Missing covariate data in medical research: To impute is better than to ignore

Janssen, Kristel J.M.; Donders, A. Rogier T.; Harrell, Frank E.; Vergouwe, Yvonne; Chen, Qingxia; Grobbee, Diederick E.; Moons, Karel G.M.

doi:10.1016/j.jclinepi.2009.12.008

Cited by 489 publications

(381 citation statements)

References 22 publications

(8 reference statements)

Supporting

Mentioning

364

Contrasting

Unclassified

Order By: Relevance

“…Parameter estimates are then combined across these datasets with standard errors adjusted for variability due to missing data (Rubin Donald, 1987;Schafer, 1997). MI suffers from less parameter estimate bias, provides superior statistical power and takes better account of missing data sampling variability than casewise deletion or alternative missing data approaches (Sterne et al, 2009;Janssen et al, 2010). Previously observed data on variables in the analysis make the MAR assumption more realistic (Newsom et al, 2013) and were included along 5 with current demographics in the imputation model.…”

Section: Discussionmentioning

confidence: 99%

How financial strain affects health: Evidence from the Dutch National Bank Household Survey

Prentice

McKillop

French

2017

Social Science & Medicine

View full text Add to dashboard Cite

The mechanisms by which financial strain affects health are not well understood. In this paper, we conduct a longitudinal mediation analysis of the Dutch National Bank 5Household Survey. To quantify the relative importance of biological and nonbiological pathways from financial strain to health, we consider smoking, heavy drinking and being overweight as plausible behavioural responses to financial strain but find that only 4.9% of the response of self-reported health to financial strain is mediated by these behaviours. Further analysis indicates that although financial strain increases 10 impulsivity this has little effect on unhealthy behaviours. Economic stresses therefore appear to be distinct from other forms of stress in the relatively minor influence of nonbiological pathways to ill-health.

show abstract

Section: Discussionmentioning

confidence: 99%

How financial strain affects health: Evidence from the Dutch National Bank Household Survey

Prentice

McKillop

French

2017

Social Science & Medicine

View full text Add to dashboard Cite

show abstract

“…Missing outcome data for remaining participants were imputed using a regression model with key predictor variables (BMI, age, fasting values, ethnicity, and treatment) for each time point and outcome. Imputation was used to correct for verification bias (29). Across all experimental conditions, 11% of data values (378 of 3,472) were missing and imputed (Supplementary Table 2).…”

Section: Statistical Analysesmentioning

confidence: 99%

Breaking Up Prolonged Sitting With Standing or Walking Attenuates the Postprandial Metabolic Response in Postmenopausal Women: A Randomized Acute Study

et al. 2015

View full text Add to dashboard Cite

OBJECTIVETo determine whether breaking up prolonged sitting with short bouts of standing or walking improves postprandial markers of cardiometabolic health in women at high risk of type 2 diabetes. RESEARCH DESIGN AND METHODSTwenty-two overweight/obese, dysglycemic, postmenopausal women (mean 6 SD age 66.6 6 4.7 years) each participated in two of the following treatments: prolonged, unbroken sitting (7.5 h) or prolonged sitting broken up with either standing or walking at a self-perceived light intensity (for 5 min every 30 min). Both allocation and treatment order were randomized. The incremental area under the curves (iAUCs) for glucose, insulin, nonesterified fatty acids (NEFA), and triglycerides were calculated for each treatment condition (mean 6 SEM). The following day, all participants underwent the 7.5-h sitting protocol. RESULTSCompared with a prolonged bout of sitting (iAUC 5.3 6 0.8 mmol/L · h), both standing (3.5 6 0.8 mmol/L · h) and walking (3.8 6 0.7 mmol/L · h) significantly reduced the glucose iAUC (both P < 0.05). When compared with prolonged sitting (548.2 6 71.8 mU/L · h), insulin was also reduced for both activity conditions (standing, 437.2 6 73.5 mU/L · h; walking, 347.9 6 78.7 mU/L · h; both P < 0.05). Both standing (21.0 6 0.2 mmol/L · h) and walking (20.8 6 0.2 mmol/L · h) attenuated the suppression of NEFA compared with prolonged sitting (21.5 6 0.2 mmol/L · h) (both P < 0.05). There was no significant effect on triglyceride iAUC. The effects on glucose (standing and walking) and insulin (walking only) persisted into the following day. CONCLUSIONSBreaking up prolonged sitting with 5-min bouts of standing or walking at a selfperceived light intensity reduced postprandial glucose, insulin, and NEFA responses in women at high risk of type 2 diabetes. This simple, behavioral approach could inform future public health interventions aimed at improving the metabolic profile of postmenopausal, dysglycemic women.

show abstract

“…CCA makes the strict assumption that data are MCAR whereas MI requires the more liberal assumption that missingness is at random, conditional on observed data. Emerging research suggests the superiority of MI [17,18]. Beyond its familiarity to authors and readers, there is little basis for preferring a technique that requires more restrictive assumptions over one with more relaxed assumptions.…”

Section: Secondary and Sensitivity Analysesmentioning

confidence: 99%