Performance of Linear Mixed Models in Estimating Structural Rates of Glaucoma Progression Using Varied Random Effect Distributions.

Purpose: To compare how linear mixed models (LMMs) using Gaussian, Student t, and log-gamma (LG) random effect distributions estimate rates of structural loss in a glaucomatous population using OCT and to compare model performance to ordinary least squares (OLS) regression. Design: Retrospective cohort study. Subjects: Patients in the Bascom Palmer Glaucoma Repository (BPGR). Methods: Eyes with ≥ 5 reliable peripapillary retinal nerve fiber layer (RNFL) OCT tests over ≥ 2 years were identified from the BPGR. Retinal nerve fiber layer thickness values from each reliable test (signal strength ≥ 7/10) and associated time points were collected. Data were modeled using OLS regression as well as LMMs using different random effect distributions. Predictive modeling involved constructing LMMs with (n - 1) tests to predict the RNFL thickness of subsequent tests. A total of 1200 simulated eyes of different baseline RNFL thickness values and progression rates were developed to evaluate the likelihood of declared progression and predicted rates. Main Outcome Measures: Model fit assessed by Watanabe-Akaike information criterion (WAIC) and mean absolute error (MAE) when predicting future RNFL thickness values; log-rank test and median time to progression with simulated eyes. Results: A total of 35 862 OCT scans from 5766 eyes of 3491 subjects were included. The mean follow-up period was 7.0 ± 2.3 years, with an average of 6.2 ± 1.4 tests per eye. The Student t model produced the lowest WAIC. In predictive models, all LMMs demonstrated a significant reduction in MAE when estimating future RNFL thickness values compared with OLS (P < 0.001). Gaussian and Student t models were similar and significantly better than the LG model in estimating future RNFL thickness values (P < 0.001). Simulated eyes confirmed LMM performance in declaring progression sooner than OLS regression among moderate and fast progressors (P < 0.01). Conclusions: LMMs outperformed conventional approaches for estimating rates of OCT RNFL thickness loss in a glaucomatous population. The Student t model provides the best model fit for estimating rates of change in RNFL thickness, although the use of the Gaussian or Student t distribution in models led to similar improvements in accurately estimating RNFL loss. Financial Disclosures: Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

"Back to the future" projections for COVID-19 surges.

We argue that information from countries who had earlier COVID-19 surges can be used to inform another country's current model, then generating what we call back-to-the-future (BTF) projections. We show that these projections can be used to accurately predict future COVID-19 surges prior to an inflection point of the daily infection curve. We show, across 12 different countries from all populated continents around the world, that our method can often predict future surges in scenarios where the traditional approaches would always predict no future surges. However, as expected, BTF projections cannot accurately predict a surge due to the emergence of a new variant. To generate BTF projections, we make use of a matching scheme for asynchronous time series combined with a response coaching SIR model.

Predictors of outcomes in patients with obesity following mitral valve surgery.

Objective: Few studies have assessed the outcomes of mitral valve surgery in patients with obesity. We sought to study factors that determine the in-hospital outcomes of this population to help clinicians provide optimal care. Methods: A retrospective analysis of adult patients with obesity who underwent open mitral valve replacement or repair between January 1, 2012, and December 31, 2020, was conducted using the National Inpatient Sample. Weighted logistic regression and random forest analyses were performed to assess factors associated with mortality and the interaction of each variable. Results: Of the 48,775 patients with obesity, 34% had morbid obesity (body mass index ≥40), 55% were women, 66% underwent elective surgery, and 55% received isolated open mitral valve replacement or repair. In-hospital mortality was 5.0% (n = 2430). After adjusting for important covariates, a greater risk of mortality was associated with older patients (adjusted odds ratio [aOR], 1.24; 95% CI, 1.08-1.43), higher Elixhauser comorbidity score (aOR, 2.10; 95% CI, 1.87-2.36), prior valve surgery (aOR, 1.63; 95% CI, 1.01-2.63), and more than 2 concomitant procedures (aOR, 2.83; 95% CI, 2.07-3.85). Lower mortality was associated with elective admissions (aOR, 0.70; 95% CI, 0.56-0.87) and valve repair (aOR, 0.58; 95% CI, 0.46-0.73). Machine learning identified several interactions associated with early mortality, such as Elixhauser score, female sex, body mass index ≥40, and kidney failure. Conclusions: The complexity of presentation, comorbidities in older and female patients, and morbid obesity are independently associated with an increased risk of mortality in patients undergoing open mitral valve replacement or repair. Morbid obesity and sex disparity should be recognized in this population, and physicians should consider older patients and females with multiple comorbidities for earlier and more opportune treatment windows.

Correcting prevalence estimation for biased sampling with testing errors.

Sampling for prevalence estimation of infection is subject to bias by both oversampling of symptomatic individuals and error-prone tests. This results in naïve estimators of prevalence (ie, proportion of observed infected individuals in the sample) that can be very far from the true proportion of infected. In this work, we present a method of prevalence estimation that reduces both the effect of bias due to testing errors and oversampling of symptomatic individuals, eliminating it altogether in some scenarios. Moreover, this procedure considers stratified errors in which tests have different error rate profiles for symptomatic and asymptomatic individuals. This results in easily implementable algorithms, for which code is provided, that produce better prevalence estimates than other methods (in terms of reducing and/or removing bias), as demonstrated by formal results, simulations, and on COVID-19 data from the Israeli Ministry of Health.

Geo-spatial risk factor analysis for drug overdose death in South Florida from 2014 to 2019, and the independent contribution of social determinants of health.

PURPOSE: The physical environment and social determinants of health have been shown to influence health behaviors including drug use and fatal drug overdose. The current research examines the effects of the built environment, social determinants of health measures and aggregated risk from the built environment at neighborhood-level on drug overdose death locations in Miami-Dade County, Florida. METHODS: Risk Terrain Modeling (RTM) was used to assess the place features risk factors that significantly increase the risk of drug overdose death spatially in Miami-Dade County ZIP Code Tabulation Areas, Florida from 2014 to 2019. An aggregated neighborhood risk of fatal drug overdose measure was developed by averaging the risk per grid cell from the RTM within census block groups each year. Six logistic and zero-inflated regression models were built to examine the effects of three indices of incident-specific social determinants of health (IS-SDH) measures and aggregated risk measures separately, and simultaneously on drug overdose death locations each year. RESULTS: Seven place features including parks, bus stops, restaurants and grocery stores were significantly related to the occurrence of fatal drug overdoses. When examined separately, one or more indices of the IS-SDH were significant covariates of drug overdose locations in some years. When examined simultaneously, the three indices of the IS-SDH and aggregated risk of fatal drug overdose measure could be all significant in certain years. CONCLUSIONS: The patterns of high-risk areas and place features identified from the RTM related to drug overdose deaths may be used to inform the placement of treatment and prevention resources. A multi-factor approach that combines an aggregated neighborhood risk measure reflecting the risk from the built environment and the incident-specific social determinants of health measures can be used to identify the drug overdose death locations in certain years.

Comparing Bayesian Variable Selection to Lasso Approaches for Applications in Psychology.

In the current paper, we review existing tools for solving variable selection problems in psychology. Modern regularization methods such as lasso regression have recently been introduced in the field and are incorporated into popular methodologies, such as network analysis. However, several recognized limitations of lasso regularization may limit its suitability for psychological research. In this paper, we compare the properties of lasso approaches used for variable selection to Bayesian variable selection approaches. In particular we highlight advantages of stochastic search variable selection (SSVS), that make it well suited for variable selection applications in psychology. We demonstrate these advantages and contrast SSVS with lasso type penalization in an application to predict depression symptoms in a large sample and an accompanying simulation study. We investigate the effects of sample size, effect size, and patterns of correlation among predictors on rates of correct and false inclusion and bias in the estimates. SSVS as investigated here is reasonably computationally efficient and powerful to detect moderate effects in small sample sizes (or small effects in moderate sample sizes), while protecting against false inclusion and without over-penalizing true effects. We recommend SSVS as a flexible framework that is well-suited for the field, discuss limitations, and suggest directions for future development.

Improved Prediction of Perimetric Loss in Glaucomatous Eyes Using Latent Class Mixed Modeling.

PURPOSE: To evaluate whether the identification of distinct classes within a population of glaucoma patients improves estimates of future perimetric loss. DESIGN: Longitudinal cohort study. PARTICIPANTS: A total of 6558 eyes of 3981 subjects from the Duke Ophthalmic Registry with ≥ 5 reliable standard automated perimetry (SAP) tests and ≥ 2 years of follow-up. METHODS: Standard automated perimetry mean deviation (MD) values were extracted with associated timepoints. Latent class mixed models (LCMMs) were used to identify distinct subgroups (classes) of eyes according to rates of perimetric change over time. Rates for individual eyes were then estimated by considering both individual eye data and the most probable class membership for that eye. Data were split into training (80%) and test sets (20%), and test set mean squared prediction errors (MSPEs) were estimated using LCMM and ordinary least squares (OLS) regression. MAIN OUTCOME MEASURES: Rates of change in SAP MD in each class and MSPE. RESULTS: The dataset contained 52 900 SAP tests with an average of 8.1 ± 3.7 tests per eye. The best-fitting LCMM contained 5 classes with rates of -0.06, -0.21, -0.87, -2.15, and +1.28dB/year (80.0%, 10.2%, 7.5%, 1.3%, and 1.0% of the population, respectively) labeled as slow, moderate, fast, catastrophic progressors, and "improvers" respectively. Fast and catastrophic progressors were older (64.1 ± 13.7 and 63.5 ± 16.9 vs. 57.8 ± 15.8, P < 0.001) and had generally mild-moderate disease at baseline (65.7% and 71% vs. 52%, P < 0.001) than slow progressors. The MSPE was significantly lower for LCMM compared to OLS, regardless of the number of tests used to obtain the rate of change (5.1 ± 0.6 vs. 60.2 ± 37.9, 4.9 ± 0.5 vs. 13.4 ± 3.2, 5.6 ± 0.8 vs. 8.1 ± 1.1, 3.4 ± 0.3 vs. 5.5 ± 1.1 when predicting the fourth, fifth, sixth, and seventh visual fields (VFs) respectively; P < 0.001 for all comparisons). MSPE of fast and catastrophic progressors was significantly lower with LCMM versus OLS (17.7 ± 6.9 vs. 48.1 ± 19.7, 27.1 ± 8.4 vs. 81.3 ± 27.1, 49.0 ± 14.7 vs. 183.9 ± 55.2, 46.6 ± 16.0 vs. 232.4 ± 78.0 when predicting the fourth, fifth, sixth, and seventh VFs respectively; P < 0.001 for all comparisons). CONCLUSIONS: Latent class mixed model successfully identified distinct classes of progressors within a large glaucoma population that seemed to reflect subgroups observed in clinical practice. Latent class mixed models were superior to OLS regression in predicting future VF observations. FINANCIAL DISCLOSURE(S): Proprietary or commercial disclosuremay be found after the references.

Disparities in Survival Due to Social Determinants of Health and Access to Treatment in US Patients With Operable Malignant Pleural Mesothelioma.

Importance: Outcomes of localized malignant pleural mesothelioma (MPM) remain poor despite multimodality therapy. It is unclear what role disparities have in the overall survival (OS) of patients with operable MPM. Objective: To examine survival disparities associated with social determinants of health (SDOHs) and treatment access in patients with malignant pleural mesothelioma. Design, Setting, and Participants: In this observational, retrospective cohort study, patients with MPM diagnosed between January 1, 2004, and December 31, 2017, were identified from the National Cancer Database with a maximum follow-up time of 13.6 years. The analysis was conducted from February 16, 2022, to July 29, 2022. Patients were included if they were diagnosed with potentially resectable clinical stage I to IIIA MPM, had epithelioid and biphasic histologic subtypes, and received chemotherapy. Patients were excluded if they could not receive curative surgery, were 75 years or older, or had metastasis, unknown stage, or tumor extension to the chest wall, mediastinal tissues, or organs. Exposures: Chemotherapy alone vs chemotherapy with curative surgery in the form of pleurectomy and decortication or extrapleural pneumonectomy. Main Outcomes and Measures: The primary end point was OS. Cox proportional hazards regression models were used to determine hazard ratios (HRs) for OS, including univariable and multivariable models controlling for potential confounders, including demographic, comorbidity, clinical, treatment, tumor, and hospital-related variables, as well as SDOHs. Results: A total of 1389 patients with MPM were identified (median [IQR] age, 66 [61-70] years; 1024 [74%] male; 12 [1%] Asian, 49 [3%] Black, 74 [5%] Hispanic, 1233 [89%] White, and 21 [2%] of other race). The median OS was 1.7 years (95% CI, 1.6-1.8). Risk factors associated with worse OS included older age, male sex, Black race, low income, and low educational attainment. Factors associated with greater odds of survival included receipt of surgical therapy, recent year of treatment, increased distance to travel, and treatment at high-volume academic hospitals. The risk factors most strongly associated with poor OS included Black race (HR, 1.96; 95% CI, 1.43-2.69) and male sex (HR, 1.60; 95% CI, 1.38-1.86). Surgical treatment in addition to systemic chemotherapy (HR, 0.70; 95% CI, 0.61-0.81) was independently associated with improved OS, as were chemotherapy initiation (HR, 0.93; 95% CI, 0.87-0.99) and greater travel distance from the hospital (HR, 0.92; 95% CI, 0.86-0.98). Conclusions and Relevance: In this retrospective cohort study of patients with operable MPM, there was significant variability in access to care by SDOHs. Addressing disparities in access to multimodality therapy can help ensure equity of care for patients with MPM.

Partially Recursively Induced Structured Moderation (PRISM) for modeling racial differences in endometrial cancer survival.

PURPOSE: Health disparities are driven by a complex interplay of determinants operating across multiple levels of influence. However, while recognized conceptually, much disparities research fails to capture this inherent complexity in study focus and/or design; little of such work accounts for the interplay across the multiple levels of influence from structural (contextual) to biological or clinical. We developed a novel modeling framework that addresses these challenges and provides new insights. METHODS: We used data from the Florida Cancer Data System on endometrial cancer patients and geocoded-derived social determinants of health to demonstrate the applicability of a new modeling paradigm we term PRISM regression. PRISM is a new highly interpretable tree-based modeling framework that allows for automatic discovery of potentially non-linear hierarchical interactions between health determinants at multiple levels and differences in survival outcomes between groups of interest, including through a new specific area-level disparity estimate (SPADE) incorporating these multilevel influences. RESULTS: PRISM demonstrates that hierarchical influences on racial disparity in endometrial cancer survival appear to be statistically relevant and that these better predict survival differences than only using individual level determinants. The interpretability of the models allows more careful inspection of the nature of these hierarchical effects on disparity. Additionally, SPADE estimates show distinct geographical patterns across census tracts in Florida. CONCLUSION: PRISM can provide a powerful new modeling framework with which to better understand racial disparities in cancer survival.

High Dimensional Mode Hunting Using Pettiest Components Analysis.

Principal components analysis has been used to reduce the dimensionality of datasets for a long time. In this paper, we will demonstrate that in mode detection the components of smallest variance, the pettiest components, are more important. We prove that for a multivariate normal or Laplace distribution, we obtain boxes of optimal volume by implementing "pettiest component analysis," in the sense that their volume is minimal over all possible boxes with the same number of dimensions and fixed probability. This reduction in volume produces an information gain that is measured using active information. We illustrate our results with a simulation and a search for modal patterns of digitized images of hand-written numbers using the famous MNIST database; in both cases pettiest components work better than their competitors. In fact, we show that modes obtained with pettiest components generate better written digits for MNIST than principal components.

Rates of Glaucoma Progression Derived from Linear Mixed Models Using Varied Random Effect Distributions.

PURPOSE: To compare the ability of linear mixed models with different random effect distributions to estimate rates of visual field loss in glaucoma patients. METHODS: Eyes with five or more reliable standard automated perimetry (SAP) tests were identified from the Duke Glaucoma Registry. Mean deviation (MD) values from each visual field and associated timepoints were collected. These data were modeled using ordinary least square (OLS) regression and linear mixed models using the Gaussian, Student's t, or log-gamma (LG) distributions as the prior distribution for random effects. Model fit was compared using the Watanabe-Akaike information criterion (WAIC). Simulated eyes of varying initial disease severity and rates of progression were created to assess the accuracy of each model in predicting the rate of change and likelihood of declaring progression. RESULTS: A total of 52,900 visual fields from 6558 eyes of 3981 subjects were included. Mean follow-up period was 8.7 ± 4.0 years, with an average of 8.1 ± 3.7 visual fields per eye. The LG model produced the lowest WAIC, demonstrating optimal model fit. In simulations, the LG model declared progression earlier than OLS (P < 0.001) and had the greatest accuracy in predicted slopes (P < 0.001). The Gaussian model significantly underestimated rates of progression among fast and catastrophic progressors. CONCLUSIONS: Linear mixed models using the LG distribution outperformed conventional approaches for estimating rates of SAP MD loss in a population with glaucoma. TRANSLATIONAL RELEVANCE: Use of the LG distribution in models estimating rates of change among glaucoma patients may improve their accuracy in rapidly identifying progressors at high risk for vision loss.

A Formal Framework for Knowledge Acquisition: Going beyond Machine Learning.

Philosophers frequently define knowledge as justified, true belief. We built a mathematical framework that makes it possible to define learning (increasing number of true beliefs) and knowledge of an agent in precise ways, by phrasing belief in terms of epistemic probabilities, defined from Bayes' rule. The degree of true belief is quantified by means of active information I+: a comparison between the degree of belief of the agent and a completely ignorant person. Learning has occurred when either the agent's strength of belief in a true proposition has increased in comparison with the ignorant person (I+>0), or the strength of belief in a false proposition has decreased (I+<0). Knowledge additionally requires that learning occurs for the right reason, and in this context we introduce a framework of parallel worlds that correspond to parameters of a statistical model. This makes it possible to interpret learning as a hypothesis test for such a model, whereas knowledge acquisition additionally requires estimation of a true world parameter. Our framework of learning and knowledge acquisition is a hybrid between frequentism and Bayesianism. It can be generalized to a sequential setting, where information and data are updated over time. The theory is illustrated using examples of coin tossing, historical and future events, replication of studies, and causal inference. It can also be used to pinpoint shortcomings of machine learning, where typically learning rather than knowledge acquisition is in focus.

Determinants of stage at diagnosis of HPV-related cancer including area deprivation and clinical factors.

BACKGROUND: Collecting social determinants of health in electronic health records is time-consuming. Meanwhile, an Area Deprivation Index (ADI) aggregates sociodemographic information from census data. The objective of this study was to ascertain whether ADI is associated with stage of human papillomavirus (HPV)-related cancer at diagnosis. METHODS: We tested for the association between the stage of HPV-related cancer presentation and ADI as well as the association between stage and the value of each census-based measure using ordered logistic regression, adjusting for age, race and sex. RESULTS: Among 3247 cases of HPV-related cancers presenting to an urban academic medical center, the average age at diagnosis was 57. The average stage at diagnosis was Surveillance, Epidemiology and End Results Stage 3. In the study population, 43% of patients were female and 87% were white. In this study population, there was no association between stage of HPV-related cancer presentation and either aggregate or individual census variables. CONCLUSIONS: These results may reflect insufficient sample size, a lack of socio-demographic diversity in our population, or suggest that simplifying social determinants of health into a single geocoded index is not a reliable surrogate for assessing a patient's risk for HPV-related cancer.

A simple correction for COVID-19 sampling bias.

COVID-19 testing has become a standard approach for estimating prevalence which then assist in public health decision making to contain and mitigate the spread of the disease. The sampling designs used are often biased in that they do not reflect the true underlying populations. For instance, individuals with strong symptoms are more likely to be tested than those with no symptoms. This results in biased estimates of prevalence (too high). Typical post-sampling corrections are not always possible. Here we present a simple bias correction methodology derived and adapted from a correction for publication bias in meta analysis studies. The methodology is general enough to allow a wide variety of customization making it more useful in practice. Implementation is easily done using already collected information. Via a simulation and two real datasets, we show that the bias corrections can provide dramatic reductions in estimation error.

Predicting DNA methylation from genetic data lacking racial diversity using shared classified random effects.

Public genomic repositories are notoriously lacking in racially and ethnically diverse samples. This limits the reaches of exploration and has in fact been one of the driving factors for the initiation of the All of Us project. Our particular focus here is to provide a model-based framework for accurately predicting DNA methylation from genetic data using racially sparse public repository data. Epigenetic alterations are of great interest in cancer research but public repository data is limited in the information it provides. However, genetic data is more plentiful. Our phenotype of interest is cervical cancer in The Cancer Genome Atlas (TCGA) repository. Being able to generate such predictions would nicely complement other work that has generated gene-level predictions of gene expression for normal samples. We develop a new prediction approach which uses shared random effects from a nested error mixed effects regression model. The sharing of random effects allows borrowing of strength across racial groups greatly improving predictive accuracy. Additionally, we show how to further borrow strength by combining data from different cancers in TCGA even though the focus of our predictions is DNA methylation in cervical cancer. We compare our methodology against other popular approaches including the elastic net shrinkage estimator and random forest prediction. Results are very encouraging with the shared classified random effects approach uniformly producing more accurate predictions - overall and for each racial group.

A simple correction for COVID-19 sampling bias.

COVID-19 testing has become a standard approach for estimating prevalence which then assist in public health decision making to contain and mitigate the spread of the disease. The sampling designs used are often biased in that they do not reflect the true underlying populations. For instance, individuals with strong symptoms are more likely to be tested than those with no symptoms. This results in biased estimates of prevalence (too high). Typical post-sampling corrections are not always possible. Here we present a simple bias correction methodology derived and adapted from a correction for publication bias in meta analysis studies. The methodology is general enough to allow a wide variety of customization making it more useful in practice. Implementation is easily done using already collected information. Via a simulation and two real datasets, we show that the bias corrections can provide dramatic reductions in estimation error.

Contextualizing covid-19 spread: a county level analysis, urban versus rural, and implications for preparing for the next wave.

Paraphrasing [Morano and Holt, 2017], contextual determinants of health including social, environmental, healthcare and others, are a so-called deck of cards one is dealt. The ability to modify health outcomes varies then based upon how one's hand is played. It is thus of great interest to understand how these determinants associate with the emerging pandemic covid-19. To this end, we conducted a deep-dive analysis into this problem using a recently curated public dataset on covid-19 that connects infection spread over time to a rich collection of contextual determinants for all counties of the U.S and Washington, D.C. Using random forest machine learning methodology, we identified a relevant constellation of contextual factors of disease spread which manifest differently for urban and rural counties. The findings also have clear implications for better preparing for the next wave of disease.

Mode hunting through active information.

We propose a new method to find modes based on active information. We develop an algorithm called active information mode hunting (AIMH) that, when applied to the whole space, will say whether there are any modes present and where they are. We show AIMH is consistent and, given that information increases where probability decreases, it helps to overcome issues with the curse of dimensionality. The AIMH also reduces the dimensionality with no resource to principal components. We illustrate the method in three ways: with a theoretical example (showing how it performs better than other mode hunting strategies), a real dataset business application, and a simulation.

Discordancy Partitioning for Validating Potentially Inconsistent Pharmacogenomic Studies.

The Genomics of Drug Sensitivity in Cancer (GDSC) and Cancer Cell Line Encyclopedia (CCLE) are two major studies that can be used to mine for therapeutic biomarkers for cancers of a large variety. Model validation using the two datasets however has proved challenging. Both predictions and signatures do not consistently validate well for models built on one dataset and tested on the other. While the genomic profiling seems consistent, the drug response data is not. Some efforts at harmonizing experimental designs has helped but not entirely removed model validation difficulties. In this paper, we present a partitioning strategy based on a data sharing concept which directly acknowledges a potential lack of concordance between datasets and in doing so, also allows for extraction of reproducible novel gene-drug interaction signatures as well as accurate test set predictions. We demonstrate these properties in a re-analysis of the GDSC and CCLE datasets.