A Bayesian semiparametric approach for inference on the population partly conditional mean from longitudinal data with dropout

Five-year changes in episodic and semantic memory were examined in a sample of 829 participants (35-80 years). A cohort-matched sample (N=967) was assessed to control for practice effects. For episodic memory, cross-sectional analyses indicated gradual age-related decrements, whereas the longitudinal data revealed no decrements before age 60, even when practice effects were adjusted for. Longitudinally, semantic memory showed minor increments until age 55, with smaller decrements in old age as compared with episodic memory. Cohort differences in educational attainment appear to account for the discrepancies between cross-sectional and longitudinal data. Collectively, the results show that age trajectories for episodic and semantic memory differ and underscore the need to control for cohort and retest effects in cross-sectional and longitudinal studies, respectively.

BackgroundRepeated cognitive testing can boost performance due to practice effects (PEs). Follow‐up scores are rarely adjusted for PEs, but such correction can be highly important as it has been shown to result in earlier detection of MCI and increased validity based on concordance with Alzheimer's biomarkers. However, it remains unclear to what extent PEs persist across multiple follow‐ups and long durations, which are further complicated by normative age‐related decline. We examined PEs across 17 years from midlife to old age in a nonclinical sample.MethodMen (N = 1,608) in the longitudinal Vietnam Era Twin Study of Aging (VETSA) completed neuropsychological batteries over 4 waves: wave 1(N = 1,290, mean age=56); wave 2, mean ages=62); wave 3 (mean age=68); wave 4 (mean age=73). Waves 2 and 3 also included age‐matched attrition replacements. By comparing returnees’ performance to replacements being assessed for the first time, we estimated PEs for 30 tests at each wave using generalized estimating equations, accounting for normative age effects and differential patterns of missingness. We calculated unadjusted and PE‐adjusted cognitive scores (episodic memory, executive function, fluency, processing speed, visual memory, and visuospatial). At each wave we compared MCI prevalence based on unadjusted versus PE‐adjusted scores.ResultAmong returnees completing all 4 assessments, we found significant PEs for 11 tests at wave 2, 8 tests at wave 3, and 5 tests at wave 4. PEs were most apparent among episodic and visual memory tests. PE‐adjusted cognitive factor scores were significantly lower than unadjusted factor scores at all follow‐ups for all domains except fluency. MCI prevalence increased up to 20% after PE adjustment (from 14% to 18% at wave 4), indicating earlier detection.ConclusionPEs persist across multiple assessments and decades, even with long testing intervals. Importantly, even very small PEs can shift scores below the impairment threshold, resulting in earlier detection and diagnosis of MCI. Additionally, we previously showed that PE‐adjusted MCI diagnosis increases accuracy based on biomarker concordance and would dramatically reduce costs and length of clinical trials by reducing necessary sample size and increasing power. These results underscore the importance of accounting for PEs in longitudinal studies.

Background: Tumor fraction (TF) in cell-free DNA (cfDNA) is a quantitative indicator of circulating tumor DNA (ctDNA) and has become an important biomarker for disease monitoring and tracking therapeutic responses. We developed a longitudinal TF estimation method that does not require matched normal control samples and analytically validated its ability to accurately estimate TF when longitudinal blood samples are available. Methods: Both reference materials and clinical blood samples were used in the analytical validation. We first determined the limit of detection (LoD) of TF for our PredicineCARE cfDNA assay at targeted sequencing depth of 20, 000x using serially diluted reference materials with 12 gradient values from 0 to 4% TF. We further used a sequentially diluted clinical sample (with known TF) with healthy donor cfDNA to a gradient of expected TF values. We evaluated the variant calling and TF calculation algorithm by treating each dilution separately or treating the immediately nearby samples as longitudinal samples. We further evaluated our longitudinal-aware analysis method on 20 clinical subjects tested with PredicineCARE cfDNA assay with ≥2 time points to evaluate the improvement in TF estimation compared to individual estimation. Results: With the titrated reference materials and manufacturer-labelled somatic variants, we achieved an LoD of 0.01% TF for 30ng cfDNA input and a sequencing depth at 20, 000x using PredicineCARE cfDNA assay. When approaching the LoD, the number of reads for the somatic variants becomes more stochastic, which resulted in big variances in the calculated TF values. When independently estimating the TF without referring to matched normal or longitudinal samples, the LoD was 0.5% TF. When the longitudinal sample information was included, the LoD went down to 0.01% TF, contingent on the number of trackable variants. Our simulated results showed that when there are 4 or more tracking variants between longitudinal samples, the LoD can go down to 0.01% TF most of the time. We further applied this longitudinal sample aware dynamic TF estimation method to 20 clinical subjects tested with PredicineCARE cfDNA assay and with ≥2 time points, most showed improved TF estimation with ≥50% of the subjects having questionable somatic/germline calls if called separately. Conclusions: We developed a dynamic TF estimation methodology that leverages longitudinal samples to improve somatic/germline variant calling and exclude CHIP mutations. Our results demonstrated that with longitudinal samples, the accuracy in TF estimation can approach the ideal situation where all somatic/germline status of the variants are known. Notably, longer interval between time points and a higher number of tracking somatic mutations (≥2) in longitudinal samples generally enhance the effectiveness of this dynamic estimation methodology. Citation Format: Junmei Wang, Wanshu He, Lisha Zhu, Yong Huang, Xiaohong Wang, Shidong Jia, Pan Du. Development and analytical validation of a highly sensitive tumor fraction estimation method using blood samples without matched normal controls [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 5895.

Longitudinal data, comprising repeated measurements of the same individuals over time, arise frequently in cardiology and the biomedical sciences in general. For example, Frison and Pocock1 used repeated measurements of the liver enzyme creatine kinase in serum of cardiac patients to study changes in liver function over a 12-month study period. The main goal, indeed the raison d’etre , of a longitudinal study is characterization of changes in the response of interest over time. Ordinarily, changes in the response are also related to selected covariates. For example, Frison and Pocock1 compared changes in creatine kinase between patients randomized to active drug and placebo. The past 25 years have witnessed remarkable developments in statistical methods for the analysis of longitudinal data. Despite these important advances, researchers in the biomedical sciences have been somewhat slow to adopt these methods and often rely on statistical techniques that fail to adequately account for longitudinal study designs. The goal of the present report is to provide an overview of some recently developed methods for longitudinal analyses that are more appropriate, with a focus on 2 methods for continuous responses: the analysis of response profiles and linear mixed-effects models. The analysis of response profiles is better suited to settings with a relatively small number of repeated measurements, obtained on a common set of occasions, whereas linear mixed-effects models are suitable in more general settings. Before describing these methods, we review some of the defining features of longitudinal studies and highlight the main aspects of longitudinal data that complicate their analysis. ### Covariance Structure A common feature of repeated measurements on an individual is correlation; that is, knowledge of the value of the response on one occasion provides information about the likely value of the response on a future occasion. Another common feature of longitudinal data is heterogeneous …

Immigrant Earnings Growth: Selection Bias or Real Progress?

Conducting longitudinal research has been encouraged among family scholars as a way to better understand family process and to measure changes in family structure and relationships over time. When doing longitudinal research, a common problem is the loss of sample members between the first wave of data collection and subsequent waves. Some subjects move between data points and cannot be located, whereas others refuse to continue their participation. Some members of the original sample, especially if it includes older adults, will have died or have become too disabled to participate further in the study. These losses represent an attrition of the original sample over time. These various forms of attrition of a sample result in a potential threat of bias if those who drop out have unique characteristics such that the remaining sample ceases to be representative of the original sample. This bias, known as attrition bias, is assumed to exist when there are significant differences between the initial and later samples. Because the loss of sample members between the first wave of data collection and subsequent waves is relatively common, researchers in longitudinal studies should check for attrition bias. Unchecked bias may be the major threat to longitudinal research (Markides, Dickson, & Pappas, 1982; Norris, 1985). Larzelere and Klein (1987) listed sample attrition as one of the three major problems that researchers face in analyzing longitudinal data. Despite the threat of attrition bias, however, it has received little attention in longitudinal family research. Indeed, few family researchers address this issue when reporting the results of their analyses of longitudinal data. There are two ways attrition can bias a sample. Firs, it may alter the characteristics of the sample, making it no longer representative of the original sample. Because the results of later waves of data are not generalizable to the original population that was sampled, the external validity of the study is threatened. For example, a longitudinal study examining the effects of divorce on the mental health of former spouses may fail to retain those subjects who have become too depressed to respond to the questionnaire. The nonparticipation of this subsample could bias the findings towards a minimization of depression as an outcome of divorce. The second way that selective attrition can bias longitudinal data is by altering the covariance of variables (Goudy, 1985; Norris, 1987). This problem occurs when the underrepresentation of some groups in the longitudinal sample leads to correlations between variables that are different than the true correlations in the original sample. Referring back to the previous example, the underrepresentation of depressed people in the second wave of the study may distort the statistical relation between length of marriage and depression. This poses a potentially severe threat to theory building because of the effect of these distortions on internal validity. TESTING FOR ATTRITION BIAS The most common method for detecting attrition bias in the characteristics of the sample is to use t tests to compare those subjects who responded to all waves of the study with those who dropped out after only one wave. The means of important variables from the first wave, such as socioeconomic status (SES), age, marital satisfaction, and depression, are compared between the two samples to determine if the differences are statistically significantly different. Significant differences in the means of one or more variables indicates the existence of attrition bias. Differences in categorical variables, such as race and marital status, can be determined by using the chi-square statistic. Logit analysis is a second method for detecting differences in characteristics of the sample. A logit equation is developed to estimate the probability that each first wave respondent will participate in later waves. …

Predicting medium-term survival after admission is necessary for identifying end-of-life patients who may benefit from goals of care (GOC) discussions. Considering that several patients have multiple hospital admissions, this study leverages patients' longitudinal data and information collected routinely at admission to predict the Hospital One-year Mortality Risk. We propose the Ensemble Longitudinal Network (ELN) to predict one-year mortality using patients' longitudinal records. The model was evaluated: (i) with only predictors reported upon admission (AdmDemo); and (ii) also with diagnoses available later during patients' stay (AdmDemoDx). Using records of 123,646 patients with 250,812 hospitalizations from 2011 to 2021, our dataset was split into a learning set (2011-2017) to compare models with and without longitudinal information using nested cross-validation, and a holdout set (2017-2021) to assess clinical utility towards GOC discussions. The ELN achieved a significant increase in predictive performance using longitudinal information (p-value < 0.05) for both the AdmDemo and AdmDemoDx predictors. For randomly selected hospitalizations in the holdout set, the ELN showed: (i) AUROCs of 0.83 (AdmDemo) and 0.87 (AdmDemoDx); and (ii) superior decision-making properties, notably with an increase in precision from 0.25 for the standard process to 0.28 (AdmDemo) and 0.36 (AdmDemoDx). Feature importance analysis confirmed that the utility of the longitudinal information increases with the number of patient hospitalizations. Integrating patients' longitudinal data provides better insights into the severity of illness and the overall patient condition, in particular when limited information is available during their stay.

In this chapter, our focus is on panel conditioning with respect to attitude questions. Our methodological approach is different from the majority of previous studies in this area, in that we do not attempt to estimate biases in marginal and associational distributions through comparison with a fresh cross-sectional sample. Rather, our approach is based on testing hypotheses on a single dataset, derived from an explicit theoretical model of the psychological mechanism underlying conditioning effects in repeated measures of the attitude. We refer to this theoretical account as the cognitive stimulus (CS) model. Specifically, we use a range of empirical indicators to evaluate the theory that repeatedly administering attitude questions serves to stimulate respondents to reflect and deliberate more closely on the issues to which the questions pertain. This, in turn, results in stronger and more internally consistent attitudes in the later waves of a panel. The chapter proceeds in the following manner. First, we review the existing literature on panel conditioning effects. Next, we set out in more detail the rationale underlying the CS hypothesis.We then use data from the first ten waves of the British Household Panel Study (BHPS) to test four inter-related hypotheses expressed as empirical expectations of the CS model. We conclude with a discussion of the implications of our findings for the validity of attitude measures in panel surveys.

This study of the general population examined interactions of the gene Apolipoprotein E (APOE) and/or lipid levels, and their effects on cognitive change. A MANCOVA model based on longitudinal data (with a 5 year follow-up) obtained from the Betula study (n = 1777; age 35–85 years) was used. The significant two-way and three-way interaction effects detected were equally frequent in tests of episodic and semantic memory. A difference in the distribution of interaction effects on episodic and semantic memory decline was also found. Men demonstrated the worst cognitive development as shown by significant two-way interaction effects on episodic memory whereas two-way interaction effects among women resulted in the worst semantic memory development. This result is discussed from the viewpoint that tests of episodic and semantic memory have different cognitive demands.This study focuses on how interaction effects of the gene APOE and vascular risk factors (such as lipid levels) affect cognitive abilities and also whether the interaction effects vary across age and sex. In this study, the main focus is on interaction effects as a phenomenon in itself.

BackgroundNeural changes occur early in Alzheimer’s disease (AD). Longitudinal data on the relationship between AD pathology and brain activity, however, are sparse. Using an established cohort of cognitively unimpaired older adults (PREVENT‐AD) at risk of AD (due to familial incidence), we assessed the relationship between functional magnetic resonance imaging (fMRI) activity during episodic memory retrieval and cerebrospinal fluid (CSF) markers of tau and β‐amyloid. We aimed to test whether baseline fMRI signal predicts changes in CSF markers and vice versa.MethodThree‐hundred and three participants (64.7±8.3yrs, 82 male) underwent task‐based fMRI of associative episodic memory retrieval at baseline, with longitudinal fMRI data for up to five timepoints with a 3, 12, 24, and 48‐month follow‐up. Region of Interest (ROI) analysis assessed retrieval activity (Fig.1A; t‐contrast differentiating hits for old stimuli from the correct rejection of new stimuli) in the bilateral precuneus and medial temporal lobe (MTL, Fig.1B‐1C). In a subset of participants (n = 117, 30 male), CSF was sampled to quantify t‐tau, p‐tau 181, and the ratio of β‐amyloid 1‐42 to p‐tau 181 (with another measurement 36‐months post‐baseline). Data were analyzed with random‐intercept linear mixed models, assessing changes in ROI activity and CSF markers over time. We tested whether baseline ROI activity predicted change in CSF markers and whether baseline CSF measures predicted change in ROI activity over time.ResultResults indicate a significant decrease in precuneus activity over time, but not MTL, and an increase in t‐tau and p‐tau 181 (Fig.2A‐2D). No change was observed for β‐amyloid/p‐tau 181 ratio. We also observed a significant time by baseline MTL activity interaction on t‐tau (Fig.3) with higher baseline activity being associated with lower t‐tau in CSF over time. We observed no relationship between baseline CSF markers and functional activity.ConclusionWhile preliminary, these results suggest that baseline MTL retrieval activity may predict t‐tau (a marker of neurodegeneration) accumulation in the CSF of healthy adults at risk of AD. Further analyses of whole‐brain episodic memory responses, episodic memory task‐performance over time, and of positron emission tomography (PET) measures of regional tau and β‐amyloid, are critical to extend these findings further.

Intelligence and selective attrition in a nationally representative and longitudinal sample of Americans

Background:Repeated cognitive testing can boost scores due to practice effects (PEs), yet it remains unclear whether PEs persist across multiple follow-ups and long durations. We examined PEs across multiple assessments from midlife to old age in a nonclinical sample.Method:Men (N=1,608) in the Vietnam Era Twin Study of Aging (VETSA) underwent neuropsychological assessment comprising 30 measures across 4 waves (~6-year testing intervals) spanning up to 20 years. We leveraged age-matched replacement participants to estimate PEs at each wave. We compared cognitive trajectories and MCI prevalence using unadjusted versus PE-adjusted scores.Results:Across follow-ups, a range of 7–12 tests (out of 30) demonstrated significant PEs, especially in episodic memory and visuospatial domains. Adjusting for PEs resulted in improved detection of cognitive decline and MCI, with up to 20% higher MCI prevalence.Conclusion:PEs persist across multiple assessments and decades underscoring the importance of accounting for PEs in longitudinal studies.

Objectives: This randomized controlled trial aims to investigate whether a single dose or six-week consumption of polyphenol-rich dealcoholized muscadine wine (DMW) impact cognitive performance and mood among healthy adults.Methods: Fourteen healthy participants, aged 50 to 65, were randomly assigned to consume 300 mL of DMW or placebo daily for six weeks using a crossover design separated by a 3-week washout.Cognitive performance and mood were assessed at baseline, 4 hours after the first dose of beverage consumption, and at the end of each 6-week phase.Attention, inhibitory control, cognitive flexibility, episodic memory, working memory, and processing speed were assessed using the NIH Toolbox and the Rey Auditory Verbal Learning Test.Mood was assessed using the NIH Toolbox and Beck's Depression Inventory.Plasma content of cortisol, TNF-, BDNF, CRP, and LBP were quantified using ELISA kits.Linear mixed models, with participants as a random effect, were used to analyze cognitive and mood outcomes, as well as plasma biomarker levels.Practice effects were assessed by the main effect of the period, while carry-over effects were examined via the period beverage interaction.Response variables that did not meet normality assumptions were log-transformed before analysis.Statistical significance was set at p 0.05.Results: DMW did not influence cognitive performance, mood, or biomarkers of stress and inflammation compared to placebo.Significant carry-over and practice effects were observed in cognitive assessments that measured verbal learning, working memory, episodic memory, and cognitive flexibility.A declining trend in cognitive performance was observed during the acute afternoon visits compared to morning visits in both DMW and placebo groups.This reduction in verbal learning, processing speed, and episodic memory in the afternoon coincided with lower plasma cortisol concentration, suggesting fatigue may be a primary contributing factor.Conclusions: One-dose or six-week consumption of DMW did not improve cognition or mood among healthy adults aged 50-65.However, the significant practice and carry-over effect could mask the benefit of DMW in this healthy population.

Objectives Episodic memory is typically declining in older age but some Superagers have high performance levels. A superager level of performance may be preceded by different trajectories. The purpose of this study was to use longitudinal data to assign Superagers to stable or declining trajectories and then consider variation in memory trajectories in analyses of predictors of Superaging. Methods In the longitudinal Betula study we identified 139 Superagers (Mean age=79 years) with as good or higher episodic memory as the average of 300 50–60 years old individuals. Episodic-memory trajectories of 125 Superagers were defined from up to 25 years of longitudinal data. The Betula database provided information on possible predictors of Superaging, including education, cognition, polygenic scores, health, lifestyle, and structural brain integrity. Results The majority of Superagers were on stable trajectories from initially high to average levels, but some were on declining trajectories from high levels. Similar longitudinal profiles were seen on a word-fluency task. Analyses of predictors of Superaging revealed that education and a polygenic score for cognition were related to initial memory level, and that a polygenic score for dementia was related to rate of change. Most but not all superagers with high-stable memory had favorable education and polygenic predictor scores, suggesting alternative pathways to Superaging. Longitudinal imaging data revealed less atrophy in entorhinal cortex and hippocampus for Superagers. Conclusion A superager performance level can reflect well-maintained memory or decline from high initial levels, with distinct factors explaining variability in initial level or stability over time. Medial-temporal lobe brain maintenance characterizes episodic-memory Superagers.

Objective:Practice effects (PE) on cognitive testing impede our ability to accurately assess change. In particular, they hamper the detection of mild cognitive impairment (MCI) and progression to dementia by delaying the point at which test scores fall below diagnostic impairment cutoffs. When decline over time is expected, as with older adults or progressive diseases, failure to adequately address PEs may lead to inaccurate conclusions because PEs artificially boost scores while pathology-related or age-related decline reduces scores. The participant-replacement method accounts for PEs by comparing performance of demographically-matched replacement participants to returnees who have been tested previously. Unlike most methods, the participant-replacement method can separate pathology- or age-related decline from PEs; however, this method has only been used across two timepoints. Neuropsychologists tend to think that PEs level out after the first follow-up, but this issue has not been evaluated in models that allow PEs in the presence of overall decline. Including more than two timepoints makes it possible to determine if PEs level out after the first follow-up, but it is analytically challenging because individuals may not be assessed at every timepoint.Participants and Methods:We examined 1190 older adults in the Alzheimer’s Disease Neuroimaging Initiative who were cognitively unimpaired (n=809) or had MCI (n=381) at baseline. Participants completed six neuropsychological measures (Trails A, Trials B, Boston Naming Test, Category Fluency, Logical Memory, Rey Auditory Verbal Learning Task) at three timepoints (baseline, 12-month, 24-month). We implemented the participant-replacement method using generalized estimating equations in comparisons of matched returnees and replacements to calculate PEs. Propensity scores matched individuals on age, education, sex, and an estimate of premorbid functioning. Generalized estimating equations modeled PEs and age-related decline separately for each cognitive measure.Results:We observed significant PEs for 5 of the 6 measures in the cognitively unimpaired group and 4 measures in the baseline MCI group. PEs did not uniformly decrease across time; some—specifically on episodic memory measures—continued to increase beyond the first follow-up for both groups of participants. Without accounting for PEs, cognitive function appeared to improve or stay the same. In contrast, when PEs were included in the models, cognitive function appeared to decline or stay the same across time.Conclusions:The replacement method of PE adjustment revealed significant PEs across two follow-ups. PEs for episodic memory, in particular, did not level out, but actually increased after the first follow-up, two years after baseline. As expected in these older adults, accounting for PEs revealed cognitive decline, in some cases, even when PE-unadjusted scores improved. This method of assessing PEs, in turn, means earlier detection of cognitive deficits, including progression to MCI, and more accurate characterization of longitudinal change.