Excess natural-cause mortality in US counties and its association with reported COVID-19 deaths.

In the United States, estimates of excess deaths attributable to the COVID-19 pandemic have consistently surpassed reported COVID-19 death counts. Excess deaths reported to non-COVID-19 natural causes may represent unrecognized COVID-19 deaths, deaths caused by pandemic health care interruptions, and/or deaths from the pandemic's socioeconomic impacts. The geographic and temporal distribution of these deaths may help to evaluate which explanation is most plausible. We developed a Bayesian hierarchical model to produce monthly estimates of excess natural-cause mortality for US counties over the first 30 mo of the pandemic. From March 2020 through August 2022, 1,194,610 excess natural-cause deaths occurred nationally [90% PI (Posterior Interval): 1,046,000 to 1,340,204]. A total of 162,886 of these excess natural-cause deaths (90% PI: 14,276 to 308,480) were not reported to COVID-19. Overall, 15.8 excess deaths were reported to non-COVID-19 natural causes for every 100 reported COVID-19 deaths. This number was greater in nonmetropolitan counties (36.0 deaths), the West (Rocky Mountain states: 31.6 deaths; Pacific states: 25.5 deaths), and the South (East South Central states: 26.0 deaths; South Atlantic states: 25.0 deaths; West South Central states: 24.2 deaths). In contrast, reported COVID-19 death counts surpassed estimates of excess natural-cause deaths in metropolitan counties in the New England and Middle Atlantic states. Increases in reported COVID-19 deaths correlated temporally with increases in excess deaths reported to non-COVID-19 natural causes in the same and/or prior month. This suggests that many excess deaths reported to non-COVID-19 natural causes during the first 30 mo of the pandemic in the United States were unrecognized COVID-19 deaths.

Dynamics of racial disparities in all-cause mortality during the COVID-19 pandemic.

As research documenting disparate impacts of COVID-19 by race and ethnicity grows, little attention has been given to dynamics in mortality disparities during the pandemic and whether changes in disparities persist. We estimate age-standardized monthly all-cause mortality in the United States from January 2018 through February 2022 for seven racial/ethnic populations. Using joinpoint regression, we quantify trends in race-specific rate ratios relative to non-Hispanic White mortality to examine the magnitude of pandemic-related shifts in mortality disparities. Prepandemic disparities were stable from January 2018 through February 2020. With the start of the pandemic, relative mortality disadvantages increased for American Indian or Alaska Native (AIAN), Native Hawaiian or other Pacific Islander (NHOPI), and Black individuals, and relative mortality advantages decreased for Asian and Hispanic groups. Rate ratios generally increased during COVID-19 surges, with different patterns in the summer 2021 and winter 2021/2022 surges, when disparities approached prepandemic levels for Asian and Black individuals. However, two populations below age 65 fared worse than White individuals during these surges. For AIAN people, the observed rate ratio reached 2.25 (95% CI = 2.14, 2.37) in October 2021 vs. a prepandemic mean of 1.74 (95% CI = 1.62, 1.86), and for NHOPI people, the observed rate ratio reached 2.12 (95% CI = 1.92, 2.33) in August 2021 vs. a prepandemic mean of 1.31 (95% CI = 1.13, 1.49). Our results highlight the dynamic nature of racial/ethnic disparities in mortality and raise alarm about the exacerbation of mortality inequities for Indigenous groups due to the pandemic.

Considerations for Use of Blood-Based Biomarkers in Epidemiologic Dementia Research.

Dementia represents a growing public health burden with large social, racial, and ethnic disparities. The etiology of dementia is poorly understood, and the lack of robust biomarkers in diverse, population-representative samples is a barrier to moving dementia research forward. Existing biomarkers and other measures of pathology-derived from neuropathology, neuroimaging, and cerebrospinal fluid samples-are commonly collected from predominantly White and highly educated samples drawn from academic medical centers in urban settings. Blood-based biomarkers are noninvasive and less expensive, offering promise to expand our understanding of the pathophysiology of dementia, including in participants from historically excluded groups. Although largely not yet approved by the Food and Drug Administration or used in clinical settings, blood-based biomarkers are increasingly included in epidemiologic studies on dementia. Blood-based biomarkers in epidemiologic research may allow the field to more accurately understand the multifactorial etiology and sequence of events that characterize dementia-related pathophysiological changes. As blood-based dementia biomarkers continue to be developed and incorporated into research and practice, we outline considerations for using them in dementia epidemiology, and illustrate key concepts with Alzheimer's Disease Neuroimaging Initiative (2003-present) data. We focus on measurement, including both validity and reliability, and on the use of dementia blood-based biomarkers to promote equity in dementia research and cognitive aging. This article is part of a Special Collection on Mental Health.

A Critique and Examination of the Polysocial Risk Score Approach: Predicting Cognition in the Health and Retirement Study.

Polysocial risk scores were recently proposed as a strategy to improve clinical relevance of knowledge about social determinants of health. The objective of this paper was to assess if the polysocial risk score model improves prediction of cognition and all-cause mortality in middle-aged and older adults beyond simpler models including a smaller set of key social determinants of health. We used a sample of 13,773 individuals aged 50+ at baseline from the 2006 to 2018 waves of the Health and Retirement Study, a US population-based longitudinal cohort. Four linear mixed models were compared: two simple models including a priori selected covariates and two polysocial risk score models which used LASSO regularization to select covariates among 9 or 21 candidate social predictors. All models included age. Predictive accuracy was assessed via R-squared and root mean-squared prediction error (RMSPE) using training/test split and cross-validation. For predicting cognition, the simple model including age, race, gender, and education had an R-squared of 0.31 and an RMSPE of 0.880. Compared with this, the most complex polysocial risk score selected 12 predictors (R-squared=0.35 and RMSPE=0.858; 2.2% improvement). For all-cause mortality, the simple model including age, race, gender, and education had an AUROC of 0.747, while the most complex polysocial risk score did not demonstrate improved performance (AUROC = 0.745). Models built on a smaller set of key social determinants performed comparably to models built on a more complex set of social "risk factors".

Mediating pathways between neighborhood disadvantage and cardiovascular risk: Quasi-experimental evidence from a Danish refugee dispersal policy.

Research has documented that neighborhood disadvantage is associated with increased cardiovascular disease risk, but it is unclear which mechanistic pathways mediate this association across the life course. Leveraging a natural experiment in which refugees to Denmark were quasi-randomly assigned to neighborhoods across the country during 1986-1998 and using 30 years of follow-up data from population and health registers, we assessed whether and how individual-level poverty, unstable employment, and poor mental health mediate the relation between neighborhood disadvantage and the risk of hypertension, hyperlipidemia, and type 2 diabetes among Danish refugees (N= 40,811). Linear probability models using the discrete time-survival framework showed that neighborhood disadvantage was associated with increased risk of hypertension (0.05 percentage points [pp] per year [95%CI -0.00, 0.10]); hyperlipidemia (0.03 pp per year [95%CI -0.01, 0.07]), and diabetes (0.01 pp per year (95%CI -0.02, 0.03)). The Baron-Kenny product-of-coefficients method for counterfactual mediation analysis indicated that cumulative income mediated 6%-28% of the disadvantage effect on these outcomes. We find limited evidence of mediation by unstable employment and poor mental health. This study informs our theoretical understanding of the pathways linking neighborhood disadvantage with cardiovascular disease risk and identifies income security as a promising point of intervention in future research.

School racial composition, effect modification by caring teacher/staff presence, and mid/late-life depressive symptoms: findings from the Study of Healthy Aging among African Americans.

For Black students in the United States, attending schools with a higher proportion of White students is associated with worse mental and physical health outcomes in adolescence/early adulthood. No prior studies evaluate K-12 school racial composition and later-life mental health. In a cohort of Black adults ages 50+ in Northern California who retrospectively self-reported school racial composition for grades 1, 6, 9, and 12, we assessed the association between attending a school with mostly Black students vs. not and mid/late-life depressive symptoms (8-item PROMIS depression score, standardized to US adult population) using age-, sex/gender-, southern US birth-, and parental education-adjusted generalized estimating equations, and assessed effect modification by caring teacher/staff presence. Later-life depressive symptoms were lower among those who attended schools with mostly Black students in grades 1 and 6 (b=-0.12, 95% CI: -0.23, 0.00 and b=-0.11, 95% CI: -0.22, 0.00, respectively). In grade 6, this difference was larger for students without an adult at school who cared about them (b=-0.29, 95% CI: -0.51, -0.07 vs. b=-0.04, 95% CI: -0.17, 0.09). Among Black Americans, attending early school with mostly Black students may have later life mental health benefits; this protective association appears more important for students without caring teachers/staff.

Comparison of approaches to control for intracranial volume in research on the association of brain volumes with cognitive outcomes.

Most neuroimaging studies linking regional brain volumes with cognition correct for total intracranial volume (ICV), but methods used for this correction differ across studies. It is unknown whether different ICV correction methods yield consistent results. Using a brain-wide association approach in the MRI substudy of UK Biobank (N = 41,964; mean age = 64.5 years), we used regression models to estimate the associations of 58 regional brain volumetric measures with eight cognitive outcomes, comparing no correction and four ICV correction approaches. Approaches evaluated included: no correction; dividing regional volumes by ICV (proportional approach); including ICV as a covariate in the regression (adjustment approach); and regressing the regional volumes against ICV in different normative samples and using calculated residuals to determine associations (residual approach). We used Spearman-rank correlations and two consistency measures to quantify the extent to which associations were inconsistent across ICV correction approaches for each possible brain region and cognitive outcome pair across 2320 regression models. When the association between brain volume and cognitive performance was close to null, all approaches produced similar estimates close to the null. When associations between a regional volume and cognitive test were not null, the adjustment and residual approaches typically produced similar estimates, but these estimates were inconsistent with results from the crude and proportional approaches. For example, when using the crude approach, an increase of 0.114 (95% confidence interval [CI]: 0.103-0.125) in fluid intelligence was associated with each unit increase in hippocampal volume. However, when using the adjustment approach, the increase was 0.055 (95% CI: 0.043-0.068), while the proportional approach showed a decrease of -0.025 (95% CI: -0.035 to -0.014). Different commonly used methods to correct for ICV yielded inconsistent results. The proportional method diverges notably from other methods and results were sometimes biologically implausible. A simple regression adjustment for ICV produced biologically plausible associations.

Representativeness of Participants in the ACCORD Trial Compared to Middle-aged and Older Adults Living with Diabetes in the United States.

BACKGROUND: We evaluated whether participants in the landmark Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial represent US adults aged ≥40 with diabetes. METHODS: Using the nationally representative 2017-2020 prepandemic National Health and Nutrition Examination Survey data, we made operational definitions of ACCORD eligibility criteria. We calculated the percentage of individuals aged ≥40 with diabetes and HbA1c ≥ 6.0% or ≥ 7.5% who met operational ACCORD eligibility criteria. RESULTS: Applying survey sampling weights to 715 National Health and Nutrition Examination Survey participants aged ≥40 with diabetes and HbA1c ≥ 6.0% (representing 29,717,406 individuals), 12% (95% confidence interval [CI] = 8%, 18%) met the operational ACCORD eligibility criteria. Restricting to HbA1c ≥ 7.5%, 39% (95% CI = 28%, 51%) of respondents met the operational ACCORD eligibility criteria. CONCLUSIONS: ACCORD represented a minority of US middle-aged and older adults with diabetes. Given the differential risk profile between ACCORD participants and the general population with diabetes, extrapolating the trial findings may not be appropriate.

Overcoming data gaps in life course epidemiology by matching across cohorts.

Lifecourse epidemiology is hampered by the absence of large studies with exposures and outcomes measured at different life stages on the same individuals. We describe when the effect of an exposure (A) on an outcome (Y) in a target population is identifiable in a combined ("synthetic") cohort created by pooling an early-life cohort including measures of A with a late-life cohort including measures of Y. We enumerate causal assumptions needed for unbiased effect estimation in the synthetic cohort and illustrate by simulating target populations under four causal models. From each target population, we randomly sampled early- and late-life cohorts and created a synthetic cohort by matching individuals from the two cohorts based on mediators and confounders. We estimated the effect of A on Y in the synthetic cohort, varying matching variables, the match ratio, and the strength of association between matching variables and A. Finally, we compared bias in the synthetic cohort estimates when matching variables did not d-separate A and Y to the bias expected in the original cohort. When the set of matching variables includes all variables d-connecting exposure and outcome (i.e., variables blocking all back- and front-door pathways), the synthetic cohort yields unbiased effect estimates. Even when matching variables did not fully account for confounders, the synthetic cohort estimate was sometimes less biased than comparable estimates in the original cohort. Methods based on merging cohorts may hasten the evaluation of early- and mid-life determinants of late-life health but rely on available measures of both confounders and mediators.

The Association Between Physical Activity and Cognition in a Racially/Ethnically Diverse Cohort of Older Adults: Results From the Kaiser Healthy Aging and Diverse Life Experiences Study.

OBJECTIVE: Most prior research on physical activity (PA) and cognition is based on predominantly white cohorts and focused on associations of PA with mean (average) cognition versus the distribution of cognition. Quantile regression offers a novel way to quantify how PA affects cognition across the entire distribution. METHODS: The Kaiser Healthy Aging and Diverse Life Experiences study includes 30% white, 19% black, 25% Asian, and 26% Latinx adults age 65+ living in Northern California (n = 1600). The frequency of light or heavy PA was summarized as 2 continuous variables. Outcomes were z-scored executive function, semantic memory, and verbal episodic memory. We tested associations of PA with mean cognition using linear regression and used quantile regression to estimate the association of PA with the 10th-90th percentiles of cognitive scores. RESULTS: Higher levels of PA were associated with higher mean semantic memory (b = 0.10; 95% CI: 0.06, 0.14) and executive function (b = 0.05; 95% CI: 0.01, 0.09). Associations of PA across all 3 cognitive domains were stronger at low quantiles of cognition. CONCLUSION: PA is associated with cognition in this racially/ethnically diverse sample and may have larger benefits for individuals with low cognitive scores, who are most vulnerable to dementia.

Estimated effects of amyloid reduction on cognitive change: A Bayesian update across a range of priors.

INTRODUCTION: The results of the CLARITY-AD, GRADUATE I and II, and TRAILBLAZER-ALZ 2 trials have rekindled discussion on the impact of amyloid-targeting drugs. We use a Bayesian approach to quantify how rational observers would have updated their prior beliefs based on new trial results. METHODS: We used publicly available data from the CLARITY-AD, GRADUATE I and II, and TRAILBLAZER-ALZ 2 trials to estimate the effect of reducing amyloid on the clinical dementia rating scale, sum of boxes (CDR-SB) score. A range of prior positions were then updated according to Bayes' theorem using these estimates. RESULTS: After updating with new trial data, a wide range of starting positions resulted in credible intervals that did not include no effect of amyloid reduction on CDR-SB score. DISCUSSION: For a range of starting beliefs and assuming the veracity of the underlying data, rational observers would conclude there is a small benefit of amyloid reductions on cognition. This benefit must be weighed against opportunity cost and side-effect risk. HIGHLIGHTS: The results of recent trials of amyloid-targeting drugs have rekindled discussion on the impact of amyloid reductions achieved with amyloid-targeting drugs on cognition. Prior to the announcement of trial results, beliefs about the effects of altering amyloid levels varied. For a range of starting beliefs, one would conclude there is a small benefit of amyloid reductions due to amyloid-targeting drugs on cognition. The perceived value of individual drugs must balance the magnitude of this benefit against opportunity cost and risk of side effects.

Association of cancer history with structural brain aging markers of Alzheimer's disease and related dementias risk.

INTRODUCTION: Cancer survivors are less likely than comparably aged individuals without a cancer history to develop Alzheimer's disease and related dementias (ADRD). METHODS: In the UK Biobank, we investigated associations between cancer history and five structural magnetic resonance imaging (MRI) markers for ADRD risk, using linear mixed-effects models to assess differences in mean values and quantile regression to examine whether associations varied across the distribution of MRI markers. RESULTS: Cancer history was associated with smaller mean hippocampal volume (b = -19 mm3 , 95% CI = -36, -1) and lower mean cortical thickness in the Alzheimer's disease signature region (b = -0.004 mm, 95% CI = -0.007, -0.000). Quantile regressions indicated individuals most vulnerable to ADRD were more affected by cancer history. DISCUSSION: Some brain MRI markers associated with ADRD risk were elevated in adults with a history of cancer. The magnitude of the adverse associations varied across quantiles of neuroimaging markers, and the pattern suggests possible harmful associations for individuals already at high ADRD risk. HIGHLIGHTS: We found no evidence of an inverse association between cancer history and ADRD-related neurodegeneration. Cancer history was associated with smaller mean hippocampal volume and lower mean cortical thickness in the Alzheimer's disease signature region. Quantile regressions indicated individuals most vulnerable to ADRD were more affected by cancer history.

Timing and level of educational attainment and late-life cognition in the KHANDLE study.

INTRODUCTION: The timing of educational attainment may modify its effects on late-life cognition, yet most studies evaluate education only at a single time point. METHODS: Kaiser Healthy Aging and Diverse Life Experiences (KHANDLE) Study cohort participants (N = 554) reported educational attainment (dichotomized at any college education) at two time points, and we classified them as having low, high, or later-life high educational attainment. Linear mixed-effects models estimated associations between educational attainment change groups and domain-specific cognitive outcomes (z-standardized). RESULTS: Compared to low educational attainment, high (ß= 0.59 SD units; 95% confidence interval [CI]: 0.39, 0.79) and later-life high educational attainment (ß = 0.22; 95% CI: 0.00, 0.44) were associated with higher executive function. Only high educational attainment was associated with higher verbal episodic memory (ß = 0.27; 95% CI: 0.06, 0.48). DISCUSSION: Level and timing of educational attainment are both associated with domain-specific cognition. A single assessment for educational attainment may inadequately characterize protective associations with late-life cognition. HIGHLIGHTS: Few studies have examined both level and timing of educational attainment on cognition. Marginalized populations are more likely to attain higher education in adulthood. Higher educational attainment in late life is also associated with higher cognition.

Genetic risk score for Alzheimer's disease predicts brain volume differences in mid and late life in UK biobank participants.

INTRODUCTION: We estimated the ages when associations between Alzheimer's disease (AD) genes and brain volumes begin among middle-aged and older adults. METHODS: Among 45,616 dementia-free participants aged 45-80, linear regressions tested whether genetic risk score for AD (AD-GRS) had age-dependent associations with 38 regional brain magnetic resonance imaging volumes. Models were adjusted for sex, assessment center, genetic ancestry, and intracranial volume. RESULTS: AD-GRS modified the estimated effect of age (per decade) on the amygdala (-0.41 mm3 [-0.42, -0.40]); hippocampus (-0.45 mm3 [-0.45, -0.44]), nucleus accumbens (-0.55 mm3 [-0.56, -0.54]), thalamus (-0.38 mm3 [-0.39, -0.37]), and medial orbitofrontal cortex (-0.23 mm3 [-0.24, -0.22]). Trends began by age 45 for the nucleus accumbens and thalamus, 48 for the hippocampus, 51 for the amygdala, and 53 for the medial orbitofrontal cortex. An AD-GRS excluding apolipoprotein E (APOE) was additionally associated with entorhinal and middle temporal cortices. DISCUSSION: APOE and other genes that increase AD risk predict lower hippocampal and other brain volumes by middle age.

My parent, myself, or my child: whose education matters most for trajectories of cognitive aging in middle age?

The growing body of evidence linking intergenerational education and late-life cognitive decline is almost exclusively from high-income countries, despite rapid intergenerational changes in education in low- and middle-income countries (LMICs). We used data from the Mexican Health and Aging Study (n = 8,822), a cohort of Mexican adults aged > 50 years (2001- 2018) to evaluate whether parental (none vs. any formal schooling), own (< primary school vs. primary completion), or adult child (< high school vs. high school completion) education was associated with verbal memory z-scores. We used linear mixed models with inverse probability of attrition weights. Educational attainment in all three generations was associated with baseline verbal memory scores, independent of the prior generation's education. Lower parental (ß= -0.005; 95% CI: -0.009, -0.002) and respondents' education (ß= -0.013; 95% CI: -0.017, -0.010) were associated with faster decline in delayed (not immediate) verbal memory z-scores. Associations between adult child education and respondent's verbal memory decline varied by exposure specification. Educational attainment of parents and adult children may influence the cognitive aging of middle-aged and older adults in LMICs. These results have important implications given recent structural shifts in educational attainment in many LMICs.

Counterpoint: Preprints and the Future of Scientific Publishing-In Favor of Relevance.

Peer-reviewed journals provide an invaluable but inadequate vehicle for scientific communication. Preprints are now an essential complement to peer-reviewed publications. Eschewing preprints will slow scientific progress and reduce the public health impact of epidemiologic research. The coronavirus disease 2019 (COVID-19) pandemic highlighted long-standing limitations of the peer-review process. Preprint servers, such as bioRxiv and medRxiv, served as crucial venues to rapidly disseminate research and provide detailed backup to sound-bite science that is often communicated through the popular press or social media. The major criticisms of preprints arise from an unjustified optimism about peer review. Peer review provides highly imperfect sorting and curation of research and only modest improvements in research conduct or presentation for most individual papers. The advantages of peer review come at the expense of months to years of delay in sharing research methods or results. For time-sensitive evidence, these delays can lead to important missteps and ill-advised policies. Even with research that is not intrinsically urgent, preprints expedite debate, expand engagement, and accelerate progress. The risk that poor-quality papers will have undue influence because they are posted on a preprint server is low. If epidemiology aims to deliver evidence relevant for public health, we need to embrace strategic uses of preprint servers.

Comparison of Imputation Strategies for Incomplete Longitudinal Data in Life-Course Epidemiology.

Incomplete longitudinal data are common in life-course epidemiology and may induce bias leading to incorrect inference. Multiple imputation (MI) is increasingly preferred for handling missing data, but few studies explore MI-method performance and feasibility in real-data settings. We compared 3 MI methods using real data under 9 missing-data scenarios, representing combinations of 10%, 20%, and 30% missingness and missing completely at random, at random, and not at random. Using data from Health and Retirement Study (HRS) participants, we introduced record-level missingness to a sample of participants with complete data on depressive symptoms (1998-2008), mortality (2008-2018), and relevant covariates. We then imputed missing data using 3 MI methods (normal linear regression, predictive mean matching, variable-tailored specification), and fitted Cox proportional hazards models to estimate effects of 4 operationalizations of longitudinal depressive symptoms on mortality. We compared bias in hazard ratios, root mean square error, and computation time for each method. Bias was similar across MI methods, and results were consistent across operationalizations of the longitudinal exposure variable. However, our results suggest that predictive mean matching may be an appealing strategy for imputing life-course exposure data, given consistently low root mean square error, competitive computation times, and few implementation challenges.

Comparison of Mammography AI Algorithms with a Clinical Risk Model for 5-year Breast Cancer Risk Prediction: An Observational Study.

Background Although several clinical breast cancer risk models are used to guide screening and prevention, they have only moderate discrimination. Purpose To compare selected existing mammography artificial intelligence (AI) algorithms and the Breast Cancer Surveillance Consortium (BCSC) risk model for prediction of 5-year risk. Materials and Methods This retrospective case-cohort study included data in women with a negative screening mammographic examination (no visible evidence of cancer) in 2016, who were followed until 2021 at Kaiser Permanente Northern California. Women with prior breast cancer or a highly penetrant gene mutation were excluded. Of the 324 009 eligible women, a random subcohort was selected, regardless of cancer status, to which all additional patients with breast cancer were added. The index screening mammographic examination was used as input for five AI algorithms to generate continuous scores that were compared with the BCSC clinical risk score. Risk estimates for incident breast cancer 0 to 5 years after the initial mammographic examination were calculated using a time-dependent area under the receiver operating characteristic curve (AUC). Results The subcohort included 13 628 patients, of whom 193 had incident cancer. Incident cancers in eligible patients (additional 4391 of 324 009) were also included. For incident cancers at 0 to 5 years, the time-dependent AUC for BCSC was 0.61 (95% CI: 0.60, 0.62). AI algorithms had higher time-dependent AUCs than did BCSC, ranging from 0.63 to 0.67 (Bonferroni-adjusted P < .0016). Time-dependent AUCs for combined BCSC and AI models were slightly higher than AI alone (AI with BCSC time-dependent AUC range, 0.66-0.68; Bonferroni-adjusted P < .0016). Conclusion When using a negative screening examination, AI algorithms performed better than the BCSC risk model for predicting breast cancer risk at 0 to 5 years. Combined AI and BCSC models further improved prediction. © RSNA, 2023 Supplemental material is available for this article.

Propensity Scores in Health Disparities Research: The Example of Cognitive Aging and the Hispanic Paradox.

BACKGROUND: Individuals of Mexican ancestry in the United States experience substantial socioeconomic disadvantages compared with non-Hispanic white individuals; however, some studies show these groups have similar dementia risk. Evaluating whether migration selection factors (e.g., education) associated with risk of Alzheimer disease and related dementia (ADRD) explain this paradoxical finding presents statistical challenges. Intercorrelation of risk factors, common with social determinants, could make certain covariate patterns very likely or unlikely to occur for particular groups, which complicates their comparison. Propensity score (PS) methods could be leveraged here to diagnose nonoverlap and help balance exposure groups. METHODS: We compare conventional and PS-based methods to examine differences in cognitive trajectories between foreign-born Mexican American, US-born Mexican American, and US-born non-Hispanic white individuals in the Health and Retirement Study (1994-2018). We examined cognition using a global measure. We estimated trajectories of cognitive decline from linear mixed models adjusted for migration selection factors also associated with ADRD risk conventionally or with inverse probability weighting. We also employed PS trimming and match weighting. RESULTS: In the full sample, where PS overlap was poor, unadjusted analyses showed both Mexican ancestry groups had worse baseline cognitive scores but similar or slower rates of decline compared with non-Hispanic white adults; adjusted findings were similar, regardless of method. Focusing analyses on populations where PS overlap was improved (PS trimming and match weighting) did not alter conclusions. CONCLUSIONS: Attempting to equalize groups on migration selection and ADRD risk factors did not explain paradoxical findings for Mexican ancestry groups in our study.

Regression discontinuity design to evaluate the effect of statins on myocardial infarction in electronic health records.

Regression discontinuity design (RDD) is a quasi-experimental method intended for causal inference in observational settings. While RDD is gaining popularity in clinical studies, there are limited real-world studies examining the performance on estimating known trial casual effects. The goal of this paper is to estimate the effect of statins on myocardial infarction (MI) using RDD and compare with propensity score matching and Cox regression. For the RDD, we leveraged a 2008 UK guideline that recommends statins if a patient's 10-year cardiovascular disease (CVD) risk score > 20%. We used UK electronic health record data from the Health Improvement Network on 49,242 patients aged 65 + in 2008-2011 (baseline) without a history of CVD and no statin use in the two years prior to the CVD risk score assessment. Both the regression discontinuity (n = 19,432) and the propensity score matched populations (n = 24,814) demonstrated good balance of confounders. Using RDD, the adjusted point estimate for statins on MI was in the protective direction and similar to the statin effect observed in clinical trials, although the confidence interval included the null (HR = 0.8, 95% CI 0.4, 1.4). Conversely, the adjusted estimates using propensity score matching and Cox regression remained in the harmful direction: HR = 2.42 (95% CI 1.96, 2.99) and 2.51 (2.12, 2.97). RDD appeared superior to other methods in replicating the known protective effect of statins with MI, although precision was poor. Our findings suggest that, when used appropriately, RDD can expand the scope of clinical investigations aimed at causal inference by leveraging treatment rules from everyday clinical practice.