Multiple imputation of completely missing repeated measures data within person from a complex sample: application to accelerometer data in the National Health and Nutrition Examination Survey

Attention Aware Deep Learning Approaches for an Efficient Stress Classification Model

In today's world, stress is a major factor for various diseases in modern societies which affects the day-to-day activities of human beings. The measurement of stress is a contributing factor for governments and societies that impacts the quality of daily lives. The strategy of stress monitoring systems requires an accurate stress classification technique which is identified via the reactions of the body to regulate itself to changes within the environment through mental and emotional responses. Therefore, this research proposed a novel deep learning approach for the stress classification system. In this paper, we presented an Enhanced Long Short-Term Memory(E-LSTM) based on the feature attention mechanism that focuses on determining and categorizing the stress polarity using sequential modeling and word-feature seizing. The proposed approach integrates pre-feature attention in E-LSTM to identify the complicated relationship and extract the keywords through an attention layer for stress classification. This research has been evaluated using a selected dataset accessed from the sixth Korea National Health and Nutrition Examination Survey conducted from 2013 to 2015 (KNHANES VI) to analyze health-related stress data. Statistical performance of the developed approach was analyzed based on the nine features of stress detection, and we compared the effectiveness of the developed approach with other different stress classification approaches. The experimental results shown that the developed approach obtained accuracy, precision, recall and a F1-score of 75.54%, 74.26%, 72.99% and 74.58%, respectively. The feature attention mechanism-based E-LSTM approach demonstrated superior performance in stress detection classification when compared to other classification methods including naïve Bayesian, SVM, deep belief network, and standard LSTM. The results of this study demonstrated the efficiency of the proposed approach in accurately classifying stress detection, particularly in stress monitoring systems where it is expected to be effective for stress prediction.

Effects of stopping criterion on the growth of trees in regression random forests

Random forests are a powerful machine learning tool that capture complex relationships between independent variables and an outcome of interest. Trees built in a random forest are dependent on several hyperparameters, one of the more critical being the node size. The original algorithm of Breiman, controls for node size by limiting the size of the parent node, so that a node cannot be split if it has less than a specified number of observations. We propose that this hyperparameter should instead be defined as the minimum number of observations in each terminal node. The two existing random forest approaches are compared in the regression context based on estimated generalization error, bias-squared, and variance of resulting predictions in a number of simulated datasets. Additionally the two approaches are applied to type 2 diabetes data obtained from the National Health and Nutrition Examination Survey. We have developed a straightforward method for incorporating weights into the random forest analysis of survey data. Our results demonstrate that generalization error under the proposed approach is competitive to that attained from the original random forest approach when data have large random error variability. The R code created from this work is available and includes an illustration.

A SEMIPARAMETRIC MULTIPLE IMPUTATION APPROACH TO FULLY SYNTHETIC DATA FOR COMPLEX SURVEYS

Data synthesis is an effective statistical approach for reducing data disclosure risk. Generating fully synthetic data might minimize such risk, but its modeling and application can be difficult for data from large, complex surveys. This article extended the two-stage imputation to simultaneously impute item missing values and generate fully synthetic data. A new combining rule for making inferences using data generated in this manner was developed. Two semiparametric missing data imputation models were adapted to generate fully synthetic data for skewed continuous variable and sparse binary variable, respectively. The proposed approach was evaluated using simulated data and real longitudinal data from the Health and Retirement Study. The proposed approach was also compared with two existing synthesis approaches: (1) parametric regressions models as implemented in IVEware; and (2) nonparametric Classification and Regression Trees as implemented in synthpop package for R using real data. The results show that high data utility is maintained for a wide variety of descriptive and model-based statistics using the proposed strategy. The proposed strategy also performs better than existing methods for sophisticated analyses such as factor analysis.

Association of daily step count and serum testosterone among men in the United States

PURPOSE

To describe the association between daily activity (i.e., daily step counts and accelerometer intensity measures) and serum TT levels in a representative sample of US adults aged 18 years or older.

METHODS

A retrospective cohort study was carried out utilizing the NHANES (National Health and Nutrition Examination Survey) 2003-2004 cycle. Physical activity was measured with a waist-worn uniaxial accelerometer (AM-7164; ActiGraph) for up to 7 days using a standardized protocol. Using linear and multivariable logistic regression controlling for relevant social, demographic, lifestyle, and comorbidity characteristics, we assessed the association between daily step counts and TT.

RESULTS

A total of 279 subjects with a median age 46 (IQR: 33-56) were included in the analysis. 23.3% of the cohort had a low serum TT level (TT < 350 ng/dl). Compared to men who took <4000 steps per day, men who took >4000 or >8000 steps/day had a lower odd of being hypogonadal (OR 0.14, 95% CI: 0.07-0.49 and 0.08, 95%CI: 0.02-0.44, respectively). While a threshold effect was noted on average, TT increased 7 ng/dL for each additional 1000 steps taken daily (β-estimate: 0.007, 95% CI: 0.002-0.013).

CONCLUSIONS

Patients with the lowest daily step counts had higher odds of being hypogonadal. The current work supports a possible association between daily steps, total testosterone, and hypogonadism for men in the US.

Data Imputation and Body Weight Variability Calculation Using Linear and Nonlinear Methods in Data Collected From Digital Smart Scales: Simulation and Validation Study

Background

Body weight variability (BWV) is common in the general population and may act as a risk factor for obesity or diseases. The correct identification of these patterns may have prognostic or predictive value in clinical and research settings. With advancements in technology allowing for the frequent collection of body weight data from electronic smart scales, new opportunities to analyze and identify patterns in body weight data are available.

Objective

This study aims to compare multiple methods of data imputation and BWV calculation using linear and nonlinear approaches

Methods

In total, 50 participants from an ongoing weight loss maintenance study (the NoHoW study) were selected to develop the procedure. We addressed the following aspects of data analysis: cleaning, imputation, detrending, and calculation of total and local BWV. To test imputation, missing data were simulated at random and using real patterns of missingness. A total of 10 imputation strategies were tested. Next, BWV was calculated using linear and nonlinear approaches, and the effects of missing data and data imputation on these estimates were investigated.

Results

Body weight imputation using structural modeling with Kalman smoothing or an exponentially weighted moving average provided the best agreement with observed values (root mean square error range 0.62%-0.64%). Imputation performance decreased with missingness and was similar between random and nonrandom simulations. Errors in BWV estimations from missing simulated data sets were low (2%-7% with 80% missing data or a mean of 67, SD 40.1 available body weights) compared with that of imputation strategies where errors were significantly greater, varying by imputation method.

Conclusions

The decision to impute body weight data depends on the purpose of the analysis. Directions for the best performing imputation methods are provided. For the purpose of estimating BWV, data imputation should not be conducted. Linear and nonlinear methods of estimating BWV provide reasonably accurate estimates under high proportions (80%) of missing data.

A novel scaling methodology to reduce the biases associated with missing data from commercial activity monitors

Background

Commercial physical activity monitors have wide utility in the assessment of physical activity in research and clinical settings, however, the removal of devices results in missing data and has the potential to bias study conclusions. This study aimed to evaluate methods to address missingness in data collected from commercial activity monitors.

Methods

This study utilised 1526 days of near complete data from 109 adults participating in a European weight loss maintenance study (NoHoW). We conducted simulation experiments to test a novel scaling methodology (NoHoW method) and alternative imputation strategies (overall/individual mean imputation, overall/individual multiple imputation, Kalman imputation and random forest imputation). Methods were compared for hourly, daily and 14-day physical activity estimates for steps, total daily energy expenditure (TDEE) and time in physical activity categories. In a second simulation study, individual multiple imputation, Kalman imputation and the NoHoW method were tested at different positions and quantities of missingness. Equivalence testing and root mean squared error (RMSE) were used to evaluate the ability of each of the strategies relative to the true data.

Results

The NoHoW method, Kalman imputation and multiple imputation methods remained statistically equivalent (p<0.05) for all physical activity metrics at the 14-day level. In the second simulation study, RMSE tended to increase with increased missingness. Multiple imputation showed the smallest RMSE for Steps and TDEE at lower levels of missingness (<19%) and the Kalman and NoHoW methods were generally superior for imputing time in physical activity categories.

Conclusion

Individual centred imputation approaches (NoHoW method, Kalman imputation and individual Multiple imputation) offer an effective means to reduce the biases associated with missing data from activity monitors and maximise data retention.

Association of Daily Step Count and Step Intensity With Mortality Among US Adults

IMPORTANCE

It is unclear whether the number of steps per day and the intensity of stepping are associated with lower mortality.

OBJECTIVE

Describe the dose-response relationship between step count and intensity and mortality.

DESIGN, SETTING, AND PARTICIPANTS

Representative sample of US adults aged at least 40 years in the National Health and Nutrition Examination Survey who wore an accelerometer for up to 7 days ( from 2003-2006). Mortality was ascertained through December 2015.

EXPOSURES

Accelerometer-measured number of steps per day and 3 step intensity measures (extended bout cadence, peak 30-minute cadence, and peak 1-minute cadence [steps/min]). Accelerometer data were based on measurements obtained during a 7-day period at baseline.

MAIN OUTCOMES AND MEASURES

The primary outcome was all-cause mortality. Secondary outcomes were cardiovascular disease (CVD) and cancer mortality. Hazard ratios (HRs), mortality rates, and 95% CIs were estimated using cubic splines and quartile classifications adjusting for age; sex; race/ethnicity; education; diet; smoking status; body mass index; self-reported health; mobility limitations; and diagnoses of diabetes, stroke, heart disease, heart failure, cancer, chronic bronchitis, and emphysema.

RESULTS

A total of 4840 participants (mean age, 56.8 years; 2435 [54%] women; 1732 [36%] individuals with obesity) wore accelerometers for a mean of 5.7 days for a mean of 14.4 hours per day. The mean number of steps per day was 9124. There were 1165 deaths over a mean 10.1 years of follow-up, including 406 CVD and 283 cancer deaths. The unadjusted incidence density for all-cause mortality was 76.7 per 1000 person-years (419 deaths) for the 655 individuals who took less than 4000 steps per day; 21.4 per 1000 person-years (488 deaths) for the 1727 individuals who took 4000 to 7999 steps per day; 6.9 per 1000 person-years (176 deaths) for the 1539 individuals who took 8000 to 11 999 steps per day; and 4.8 per 1000 person-years (82 deaths) for the 919 individuals who took at least 12 000 steps per day. Compared with taking 4000 steps per day, taking 8000 steps per day was associated with significantly lower all-cause mortality (HR, 0.49 [95% CI, 0.44-0.55]), as was taking 12 000 steps per day (HR, 0.35 [95% CI, 0.28-0.45]). Unadjusted incidence density for all-cause mortality by peak 30 cadence was 32.9 per 1000 person-years (406 deaths) for the 1080 individuals who took 18.5 to 56.0 steps per minute; 12.6 per 1000 person-years (207 deaths) for the 1153 individuals who took 56.1 to 69.2 steps per minute; 6.8 per 1000 person-years (124 deaths) for the 1074 individuals who took 69.3 to 82.8 steps per minute; and 5.3 per 1000 person-years (108 deaths) for the 1037 individuals who took 82.9 to 149.5 steps per minute. Greater step intensity was not significantly associated with lower mortality after adjustment for total steps per day (eg, highest vs lowest quartile of peak 30 cadence: HR, 0.90 [95% CI, 0.65-1.27]; P value for trend = .34).

CONCLUSIONS AND RELEVANCE

Based on a representative sample of US adults, a greater number of daily steps was significantly associated with lower all-cause mortality. There was no significant association between step intensity and mortality after adjusting for total steps per day.

Chronic Disease Prediction Using Character-Recurrent Neural Network in The Presence of Missing Information

The aim of this study was to predict chronic diseases in individual patients using a character-recurrent neural network (Char-RNN), which is a deep learning model that treats data in each class as a word when a large portion of its input values is missing. An advantage of Char-RNN is that it does not require any additional imputation method because it implicitly infers missing values considering the relationship with nearby data points. We applied Char-RNN to classify cases in the Korea National Health and Nutrition Examination Survey (KNHANES) VI as normal status and five chronic diseases: hypertension, stroke, angina pectoris, myocardial infarction, and diabetes mellitus. We also employed a multilayer perceptron network for the same task for comparison. The results show higher accuracy for Char-RNN than for the conventional multilayer perceptron model. Char-RNN showed remarkable performance in finding patients with hypertension and stroke. The present study utilized the KNHANES VI data to demonstrate a practical approach to predicting and managing chronic diseases with partially observed information.

Environmental and physical factors predisposing middle-aged and older Japanese adults to falls and fall-related fractures in the home

AIM

To identify environmental and physical factors that predispose middle-aged and older Japanese adults to falls and fall-related fractures in the home.

METHODS

A cross-sectional survey was carried out in 2014. Self-administered questionnaires were distributed to 15 000 community-dwelling adults in Japan. The overall crude response rate was 13%. Response data were analyzed from 1561 individuals aged ≥40 years using multiple imputation to analyze missing data. We evaluated falls without fractures and fall-related fractures during the previous 3 years according to demographic, physical and environmental factors, including age, sex, long-term care insurance certification, type of house and barrier-free housing.

RESULTS

Of the 1561 adults (mean age 68.1 ± 13.0 years), 28% experienced a fall in the home. Among the individuals who experienced a fall, 11% experienced fall-related fractures. These individuals were more likely to be women (OR 2.4, 95.0% CI 1.1-5.1), have LTCI certification (OR 3.9, 95.0% CI 1.6-9.4) and be living in a barrier home (OR 4.0, 95.0% CI 1.6-9.8), after adjustment for covariates.

CONCLUSIONS

Environmental factors, such as living in a barrier home, are critical for fall-related fractures, in addition to demographic and physical factors. A multidisciplinary approach that considers both physical and environmental factors is necessary for reducing the incidence of fall-related fractures among middle-aged and older Japanese adults. Geriatr Gerontol Int 2018; 18: 1372-1377.

Longitudinal associations of high-density lipoprotein cholesterol or low-density lipoprotein cholesterol with metabolic syndrome in the Chinese population: a prospective cohort study

OBJECTIVE

Currently, most studies only reveal the relationship between baseline high-density lipoprotein cholesterol (HDL-c) or low-density lipoprotein cholesterol (LDL-c) levels and metabolic syndrome (MetS). The relationship between dynamic changes in HDL-c or LDL-c and MetS remains unclear. We aimed to gain a deeper understanding of the relationship between the dynamic changes in HDL-c or LDL-c and MetS.

DESIGN

A prospective study.

SETTING

The Medical Centre of the Second Hospital affiliated with Dalian Medical University from 2010 to 2016.

PARTICIPANTS

A total of 4542 individuals who were initially MetS-free and completed at least two follow-up examinations as part of the longitudinal population were included.

METHODS

The Joint Interim Statement criteria 2009 were used to define MetS. We used the Joint model to estimate the relative risks (RRs) of incident MetS.

RESULTS

The cumulative incidence of MetS was 17.81% and was 14.86% in men and 5.36% in women during the 7 years of follow-up. In the Joint models, the RRs of the longitudinal decrease in HDL-c and the longitudinal increase in LDL-c for the development of MetS were 18.8781-fold (95% CI 12.5156 to 28.4900) and 1.3929-fold (95% CI 1.2283 to 1.5795), respectively.

CONCLUSIONS

The results highlight that the dynamic longitudinal decrement of HDL-c or the increment of LDL-c is associated with an elevated risk of MetS.