American Journal of Epidemiology

Background Population research increasingly relies on integrating data across studies to understand patterns of health and disease at broader scales. Such integration expands sample size and supports multimodal analyses, but independent cohort studies are often not directly comparable due to differences in instruments, timing, and measurement structure. Data harmonization addresses this challenge by transforming heterogeneous variables into comparable measures, allowing datasets collected under different protocols to be analyzed jointly, but published harmonization efforts seldom include an analysis conducted on the integrated dataset itself. Methods We developed an integrated dataset, combining two longitudinal cohorts, the Boston Puerto Rican Health Study (BPRHS) and the Puerto Rico Observational Study of Psychosocial, Environmental, and Chronic Disease Trends (PROSPECT). Comparability between variables was established through an explicit framework distinguishing availability from construct comparability, and the evidence supporting comparability is reported by class of feature, including tests of measurement invariance and differential test functioning for the multi-item psychosocial scales. To support transparency and reuse, we provide an open-source Python pipeline and structured metadata describing variable alignment and transformations. As a worked example of downstream use, we conducted a Random Forest analysis predicting diabetes status. Results The integrated dataset comprises 539 harmonized variables across four BPRHS visits and two PROSPECT visits, each documented with its source variable, question wording, and transformation. In the worked example, the integrated dataset retained predictive signal relative to cohort-specific models and improved performance for the smaller cohort. Conclusions These findings illustrate how transparent, documented cross-cohort integration can support predictive population health research while making the basis for comparability explicit.

Background. Mendelian randomization (MR) studies often evaluate exposures such as LDL cholesterol (LDL-C). The widespread use of lipid lowering medications complicates the interpretation and the validity of MR estimates. Methods. We describe two causal estimands in populations with medication use: a lifetime effect, and a lifetime effect under no medication use. In simulations, we compared the common approach of excluding medication users with estimation based on inverse probability (IP) weighting to adjust for medication use. We applied both approaches to a MR analysis of LDL-C and coronary artery disease in the Million Veteran Program (MVP), a large prospective cohort of U.S. veterans with linked electronic health record and genetic data. Results. In simulations, MR analyses that did not adjust for medication use estimated a lifetime effect that reflected a valid estimate of a total effect that included both the harms of higher LDL-C and the benefits of statins. To estimate the effect under no medication use, excluding statin users introduced selection bias. Alternatively, IP weighting could address bias from incident statin users, but could not address bias related to prevalent medication use. Estimates from MVP data varied considerably, reflecting the importance of these analytic choices. Conclusions. Different medication adjustment strategies in MR studies implicitly target different causal estimands and are subject to distinct biases. Transparent analytical choices and careful interpretation are essential for informative MR results in the context of widespread medication use.

Background and Objective. Missing data frequently occurs in health databases and can bias analyses if not correctly dealt with. Using real-world data, we compared complete-case and multiple-imputation methods for recovering true parameters of a multivariable logistic regression model for the association between maternal glucose levels during pregnancy and child excess weight at preschool age, where missing values were present in as much as 30% of our sample. Methods. This study utilized a cohort of 130,424 children with complete preschool-age body mass index (BMI) measurements from the Calgary and Edmonton health regions of Alberta, Canada. In the complete BMI data, we introduced missingness through deletion following three distinct mechanisms: missing completely at random (MCAR), at random (MAR), and not at random (MNAR). To handle the missing data created, we employed complete-case and multiple-imputation methods. Maternal glucose levels during pregnancy were categorized into five groups and its association with child excess weight at pre-school age was determined based on a logistic regression model using the full observed data (yielding true values), observed data that was not deleted (complete-case estimates), and imputed data (multiple-imputation estimates). The accuracy of complete-case and multiple-imputation estimates were evaluated against the true values. Finally, we conducted a sensitivity analysis for the MNAR mechanism using pattern-mixture models with an additive shift. Results. Under MCAR and MAR, multiple-imputation generally outperformed complete-case, yielding smaller absolute and relative bias. Both methods achieved high significance ([≥] 0.96) for most effects. Mean squared errors for multiple-imputation and complete-case were similar missing completely at random, missing at random, and coverage was consistently high ([≥] 0.99). Under MNAR, both complete-case and multiple-imputation showed poor performance regarding bias and statistical significance. Sensitivity analysis using pattern-mixture models indicated performance varied by specific effect. Conclusions. Under MCAR and MAR, multiple-imputation introduced higher bias but demonstrated superior overall performance based on mean squared error and restored statistical power. Conversely, both methods failed under MNAR, where pattern-mixture modeling sensitivity analyses revealed highly variable, effect-specific performance due to unverifiable shift assumptions. When faced with missing data, researchers should assess missingness mechanisms, report both complete-case and multiple-imputation estimates under MCAR/MAR while accounting for power-versus-bias tradeoffs, and employ pattern-mixture sensitivity analyses to test robustness when MNAR is plausible.

Background: Excess mortality estimates depend on an expected baseline, an unobservable quantity typically derived from models with prescribed functional forms and short reference periods that exclude exceptional events. Singular Spectrum Analysis (SSA) offers a nonparametric alternative but requires subjective component selection and is sensitive to outliers. We combined SSA with automated component selection and outlier interpolation to estimate long-term baselines and compile a catalog of excess mortality events (EMEs). Methods: We analyzed weekly all-cause mortality (2015-2024) for six metropolitan areas with contrasting climatic, demographic, and epidemiological contexts: Brussels, Antwerp, and Liege (Belgium), and Athens, Thessaloniki, and Larisa (Greece). SSA was applied to standardized log-mortality rates; components were selected using Monte Carlo SSA, and COVID-19 periods were interpolated iteratively. EMEs were defined as weeks exceeding a quasi-Poisson two-z-score threshold, then screened for coincidence with heat waves, cold waves, COVID-19 periods, and influenza epidemics. Results: Baselines were consistent within countries: annual cycles in Belgium, with an additional significant semiannual cycle in Greece reflecting summer mortality. We cataloged 170 EMEs spanning 456 weeks and 14,210 excess deaths. Pandemic weeks accounted for 77% of excess mortality, and 96 weeks coincided with temperature extremes; 56 EMEs, representing 3.7% of excess mortality, remained unmatched. Conclusions: SSA yields reproducible, low-assumption baselines that capture country-specific seasonality. Because flexible baselines can absorb recurrent seasonal mortality, component selection is a substantive choice that should be reported explicitly, together with its sensitivity.

Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy (MAIHDA) describes health inequalities through three components: (i) specific contextual effects (SCE), (ii) general contextual effects (GCE), and (iii) discriminatory accuracy of the context. We present Simple-Means MAIHDA (S-MAIHDA), which estimates each stratum directly from its observed individuals, with no distributional assumption. The observed proportions are unbiased whatever the stratum size, and their confidence intervals report the uncertainty honestly. S-MAIHDA operationalises the three components on the probability scale. The SCE are the raw and standardised stratum prevalences and the modification of the sociodemographic average differences by the area. The GCE are the variance partition coefficient (VPC) and the contextual structuring of the between-stratum inequality, expressed as the contextual clustering of inequalities, the additive sociodemographic differences, and the contextual modification of inequalities (CMI). The contextual discriminatory accuracy is expressed by the area under the ROC curve (AUC), and the sensitivity and specificity at the population prevalence as the threshold for a possible intervention. Because its estimates are the observed data themselves, S-MAIHDA is the canonical description, and the compare diagnostic quantifies how Random-Effects MAIHDA (RE-MAIHDA), the usual implementation, departs from it: RE shrinkage pulls small strata towards the overall mean and can hide the very inequalities the analysis seeks. The approach is implemented in the smaihda Stata command and reproduced in free Python code. We illustrate S-MAIHDA on register data from Malmo, Sweden (43,291 individuals; 300 area-sociodemographic strata), showing how the three components separate two contrasting outcomes: psychotropic medication use, almost purely sociodemographic, stable across areas, with weak contextual structuring (VPC {approx} 4%, CMI {approx} 0%); and choice of a private general practitioner, strongly geographical (VPC {approx} 11%, CMI {approx} 17%), with the sociodemographic differences reshaped and amplified in wealthy areas. RE-MAIHDA attenuated inequalities. For describing inequalities, S-MAIHDA preserves what the data show.

Background and Objectives: Subjective cognitive decline (SCD), self-reported worsening confusion or memory over the past year, is a common early marker of cognitive concern with relevance for Alzheimer's disease prevention and population health. Population-based machine learning benchmarks that respect temporal drift in public health surveillance remain limited. We developed a reusable multi-language prediction and interpretability framework for SCD using Behavioral Risk Factor Surveillance System (BRFSS) Cognitive Decline data. Methods: We analyzed pooled national (n = 298,944) and New York (n = 30,366) cohorts with chronological train (2015-2019), validation (national: 2020-2022; New York: 2020-2021), and locked test (2023-2024) splits. Nested LASSO identified stable predictors. Sixteen machine learning algorithms were compared under year-grouped cross-validation with SMOTE restricted to training folds. Four end-to-end Python/R pipelines (single-model or soft-voting) used validation-only isotonic calibration and Youden thresholding. Primary reporting pipelines were prespecified before test unlock (national: R tidymodels single-model; New York: Python single-model); algorithms within each pipeline were chosen by validation ROC-AUC. Post-hoc GLMs (national unweighted; New York design-weighted) and two training-only knowledge-graph layers supported interpretability. Results: Locked-test discrimination was consistent across implementations (ROC-AUC approximately 0.76-0.77). Prespecified pipelines achieved test ROC-AUC 0.770 (95% CI 0.767-0.773) nationally (R gradient boosting) and 0.762 (95% CI 0.746-0.777) in New York (Python AdaBoost). Soft-voting pipelines performed similarly (national 0.770; New York 0.757) and were treated as sensitivity benchmarks. Predicted probabilities were reasonably calibrated (Brier 0.118 nationally; 0.112 in New York), and higher scores among SCD-positive respondents persisted across survey years. Difficulty deciding, mental health, and functional health items ranked highest across permutation importance, SHAP, and GLMs. Respondents who reported no difficulty deciding (DECIDE = 2) had substantially lower odds of SCD than those who reported difficulty (DECIDE = 1; aOR approximately 0.13; FDR < 0.05). Training-only knowledge graphs likewise placed difficulty deciding nearest to SCD in both cohorts. Conclusions: A temporally locked, multi-pipeline BRFSS benchmark yields stable future-year SCD risk ranking, usable calibrated probability scores that remain separated by SCD status across survey years, and convergent interpretability signals. The open implementation supports reproducible surveillance-oriented machine learning for cognitive health.

Background: Reliable estimation of excess mortality is central to population health surveillance. We introduce NeMMo (New Mortality Model), an evolution of the EuroMOMO model for estimating weekly all-cause expected mortality, and assess its behaviour and performance on empirical data. Methods: NeMMo incorporates population offsets, stratifies observed deaths by age group and models seasonality using a periodic B-spline rather than a Serfling-type sinusoidal function. Baseline weeks are selected by a data-driven procedure minimizing the skewness of the residuals before refitting the model, instead of relying solely on fixed calendar windows. NeMMo enables pooling across age groups, direct age standardization and incorporation of external predictors. We applied NeMMo and EuroMOMO to mortality and population data downloaded from Eurostat for 31 countries from 2015 onwards, excluding the COVID-19 pandemic period from baseline estimation. Results: For most countries NeMMo produced a higher expected mortality baseline that better tracked observed deaths, as well as tighter prediction intervals and higher maximum Z-scores, suggesting improved discrimination of mortality excesses. Z-scores and P-scores during non-pandemic weeks were closer to zero with NeMMo than with EuroMOMO but further elevated during pandemic weeks, providing greater separation between pandemic and non-pandemic mortality. Incorporating population offsets resulted in negative linear trends across all countries, consistent with declining mortality after accounting for demographic changes. The periodic B-spline identified substantial heterogeneity in the shape and timing of seasonal mortality that was not captured by a sinusoidal function. Conclusions: NeMMo provides a flexible and parsimonious framework for all-cause mortality surveillance that improves the established EuroMOMO model and offers theoretical, empirical and practical advantages. It is thus suitable both for detecting short-term spikes and for the long-term, age-adjusted quantification and comparison of mortality excesses that has become increasingly important since the COVID-19 pandemic. The accompanying 'nemmo' package for R facilitates its widespread adoption and application.

Background: Large-scale prevention and population health monitoring require measurement approaches that are both feasible and informative. Although several self-measurable anthropometric and fitness indicators have been associated with cardiometabolic risk, it remains unclear whether combining multiple measurements provides meaningful improvements over simpler approaches. We evaluated whether a parsimonious set of self-measurable indicators can achieve classification performance comparable to a full candidate set and quantified the incremental value of additional measurements. Methods: Using data from 8,275 adults in the NHANES 1999-2004 cohorts, we evaluated a predefined minimal set of four self-measurable anthropometric and fitness indicators (body mass index (BMI), waist-to-height ratio (WHtR), mid-upper arm circumference (MUAC), and VO2max (as a proxy for the 6-minute walk test) as candidate indicators of cardiometabolic risk. Their ability to reflect underlying clinical risk factors related to adiposity, glucose and lipid metabolism, and physical fitness was assessed using nested logistic regression models, likelihood ratio tests, discrimination metrics, and decision tree analyses. Results: WHtR consistently showed the strongest discriminative performance, with {Delta}PR-AUC values for BMI versus WHtR ranging from -0.002 to -0.037, and emerged as the primary splitting variable. Adding BMI to WHtR resulted in small gains in PR-AUC for most outcomes, ranging from 0.000 to 0.008, except for triglycerides where the gain was larger ({Delta}PR-AUC=0.039). Further inclusion of MUAC and VO2max provided limited additional value overall, with evidence of variation across outcomes and sex stratified analyses. Conclusion: Most classification performance was achieved using a limited number of simple self-measurable indicators, with little additional benefit from incorporating further measurements. These findings suggest that parsimonious measurement strategies may provide a feasible approach for cardiometabolic risk classification in population health and prevention settings while reducing measurement burden.

Difficulties ascertaining the true burden of infection and mortality from SARS-CoV-2 hindered disease surveillance and healthcare resource allocation throughout the COVID-19 pandemic. Although all-cause excess mortality estimates have long been deployed to avoid the challenges of case detection when testing availability is variable, aggregation of mortality by cause of death and spatial unit may mask underlying patterns and hide underlying associations with social factors. Furthermore, some noinfectious causes of death may have decreased during the pandemic, resulting in all-cause excess mortality capturing a net pandemic effect rather than the direct burden of Covid-19. Using mortality data in the state of Michigan from January 2015 - November 2019, we calculated the expected mortality for December 2019 - November 2022 using negative binomial regression. Comparing these estimates to the observed mortality, we calculated the burden of excess mortality during the winter 2019 influenza season and COVID-19 pandemic, as divided into 7 distinct periods based on dominant viral strains and pharmaceutical and non-pharmaceutical interventions implemented. All analyses were performed with the data aggregated by cause of death and to the county level and repeated at both the county and census tract level disaggregated into two crude cause of death categories: acute respiratory infections (ARIs) and all other causes (non-acute respiratory infections, non-ARIs). We examined spatial heterogeneity of excess mortality by cause using the Theil index to determine if variation in excess mortality was driven by differences between or within counties. The excess mortality rate from ARI peaked in the first phase of the pandemic, with a ratio of 10.7 observed to expected deaths (95% CrI: 10.3 -1.1). However, the all- cause excess mortality rate rose only to 1.20 excess deaths per expected (95% CrI: 1.17 - 1.22) and the non-ARI excess mortality rate was unchanged. Comparing ARI excess mortality rates calculated at the county vs. census tract level demonstrated large and unpredictable variation, with county level estimates ranging from 0.42 to 1.2 times the tract level estimates. However, county level estimates of non-ARI and all- cause excess mortality ratios were consistently closer to tract level estimates. Within county variation of ARI mortality decreased during the pandemic and slowly returned to reference levels, suggesting spatial patterning of ARI mortality became more similar across census tracts as rates rose statewide. Social vulnerability was associated with an increased ARI excess mortality rate ratio at the census tract level during the first pandemic phase (EMRR: 1.1, 95% CrI: 1.08 - 1.13), but this effect was attenuated when aggregated to all-cause mortality or to the county level. Our findings indicate that using county-level all-cause excess mortality as a proxy for ARI excess mortality obscuredthe burden of ARI death, particularly during the most acute phases of the COVID-19 pandemic. Relying on aggregated metric of all-cause mortality limits the power to detect large shifts in cause-specific mortality, and that limitation can be mitigated by even using coarse categories such as ARI vs. non-ARI deaths. Similarly, finer scale spatial units allow for the detection of local trends necessary to identify associated social factors.

Background: Waning vaccine protection and variable epidemic timing create a tradeoff between vaccinating before influenza circulation and preserving protection when disease burden is greatest. Existing guidance does not quantify the week that best balances this tradeoff. Objective: To estimate the optimal week for influenza vaccination while accounting for uncertainty in epidemic timing and vaccine protection. Design: Simulation-based probabilistic decision analysis. Setting: United States outpatient influenza surveillance from the 50 states and District of Columbia during the 2010/11 through 2018/19 and 2023/24 through 2025/26 seasons. Participants: People in the United States who would receive 1 influenza vaccine dose. Interventions: Vaccination during Morbidity and Mortality Weekly Report (MMWR) weeks 36 through 12. Measurements: We combined influenza-like illness (ILI) syndromic surveillance and influenza-specific data with uncertainty in epidemic timing, initial vaccine effectiveness, waning, and immune-response lag. For each candidate week, the model estimated direct protection against outpatient surveillance burden. Results: The primary model identified the optimal time to vaccinate as week 47, with weeks 47-48 being of similar utility. Our influenza-weighted outpatient analysis (ILI+) selected week 48. Compared with week 44, near the end of the current September-October guidance window, week 48 reduced modeled influenza-weighted outpatient burden by 50 visit equivalents per 10,000 weekly encounters. Limitations: The model assumed vaccination would occur, used outpatient surveillance burden rather than severe outcomes, and did not incorporate transmission effects. Conclusion: Under the primary waning model, these findings favor vaccinating during weeks 47-48, approximately late November through early December, for most one-dose recipients.

Longitudinal household studies, combined with mathematical modeling, are widely used to characterize the drivers of respiratory pathogen transmission, including the effects of age and symptoms. In practice, household recruitment protocols vary across studies, potentially introducing biases into observed data. However, these biases are typically overlooked in statistical inference, and their impact on parameter estimates remains unknown. Here, we use synthetic household outbreak data simulated under different recruitment protocols to evaluate how recruiting through infected children affects estimates of age-specific infectiousness and susceptibility. We show that, under child-based recruitment, the standard likelihood, which accounts only for transmission dynamics, leads to underestimating child infectiousness and overestimating child susceptibility by more than 30%. We then propose a novel estimation framework that explicitly incorporates the household recruitment process into the likelihood and show that it substantially reduces these biases. Applying this new approach to a French household study conducted during the COVID-19 pandemic, we estimated that children under 6 had 49% lower infectiousness than teenagers and adults during the Alpha wave, whereas no difference was observed during the Omicron wave. This study demonstrates that ignoring recruitment protocols can bias key epidemiological parameter estimates and highlights the importance of accounting for study design.

Tracking how an infectious disease spreads in time and space relies on several distinct sources of surveillance data, reported case counts, viral concentrations in wastewater, seroprevalence surveys, and pathogen genomic sequences, each of which is imperfect and captures only part of the underlying transmission process. These data streams are typically analyzed separately or with highly parameterized, disease-specific models, making it difficult to combine their complementary strengths. Here we present MASCOT-DataStreams (MASCOT-DS), a BEAST2 software package that extends the structured coalescent model MASCOT to jointly infer prevalence over time and transmission rates between locations from any combination of case counts, wastewater concentrations, seroprevalence surveys, and pathogen phylogenies. Using simulated outbreaks in structured populations, we show that MASCOT-DS accurately recovers true prevalence trajectories and between-location migration rates. We then apply MASCOT-DS to genomic, case count, wastewater, and seroprevalence data from the SARS-CoV-2 Epsilon wave (winter 2020-21) in three San Francisco Bay Area counties, reconstructing county-level prevalence dynamics and quantifying transmission within and into the region. By systematically removing individual data streams, we find that genomic data are uniquely required to estimate transmission between locations, while seroprevalence data are essential for anchoring the overall magnitude of an outbreak; case counts and wastewater concentrations play largely interchangeable roles in capturing outbreak shape. These results demonstrate that integrating complementary epidemiological data streams substantially increases the certainty of transmission dynamics estimates compared to relying on any single data stream, and provides a framework for evaluating the added value of different surveillance strategies.

Doubly robust survival estimators require nuisance models that the available information can support. We examine design and estimator dependence in TCGA breast cancer data. Treatment indicators disagree for 337 of 1076 patients; 67.4% have unobserved five-year binary outcomes; and receptor imputation expands the triple-negative subgroup from 156 directly measured patients to 226. We compare chemotherapy initiation by day 90 with non-initiation, using adjusted restricted mean survival time (RMST) over the next 1095 days. Weighting-only estimates are 55 days (95% bootstrap interval: -24, 133) in triple- negative disease (16 deaths) and 38 days (12, 72) in the whole cohort (72 deaths). A Cox outcome-model benchmark yields 108 and 35 days, respectively, demonstrating material model dependence in the smaller population. A parameter-to-information ratio summarizes the burden on complete-outcome regressions within cross-fitting. Its simulation-calibrated warning level is 0.31 (0.28 - 0.35) for the original regime; additional simulations examine covariate-dependent censoring, delayed benefit and nonlinear misspecification. The ratio does not certify estimator validity or replace weight diagnostics. The contribution is a reproducible workflow that separates design, nuisance-model support and estimator stability before interpreting observational survival contrasts.

Background: Retirement is a major life transition that may influence cognitive aging. Retirement pathways are becoming increasingly diversified, yet evidence on how heterogeneous retirement trajectories relate to dementia risk remains limited. This study used sequence and cluster analysis to characterize thousands of transition-to-retirement trajectories, including retirement timing and work intensity, to estimate associations with dementia risk. Methods: We used data from 13,879 participants in the U.S. Health and Retirement Study. Our exposure was transition-to-retirement trajectories between ages 51 and 75 identified using sequence and cluster analysis. Our outcome was a validated algorithmically derived dementia probability measure. We estimated associations between trajectory clusters and dementia risk beginning at age 76 using generalized estimating equations, adjusting for sociodemographic characteristics. Inverse probability weights were used to account for attrition. Results: We identified ten distinct transition-to-retirement trajectories characterized by differences in timing of retirement, part-time work, disability, and labor force attachment. Compared with individuals retiring after full-time work in their early-to-mid 60s, dementia risk was similar among those retiring earlier or later from full-time work. In contrast, individuals with sustained part-time employment between ages 51 and 75 had lower odds of dementia (OR: 0.62, 95% CI: 0.48, 0.80), while trajectories characterized by prolonged disability were associated with higher odds of dementia (OR: 1.53, 95% CI: 1.22, 1.91). Conclusions: Retirement timing alone showed limited association with later-life dementia risk, whereas sustained part-time employment and disability-related trajectories were more strongly associated with dementia risk. These findings suggest that employment intensity and health-related exit from the labor force may be more relevant for dementia risk than retirement timing, and that gradual and flexible work-to-retirement transitions may benefit brain health. Background: Retirement is a major life transition that may influence cognitive aging. Retirement pathways are becoming increasingly diversified, yet evidence on how heterogeneous retirement trajectories relate to dementia risk remains limited. This study used sequence and cluster analysis to characterize thousands of transition-to-retirement trajectories, including retirement timing and work intensity, to estimate associations with dementia risk. Methods: We used data from 13,879 participants in the U.S. Health and Retirement Study. Our exposure was transition-to-retirement trajectories between ages 51 and 75 identified using sequence and cluster analysis. Our outcome was a validated algorithmically derived dementia probability measure. We estimated associations between trajectory clusters and dementia risk beginning at age 76 using generalized estimating equations, adjusting for sociodemographic characteristics. Inverse probability weights were used to account for attrition. Results: We identified ten distinct transition-to-retirement trajectories characterized by differences in timing of retirement, part-time work, disability, and labor force attachment. Compared with individuals retiring after full-time work in their early-to-mid 60s, dementia risk was similar among those retiring earlier or later from full-time work. In contrast, individuals with sustained part-time employment between ages 51 and 75 had lower odds of dementia (OR: 0.62, 95% CI: 0.48, 0.80), while trajectories characterized by prolonged disability were associated with higher odds of dementia (OR: 1.53, 95% CI: 1.22, 1.91). Conclusions: Retirement timing alone showed limited association with later-life dementia risk, whereas sustained part-time employment and disability-related trajectories were more strongly associated with dementia risk. These findings suggest that employment intensity and health-related exit from the labor force may be more relevant for dementia risk than retirement timing, and that gradual and flexible work-to-retirement transitions may benefit brain health.

Many urban health outcomes are shaped by environmental stressors that occur together rather than in isolation, yet methods for measuring such co-occurrence at the neighbourhood scale remain underdeveloped. We developed a multi-metric framework for joint co-exposure assessment and applied it to characterise the joint spatial distribution of summer surface heat and fine particulate matter (PM2.5) across 42,304 census tracts in 48 large US metropolitan areas during summers 2015 to 2020, covering approximately 174.6 million residents. The framework combines a composite co-exposure index, a joint exceedance indicator, a conditional exceedance ratio that compares observed joint occurrence to within-group statistical independence, and an upper tail dependence parameter estimated using both the non-parametric Caperaa-Fougeres-Genest estimator and a Gumbel copula, with bias-corrected and accelerated (BCa) confidence intervals obtained from a 5,000-replicate metropolitan-area block bootstrap. Among residents of predominantly Black tracts, 13.21% lived in neighbourhoods that simultaneously exceeded the within-metropolitan-area 80th percentile for both heat and PM2.5, compared with 3.33% of residents of predominantly White tracts; the corresponding heat-only and PM2.5-only ratios were 2.88 and 2.48. Residents of Home Owners Loan Corporation grade D tracts had 3.97 times the odds (95% confidence interval 2.79 to 5.66) of joint hotspot residence compared with grade A residents after adjustment for contemporary tract racial composition, poverty, renter-occupancy, and pre-1960 housing. The within-group conditional exceedance ratio at the 80th percentile was 2.29 in predominantly White tracts (95% BCa CI 1.81 to 2.78), 1.27 in predominantly Black tracts (0.71 to 1.56), and 1.13 in predominantly Hispanic tracts (0.70 to 1.41); the White interval excluded one while the Black and Hispanic intervals included one, which we interpret as power-limited given fewer contributing CBSAs. Magnitudes attenuated under near-surface air temperature surfaces but the direction and statistical significance of the primary findings were preserved. The framework is portable to other compound-exposure questions and supports cumulative-impact assessment.

Selection bias is a common concern in causal inference. Two mechanisms can give rise to selection bias: collider bias or stratifying on an effect modifier of the exposure-outcome causal effect. Research into the impact of selection bias on instrumental variable (IV) analyses has focused primarily on collider bias. Here, we show that IV estimates can be biased if selection is an effect modifier (or caused by an effect modifier) of the exposure-outcome effect. In a simulation study, we investigate the size and direction of the bias when estimating effects on the mean difference or odds ratio scale. Bias can also occur due to heterogeneity in the instrument-exposure relationship, although its magnitude is typically smaller. Finally, we discuss methods to deal with this bias in practice, including inverse probability weighting (IPW).

Forecasting influenza hospitalizations informs public health preparedness, yet questions remain about which types of forecasts best guide action. We evaluate categorical trend forecasts, which communicate probabilities of upcoming increases or decreases in epidemic trajectories, submitted to CDC's FluSight Forecasting Challenge between Fall-2024 and Spring-2026. Teams submitted probability distributions over five categories describing direction and magnitude of week-over-week changes in laboratory-confirmed influenza hospital admissions. We assessed performance using Ranked Probability Skill Score, Brier Skill Score, and measures of forecast-observation agreement. Most models outperformed an equal-probability baseline; the FluSight ensemble ranked among the top three in the 2024-25 and 2025-26 seasons. Forecasts were most accurate during stable periods and least during periods of rapid change, with most models underestimating observed trends. Conclusions were robust to choice of scoring metric and reference model. These results support categorical trend ensembles as an approach to communicating infectious disease forecasts that may inform public health decision-making.

Wearable (accelerometer) data and machine-learning allow objective assessment of the amount of daily physical activity. However, wearable-derived human activity is subject to measurement error. No studies have corrected the dose-response association between physical activity and survival time to chronic diseases, including cardiovascular disease (CVD). The objective is to estimate the measurement error-corrected association between CVD events and multiple measures of daily duration of light and total physical activity, derived from machine-learning and conventional accelerometer-processing methods. Our method combined an accelerated failure time model, spline, and simulation-extrapolation (SIMEX). The method recovered the true dose-response non-linear association in simulated data, while the naive model failed to capture it due to substantial attenuation. Application to the UK Biobank accelerometer cohort also showed an increased protective association of total physical activity after SIMEX correction (Time Ratio [TR] = 1.56, 95% CI: 1.28-1.82 vs. TR = 1.38, 95% CI: 1.24-1.54 for SIMEX-corrected vs. uncorrected dose-response association between the 95th and 5th percentiles of total activity), with a similar increase for light physical activity. Sensitivity analysis indicates that the female population experiences a substantially larger protective association after SIMEX correction than males. Dose-response survival analysis is a widely used analytical method in physical activity epidemiology and benefits from measurement error correction.

Surrogate endpoints, or measures used in place of the true outcome of interest, have relevance across multiple disease areas. The Prentice Criteria, proposed in 1989, assess surrogacy by evaluating how the treatment's effect on the true outcome operates through the potential surrogate. Using the A4 Study of solanezumab, we evaluate multiple formulations of the Prentice Criteria using causal mediation. We estimated direct and indirect effects of solanezumab on cognitive decline through cerebral amyloid across different, but reasonable, measures of exposure, mediator, outcome, and covariate adjustment. Unsurprisingly given that solanezumab did not show benefit, estimated indirect effects were close to zero. These results provide little evidence of meaningful mediation for memory and global cognition, but precision varied substantially. Confidence interval widths varied by up to a factor of 17 across specifications. Causal mediation analysis of individual-level randomized trial data may contribute to quantitative surrogate endpoint evaluation, but our findings indicate this is only the case when analytic choices are biologically justified, prespecified, and interpreted with attention to uncertainty.

Estimating seroprevalence for an infection -- the proportion of individuals with antibodies -- and its variability between subpopulations is a key first step for research and the allocation of resources for treatment and prevention. The relevant raw data is often serological assay measurements that serve as a proxy for antibody level, such as optical density values from enzyme-linked immunosorbent assays (ELISA). Analysis pipelines typically proceed through sequential steps of fitting calibration data independently in each separate batch, transforming assay measurements to antibody levels via each fitted calibration relationship, classifying antibody levels into discrete serostatus by comparison to a threshold, and finally comparing seropositive proportions between subpopulations. Such approaches have numerous limitations including loss of information, overconfidence (discarding uncertainty), and inappropriate choice of threshold. We developed a Bayesian statistical model that replaces the sequential steps of the typical pipeline by multiple levels within a single hierarchical model, treating each sample's antibody level as a model parameter rather than directly observed data. This integrates the connections from the raw assay measurements all the way through to subpopulation variability in seroprevalence, allowing partial pooling of information between related variables and the propagation of uncertainty from each part of the model throughout the whole of the rest of the model. We calculate observation probabilities for all assay measurements from both samples and calibration data, allowing easy identification of outlying calibration or sample measurements. We replace a single hard classification threshold for disease status by continuous probabilities that are adapted to each subpopulation. We allow a flexible multivariate random-effects logistic regression to capture variability in seroprevalence between subpopulations. Using simulated data we found that the typical stepwise approach gave prevalence estimates far from the true values with narrow confidence intervals. Our method, dvsb, had markedly greater accuracy. We report the run time and convergence of dvsb when applied to a real dataset for Lassa fever IgG antibodies measured with ELISA, comprising 72,863 measurements for 21,391 unique samples (results reported elsewhere). dvsb is available at https://github.com/BDI-pathogens/dvsb.