JoAnn E Manson’s research while affiliated with Harvard University and other places

Recent advancements in machine learning (ML) for analyzing heterogeneous treatment effects (HTE) are gaining prominence within the medical and epidemiological communities, offering potential breakthroughs in the realm of precision medicine by enabling the prediction of individual responses to treatments. This paper introduces the methodological frameworks used to study HTEs, particularly based on a single randomized controlled trial (RCT). We focus on methods to estimate conditional average treatment effect (CATE) for multiple covariates, aiming to predict individualized treatment effects. We explore a range of methodologies from basic frameworks like the T-learner, S-learner, and Causal Forest, to more advanced ones such as the DR-learner and R-learner, as well as cross-validation for CATE estimation to enhance statistical efficiency by estimating CATE for all RCT participants. We also provide a practical application of these approaches using the Preventing Overweight Using Novel Dietary Strategies (POUNDS Lost) trial, which compared the effects of high versus low-fat diet interventions on 2-year weight changes. We compared different sets of covariates for CATE estimation, showing that the DR- and R-learners are useful for the estimation of CATE in high-dimensional settings. This paper aims to explain the theoretical underpinnings and methodological nuances of ML-based HTE analysis without relying on technical jargon, making these concepts more accessible to the clinical and epidemiological research communities.

Background/Objectives: While previous study results have suggested an elevated risk of type 2 diabetes with potato consumption, limited and inconsistent results are available on the association of potato consumption with the risk of cardiovascular disease (CVD) and hypertension (HTN). We assessed the associations of (i) total potato consumption with the risk of CVD and HTN as the primary aim and (ii) fried potatoes and combined baked, boiled, and mashed potatoes with the risk of CVD and HTN as the secondary aim. Methods: We conducted a meta-analysis using data from seven cohorts for CVD (n = 110,063) and five cohorts for HTN (n = 67,146). Cox regression was used to estimate multivariable adjusted hazard ratios separately in each cohort and the cohort-specific results were meta-analyzed using an inverse-variance weighted method. Results: The mean age ranged from 25 to 72 years, 65% of the respondents were women, and the mean consumption of total potatoes ranged from 1.9 to 4.3 times per week. In the primary analysis, total potato intake was not associated with the risk of either CVD or HTN: multivariable adjusted HR (95% CI) comparing 5+ servings/week to no potato intake: 0.96 (0.89–1.04) for CVD and 1.04 (0.99–1.08) for HTN. In secondary analyses, the consumption of combined baked, boiled, and mashed potatoes was not associated with CVD or HTN; while fried potato consumption was not associated with CVD risk, there was a 10% higher risk of HTN (95% CI: 4% to 17%) comparing 1+ servings/week to no fried potato intake. Conclusions: While the consumption of total potato was not associated with the risk of CVD or HTN risk, a modest elevated risk of HTN but not CVD was observed only with fried potato consumption.

Background The importance of integrating physical and psychosocial factors in assessing frailty -health outcomes has been increasingly acknowledged, while the related evidence is lacking. We sought to investigate the associations of joint physical-psychosocial frailty with risk of premature mortality and evaluate the relative importance of individual physical and psychosocial factors. Design A total of 381,295 participants with no history of cancer or cardiovascular disease (CVD) were recruited from the UK Biobank cohort. The physical-psychosocial frailty was evaluated based on seven indicators including weight loss, exhaustion, physical activity, walking pace, grip strength, social isolation, and loneliness. The outcomes were premature mortality from all causes, cancer, CVD, and other causes. Cox proportional hazards models were used to assess the associations between the physical-psychosocial frailty and premature mortality. Key Results During a median follow-up period of 12.7 years, we recorded 20,328 premature deaths. Each additional increment in the physical-psychosocial frailty index was associated with a 26% (HR 1.26, 95% CI 1.24–1.28), 10% (HR 1.10, 95% CI 1.08–1.12), 30% (HR 1.30, 95% CI 1.26–1.33), and 44% (HR 1.44, 95% CI 1.41–1.47) higher risk of all-cause, cancer, cardiovascular, and other-cause premature mortality, respectively. Compared with participants with the physical-psychosocial frailty index of 0, those with the index ≥ 4 had a 2.67 (95% CI 2.49–2.87)-fold higher risk of all-cause premature mortality. Slow walking pace and social isolation were the top two strongest predictors for all-cause premature mortality. In addition, we found that lower body mass index (BMI), age, smoking status, and dietary quality modified the associations of physical-psychosocial frailty with all-cause premature mortality ( P -interaction < 0.05). Conclusions In this cohort study of UK Biobank participants, joint physical-psychosocial frailty is significantly associated with risks of all-cause and cause-specific premature mortality, highlighting the importance to jointly assess physical and psychosocial factors in determining aging-related health.

Background There is an urgent need to identify novel, accessible and affordable strategies to prevent cognitive decline and progression in the Alzheimer disease and related dementias (ADRD) continuum. Vitamin D3 and marine omega‐3 fatty acids (omega‐3s) supplements show promise for cognitive protection, with potential variations in their effects by sex or race. However, to date, no randomized clinical trials (RCTs) have tested their impact on emerging plasma‐based biomarkers with potential utility to predict ADRD pathogenesis. Methods Vitamin D and Omega‐3 Trial (VITAL) is a completed, nation‐wide, 2x2 factorial placebo‐controlled RCT testing vitamin D3 (2000 IU/d) and omega‐3s (1 g/d) for cancer and cardiovascular disease prevention. We included a Boston‐area sub‐cohort of 929 randomized VITAL participants who provided blood samples at baseline, 2‐year, and/or 4‐year follow‐up. Plasma ADRD biomarkers, including an N‐terminal tau fragment (NT1), amyloid‐β (Aβ)‐40, Aβ‐42, neurofilament‐light (NfL), and glial fibrillary acidic protein (GFAP), were measured. We used multivariable‐adjusted repeated measures models for data analysis; sex and race were pre‐specified effect modifiers. Results Among 929 participants, the mean age was 65 years; 49.2% were females; 17.3% were from racial and/or ethnic minority backgrounds, including 8.2% Black adults. Neither vitamin D3 nor omega‐3s, compared to placebo, significantly reduced ADRD biomarkers overall across 4 years of treatment; however, there was a trend for reduction in Aβ‐40:Aβ‐42 ratio over 4 years for vitamin D3 versus placebo [percent difference (95% confidence interval [CI]): ‐1.2 (‐2.7, 0.3)]. Subgroup analyses uncorrected for multiple‐testing suggested interactions by sex and race. Vitamin D3 versus placebo resulted in a reduction in NT1 among males (‐2.7%) but not females (p‐interaction=0.08). Among Black participants, vitamin D3 versus placebo resulted in a 16% reduction in NfL levels [95% CI, ‐30.5% to 1.1%; p‐interaction=0.06], while omega‐3s versus placebo showed a 12.4% reduction in GFAP levels [95% CI, ‐21.5% to ‐2.2%; p‐interaction=0.049]. Conclusion In this RCT sub‐cohort of 929 older adults, neither vitamin D3 nor omega‐3s supplements significantly reduced selected plasma ADRD biomarkers over 4 years. We observed potential differences by sex and race in reductions of some ADRD biomarkers in response to these supplements which warrant further investigation in a larger sample.

Importance Classification of persons with long COVID (LC) or post–COVID-19 condition must encompass the complexity and heterogeneity of the condition. Iterative refinement of the classification index for research is needed to incorporate newly available data as the field rapidly evolves. Objective To update the 2023 research index for adults with LC using additional participant data from the Researching COVID to Enhance Recovery (RECOVER-Adult) study and an expanded symptom list based on input from patient communities. Design, Setting, and Participants Prospective, observational cohort study including adults 18 years or older with or without known prior SARS-CoV-2 infection who were enrolled at 83 sites in the US and Puerto Rico. Included participants had at least 1 study visit taking place 4.5 months after first SARS-CoV-2 infection or later, and not within 30 days of a reinfection. The study visits took place between October 2021 and March 2024. Exposure SARS-CoV-2 infection. Main Outcomes and Measures Presence of LC and participant-reported symptoms. Results A total of 13 647 participants (11 743 with known SARS-CoV-2 infection and 1904 without known prior SARS-CoV-2 infection; median age, 45 years [IQR, 34-69 years]; and 73% were female) were included. Using the least absolute shrinkage and selection operator analysis regression approach from the 2023 model, symptoms contributing to the updated 2024 index included postexertional malaise, fatigue, brain fog, dizziness, palpitations, change in smell or taste, thirst, chronic cough, chest pain, shortness of breath, and sleep apnea. For the 2024 LC research index, the optimal threshold to identify participants with highly symptomatic LC was a score of 11 or greater. The 2024 index classified 20% of participants with known prior SARS-CoV-2 infection and 4% of those without known prior SARS-CoV-2 infection as having likely LC (vs 21% and 5%, respectively, using the 2023 index) and 39% of participants with known prior SARS-CoV-2 infection as having possible LC, which is a new category for the 2024 model. Cluster analysis identified 5 LC subtypes that tracked quality-of-life measures. Conclusions and Relevance The 2024 LC research index for adults builds on the 2023 index with additional data and symptoms to help researchers classify symptomatic LC and its symptom subtypes. Continued future refinement of the index will be needed as the understanding of LC evolves.

Aims/hypothesis Vitamin D insufficiency is associated with an elevated risk of type 2 diabetes, but evidence from randomised trials on the benefits of vitamin D supplementation is limited, especially for average-risk populations. The Finnish Vitamin D Trial (FIND) investigated the effects of vitamin D3 supplementation at two different doses on the incidence of type 2 diabetes in a generally healthy older adult population. Methods FIND was a 5 year randomised placebo-controlled, parallel-arm trial among 2271 male and female participants aged ≥60 years and ≥65 years, respectively, from a general Finnish population who were free of CVD or cancer and did not use diabetes medications. The study had three arms: placebo, 1600 IU/day of vitamin D3 or 3200 IU/day of vitamin D3. A non-study group statistician carried out sex-stratified simple randomisation in a 1:1:1 ratio, based on computerised random number generation. The participants, investigators and study staff were masked to group assignment. National health registries were used to collect event data. A representative subcohort of 505 participants had more detailed in-person investigations at months 0, 6, 12 and 24. Results During the mean follow-up of 4.2 years, there were 38 (5.0%), 31 (4.2%) and 36 (4.7%) type 2 diabetes events in the placebo (n=760), 1600 IU/day vitamin D3 (n=744; vs placebo: HR 0.81; 95% CI 0.50, 1.30) and 3200 IU/day vitamin D3 (n=767; vs placebo: HR 0.92, 95% CI 0.58, 1.45) arms, respectively (p-trend=0.73). When the two vitamin D3 arms were combined and compared with the placebo arm, the HR was 0.86 (95% CI 0.58, 1.29). In the analyses stratified by BMI (<25 kg/m² [n=813, number of type 2 diabetes events=12], 25–30 kg/m² [n=1032, number of events=38], ≥30 kg/m² [n=422, number of events=54]), the HRs in the combined vitamin D3 arms vs the placebo were 0.43 (95% CI 0.14, 1.34), 0.97 (0.50, 1.91) and 1.00 (0.57, 1.75), respectively (p-interaction <0.001). In the subcohort, the mean (SD) baseline serum 25-hydroxyvitamin D3 (25(OH)D3) concentration was 74.5 (18.1) nmol/l. After 12 months, the concentrations were 72.6 (17.7), 99.3 (20.8) and 120.9 (22.1) nmol/l in the placebo, 1600 IU/day vitamin D3 and 3200 IU/day vitamin D3 arms, respectively. In the subcohort, no differences were observed in changes in plasma glucose or insulin concentrations, BMI or waist circumference during the 24 month follow-up (p values ≥0.19). Conclusion/interpretation Among generally healthy older adults who are not at high risk for diabetes and who have serum 25(OH)D3 levels that are sufficient for bone health, vitamin D3 supplementation did not significantly reduce the risk of developing diabetes. Trial registration ClinicalTrials.gov NCT01463813. Graphical Abstract

Background Findings from Kronos Early Estrogen Prevention Study (KEEPS)-Cog trial suggested no cognitive benefit or harm after 48 months of menopausal hormone therapy (mHT) initiated within 3 years of final menstrual period. To clarify the long-term effects of mHT initiated in early postmenopause, the observational KEEPS Continuation Study reevaluated cognition, mood, and neuroimaging effects in participants enrolled in the KEEPS-Cog and its parent study the KEEPS approximately 10 years after trial completion. We hypothesized that women randomized to transdermal estradiol (tE2) during early postmenopause would show cognitive benefits, while oral conjugated equine estrogens (oCEE) would show no effect, compared to placebo over the 10 years following randomization in the KEEPS trial. Methods and findings The KEEPS-Cog (2005–2008) was an ancillary study to the KEEPS (NCT00154180), in which participants were randomized into 3 groups: oCEE (Premarin, 0.45 mg/d), tE2 (Climara, 50 μg/d) both with micronized progesterone (Prometrium, 200 mg/d for 12 d/mo) or placebo pills and patch for 48 months. KEEPS Continuation (2017–2022), an observational, longitudinal cohort study of KEEPS clinical trial, involved recontacting KEEPS participants approximately 10 years after the completion of the 4-year clinical trial to attend in-person research visits. Seven of the original 9 sites participated in the KEEPS Continuation, resulting in 622 women of original 727 being invited to return for a visit, with 299 enrolling across the 7 sites. KEEPS Continuation participants repeated the original KEEPS-Cog test battery which was analyzed using 4 cognitive factor scores and a global cognitive score. Cognitive data from both KEEPS and KEEPS Continuation were available for 275 participants. Latent growth models (LGMs) assessed whether baseline cognition and cognitive changes during KEEPS predicted cognitive performance at follow-up, and whether mHT randomization modified these relationships, adjusting for covariates. Similar health characteristics were observed at KEEPS randomization for KEEPS Continuation participants and nonparticipants (i.e., women not returning for the KEEPS Continuation). The LGM revealed significant associations between intercepts and slopes for cognitive performance across almost all domains, indicating that cognitive factor scores changed over time. Tests assessing the effects of mHT allocation on cognitive slopes during the KEEPS and across all years of follow-up including the KEEPS Continuation visit were all statistically nonsignificant. The KEEPS Continuation study found no long-term cognitive effects of mHT, with baseline cognition and changes during KEEPS being the strongest predictors of later performance. Cross-sectional comparisons confirmed that participants assigned to mHT in KEEPS (oCEE and tE2 groups) performed similarly on cognitive measures to those randomized to placebo, approximately 10 years after completion of the randomized treatments. These findings suggest that mHT poses no long-term cognitive harm; conversely, it provides no cognitive benefit or protective effects against cognitive decline. Conclusions In these KEEPS Continuation analyses, there were no long-term cognitive effects of short-term exposure to mHT started in early menopause versus placebo. These data provide reassurance about the long-term neurocognitive safety of mHT for symptom management in healthy, recently postmenopausal women, while also suggesting that mHT does not improve or preserve cognitive function in this population.

Background Obesity, defined by body mass index (BMI) alone, is a metabolically heterogeneous disorder with distinct cardiovascular manifestations across individuals. This study aimed to investigate the associations of a proteomic signature of BMI with risk of major subtypes of cardiovascular disease (CVD). Methods A total of 40 089 participants from UK Biobank, free of CVD at baseline, had complete data on proteomic data measured by the Olink assay. A BMI-proteomic score (pro-BMI score) was calculated from 67 pre-identified plasma proteins associated with BMI. Results A higher pro-BMI score was significantly associated with higher risks of ischemic heart disease (IHD) and heart failure (HF), but not with risk of stroke. Comparing the highest with the lowest quartiles, the adjusted hazard ratio (HR) for IHD was 1.49 (95% CI, 1.32–1.67) (P-trend < 0.001), and the adjusted HR for HF was 1.52 (95% CI, 1.25–1.85) (P-trend < 0.001). Further analyses showed that the association of pro-BMI score with HF risk was largely driven by the actual BMI, whereas the association of the pro-BMI score with IHD risk was independent of actual BMI and waist-to-hip ratio (WHR). The association between pro-BMI score and IHD risk appeared to be stronger in the normal BMI group than other BMI groups (P-interaction = 0.004) and stronger in the normal WHR group than the high WHR group (P-interaction = 0.049). Conclusions Higher pro-BMI score is significantly associated with higher IHD risk, independent of actual BMI levels. Our findings suggest that plasma proteins hold promise as complementary markers for diagnosing obesity and may facilitate personalized interventions.

Whole genome sequences (WGS) enable discovery of rare variants which may contribute to missing heritability of coronary artery disease (CAD). To measure their contribution, we apply the GREML-LDMS-I approach to WGS of 4949 cases and 17,494 controls of European ancestry from the NHLBI TOPMed program. We estimate CAD heritability at 34.3% assuming a prevalence of 8.2%. Ultra-rare (minor allele frequency ≤ 0.1%) variants with low linkage disequilibrium (LD) score contribute ~50% of the heritability. We also investigate CAD heritability enrichment using a diverse set of functional annotations: i) constraint; ii) predicted protein-altering impact; iii) cis-regulatory elements from a cell-specific chromatin atlas of the human coronary; and iv) annotation principal components representing a wide range of functional processes. We observe marked enrichment of CAD heritability for most functional annotations. These results reveal the predominant role of ultra-rare variants in low LD on the heritability of CAD. Moreover, they highlight several functional processes including cell type-specific regulatory mechanisms as key drivers of CAD genetic risk.

Disclosure: E.R. Uddenberg: None. N. Safwan: None. G. Cook-Wiens: None. O. Obrutu: None. M. Saadedine: None. M.D. Hurtado: None. S. Faubion: None. M.D. Pisarska: None. S.L. Berga: None. J.E. Manson: None. C.N. Bairey Merz: None. C.L. Shufelt: None. Background: Functional hypothalamic amenorrhea (FHA) occurs in women of reproductive age and is a common form of secondary amenorrhea that results in hypoestrogenemia and hypercortisolemia. The underlying cause of FHA is due to a mismatch between energetic intake and expenditure that often manifests as disordered eating, excessive exercise, and psychosocial stress. Previous research identified that 1 in 3 women with FHA had endothelial dysfunction, a preclinical marker for cardiovascular disease. We sought to evaluate the nutrition content in young women with FHA compared to eumenorrheic controls and further assess the relationship to endothelial dysfunction. Methods: We enrolled 30 women with FHA and 29 eumenorrheic controls, not on hormones. FHA was defined as ≥3 consecutive months of amenorrhea, estradiol <50 pg/mL, FSH<10 mIU/L, and LH<10 mIU/L, and exclusion of polycystic ovary syndrome, hyperprolactinemia, thyroid dysfunction, and pregnancy. Eumenorrheic controls had monthly menses with ovulation (progesterone ≥4 ng/mL). Vascular function was measured using EndoPAT 2000 (Itamar® Medical Ltd) to calculate the reactive hyperemic index (RHI). An RHI ≤1.67 indicates endothelial dysfunction. Nutritional intake was collected using a 3-day food diary and analyzed using ESHA Research Food Processor® Nutrition Analysis. Percent (%) recommendation values for nutrition content comparison were calculated based on height, weight, and age. Statistical analysis included Analysis of Variance and Kruskal Wallis test with nonparametric variables reported as median [IQR]. Results: The mean age and BMI of FHA and controls were 26.4 ± 6.2 yrs and 30.3 ± 3.7 yrs (p=0.008), and 21.6 ± 6.2 and 21.8 ± 2.0 kg/m2 (p=0.16), respectively. The median months of FHA amenorrhea was 12 months [5.0, 34.0]. After adjusting for age, there was no difference in caloric intakes in women with FHA compared to controls (1783.9 [1465, 2011] kcal vs. 1731.9 [1578, 1982] kcal, p=0.73); however FHA has higher intake of grams of protein (84.2 [74, 112] g vs. 74 [59, 92] g, p=0.0036), and higher grams of dietary fiber (30.1 [22, 43] g vs. 17.3 [15, 23] g, p=0.0037), respectively. In women with FHA, endothelial dysfunction was associated with a higher % recommended calories (p=0.0098), calories from fat (p=0.02), grams of saturated fat (p=0.03), and grams of carbohydrates (p=0.049). Conclusions: Our results demonstrate that women with FHA consumed similar daily calories, although more grams of protein and dietary fiber than controls. In women with FHA, higher calories, calories from fat, grams of saturated fat and carbohydrates were associated with endothelial dysfunction. Future studies should aim to identify other factors related to endothelial dysfunction among women with FHA and whether some FHA phenotypes are associated with more endothelial dysfunction than others. Presentation: 6/1/2024