Multiple imputation methods for handling missing values in longitudinal studies with sampling weights: Comparison of methods implemented in Stata

Temporal trends and disparities of population attributable fractions of modifiable risk factors for dementia in China: a time-series study of the China health and retirement longitudinal study (2011-2018)

Background

In China, dementia poses a significant public health challenge, exacerbated by an ageing population and lifestyle changes. This study assesses the temporal trends and disparities in the population-attributable fractions (PAFs) of modifiable risk factors (MRFs) for new-onset dementia from 2011 to 2018.

Methods

We used data from the China Health and Retirement Longitudinal Study (CHARLS), covering 75,214 person-waves. We calculated PAFs for 12 MRFs identified by the Lancet Commission (including six early-to mid-life factors and six late-life factors). We also determined the individual weighted PAFs (IW-PAFs) for each risk factor. Subgroup analyses were conducted by sex, socio-economic status (SES), and geographic location.

Findings

The overall PAF for dementia MRFs had a slight increase from 45.36% in 2011 to 52.46% in 2018, yet this change wasn't statistically significant. During 2011-2018, the most contributing modifiable risk was low education (average IW-PAF 11.3%), followed by depression, hypertension, smoking, and physical inactivity. Over the eight-year period, IW-PAFs for risk factors like low education, hypertension, hearing loss, smoking, and air pollution showed decreasing trends, while others increased, but none of these changes were statistically significant. Sex-specific analysis revealed higher IW-PAFs for traumatic brain injury (TBI), social isolation, and depression in women, and for alcohol and smoking in men. The decline in IW-PAF for men's hearing loss were significant. Lower-income individuals had higher overall MRF PAFs, largely due to later-life factors like depression. Early-life factors, such as TBI and low education, also contributed to SES disparities. Rural areas reported higher overall MRF PAFs, driven by factors like depression, low education, and hearing loss. The study also found that the gap in MRF PAFs across different SES groups or regions either remained constant or increased over the study period.

Interpretation

The study reveals a slight but non-significant increase in dementia's MRF PAF in China, underscoring the persistent relevance of these risk factors. The findings highlight the need for targeted public health strategies, considering the demographic and regional differences, to effectively tackle and reduce dementia risk in China's diverse population.

Funding

This work was supported by the PKU Young Scholarship in Global Health and Development.

Lifetime costs of overweight and obesity in Italy

We use longitudinal electronic clinical data on a large representative sample of the Italian population to estimate the lifetime profile costs of different BMI classes - normal weight, overweight, and obese (I, II, and III) - in a primary care setting. Our research reveals that obese patients generate the highest cost differential throughout their lives compared to normal weight patients. Moreover, we show that overweight individuals spend less than those with normal weight, primarily due to reduced expenditures beginning in early middle age. Our estimates could serve as a vital benchmark for policymakers looking to prioritize public interventions that address the obesity pandemic while considering the increasing obesity rates projected by the OECD until 2030.

A comparison of strategies for selecting auxiliary variables for multiple imputation

Multiple imputation (MI) is a popular method for handling missing data. Auxiliary variables can be added to the imputation model(s) to improve MI estimates. However, the choice of which auxiliary variables to include is not always straightforward. Several data-driven auxiliary variable selection strategies have been proposed, but there has been limited evaluation of their performance. Using a simulation study we evaluated the performance of eight auxiliary variable selection strategies: (1, 2) two versions of selection based on correlations in the observed data; (3) selection using hypothesis tests of the "missing completely at random" assumption; (4) replacing auxiliary variables with their principal components; (5, 6) forward and forward stepwise selection; (7) forward selection based on the estimated fraction of missing information; and (8) selection via the least absolute shrinkage and selection operator (LASSO). A complete case analysis and an MI analysis using all auxiliary variables (the "full model") were included for comparison. We also applied all strategies to a motivating case study. The full model outperformed all auxiliary variable selection strategies in the simulation study, with the LASSO strategy the best performing auxiliary variable selection strategy overall. All MI analysis strategies that we were able to apply to the case study led to similar estimates, although computational time was substantially reduced when variable selection was employed. This study provides further support for adopting an inclusive auxiliary variable strategy where possible. Auxiliary variable selection using the LASSO may be a promising alternative when the full model fails or is too burdensome.

On the use of multiple imputation to address data missing by design as well as unintended missing data in case-cohort studies with a binary endpoint

BACKGROUND

Case-cohort studies are conducted within cohort studies, with the defining feature that collection of exposure data is limited to a subset of the cohort, leading to a large proportion of missing data by design. Standard analysis uses inverse probability weighting (IPW) to address this intended missing data, but little research has been conducted into how best to perform analysis when there is also unintended missingness. Multiple imputation (MI) has become a default standard for handling unintended missingness and is typically used in combination with IPW to handle the intended missingness due to the case-control sampling. Alternatively, MI could be used to handle both the intended and unintended missingness. While the performance of an MI-only approach has been investigated in the context of a case-cohort study with a time-to-event outcome, it is unclear how this approach performs with a binary outcome.

METHODS

We conducted a simulation study to assess and compare the performance of approaches using only MI, only IPW, and a combination of MI and IPW, for handling intended and unintended missingness in the case-cohort setting. We also applied the approaches to a case study.

RESULTS

Our results show that the combined approach is approximately unbiased for estimation of the exposure effect when the sample size is large, and was the least biased with small sample sizes, while MI-only and IPW-only exhibited larger biases in both sample size settings.

CONCLUSIONS

These findings suggest that a combined MI/IPW approach should be preferred to handle intended and unintended missing data in case-cohort studies with binary outcomes.

Keeping kin close? Geographies of family networks by race and income, 1981-2017

Objective

This study examined changes in geographic proximity to family members among race and income groups in the United States from 1981 to 2017.

Background

Close geographic proximity to family members can facilitate mutual support and strengthen family bonds. Some scholars argue that institutional sources of support have replaced many core family functions, which might mean that households are likely to live increasingly farther away from family. Advancing technology and changing labor market opportunities might reinforce this pattern. Yet, the ongoing cultural and emotional salience of family might curtail the effects of these factors on the increasing distance to family.

Method

We conducted a quantitative analysis of longitudinal data from the Panel Study of Income Dynamics (PSID). We utilized the multigenerational structure of the PSID and restricted-use geocodes to map kin proximity at every interview from 1981 to 2017. We cross-classified our sample by race and income, focusing on Black and White respondents across income quartiles (n = 171,501 person-periods).

Results

High-income White respondents showed the greatest increases in distance from kin over time, whereas proximity to kin among other race-income groups was relatively stable.

Conclusion

Proximate kin has become less central in the lives of high-income White households over time, whereas close proximity to kin has been the norm over time for other racial and income groups. These results have implications for racial and income differences in kin relations over time.

Repercussions of absolute and time-rated BMI "yo-yo" fluctuations on cardiovascular stress-related morbidities within the vascular-metabolic CUN cohort

AIMS

The association between body mass index (BMI) fluctuation and BMI fluctuation rate with cardiovascular stress morbidities in a Caucasian European cohort was evaluated to ascertain the impact of weight cycling.

METHODS

A total of 4,312 patients of the Vascular-Metabolic CUN cohort (VMCUN cohort) were examined and followed up during 9.35 years ( ± 4.39). Cox proportional hazard ratio analyses were performed to assess the risk of developing cardiovascular stress-related diseases (CVDs) across quartiles of BMI fluctuation, measured as the average successive variability (ASV) (ASV = |BMIt0 - BMIt1| + |BMIt1 - BMIt2| + |BMIt2-BMIt3| +…+ |BMItn - 1 - BMItn|/n - 1), and quartiles of BMI fluctuation rate (ASV/year).

RESULTS

There were 436 incident cases of CVD-associated events involving 40,323.32 person-years of follow-up. A progressively increased risk of CVD in subjects with greater ASV levels was found. Also, a higher level of ASV/year was significantly associated with an increased risk of developing CVD stress independent of confounding factors with a value of 3.71 (95% CI: 2.71-5.07) for those in the highest quartile and 1.82 (95% CI: 1.33-2.50) for those in the third quartile.

CONCLUSIONS

The BMI fluctuation rate seems to be a better predictor than BMI fluctuation of the potential development of cardiovascular stress morbidities. The time-rated weight fluctuations are apparently more determinant in increasing the risk of a CVD than the weight fluctuation itself, which is remarkable in subjects under "yo-yo" weight patterns for precision medicine.

Standard multiple imputation of survey data didn't perform better than simple substitution in enhancing an administrative dataset: the example of self-rated health in England

Background

Health surveys provide a rich array of information but on relatively small numbers of individuals and evidence suggests that they are becoming less representative as response levels fall. Routinely collected administrative data offer more extensive population coverage but typically comprise fewer health topics. We explore whether data combination and multiple imputation of health variables from survey data is a simple and robust way of generating these variables in the general population.

Methods

We use the UK Integrated Household Survey and the English 2011 population census both of which included self-rated general health. Setting aside the census self-rated health data we multiply imputed self-rated health responses for the census using the survey data and compared these with the actual census results in 576 unique groups defined by age, sex, housing tenure and geographic region.

Results

Compared with original census data across the groups, multiply imputed proportions of bad or very bad self-rated health were not a markedly better fit than those simply derived from the survey proportions.

Conclusion

While multiple imputation may have the potential to augment population data with information from surveys, further testing and refinement is required.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12982-021-00099-z.

A survey on missing data in machine learning

Machine learning has been the corner stone in analysing and extracting information from data and often a problem of missing values is encountered. Missing values occur because of various factors like missing completely at random, missing at random or missing not at random. All these may result from system malfunction during data collection or human error during data pre-processing. Nevertheless, it is important to deal with missing values before analysing data since ignoring or omitting missing values may result in biased or misinformed analysis. In literature there have been several proposals for handling missing values. In this paper, we aggregate some of the literature on missing data particularly focusing on machine learning techniques. We also give insight on how the machine learning approaches work by highlighting the key features of missing values imputation techniques, how they perform, their limitations and the kind of data they are most suitable for. We propose and evaluate two methods, the k nearest neighbor and an iterative imputation method (missForest) based on the random forest algorithm. Evaluation is performed on the Iris and novel power plant fan data with induced missing values at missingness rate of 5% to 20%. We show that both missForest and the k nearest neighbor can successfully handle missing values and offer some possible future research direction.