Accounting for missing data caused by drug cessation in observational comparative effectiveness research: a simulation study

ObjectivesTo assess the performance of statistical methods used to compare the effectiveness between drugs in an observational setting in the presence of attrition.MethodsIn this simulation study, we compared the estimations of low disease activity (LDA) at 1 year produced by complete case analysis (CC), last observation carried forward (LOCF), LUNDEX, non-responder imputation (NRI), inverse probability weighting (IPW) and multiple imputations of the outcome. All methods were adjusted for confounders. The reasons to stop the treatments were included in the multiple imputation method (confounder-adjusted response rate with attrition correction, CARRAC) and were either included (IPW2) or not (IPW1) in the IPW method. A realistic simulation data set was generated from a real-world data collection. The amount of missing data caused by attrition and its dependence on the ‘true’ value of the data missing were varied to assess the robustness of each method to these changes.ResultsLUNDEX and NRI strongly underestimated the absolute LDA difference between two treatments, and their estimates were highly sensitive to the amount of attrition. IPW1 and CC overestimated the absolute LDA difference between the two treatments and the overestimation increased with increasing attrition or when missingness depended on disease activity at 1 year. IPW2 and CARRAC produced unbiased estimations, but IPW2 had a greater sensitivity to the missing pattern of data and the amount of attrition than CARRAC.ConclusionsOnly multiple imputation and IPW2, which considered both confounding and treatment cessation reasons, produced accurate comparative effectiveness estimates.

Predictors of response and survival in a large cohort of 319 Waldenström macroglobulinemia patients treated with ibrutinib monotherapy

Bruton tyrosine kinase (BTK) inhibitors are the only FDA-approved treatments for Waldenström macroglobulinemia (WM). Factors prognostic of survival and predictive of response to BTK inhibitors remained to be clarified. We evaluated 319 patients with WM to identify predictive and prognostic factors on ibrutinib monotherapy. Logistic and Cox proportional-hazard regression models were fitted for response and survival. Multiple imputation analyses were used to address bias associated with missing data. Major (partial response or better) and deep responses (very good partial response or better) were attained in 78% and 28% of patients. CXCR4 mutations were associated with lower odds of major (OR 0.2, 95% CI 0.1-0.5; p<0.001) and deep response (OR 0.3, 95% CI 0.2-0.6; p=0.001). CXCR4 mutations (HR 2.0, 95% CI 1.2-3.4; p=0.01) and platelet count 100 K/uL or less (HR 2.5, 95% CI 1.3-4.9; p=0.007) were associated with worse progression-free survival (PFS). We proposed a scoring system using these two factors. The median PFS for patients with zero, one and two risk factors were not reached, 5 years and 3 years (p<0.001). Patients with two risk factors had HR 2.2 (95% CI 1.3-3.8; p=0.004) compared with one factor, and patients with one factor had HR 2.3 (95% CI 1.1-5.1; p=0.03) compared with zero factors. Age 65 years or older was the only factor associated with overall survival (HR 3.2, 95% CI 1.4-7.0; p=0.005). Multiple imputation analyses did not alter our results. Our study confirms the predictive and prognostic value of CXCR4 mutations in patients with WM treated with ibrutinib monotherapy.

Improvement of random forest by multiple imputation applied to tower crane accident prediction with missing data

PurposeThis research is aimed at predicting tower crane accident phases with incomplete data.Design/methodology/approachThe tower crane accidents are collected for prediction model training. Random forest (RF) is used to conduct prediction. When there are missing values in the new inputs, they should be filled in advance. Nevertheless, it is difficult to collect complete data on construction site. Thus, the authors use multiple imputation (MI) method to improve RF. Finally the prediction model is applied to a case study.FindingsThe results show that multiple imputation RF (MIRF) can effectively predict tower crane accident when the data are incomplete. This research provides the importance rank of tower crane safety factors. The critical factors should be focused on site, because the missing data affect the prediction results seriously. Also the value of critical factors influences the safety of tower crane.Practical implicationThis research promotes the application of machine learning methods for accident prediction in actual projects. According to the onsite data, the authors can predict the accident phase of tower crane. The results can be used for tower crane accident prevention.Originality/valuePrevious studies have seldom predicted tower crane accidents, especially the phase of accident. This research uses tower crane data collected on site to predict the phase of the tower crane accident. The incomplete data collection is considered in this research according to the actual situation.

Multiple Imputation by Chained Equations–K-Nearest Neighbors and Deep Neural Network Architecture for Kidney Disease Prediction

Chronic kidney disease (CKD) is a health concern that affects people all over the world. Kidney dysfunction or impaired kidney functions are the causes of CKD. The machine learning-based prediction models are used to determine the risk level of CKD and assist healthcare practitioners in delaying and preventing the disease’s progression. The researchers proposed many prediction models for determining the CKD risk level. Although these models performed well, their precision is limited since they do not handle missing values in the clinical dataset adequately. The missing values of a clinical dataset can degrade the training outcomes that leads to false predictions. Thus, imputing missing values increases the prediction model performance. This proposed work developed a novel imputation technique by combining Multiple Imputation by Chained Equations and [Formula: see text]-Nearest Neighbors (MICE–KNN) for imputing the missing values. The experimental results show that MICE–KNN accurately predicts the missing values, and the Deep Neural Network (DNN) improves the prediction performance of the CKD model. Various metrics like mean absolute error, accuracy, specificity, Matthews correlation coefficient, the area under the curve, [Formula: see text]-score, sensitivity, and precision have been used to evaluate the proposed CKD model performance. The performance analysis exhibits that MICE–KNN with deep learning outperforms other classifiers. According to our experimental study, the MICE–KNN imputation algorithm with DNN is more appropriate for predicting the kidney disease.

Projections for obesity, smoking and hypertension based on multiple imputation

Aims: Information on the future development of prevalences of risk factors and health indicators is needed to prepare for the forthcoming burden of disease in the population and to allocate resources properly for prevention. We aim to present how multiple imputation can be used flexibly to project future prevalences. Methods: The proposed approach uses data on repeated cross-sectional surveys from different years. We create future samples with age and sex distributions corresponding to the official national population forecasts. Then, the risk factors are simulated using multiple imputation by chained equations. Finally, the imputations are pooled to obtain the prevalences of interest. Covariates, such as sociodemographic variables as well as their possible interactions and non-linear terms, can be included in the modelling. The future development of these covariates is also projected simultaneously. We apply the procedure to data from five Finnish health examination surveys conducted between 1997 and 2017, and project the prevalences of obesity, smoking and hypertension to 2020 and 2025. Results: The prevalence of obesity is projected to increase to 24% for both men and women in 2025. The prevalences of hypertension and smoking are expected to continue decreasing, and the differences between men and women are projected to remain so that men will have higher prevalences. Conclusions: Simulation of future observations by multiple imputation can be used as a flexible yet relatively easy-to-use projection method.