Selection Mechanisms and Their Consequences: Understanding and Addressing Selection Bias

Modeling the relative influence of socio-demographic variables on post-acute COVID-19 quality of life

Background

Post-acute sequelae of SARS-CoV-2, referred to as "long COVID", are a globally pervasive threat. While their many clinical determinants are commonly considered, their plausible social correlates are often overlooked.

Methods

Here, we use data from a multinational prospective cohort study to compare social and clinical predictors of differences in quality of life with long COVID. We further measure the extent to which clinical intermediates may explain relationships between social variables and quality of life with long COVID.

Findings

Beyond age, neuropsychological and rheumatological comorbidities, educational attainment, employment status, and female sex were important predictors of long COVID-associated quality of life days (long COVID QALDs). Furthermore, most of their associations could not be attributed to key long COVID-predicting comorbidities. In Norway, 90% (95% CI: 77%, 100%) of the adjusted association between belonging to the top two quintiles of educational attainment and long COVID QALDs was not explained by these clinical intermediates. The same was true for 86% (73%, 100%) and 93% (80%,100%) of the adjusted association between full-time employment and long COVID QALDs in the United Kingdom (UK) and Russia. Additionally, 77% (46%,100%) and 73% (52%, 94%) of the adjusted associations between female sex and long COVID QALDs in Norway and the UK were unexplained by the clinical mediators.

Interpretation

Our findings highlight that socio-economic proxies and sex are key predictors of long COVID QALDs and that other (non-clinical) mechanisms drive their observed relationships. Importantly, we outline a multi-method, adaptable causal approach for evaluating the isolated contributions of social disparities to experiences with long COVID.

Funding

UK Foreign, Commonwealth and Development Office; Wellcome Trust; Bill & Melinda Gates Foundation; Oxford COVID-19 Research Response Funding; UK National Institute for Health and Care Research; UK Medical Research Council; Public Health England; Liverpool Experimental Cancer Medicine Centre; Research Council of Norway; Vivaldi Invest A/S; South Eastern Norway Health Authority.

The associations between religious/spiritual beliefs and behaviours and study participation in a prospective cohort study (ALSPAC) in Southwest England

Background

Longitudinal studies are key to understanding risk factors for health, well-being, and disease, yet associations may be biased if study invitation and participation are non-random. Religious/spiritual beliefs and behaviours (RSBB) are increasingly recognised as having potentially important relationships with health. However, it is unclear whether RSBB is associated with study participation. We examine whether RSBB is associated with participation in the longitudinal birth cohort ALSPAC (Avon Longitudinal Study of Parents and Children).

Methods

Three RSBB factors were used: religious belief (belief in God/a divine power; yes/not sure/no), religious affiliation (Christian/none/other), and religious attendance (frequency of attendance at a place of worship). Participation was measured in three ways: i) total number of questionnaires/clinics completed (linear and ordinal models); ii) completion of the most recent questionnaire (logistic model); and iii) length of participation (survival model). Analyses were repeated for the ALSPAC mothers, their partners, and the study children, and were adjusted for relevant socio-demographic confounders.

Results

Religious attendance was positively associated with participation in all adjusted models in all three cohorts. For example, study mothers who attended a place of worship at least once a month on average completed two more questionnaires (out of a possible 50), had 50% greater odds of having completed the most recent questionnaire, and had 25% reduced risk of drop-out, relative to those who did not attend. In the adjusted analyses, religious belief and attendance were not associated with participation. However, the majority of unadjusted models showed associations between RSBB and participation.

Conclusion

After adjusting for confounders, religious attendance - not religious belief or affiliation - was associated with participation in ALSPAC. These results indicate that use of RSBB variables (and religious attendance in particular) may result in selection bias and spurious associations; these potential biases should be explored and discussed in future studies using these data.

Inverse probability weighting for self-selection bias correction in the investigation of social inequality in mortality

BACKGROUND

Empirical evaluation of inverse probability weighting (IPW) for self-selection bias correction is inaccessible without the full source population. We aimed to: (i) investigate how self-selection biases frequency and association measures and (ii) assess self-selection bias correction using IPW in a cohort with register linkage.

METHODS

The source population included 17 936 individuals invited to the Copenhagen Aging and Midlife Biobank during 2009-11 (ages 49-63 years). Participants counted 7185 (40.1%). Register data were obtained for every invited person from 7 years before invitation to the end of 2020. The association between education and mortality was estimated using Cox regression models among participants, IPW participants and the source population.

RESULTS

Participants had higher socioeconomic position and fewer hospital contacts before baseline than the source population. Frequency measures of participants approached those of the source population after IPW. Compared with primary/lower secondary education, upper secondary, short tertiary, bachelor and master/doctoral were associated with reduced risk of death among participants (adjusted hazard ratio [95% CI]: 0.60 [0.46; 0.77], 0.68 [0.42; 1.11], 0.37 [0.25; 0.54], 0.28 [0.18; 0.46], respectively). IPW changed the estimates marginally (0.59 [0.45; 0.77], 0.57 [0.34; 0.93], 0.34 [0.23; 0.50], 0.24 [0.15; 0.39]) but not only towards those of the source population (0.57 [0.51; 0.64], 0.43 [0.32; 0.60], 0.38 [0.32; 0.47], 0.22 [0.16; 0.29]).

CONCLUSIONS

Frequency measures of study participants may not reflect the source population in the presence of self-selection, but the impact on association measures can be limited. IPW may be useful for (self-)selection bias correction, but the returned results can still reflect residual or other biases and random errors.

Missing data and missed infections: investigating racial and ethnic disparities in SARS-CoV-2 testing and infection rates in Holyoke, Massachusetts

Routinely collected testing data have been a vital resource for public health response during the COVID-19 pandemic and have revealed the extent to which Black and Hispanic persons have borne a disproportionate burden of SARS-CoV-2 infections and hospitalizations in the United States. However, missing race and ethnicity data and missed infections due to testing disparities limit the interpretation of testing data and obscure the true toll of the pandemic. We investigated potential bias arising from these 2 types of missing data through a case study carried out in Holyoke, Massachusetts, during the prevaccination phase of the pandemic. First, we estimated SARS-CoV-2 testing and case rates by race and ethnicity, imputing missing data using a joint modeling approach. We then investigated disparities in SARS-CoV-2 reported case rates and missed infections by comparing case rate estimates with estimates derived from a COVID-19 seroprevalence survey. Compared with the non-Hispanic White population, we found that the Hispanic population had similar testing rates (476 tested per 1000 vs 480 per 1000) but twice the case rate (8.1% vs 3.7%). We found evidence of inequitable testing, with a higher rate of missed infections in the Hispanic population than in the non-Hispanic White population (79 infections missed per 1000 vs 60 missed per 1000).

Causal Inference Challenges in the Relationship Between Social Determinants and Cardiovascular Outcomes

The effects of social determinants on cardiovascular outcomes are frequently estimated in epidemiologic analyses, but the profound causal and statistical challenges of this research program are not widely discussed. Here, we carefully review definitions and measures for social determinants of cardiovascular health and then examine the various assumptions required for valid causal inference in multivariable analyses of observational data, such as what one would typically encounter in cohorts, population surveys, health care databases, and vital statistics databases. We explain the necessity of the "well-defined exposure" and show how this goal relates to the "consistency assumption" that is necessary for valid causal inference. Well-defined exposure is especially challenging for social determinants of health because they are seldom simple atomistic interventions that are easily conceptualized and measured. We then review threats to valid inference that arise from confounding, selection bias, information bias, and positivity violations. Other causal considerations are reviewed and explained, such as correct model specification, absence of immortal time, and avoidance of the "Table 2 Fallacy," and their application to social determinants of cardiovascular outcomes are discussed. Fruitful approaches, including focusing on policy interventions and the "target trial" frameworks are proposed and provide a pathway for a more efficacious research program that can more reliably improve population health. Valid causal inference in this setting is quite challenging, but-with clever design and thoughtful analysis-the important role of social factors in patterning cardiovascular outcomes can be quantified and reported.

Insomnia symptom prevalence in England: a comparison of cross-sectional self-reported data and primary care records in the UK Biobank

OBJECTIVES

We aimed to use a large dataset to compare self-reported and primary care measures of insomnia symptom prevalence in England and establish whether they identify participants with similar characteristics.

DESIGN

Cross-sectional study with linked electronic health records (EHRs).

SETTING

Primary care in England.

PARTICIPANTS

163 748 UK Biobank participants in England (aged 38-71 at baseline) with linked primary care EHRs.

OUTCOME MEASURES

We compared the percentage of those self-reporting 'usually' having insomnia symptoms at UK Biobank baseline assessment (2006-2010) to those with a Read code for insomnia symptoms in their primary care records prior to baseline. We stratified prevalence in both groups by sociodemographic, lifestyle, sleep and health characteristics.

RESULTS

We found that 29% of the sample self-reported having insomnia symptoms, while only 6% had a Read code for insomnia symptoms in their primary care records. Only 10% of self-reported cases had an insomnia symptom Read code, while 49% of primary care cases self-reported having insomnia symptoms. In both primary care and self-reported data, prevalence of insomnia symptom cases was highest in females, older participants and those with the lowest household incomes. However, while snorers and risk takers were more likely to be a primary care case, they were less likely to self-report insomnia symptoms than non-snorers and non-risk takers.

CONCLUSIONS

Only a small proportion of individuals experiencing insomnia symptoms have an insomnia symptom Read code in their primary care record. However, primary care data do provide a clinically meaningful measure of insomnia prevalence. In addition, the sociodemographic characteristics of people attending primary care with insomnia were consistent with those with self-reported insomnia, thus primary care records are a valuable data source for studying risk factors for insomnia. Further studies should replicate our findings in other populations and examine ways to increase discussions about sleep health in primary care.

Never miss a beep: Using mobile sensing to investigate (non-)compliance in experience sampling studies

Given the increasing number of studies in various disciplines using experience sampling methods, it is important to examine compliance biases because related patterns of missing data could affect the validity of research findings. In the present study, a sample of 592 participants and more than 25,000 observations were used to examine whether participants responded to each specific questionnaire within an experience sampling framework. More than 400 variables from the three categories of person, behavior, and context, collected multi-methodologically via traditional surveys, experience sampling, and mobile sensing, served as predictors. When comparing different linear (logistic and elastic net regression) and non-linear (random forest) machine learning models, we found indication for compliance bias: response behavior was successfully predicted. Follow-up analyses revealed that study-related past behavior, such as previous average experience sampling questionnaire response rate, was most informative for predicting compliance, followed by physical context variables, such as being at home or at work. Based on our findings, we discuss implications for the design of experience sampling studies in applied research and future directions in methodological research addressing experience sampling methodology and missing data.

Contralateral knee pain exacerbation after total knee arthroplasty and its impact on functional activity

INTRODUCTION

The purposes of the present study were to (1) describe the prevalence of contralateral knee pain exacerbation after total knee arthroplasty (TKA), (2) explore the risk factors for pain exacerbation, and (3) verify the association of contralateral knee pain with future functional activity.

MATERIALS AND METHOD

We consecutively recruited outpatients with osteoarthritis of both knees who had primary TKA planned. The contralateral knee pain using a Numerical Rating Scale (NRS) and the functional activities subdomain of the new Knee Society Knee Scoring System (KSS) were assessed preoperatively and at 1, 3, and 6 months postoperatively. Among patients with < 5 NRS points preoperatively, we described the frequency of the contralateral knee pain exacerbation, defined as a ≥ 2-point increase from preoperative pain at each postoperative visit. An exploratory analysis was performed to identify preoperative risk factors for contralateral knee pain exacerbation. A linear mixed model was fit to examine the association of the contralateral knee pain with KSS functional activities at subsequent visits.

RESULTS

Among 315 patients, 14.6%, 24.1%, and 27.6% of patients experienced contralateral knee pain exacerbation at 1, 3, and 6 months postoperatively, respectively. The identified preoperative risk factors were low quadriceps strength and higher Kellgren-Lawrence grade on the non-operative knee, along with severe pain on the operative knee. The magnitude of the association between contralateral knee pain and worsening KSS functional activities increased with subsequent visits (p for interaction < 0.001).

CONCLUSION

The frequency and impact of pain exacerbation on the contralateral knee increase after TKA and should be carefully evaluated for a prolonged period of time.

Placental Abruption and Cardiovascular Event Risk (PACER): Design, data linkage, and preliminary findings

BACKGROUND

Obstetrical complications impact the health of mothers and offspring along the life course, resulting in an increased burden of chronic diseases. One specific complication is abruption, a life-threatening condition with consequences for cardiovascular health that remains poorly studied.

OBJECTIVES

To describe the design and data linkage algorithms for the Placental Abruption and Cardiovascular Event Risk (PACER) cohort.

POPULATION

All subjects who delivered in New Jersey, USA, between 1993 and 2020.

DESIGN

Retrospective, population-based, birth cohort study.

METHODS

We linked the vital records data of foetal deaths and live births to delivery and all subsequent hospitalisations along the life course for birthing persons and newborns. The linkage was based on a probabilistic record-matching algorithm.

PRELIMINARY RESULTS

Over the 28 years of follow-up, we identified 1,877,824 birthing persons with 3,093,241 deliveries (1.1%, n = 33,058 abruption prevalence). The linkage rates for live births-hospitalisations and foetal deaths-hospitalisations were 92.4% (n = 2,842,012) and 70.7% (n = 13,796), respectively, for the maternal cohort. The corresponding linkage rate for the live births-hospitalisations for the offspring cohort was 70.3% (n = 2,160,736). The median (interquartile range) follow-up for the maternal and offspring cohorts was 15.4 (8.1, 22.4) and 14.4 (7.4, 21.0) years, respectively. We will undertake multiple imputations for missing data and develop inverse probability weights to account for selection bias owing to unlinked records.

CONCLUSIONS

Pregnancy offers a unique window to study chronic diseases along the life course and efforts to identify the aetiology of abruption may provide important insights into the causes of future CVD. This project presents an unprecedented opportunity to understand how abruption may predispose women and their offspring to develop CVD complications and chronic conditions later in life.

Selection Bias in Health Research: Quantifying, Eliminating, or Exacerbating Health Disparities?

Purpose of review

To summarize recent literature on selection bias in disparities research addressing either descriptive or causal questions, with examples from dementia research.

Recent findings

Defining a clear estimand, including the target population, is essential to assess whether generalizability bias or collider-stratification bias are threats to inferences. Selection bias in disparities research can result from sampling strategies, differential inclusion pipelines, loss to follow-up, and competing events. If competing events occur, several potentially relevant estimands can be estimated under different assumptions, with different interpretations. The apparent magnitude of a disparity can differ substantially based on the chosen estimand. Both randomized and observational studies may misrepresent health disparities or heterogeneity in treatment effects if they are not based on a known sampling scheme.

Conclusion

Researchers have recently made substantial progress in conceptualization and methods related to selection bias. This progress will improve the relevance of both descriptive and causal health disparities research.

Discriminative machine learning for maximal representative subsampling

Biased population samples pose a prevalent problem in the social sciences. Therefore, we present two novel methods that are based on positive-unlabeled learning to mitigate bias. Both methods leverage auxiliary information from a representative data set and train machine learning classifiers to determine the sample weights. The first method, named maximum representative subsampling (MRS), uses a classifier to iteratively remove instances, by assigning a sample weight of 0, from the biased data set until it aligns with the representative one. The second method is a variant of MRS - Soft-MRS - that iteratively adapts sample weights instead of removing samples completely. To assess the effectiveness of our approach, we induced artificial bias in a public census data set and examined the corrected estimates. We compare the performance of our methods against existing techniques, evaluating the ability of sample weights created with Soft-MRS or MRS to minimize differences and improve downstream classification tasks. Lastly, we demonstrate the applicability of the proposed methods in a real-world study of resilience research, exploring the influence of resilience on voting behavior. Through our work, we address the issue of bias in social science, amongst others, and provide a versatile methodology for bias reduction based on machine learning. Based on our experiments, we recommend to use MRS for downstream classification tasks and Soft-MRS for downstream tasks where the relative bias of the dependent variable is relevant.

Data Missingness Reporting and Use of Methods to Address It in Critical Care Cohort Studies

IMPORTANCE

Failure to recognize and address data missingness in cohort studies may lead to biased results. Although Strengthening the Reporting of Observational Studies in Epidemiology reporting guidelines advocate data missingness reporting, the degree to which missingness is reported and addressed in the critical care literature remains unclear.

OBJECTIVES

To review published ICU cohort studies to characterize data missingness reporting and the use of methods to address it.

DESIGN SETTING AND PARTICIPANTS

We searched the 2022 table of contents of 29 critical care/critical care subspecialty journals having a 2021 impact factor greater than or equal to 3 to identify published prospective clinical or retrospective database cohort studies enrolling greater than or equal to 100 patients.

MAIN OUTCOMES AND MEASURES

In duplicate, two trained researchers conducted a manuscript/supplemental material PDF word search for "missing*" and extracted study type, patient age, ICU type, sample size, missingness reporting, and the use of methods to address it.

RESULTS

A total of 656 studies were reviewed. Of the 334 of 656 (50.9%) studies mentioning missingness, missingness was reported for greater than or equal to 1 variable in 234 (70.1%) and it exceeded 5% for at least one variable in 160 (47.9%). Among the 334 studies mentioning missingness, 88 (26.3%) used exclusion criteria, 36 (10.8%) used complete-case analysis, and 164 (49.1%) used a formal method to avoid missingness. In these 164 studies, imputation only was used in 100 (61.0%), an analytic strategy only in 24 (14.6%), and both in 40 (24.4%). Only missingness greater than 5% (in ≥ 1 variable) was independently associated with greater use of a missingness method (adjusted odds ratio 2.91; 95% CI, 1.85-4.60). Among 140 studies using imputation, multiple imputation was used in 87 studies (62.1%) and simple imputation in 49 studies (35.0%). For the 64 studies using an analytic method, 12 studies (18.8%) assigned missingness as an unknown category, whereas sensitivity analysis was used in 47 studies (73.4%).

CONCLUSIONS AND RELEVANCE

Among published critical care cohort studies, only half mentioned result missingness, one-third reported actual missingness and only one-quarter used a method to manage missingness. Educational strategies to promote missingness reporting and resolution methods are required.

Pre-pubertal accelerometer-assessed physical activity and timing of puberty in British boys and girls: the Millennium Cohort Study

BACKGROUND

Early puberty timing is associated with adverse health outcomes. We aimed to examine prospective associations between objectively measured physical activity and puberty timing in boys and girls.

METHODS

In the UK Millennium Cohort Study, physical activity volume and intensities at 7 years were measured using accelerometers. Status of several pubertal traits and age at menarche were reported at 11, 14 and 17 years. Age at menarche in girls was categorized into tertiles. Other puberty traits were categorized into earlier or later than the median ages calculated from probit models, separately in boys and girls. Multivariable regression models, with adjustment for maternal and child characteristics including body mass index (BMI) at age 7 years as potential confounders, were performed to test the associations of total daily activity counts and fractions of activity counts across intensities (in compositional models) with puberty timing, separately in boys (n = 2531) and girls (n = 3079).

RESULTS

Higher total daily activity counts were associated with lower risks for earlier (vs later) growth spurt, body hair growth, skin changes and menarche in girls, and more weakly with lower risks for earlier skin changes and voice breaking in boys (odds ratios = 0.80-0.87 per 100 000 counts/day). These associations persisted on additional adjustment for BMI at 11 years as a potential mediator. No association with puberty timing was seen for any physical activity intensity (light, moderate or vigorous).

CONCLUSIONS

More physical activity regardless of intensity may contribute to the avoidance of earlier puberty timing, independently of BMI, particularly in girls.

Food additive emulsifiers and risk of cardiovascular disease in the NutriNet-Santé cohort: prospective cohort study

OBJECTIVE

To assess the associations between exposure to food additive emulsifiers and risk of cardiovascular disease (CVD).

DESIGN

Prospective cohort study.

SETTING

French NutriNet-Santé study, 2009-21.

PARTICIPANTS

95 442 adults (>18 years) without prevalent CVD who completed at least three 24 hour dietary records during the first two years of follow-up.

MAIN OUTCOME MEASURES

Associations between intake of food additive emulsifiers (continuous (mg/day)) and risk of CVD, coronary heart disease, and cerebrovascular disease characterised using multivariable proportional hazard Cox models to compute hazard ratios for each additional standard deviation (SD) of emulsifier intake, along with 95% confidence intervals.

RESULTS

Mean age was 43.1 (SD 14.5) years, and 79.0% (n=75 390) of participants were women. During follow-up (median 7.4 years), 1995 incident CVD, 1044 coronary heart disease, and 974 cerebrovascular disease events were diagnosed. Higher intake of celluloses (E460-E468) was found to be positively associated with higher risks of CVD (hazard ratio for an increase of 1 standard deviation 1.05, 95% confidence interval 1.02 to 1.09, P=0.003) and coronary heart disease (1.07, 1.02 to 1.12, P=0.004). Specifically, higher cellulose E460 intake was linked to higher risks of CVD (1.05, 1.01 to 1.09, P=0.007) and coronary heart disease (1.07, 1.02 to 1.12, P=0.005), and higher intake of carboxymethylcellulose (E466) was associated with higher risks of CVD (1.03, 1.01 to 1.05, P=0.004) and coronary heart disease (1.04, 1.02 to 1.06, P=0.001). Additionally, higher intakes of monoglycerides and diglycerides of fatty acids (E471 and E472) were associated with higher risks of all outcomes. Among these emulsifiers, lactic ester of monoglycerides and diglycerides of fatty acids (E472b) was associated with higher risks of CVD (1.06, 1.02 to 1.10, P=0.002) and cerebrovascular disease (1.11, 1.06 to 1.16, P<0.001), and citric acid ester of monoglycerides and diglycerides of fatty acids (E472c) was associated with higher risks of CVD (1.04, 1.02 to 1.07, P=0.004) and coronary heart disease (1.06, 1.03 to 1.09, P<0.001). High intake of trisodium phosphate (E339) was associated with an increased risk of coronary heart disease (1.06, 1.00 to 1.12, P=0.03). Sensitivity analyses showed consistent associations.

CONCLUSION

This study found positive associations between risk of CVD and intake of five individual and two groups of food additive emulsifiers widely used in industrial foods.

TRIAL REGISTRATION

ClinicalTrials.gov NCT03335644.

Demographic and socioeconomic predictors of religious/spiritual beliefs and behaviours in a prospective cohort study (ALSPAC) in Southwest England: Results from the parental generation

Background: We explored associations between possible demographic and socioeconomic causes of religious/spiritual beliefs and behaviours (RSBB) in the parental generation of the Avon Longitudinal Study of Parents and Children (ALSPAC). Methods : We used a prospective birth cohort study (ALSPAC) in Southwest England with 14,157 enrolled mothers and 14,154 associated partners. Three RSBB outcome measures collected during pregnancy were examined: religious belief (belief in God/a divine power; yes/not sure/no), religious affiliation (Christian/none/other) and religious attendance (frequency of attendance at a place of worship). Multiple demographic and socioeconomic exposures were assessed (23 in mothers and 22 in partners). We explored age-adjusted associations between each exposure and outcome using multinomial regression, in addition to exposure-age interactions. Results: Many demographic and socioeconomic factors were associated with RSBB, including age, ethnicity, marital status, education, income and deprivation. Overall, higher socioeconomic position was associated with increased levels of RSBB, particularly regarding religious attendance. For instance, compared to mothers with the lowest level of educational attainment, a degree-level education was associated with a six-fold increase in the relative risk ratio of religious attendance at least once a week, relative to not attending at all (RRR=5.90; 95% CI=[4.44; 7.86]). The magnitude of these associations often varied by outcome, e.g., income was associated with religious attendance, but only weakly with religious affiliation. Although results were demographically and socially patterned, overall effect sizes were relatively small, with a largest pseudo- R 2 value of 2.4%. Patterns of association were similar for mothers and partners. Conclusion: The observed positive association between socioeconomic position and RSBB is contrary to much previous theoretical and empirical work. Potential reasons for these differences are discussed, including cross-cultural variation in religiosity and state support, and differences between RSBB measures. This descriptive paper can also help inform future studies using these data regarding the consideration of appropriate confounders.

Demographic and socioeconomic predictors of religious/spiritual beliefs and behaviours in a prospective cohort study (ALSPAC) in Southwest England: Results from the parental generation

Background: We explored associations between possible demographic and socioeconomic causes of religious/spiritual beliefs and behaviours (RSBB) in the parental generation of the Avon Longitudinal Study of Parents and Children (ALSPAC). Methods : We used a prospective birth cohort study (ALSPAC) in Southwest England with 14,157 enrolled mothers and 14,154 associated partners. Three RSBB outcome measures collected during pregnancy were examined: religious belief (belief in God/a divine power; yes/not sure/no), religious affiliation (Christian/none/other) and religious attendance (frequency of attendance at a place of worship). Multiple demographic and socioeconomic exposures were assessed (23 in mothers and 22 in partners). We explored age-adjusted associations between each exposure and outcome using multinomial regression, in addition to exposure-age interactions. Results: Many demographic and socioeconomic factors were associated with RSBB, including age, ethnicity, marital status, education, income and deprivation. Overall, higher socioeconomic position was associated with increased levels of RSBB, particularly regarding religious attendance. For instance, compared to mothers with the lowest level of educational attainment, a degree-level education was associated with a six-fold increase in the relative risk ratio of religious attendance at least once a week, relative to not attending at all (RRR=5.90; 95% CI=[4.44; 7.86]). The magnitude of these associations often varied by outcome, e.g., income was associated with religious attendance, but only weakly with religious affiliation. Although results were demographically and socially patterned, overall effect sizes were relatively small, with a largest pseudo- R 2 value of 2.4%. Patterns of association were similar for mothers and partners. Conclusion: The observed positive association between socioeconomic position and RSBB is contrary to much previous theoretical and empirical work. Potential reasons for these differences are discussed, including cross-cultural variation in religiosity and state support, and differences between RSBB measures. This descriptive paper can also help inform future studies using these data regarding the consideration of appropriate confounders.

Decreasing Survey Response Rates in the Time of COVID-19: Implications for Analyses of Population Health and Health Inequities

Objectives. To examine whether, and if so how, US national and state survey response rates changed after the onset of the COVID-19 pandemic. Methods. We compared the change in response rates between 2020 and 2019 of 6 (3 social and economic, 3 health focused) major US national surveys (2 with state response rates). Results. All the ongoing surveys except 1 reported relative decreases (∼29%) in response rates. For example, the household response rate to the US Census American Community Survey decreased from 86.0% in 2019 to 71.2% in 2020, and the response rate of the US National Health Interview Survey decreased from 60.0% to 42.7% from the first to the second quarter of 2020. For all surveys, the greatest decreases in response rates occurred among persons with lower income and lower education. Conclusions. Socially patterned decreases in response rates pose serious challenges and must be addressed explicitly in all studies relying on data obtained since the onset of the pandemic. Public Health Implications. Artifactual reduction of estimates of the magnitude of health inequities attributable to differential response rates could adversely affect efforts to reduce these inequities. (Am J Public Health. 2023;113(6):667-670. https://doi.org/10.2105/AJPH.2023.307267).

The M-Value: A Simple Sensitivity Analysis for Bias Due to Missing Data in Treatment Effect Estimates

Complete-case analyses can be biased if missing data are not missing completely at random. We propose simple sensitivity analyses that apply to complete-case estimates of treatment effects; these analyses use only simple summary data and obviate specifying the precise mechanism of missingness and making distributional assumptions. Bias arises when treatment effects differ between retained and nonretained participants or, among retained participants, the estimate is biased because conditioning on retention has induced a noncausal path between the treatment and outcome. We thus bound the overall treatment effect on the difference scale by specifying: 1) the unobserved treatment effect among nonretained participants; and 2) the strengths of association that unobserved variables have with the exposure and with the outcome among retained participants ("induced confounding associations"). Working with the former sensitivity parameter subsumes certain existing methods of worst-case imputation while also accommodating less-conservative assumptions (e.g., that the treatment is not detrimental on average even among nonretained participants). As an analog to the E-value for confounding, we propose the M-value, which represents, for a specified treatment effect among nonretained participants, the strength of induced confounding associations required to reduce the treatment effect to the null or to any other value. These methods could help characterize the robustness of complete-case analyses to potential bias due to missing data.

Exploring the impact of selection bias in observational studies of COVID-19: a simulation study

BACKGROUND

Non-random selection of analytic subsamples could introduce selection bias in observational studies. We explored the potential presence and impact of selection in studies of SARS-CoV-2 infection and COVID-19 prognosis.

METHODS

We tested the association of a broad range of characteristics with selection into COVID-19 analytic subsamples in the Avon Longitudinal Study of Parents and Children (ALSPAC) and UK Biobank (UKB). We then conducted empirical analyses and simulations to explore the potential presence, direction and magnitude of bias due to this selection (relative to our defined UK-based adult target populations) when estimating the association of body mass index (BMI) with SARS-CoV-2 infection and death-with-COVID-19.

RESULTS

In both cohorts, a broad range of characteristics was related to selection, sometimes in opposite directions (e.g. more-educated people were more likely to have data on SARS-CoV-2 infection in ALSPAC, but less likely in UKB). Higher BMI was associated with higher odds of SARS-CoV-2 infection and death-with-COVID-19. We found non-negligible bias in many simulated scenarios.

CONCLUSIONS

Analyses using COVID-19 self-reported or national registry data may be biased due to selection. The magnitude and direction of this bias depend on the outcome definition, the true effect of the risk factor and the assumed selection mechanism; these are likely to differ between studies with different target populations. Bias due to sample selection is a key concern in COVID-19 research based on national registry data, especially as countries end free mass testing. The framework we have used can be applied by other researchers assessing the extent to which their results may be biased for their research question of interest.

Toward a Clearer Definition of Selection Bias When Estimating Causal Effects

Selection bias remains a subject of controversy. Existing definitions of selection bias are ambiguous. To improve communication and the conduct of epidemiologic research focused on estimating causal effects, we propose to unify the various existing definitions of selection bias in the literature by considering any bias away from the true causal effect in the referent population (the population before the selection process), due to selecting the sample from the referent population, as selection bias. Given this unified definition, selection bias can be further categorized into two broad types: type 1 selection bias owing to restricting to one or more level(s) of a collider (or a descendant of a collider) and type 2 selection bias owing to restricting to one or more level(s) of an effect measure modifier. To aid in explaining these two types-which can co-occur-we start by reviewing the concepts of the target population, the study sample, and the analytic sample. Then, we illustrate both types of selection bias using causal diagrams. In addition, we explore the differences between these two types of selection bias, and describe methods to minimize selection bias. Finally, we use an example of "M-bias" to demonstrate the advantage of classifying selection bias into these two types.

Causal Modeling to Mitigate Selection Bias and Unmeasured Confounding in Internet-Based Epidemiology of COVID-19: Model Development and Validation

BACKGROUND

Selection bias and unmeasured confounding are fundamental problems in epidemiology that threaten study internal and external validity. These phenomena are particularly dangerous in internet-based public health surveillance, where traditional mitigation and adjustment methods are inapplicable, unavailable, or out of date. Recent theoretical advances in causal modeling can mitigate these threats, but these innovations have not been widely deployed in the epidemiological community.

OBJECTIVE

The purpose of our paper is to demonstrate the practical utility of causal modeling to both detect unmeasured confounding and selection bias and guide model selection to minimize bias. We implemented this approach in an applied epidemiological study of the COVID-19 cumulative infection rate in the New York City (NYC) spring 2020 epidemic.

METHODS

We collected primary data from Qualtrics surveys of Amazon Mechanical Turk (MTurk) crowd workers residing in New Jersey and New York State across 2 sampling periods: April 11-14 and May 8-11, 2020. The surveys queried the subjects on household health status and demographic characteristics. We constructed a set of possible causal models of household infection and survey selection mechanisms and ranked them by compatibility with the collected survey data. The most compatible causal model was then used to estimate the cumulative infection rate in each survey period.

RESULTS

There were 527 and 513 responses collected for the 2 periods, respectively. Response demographics were highly skewed toward a younger age in both survey periods. Despite the extremely strong relationship between age and COVID-19 symptoms, we recovered minimally biased estimates of the cumulative infection rate using only primary data and the most compatible causal model, with a relative bias of +3.8% and -1.9% from the reported cumulative infection rate for the first and second survey periods, respectively.

CONCLUSIONS

We successfully recovered accurate estimates of the cumulative infection rate from an internet-based crowdsourced sample despite considerable selection bias and unmeasured confounding in the primary data. This implementation demonstrates how simple applications of structural causal modeling can be effectively used to determine falsifiable model conditions, detect selection bias and confounding factors, and minimize estimate bias through model selection in a novel epidemiological context. As the disease and social dynamics of COVID-19 continue to evolve, public health surveillance protocols must continue to adapt; the emergence of Omicron variants and shift to at-home testing as recent challenges. Rigorous and transparent methods to develop, deploy, and diagnosis adapted surveillance protocols will be critical to their success.

Religious/spiritual beliefs and behaviours and study participation in a prospective cohort study (ALSPAC) in Southwest England

Background Longitudinal studies are key to understanding risk factors for health, well-being, and disease, yet associations may be biased if study invitation and participation are non-random. Religious/spiritual beliefs and behaviours (RSBB) are increasingly recognised as having potentially important relationships with health. However, it is unclear whether RSBB is associated with study participation. We examine whether RSBB is associated with participation in the longitudinal birth cohort ALSPAC (Avon Longitudinal Study of Parents and Children). Methods Three RSBB factors were used: religious belief (belief in God/a divine power; yes/not sure/no), religious affiliation (Christian/none/other), and religious attendance (frequency of attendance at a place of worship). Participation was measured in three ways: i) total number of questionnaires/clinics completed; ii) completion of the most recent questionnaire (in 2020); and iii) length of participation. Analyses were repeated for the ALSPAC mothers, their partners, and the study children, and were adjusted for relevant socio-demographic confounders. Results Religious attendance was positively associated with participation in all adjusted models in all three cohorts. For example, study mothers who attended a place of worship at least once a month on average completed two more questionnaires (out of a possible 50), had 50% greater odds of having completed the most recent questionnaire, and had 25% reduced risk of drop-out, relative to those who did not attend a place of worship. In the adjusted analyses, religious belief and attendance were not associated with participation. However, the majority of unadjusted models showed associations between RSBB and participation. Conclusion After adjusting for confounders, religious attendance –  not religious belief or affiliation – was associated with participation in ALSPAC. These results indicate that use of RSBB variables (and religious attendance in particular) may result in selection bias and spurious associations; these potential biases should be explored and discussed in future studies using these data.

On selection bias in comparison measures of smartphone-generated population mobility: an illustration of no-bias conditions with a commercial data source

PURPOSE

Passively generated cell-phone location ("mobility") data originally intended for commercial use has become frequently used in epidemiologic research, notably during the COVID-19 pandemic to study the impact of physical-distancing recommendations on aggregate population behavior (e.g., average daily mobility). Given the opaque nature of how individuals are selected into these datasets, researchers have cautioned that their use may give rise to selection bias, yet little guidance exists for assessing this potential threat to validity in mobility-data research. Through an example analysis of cell-phone-derived mobility data, we present a set of conditions to guide the assessment of selection bias in measures comparing aggregate mobility patterns over time and between groups.

METHODS

We specifically consider bias in measures comparing group-level mobility in the same group (difference, ratio, percent difference) and between groups (difference in differences, ratio of ratios, ratio of percent differences). We illustrate no-bias conditions in these measures through an example comparing block-group-level mobility between income groups in United States metro areas before (January 1st-March 10, 2020) and after (March 11th-April 19th, 2020) the day COVID-19 was declared a pandemic.

RESULTS

Within-group contrasts describing mobility over time, especially for the higher-income decile, were expected to be most resistant to bias during the example study period.

CONCLUSIONS

The presented conditions can be used to assess the susceptibility to selection bias of group-level measures comparing mobility. Importantly, they can be used even without knowledge of the degree of bias in each group at each time point. We further highlight links between no-bias principles originating in epidemiology and economics, showing that certain assumptions (e.g., parallel trends) can apply to biases beyond their original application.

The Avon Longitudinal Study of Parents and Children - A resource for COVID-19 research: Home-based antibody testing results, October 2020. An emphasis on self-screening at a population level

The Avon Longitudinal Study of Parents and Children (ALSPAC) is a prospective population-based cohort study which recruited pregnant women in 1990-1992 and has followed these women, their partners (Generation 0; G0) and offspring (Generation 1; G1) ever since. The study reacted rapidly to the COVID-19 pandemic, deploying online questionnaires in March and May 2020. Home-based antibody tests and a further questionnaire were sent to 5220 participants during a two-week period of October 2020.  4.2% (n=201) of participants reported a positive antibody test (3.2% G0s [n=81]; 5.6% G1s [n=120]). 43 reported an invalid test, 7 did not complete and 3 did not report their result. Participants uploaded a photo of their test to enable validation: all positive tests, those where the participant could not interpret the result and a 5% random sample were manually checked against photos. We report 92% agreement (kappa=0.853). Positive tests were compared to additional COVID-19 status information: 58 (1.2%) participants reported a previous positive test, 73 (1.5%) reported that COVID-19 was suspected by a doctor, but not tested and 980 (20.4%) believed they had COVID-19 due to their own suspicions.  Of those reporting a positive result on our antibody test, 55 reported that they did not think they had had COVID-19. Results from antibody testing and questionnaire data will be complemented by health record linkage and results of other biological testing- uniting Pillar testing data with home testing and self-report. Data have been released as an update to the original datasets released in July 2020. It comprises: 1) a standard dataset containing all participant responses to all three questionnaires with key sociodemographic factors and 2) as individual participant-specific release files enabling bespoke research across all areas supported by the study. This data note describes the antibody testing, associated questionnaire and the data obtained from it.

The Avon Longitudinal Study of Parents and Children - A resource for COVID-19 research: Home-based antibody testing results, October 2020

The Avon Longitudinal Study of Parents and Children (ALSPAC) is a prospective population-based cohort study which recruited pregnant women in 1990-1992 and has followed these women, their partners (Generation 0; G0) and offspring (Generation 1; G1) ever since. The study reacted rapidly to the COVID-19 pandemic, deploying online questionnaires in March and May 2020. Home-based antibody tests and a further questionnaire were sent to 5220 participants during a two-week period of October 2020. 4.2% (n=201) of participants reported a positive antibody test (3.2% G0s [n=81]; 5.6% G1s [n=120]). 43 reported an invalid test, 7 did not complete and 3 did not report their result. Participants uploaded a photo of their test to enable validation: all positive tests, those where the participant could not interpret the result and a 5% random sample were manually checked against photos. We report 92% agreement (kappa=0.853). Positive tests were compared to additional COVID-19 status information: 58 (1.2%) participants reported a previous positive test, 73 (1.5%) reported that COVID-19 was suspected by a doctor, but not tested and 980 (20.4%) believed they had COVID-19 due to their own suspicions.  Of those reporting a positive result on our antibody test, 55 reported that they did not think they had had COVID-19. Results from antibody testing and questionnaire data will be complemented by health record linkage and results of other biological testing- uniting Pillar testing data with home testing and self-report. Data have been released as an update to the original datasets released in July 2020. It comprises: 1) a standard dataset containing all participant responses to all three questionnaires with key sociodemographic factors and 2) as individual participant-specific release files enabling bespoke research across all areas supported by the study. This data note describes the antibody testing, associated questionnaire and the data obtained from it.

Emulating a Randomised Controlled Trial With Observational Data: An Introduction to the Target Trial Framework

Randomised controlled trials (RCTs) are often considered to be the highest quality of evidence owing to the absence of baseline confounding, the simplicity of analyses, and direct estimation of causal effects. However, observational studies can be designed to mimic RCTs and estimate causal treatment effects. In this review, we describe the target trial framework to illustrate how observational studies can successfully emulate RCTs. We focus on key design elements of RCTs and how to emulate them with observational data. These elements include 1) eligibility criteria, 2) treatment assignment and randomisation, 3) specification of "time zero", 4) outcomes, 5) follow-up, 6) causal contrasts (intention-to-treat vs per-protocol), and 7) statistical analyses. In addition, we describe the design of an example target trial created to emulate the Trial to Assess Improvement in Therapeutic Outcomes by Optimizing Platelet Inhibition With Prasugrel Thrombolysis in Myocardial Infarction (TRITON-TIMI) 38 trial and compare effect estimates. Overall, careful design of a target trial using observational data can produce causal effect estimates that are often comparable to RCTs.