planned end-of-life care, is intended to be equally accessible regardless of socioeconomic position. This study examined social and socioeconomic differences in DRTI among Danish patients who are terminally ill.MethodsThis cross-sectional study based on individual-level nationwide data included all patients dying from cancer, dementia, ischaemic heart disease, chronic obstructive pulmonary disease, chronic liver disease, congestive heart failure, diabetes or stroke in 2006–2015 (n=307 188). We analysed associations between social and socioeconomic position (education, income, cohabiting status, migrant status and employment) and DRTI. Prevalence ratios (PR) and 95% CIs were estimated using log-linear models adjusted for age, gender, comorbidity, cause of death and residence.ResultsOverall, 27.9% of patients received DRTI (n=85 616). A substantial difference in likelihood of receiving DRTI was observed among patients with a social and socioeconomic profile associated with the highest versus lowest probability of DRTI (adjusted PR 1.44, 95% CI 1.18 to 1.75). The probability of DRTI was higher among patients with high income compared with low income (adjusted PR 1.22, 95% CI 1.17 to 1.26). Also, living with a partner and being immigrant or descendant of such were associated with higher probability of DRTI compared with living alone and of Danish origin, whereas employment was associated with lower probability of DRTI compared with retirement.ConclusionSocial and socioeconomic position was associated with the likelihood of receiving DRTI, which indicates that planned end-of-life care is not equally accessible in Denmark.

show abstract

“…Sensitivity analyses using marginal structural models and a disease risk score analysis diminished the possibility of residual and time dependent confounding due to the long follow-up period 10. Multiple imputation11 also suggested that missing data had little impact on the observed associations. A frequent limitation of pharmacoepidemiology studies is the potential for confounding by indication 12.…”

mentioning

confidence: 99%