Constructing Causal Life-Course Models: Comparative Study of Data-Driven and Theory-Driven Approaches.

Life-course epidemiology relies on specifying complex (causal) models that describe how variables interplay over time. Traditionally, such models have been constructed by perusing existing theory and previous studies. By comparing data-driven and theory-driven models, we investigated whether data-driven causal discovery algorithms can help in this process. We focused on a longitudinal data set on a cohort of Danish men (the Metropolit Study, 1953-2017). The theory-driven models were constructed by 2 subject-field experts. The data-driven models were constructed by use of the temporal Peter-Clark (TPC) algorithm. The TPC algorithm utilizes the temporal information embedded in life-course data. We found that the data-driven models recovered some, but not all, causal relationships included in the theory-driven expert models. The data-driven method was especially good at identifying direct causal relationships that the experts had high confidence in. Moreover, in a post hoc assessment, we found that most of the direct causal relationships proposed by the data-driven model but not included in the theory-driven model were plausible. Thus, the data-driven model may propose additional meaningful causal hypotheses that are new or have been overlooked by the experts. In conclusion, data-driven methods can aid causal model construction in life-course epidemiology, and combining both data-driven and theory-driven methods can lead to even stronger models.

What Is the Causal Interpretation of Sibling Comparison Designs?

Sibling comparison designs have long been used to assess causal effects of exposures for which randomized studies are impossible and measurement of all relevant confounding is unobtainable. The idea is to utilize the fact that siblings often share a lot of unobserved variables. Therefore, it is proposed that in certain cases, comparing siblings is equivalent to comparing exchangeable individuals, which is the foundation for causal inference based on randomized controlled trials (RCTs). However, this intuition-and the publication of highly important sibling studies-vastly predate modern causal inference theory. Full causal descriptions of sibling comparison designs are essentially nonexistent, and therefore it is not clear exactly how or if we can interpret their estimated effects as causal. We fill this theoretical gap by proposing a counterfactual-based framework for sibling comparison designs. Moreover, we employ this framework to derive precise causal interpretations for three commonly used sibling model estimators stemming from fixed-effects ordinary least squares (OLS), conditional logistic regression, and stratified Cox regression. We establish that, for the latter two, the obtained effect parameter describes a causal effect on the full sibling group, not the individuals, and thus it does not correspond to the prevailing intuition from the RCT analogue. For fixed-effects OLS estimation, the parameter describes a causal effect on an individual, but may depend on an intervention on the whole sibling group. OLS estimation thus results in an estimator that can be given a simple causal interpretation that is similar, but not equal to, the RCT parallel. See video abstract at, http://links.lww.com/EDE/B618.

Relation Between Infant Microbiota and Autism?: Results from a National Cohort Sibling Design Study.

BACKGROUND: Hypotheses concerning adverse effects of changes in microbiota have received much recent attention, but unobserved confounding makes them difficult to test. We investigated whether surrogate markers for potential adverse microbiota change in infancy affected autism risk, addressing unobserved confounding using a sibling study design. METHODS: This is a population-based, prospective cohort study including all singleton live births in Denmark from 1997 to 2010. The exposure variables were cesarean delivery and antibiotic use in the first 2 years of life. The outcome was a subsequent autism diagnosis. We used the between- and within-sibling model and compared it with sibling-stratified Cox models and simpler standard Cox models that ignored sibship. RESULTS: Of our study population including 671,606 children, who were followed for up to 15 years (7,341,133 person-years), 72% received antibiotics, 17.5% were delivered by cesarean, and 1.2% (8,267) developed autism. The standard Cox models predicted that both cesarean (compared with vaginal) delivery and antibiotics increased the risk of autism. In the sibling-stratified Cox model, only broader spectrum antibiotics were associated with increased risk of autism: hazard ratio (HR) = 1.16 (95% confidence interval = 1.01, 1.36). The between-within model estimated no exposure effects: intrapartum cesarean HR = 1.06 (0.89, 1.26); prelabor cesarean HR = 0.97 (0.83, 1.15); exclusively penicillin HR = 1.05 (0.93, 1.18); and broader spectrum antibiotics HR = 1.05 (0.95, 1.16). CONCLUSIONS: The between-within model rendered more precise estimates than sibling-stratified Cox models, and we believe that it also provided more valid estimates. Results from these preferred models do not support a causal relation between antibiotic treatment during infancy, cesarean delivery, and autism. See video abstract at, http://links.lww.com/EDE/B432.

Investigating the effects of cesarean delivery and antibiotic use in early childhood on risk of later attention deficit hyperactivity disorder.

BACKGROUND: Increasing attention deficit hyperactivity disorder (ADHD) incidence has been proposed to be caused by factors influencing microbiota in early life. We investigated the potential causality between ADHD and two surrogate markers for changes in children's microbiota: birth delivery mode and early childhood antibiotic use. METHOD: This population-based, prospective cohort study linked nationwide registers of data for native Danish singleton live births in Denmark from 1997 to 2010. Exposure variables were delivery mode and antibiotic use during the first 2 years of life. The main outcome measure was ADHD diagnosis or redeemed ADHD medication prescriptions. For statistical analysis, we used both advanced sibling models and a more traditional approach. RESULTS: We included 671,592 children, followed from their second birthday in the period 1999-2014 for 7,300,522 person-years. ADHD was diagnosed in 17,971. In total, 17.5% were born by cesarean delivery, and 72% received antibiotic treatment within their first 2 years of life. In the adjusted between-within sibling survival model, mode of delivery or antibiotics had no effect on ADHD when compared with vaginal delivery or no antibiotic treatment as hazard ratios were 1.09 (95% confidence interval 0.97-1.24) for intrapartum cesarean, 1.03 (0.91-1.16) for prelabor cesarean, 0.98 (0.90-1.07) for penicillin, and 0.99 (0.92-1.06) for broader spectrum antibiotics. In a sibling-stratified Cox regression, intrapartum cesarean was associated with increased ADHD risk, but other exposures were not. In a descriptive, nonstratified Cox model, we found increased risk for ADHD for all exposures. CONCLUSIONS: Detailed family confounder control using the superior between-within model indicates that cesarean delivery or use of antibiotics during the first 2 years of life does not increase ADHD risk. Therefore, our study suggests that changes in children's microbiota related to cesarean delivery or antibiotic use, do not cause ADHD.

Exploratory data structure comparisons: three new visual tools based on principal component analysis.

Datasets are sometimes divided into distinct subsets, e.g. due to multi-center sampling, or to variations in instruments, questionnaire item ordering or mode of administration, and the data analyst then needs to assess whether a joint analysis is meaningful. The Principal Component Analysis-based Data Structure Comparisons (PCADSC) tools are three new non-parametric, visual diagnostic tools for investigating differences in structure for two subsets of a dataset through covariance matrix comparisons by use of principal component analysis. The PCADCS tools are demonstrated in a data example using European Social Survey data on psychological well-being in three countries, Denmark, Sweden, and Bulgaria. The data structures are found to be different in Denmark and Bulgaria, and thus a comparison of for example mean psychological well-being scores is not meaningful. However, when comparing Denmark and Sweden, very similar data structures, and thus comparable concepts of well-being, are found. Therefore, inter-country comparisons are warranted for these countries.

Is cesarean section a cause of affective disorders?-A national cohort study using sibling designs.

BACKGROUND: The gut microbiota of children delivered by cesarean section differs from that of children delivered vaginally. In light of the gut-brain axis hypothesis, cesarean section may influence risk of affective disorders. METHODS: Population based prospective cohort study included Danish children born 1982 through 2001, with follow-up until 2015. The effect of delivery mode on the risk of affective disorders was assessed using a standard Cox model and two types of Cox sibling models. Diagnostic codes or prescriptions for antidepressants and lithium were used to define cases of affective disorders. RESULTS: 1,009,444 children were followed for 8,880,794 person-years from the age of 13 years, with relevant covariates available from birth. There are strong calendar time trends in the occurrence of affective disorders with an increasingly younger age at first diagnosis and with a hotspot between the years 2007-2012. Fully adjusted standard Cox models showed an increased risk of affective disorders for both pre-labor (hazard ratio [HR], 1.11; 95% confidence interval [CI], 1.08-1.15) and intrapartum (HR, 1.07; 95% CI, 1.05-1.10) cesarean section, compared to vaginal delivery. This effect disappeared in the between-within sibling model for pre-labor (HR, 1.00; 95% CI, 0.94-1.07) but not intrapartum (HR, 1.05; 95% CI, 1.00-1.12) cesarean section. LIMITATIONS: Interpretation of results from sibling models may not be relevant to children without siblings. CONCLUSIONS: These results do not support the hypothesis that a delivery-mode dependent change in gut microbiota is a cause of subsequent affective disorders, despite an apparent association with delivery mode.

Evaluation of adding the community reinforcement approach to motivational enhancement therapy for adults aged 60 years and older with DSM-5 alcohol use disorder: a randomized controlled trial.

AIM: To examine whether adding the Community Reinforcement Approach for Seniors (CRA-S) to Motivational Enhancement Therapy (MET) increases the probability of treatment success in people aged ≥ 60 years with alcohol use disorder (AUD). DESIGN: A single blind multi-centre multi-national randomized (1 : 1) controlled trial. SETTING: Out-patient settings (municipal alcohol treatment clinics in Denmark, specialized addiction care facilities in Germany and a primary care clinic in the United States). PARTICIPANTS: Between January 2014 and May 2016, 693 patients aged 60+ years and fulfilling DSM-5 criteria for AUD participated in comparing MET (n = 351) and MET + CRA-S (n = 342). INTERVENTION AND COMPARATOR: MET (comparator) included four manualized sessions aimed at increasing motivation to change and establishing a change plan. CRA-S (intervention) consisted of up to eight further optional, manualized sessions aimed at helping patients to implement their change plan. CRA-S included a specially designed module on coping with age and age-related problems. MEASUREMENTS: The primary outcome was either total alcohol abstinence or an expected blood alcohol concentration of ≤ 0.05% during the 30 days preceding the 26 weeks follow-up (defined as success) or blood alcohol concentration of > 0.05% during the follow-up period (defined as failure). This was assessed by self-report using the Form 90 instrument. The main analysis involved complete cases. FINDINGS: The follow-up rate at 26 weeks was 76.2% (76.9% in the MET group and 76.0% in the MET + CRA-S group). The success rate in the MET group was 48.9% [95% confidence interval (CI) = 42.9-54.9%] versus 52.3% (95% CI = 46.2-58.3%) in the MET + CRA-S group. The odds of success in the two conditions did not differ (odds ratio = 1.22. 95% CI = 0.86-1.75, P = 0.26, Bayes factor = 0.10). Sensitivity analyses involving alternative approaches to missing values did not change the results. CONCLUSIONS: In older adults with an alcohol use disorder diagnosis, adding the 'community reinforcement approach for seniors' intervention to brief out-patient motivational enhancement therapy treatment did not improve drinking outcome.