How much missing data is too much? A single study exploration

Missing data (MD) is a common and inevitable problem facing data mining (DM)-based decision systems in e-health since many medical historical datasets contain a huge number of missing values. Therefore, a pre-processing stage is usually required to deal with missing values before building any DM-based decision system. The purpose of this paper is to evaluate the impact of MD techniques on classification systems in cardiovascular dysautonomias diagnosis. We analyzed and compared the accuracy rates of four classification techniques: random forest (RF), support vector machines (SVM), C4.5 decision tree, and Naive Bayes (NB), using two MD techniques: deletion or imputation with k-nearest neighbors (KNN). A total of 216 experiments were therefore carried out using three missingness mechanisms (MCAR: missing completely at random, MAR: missing at random and NMAR: not missing at random), two MD techniques (deletion and KNN imputation), nine MD percentages from 10 to 90% over a dataset collected from the autonomic nervous system (ANS) unit of the University Hospital Avicenne in Morocco. The results obtained suggest that using KNN imputation rather than deletion enhances the accuracy rates of the four classifiers. Moreover, the MD percentages have a negative impact on the performance of classification techniques regardless of the MD mechanisms and MD techniques used. In fact, the accuracy rates of the four classifiers decrease as the MD percentage increases. Graphical abstract.

BackgroundMissing data are a common problem in prospective studies with a long follow-up, and the volume, pattern and reasons for missing data may be relevant when estimating the cost of illness. We aimed to evaluate the effects of different methods for dealing with missing longitudinal cost data and for costing caregiver time on total societal costs in Alzheimer’s disease (AD).MethodsGERAS is an 18-month observational study of costs associated with AD. Total societal costs included patient health and social care costs, and caregiver health and informal care costs. Missing data were classified as missing completely at random (MCAR), missing at random (MAR) or missing not at random (MNAR). Simulation datasets were generated from baseline data with 10–40 % missing total cost data for each missing data mechanism. Datasets were also simulated to reflect the missing cost data pattern at 18 months using MAR and MNAR assumptions. Naïve and multiple imputation (MI) methods were applied to each dataset and results compared with complete GERAS 18-month cost data. Opportunity and replacement cost approaches were used for caregiver time, which was costed with and without supervision included and with time for working caregivers only being costed.ResultsTotal costs were available for 99.4 % of 1497 patients at baseline. For MCAR datasets, naïve methods performed as well as MI methods. For MAR, MI methods performed better than naïve methods. All imputation approaches were poor for MNAR data. For all approaches, percentage bias increased with missing data volume. For datasets reflecting 18-month patterns, a combination of imputation methods provided more accurate cost estimates (e.g. bias: −1 % vs −6 % for single MI method), although different approaches to costing caregiver time had a greater impact on estimated costs (29–43 % increase over base case estimate).ConclusionsMethods used to impute missing cost data in AD will impact on accuracy of cost estimates although varying approaches to costing informal caregiver time has the greatest impact on total costs. Tailoring imputation methods to the reason for missing data will further our understanding of the best analytical approach for studies involving cost outcomes.Electronic supplementary materialThe online version of this article (doi:10.1186/s12874-016-0188-1) contains supplementary material, which is available to authorized users.

BackgroundMissing data due to attrition are rampant in substance abuse clinical trials. However, missing data are often ignored in the presentation of substance abuse clinical trials. This paper demonstrates missing data methods which may be used for hypothesis testing.MethodsMethods involving stratifying and weighting individuals based on missing data pattern are shown to produce tests that are robust to missing data mechanisms in terms of Type I error and power. In this article, we describe several methods of combining data that may be used for testing hypotheses of the treatment effect. Furthermore, illustrations of each test's Type I error and power under different missing data percentages and mechanisms are quantified using a Monte-Carlo simulation study.ResultsType I error rates were similar for each method, while powers depended on missing data assumptions. Specifically, power was greatest for the weighted, compared to un-weighted methods, especially for greater missing data percentages.ConclusionResults of this study as well as extant literature demonstrate the need for standards of design and analysis specific to substance abuse clinical trials. Given the known substantial attrition rates and concern for the missing data mechanism in substance abuse clinical trials, investigators need to incorporate missing data methods a priori. That is, missing data methods should be specified at the outset of the study and not after the data have been collected.

Abstract. The integrity of the Baseline Surface Radiation Network (BSRN) radiation monthly averages are assessed by investigating the impact on monthly means due to the frequency of data gaps caused by missing or discarded high time resolution data. The monthly statistics, especially means, are considered to be important and useful values for climate research, model performance evaluations and for assessing the quality of satellite (time- and space-averaged) data products. The study investigates the spread in different algorithms that have been applied for the computation of monthly means from 1-min values. The paper reveals that the computation of monthly means from 1-min observations distinctly depends on the method utilized to account for the missing data. The intra-method difference generally increases with an increasing fraction of missing data. We found that a substantial fraction of the radiation fluxes observed at BSRN sites is either missing or flagged as questionable. The percentage of missing data is 4.4%, 13.0%, and 6.5% for global radiation, direct shortwave radiation, and downwelling longwave radiation, respectively. Most flagged data in the shortwave are due to nighttime instrumental noise and can reasonably be set to zero after correcting for thermal offsets in the daytime data. The study demonstrates that the handling of flagged data clearly impacts on monthly mean estimates obtained with different methods. We showed that the spread of monthly shortwave fluxes is generally clearly higher than for downwelling longwave radiation. Overall, BSRN observations provide sufficient accuracy and completeness for reliable estimates of monthly mean values. However, the value of future data could be further increased by reducing the frequency of data gaps and the number of outliers. It is shown that two independent methods for accounting for the diurnal and seasonal variations in the missing data permit consistent monthly means to within less than 1 W m−2 in most cases. The authors suggest using a standardized method for the computation of monthly means which addresses diurnal variations in the missing data in order to avoid a mismatch of future published monthly mean radiation fluxes from BSRN. The application of robust statistics would probably lead to less biased results for data records with frequent gaps and/or flagged data and outliers. The currently applied empirical methods should, therefore, be completed by the development of robust methods.

Breast cancer is the most common cancer in women in the United States and constitutes a major source of medical and psychological morbidity. The psychosocial sequelae, while not always severe, can last up to a year after diagnosis. This may be a time when spiritual and social resources become important. It is known that practical and emotional support during treatment for breast cancer benefits the patient's mood and quality of life, but it is unclear as to how spirituality and spiritual well-being may benefit the patient. This study examined this question by exploring the relationships between spirituality, spiritual well-being, physical well-being, functional well-being, mood, and adjustment style in a sample of 191 women recently diagnosed with breast cancer or who had metastatic cancer. The women were asked to complete questionnaires pertaining to the above topics at enrollment to a group intervention study. Measures of spirituality, spiritual well-being, physical well-being, functional well-being, mood, and adjustment style taken at the same point in time were then correlated with each other. There were more significant correlations of spirituality and spiritual well-being with functional well-being than physical well-being, but items pertaining to meaning and peace tended to correlate significantly with physical well-being. Spirituality also correlated significantly with several coping styles, but not avoidance as has been previously suggested. Regression analyses were also performed to find the best combination of variables to predict physical and functional well-being. A combination of social well-being and several questions pertaining to peacefulness accounted for 18% of the variance in physical well-being. However, a combination of social well-being and the spiritual scales accounted for 46% of the variance in functional well-being. When the spiritual scales were examined alone, they accounted for 40% of the variance in functional well-being. The results of this study confirm the importance of spirituality and spiritual well-being in both physical and functional well-being. Asking patients about the role of spirituality in their lives may be a useful marker to predict patient's ability to cope with stress in their lives and of their quality of life.

Recently, there has been much interest in detecting and measuring patterns of change in disparity. Although most studies have used one or two measures of disparity to quantify and characterize the occupation of morphospace, multiple measures may be necessary to fully detect changes in patterns of morphospace occupation. Also, the ability to detect morphological trends and occupation patterns within morphospace depends on using the appropriate measure(s) of disparity. In this study, seven measures were used to determine and characterize sensitivity to sample size of the data, number of morphological characters, percentage of missing data, and changes in morphospace occupation pattern. These consist of five distance measures—sum of univariate variances, total range, mean distance, principal coordinate analysis volume, average pairwise dissimilarity—and two non-distance measures—participation ratio and number of unique pairwise character combinations. Evaluation of each measure with respect to sensitivity to sample size, number of morphological characters, and percentage of missing data was accomplished by using both simulated and Ordovician crinoid data. For simulated data, each measure of disparity was evaluated for its response to changes of morphospace occupation pattern, and with respect to simulated random and nonrandom extinction events. Changes in disparity were also measured within the Crinoidea across the Permian extinction event.Although all measures vary in sensitivity with respect to species sample size, number of morphological characters, and percentage of missing data, the non-distance measures overall produce the lowest estimates of variance (in bootstrap analyses). The non-distance measures appear to be relatively insensitive to changes in morphospace occupation pattern. All measures, except average pairwise dissimilarity, detect changes in occupation pattern in simulated nonrandom extinction events, but all measures, except number of unique pairwise character combinations and principal coordinate analysis volume, are relatively insensitive to changes in pattern in simulated random extinction events. The distance measures report similar changes in disparity over the Permian extinction event, whereas the non-distance measures differ. This study suggests that each measure of disparity is designed for different purposes, and that by using a combination of techniques a clearer picture of disparity should emerge.

Background Mortality rates after hospitalisation for acute myocardial infarction (AMI) have fallen substantially, but vary by hospital. Whilst previous studies have reported the extent to which the residual (unexplained) variance in outcomes from AMI can be attributed to patients and hospitals, there are no reports quantifying hospital level residual variation over time. Table 1 Patient characteristics and data misingness for 2004 2004 STEMI NSTEMI Missing n=81669 Male (n=22858) Female (n=9385) Male (n=31026) Female (n=18299) 0.1% Mean age in years (SD) 62.8 (13.1) 71.4 (13.6) 69.4 (13.3) 75.8 (12.6) 0.3% Mean SBP (SD) 134.3 (27.5) 133.6 (29.9) 140.2 (27.2) 142.9 (29.7) 15.7% Mean heart rate (SD) 77.7 (19.7) 80.7 (21.6) 81.2 (22.3) 85.7 (22.9) 15.8% ST deviation (%) 20665 (94.2%) 8322 (92.5%) 7884 (26.7%) 4583 (26.3%) 4.6% Elevated enzymes (%) 21601 (94.5%) 8903 (94.9%) 29356 (94.6%) 17297 (94.5%) 0.0% Cardiac arrest (%) 2302 (10.8%) 992 (11.3%) 1026 (3.4%) 584 (3.3%) 4.1% Chronic renal failure (%) 583 (3.0%) 327 (4.1%) 2574 (8.8%) 1516 (8.8%) 9.4% Loop diuretic (%) 3059 (16.8%) 1909 (25.6%) 7798 (28.5%) 6308 (38.7%) 15.1% Methods We studied 698 092 patients admitted to 247 hospitals in England and Wales from 2004 through 2010 inclusive with ST-elevation myocardial infarction (STEMI) and non ST-elevation myocardial infarction (NSTEMI). Multiple imputation (MI) was used to handle missing data. Bayesian multi-level modelling was applied to each year of data standardising for all-cause in-hospital and 30-day mortality using covariates from the adjusted mini-GRACE risk score. Hospital volume effects were modelled. Hospital-level effects were, therefore, the proportion of variance in mortality explained by the hospital (but not attributable to case mix). Models were fitted using an uninformative half-Cauchy prior distribution over 100 000 MCMC iterations, the posterior distributions of 15 multiply imputed datasets pooled and the hospital-level effects reported with 95% credible limits (CI). Results Tables 1 and 2 show the patient characteristics by STEMI and NSTEMI, along with the percentages of missing data before multiple imputation, for 2004 and 2010. The IQR of adjusted 30-day mortality rates was 8.3% to 12.6% in 2004 and 5.5% to 8.5% in 2010. Table 2 Patient characteristics and data missingness for 2010 2010 STEMI NSTEMI Missing n=72845 Male (n=20421) Female (n=9352) Male (n=25936) Female (16393) 1.0% Mean age in years (SD) 64 (12.8) 72.5 (12.5) 69.5 (12.9) 76 (11.9) 0.3%

Studies in many different fields of research suffer from the problem of missing data. With missing data, statistical tests will lose power, results may be biased, or analysis may not be feasible at all. There are several ways to handle the problem, for example through imputation. With imputation, missing values are replaced with estimated values according to an imputation method or model. In the k-nearest neighbour (k-NN) method, a case is imputed using values from the k most similar cases. In this paper, we present an evaluation of the k-NN method using Likert data in a software engineering context. We simulate the method with different values of k and for different percentages of missing data. Our findings indicate that it is feasible to use the k-NN method with Likert data. We suggest that a suitable value of k is approximately the square root of the number of complete cases. We also show that by relaxing the method rules with respect to selecting neighbours, the ability of the method remains high for large amounts of missing data without affecting the quality of the imputation.

Quality of Life and Depression in CKD: Improving Hope and Health

Road link traffic speed pattern mining in probe vehicle data via soft computing techniques

Trial Quality in Nephrology: How Are We Measuring Up?

Objective: Data being missed is an unavoidable problem in cohort studies. This paper compares the imputation effect of eight common missing data imputation methods involved in cutting longitudinal data through simulation study to provide a valuable reference for the treatment of missing data in longitudinal studies. Methods: The simulation study is based on R language software and generates missing longitudinal data by the Monte Carlo method. By comparing the average absolute deviation, average relative deviation, and TypeⅠerror from the regression analysis of different imputation methods, the imputation effect of varying imputation methods on missing longitudinal data and the influence on subsequent multivariate analysis are evaluated. Results: The mean imputation, k nearest neighbor (KNN), regression imputation, and random forest all have a similar imputation effect, which is also steady. However, the hot deck is inferior to the above imputation methods. K-means clustering and expectation maximization (EM) algorithm are among the worst and unstable. Mean imputation, EM algorithm, random forest, KNN, and regression imputation can control TypeⅠerror. Still, multiple imputations, hot deck, and K-means clustering cannot effectively manage the TypeⅠerror. Conclusions: For missing data in longitudinal studies, mean imputation, KNN, regression imputation, and random forest can be used as better imputation methods under the mechanism of missing at random. When the missing ratio is not too large, multiple imputations and hot deck can also perform well, but K-means clustering and EM algorithm are not recommended.

Gene expression data often contain missing expression values. Effective missing value estimation methods are needed since many algorithms for gene expression data analysis require a complete matrix of gene array values. In this paper, imputation methods based on the least squares formulation are proposed to estimate missing values in the gene expression data, which exploit local similarity structures in the data as well as least squares optimization process. The proposed local least squares imputation method (LLSimpute) represents a target gene that has missing values as a linear combination of similar genes. The similar genes are chosen by k-nearest neighbors or k coherent genes that have large absolute values of Pearson correlation coefficients. Non-parametric missing values estimation method of LLSimpute are designed by introducing an automatic k-value estimator. In our experiments, the proposed LLSimpute method shows competitive results when compared with other imputation methods for missing value estimation on various datasets and percentages of missing values in the data. The software is available at http://www.cs.umn.edu/~hskim/tools.html hpark@cs.umn.edu

Estimates of longitudinal change in the parameters of latent (i.e. unobserved) variables, including means, are affected by non-response on the items or indicators of the latent variable. This study used Monte Carlo simulation and a numeric example to compare four ordinal item non-response methods: non-negative matrix factorization (NNMF), multiple imputation with conditional proportional odds model (POM), full information maximum likelihood (FIML) and complete-case analysis, when estimating the longitudinal change in latent variable means. The mean squared error for the NNMF method was more than 40% lower than for the FIML and POM methods when the latent variable correlations over time were strong, percentage of missing data was 25% or more, and sample size was large. The NNMF method is a promising method to address item non-response. It is relatively efficient when sample size is large, and the percentage of missing data is high but has limitations under other data-analytic conditions.

Background: In Israel all residents are covered by universal health care and yet disparities in health care access and utilization exist. Jerusalem is the largest and poorest city in Israel, with significant with significant Jewish ultra-orthodox and Arab communities. These groups have shared reproductive factors, similar patriarchal community structures and often live in poverty. No research has been performed on these populations in the advanced breast cancer (ABC) setting with respect to quality of life (QOL) and psychosocial wellbeing. Methods: Women with ABC were prospectively recruited after consenting and self-identified their religion & level of religiosity. The following measurement tools were implemented: QOL - Functional Assessment of Cancer Therapy-Breast (FACT-B) Version 4, Spiritual wellbeing - The FACIT-Sp, Depression and Anxiety - Patient-Reported Outcomes Measurement Information System and demographic questionnaire. Comparisons between groups were conducted and the role of religion and depression in the explanation of differences of QOL between the groups was tested by moderated mediation model. Results: 160 ABC patients were recruited. 39 (24.4%) identified themselves as Ultra-Orthodox Jews, 35 (21.9%) as Arabs and 81 (50.6%) as non-ultra-orthodox Jewish. Groups differed in their health related QOL(F(2,150) = 10.49, p<.001); Spiritual wellbeing (F(2,150) = 26.09, p<.001p<.001) and depression (F(2,134) = 9.61, p<.001). Ultra-Orthodox patients reported higher levels of QOL and spiritual wellbeing compared to the other groups and lower levels of depression compared to the Arab patients. Arab patients reported lower levels of QOL and higher levels of depression compared to the other groups. In addition, they reported lower levels of spiritual wellbeing compared to Ultra-Orthodox patents. There were significant differences between groups in physical (F(2,152) = 13.96, p<.001); functional (F(2,151) = 7.12, p<.001); emotional (F(2,151) = 8.81, p<.001) and breast cancer wellbeing (F(2,152) = 10.21, p<.001) as measured by FACT-B. Throughout the different domains of QOL Arab patients did worse (except the social/family subscale) than the Ultra – Orthodox patients. In addition, they reported on lower levels of QOL than the general population in total QOL, physical and functional wellbeing. Ultra – orthodox patients reported on higher levels of QOL in comparison to the general population in Total QOL, emotional well- being and breast cancer wellbeing. The study’s moderated mediation model was found to be significant (F (2,126) = 121.16, R-sq = 65.8%, p < 0.000). The differences between Arab and Jewish patients in Health related QOL were found to be mediated by depression, with group-depression relationship moderated by religiosity levels. Conclusions: Significant health care disparities exist amongst women with ABC in terms of QOL, psychosocial and spiritual wellbeing. Although the current study examined two minority groups who have shared low socio-economic status, similar reproductive habits (multiple pregnancies from an early age), paternalistic and patriarchal community structure and challenges in health care access for cultural-ethnic-social reasons, different patterns of well-being were found. Ultra-Orthodox patients reported superior health related QOL, spiritual wellbeing, and lower levels of depression compared to other groups. Arab patients reported poorer health related QOL, higher levels of depression and lower levels of spiritual wellbeing and religiosity compared to Ultra-Orthodox patients. Further research is needed to create effective, culturally appropriate interventions. Citation Format: Michal Braun, Gil Goldzweig, Noa Shafri, Yehosua Akerman, Daniel Yusovich, Inbal Fuchs, Shani Paluch-Shimon. Quality of life and psychosocial disparities between ethnically and religiously diverse population with advanced breast cancer [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2024; 2024 Dec 10-13; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(12 Suppl):Abstract nr P5-04-11.