Avoiding bias due to perfect prediction in multiple imputation of incomplete categorical variables

Analyzing Coarsened and Missing Data by Imputation Methods

In various missing data problems, values are not entirely missing, but are coarsened. For coarsened observations, instead of observing the true value, a subset of values - strictly smaller than the full sample space of the variable - is observed to which the true value belongs. In our motivating example for patients with endometrial carcinoma, the degree of lymphovascular space invasion (LVSI) can be either absent, focally present, or substantially present. For a subset of individuals, however, LVSI is reported as being present, which includes both non-absent options. In the analysis of such a dataset, difficulties arise when coarsened observations are to be used in an imputation procedure. To our knowledge, no clear-cut method has been described in the literature on how to handle an observed subset of values, and treating them as entirely missing could lead to biased estimates. Therefore, in this paper, we evaluated the best strategy to deal with coarsened and missing data in multiple imputation. We tested a number of plausible ad hoc approaches, possibly already in use by statisticians. Additionally, we propose a principled approach to this problem, consisting of an adaptation of the SMC-FCS algorithm (SMC-FCS  CoCo $$ {}_{\mathrm{CoCo}} $$ : Coarsening compatible), that ensures that imputed values adhere to the coarsening information. These methods were compared in a simulation study. This comparison shows that methods that prevent imputations of incompatible values, like the SMC-FCS  CoCo $$ {}_{\mathrm{CoCo}} $$ method, perform consistently better in terms of a lower bias and RMSE, and achieve better coverage than methods that ignore coarsening or handle it in a more naïve way. The analysis of the motivating example shows that the way the coarsening information is handled can matter substantially, leading to different conclusions across methods. Overall, our proposed SMC-FCS  CoCo $$ {}_{\mathrm{CoCo}} $$ method outperforms other methods in handling coarsened data, requires limited additional computation cost and is easily extendable to other scenarios.

Exploring the effect of COVID-19 restrictions on the social functioning scale in a clinical trial of antipsychotic reduction: using multiple imputation to target a hypothetical estimand

OBJECTIVES

Many trials are affected by unforeseen events after recruitment has commenced. The aim of this study is to explore a hypothetical strategy for dealing with an intercurrent event that occurred during trial follow-up; COVID-19 restrictions.

STUDY DESIGN AND SETTING

Secondary analysis of a randomized controlled trial (RCT) in schizophrenia, comparing antipsychotic reduction vs maintenance medication on the social functioning scale (SFS) score at 12 months' follow-up. A hypothetical analysis strategy was used to estimate the treatment effect in a COVID-19 restriction-free world. Outcome data were set to missing, and multiple imputation was used to replace values affected by COVID-19.

RESULTS

The trial randomized 253 participants, 187 participants had an SFS score at 12 months, and 75 of those were collected during COVID-19 restrictions. In the original complete case regression analysis, targeting a treatment policy estimand, the treatment effect was estimated to be 0.51 (95% CI -1.33, 2.35) points higher in the reduction group. After multiple imputation, targeting the hypothetical estimand, the mean SFS score was -3.01 (95% CI -7.22, 1.20) points lower in the reduction group, but varied with different assumptions about the timing of events and in sensitivity analyses to increase the size of difference between randomized groups.

CONCLUSION

We demonstrated how the intervention effect can change when estimating the intervention effect in a pandemic world (treatment policy estimand) vs a pandemic restriction-free world (hypothetical estimand), and that estimates are sensitive to imputation and input assumptions. Trialists should be aware of potential intercurrent events and plan the analysis to take them into account.

PLAIN LANGUAGE SUMMARY

Many medical research studies that enable us to find out how well things work had to change due to COVID-19 restrictions. This may have altered the results. We used data from a randomized controlled trial (RCT) to examine whether there was evidence for this. The main outcome included questions on how often participants went to the cinema, swimming, to church or saw friends or relatives. Many of these activities were not possible during COVID-19 restrictions and became possible again over time. We used statistical methods to replace data that were collected during COVID-19 restrictions with the best possible estimate if COVID-19 had not happened. We found that the trial results were likely to have been different if the effect of COVID-19 restrictions were taken away. It is likely that most studies will be interested in results that do not include data collected during COVID-19.

Optimization of school physical education schedules to enhance long-term public health outcomes

Introduction

Optimizing school physical education (PE) schedules is crucial for enhancing public health outcomes, particularly among school-aged children.

Methods

Therefore, in this study, a weighted fitness function is developed to evaluate health fitness scores. This function integrates multiple health metrics such as BMI reduction, fitness improvement, calories burned, and heart rate reduction. Six optimization algorithms such as Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), Simulated Annealing (SA), Differential Evolution (DE), and Artificial Bee Colony (ABC) optimization algorithms are utilized to optimize PE schedules based on the designed weighted fitness function. Using a dataset of 1,360 student entries, the study incorporates health metrics such as BMI reduction, fitness score improvement, caloric expenditure, and heart rate reduction into a weighted fitness function for optimization.

Results

The results show that ACO achieved the highest allocation of PE time (9.91 h/week), the most significant caloric expenditure (370 kcal/session), and the greatest reduction in heart rate (8.5 bpm). GA excelled in the reduction of BMI, achieving a decrease of 10.63 units.

Discussion

These analyses reveal the transformative potential of optimized PE schedules in reducing the burden of lifestyle-related diseases, promoting equitable health outcomes, and supporting cognitive and mental well-being. Finally, recommendations are provided for policymakers and stakeholders to implement data-driven PE programs that maximize long-term public health benefits.

Analysis and prediction of atmospheric ozone concentrations using machine learning

Atmospheric ozone chemistry involves various substances and reactions, which makes it a complex system. We analyzed data recorded by Switzerland's National Air Pollution Monitoring Network (NABEL) to showcase the capabilities of machine learning (ML) for the prediction of ozone concentrations (daily averages) and to document a general approach that can be followed by anyone facing similar problems. We evaluated various artificial neural networks and compared them to linear as well as non-linear models deduced with ML. The main analyses and the training of the models were performed on atmospheric air data recorded from 2016 to 2023 at the NABEL station Lugano-Università in Lugano, TI, Switzerland. As a first step, we used techniques like best subset selection to determine the measurement parameters that might be relevant for the prediction of ozone concentrations; in general, the parameters identified by these methods agree with atmospheric ozone chemistry. Based on these results, we constructed various models and used them to predict ozone concentrations in Lugano for the period between January 1, 2024, and March 31, 2024; then, we compared the output of our models to the actual measurements and repeated this procedure for two NABEL stations situated in northern Switzerland (Dübendorf-Empa and Zürich-Kaserne). For these stations, predictions were made for the aforementioned period and the period between January 1, 2023, and December 31, 2023. In most of the cases, the lowest mean absolute errors (MAE) were provided by a non-linear model with 12 components (different powers and linear combinations of NO2, NOX, SO2, non-methane volatile organic compounds, temperature and radiation); the MAE of predicted ozone concentrations in Lugano was as low as 9 μgm-3. For the stations in Zürich and Dübendorf, the lowest MAEs were around 11 μgm-3 and 13 μgm-3, respectively. For the tested periods, the accuracy of the best models was approximately 1 μgm-3. Since the aforementioned values are all lower than the standard deviations of the observations we conclude that using ML for complex data analyses can be very helpful and that artificial neural networks do not necessarily outperform simpler models.

Comparison of imputation methods for univariate categorical longitudinal data

The life course paradigm emphasizes the need to study not only the situation at a given point in time, but also its evolution over the life course in the medium and long term. These trajectories are often represented by categorical data. This article aims to provide a comprehensive review of the multiple imputation methods proposed so far in the context of univariate categorical data and to assess their practical relevance through a simulation study based on real data. The primary goal is to provide clear methodological guidelines and improve the handling of missing data in life course research. In parallel, we develop the MICT-timing algorithm, which is an extension of the MICT algorithm. This innovative multiple imputation method improves the quality of imputation in trajectories subject to time-varying transition rates, a situation often encountered in life course data.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11135-024-02028-z.

Identifying trajectories of change in sleep disturbance during psychological treatment for depression

BACKGROUND

Sleep disturbance may impact response to psychological treatment for depression. Understanding how sleep disturbance changes during the course of psychological treatment, and identifying the risk factors for sleep disturbance response may inform clinical decision-making.

METHOD

This analysis included 18,915 patients receiving high-intensity psychological therapy for depression from one of eight London-based Improving Access to Psychological Therapies (IAPT) services between 2011 and 2020. Distinct trajectories of change in sleep disturbance were identified using growth mixture modelling. The study also investigated associations between identified trajectory classes, pre-treatment patient characteristics, and eventual treatment outcomes from combined PHQ-9 and GAD-7 metrics used by the services.

RESULTS

Six distinct trajectories of sleep disturbance were identified: two demonstrated improvement, while one showed initial deterioration and the other three groups displayed only limited change in sleep disturbance, each with varying baseline sleep disturbance. Associations with trajectory class membership were found based on: gender, ethnicity, employment status, psychotropic medication use, long-term health condition status, severity of depressive symptoms, and functional impairment. Groups that showed improvement in sleep had the best eventual outcomes from depression treatment, followed by groups that consistently slept well.

LIMITATION

Single item on sleep disturbance used, no data on treatment adherence.

CONCLUSIONS

These findings reveal heterogeneity in the course of sleep disturbance during psychological treatment for depression. Closer monitoring of changes in sleep disturbance during treatment might inform treatment planning. This includes decisions about when to incorporate sleep management interventions, and whether to change or augment therapy with interventions to reduce sleep disturbance.

Imputation methods for mixed datasets in bioarchaeology

Missing data is a prevalent problem in bioarchaeological research and imputation could provide a promising solution. This work simulated missingness on a control dataset (481 samples × 41 variables) in order to explore imputation methods for mixed data (qualitative and quantitative data). The tested methods included Random Forest (RF), PCA/MCA, factorial analysis for mixed data (FAMD), hotdeck, predictive mean matching (PMM), random samples from observed values (RSOV), and a multi-method (MM) approach for the three missingness mechanisms (MCAR, MAR, and MNAR) at levels of 5%, 10%, 20%, 30%, and 40% missingness. This study also compared single imputation with an adapted multiple imputation method derived from the R package "mice". The results showed that the adapted multiple imputation technique always outperformed single imputation for the same method. The best performing methods were most often RF and MM, and other commonly successful methods were PCA/MCA and PMM multiple imputation. Across all criteria, the amount of missingness was the most important parameter for imputation accuracy. While this study found that some imputation methods performed better than others for the control dataset, each imputation method has advantages and disadvantages. Imputation remains a promising solution for datasets containing missingness; however when making a decision it is essential to consider dataset structure and research goals.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12520-024-02078-2.

Racial and ethnic differences in comorbid psychosis: a population-based study

Introduction

Differences in the prevalence of psychiatric conditions such as psychosis as well as patterns of comorbidity for psychosis have been reported between racial and ethnic groups. It is unclear whether those differences are consistent for comorbid psychosis.

Methods

Self-reported diagnostic data from American adults ages 18-99 participating in the Collaborative Psychiatric Epidemiology Surveys (CPES) (N ~ 11,844) were used to test the association between four racial and ethnic group categories (White, Asian, Hispanic, Black) and comorbid psychosis. Comorbid psychosis was measured as a 4-level categorical variable (No mental illness nor psychosis, Mental Illness, Psychosis only, comorbid psychosis (i.e., Psychosis + Mental Illness). Chi-square tests were used to determine significant differences in the prevalence of comorbid psychosis by race and ethnicity. A multinomial logistic regression was used to test the association between racial and ethnic classifications and comorbid psychosis after adjusting for common demographic characteristics (i.e., education, sex, income, and age).

Results

Relative to White participants, Hispanic and Asian participants were less likely to be affected with comorbid psychosis. (Adjusted Odds Ratio, AORAsian = 0.32, CI = 0.22 - 0.47, p <0.0001, AORHispanic = 0.66, CI = 0.48 - 0.92, p = 0.012). Relative to White participants there was not significant association for comorbid psychosis in Black participants (AORBlack = 0.91, CI = 0.70 - 1.20, p = 0.52) In contrast Hispanic and Black participants were more likely to report psychosis alone (AORHispanic = 1.94, CI = 1.27-2.98, p = 0.002, AORBlack = 1.86, 1.24-2.82, p = 0.003) compared to White participants.

Conclusion

There were different patterns of associations by race and ethnicity for psychosis and comorbid psychosis. The lower prevalence of comorbid psychosis in non-White groups may be due to underdiagnosis or underreporting of other mental disorders.

Older men and loneliness: a cross-sectional study of sex differences in the English Longitudinal Study of Ageing

BACKGROUND

Research into men and masculinities suggests men may be more reluctant than women to state they are lonely, more reliant on partners/spouses and/or alcohol to tackle it, and that this may be a result of poorer social relationships. Ageing is often associated with loneliness, and research has indicated gendered results in older people, but existing evidence lacks generalisability and cultural context. This study tests hypotheses on sex differences in loneliness in older England-based men and women.

METHODS

We conducted a cross-sectional study using a sample of 6936 respondents aged 50 + from the English Longitudinal Study of Ageing (wave 8). Multiple imputation with chained equations was conducted to handle missing data. Multivariate regression was used to investigate the impact of sex on a direct question on loneliness whilst controlling for the University of California loneliness (UCLA) scale. Multivariate regression with interaction terms were used to examine sex differences in loneliness and alcohol consumption, partner status, and social relationships.

RESULTS

Older men were less likely than older women to state they are lonely even when controlling for UCLA score. Older men showed a greater association between loneliness and alcohol consumption, but only when measuring the number of units consumed in the last week, and not using a less precise measure of the past year. Older men who cohabited with a partner were less lonely than cohabiting older women, whereas previously married but not cohabiting older men were lonelier than their female counterparts. However, never married older men were less lonely than never married older women. Evidence was found to suggests older men's worse friendships mediated this association, but social isolation and number of close relationships did not. Severe isolation predicted greater loneliness in older women, but not older men.

CONCLUSIONS

Cultural ideals of masculinity and older men's poorer quality friendships may explain their reluctance to directly state loneliness, greater dependency on partners/spouses, and use of alcohol. Severely isolated older men may under-report loneliness on the UCLA scale as well as a direct question.

The Association of Emotional Support, HIV Stigma, and Home Environment With Disclosure Efficacy and Perceived Disclosure Outcomes in Young People Living With HIV in Zambia: A Cross-Sectional Study

ABSTRACT

This study examined the association of various forms of social support, attitudes toward living at home, and HIV stigma experiences with HIV self-disclosure efficacy and perceived negative disclosure outcomes. We analyzed cross-sectional data from 120 young people with HIV (YPWH) aged 18-21 years receiving outpatient care in Eastern Province, Zambia. Perceived negative disclosure outcomes and disclosure self-efficacy were measured using an adapted version of the Adolescent HIV Disclosure Cognitions and Affect Scale. Explanatory variables included parental or caregiver support, emotional support, instrumental support, HIV stigma experiences, and attitudes toward living at home. Findings suggest that YPWH's confidence in their ability to self-disclose their HIV status and their assessment of negative outcomes associated with HIV disclosure are influenced by emotional support, experiences of HIV stigma, and the quality of the home environment.

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.

Food insecurity and depressive symptoms among young people living with HIV in Eastern Zambia

Background: Mental health problems are common among people living with HIV/AIDS and contribute to poor HIV-related outcomes, including AIDS-related mortality. We examined the association between severe food insecurity and depressive symptoms in young people living with HIV (YPLH) in Zambia. Methods: We sampled 120 youth living with HIV aged 18-21 years in the Eastern Province of Zambia. Household food insecurity was measured using the Household Food Insecurity Access Scale (HFIAS). Mental health was assessed using the Children's Depression Inventory-Short Form. We fitted linear regression models to assess whether food insecurity is associated with depressive symptoms. Results: The mean age was 19 years, and 63% were female. Overall, 43% were severely food insecure. After adjusting for sociodemographic variables and other confounders, severely food insecure participants were more likely to report depressive symptoms (β = 0.81, 95% Confidence Interval [CI] 0.07-1.55) and engagement in sex work (β = 1.78, 95% CI 0.32-3.25). Conclusion: Almost half of the sample reported severe food insecurity which was associated with depressive symptoms. Interventions aimed at improving food insecurity may have beneficial effects on mental health and HIV outcomes among young people living with HIV in resource-limited settings.

A longitudinal transition imputation model for categorical data applied to a large registry dataset

Imputation of longitudinal categorical covariates with several waves and many predictors is cumbersome in terms of implausible transitions, colinearity, and overfitting. We designed a simulation study with data obtained from a general practitioners' morbidity registry in Belgium for three waves, with smoking as the longitudinal covariate of interest. We set varying proportions of data on smoking to missing completely at random and missing not at random with proportions of missingness equal to 10%, 30%, 50%, and 70%. This study proposed a 3-stage approach that allows flexibility when imputing time-dependent categorical covariates. First, multiple imputation using fully conditional specification or multiple imputation for the predictor variables was deployed using the wide format such that previous and future information of the same patient was utilized. Second, a joint Markov transition model for initial, forward, backward, and intermittent probabilities was developed for each imputed dataset. Finally, this transition model was used for imputation. We compared the performance of this methodology with an analyses of the complete data and with listwise deletion in terms of bias and root mean square error. Next, we applied this methodology in a clinical case for years 2017 to 2021, where we estimated the effect of several covariates on the pneumococcal vaccination. This methodological framework ensures that the plausibility of transitions is preserved, overfitting and colinearity issues are resolved, and confounders can be utilized. Finally, a companion R package was developed to enable the replication and easy application of this methodology.

Associations between religious/spiritual beliefs and behaviours and dietary patterns: analysis of the parental generation in a prospective cohort study (ALSPAC) in Southwest England

OBJECTIVE

Religious/spiritual beliefs and behaviours (RSBB) have been associated with health outcomes, with diet a potential mediator of this relationship. We therefore explored whether RSBB were associated with differences in diet.

DESIGN

Dietary patterns and nutrient intakes were derived from food frequency questionnaire completed by pregnant women in 1991-1992 (mean age = 28·3 years, range = 15-46) and by the mothers and partners 4 years post-partum (mothers mean age = 32·3, range = 19-49; partners mean age = 34·5, range = 18-74). RSBB exposures measured in pregnancy included religious belief, affiliation and attendance. We first explored whether RSBBs were associated with dietary patterns in confounder-adjusted linear regression models. If associations were found, we examined whether RSBB were associated with nutrient intake (linear regression) and following nutrient intake guidelines (logistic regression).

SETTING

Prospective birth cohort study in Southwest England (Avon Longitudinal Study of Parents and Children; ALSPAC).

PARTICIPANTS

13 689 enrolled mothers and their associated partners.

RESULTS

In pregnant women, RSBB were associated with higher 'traditional' (i.e. 'meat and two veg') and lower 'vegetarian' dietary pattern scores. Religious attendance and non-Christian religious affiliation were associated with higher 'health-conscious' dietary pattern scores. Religious attendance was associated with increased micronutrient intake and following recommended micronutrient intake guidelines, with weaker effects for religious belief and affiliation. Comparable patterns were observed for mothers and partners 4 years post-partum, although associations between RSBB and nutrient intakes were weaker for partners.

CONCLUSIONS

RSBBs are associated with broad dietary patterns and nutrient intake in this cohort. If these reflect causal relationships, diet may potentially mediate the pathway between RSBB and health.

Logistic regression vs. predictive mean matching for imputing binary covariates

Multivariate imputation using chained equations (MICE) is a popular algorithm for imputing missing data that entails specifying multivariate models through conditional distributions. For imputing missing continuous variables, two common imputation methods are the use of parametric imputation using a linear model and predictive mean matching. When imputing missing binary variables, the default approach is parametric imputation using a logistic regression model. In the R implementation of MICE, the use of predictive mean matching can be substantially faster than using logistic regression as the imputation model for missing binary variables. However, there is a paucity of research into the statistical performance of predictive mean matching for imputing missing binary variables. Our objective was to compare the statistical performance of predictive mean matching with that of logistic regression for imputing missing binary variables. Monte Carlo simulations were used to compare the statistical performance of predictive mean matching with that of logistic regression for imputing missing binary outcomes when the analysis model of scientific interest was a multivariable logistic regression model. We varied the size of the analysis samples (N = 250, 500, 1,000, 5,000, and 10,000) and the prevalence of missing data (5%-50% in increments of 5%). In general, the statistical performance of predictive mean matching was virtually identical to that of logistic regression for imputing missing binary variables when the analysis model was a logistic regression model. This was true across a wide range of scenarios defined by sample size and the prevalence of missing data. In conclusion, predictive mean matching can be used to impute missing binary variables. The use of predictive mean matching to impute missing binary variables can result in a substantial reduction in computer processing time when conducting simulations of multiple imputation.

Multiple imputation for non-monotone missing not at random data using the no self-censoring model

Although approaches for handling missing data from longitudinal studies are well-developed when the patterns of missingness are monotone, fewer methods are available for non-monotone missingness. Moreover, the conventional missing at random assumption-a natural benchmark for monotone missingness-does not model realistic beliefs about the non-monotone missingness processes (Robins and Gill, 1997). This has provided the impetus for alternative non-monotone missing not at random mechanisms. The "no self-censoring" model is such a mechanism and assumes the probability an outcome variable is missing is independent of its value when conditioning on all other possibly missing outcome variables and their missingness indicators. As an alternative to "weighting" methods that become computationally demanding with increasing number of outcome variables, we propose a multiple imputation approach under no self-censoring. We focus on the case of binary outcomes and present results of simulation and asymptotic studies to investigate the performance of the proposed imputation approach. We describe a related approach to sensitivity analysis to departure from no self-censoring. We discuss the relationship between missing at random and no self-censoring and prove that one is not a special case of the other. Finally, we discuss extensions to non-binary data settings. The proposed methods are illustrated with application to a substance use disorder clinical trial.

Enhanced Street Crossing Features are Associated with Higher Post-Stroke Physical Quality of Life

BACKGROUND

Features of the physical environment may affect post-stroke recovery, but empirical evidence is limited. This study examines associations between features of the physical environment and post-stroke physical quality of life (PH-QOL).

METHODS

The study sample included stroke survivors enrolled in the Caring for Adults Recovering from the Effects of Stroke project, a prospective cohort. Features of the physical environment surrounding participants' home addresses were audited using Google Earth. Audits captured information about crossings (e.g. curb-cuts; range 0-4), street segments (e.g. sidewalks; range 0-17.5), and a route (e.g. parks; range 0-27) near participants' home. Summary scores were categorized into tertials representing "few," "some," and "many" pedestrian-friendly features. Post-stroke PH-QOL was measured by the SF-12 (range 0-100) around 6 to 12-, 18-, 27-, and 36-months post-stroke. Linear mixed models were used to estimate PH-QOL over time. Chained multiple imputation was used to account for missing data.

RESULTS

Two hundred and seventy-five participants were eligible, among whom 210 had complete data. Most participants lived in areas with "few" features to promote outdoor mobility. Participants living in environments with "some" crossing features had a 4.90 (95% CI: 2.32, 7.48) higher PH-QOL score across the observation period in comparison to participants living in environments with "few" crossing features. Features of the physical environment along street segments and routes were not associated with post-stroke PH-QOL.

CONCLUSION

Crossing features are associated with post-stroke PH-QOL. Modifying features of the physical environment at nearby crossings, such as curb-cuts, may be a promising strategy for increasing PH-QOL.

Modifiable lifestyle risk factors and survival after diagnosis with multiple myeloma

BACKGROUND

While remaining incurable, median overall survival for MM now exceeds 5 years. Yet few studies have investigated how modifiable lifestyle factors influence survival. We investigate whether adiposity, diet, alcohol, or smoking are associated with MM-related fatality.

RESEARCH DESIGN AND METHODS

We recruited 760 incident cases of MM via cancer registries in two Australian states during 2010-2016. Participants returned questionnaires on health and lifestyle. Follow-up ended in 2020. Flexible parametric survival models were used to estimate hazard ratios (HR) and 95% confidence intervals (CI) for lifestyle exposures and risk of all-cause and MM-specific fatality.

RESULTS

Higher pre-diagnosis Alternative Healthy Eating Index (AHEI) scores were associated with reduced MM-specific fatality (per 10-unit score, HR = 0.84, 95%CI = 0.70-0.99). Pre-diagnosis alcohol consumption was inversely associated with MM-specific fatality, compared with nondrinkers (0.1-20 g per day, HR = 0.59, 95%CI = 0.39-0.90; >20 g per day, HR = 0.67, 95%CI = 0.40-1.13). Tobacco smoking was associated with increased all-cause fatality compared with never smoking (former smokers: HR = 1.44, 95%CI = 1.10-1.88; current smokers: HR = 1.30, 95%CI = 0.80-2.10). There was no association between pre-enrollment body mass index (BMI) and MM-specific or all-cause fatality.

CONCLUSIONS

Our findings support established recommendations for healthy diets and against smoking. Higher quality diet, as measured by the AHEI, may improve survival post diagnosis with MM.

Impact of lifestyle and socioeconomic position on use of non-steroidal anti-inflammatory drugs: A population-based cohort study

PURPOSE

Lifestyle and socioeconomic position may confound the link between non-steroidal anti-inflammatory drugs (NSAIDs) and cardiovascular events, if associated with NSAID use. We examined this association.

METHODS

We conducted a cohort study of all adult first-time responders to the Danish National Health Surveys of 2010, 2013, or 2017 without an NSAID prescription within 3 months before survey completion (n = 407 395). Study exposures were weight, smoking status, alcohol consumption, binge drinking frequency, physical activity level, marital status, highest achieved level of education, income, and employment status. We used a Cox model to compute hazard ratios of time to first redemption of an NSAID prescription and a cumulative odds model to compute odds ratios (ORs) of redeeming one additional NSAID prescription in the year after survey completion.

RESULTS

Total follow-up time was 1 931 902 years. The odds of redeeming one additional NSAID prescription in the year after survey completion varied within all categories of lifestyle and socioeconomic position. The largest ORs were observed within categories of weight (1.70, 95% CI: 1.65-1.74 for obesity vs. normal weight), smoking status (1.24, 95% CI: 1.21-1.27 for current vs. never use), and education (1.44, 95% CI: 1.39-1.49 for primary or other vs. university or higher education). The Cox model showed consistent results.

CONCLUSIONS

Markers of unhealthy lifestyle and low socioeconomic position were associated with initiation and prolonged NSAID use. Consideration of lifestyle and socioeconomic markers as potential confounders in NSAID studies is therefore recommended.

Logistic Regression With Machine Learning Sheds Light on the Problematic Sexual Behavior Phenotype

OBJECTIVES

There has been a longstanding debate about whether the mechanisms involved in problematic sexual behavior (PSB) are similar to those observed in addictive disorders, or related to impulse control or to compulsivity. The aim of this report was to contribute to this debate by investigating the association between PSB, addictive disorders (internet addiction, compulsive buying), measures associated with the construct known as reward deficiency (RDS), and obsessive-compulsive disorder (OCD).

METHODS

A Canadian university Office of the Registrar invited 68,846 eligible students and postdoctoral fellows. Of 4710 expressing interest in participating, 3359 completed online questionnaires, and 1801 completed the Mini-International Neuropsychiatric Interview. PSB was measured by combining those screening positive (score at least 6) on the Sexual Addiction Screening Test-Revised Core with those self-reporting PSB. Current mental health condition(s) and childhood trauma were measured by self-report. OCD was assessed by a combination of self-report and Mini-International Neuropsychiatric Interview data.

RESULTS

Of 3341 participants, 407 (12.18%) screened positive on the Sexual Addiction Screening Test-Revised Core. On logistic regression, OCD, attention deficit, internet addiction, a family history of PSB, childhood trauma, compulsive buying, and male gender were associated with PSB. On multiple correspondence analysis, OCD appeared to cluster separately from the other measures, and the pattern of data differed by gender.

CONCLUSIONS

In our sample, factors that have previously been associated with RDS and OCD are both associated with increased odds of PSB. The factors associated with RDS appear to contribute to a separate data cluster from OCD and to lie closer to PSB.

Association of intraoperative dexmedetomidine use with postoperative hypotension in unilateral hip and knee arthroplasties: a historical cohort study

PURPOSE

Dexmedetomidine is frequently used as a sedative agent for orthopedic surgery patients undergoing total hip or knee arthroplasty. Although the benefits of dexmedetomidine are well described in the literature, there is also potential for harm, especially regarding the hemodynamic effects of dexmedetomidine in the postoperative setting.

METHODS

This historical cohort study included all primary unilateral total hip or knee arthroplasties conducted from April 2017 to February 2020 in a single, university-affiliated, tertiary care centre (Jewish General Hospital, Montreal, QC, Canada). We used multivariable logistic regression to analyze the predictors for postoperative hypotension, defined as a systolic blood pressure < 90 mm Hg or any systolic blood pressure while on a vasopressor infusion in the postanesthesia care unit. Models were validated using calibration and discrimination with bootstrapping technique.

RESULTS

One thousand five hundred and eighty-eight patients were included in this study. Postoperative hypotension occurred in 413 (26%) patients. Statistically significant predictors for postoperative hypotension included female sex (adjusted odds ratio [aOR], 3.24; 95% confidence interval [CI], 2.29 to 4.58), a history of transient ischemic attack or cerebrovascular accident (aOR, 1.97; 95% CI, 1.04 to 3.72), and intraoperative dexmedetomidine use (aOR, 2.61; 95% CI, 1.99 to 3.42). Moreover, the risk of postoperative hypotension was approximately two times higher than baseline, with a total intraoperative dexmedetomidine dose above 50 μg (relative risk, 1.99; 95% CI, 1.63 to 2.44; P < 0.001). A higher preoperative systolic blood pressure (aOR, 0.98; 95% CI, 0.97 to 0.99) was a protective factor for postoperative hypotension.

CONCLUSION

In this historical cohort study, dexmedetomidine was a strong risk factor for postoperative hypotension in total hip or knee arthroplasty patients. Dexmedetomidine, and particularly at high cumulative doses above 50 μg, should be administered judiciously in high-risk surgical patients to minimize the risk of postoperative hypotension.

The Association of Perceived, Internalized, and Enacted HIV Stigma With Medication Adherence, Barriers to Adherence, and Mental Health Among Young People Living With HIV in Zambia

Few studies have examined the independent effects of different manifestations of HIV stigma experiences on health outcomes among youth living with HIV in low- and middle-income countries. We examined the association of internalized, enacted, and perceived HIV stigmas with medication adherence, self-esteem, depression, and barriers to adherence. Young people living with HIV aged 18-21 years (N = 120) were purposively sampled from two health facilities in Eastern Province, Zambia, and completed self-report measures. Results indicated heterogeneous associations. Internalized HIV stigma was positively associated with depression and negatively associated with adherence, adherence motivation, behavioral adherence skills, and self-esteem. Perceived stigma was negatively associated with self-esteem. No significant association was observed between enacted stigma and health outcomes. The complexity of HIV stigma requires a precise explication of the associations among different HIV stigma experiences and outcomes, which can inform the development of stigma reduction interventions targeting one or more stigma experiences.

Disparities in Lung-Protective Ventilation in the United States

Background The objective of our study was to determine whether disparities exist in the use of lung-protective ventilation for critically ill mechanically ventilated patients in the United States based on gender, race/ethnicity, or insurance status. Methods This was a secondary data analysis of a prospective multicenter cohort study conducted from 2010 to 2012. The outcome of interest was the proportion of patients receiving tidal volume > 8 mL/kg predicted body weight (PBW). Results There were 1,595 patients in our primary analysis (710 women, 885 men). Women were more likely to receive tidal volumes > 8 mL/kg PBW than men (odds ratio [OR] = 3.42, 95% confidence interval [CI] = 2.67-4.40), a finding largely but not completely explained by gender differences in height. The underinsured were significantly more likely to receive tidal volume > 8 mL/kg PBW than the insured in multivariable analysis (OR = 1.54, 95% CI = 1.16-2.04). The prescription of > 8 mL/kg PBW tidal volume did not differ by racial or ethnic categories. Conclusions In this prospective nationwide cohort of critically ill mechanically ventilated patients, women and the underinsured were less likely than their comparators to receive lung-protective ventilation, with no apparent differences based on race/ethnicity alone.

Identifying Predictors of Inpatient Verbal Aggression in a Forensic Psychiatric Setting Using a Tree-based Modeling Approach

Inpatient violence poses a great risk to the health and well-being of other patients and members of staff. Previous research has shown that prevalence rates of violent behavior are particularly high in forensic psychiatric settings. Thus, the reliable identification of forensic inpatients who are particularly at risk for violent behavior is an important aspect of risk management. In the present study, we analyzed clinicians' assessments of N = 504 male and female inpatients of German forensic mental health institutions in order to identify risk factors for verbal institutional violence. Using a tree-based modeling approach, we found the following variables to be predictors of verbal aggression: gender, insight into the illness, number of prior admissions to psychiatric hospitals, and insight into the iniquity of the offence. A high number of prior admissions to psychiatric hospitals seems to be a risk factor for verbal aggression amongst men whereas it showed the opposite effect amongst women. Our results highlight the importance of dynamic risk factors, such as poor insight into the own illness, in the prediction of violent incidents. With regard to future research, we argue for a stronger emphasis on nonparametric models as well as on potential interaction effects of risk and protective factors.

AZTEC-azithromycin therapy for prevention of chronic lung disease of prematurity: a statistical analysis plan for clinical outcomes

BACKGROUND

The AZTEC trial is a multi-centre, randomised, placebo-controlled trial of azithromycin to improve survival without development of chronic lung disease of prematurity (CLD) in preterm infants. The statistical analysis plan for the clinical outcomes of the AZTEC trial is described.

METHODS AND DESIGN

A double-blind, randomised, placebo-controlled trial of a 10-day course of intravenous azithromycin (20 mg/kg for 3 days; 10 mg/kg for 7 days) administered to preterm infants born at < 30 weeks' gestational age across UK tertiary neonatal units. Following parental consent, infants are randomly allocated to azithromycin or placebo, with allocated treatment starting within 72 h of birth. The primary outcome is survival without moderate/severe CLD at 36 weeks' postmenstrual age (PMA). Serial respiratory fluid and stool samples are being collected up to 21 days of life. The target sample size is 796 infants, which is based on detecting a 12% absolute difference in survival without moderate/severe CLD at 36 weeks' PMA (90% power, two-sided alpha of 0.05) and includes 10% loss to follow-up.

RESULTS

Baseline demographic and clinical characteristics will be summarised by treatment arm and in total. Categorical data will be summarised by numbers and percentages. Continuous data will be summarised by mean, standard deviation, if data are normal, or median, interquartile range, if data are skewed. Tests of statistical significance will not be undertaken for baseline characteristics. The primary analysis, on the intention to treat (ITT) population, will be analysed using multilevel logistic regression, within a multiple imputation framework. Adjusted odds ratios, 95% confidence intervals, and p-values will be presented. For all other analyses, the analysis population will be based on the complete case population, which is a modified ITT population. All analyses will be adjusted for gestational age and treatment arm and account for any clustering by centre and/or multiple births as a random effect.

CONCLUSION

We describe the statistical analysis plan for the AZTEC trial, including the analysis principles, definitions of the key clinical outcomes, methods for primary analysis, pre-specified subgroup analysis, sensitivity analysis, and secondary analysis. The plan has been finalised prior to the completion of recruitment.

TRIAL REGISTRATION

ISRCTN registry ISRCTN11650227. Registered on 31 July 2018.

MISL: Multiple imputation by super learning

Multiple imputation techniques are commonly used when data are missing, however, there are many options one can consider. Multivariate imputation by chained equations is a popular method for generating imputations but relies on specifying models when imputing missing values. In this work, we introduce multiple imputation by super learning, an update to the multivariate imputation by chained equations method to generate imputations with ensemble learning. Ensemble methodologies have recently gained attention for use in inference and prediction as they optimally combine a variety of user-specified parametric and non-parametric models and perform well when estimating complex functions, including those with interaction terms. Through two simulations we compare inferences made using the multiple imputation by super learning approach to those made with other commonly used multiple imputation methods and demonstrate multiple imputation by super learning as a superior option when considering characteristics such as bias, confidence interval coverage rate, and confidence interval width.

Neonatal sepsis and mortality in low-income and middle-income countries from a facility-based birth cohort: an international multisite prospective observational study

BACKGROUND

Neonatal sepsis is a primary cause of neonatal mortality and is an urgent global health concern, especially within low-income and middle-income countries (LMICs), where 99% of global neonatal mortality occurs. The aims of this study were to determine the incidence and associations with neonatal sepsis and all-cause mortality in facility-born neonates in LMICs.

METHODS

The Burden of Antibiotic Resistance in Neonates from Developing Societies (BARNARDS) study recruited mothers and their neonates into a prospective observational cohort study across 12 clinical sites from Bangladesh, Ethiopia, India, Pakistan, Nigeria, Rwanda, and South Africa. Data for sepsis-associated factors in the four domains of health care, maternal, birth and neonatal, and living environment were collected for all mothers and neonates enrolled. Primary outcomes were clinically suspected sepsis, laboratory-confirmed sepsis, and all-cause mortality in neonates during the first 60 days of life. Incidence proportion of livebirths for clinically suspected sepsis and laboratory-confirmed sepsis and incidence rate per 1000 neonate-days for all-cause mortality were calculated. Modified Poisson regression was used to investigate factors associated with neonatal sepsis and parametric survival models for factors associated with all-cause mortality.

FINDINGS

Between Nov 12, 2015 and Feb 1, 2018, 29 483 mothers and 30 557 neonates were enrolled. The incidence of clinically suspected sepsis was 166·0 (95% CI 97·69-234·24) per 1000 livebirths, laboratory-confirmed sepsis was 46·9 (19·04-74·79) per 1000 livebirths, and all-cause mortality was 0·83 (0·37-2·00) per 1000 neonate-days. Maternal hypertension, previous maternal hospitalisation within 12 months, average or higher monthly household income, ward size (>11 beds), ward type (neonatal), living in a rural environment, preterm birth, perinatal asphyxia, and multiple births were associated with an increased risk of clinically suspected sepsis, laboratory-confirmed sepsis, and all-cause mortality. The majority (881 [72·5%] of 1215) of laboratory-confirmed sepsis cases occurred within the first 3 days of life.

INTERPRETATION

Findings from this study highlight the substantial proportion of neonates who develop neonatal sepsis, and the high mortality rates among neonates with sepsis in LMICs. More efficient and effective identification of neonatal sepsis is needed to target interventions to reduce its incidence and subsequent mortality in LMICs.

FUNDING

Bill & Melinda Gates Foundation.

Planning a method for covariate adjustment in individually randomised trials: a practical guide

BACKGROUND

It has long been advised to account for baseline covariates in the analysis of confirmatory randomised trials, with the main statistical justifications being that this increases power and, when a randomisation scheme balanced covariates, permits a valid estimate of experimental error. There are various methods available to account for covariates but it is not clear how to choose among them.

METHODS

Taking the perspective of writing a statistical analysis plan, we consider how to choose between the three most promising broad approaches: direct adjustment, standardisation and inverse-probability-of-treatment weighting.

RESULTS

The three approaches are similar in being asymptotically efficient, in losing efficiency with mis-specified covariate functions and in handling designed balance. If a marginal estimand is targeted (for example, a risk difference or survival difference), then direct adjustment should be avoided because it involves fitting non-standard models that are subject to convergence issues. Convergence is most likely with IPTW. Robust standard errors used by IPTW are anti-conservative at small sample sizes. All approaches can use similar methods to handle missing covariate data. With missing outcome data, each method has its own way to estimate a treatment effect in the all-randomised population. We illustrate some issues in a reanalysis of GetTested, a randomised trial designed to assess the effectiveness of an electonic sexually transmitted infection testing and results service.

CONCLUSIONS

No single approach is always best: the choice will depend on the trial context. We encourage trialists to consider all three methods more routinely.

Prospective Associations between Physical Activity and Memory in the Canadian Longitudinal Study on Aging: Examining Social Determinants

Objectives

To examine associations between physical activity (PA) and prospectively assessed memory in a cohort of cognitively healthy adults, after accounting for understudied social determinants.

Methods

We used data from the Canadian Longitudinal Study on Aging (CLSA). PA (exposure) and memory (outcome) were assessed using validated measures in 2013–2015 and 2015–2018, respectively. Respondents reported their daily number of hours spent engaging in five different PAs. We conducted multiple imputation and used linear regression (n = 41,394), adjusting for five categories of covariates: demographics, sensory health characteristics, health behaviors, health status, and social determinants (sex/gender, education, income, social support, perceived social standing, race, and sexual orientation).

Results

In crude models, nearly every intensity and duration of PA was associated with better memory. In fully adjusted models, protective associations were attenuated; however, some associations held: all durations of walking, most durations of light activities, moderate activities for ≥1 hour, and strenuous activities for 1 to <2 hours.

Discussion

Some forms of PA may be associated with better memory. The benefits of higher intensity PA may only be realized after social determinants are addressed.

A note on imputing squares via polynomial combination approach

The polynomial combination (PC) method, proposed by Vink and Van Buuren, is a hot-deck multiple imputation method for imputation models containing squared terms. The method yields unbiased regression estimates and preserves the quadratic relationships in the imputed data for both MCAR and MAR mechanisms. However, Vink and Van Buuren never studied the coverage rate of the PC method. This paper investigates the coverage of the nominal 95% confidence intervals for the polynomial combination method and improves the algorithm to avoid the perfect prediction issue. We also compare the original and the improved PC method to the substantive model compatible fully conditional specification method proposed by Bartlett et al. and elucidate the two imputation methods’ characters.

A comparison of methods for analyzing a binary composite endpoint with partially observed components in randomized controlled trials

Composite endpoints are commonly used to define primary outcomes in randomized controlled trials. A participant may be classified as meeting the endpoint if they experience an event in one or several components (eg, a favorable outcome based on a composite of being alive and attaining negative culture results in trials assessing tuberculosis treatments). Partially observed components that are not missing simultaneously complicate the analysis of the composite endpoint. An intuitive strategy frequently used in practice for handling missing values in the components is to derive the values of the composite endpoint from observed components when possible, and exclude from analysis participants whose composite endpoint cannot be derived. Alternatively, complete record analysis (CRA) (excluding participants with any missing components) or multiple imputation (MI) can be used. We compare a set of methods for analyzing a composite endpoint with partially observed components mathematically and by simulation, and apply these methods in a reanalysis of a published trial (TOPPS). We show that the derived composite endpoint can be missing not at random even when the components are missing completely at random. Consequently, the treatment effect estimated from the derived endpoint is biased while CRA results without the derived endpoint are valid. Missing at random mechanisms require MI of the components. We conclude that, although superficially attractive, deriving the composite endpoint from observed components should generally be avoided. Despite the potential risk of imputation model mis-specification, MI of missing components is the preferred approach in this study setting.

Examining the potential link between child maltreatment and callous-unemotional traits in children and adolescents: A multilevel analysis

BACKGROUND

There is a great deal of research indicating that callous-unemotional traits in childhood are among the strongest predictors of adult psychopathy and psychopathic traits. As a result, there has been a recent surge of studies examining potential risk factors that may be related to the development of callous-unemotional traits.

OBJECTIVE

The current study sought to extend prior research examining potential risk factors for the development of callous-unemotional traits by estimating the extent to which child maltreatment related to callous-unemotional traits in children and adolescents.

PARTICIPANTS

To do so, the study uses a longitudinal sample of 4579 male and female youths drawn from the National Survey of Child and Adolescent Well-Being (NSCAW I) across four waves of data. Data collection ran from November 1999 to December 2006.

METHODS

A series of multilevel random-effects models were estimated in order to examine the association between child maltreatment and callous-unemotional traits.

RESULTS

The results of the analyses revealed a significant association between child maltreatment and callous-unemotional traits across all the models. Additionally, our models demonstrated that the association between child maltreatment and callous-unemotional traits may be dependent upon the biological sex of the individual with child maltreatment having a stronger effect on males than females (β = 0.15*).

CONCLUSIONS

Overall, our analyses lend support to prior research examining child maltreatment as a risk factor for the development of callous-unemotional traits in youth. We conclude by discussing the implications of our study and considerations for future research.

Multiple imputation of semi-continuous exposure variables that are categorized for analysis

Semi-continuous variables are characterized by a point mass at one value and a continuous range of values for remaining observations. An example is alcohol consumption quantity, with a spike of zeros representing non-drinkers and positive values for drinkers. If multiple imputation is used to handle missing values for semi-continuous variables, it is unclear how this should be implemented within the standard approaches of fully conditional specification (FCS) and multivariate normal imputation (MVNI). This question is brought into focus by the use of categorized versions of semi-continuous exposure variables in analyses (eg, no drinking, drinking below binge level, binge drinking, heavy binge drinking), raising the question of how best to achieve congeniality between imputation and analysis models. We performed a simulation study comparing nine approaches for imputing semi-continuous exposures requiring categorization for analysis. Three methods imputed the categories directly: ordinal logistic regression, and imputation of binary indicator variables representing the categories using MVNI (with two variants). Six methods (predictive mean matching, zero-inflated binomial imputation, and two-part imputation methods with variants in FCS and MVNI) imputed the semi-continuous variable, with categories derived after imputation. The ordinal and zero-inflated binomial methods had good performance across most scenarios, while MVNI methods requiring rounding after imputation did not perform well. There were mixed results for predictive mean matching and the two-part methods, depending on whether the estimands were proportions or regression coefficients. The results highlight the need to consider the parameter of interest when selecting an imputation procedure.

Missing data methods for intensive care unit SOFA scores in electronic health records studies: results from a Monte Carlo simulation

Aim: Missing data cause problems through decreasing sample size and the potential for introducing bias. We tested four missing data methods on the Sequential Organ Failure Assessment (SOFA) score, an intensive care research severity adjuster. Methods: Simulation study using 2015-2017 electronic health record data, where the complete dataset was sampled, missing SOFA score elements imposed and performance examined of four missing data methods - complete case analysis, median imputation, zero imputation (recommended by SOFA score creators) and multiple imputation (MI) - on the outcome of in-hospital mortality. Results: MI performed well, whereas other methods introduced varying amounts of bias or decreased sample size. Conclusion: We recommend using MI in analyses where SOFA score component values are missing in administrative data research.

Pretreatment Risk Stratification for Endoscopic Kidney-sparing Surgery in Upper Tract Urothelial Carcinoma: An International Collaborative Study

BACKGROUND

Several groups have proposed features to identify low-risk patients who may benefit from endoscopic kidney-sparing surgery in upper tract urothelial carcinoma (UTUC).

OBJECTIVE

To evaluate standard risk stratification features, develop an optimal model to identify ≥pT2/N+ stage at radical nephroureterectomy (RNU), and compare it with the existing unvalidated models.

DESIGN, SETTING, AND PARTICIPANTS

This was a collaborative retrospective study that included 1214 patients who underwent ureterorenoscopy with biopsy followed by RNU for nonmetastatic UTUC between 2000 and 2017.

OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS

We performed multiple imputation of chained equations for missing data and multivariable logistic regression analysis with a stepwise selection algorithm to create the optimal predictive model. The area under the curve and a decision curve analysis were used to compare the models.

RESULTS AND LIMITATIONS

Overall, 659 (54.3%) and 555 (45.7%) patients had ≤pT1N0/Nx and ≥pT2/N+ disease, respectively. In the multivariable logistic regression analysis of our model, age (odds ratio [OR] 1.02, 95% confidence interval [CI] 1.0-1.03, p = 0.013), high-grade biopsy (OR 1.81, 95% CI 1.37-2.40, p < 0.001), biopsy cT1+ staging (OR 3.23, 95% CI 1.93-5.41, p < 0.001), preoperative hydronephrosis (OR 1.37 95% CI 1.04-1.80, p = 0.024), tumor size (OR 1.09, 95% CI 1.01-1.17, p = 0.029), invasion on imaging (OR 5.10, 95% CI 3.32-7.81, p < 0.001), and sessile architecture (OR 2.31, 95% CI 1.58-3.36, p < 0.001) were significantly associated with ≥pT2/pN+ disease. Compared with the existing models, our model had the highest performance accuracy (75% vs 66-71%) and an additional clinical net reduction (four per 100 patients).

CONCLUSIONS

Our proposed risk-stratification model predicts the risk of harboring ≥pT2/N+ UTUC with reliable accuracy and a clinical net benefit outperforming the current risk-stratification models.

PATIENT SUMMARY

We developed a risk stratification model to better identify patients for endoscopic kidney-sparing surgery in upper tract urothelial carcinoma.

Antifungal prophylaxis for prevention of COVID-19-associated pulmonary aspergillosis in critically ill patients: an observational study

BACKGROUND

Coronavirus disease 19 (COVID-19)-associated pulmonary aspergillosis (CAPA) emerged as important fungal complications in patients with COVID-19-associated severe acute respiratory failure (ARF). Whether mould active antifungal prophylaxis (MAFP) can prevent CAPA remains elusive so far.

METHODS

In this observational study, we included all consecutive patients admitted to intensive care units with COVID-19-associated ARF between September 1, 2020, and May 1, 2021. We compared patients with versus without antifungal prophylaxis with respect to CAPA incidence (primary outcome) and mortality (secondary outcome). Propensity score adjustment was performed to account for any imbalances in baseline characteristics. CAPA cases were classified according to European Confederation of Medical Mycology (ECMM)/International Society of Human and Animal Mycoses (ISHAM) consensus criteria.

RESULTS

We included 132 patients, of whom 75 (57%) received antifungal prophylaxis (98% posaconazole). Ten CAPA cases were diagnosed, after a median of 6 days following ICU admission. Of those, 9 CAPA cases were recorded in the non-prophylaxis group and one in the prophylaxis group, respectively. However, no difference in 30-day ICU mortality could be observed. Thirty-day CAPA incidence estimates were 1.4% (95% CI 0.2-9.7) in the MAFP group and 17.5% (95% CI 9.6-31.4) in the group without MAFP (p = 0.002). The respective subdistributional hazard ratio (sHR) for CAPA incidence comparing the MAFP versus no MAFP group was of 0.08 (95% CI 0.01-0.63; p = 0.017).

CONCLUSION

In ICU patients with COVID-19 ARF, antifungal prophylaxis was associated with significantly reduced CAPA incidence, but this did not translate into improved survival. Randomized controlled trials are warranted to evaluate the efficacy and safety of MAFP with respect to CAPA incidence and clinical outcomes.

Walking and perceived lack of safety: Correlates and association with health outcomes for people living with HIV in rural Zambia

INTRODUCTION

Geographic inaccessibility disproportionately affects health outcomes of rural populations due to lack of suitable transport, prolonged travel time, and poverty. Rural patients are left with few transport options to travel to a health facility. One common option is to travel by foot, which may present additional challenges, such as perceived lack of safety while transiting. We examined the correlates of perceived lack of safety when walking to a health facility and its association with treatment and psychosocial outcomes among adults living with HIV.

METHODS

Data were collected from 101 adults living with HIV in Eastern Province, Zambia. All participants were receiving antiretroviral therapy at one of two health clinics. Perceived lack of safety was measured by asking respondents whether they felt unsafe traveling to and from the health facility in which they were receiving their HIV care. Outcomes included medication adherence, perceived stress, hope for the future, and barriers to pill taking. Linear and logistic regression methods were used to examine the correlates of perceived safety and its association with health outcomes.

RESULTS

Being older, a woman, having a primary education, living farther from a health facility, traveling longer to reach a health facility, and owing money were associated with higher likelihood of feeling unsafe when traveling by foot to health facility. Perceived lack of safety was associated with medication nonadherence, higher level of stress, lower level of agency, and more barriers to pill taking.

CONCLUSIONS

Perceived lack of safety when traveling by foot to a health facility may be a barrier to better treatment and psychosocial outcomes, especially among rural patients. Practitioners and policymakers should consider implementation of differentiated HIV service delivery models to reduce frequent travel to health facilities and to alleviate ART patients' worry about lack of safety when traveling by foot to a health facility.

CRISTAL (a cluster-randomised, crossover, non-inferiority trial of aspirin compared to low molecular weight heparin for venous thromboembolism prophylaxis in hip or knee arthroplasty, a registry nested study): statistical analysis plan

BACKGROUND

This a priori statistical analysis plan describes the analysis for CRISTAL.

METHODS

CRISTAL (cluster-randomised, crossover, non-inferiority trial of aspirin compared to low molecular weight heparin for venous thromboembolism prophylaxis in hip or knee arthroplasty, a registry nested study) aims to determine whether aspirin is non-inferior to low molecular weight heparin (LMWH) in preventing symptomatic venous thromboembolism (VTE) following hip arthroplasty (HA) or knee arthroplasty (KA). The study is nested within the Australian Orthopaedic Association National Joint Replacement Registry. The trial was commenced in April 2019 and after an unplanned interim analysis, recruitment was stopped (December 2020), as the stopping rule was met for the primary outcome. The clusters comprised hospitals performing > 250 HA and/or KA procedures per annum, whereby all adults (> 18 years) undergoing HA or KA were recruited. Each hospital was randomised to commence with aspirin, orally, 85-150 mg daily or LMWH (enoxaparin), 40 mg, subcutaneously, daily within 24 h postoperatively, for 35 days after HA and 14 days after KA. Crossover was planned once the registration target was met for the first arm. The primary end point is symptomatic VTE within 90 days. Secondary outcomes include readmission, reoperation, major bleeding and death within 90 days, and reoperation and patient-reported pain, function and health status at 6 months. The main analyses will focus on the primary and secondary outcomes for patients undergoing elective primary total HA and KA for osteoarthritis. The analysis will use an intention-to-treat approach with cluster summary methods to compare treatment arms. As the trial stopped early, analyses will account for incomplete cluster crossover and unequal cluster sizes.

CONCLUSIONS

This paper provides a detailed statistical analysis plan for CRISTAL.

TRIAL REGISTRATION

Australian and New Zealand Clinical Trials Registry ACTRN12618001879257 . Registered on 19/11/2018.

Clinical and sociodemographic factors associated with late stage cervical cancer diagnosis in Botswana

BACKGROUND

Cervical cancer is the leading cause of female cancer mortality in Botswana with the majority of cervical cancer patients presenting with late-stage disease. The identification of factors associated with late-stage disease could reduce the cervical cancer burden. This study aims to identify potential patient level clinical and sociodemographic factors associated with a late-stage diagnosis of cervical cancer in Botswana in order to help inform future interventions at the community and individual levels to decrease cervical cancer morbidity and mortality.

RESULTS

There were 984 women diagnosed with cervical cancer from January 2015 to March 2020 at two tertiary hospitals in Gaborone, Botswana. Four hundred forty women (44.7%) presented with late-stage cervical cancer, and 674 women (69.7%) were living with HIV. The mean age at diagnosis was 50.5 years. The association between late-stage (III/IV) cervical cancer at diagnosis and patient clinical and sociodemographic factors was evaluated using multivariable logistic regression with multiple imputation. Women who reported undergoing cervical cancer screening had lower odds of late-stage disease at diagnosis (OR: 0.63, 95% CI 0.47-0.84) compared to those who did not report screening. Women who had never been married had increased odds of late-stage disease at diagnosis (OR: 1.35, 95% CI 1.02-1.86) compared to women who had been married. Women with abnormal vaginal bleeding had higher odds of late-stage disease at diagnosis (OR: 2.32, 95% CI 1.70-3.16) compared to those without abnormal vaginal bleeding. HIV was not associated with a diagnosis of late-stage cervical cancer. Rural women who consulted a traditional healer had increased odds of late-stage disease at diagnosis compared to rural women who had never consulted a traditional healer (OR: 1.61, 95% CI 1.02-2.55).

CONCLUSION

Increasing education and awareness among women, regardless of their HIV status, and among providers, including traditional healers, about the benefits of cervical cancer screening and about the importance of seeking prompt medical care for abnormal vaginal bleeding, while also developing support systems for unmarried women, may help reduce cervical cancer morbidity and mortality in Botswana.

Missing data: A statistical framework for practice

Missing data are ubiquitous in medical research, yet there is still uncertainty over when restricting to the complete records is likely to be acceptable, when more complex methods (e.g. maximum likelihood, multiple imputation and Bayesian methods) should be used, how they relate to each other and the role of sensitivity analysis. This article seeks to address both applied practitioners and researchers interested in a more formal explanation of some of the results. For practitioners, the framework, illustrative examples and code should equip them with a practical approach to address the issues raised by missing data (particularly using multiple imputation), alongside an overview of how the various approaches in the literature relate. In particular, we describe how multiple imputation can be readily used for sensitivity analyses, which are still infrequently performed. For those interested in more formal derivations, we give outline arguments for key results, use simple examples to show how methods relate, and references for full details. The ideas are illustrated with a cohort study, a multi-centre case control study and a randomised clinical trial.

Convalescent plasma therapy and mortality in COVID-19 patients admitted to the ICU: a prospective observational study

Background

This study aimed to quantify the potential survival benefit of convalescent plasma therapy (CVP) in critically ill patients with acute respiratory failure related to coronavirus disease-2019 (COVID-19).

Methods

This is a single-center prospective observational cohort study in COVID-19 patients with acute respiratory failure. Immediately after intensive care unit (ICU) admission patients were allocated to CVP treatment following pre-specified criteria to rapidly identify those patients potentially susceptible for this treatment. A propensity score adjustment [inverse probability of treatment weighted (IPTW) analysis] was implemented to account rigorously for imbalances in prognostic variables between the treatment groups.

Results

We included 120 patients of whom 48 received CVP. Thirty percent were female with a median age of 66 years [25th–75th percentile 54–75]. Eighty-eight percent of patients presented with severe acute respiratory failure as displayed by a median paO₂/FiO₂ ratio (Horowitz Index) of 92 [77–150]. All patients required any kind of ventilatory support with more than half of them (52%) receiving invasive ventilation. Thirty-day ICU overall survival (OS) was 69% in the CVP group and 54% in the non-CVP group (log-rank p = 0.049), respectively. After weighing the time-to-event data for the IPTW, the favorable association between CVP and OS became even stronger (log-rank p = 0.035). Moreover, an exploratory analysis showed an overall survival benefit of CVP therapy for patients with non-invasive ventilation (Hazard ratio 0.12 95% CI 0.03–0.57, p = 0.007)

Conclusion

Administration of CVP in patients with acute respiratory failure related to COVID-19 is associated with improved ICU survival rates.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13613-021-00867-9.

Young adult cancer risk behaviours originate in adolescence: a longitudinal analysis using ALSPAC, a UK birth cohort study

BACKGROUND

An estimated 40% of cancer cases in the UK in 2015 were attributable to cancer risk behaviours. Tobacco smoking, alcohol consumption, obesity, and unprotected sexual intercourse are known causes of cancer and there is strong evidence that physical inactivity is associated with cancer. These cancer risk behaviours co-occur however little is known about how they pattern longitudinally across adolescence and early adulthood. Using data from ALSPAC, a prospective population-based UK birth cohort study, we explored patterns of adolescent cancer risk behaviours and their associations with cancer risk behaviours in early adulthood.

METHODS

Six thousand three hundred fifty-one people (46.0% of ALSPAC participants) provided data on all cancer risk behaviours at one time during adolescence, 1951 provided data on all cancer risk behaviours at all time points. Our exposure measure was quartiles of a continuous score summarising cumulative exposure to cancer risk behaviours and longitudinal latent classes summarising distinct categories of adolescents exhibiting similar patterns of behaviours, between age 11 and 18 years. Using both exposure measures, odds of harmful drinking (Alcohol Use Disorders Identification Test-C ≥ 8),daily tobacco smoking, nicotine dependence (Fagerström test ≥4), obesity (BMI ≥30), high waist circumference (females: ≥80 cm and males: ≥94 cm, and high waist-hip ratio (females: ≥0.85 and males: ≥1.00) at age 24 were estimated using logistic regression analysis.

RESULTS

We found distinct groups of adolescents characterised by consistently high and consistently low engagement in cancer risk behaviours. After adjustment, adolescents in the top quartile had greater odds of all outcomes in early adulthood: nicotine dependency (odds ratio, OR = 5.37, 95% confidence interval, CI = 3.64-7.93); daily smoking (OR = 5.10, 95% CI =3.19-8.17); obesity (OR = 4.84, 95% CI = 3.33-7.03); high waist circumference (OR = 2.48, 95% CI = 1.94-3.16); harmful drinking (OR = 2.04, 95% CI = 1.57-2.65); and high waist-hip ratio (OR = 1.88, 95% CI = 1.30-2.71), compared to the bottom quartile. In latent class analysis, adolescents characterised by consistently high-risk behaviours throughout adolescence were at higher risk of all cancer risk behaviours at age 24, except harmful drinking.

CONCLUSIONS

Exposure to adolescent cancer risk behaviours greatly increased the odds of cancer risk behaviours in early adulthood. Interventions to reduce these behaviours should target multiple rather than single risk behaviours and should focus on adolescence.

Practical strategies for handling breakdown of multiple imputation procedures

Multiple imputation is a recommended method for handling incomplete data problems. One of the barriers to its successful use is the breakdown of the multiple imputation procedure, often due to numerical problems with the algorithms used within the imputation process. These problems frequently occur when imputation models contain large numbers of variables, especially with the popular approach of multivariate imputation by chained equations. This paper describes common causes of failure of the imputation procedure including perfect prediction and collinearity, focusing on issues when using Stata software. We outline a number of strategies for addressing these issues, including imputation of composite variables instead of individual components, introducing prior information and changing the form of the imputation model. These strategies are illustrated using a case study based on data from the Longitudinal Study of Australian Children.

[Leisure Opportunities for Autistic Children and Adolescents: How Satisfied are Parents and what do they Think their Children Would Like?]

Leisure Opportunities for Autistic Children and Adolescents: How Satisfied are Parents and what do they Think their Children Would Like? Like all children and youth, autistic children and adolescents have the right to participate in cultural life including recreational, leisure and sporting activities. Therefore, arrangements have to be made to ensure access to these areas for these particular population (Art. 30 UN-BRK, §§76, 78 BTHG). Thus far, no research has been conducted that examines the need for leisure time facilities for autistic children and/or adolescents in Germany. Therefore, it is unclear if an equal participation in leisure time opportunities is provided. This study presents a first assessment of this topic based on an online-survey of parents of autistic children and adolescents (N = 327). The results indicate, that there is a substantial need for leisure opportunities. Parents of children with infantile and atypical autism show a greater need for more leisure opportunities than parents of children with Asperger Syndrome. In general, the findings show that an equal participation has not been achieved yet. Overall, there is a need for (small) group and sporting activities and especially younger parents asked for specific experiential environments. Possible approaches to a need-oriented organization of leisure facilities for autistic children and adolescents and for an enhancement of leisure opportunities are discussed. The paper further outlines the need for more research in order to provide an empirical basis for equal participation possibilities in public life for autistic children and adolescents.

Hyperglycemia and Adverse Pregnancy Outcome Follow-Up Study: newborn anthropometrics and childhood glucose metabolism

AIMS/HYPOTHESIS

We aimed to examine associations of newborn anthropometric measures with childhood glucose metabolism with the hypothesis that greater newborn birthweight, adiposity and cord C-peptide are associated with higher childhood glucose levels and lower insulin sensitivity.

METHODS

Data from the international, multi-ethnic, population-based Hyperglycemia and Adverse Pregnancy Outcome (HAPO) Study and the HAPO Follow-Up Study were used. The analytic cohort included 4155 children (mean age [SD], 11.4 [1.2] years; 51.0% male). Multiple linear regression was used to examine associations of primary predictors, birthweight, newborn sum of skinfolds (SSF) and cord C-peptide, from HAPO with continuous child glucose outcomes from the HAPO Follow-Up Study.

RESULTS

In an initial model that included family history of diabetes and maternal BMI during pregnancy, birthweight and SSF demonstrated a significant, inverse association with 30 min and 1 h plasma glucose levels. In the primary model, which included further adjustment for maternal sum of glucose z scores from an oral glucose tolerance test during pregnancy, the associations were strengthened, and birthweight and SSF were inversely associated with fasting, 30 min, 1 h and 2 h plasma glucose levels. Birthweight and SSF were also associated with higher insulin sensitivity (Matsuda index) (β = 1.388; 95% CI 0.870, 1.906; p < 0.001; β = 0.792; 95% CI 0.340, 1.244; p < 0.001, for birthweight and SSF higher by 1 SD, respectively) in the primary model, while SSF, but not birthweight, was positively associated with the disposition index, a measure of beta cell compensation for insulin resistance (β = 0.034; 95% CI 0.012, 0.056; p = 0.002). Cord C-peptide levels were inversely associated with Matsuda index (β = -0.746; 95% CI -1.188, -0.304; p < 0.001 for cord C-peptide higher by 1 SD) in the primary model.

CONCLUSIONS/INTERPRETATION

This study demonstrates that higher birthweight and SSF are associated with greater childhood insulin sensitivity and lower glucose levels following a glucose load, associations that were further strengthened after adjustment for maternal glucose levels during pregnancy. Graphical abstract.

The virtue of optimistic realism - expectation fulfillment predicts patient-rated global effectiveness of total hip arthroplasty

BACKGROUND

Emerging evidence highlights the importance of preoperative expectations in predicting patient-reported outcomes of orthopedic surgeries. To date, it is still a matter of controversy whether patient satisfaction can be maximized by promoting either optimistic or realistic outcome expectations before surgery. Adjusting overly optimistic outcome expectancies in favor of a more realistic outlook on the limitations of total hip arthroplasty could reduce the risk of disappointment and lead to greater satisfaction with surgery outcomes. Our prospective cohort study was aimed at comparing the relative predictive influence of baseline expectations, expectation fulfillment and symptomatic improvement on the global effectiveness of total hip arthroplasty.

METHODS

Ninety patients (49 female, 41 male; mean age: 63 ± 12.87 years) fulfilled inclusion criteria and completed a comprehensive preoperative assessment comprising sociodemographic, clinical, functional and psychological phenotypes. Moreover, the strengths of preoperative expectations for improvements in eight pain-related and functional domains were recorded on a 5-point Likert-scale. At 12 months after surgery, patients were asked to rate perceived improvements in each of these domains as well as the global effectiveness of the total hip replacement on a 5-point Likert-scale. To evaluate the relative impact of preoperative expectations, symptom improvement and the fulfillment of expectations on the global effectiveness of surgery, a sequential multiple regression analysis was performed.

RESULTS

Compared with the actual improvement at 12-months follow-up, prior expectations had been overly optimistic in about 28% of patients for hip pain, in about 45% for walking ability and around 60% for back pain, independence in everyday life, physical exercise, general function social interactions and mental well-being. An optimistic hip pain expectation, walking ability at baseline and the fulfillment of expectations for walking ability, general function and independence in everyday life were found to independently predict global effectiveness ratings.

CONCLUSIONS

Positive expectation about pain and the fulfillment of expectations concerning functional domains predicted higher global effectiveness ratings. In line with many authors investigating the relationship between the fulfillment of expectations and satisfaction with medical interventions, we suggest that professionals should explicitly address their patients' expectations during the preoperative education and consultation.

Handling Complex Missing Data Using Random Forest Approach for an Air Quality Monitoring Dataset: A Case Study of Kuwait Environmental Data (2012 to 2018)

In environmental research, missing data are often a challenge for statistical modeling. This paper addressed some advanced techniques to deal with missing values in a data set measuring air quality using a multiple imputation (MI) approach. MCAR, MAR, and NMAR missing data techniques are applied to the data set. Five missing data levels are considered: 5%, 10%, 20%, 30%, and 40%. The imputation method used in this paper is an iterative imputation method, missForest, which is related to the random forest approach. Air quality data sets were gathered from five monitoring stations in Kuwait, aggregated to a daily basis. Logarithm transformation was carried out for all pollutant data, in order to normalize their distributions and to minimize skewness. We found high levels of missing values for NO2 (18.4%), CO (18.5%), PM10 (57.4%), SO2 (19.0%), and O3 (18.2%) data. Climatological data (i.e., air temperature, relative humidity, wind direction, and wind speed) were used as control variables for better estimation. The results show that the MAR technique had the lowest RMSE and MAE. We conclude that MI using the missForest approach has a high level of accuracy in estimating missing values. MissForest had the lowest imputation error (RMSE and MAE) among the other imputation methods and, thus, can be considered to be appropriate for analyzing air quality data.