Use of the mean, hot deck and multiple imputation techniques to predict outcome in intensive care unit patients in Colombia

Abstract

A cohort of intensive care unit (ICU) patients in 20 Colombian ICUs is used to describe the application of three imputation techniques: single, hot deck and multiple imputation. These strategies were used to impute the missing data in the variables used to construct APACHE II scores, a scoring system for the ICU patients that provides an unbiased standardized estimate of the probability of hospital death. Imputed APACHE II scores were then used in the APACHE II model to estimate adjusted hospital mortality rates. The area under the receiver operating characteristic (ROC) curve was used to compare imputation strategies with respect to predictive power. While statistically significant differences were found for the area under the ROC curve, these differences were not clinically significant.

Substantial sick-leave costs savings due to a graded activity intervention for workers with non-specific sub-acute low back pain

“…The HCU costs and lost productivity costs were analyzed using a bias-corrected and accelerated bootstrapping method with 2,000 replications. The mean costs of HCU were analyzed for the entire study population after imputation of missing data via the hot deck method and after interpolation for non-observed months [20].…”

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

“…This process continues until all censored values for the variable or variables of interest have been replaced. 28 This technique generally underestimates the variance of the parameter estimate because the variation introduced into the imputed analysis is only consistent with the ranges from complete cases identified as being similar to those with missing data. 28 Variance is also underestimated because only a single specific value is imputed for each censored value.…”

“…28 This technique generally underestimates the variance of the parameter estimate because the variation introduced into the imputed analysis is only consistent with the ranges from complete cases identified as being similar to those with missing data. 28 Variance is also underestimated because only a single specific value is imputed for each censored value. The principal advantages of this technique are that it does not require parametric assumptions or careful modeling to identify values to impute and that variation of imputed values may better reflect the distributional properties of the variable.…”

“…10,29 Disadvantages of this technique include the potential for bias because imputed data result solely from complete cases, the need to potentially adjust for variance as mentioned with previous techniques, the need for an adequate number of complete cases to match against (i.e., large sample size with little or modest amounts of missing data), and that there are no clear criteria to guide selection of complete cases to match against cases with missing data. 28 It is this last selection process that is used to address the MAR assumption.…”

The analysis of missing data in public use survey databases : a survey of statistical methods.

“…As a result, there are fewer tendencies towards the mean of the sample. The other main advantage of this non-parametric technique is that it does not require strong distribution assumptions or careful modeling to develop selection criteria for imputing a value (Perez A and Rodolfo JD, 2002 (Perez A and Rodolfo JD, 2002).…”