<i>Advanced Statistics:</i> Missing Data in Clinical Research—Part 2: Multiple Imputation

Newgard, Craig D.; Haukoos, Jason S.

doi:10.1111/j.1553-2712.2007.tb01856.x

Cited by 145 publications

(185 citation statements)

References 37 publications

Supporting

Mentioning

184

Contrasting

Unclassified

Order By: Relevance

“…18 -21 In large populations (n Ͼ 1000) and with 20 databases imputed, as in our study, this technique has been shown to handle up to 90% missing values with a validity and precision of the imputed values almost identical to the "true" values. 22,23 To avoid potential artifacts, we measured ACR in three nonfrozen urine samples within 5 d. This is highly recommended 24,25 and could give more accurate results than a single measurement. Other limitations may be attributed to the imprecision of the Modification of Diet in Renal Disease (MDRD) formula in the range of near-normal values leading to risk category misclassification of some participants.…”

Section: Discussionmentioning

confidence: 99%

“…Using the Stata command "ice," we created 20 complete data sets to achieve maximum accuracy. 22,23 Subsequently, the Stata command "micombine" was used together with standard statistical methods, giving unbiased risk estimates with correct CIs. For most individuals without diabetes and hypertension, data were missing completely at random, and for those who did not return urine samples as requested, data were assumed to be missing at random, thus meeting the assumptions of the method.…”

Section: Concise Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Combining GFR and Albuminuria to Classify CKD Improves Prediction of ESRD

Hallan

Ritz²,

Lydersen³

et al. 2009

Journal of the American Society of Nephrology

392

294

View full text Add to dashboard Cite

Despite the high prevalence of chronic kidney disease (CKD), relatively few individuals with CKD progress to ESRD. A better understanding of the risk factors for progression could improve the classification system of CKD and strategies for screening. We analyzed data from 65,589 adults who participated in the Nord-Trøndelag Health (HUNT 2) Study (1995 to 1997) and found 124 patients who progressed to ESRD after 10.3 yr of follow-up. In multivariable survival analysis, estimated GFR (eGFR) and albuminuria were independently and strongly associated with progression to ESRD: Hazard ratios for eGFR 45 to 59, 30 to 44, and 15 to 29 ml/min per 1.73 m 2 were 6.7, 18.8, and 65.7, respectively (P Ͻ 0.001 for all), and for micro-and macroalbuminuria were 13.0 and 47.2 (P Ͻ 0.001 for both). Hypertension, diabetes, male gender, smoking, depression, obesity, cardiovascular disease, dyslipidemia, physical activity and education did not add predictive information. Time-dependent receiver operating characteristic analyses showed that considering both the urinary albumin/creatinine ratio and eGFR substantially improved diagnostic accuracy. Referral based on current stages 3 to 4 CKD (eGFR 15 to 59 ml/min per 1.73 m 2 ) would include 4.7% of the general population and identify 69.4% of all individuals progressing to ESRD. Referral based on our classification system would include 1.4% of the general population without losing predictive power (i.e., it would detect 65.6% of all individuals progressing to ESRD). In conclusion, all levels of reduced eGFR should be complemented by quantification of urinary albumin to predict optimally progression to ESRD.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Concise Methodsmentioning

confidence: 99%

Combining GFR and Albuminuria to Classify CKD Improves Prediction of ESRD

Hallan

Ritz²,

Lydersen³

et al. 2009

Journal of the American Society of Nephrology

392

294

View full text Add to dashboard Cite

show abstract

“…7 This mechanism requires the pattern of censored values to be completely ''explained,'' or dependent on, observed values in the sample but not dependent on any unobserved or missing values. 8,11,12 In other words, in a given data set (Y) consisting of observed values (Y obs ) and missing values (Y mis ), MAR is present if the probability that a value is censored is dependent only on Y obs and not on Y mis .…”

Section: Marmentioning

confidence: 99%

“…8,10 As this relates to the discussion of MI in the second article of this series, the between-imputation variance is assumed to be zero because only one value exists. 7 Stochastic Regression Imputation. Stochastic regression imputation provides an additional level of sophistication to regression imputation by replacing censored values with the predicted value from a regression analysis plus its residual error.…”

Section: Weighted Complete-case Analysismentioning

confidence: 99%

See 1 more Smart Citation

Advanced Statistics: Missing Data in Clinical Research—Part 1: An Introduction and Conceptual Framework

Haukoos

Newgard

2007

Academic Emergency Medicine

Self Cite

166

View full text Add to dashboard Cite

Missing data are commonly encountered in clinical research. Unfortunately, they are often neglected or not properly handled during analytic procedures, and this may substantially bias the results of the study, reduce study power, and lead to invalid conclusions. In this two-part series, the authors will introduce key concepts regarding missing data in clinical research, provide a conceptual framework for how to approach missing data in this setting, describe typical mechanisms and patterns of censoring of data and their relationships to specific methods of handling incomplete data, and describe in detail several simple and more complex methods of handling such data. In part 1, the authors will describe relatively simple approaches to handling missing data, including complete-case analysis, available-case analysis, and several forms of single imputation, including mean imputation, regression imputation, hot and cold deck imputation, last observation carried forward, and worst case analysis. In part 2, the authors will describe in detail multiple imputation, a more sophisticated and valid method for handling missing data.ACADEMIC EMERGENCY MEDICINE 2007; 14:662-668 ª

show abstract

Physical Activity and Mortality in Individuals With Diabetes Mellitus

Sluik¹,

Buijsse²,

Muckelbauer³

et al. 2012

Arch Intern Med

252

111

View full text Add to dashboard Cite

BACKGROUND Physical activity (PA) is considered a cornerstone of diabetes mellitus management to prevent complications, but conclusive evidence is lacking. METHODS This prospective cohort study and meta-analysis of existing studies investigated the association between PA and mortality in individuals with diabetes. In the EPIC study (European Prospective Investigation Into Cancer and Nutrition), a cohort was defined of 5859 individuals with diabetes at baseline. Associations of leisure-time and total PA and walking with cardiovascular disease (CVD) and total mortality were studied using multivariable Cox proportional hazards regression models. Fixed- and random-effects meta-analyses of prospective studies published up to December 2010 were pooled with inverse variance weighting. RESULTS In the prospective analysis, total PA was associated with lower risk of CVD and total mortality. Compared with physically inactive persons, the lowest mortality risk was observed in moderately active persons: hazard ratios were 0.62 (95% CI, 0.49-0.78) for total mortality and 0.51 (95% CI, 0.32-0.81) for CVD mortality. Leisure-time PA was associated with lower total mortality risk, and walking was associated with lower CVD mortality risk. In the meta-analysis, the pooled random-effects hazard ratio from 5 studies for high vs low total PA and all-cause mortality was 0.60 (95% CI, 0.49-0.73). CONCLUSIONS Higher levels of PA were associated with lower mortality risk in individuals with diabetes. Even those undertaking moderate amounts of activity were at appreciably lower risk for early death compared with inactive persons. These findings provide empirical evidence supporting the widely shared view that persons with diabetes should engage in regular PA.

show abstract

Advanced Statistics: Missing Data in Clinical Research—Part 2: Multiple Imputation

Cited by 145 publications

References 37 publications

Combining GFR and Albuminuria to Classify CKD Improves Prediction of ESRD

Combining GFR and Albuminuria to Classify CKD Improves Prediction of ESRD

Advanced Statistics: Missing Data in Clinical Research—Part 1: An Introduction and Conceptual Framework

Physical Activity and Mortality in Individuals With Diabetes Mellitus

Contact Info

Product

Resources

About