Single World Intervention Graphs (SWIGs): A Practical Guide.

This article offers a comprehensive and user-friendly guide to visualizing causal theories using Single World Intervention Graphs (SWIGs). We begin with a discussion of the potential outcomes approach to causality and limitations of using Directed Acyclic Graphs (DAGs) under this framework. We then introduce SWIGs as a simple but powerful tool for integrating potential outcomes explicitly into causal diagrams. The article provides a step-by-step guide on transforming DAGs into SWIGs that includes practical insights into constructing SWIGs under various scenarios such as confounding, mediation, and sequential randomization. Highlighting the utility of SWIGs in practice, we illustrate their application in identifying the g-formula, showcasing their capacity to make causal estimands visually explicit. This article serves as a resource for epidemiologists and researchers interested in expanding their causal inference toolkit.

Prediction of Severe Baseline Asymptomatic Carotid Stenosis and Subsequent Risk of Stroke and Cardiovascular Disease.

BACKGROUND: Risk models to identify patients at high risk of asymptomatic carotid artery stenosis (ACAS) can help in selecting patients for screening, but long-term outcomes in these patients are unknown. We assessed the diagnostic and prognostic value of the previously published Prevalence of ACAS (PACAS) risk model to detect ACAS at baseline and to predict subsequent risk of stroke and cardiovascular disease (CVD) during follow-up. METHODS: We validated the discrimination and calibration of the PACAS risk model to detect severe (≥70% narrowing) ACAS with patients from the Reduction of Atherothrombosis for Continued Health registry. We subsequently calculated the incidence rates of stroke and CVD (fatal and nonfatal stroke or myocardial infarction or vascular death) during follow-up in 4 risk groups (low, medium, high, and very high, corresponding to sum scores of ≤9, 10-13, 14-17, and ≥18, respectively). RESULTS: Among 26 384 patients, aged between 45 and 80 years, without prior carotid procedures, 1662 (6.3%) had severe baseline ACAS. During ≈70 000 patient-years of follow-up, 1124 strokes and 2484 CVD events occurred. Discrimination of the PACAS model was 0.67 (95% CI, 0.65-0.68), and calibration showed adequate concordance between predicted and observed risks of severe baseline ACAS after recalibration. Significantly higher incidence rates of stroke (Ptrend<0.011) and CVD (Ptrend<0.0001) during follow-up were found with increasing PACAS risk groups. Among patients with high PACAS sum score of ≥14 (corresponding to 27.7% of all patients), severe baseline ACAS prevalence was 11.4%. In addition, 56.6% of incident strokes and 64.9% of incident CVD events occurred in this group. CONCLUSIONS: The PACAS risk model can reliably identify patients at high risk of severe baseline ACAS. Incidence rates of stroke and CVD during follow-up were significantly higher in patients with high PACAS sum scores. Selective screening of patients with high PACAS sum scores may help to prevent future stroke or CVD.

Sample size calculation in clinical research.

Calculation of sample size is an essential part of research study design since it affects the reliability and feasibility of the research study. In this article, we look at the principles of sample size calculation for different types of research studies.

Methodological Considerations in Evaluating Breast Cancer Screening Studies.

In evidence-based medicine frameworks, the highest level of evidence is derived from quantitative synthesis of double-masked, high-quality, randomly assigned controlled trials. Meta-analyses of randomly assigned controlled trials have demonstrated that screening mammography reduces breast cancer deaths. In the United States, every major guideline-producing organization has recommended screening mammography in average-risk women; however, there are controversies about age and frequency. Carefully controlled observational research studies and statistical modeling studies can address evidence gaps and inform evidence-based, contemporary screening practices. As breast imaging radiologists develop and evaluate existing and new screening tests and technologies, they will need to understand the key methodological considerations and scientific criteria used by policy makers and health service researchers to support dissemination and implementation of evidence-based screening tests. The Wilson and Jungner principles and the U.S. Preventive Services Task Force general analytic framework provide structured evaluations of the effectiveness of screening tests. Key considerations in both frameworks include public health significance, natural history of disease, cost-effectiveness, and characteristics of screening tests and treatments. Rigorous evaluation of screening tests using analytic frameworks can maximize the benefits of screening tests while reducing potential harms. The purpose of this article is to review key methodological considerations and analytic frameworks used to evaluate screening studies and develop evidence-based recommendations.

Deep Learning - Methods to Amplify Epidemiological Data Collection and Analyses.

Deep learning is a subfield of artificial intelligence and machine learning based mostly on neural networks and often combined with attention algorithms that has been used to detect and identify objects in text, audio, images, and video. Serghiou and Rough (Am J Epidemiol. 0000;000(00):0000-0000) present a primer for epidemiologists on deep learning models. These models provide substantial opportunities for epidemiologists to expand and amplify their research in both data collection and analyses by increasing the geographic reach of studies, including more research subjects, and working with large or high dimensional data. The tools for implementing deep learning methods are not quite yet as straightforward or ubiquitous for epidemiologists as traditional regression methods found in standard statistical software, but there are exciting opportunities for interdisciplinary collaboration with deep learning experts, just as epidemiologists have with statisticians, healthcare providers, urban planners, and other professionals. Despite the novelty of these methods, epidemiological principles of assessing bias, study design, interpretation and others still apply when implementing deep learning methods or assessing the findings of studies that have used them.

Identifying Pediatric Hypertension in Observational Data: Comparing Clinical and Claims Cohorts in Real World Data.

We conducted retrospective public health surveillance using data from 2006 to 2016 in seven integrated delivery systems from FDA's Sentinel System. We identified pediatric hypertensive patients by clinical and claims-based definitions and compared demographics, baseline profiles and follow-up time profiles. Among 3,757,803 pediatric patients aged 3 to 17 years, we identified 781,722 children and 551,246 teens with at least three blood pressure measures over 36-months. Of these, 70,315 children (9%) and 47,928 teens (8.7%) met the clinical definition for hypertension and 22,465 (2.8%) children and 60,952 (11%) of teens met the clinical definition for elevated, non-hypertensive blood pressure. Of the 3.7M patients, we identified 3,246 children and 7,293 teens with any claim for hypertension (claims definition). Evidence of hypertension claims among those meeting our clinical definition was poor; 2.2% and 7.3% of clinically hypertensive children and teens had corresponding claims for hypertension. Baseline profiles for claims-based hypertensive patients suggest greater severity of disease compared to clinical patients. Claims-based patients showed higher rates of all-cause mortality during follow-up. Pediatric hypertension in claims-based data sources is under-captured but may serve as a marker for greater disease severity. Investigators should understand coding practices when selecting real-world data sources for future pediatric hypertension work.

Invited Commentary: Raising a High-Pressure Alarm About Pediatric Hypertension.

Hypertension is a common "silent killer" in adult medicine, but epidemiologic estimates of elevated blood pressure in children and adolescents are challenged by under-diagnosis and resultant low utilization of relevant administrative or billing codes. In the article by Horgan et al (Am J Epidemiol 2024), children and adolescents with hypertension and elevated blood pressure were identified using direct assessment of blood pressure measurements available in the electronic health record from both inpatient and outpatient visits ("clinical cohort") in comparison to diagnosis codes ("claims-based cohort"). The study population included 3.75 million pediatric healthcare visits available in the US Food and Drug Administration's Sentinel System. While the study applied a relatively novel methodology to interrogate available clinical data within the EHR to better understand the prevalence of pediatric hypertension and raised concern for a higher occurrence of hypertension among children and adolescents than previously realized using claims codes, the utility of the prevalence estimates may be limited by the potential for misclassification bias inherent in EHR data. However, these data raise important concerns about relaying solely on ICD-9-CM/ICD-10-CM codes to quantify the epidemiology of pediatric hypertension and highlight opportunities to address elevated blood pressure in children that could improve long-term cardiovascular health.

The role and limitations of electronic medical records versus patient interviews for determining symptoms of, underlying comorbidities of, and medication use by patients with COVID-19.

Electronic medical records (EMRs) are important for rapidly compiling information to determine disease characteristics (eg, symptoms) and risk factors (eg, underlying comorbidities, medications) for disease-related outcomes. To assess EMR data accuracy, agreement between EMR abstractions and patient interviews was evaluated. Symptoms, medical history, and medication use among patients with COVID-19 collected from EMRs and patient interviews were compared using overall agreement (ie, same answer in EMR and interview), reported agreement (yes answer in both EMR and interview among those who reported yes in either), and κ statistics. Overall, patients reported more symptoms in interviews than in EMR abstractions. Overall agreement was high (≥50% for 20 of 23 symptoms), but only subjective fever and dyspnea had reported agreement of ≥50%. The κ statistics for symptoms were generally low. Reported medical conditions had greater agreement with all condition categories (n = 10 of 10) having ≥50% overall agreement and half (n = 5 of 10) having ≥50% reported agreement. More nonprescription medications were reported in interviews than in EMR abstractions, leading to low reported agreement (28%). Discordance was observed for symptoms, medical history, and medication use between EMR abstractions and patient interviews. Investigations using EMRs to describe clinical characteristics and identify risk factors should consider the potential for incomplete data, particularly for symptoms and medications.

Influence of incomplete death information on cumulative risk estimates in US claims data.

Administrative claims databases often do not capture date or fact of death, so studies using these data may inappropriately treat death as a censoring event-equivalent to other withdrawal reasons-rather than a competing event. We examined 1-, 3-, and 5-year inverse-probability-of-treatment weighted cumulative risks of a composite cardiovascular outcome among 34 527 initiators of telmisartan (exposure) and ramipril (referent), who were aged ≥55 years, in Optum (United States) claims data from 2003 to 2020. Differences in cumulative risks of the cardiovascular endpoint due to censoring of death (cause-specific), as compared with treating death as a competing event (subdistribution), increased with greater follow-up time and older age, where event and mortality risks were higher. Among ramipril users, 5-year cause-specific and subdistribution cumulative risk estimates per 100, respectively, were 16.4 (95% CI, 15.3-17.5) and 16.2 (95% CI, 15.1-17.3) among ages 55-64 (difference = 0.2) and were 43.2 (95% CI, 41.3-45.2) and 39.7 (95% CI, 37.9-41.4) among ages ≥75 (difference = 3.6). Plasmode simulation results demonstrated the differences in cause-specific versus subdistribution cumulative risks to increase with increasing mortality rate. We suggest researchers consider the cohort's baseline mortality risk when deciding whether real-world data with incomplete death information can be used without concern. This article is part of a Special Collection on Pharmacoepidemiology.

A call for sex-positive epidemiology.

The World Health Organization specifies that sexual health requires the potential for pleasurable and safe sexual experiences. Yet epidemiologic research into sexual pleasure and other positive sexual outcomes has been scant. In this commentary, we aim to support the development and adoption of sex-positive epidemiology, which we define as epidemiology that incorporates the study of pleasure and other positive features alongside sexually transmitted infections and other familiar negative outcomes. We first call epidemiologists' attention to the potential role that stigma plays in the suppression of sex-positive research. We further describe existing measures of sex-positive constructs that may be useful in epidemiologic research. Finally, the study of sex-positive constructs is vulnerable to biases that are well-known to epidemiologists, especially selection bias, information bias, and confounding. We outline how these biases influence existing research and identify opportunities for future research. Epidemiologists have the potential to contribute a great deal to the study of sexuality by bringing their considerable methodological expertise to long-standing challenges in the field. We hope to encourage epidemiologists to broaden their sexual health research to encompass positive outcomes and pleasure.

Comparing the Effect of Early Versus Delayed Metformin Treatment on Glycaemic Parameters Among Australian Adults With Incident Diabetes: Evidence Using a National General Practice Database.

PURPOSE: To compare the effect of early vs delayed metformin treatment for glycaemic management among patients with incident diabetes. METHODS: Cohort study using electronic health records of regular patients (1+ visits per year in 3 consecutive years) aged 40+ years with 'incident' diabetes attending Australian general practices (MedicineInsight, 2011-2018). Patients with incident diabetes were defined as those who had a) 12+ months of medical data before the first recording of a diabetes diagnosis AND b) a diagnosis of 'diabetes' recorded at least twice in their electronic medical records or a diagnosis of 'diabetes' recorded only once combined with at least 1 abnormal glycaemic result (i.e., HbA1c ≥6.5%, fasting blood glucose [FBG] ≥7.0 mmol/L, or oral glucose tolerance test ≥11.1mmol/L) in the preceding 3 months. The effect of early (<3 months), timely (3-6 months), or delayed (6-12 months) initiation of metformin treatment vs no metformin treatment within 12 months of diagnosis on HbA1c and FBG levels 3 to 24 months after diagnosis was compared using linear regression and augmented inverse probability weighted models. Patients initially managed with other antidiabetic medications (alone or combined with metformin) were excluded. FINDINGS: Of 18,856 patients with incident diabetes, 38.8% were prescribed metformin within 3 months, 3.9% between 3 and 6 months, and 6.2% between 6 and 12 months after diagnosis. The untreated group had the lowest baseline parameters (mean HbA1c 6.4%; FBG 6.9mmol/L) and maintained steady levels throughout follow-up. Baseline glycaemic parameters for those on early treatment with metformin (<3 months since diagnosis) were the highest among all groups (mean HbA1c 7.6%; FBG 8.8mmol/L), reaching controlled levels at 3 to 6 months (mean HbA1c 6.5%; FBG 6.9mmol/L) with sustained improvement until the end of follow-up (mean HbA1c 6.4%; FBG 6.9mmol/L at 18-24 months). Patients with timely and delayed treatment also improved their glycaemic parameters after initiating treatment (timely treatment: mean HbA1c 7.3% and FBG 8.3mmol/L at 3-6 months; 6.6% and 6.9mmol/L at 6-12 months; delayed treatment: mean HbA1c 7.2% and FBG 8.4mmol/L at 6-12 months; 6.7% and 7.1mmol/L at 12-18 months). Compared to those not managed with metformin, the corresponding average treatment effect for HbA1c at 18-24 months was +0.04% (95%CI -0.05;0.10) for early, +0.24% (95%CI 0.11;0.37) for timely, and +0.29% (95%CI 0.20;0.39) for delayed treatment. IMPLICATIONS: Early metformin therapy (<3 months) for patients recently diagnosed with diabetes consistently improved HbA1c and FBG levels in the first 24 months of diagnosis.

Invited Commentary: Combining Information to Answer Epidemiologic Questions About a Target Population.

Epidemiologists are attempting to address research questions of increasing complexity by developing novel methods for combining information from diverse sources. Cole et al. (Am J Epidemiol. 2023;192(3)467-474) provide 2 examples of the process of combining information to draw inferences about a population proportion. In this commentary, we consider combining information to learn about a target population as an epidemiologic activity and distinguish it from more conventional meta-analyses. We examine possible rationales for combining information and discuss broad methodological considerations, with an emphasis on study design, assumptions, and sources of uncertainty.

The Rotterdam Study. Design update and major findings between 2020 and 2024.

The Rotterdam Study is a population-based cohort study, started in 1990 in the district of Ommoord in the city of Rotterdam, the Netherlands, with the aim to describe the prevalence and incidence, unravel the etiology, and identify targets for prediction, prevention or intervention of multifactorial diseases in mid-life and elderly. The study currently includes 17,931 participants (overall response rate 65%), aged 40 years and over, who are examined in-person every 3 to 5 years in a dedicated research facility, and who are followed-up continuously through automated linkage with health care providers, both regionally and nationally. Research within the Rotterdam Study is carried out along two axes. First, research lines are oriented around diseases and clinical conditions, which are reflective of medical specializations. Second, cross-cutting research lines transverse these clinical demarcations allowing for inter- and multidisciplinary research. These research lines generally reflect subdomains within epidemiology. This paper describes recent methodological updates and main findings from each of these research lines. Also, future perspective for coming years highlighted.

Data Distribution: Normal or Abnormal?

Determining if the frequency distribution of a given data set follows a normal distribution or not is among the first steps of data analysis. Visual examination of the data, commonly by Q-Q plot, although is acceptable by many scientists, is considered subjective and not acceptable by other researchers. One-sample Kolmogorov-Smirnov test with Lilliefors correction (for a sample size ≥ 50) and Shapiro-Wilk test (for a sample size < 50) are common statistical tests for checking the normality of a data set quantitatively. As parametric tests, which assume that the data distribution is normal (Gaussian, bell-shaped), are more robust compared to their non-parametric counterparts, we commonly use transformations (e.g., log-transformation, Box-Cox transformation, etc.) to make the frequency distribution of non-normally distributed data close to a normal distribution. Herein, I wish to reflect on presenting how to practically work with these statistical methods through examining of real data sets.

Stratified analysis in comparative effectiveness studies that emulate randomized trials.

PURPOSE: For observational cohort studies that employ matching by propensity scores (PS), preliminary stratification by consequential predictors of outcome better emulates stratified randomization and potentially reduces variance and bias through relaxed dependence on modeling assumptions. We assessed the impact of pre-stratification in two real-life examples. For both, prior evidence from placebo-controlled randomized clinical trials (RCTs) suggested small or no risk reduction, but observational analysis suggested protection, presumably the result of confounding bias. STUDY DESIGN AND SETTING: The study populations consisted of Medicare beneficiaries (2014-18) with type 2 diabetes initiating either (i) empagliflozin versus dipeptidyl peptidase-4 inhibitors (DPP-4i) or (ii) empagliflozin versus glucagon-like peptide-1 receptor agonists (GLP-1RA). The outcome was myocardial infarction or stroke. We estimated hazard ratios (HR) and rate differences (RD) after controlling for 143 pre-exposure covariates via 1:1 PS matching after (1) PS estimation in the total cohort (total-cohort PS-matching) and (2) PS estimation separately by baseline cardiovascular disease (stratified PS matching). RESULTS: Stratified PS matching resulted in HRs that exceeded those from total-cohort PS-matching by 13% and 9%, respectively, for the comparisons of empagliflozin to DPP-4i and GLP-1RA. Against both comparators, HRs and RDs after stratified PS matching were closer to the null, with slightly higher variances (2%-3%) than those after total-cohort PS matching. CONCLUSION: Stratified PS matching produced effect estimates closer to the expected trial findings than total-cohort PS matching. The price paid in increased variance was minimal.

Geographic Variation, Economic Activity, and Labor Market Characteristics in Trajectories of Suicide in the United States, 2008-2020.

Suicide rates in the United States have increased over the past 15 years, with substantial geographic variation in these increases; yet there have been few attempts to cluster counties by the magnitude of suicide rate changes according to intercept and slope or to identify the economic precursors of increases. We used vital statistics data and growth mixture models to identify clusters of counties by their magnitude of suicide growth from 2008 to 2020 and examined associations with county economic and labor indices. Our models identified 5 clusters, each differentiated by intercept and slope magnitude, with the highest-rate cluster (4% of counties) being observed mainly in sparsely populated areas in the West and Alaska, starting the time series at 25.4 suicides per 100,000 population, and exhibiting the steepest increase in slope (0.69/100,000/year). There was no cluster for which the suicide rate was stable or declining. Counties in the highest-rate cluster were more likely to have agricultural and service economies and less likely to have urban professional economies. Given the increased burden of suicide, with no clusters of counties improving over time, additional policy and prevention efforts are needed, particularly targeted at rural areas in the West.

Recent Methodological Trends in Epidemiology: No Need for Data-Driven Variable Selection?

Variable selection in regression models is a particularly important issue in epidemiology, where one usually encounters observational studies. In contrast to randomized trials or experiments, confounding is often not controlled by the study design, but has to be accounted for by suitable statistical methods. For instance, when risk factors should be identified with unconfounded effect estimates, multivariable regression techniques can help to adjust for confounders. We investigated the current practice of variable selection in 4 major epidemiologic journals in 2019 and found that the majority of articles used subject-matter knowledge to determine a priori the set of included variables. In comparison with previous reviews from 2008 and 2015, fewer articles applied data-driven variable selection. Furthermore, for most articles the main aim of analysis was hypothesis-driven effect estimation in rather low-dimensional data situations (i.e., large sample size compared with the number of variables). Based on our results, we discuss the role of data-driven variable selection in epidemiology.

Confounder Adjustment Using the Disease Risk Score: A Proposal for Weighting Methods.

Propensity score analysis is a common approach to addressing confounding in nonrandomized studies. Its implementation, however, requires important assumptions (e.g., positivity). The disease risk score (DRS) is an alternative confounding score that can relax some of these assumptions. Like the propensity score, the DRS summarizes multiple confounders into a single score, on which conditioning by matching allows the estimation of causal effects. However, matching relies on arbitrary choices for pruning out data (e.g., matching ratio, algorithm, and caliper width) and may be computationally demanding. Alternatively, weighting methods, common in propensity score analysis, are easy to implement and may entail fewer choices, yet none have been developed for the DRS. Here we present 2 weighting approaches: One derives directly from inverse probability weighting; the other, named target distribution weighting, relates to importance sampling. We empirically show that inverse probability weighting and target distribution weighting display performance comparable to matching techniques in terms of bias but outperform them in terms of efficiency (mean squared error) and computational speed (up to >870 times faster in an illustrative study). We illustrate implementation of the methods in 2 case studies where we investigate placebo treatments for multiple sclerosis and administration of aspirin in stroke patients.