Multilevel Regression and Poststratification: A Modeling Approach to Estimating Population Quantities From Highly Selected Survey Samples

Bayesian modeling of spatial ordinal data from health surveys

Health surveys allow exploring health indicators that are of great value from a public health point of view and that cannot normally be studied from regular health registries. These indicators are usually coded as ordinal variables and may depend on covariates associated with individuals. In this article, we propose a Bayesian individual-level model for small-area estimation of survey-based health indicators. A categorical likelihood is used at the first level of the model hierarchy to describe the ordinal data, and spatial dependence among small areas is taken into account by using a conditional autoregressive distribution. Post-stratification of the results of the proposed individual-level model allows extrapolating the results to any administrative areal division, even for small areas. We apply this methodology to describe the geographical distribution of a self-perceived health indicator from the Health Survey of the Region of Valencia (Spain) for the year 2016.

Transitive inference in a clinical childhood sample with a focus on autism spectrum disorder

Transitive inference (TI) has a long history in the study of human development. There have, however, few pediatric studies that report clinical diagnoses have tested trial-and-error TI learning, in which participants infer item relations, rather than evaluate them explicitly from verbal descriptions. Children aged 8-10 underwent a battery of clinical assessments and received a range of diagnoses, potentially including autism spectrum disorder (ASD), attention-deficit hyperactive disorder (ADHD), anxiety disorders (AD), specific learning disorders (SLD), and/or communication disorders (CD). Participants also performed a trial-and-error learning task that tested for TI. Response accuracy and reaction time were assessed using a statistical model that controlled for diagnostic comorbidity at the group level. Participants in all diagnostic categories showed evidence of TI. However, a model comparison analysis suggested that those diagnosed with ASD succeeded in a qualitatively different way, responding more slowly to each choice and improving faster across trials than their non-ASD counterparts. Additionally, TI performance was not associated with IQ. Overall, our data suggest that superficially similar performance levels between ASD and non-ASD participants may have resulted from a difference in the speed-accuracy tradeoff made by each group. Our work provides a preliminary profile of the impact of various clinical diagnoses on TI performance in young children. Of these, an ASD diagnosis resulted in the largest difference in task strategy.

Measuring climate change perception in China using mental images: A nationwide open-ended survey

Current knowledge about public climate change perception mainly covers belief, concern, and attitudes. However, how this discourse is interpreted using individuals' own frame of reference remains largely unknown, particularly in many large emitters from non-Annex I countries such as China. This study, for the first time, performs a nationwide open-ended survey covering 4,037 respondents and collected 12,100 textual answers. Using a semiautomated coding method, we find seven mental images that exclusively represent the Chinese interpretation of the climate change issue, including global warming, distant icons, natural disasters, environmental degradation, cause, solution, and weather. Analysis of influencing factors shows that females, those with lower education levels, lower income, and older individuals tend to connect climate change with natural weather phenomena. Younger and well-educated residents in developed cities are more aware of various consequences and anthropogenic causes of climate change. People with stronger climate change beliefs, policy support, and personal experience of extreme weather are more likely to mention disastrous impacts, carbon emission as causes, and potential solutions. Employing the multilevel regression and post-stratification technique, we map the prevalence of mental images in China at the prefecture-city level. The results reveal significant geographical heterogeneity, with estimated national means ranging from a high of 55% (weather) to a low of 11% (solution). Our findings reveal diverse perspectives and a widespread misconception of climate change in China, suggesting the need for tailored clarification strategies to gain public consent.

Seroprevalence of human herpes viruses in France, 2018-2022: a multilevel regression and poststratification approach

BACKGROUND

Information related to herpes simplex virus 1 and 2 (HSV-1 and 2), varicella-zoster virus (VZV), Epstein-Barr virus (EBV), and cytomegalovirus (CMV) seroprevalence in France is either lacking, incomplete, or outdated, despite their public health burden.

METHOD

We used routinely collected serological data between 2018 and 2022 to estimate HSV-1, HSV-2, VZV, EBV, and CMV seroprevalence in France. To account for demographic differences between our analytic samples and the French population and get estimates for sparsely sampled districts and age classes, we used a multilevel regression and poststratification approach combined with Bayesian model averaging via stacking weights.

RESULTS

The observed seroprevalence (number of positive tests/number of tests) were 64.6% (93,294/144,424), 16.9% (24,316/144,159), 93.0% (141,419/152,084), 83.4% (63,199/75, 781), and 49.0% (23,276/47,525), respectively, for HSV-1, HSV-2, VZV, EBV, and CMV. Between 2018 and 2022, France had a model-based average (equal-tailed interval at 95%) expected seroprevalence equal to 61.1% (60.7,61.5), 14.5% (14.2,14.81), 89.5% (89.3,89.8), 85.6% (85.2,86.0), and 50.5% (49.3,51.7), respectively, for HSV-1, HSV-2, VZV, EBV, and CMV infections. We found an almost certain lower expected seroprevalence in Metropolitan France than in overseas territories for all viruses but VZV, for which it was almost certainly greater. The expected seroprevalences were likely greater among females for all viruses.

LIMITATIONS

Our results relied on the assumption that individuals were sampled at random conditionally to variables used to build the poststratification table.

IMPLICATIONS

The analysis highlights spatial and demographic patterns in seroprevalence that should be considered for designing tailored public health policies.

Research protocol for an observational health data analysis to assess the applicability of randomized controlled trials focusing on newly diagnosed metastatic prostate cancer using real-world data: PIONEER IMI's "big data for better outcomes" program

Background

Metastatic prostate cancer (PCa) constitutes ~5% of all new PCa diagnoses in Western countries. For most cases, primary consideration should be given to systemic therapies as the first-line approach based on evidence from randomized controlled trials (RCTs). Despite the importance of RCTs as the pinnacle of evidence in modern medicine, concerns have been raised about their applicability to real-life scenarios. These trials often feature participants who are younger with better performance statuses and prognoses compared to their real-world counterparts. The PIONEER project falls under the Innovative Medicine Initiative's (IMI) "Big Data for Better Outcomes" initiative, aimed at revolutionizing PCa care in Europe. The central focus lies in improving cancer-related outcomes, enhancing health system efficiency, and elevating the quality of health and social care. This study endeavours to evaluate the generalizability of RCT findings concerning newly diagnosed metastatic PCa.

Methods

A systematic review of the literature will be conducted to compile patient characteristics from RCTs addressing this subject within the past decade. To create a real-world benchmark, patients with recently diagnosed metastatic PCa from a network of population-based databases will serve as a comparison group. The objective is to assess the applicability of RCT results in two ways. First, a comparison will be made between the characteristics of patients with newly diagnosed metastatic PCa enroled in RCTs and those with the same condition included in our databases which might represent the real-world setting. Second, an evaluation will be undertaken to determine the proportion of real-world patients with newly diagnosed metastatic PCa who meet the criteria for RCT enrolment. This study will rely on extensive observational data, primarily sourced from population-based registries, electronic health records, and insurance claims data. The study cohort is established upon routinely gathered healthcare data, meticulously mapped to the Observational Medical Outcomes Partnership Common Data Model.

Isolated rapid eye movement sleep behaviour disorder (iRBD) in the Island Study Linking Ageing and Neurodegenerative Disease (ISLAND) Sleep Study: protocol and baseline characteristics

Isolated rapid eye movement (REM) sleep behaviour disorder (iRBD) is a sleep disorder that is characterised by dream enactment episodes during REM sleep. It is the strongest known predictor of α-synuclein-related neurodegenerative disease (αNDD), such that >80% of people with iRBD will eventually develop Parkinson's disease, dementia with Lewy bodies, or multiple system atrophy in later life. More research is needed to understand the trajectory of phenoconversion to each αNDD. Only five 'gold standard' prevalence studies of iRBD in older adults have been undertaken previously, with estimates ranging from 0.74% to 2.01%. The diagnostic recommendations for video-polysomnography (vPSG) to confirm iRBD makes prevalence studies challenging, as vPSG is often unavailable to large cohorts. In Australia, there have been no iRBD prevalence studies, and little is known about the cognitive and motor profiles of Australian people with iRBD. The Island Study Linking Ageing and Neurodegenerative Disease (ISLAND) Sleep Study will investigate the prevalence of iRBD in Tasmania, an island state of Australia, using validated questionnaires and home-based vPSG. It will also explore several cognitive, motor, olfactory, autonomic, visual, tactile, and sleep profiles in people with iRBD to better understand which characteristics influence the progression of iRBD to αNDD. This paper details the ISLAND Sleep Study protocol and presents preliminary baseline results.

Using leave-one-out cross validation (LOO) in a multilevel regression and poststratification (MRP) workflow: A cautionary tale

In recent decades, multilevel regression and poststratification (MRP) has surged in popularity for population inference. However, the validity of the estimates can depend on details of the model, and there is currently little research on validation. We explore how leave-one-out cross validation (LOO) can be used to compare Bayesian models for MRP. We investigate two approximate calculations of LOO: Pareto smoothed importance sampling (PSIS-LOO) and a survey-weighted alternative (WTD-PSIS-LOO). Using two simulation designs, we examine how accurately these two criteria recover the correct ordering of model goodness at predicting population and small-area estimands. Focusing first on variable selection, we find that neither PSIS-LOO nor WTD-PSIS-LOO correctly recovers the models' order for an MRP population estimand, although both criteria correctly identify the best and worst model. When considering small-area estimation, the best model differs for different small areas, highlighting the complexity of MRP validation. When considering different priors, the models' order seems slightly better at smaller-area levels. These findings suggest that, while not terrible, PSIS-LOO-based ranking techniques may not be suitable to evaluate MRP as a method. We suggest this is due to the aggregation stage of MRP, where individual-level prediction errors average out. We validate these results by applying to the real world National Health and Nutrition Examination Survey (NHANES) data in the United States. Altogether, these results show that PSIS-LOO-based model validation tools need to be used with caution and might not convey the full story when validating MRP as a method.

Embedded multilevel regression and poststratification: Model-based inference with incomplete auxiliary information

Health disparity research often evaluates health outcomes across demographic subgroups. Multilevel regression and poststratification (MRP) is a popular approach for small subgroup estimation as it can stabilize estimates by fitting multilevel models and adjust for selection bias by poststratifying on auxiliary variables, which are population characteristics predictive of the analytic outcome. However, the granularity and quality of the estimates produced by MRP are limited by the availability of the auxiliary variables' joint distribution; data analysts often only have access to the marginal distributions. To overcome this limitation, we embed the estimation of population cell counts needed for poststratification into the MRP workflow: embedded MRP (EMRP). Under EMRP, we generate synthetic populations of the auxiliary variables before implementing MRP. All sources of estimation uncertainty are propagated with a fully Bayesian framework. Through simulation studies, we compare different methods of generating the synthetic populations and demonstrate EMRP's improvements over alternatives on the bias-variance tradeoff to yield valid subpopulation inferences of interest. We apply EMRP to the Longitudinal Survey of Wellbeing and estimate food insecurity prevalence among vulnerable groups in New York City. We find that all EMRP estimators can correct for the bias in classical MRP while maintaining lower standard errors and narrower confidence intervals than directly imputing with the weighted finite population Bayesian bootstrap (WFPBB) and design-based estimates. Performances from the EMRP estimators do not differ substantially from each other, though we would generally recommend using the WFPBB-MRP for its consistently high coverage rates.

Characterisation of infection-induced SARS-CoV-2 seroprevalence amongst children and adolescents in North Carolina

Few prospective studies have documented the seropositivity among those children infected with severe acute respiratory syndrome coronavirus 2. From 2 April 2021 to 24 June 2021, we prospectively enrolled children between the ages of 2 and 17 years at three North Carolina healthcare systems. Participants received at least four at-home serological tests detecting the presence of antibodies against, but not differentiating between, the nucleocapsid or spike antigen. A total of 1,058 participants were enrolled in the study, completing 2,709 tests between 1 May 2021 and 31 October 2021. Using multilevel regression with poststratification techniques and considering our assay sensitivity and sensitivity, we estimated that the seroprevalence of infection-induced antibodies among unvaccinated children and adolescents aged 2-17 years in North Carolina increased from 15.2% (95% credible interval, CrI 9.0-22.0) in May 2021 to 54.1% (95% CrI 46.7-61.1) by October 2021, indicating an average infection-to-reported-case ratio of 5. A rapid rise in seropositivity was most pronounced in those unvaccinated children aged 12-17 years, based on our estimates. This study underlines the utility of serial, serological testing to inform a broader understanding of the regional immune landscape and spread of infection.

Cyr SB, Pham CD, Kreisel KM. Assessing the National Representativeness of Estimates of Antimicrobial-Resistant Urogenital Neisseria gonorrhoeae in US Men, Gonococcal Isolate Surveillance Project, 2008-2018

BACKGROUND

The percentage of Neisseria gonorrhoeae (GC) isolates with resistance or elevated minimum inhibitory concentrations to antimicrobials has steadily increased. Current estimates are based on the Gonococcal Isolate Surveillance Project (GISP), a sentinel surveillance study of male GC in the United States. This analysis seeks to assess for adjustment before treating aggregated GISP estimates as nationally representative of all reported male urogenital infections.

METHODS

We used multilevel regression with poststratification (MRP) to compute national estimates of the proportion of antimicrobial resistance (AMR) (defined as exceeding minimum inhibitory concentration thresholds) in male GC using data from 2008 to 2018 GISP and case reports. Sensitivity analyses investigated the impact of analysis assumptions and unmeasured variables. We additionally produced estimates of 2018 AMR GC cases among US men.

RESULTS

National estimates were consistent with unweighted estimates. The estimated proportion of incident AMR GC infections in men with urogenital GC in 2018 was 51.5% (95% confidence interval [CI], 50.1%-52.9%), equating to an estimated 366,300 incident AMR GC infections in US men aged 15 to 39 years. Estimates of AMR for tested antimicrobials in male GC infections in 2018 ranged from 0.16% (95% CI, 0.08%-0.24%) for ceftriaxone to 29.9% (95% CI, 28.6%-31.1%) for ciprofloxacin. Sensitivity analyses revealed that unmeasured data on sex of sex partners could substantially impact weighted estimates.

CONCLUSIONS

Antimicrobial resistance among reported incident male urogenital GC infections remains rare for ceftriaxone, the current standard of care. Aggregated GISP data are generally representative of men in the US who are reported with urogenital gonorrhea.

Geographic and demographic variation in worry about extreme heat and COVID-19 risk in summer 2020

Extreme heat is a major health hazard that is exacerbated by ongoing human-caused climate change. However, how populations perceive the risks of heat in the context of other hazards like COVID-19, and how perceptions vary geographically, are not well understood. Here we present spatially explicit estimates of worry among the U.S. public about the risks of heat and COVID-19 during the summer of 2020, using nationally representative survey data and a multilevel regression and poststratification (MRP) model. Worry about extreme heat and COVID-19 varies both across states and across demographic groups, in ways that reflect disparities in the impact of each risk. Black or African American and Hispanic or Latino populations, who face greater health impacts from both COVID-19 and extreme heat due to institutional and societal inequalities, also tend to be much more worried about both risks than white, non-Hispanic populations. Worry about heat and COVID-19 were correlated at the individual and population level, and patterns tended to be related to underlying external factors associated with the risk environment. In the face of a changing climate there is an urgent need to address disparities in heat risk and develop responses that ensure the most at-risk populations are protected.

Observed versus estimated actual trend of COVID-19 case numbers in Cameroon: A data-driven modelling

Controlling the COVID-19 outbreak remains a challenge for Cameroon, as it is for many other countries worldwide. The number of confirmed cases reported by health authorities in Cameroon is based on observational data, which is not nationally representative. The actual extent of the outbreak from the time when the first case was reported in the country to now remains unclear. This study aimed to estimate and model the actual trend in the number of COVID -19 new infections in Cameroon from March 05, 2020 to May 31, 2021 based on an observed disaggregated dataset. We used a large disaggregated dataset, and multilevel regression and poststratification model was applied prospectively for COVID-19 cases trend estimation in Cameroon from March 05, 2020 to May 31, 2021. Subsequently, seasonal autoregressive integrated moving average (SARIMA) modeling was used for forecasting purposes. Based on the prospective MRP modeling findings, a total of about 7450935 (30%) of COVID-19 cases was estimated from March 05, 2020 to May 31, 2021 in Cameroon. Generally, the reported number of COVID-19 infection cases in Cameroon during this period underestimated the estimated actual number by about 94 times. The forecasting indicated a succession of two waves of the outbreak in the next two years following May 31, 2021. If no action is taken, there could be many waves of the outbreak in the future. To avoid such situations which could be a threat to global health, public health authorities should effectively monitor compliance with preventive measures in the population and implement strategies to increase vaccination coverage in the population.

Novel estimates reveal subnational heterogeneities in disease-relevant contact patterns in the United States

Population contact patterns fundamentally determine the spread of directly transmitted airborne pathogens such as SARS-CoV-2 and influenza. Reliable quantitative estimates of contact patterns are therefore critical to modeling and reducing the spread of directly transmitted infectious diseases and to assessing the effectiveness of interventions intended to limit risky contacts. While many countries have used surveys and contact diaries to collect national-level contact data, local-level estimates of age-specific contact patterns remain rare. Yet, these local-level data are critical since disease dynamics and public health policy typically vary by geography. To overcome this challenge, we introduce a flexible model that can estimate age-specific contact patterns at the subnational level by combining national-level interpersonal contact data with other locality-specific data sources using multilevel regression with poststratification (MRP). We estimate daily contact matrices for all 50 US states and Washington DC from April 2020 to May 2021 using national contact data from the US. Our results reveal important state-level heterogeneities in levels and trends of contacts across the US over the course of the COVID-19 pandemic, with implications for the spread of respiratory diseases.

Limitations introduced by a low participation rate of SARS-CoV-2 seroprevalence data

BACKGROUND

There has been a large influx of COVID-19 seroprevalence studies, but comparability between the seroprevalence estimates has been an issue because of heterogeneities in testing platforms and study methodology. One potential source of heterogeneity is the response or participation rate.

METHODS

We conducted a review of participation rates (PR) in SARS-CoV-2 seroprevalence studies collected by SeroTracker and examined their effect on the validity of study conclusions. PR was calculated as the count of participants for whom the investigators had collected a valid sample, divided by the number of people invited to participate in the study. A multivariable beta generalized linear model with logit link was fitted to determine if the PR of international household and community-based seroprevalence studies was associated with the factors of interest, from 1 December 2019 to 10 March 2021.

RESULTS

We identified 90 papers based on screening and were able to calculate the PR for 35 out of 90 papers (39%), with a median PR of 70% and an interquartile range of 40.92; 61% of the studies did not report PR.

CONCLUSIONS

Many SARS-CoV-2 seroprevalence studies do not report PR. It is unclear what the median PR rate would be had a larger portion not had limitations in reporting. Low participation rates indicate limited representativeness of results. Non-probabilistic sampling frames were associated with higher participation rates but may be less representative. Standardized definitions of participation rate and data reporting necessary for the PR calculations are essential for understanding the representativeness of seroprevalence estimates in the population of interest.

Beyond Vaccination Rates: A Synthetic Random Proxy Metric of Total SARS-CoV-2 Immunity Seroprevalence in the Community

BACKGROUND

Explicit knowledge of total community-level immune seroprevalence is critical to developing policies to mitigate the social and clinical impact of SARS-CoV-2. Publicly available vaccination data are frequently cited as a proxy for population immunity, but this metric ignores the effects of naturally acquired immunity, which varies broadly throughout the country and world. Without broad or random sampling of the population, accurate measurement of persistent immunity post-natural infection is generally unavailable.

METHODS

To enable tracking of both naturally acquired and vaccine-induced immunity, we set up a synthetic random proxy based on routine hospital testing for estimating total immunoglobulin G (IgG) prevalence in the sampled community. Our approach analyzed viral IgG testing data of asymptomatic patients who presented for elective procedures within a hospital system. We applied multilevel regression and poststratification to adjust for demographic and geographic discrepancies between the sample and the community population. We then applied state-based vaccination data to categorize immune status as driven by natural infection or by vaccine.

RESULTS

We validated the model using verified clinical metrics of viral and symptomatic disease incidence to show the expected biologic correlation of these entities with the timing, rate, and magnitude of seroprevalence. In mid-July 2021, the estimated immunity level was 74% with the administered vaccination rate of 45% in the two counties.

CONCLUSIONS

Our metric improves real-time understanding of immunity to COVID-19 as it evolves and the coordination of policy responses to the disease, toward an inexpensive and easily operational surveillance system that transcends the limits of vaccination datasets alone.

A Nationwide Evaluation of the Prevalence of Human Papillomavirus in Brazil (POP-Brazil Study): Protocol for Data Quality Assurance and Control

BACKGROUND

The credibility of a study and its internal and external validity depend crucially on the quality of the data produced. An in-depth knowledge of quality control processes is essential as large and integrative epidemiological studies are increasingly prioritized.

OBJECTIVE

This study aimed to describe the stages of quality control in the POP-Brazil study and to present an analysis of the quality indicators.

METHODS

Quality assurance and control were initiated with the planning of this nationwide, multicentric study and continued through the development of the project. All quality control protocol strategies, such as training, protocol implementation, audits, and inspection, were discussed one by one. We highlight the importance of conducting a pilot study that provides the researcher the opportunity to refine or modify the research methodology and validating the results through double data entry, test-retest, and analysis of nonresponse rates.

RESULTS

This cross-sectional, nationwide, multicentric study recruited 8628 sexually active young adults (16-25 years old) in 119 public health units between September 2016 and November 2017. The Human Research Ethics Committee of the Moinhos de Vento Hospital approved this project.

CONCLUSIONS

Quality control processes are a continuum, not restricted to a single event, and are fundamental to the success of data integrity and the minimization of bias in epidemiological studies. The quality control steps described can be used as a guide to implement evidence-based, valid, reliable, and useful procedures in most observational studies to ensure data integrity.

INTERNATIONAL REGISTERED REPORT IDENTIFIER (IRRID)

RR1-10.2196/31365.

Routine Hospital-based SARS-CoV-2 Testing Outperforms State-based Data in Predicting Clinical Burden

Throughout the coronavirus disease 2019 (COVID-19) pandemic, government policy and healthcare implementation responses have been guided by reported positivity rates and counts of positive cases in the community. The selection bias of these data calls into question their validity as measures of the actual viral incidence in the community and as predictors of clinical burden. In the absence of any successful public or academic campaign for comprehensive or random testing, we have developed a proxy method for synthetic random sampling, based on viral RNA testing of patients who present for elective procedures within a hospital system. We present here an approach under multilevel regression and poststratification to collecting and analyzing data on viral exposure among patients in a hospital system and performing statistical adjustment that has been made publicly available to estimate true viral incidence and trends in the community. We apply our approach to tracking viral behavior in a mixed urban-suburban-rural setting in Indiana. This method can be easily implemented in a wide variety of hospital settings. Finally, we provide evidence that this model predicts the clinical burden of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) earlier and more accurately than currently accepted metrics. See video abstract at, http://links.lww.com/EDE/B859.

Improving multilevel regression and poststratification with structured priors

A central theme in the field of survey statistics is estimating population-level quantities through data coming from potentially non-representative samples of the population. Multilevel regression and poststratification (MRP), a model-based approach, is gaining traction against the traditional weighted approach for survey estimates. MRP estimates are susceptible to bias if there is an underlying structure that the methodology does not capture. This work aims to provide a new framework for specifying structured prior distributions that lead to bias reduction in MRP estimates. We use simulation studies to explore the benefit of these prior distributions and demonstrate their efficacy on non-representative US survey data. We show that structured prior distributions offer absolute bias reduction and variance reduction for posterior MRP estimates in a large variety of data regimes.

Using Multilevel Regression and Poststratification to Estimate Physical Activity Levels from Health Surveys

BACKGROUND

Large-scale health surveys often consider sociodemographic characteristics and several health indicators influencing physical activity that often vary across subpopulations. Data in a survey for some small subpopulations are often not representative of the larger population.

OBJECTIVE

We developed a multilevel regression and poststratification (MRP) model to estimate leisure-time physical activity across Brazilian state capitals and evaluated whether the MRP outperforms single-level regression estimates based on the Brazilian cross-sectional national survey VIGITEL (2018).

METHODS

We used various approaches to compare the MRP and single-level model (complete-pooling) estimates, including cross-validation with various subsample proportions tested.

RESULTS

MRP consistently had predictions closer to the estimation target than single-level regression estimations. The mean absolute errors were smaller for the MRP estimates than single-level regression estimates with smaller sample sizes. MRP presented substantially smaller uncertainty estimates compared to single-level regression estimates. Overall, the MRP was superior to single-level regression estimates, particularly with smaller sample sizes, yielding smaller errors and more accurate estimates.

CONCLUSION

The MRP is a promising strategy to predict subpopulations' physical activity indicators from large surveys. The observations present in this study highlight the need for further research, which could, potentially, incorporate more information in the models to better interpret interactions and types of activities across target populations.

Obesity and occupation in Thailand: using a Bayesian hierarchical model to obtain prevalence estimates from the National Health Examination Survey

BACKGROUND

Like many developing countries, Thailand has experienced a rapid rise in obesity, accompanied by a rapid change in occupational structure. It is plausible that these two trends are related, with movement into sedentary occupations leading to increases in obesity. National health examination survey data contains information on obesity and socioeconomic conditions that can help untangle the relationship, but analysis is challenging because of small sample sizes.

METHODS

This paper explores the relationship between occupation and obesity using data on 10,127 respondents aged 20-59 from the 2009 National Health Examination Survey. Obesity is measured using waist circumference. Modelling is carried out using an approach known as Multiple Regression with Post-Stratification (MRP). We use Bayesian hierarchical models to construct prevalence estimates disaggregated by age, sex, education, urban-rural residence, region, and occupation, and use census population weights to aggregate up. The Bayesian hierarchical model is designed to protect against overfitting and false discovery, which is particularly important in an exploratory study such as this one.

RESULTS

There is no clear relationship between the overall sedentary nature of occupations and obesity. Instead, obesity appears to vary occupation by occupation. For instance, women in professional occupations, and men who are agricultural or fishery workers, have relatively low rates of obesity.

CONCLUSION

Bayesian hierarchical models plus post-stratification offers new possibilities for using surveys to learn about complex health issues.

Addressing Public Health Emergencies via Facebook Surveys: Advantages, Challenges, and Practical Considerations

Surveys of the general population can provide crucial information for designing effective nonpharmaceutical interventions to tackle public health emergencies, such as the COVID-19 pandemic. Yet, conducting such surveys can be difficult, especially when timely data collection is required. In this viewpoint paper, we discuss our experiences with using targeted Facebook advertising campaigns to address these difficulties in relation to the COVID-19 pandemic. We describe central advantages, challenges, and practical considerations. This includes a discussion of potential sources of bias and how they can be addressed.

The COVID-19 Social Monitor longitudinal online panel: Real-time monitoring of social and public health consequences of the COVID-19 emergency in Switzerland

BACKGROUND

The COVID-19 pandemic challenges societies in unknown ways, and individuals experience a substantial change in their daily lives and activities. Our study aims to describe these changes using population-based self-reported data about social and health behavior in a random sample of the Swiss population during the COVID-19 pandemic. The aim of the present article is two-fold: First, we want to describe the study methodology. Second, we want to report participant characteristics and study findings of the first survey wave to provide some baseline results for our study.

METHODS

Our study design is a longitudinal online panel of a random sample of the Swiss population. We measure outcome indicators covering general well-being, physical and mental health, social support, healthcare use and working state over multiple survey waves.

RESULTS

From 8,174 contacted individuals, 2,026 individuals participated in the first survey wave which corresponds to a response rate of 24.8%. Most survey participants reported a good to very good general life satisfaction (93.3%). 41.4% of the participants reported a worsened quality of life compared to before the COVID-19 emergency and 9.8% feelings of loneliness.

DISCUSSION

The COVID-19 Social Monitor is a population-based online survey which informs the public, health authorities, and the scientific community about relevant aspects and potential changes in social and health behavior during the COVID-19 emergency and beyond. Future research will follow up on the described study population focusing on COVID-19 relevant topics such as subgroup differences in the impact of the pandemic on well-being and quality of life or different dynamics of perceived psychological distress.

Multilevel regression and poststratification for estimating population quantities from large health studies: a simulation study based on US population structure

INTRODUCTION

Recruiting a representative sample of participants is becoming increasingly difficult in large-scale health surveys. Multilevel regression and poststratification (MRP) has been shown to be effective in estimating population descriptive quantities in non-representative samples. We performed a simulation study, previously applied to an Australian population, this time to a US population, to assess MRP performance.

METHODS

Data were extracted from the 2017 Current Population Survey representing a population of US adult males aged 18-55 years. Simulated datasets of non-representative samples were generated. State-level prevalence estimates for a dichotomous outcome using MRP were compared with the use of sampling weights (with and without raking adjustment). We also investigated the impact on MRP performance of sample size, model misspecification, interactions and the addition of a geographic-level covariate.

RESULTS

MRP was found to achieve generally superior performance, with large gains in precision vastly outweighing the increased accuracy observed for sampling weights with raking adjustment. MRP estimates were generally robust to model misspecification. We found a tendency of MRP to over-pool between-state variation in the outcome, particularly for the least populous states and small sample sizes. The inclusion of a state-level covariate appeared to mitigate this and further improve MRP performance.

DISCUSSION

MRP has been shown to be effective in estimating population descriptive quantities in two different populations. This provides promising evidence for the general applicability of MRP to populations with different geographic structures. MRP appears to be a valuable analytic strategy for addressing potential participation bias from large-scale health surveys.

Multilevel Regression and Poststratification Versus Survey Sample Weighting for Estimating Population Quantities in Large Population Health Studies

Multilevel regression and poststratification (MRP) is a model-based approach for estimating a population parameter of interest, generally from large-scale surveys. It has been shown to be effective in highly selected samples, which is particularly relevant to investigators of large-scale population health and epidemiologic surveys facing increasing difficulties in recruiting representative samples of participants. We aimed to further examine the accuracy and precision of MRP in a context where census data provided reasonable proxies for true population quantities of interest. We considered 2 outcomes from the baseline wave of the Ten to Men study (Australia, 2013-2014) and obtained relevant population data from the 2011 Australian Census. MRP was found to achieve generally superior performance relative to conventional survey weighting methods for the population as a whole and for population subsets of varying sizes. MRP resulted in less variability among estimates across population subsets relative to sample weighting, and there was some evidence of small gains in precision when using MRP, particularly for smaller population subsets. These findings offer further support for MRP as a promising analytical approach for addressing participation bias in the estimation of population descriptive quantities from large-scale health surveys and cohort studies.

Factors Influencing Student-Athletes' Identity: A Multilevel Regression and Poststratification Approach

We considered identity variation among Brazilian university student-athletes in relation to their gender, sport type, competition level, and university type. Participants were 506 student-athletes (219 males and 287 females) from public and private Brazilian universities, competing in team and individual sports, at local, state, and national levels. We used multilevel regression and poststratification to estimate each participant’s identity from the aforementioned variables. Gender and sport type were not associated with any substantial identify variation, but there were higher values on Baller Identity Measurement Scale dimensions for student-athletes from public versus private universities, and student-athletes competing at the highest level had lower Baller Identity Measurement Scale values compared to peers competing at lower levels. Overall, university type and sport competitive level were the contextual factors that most influenced Brazilian student-athletes’ identities.

Student-Athletes' Motivation and Identity: Variation Among Brazilian and Portuguese University Student-Athletes

This study examined the variation of student-athletes' identity and motivation across Portuguese and Brazilian universities, accounting for variation in gender, student-athletes' training hours per week, sports level, student-athletes status within each university, and university type. We initially established the validity of the Baller Identity Measurement Scale questionnaire and the Student-Athletes' Motivation toward Sports and Academics Questionnaire-based observations among 441 Brazilian and Portuguese student-athletes. Then, the validated version of the questionnaires was applied to a total sample of 765 student-athletes from Brazil (n= 568) and Portugal (n = 197). We further considered individual (hours of training and student-athlete status) and contextual characteristics (university type and country). Multilevel regression and poststratification were used to estimate each student-athlete identity and motivation as a function of his or her individual and contextual characteristics. Overall, the predictions showed that cultural (country), academic (type of university), and athletic (training hours) context likely have a substantial influence on student-athletes' identity and motivation.

Bias from self selection and loss to follow-up in prospective cohort studies

Self-selection into prospective cohort studies and loss to follow-up can cause biased exposure-outcome association estimates. Previous investigations illustrated that such biases can be small in large prospective cohort studies. The structural approach to selection bias shows that general statements about bias are not possible for studies that investigate multiple exposures and outcomes, and that inverse probability of participation weighting (IPPW) but not adjustment for participation predictors generally reduces bias from self-selection and loss to follow-up. We propose to substantiate assumptions in structural models of selection bias through calculation of genetic correlations coefficients between participation predictors, outcome, and exposure, and to estimate a lower bound for bias due to self-selection and loss to follow-up by comparing effect estimates from IPP weighted and unweighted analyses. This study used data from the Norwegian Mother and Child Cohort Study and the Medical Birth Registry of Norway. Using the example of risk factors for ADHD, we find that genetic correlations between participation predictors, exposures, and outcome suggest the presence of bias. The comparison of exposure-outcome associations from regressions with and without IPPW revealed meaningful deviations. Assessment of selection bias for entire multi-exposure multi-outcome cohort studies is not possible. Instead, it has to be assessed and controlled on a case-by-case basis.

Multilevel regression and poststratification as a modeling approach for estimating population quantities in large population health studies: A simulation study

There are now a growing number of applications of multilevel regression and poststratification (MRP) in population health and epidemiological studies. MRP uses multilevel regression to model individual survey responses as a function of demographic and geographic covariates. Estimated mean outcome values for each demographic-geographic respondent subtype are then weighted by the proportions of each subtype in the population to produce an overall population-level estimate. We recently reported an extensive case study of a large nationwide survey and found that MRP performed favorably compared to conventional survey sampling weights for the estimation of population descriptive quantities in a highly selected sample. In this study, we aimed to evaluate, by way of a simulation experiment, both the accuracy and precision of MRP versus survey sampling weights in the context of large population health studies. While much of the research into MRP has been focused on U.S. political and social science, we considered an alternative population structure of smaller size and with notably fewer geographic subsets. We explored the impact on MRP performance of sample size, model misspecification, interactions, and the addition of a geographic-level covariate. MRP was found to achieve generally superior performance in both accuracy and precision at both the national and state levels. Results were generally robust to model misspecification, and MRP performance was further improved by the inclusion of a geographic-level covariate. These findings offer further evidence that MRP provides a promising analytic approach for addressing participation bias in the estimation of population descriptive quantities from large-scale health surveys and cohort studies.

Obtaining contextually relevant geographic data using Facebook recruitment in public health studies

INTRODUCTION

Online participant recruitment for public health research studies has increased dramatically in recent years, particularly as traditional recruitment strategies have waned in efficiency. The emergence of e-epidemiology offers possibilities for reaching understudied populations as well as conducting large-scale studies.

METHODS

We conducted a cross-sectional survey focused on self-reported neighborhood characteristics, perceived stress, and feasibility of obtaining work/residential addresses via online recruitment in St. Louis, Missouri, USA from February 2017 to December 2017. We report the process of using Facebook recruitment and demonstrate how this strategy can enhance collection of geospatial data to better understand context and spatial patterns of disease.

RESULTS

A total of 425 participants were recruited via Facebook advertisements. All participants reported their residential and work ZIP codes, though only 64.7% and 45.6% provided their complete residential and work street addresses, respectively. Those who reported their complete residential street addresses were more likely to be female (69.8% vs. 56.1% of males, χ² = 7.89, 1 df, p = 0.005), though no differences were observed by race, age, or employment status.

DISCUSSION

These findings indicate that valuable location data can be successfully collected via Facebook recruitment - data that could potentially include residential history or prospective follow-up time or be combined with other emerging technologies for geographic data in order to better understand the context and the effects of place on health outcomes.

CONCLUSIONS

Facebook recruitment may be an underutilized resource for obtaining accurate geospatial and contextually relevant health data and should be considered as a means for finding participants due to the cost-effectiveness, efficiency and flexibility of this recruitment approach.