Sandwich Estimator Research Articles

Empirical Sandwich Variance Estimator for Iterated Conditional Expectation g-Computation.

Iterated conditional expectation (ICE) g-computation is an estimation approach for addressing time-varying confounding for both longitudinal and time-to-event data. Unlike other g-computation implementations, ICE avoids the need to specify models for each time-varying covariate. For variance estimation, previous work has suggested the bootstrap. However, bootstrapping can be computationally intense. Here, we present ICE g-computation as a set of stacked estimating equations. Therefore, the variance for the ICE g-computation estimator can be consistently estimated using the empirical sandwich variance estimator. Performance of the variance estimator was evaluated empirically with a simulation study. The proposed approach is also demonstrated with an illustrative example on the effect of cigarette smoking on the prevalence of hypertension. In the simulation study, the empirical sandwich variance estimator appropriately estimated the variance. When comparing runtimes between the sandwich variance estimator and the bootstrap for the applied example, the sandwich estimator was substantially faster, even when bootstraps were run in parallel. The empirical sandwich variance estimator is a viable option for variance estimation with ICE g-computation.

Simplified methods for variance estimation in microbiome abundance count data analysis.

The complex nature of microbiome data has made the differential abundance analysis challenging. Microbiome abundance counts are often skewed to the right and heteroscedastic (also known as overdispersion), potentially leading to incorrect inferences if not properly addressed. In this paper, we propose a simple yet effective framework to tackle the challenges by integrating Poisson (log-linear) regression with standard error estimation through the Bootstrap method and Sandwich robust estimation. Such standard error estimates are accurate and yield satisfactory inference even if the distributional assumption or the variance structure is incorrect. Our approach is validated through extensive simulation studies, demonstrating its effectiveness in addressing overdispersion and improving inference accuracy. Additionally, we apply our approach to two real datasets collected from the human gut and vagina, respectively, demonstrating the wide applicability of our methods. The results highlight the efficacy of our covariance estimators in addressing the challenges of microbiome data analysis. The corresponding software implementation is publicly available at https://github.com/yimshi/robustestimates.

How estimating nuisance parameters can reduce the variance (with consistent variance estimation).

We often estimate a parameter of interest when the identifying conditions involve a finite-dimensional nuisance parameter . Examples from causal inference are inverse probability weighting, marginal structural models and structural nested models, which all lead to unbiased estimating equations. This article presents a consistent sandwich estimator for the variance of estimators that solve unbiased estimating equations including which is also estimated by solving unbiased estimating equations. This article presents four additional results for settings where solves (partial) score equations and does not depend on . This includes many causal inference settings where describes the treatment probabilities, missing data settings where describes the missingness probabilities, and measurement error settings where describes the error distribution. These four additional results are: (1) Counter-intuitively, the asymptotic variance of is typically smaller when is estimated. (2) If estimating is ignored, the sandwich estimator for the variance of is conservative. (3) A consistent sandwich estimator for the variance of . (4) If with the true plugged in is efficient, the asymptotic variance of does not depend on whether is estimated. To illustrate we use observational data to calculate confidence intervals for (1) the effect of cazavi versus colistin on bacterial infections and (2) how the effect of antiretroviral treatment depends on its initiation time in HIV-infected patients.

A new disease mapping method for improving data completeness of syndromic surveillance with high missing rates

AbstractSyndromic surveillance is a type of public health surveillance that utilizes nonspecific indicators or symptoms associated with a particular disease or condition to detect and track disease outbreaks early. However, data completeness has been a significant challenge for syndromic surveillance systems in many countries. Incomplete data may make it difficult to accurately identify anomalies or trends in surveillance data. In this study, a new disease mapping method based on a high‐accuracy, low‐rank tensor completion (HaLRTC) algorithm is proposed to estimate the quarterly positivity rate of the human influenza virus (IFV) based on highly insufficient 2010–2015 respiratory syndromic surveillance data from the subtropical monsoon region of China. The HaLRTC algorithm is a spatiotemporal interpolation method applied to fill in missing or incomplete data using a low‐rank tensor structure. The results show that the accuracy (R2 = 0.880, RMSE = 0.037) of the proposed method is much higher than that of three traditional disease mapping methods: Cokriging, hierarchical Bayesian, and sandwich estimation methods. This study provides a new disease mapping approach to improve the quality and completeness of data in syndrome surveillance or other familiar systems with a large proportion of missing data.

#2539 Effects of relative blood volume and ultrafiltration volume on pulse wave characteristics during hemodialysis

Abstract Background and Aims Measures of non-invasive pulse wave (PW) analysis can help to assess the status of the central cardiovascular system and to predict clinical outcomes, including morbidity and mortality. During hemodialysis (HD), fluid is extracted from blood. The resulting reduction in relative blood volume (RBV) and collected ultrafiltration volume (UFV) were hypothesized to substantially influence PW characteristics during HD. Volume overload (VO) is common in HD patients and may affect the chance of improving the cardiovascular status of patients during HD. The present study aimed to explore associations between RBV, UFV, VO and a wide range of PW parameters. Method In a single center in Vienna, Austria, 24 patients on maintenance HD underwent four HD sessions each, with whole-body bioimpedance spectroscopy (BIS) before, and automated, cuff-based PW measurements every 15 minutes, RBV- and UFV-monitoring throughout each session. Associations between RBV or UFV as independent variable and 26 different PW characteristics as dependent variables (52 separate models) were tested using confounder-adjusted generalized estimating equations with sandwich estimator and multilevel nested working covariance structure, reporting standardized effect estimates and drop-in-R² to assess variable importance. Sensitivity models including both RBV and UFV as predictors (26 models) were used to assess effects adjusted for confounding and mediation, respectively. The intra-patient correlation between BIS-derived VO and PW parameters at the start of sessions and PW parameter change by end of sessions was assessed using repeated measures correlation analysis. Results Data from 990 PW measurements were available for analysis from an 88% male cohort with median age of 66 years. In half of the PW parameters analyzed, each percent RBV was significantly associated with in summary about 4.3 to 5.7% (95% confidence intervals [CI] 0.9 to 10.46) of the PW parameter's respective standard deviation (SD), negatively so only in heart rate and R-Peak. UFV was negatively associated with a third of PW parameters (effects ranging from −2.5 to −5.47% SD [95% CIs −0.9 to −10.0]), mostly in classical peripheral and central pressures and peripheral resistance. In models including both predictors, RBV overall exhibited greater effect size and variance capture than UFV, except for peripheral resistance. On within-patient level, VO was positively correlated with initial medians of pressure excess (r=0.34, p=0.034), augmentation-index (0.34, p=0.031) and augmentation-pressure (0.32, p=0.042), increase in heart rate (0.32, p=0.04) and decreases in peripheral resistance (0.31, p=0.05), peripheral (systolic: 0.34, p=0.032; diastolic: 0.45, p=0.004; mean: 0.41, p=0.008) and central (diastolic: 0.46, p=0.003) blood pressures during HD. Conclusion The observed PW characteristics were associated with RBV and UFV during HD and may represent modifiable targets for intervention studies. Pressure excess, S-Peak, R-Peak, cardiac and preload parameters were more closely associated with RBV, while classic (especially diastolic) blood pressure parameters, ejection duration and S/D-Ratio were more associated with UFV, in models using both predictors. On average, blood pressure and peripheral resistance tended to decrease less when a patient began HD in a volume-expanded state, which may affect day-to-day feasibility of more classical pressure targets, but perhaps not of other PW parameters. Multiple PW characteristics have previously been shown to be associated with mortality and morbidity and are potential target parameters for future intervention trials. Utilizing RBV modification as an intervention may offer superior precision compared to UFV goals, requiring further studies. PW technology could also be adopted by HD-machine manufacturers at low cost, enabling more precise and individualized treatment options.

Covariate adjustment and estimation of difference in proportions in randomized clinical trials.

Difference in proportions is frequently used to measure treatment effect for binary outcomes in randomized clinical trials. The estimation of difference in proportions can be assisted by adjusting for prognostic baseline covariates to enhance precision and bolster statistical power. Standardization or g-computation is a widely used method for covariate adjustment in estimating unconditional difference in proportions, because of its robustness to model misspecification. Various inference methods have been proposed to quantify the uncertainty and confidence intervals based on large-sample theories. However, their performances under small sample sizes and model misspecification have not been comprehensively evaluated. We propose an alternative approach to estimate the unconditional variance of the standardization estimator based on the robust sandwich estimator to further enhance the finite sample performance. Extensive simulations are provided to demonstrate the performances of the proposed method, spanning a wide range of sample sizes, randomization ratios, and model specification. We apply the proposed method in a real data example to illustrate the practical utility.

Gestational Exposure to Maternal Systemic Glucocorticoids and Childhood Risk of CKD

Background & ObjectivesThe potential effects of antenatal glucocorticoid exposure on the health of children are unclear. We examined the association of gestational exposure to maternal systemic glucocorticoids (SG) and the risk of developing chronic kidney disease (CKD) in childhood. Study designRetrospective cohort study. Setting & ParticipantsNewborns cared for at the largest healthcare delivery system in Taiwan between 2004 and 2018. ExposureMaternal prescriptions for SG between the last menstrual period and birth as a proxy for gestational exposure. OutcomeIncidence of childhood CKD, including congenital anomalies of the kidney and urinary tract (CAKUT) and other kidney diseases (non-CAKUT) over 10 years. Analytical ApproachCox proportional hazards models with stabilized inverse probability of treatment weighting and robust sandwich estimator were used to estimate the average association between SG and incident CKD after adjustment for offspring characteristics (aHR). ResultsAmong 23,363 singleton-born children, gestational SG exposure was significantly associated with a higher risk of childhood CKD (aHR, 1.69 [95% CI, 1.01–2.84]). Stratified analyses showed stronger associations between SG and childhood CKD within the strata of birth <37 weeks gestational age (aHR, 2.38 [95% CI, 1.19–4.78]), male sex (aHR, 1.89 [95% CI, 1.00–3.55]), gestational exposure in the second trimester (aHR, 6.70 [95% CI, 2.17–20.64]), and total dose >24 mg hydrocortisone equivalent (aHR, 1.91 [95% CI, 1.05–3.47). LimitationsStudy was limited to the Taiwan healthcare delivery system and childhood CKD events through age 10 years. ConclusionsThe findings of this study suggest that gestational exposure to systemic glucocorticoids is associated with the occurrence of kidney disease in childhood. If these findings are confirmed, they may inform clinicians who are considering prescribing SG during pregnancy.

Bootstrapping Long-Run Covariance of Stationary Functional Time Series

A key summary statistic in a stationary functional time series is the long-run covariance function that measures serial dependence. It can be consistently estimated via a kernel sandwich estimator, which is the core of dynamic functional principal component regression for forecasting functional time series. To measure the uncertainty of the long-run covariance estimation, we consider sieve and functional autoregressive (FAR) bootstrap methods to generate pseudo-functional time series and study variability associated with the long-run covariance. The sieve bootstrap method is nonparametric (i.e., model-free), while the FAR bootstrap method is semi-parametric. The sieve bootstrap method relies on functional principal component analysis to decompose a functional time series into a set of estimated functional principal components and their associated scores. The scores can be bootstrapped via a vector autoregressive representation. The bootstrapped functional time series are obtained by multiplying the bootstrapped scores by the estimated functional principal components. The FAR bootstrap method relies on the FAR of order 1 to model the conditional mean of a functional time series, while residual functions can be bootstrapped via independent and identically distributed resampling. Through a series of Monte Carlo simulations, we evaluate and compare the finite-sample accuracy between the sieve and FAR bootstrap methods for quantifying the estimation uncertainty of the long-run covariance of a stationary functional time series.

Abstract 115: Presenting Symptoms and Neuroimaging Correlates in North American Moyamoya Vasculopathy

Introduction: Moyamoya disease (MMD) is a steno-occlusion of the supraclinoid internal carotid arteries and proximal branches with the development of collateral vessels 1 . Once thought to affect primarily females of East Asian descent, more than 11,000 hospitalizations in patients with MMD of non-Asian ancestry are reported annually in the USA, with the majority presenting with ischemic stroke. Here, we present frequent non-stroke symptomatology in North American MMD along with hemodynamic imaging profiles at presentation. Methods: Participants provided informed consent and underwent anatomical MRI with hypercapnic blood oxygenation level-dependent 2 ; time regression was applied to quantify cerebrovascular reactivity (CVR) and reactivity delay time (CVR DELAY ) 3 in right and left MCA territories. A logistic regression analysis based on the Huber-White sandwich estimator compared the CVR metrics between cohorts. Significance criterion: Bonferroni corrected two-sided p&lt;0.05. Results: 74 patients (age=11-77 years; Asian=6%, White=65%, Black=29%) diagnosed with MMD with symptom durations spanning hours to years prior to presentation were enrolled. Non-stroke symptoms and frequency prior to stroke are listed in Table 1. In more than 1/3 rd of patients, unprovoked headaches preceded stroke-like symptoms. In patients exhibiting ischemic symptoms within six months of neuroimaging, cerebrovascular reserve capacity was reduced (p=0.007), and cerebrovascular reactivity delay time increased (p&lt;0.001), in the flow territory from which symptoms were derived compared to the contralateral flow territory. Discussion: North American MMD is being observed with increasing frequency and increasing the awareness of non-stroke MMD presentations may lead to early diagnosis that can lead to proactive revascularization to minimize the risk of first stroke in these patients.

Proportional rates models for multivariate panel count data.

Multivariate panel count data arise when there are multiple types of recurrent events, and the observation for each study subject consists of the number of recurrent events of each type between two successive examinations. We formulate the effects of potentially time-dependent covariates on multiple types of recurrent events through proportional rates models, while leaving the dependence structures of the related recurrent events completely unspecified. We employ nonparametric maximum pseudo-likelihood estimation under the working assumptions that all types of events are independent and each type of event is a nonhomogeneous Poisson process, and we develop a simple and stable EM-type algorithm. We show that the resulting estimators of the regression parameters are consistent and asymptotically normal, with a covariance matrix that can be estimated consistently by a sandwich estimator. In addition, we develop a class of graphical and numerical methods for checking the adequacy of the fitted model. Finally, we evaluate the performance of the proposed methods through simulation studies and analysis of a skin cancer clinical trial.

Survival of patients with microsatellite instability-high and Lynch syndrome-associated pancreatic ductal adenocarcinomas.

640 Background: Pancreatic ductal adenocarcinoma (PDAC) remains a challenging disease due to its aggressive nature, late-stage diagnosis, and limited treatment options. There is an increased interest in cancers which harbor microsatellite instability (MSI) due to their susceptibility to immune checkpoint inhibitors. MSI in PDAC is rare, and little is known about survival in this subtype of pancreas cancers. Our aim was to assess overall survival in surgically resected high microsatellite instability (MSI-H), mismatch repair deficient (MMRd), and Lynch Syndrome (LS) associated PDACs treated in our center. Methods: Patients with surgically resected PDACs from 1990 to 2023 were identified. A total of 936 cases were sequenced utilizing MSK-IMPACT, a targeted next generation sequencing platform. Patients with germline or sporadic pathogenic variants or likely pathogenic variants in one of the DNA MMR genes MLH1, MSH2, MSH6, PMS2, or 3’ end of EpCAM were identified. MSI-H, MMRd, and LS-associated cases were matched to patients with microsatellite stable or MMR proficient, non-LS-associated PDAC in a 1:2 ratio with a direct match on age (±5 years), gender, and year of surgery (±3 years). Kaplan-Meier curves were used to visualize the data. A generalized estimating equation (GEE) cox model with robust sandwich estimator was used to compare matched cohorts. Results: 17 patients with MSI-H, MMRd or LS-associated surgically resected PDACs were identified. Five patients were classified as MSI-H, 3 patients were MSI-indeterminate, and 9 patients were microsatellite stable. Eight patients were found to have pathogenic germline variants in LS-associated genes. The remaining nine patients had sporadic pathogenic DNA MMR gene variants ( MLH1 n=3, MSH6 n=6). Median age at surgery was 66 (interquartile range (IQR) 60-77), 41% of the patients were female, and median year of surgery was 2014. MSI-H, MMRd or LS-associated PDACs were matched to 34 control patients. The median survival time of the cases was 12 years vs. 1.9 years in the control group. Five-year survival rate was 81% (IQR 60-100%) and 18% (7.4%-43%), respectively. LS-associated, MSI-H or MMRd status in resectable PDAC was a strong predictor of good prognosis (hazard ratio 0.17, 95% CI 0.07-0.38, p value &lt;0.001). Conclusions: MSI-H, MMRd or LS-associated PDACs displayed significantly better survival compared to their MSI-stable, MMR proficient, non-LS-associated counterparts, in the era before routine use of immunotherapy. It is expected that survival will be increased further with more frequent availability of immunotherapy.

Principal Component Patterns of Pediatric Respiratory Viral Testing Across Health Care Settings.

Factors prompting clinicians to request viral testing in children are unclear. We assessed patterns prompting clinicians to perform viral testing in children discharged from an emergency department (ED) or hospitalized with an acute respiratory infection (ARI). Using active ARI surveillance data collected from November 2017 through February 2020, children aged between 30 days and 17 years with fever or respiratory symptoms who had a research respiratory specimen tested were included. Children's presentation patterns from their initial evaluation at each health care setting were analyzed using principal components (PCs) analysis. PC-specific models using logistic regression with robust sandwich estimators were used to estimate odds ratios (ORs) and 95% confidence intervals (CIs) between PCs and provider-ordered viral testing. PCs were assigned respiratory virus/viruses names a priori based on the patterns represented. In total, 4107 children were enrolled and tested, with 2616 (64%) discharged from the ED and 1491 (36%) hospitalized. In the ED, children with a coviral presentation pattern had a 1.44-fold (95% CI, 1.24-1.68) increased odds of receiving a provider-ordered viral test than children showing clinical symptoms less representative of coviral-like infection. Whereas children in the ED and hospitalized with rhinovirus-like symptoms had 71% (OR, 0.29; 95% CI, 0.24-0.34) and 39% (OR, 0.61; 95% CI, 0.49-0.76) decreased odds, respectively, of receiving a provider-ordered viral test during their medical encounter. Viral tests are frequently ordered by clinicians, but presentation patterns vary by setting and influence the initiation of testing. Additional assessments of factors affecting provider decisions to use viral testing in pediatric ARI management are needed to maximize patient benefits of testing.

Detecting potential outliers in longitudinal data with time-dependent covariates.

Outliers can influence regression model parameters and change the direction of the estimated effect, over-estimating or under-estimating the strength of the association between a response variable and an exposure of interest. Identifying visit-level outliers from longitudinal data with continuous time-dependent covariates is important when the distribution of such variable is highly skewed. The primary objective was to identify potential outliers at follow-up visits using interquartile range (IQR) statistic and assess their influence on estimated Cox regression parameters. Study was motivated by a large TEDDY dietary longitudinal and time-to-event data with a continuous time-varying vitamin B12 intake as the exposure of interest and development of Islet Autoimmunity (IA) as the response variable. An IQR algorithm was applied to the TEDDY dataset to detect potential outliers at each visit. To assess the impact of detected outliers, data were analyzed using the extended time-dependent Cox model with robust sandwich estimator. Partial residual diagnostic plots were examined for highly influential outliers. Extreme vitamin B12 observations that were cases of IA had a stronger influence on the Cox regression model than non-cases. Identified outliers changed the direction of hazard ratios, standard errors, or the strength of association with the risk of developing IA. At the exploratory data analysis stage, the IQR algorithm can be used as a data quality control tool to identify potential outliers at the visit level, which can be further investigated.

Provider-ordered viral testing and antibiotic administration practices among children with acute respiratory infections across healthcare settings in Nashville, Tennessee.

Evaluate the association between provider-ordered viral testing and antibiotic treatment practices among children discharged from an ED or hospitalized with an acute respiratory infection (ARI). Active, prospective ARI surveillance study from November 2017 to February 2020. Pediatric hospital and emergency department in Nashville, Tennessee. Children 30 days to 17 years old seeking medical care for fever and/or respiratory symptoms. Antibiotics prescribed during the child's ED visit or administered during hospitalization were categorized into (1) None administered; (2) Narrow-spectrum; and (3) Broad-spectrum. Setting-specific models were built using unconditional polytomous logistic regression with robust sandwich estimators to estimate the adjusted odds ratios and 95% confidence intervals between provider-ordered viral testing (ie, tested versus not tested) and viral test result (ie, positive test versus not tested and negative test versus not tested) and three-level antibiotic administration. 4,107 children were enrolled and tested, of which 2,616 (64%) were seen in the ED and 1,491 (36%) were hospitalized. In the ED, children who received a provider-ordered viral test had 25% decreased odds (aOR: 0.75; 95% CI: 0.54, 0.98) of receiving a narrow-spectrum antibiotic during their visit than those without testing. In the inpatient setting, children with a negative provider-ordered viral test had 57% increased odds (aOR: 1.57; 95% CI: 1.01, 2.44) of being administered a broad-spectrum antibiotic compared to children without testing. In our study, the impact of provider-ordered viral testing on antibiotic practices differed by setting. Additional studies evaluating the influence of viral testing on antibiotic stewardship and antibiotic prescribing practices are needed.

Choice of link functions for generalized linear mixed models in meta-analyses of proportions

Two-step approaches for synthesizing proportions in a meta-analysis require first transforming the proportions to a scale where their distribution across studies can be approximated by a normal distribution. Commonly used transformations include the log, logit, arcsine, and Freeman-Tukey double-arcsine transformations. Alternatively, a generalized linear mixed model (GLMM) can be fit directly on the data using the exact binomial likelihood. Unlike popular two-step methods, this accounts for uncertainty in the within-study variances without a normal approximation and does not require an ad hoc correction for zero counts. However, GLMMs require choosing a link function; we illustrate how the AIC can be used to choose the best fitting link when different link functions give different results. We also highlight how misspecification of the link function can introduce bias; using an empirical sandwich estimator for the standard error may not sufficiently avoid undercoverage due to link function misspecification. We demonstrate the application of GLMMs and choice of link function using data from a systematic review on the prevalence of fever in children with COVID-19.

Investigating the association between asthma and opioid use disorder with interactions of anxiety and depression among a national sample of the US population

Introduction Previous studies have not examined the association between asthma and opioid use disorder (OUD) in a comprehensive national sample of the U.S. population. This study aims to investigate such an association. Methods This is a matched retrospective cohort study, with a follow-up period of two years, utilizing longitudinal electronic medical records of a comprehensive national healthcare database in the U.S.–Cerner-Real World DataTM. Patients selected for analysis were ≥12 years old with a hospital encounter between January 2000 and June 2020. Adjusted risk ratios (aRRs) of incident OUD for those with asthma compared to those without asthma were calculated using a modified Poisson regressions with robust standard errors via the Huber-White sandwich estimator, and results were stratified by comorbid mental illnesses. Results Individuals with asthma had a greater risk of OUD compared to those without asthma (aRR = 2.12; 95% CI 2.03–2.23). When stratified by anxiety and depression status, individuals with asthma and no anxiety or depression had a greater risk of incident OUD compared to individuals with asthma and either anxiety, depression, or both. Additionally, individuals with asthma medication had 1.29 (95% CI: 1.24, 1.35) greater overall risk for incident OUD compared to those without medication. Independent of comorbid mental illnesses, individuals with asthma medication had greater risk for incident OUD compared to those without medication among individuals without severe/obstructive asthma. Conclusions Individuals with asthma face a higher OUD risk compared to those without asthma. Comorbid mental illnesses modulate this risk. Caution is advised in opioid prescribing for asthma patients.

Description of a Clinical Score to Identify PMBL Patients at High Risk of Early-Failure after Rituximab Doxorubicin Back-Bone Chemoimmunotherapy. a FIL Real-World Study

Introduction. Primary Mediastinal B Lymphoma (PMBL) is a rare and aggressive neoplasia that typically occurs in young women and is primarily located in the mediastinum. Although the combination of anthracycline and rituximab immunochemotherapy, and PET-guided consolidation radiotherapy, results in a high cure rate, approximately 10-20% of patients show refractoriness or early failure to front-line treatment with a rapid clinical deterioration and a dismal prognosis. These cases, if promptly identified, could benefit from early treatment shift. Therefore, we addressed the development of a clinical score aimed at estimating the risk of early failure of PMBL patients treated with anthracycline-rituximab-based chemoimmunotherapy in a real-world nationwide setting. Material and Methods This is a retrospective cohort study of an unselected population of 931 adult PMBL patients treated in 37 FIL centers. To qualify for this research, patients had to have a confirmed diagnosis of PMBL, display typical symptoms and have been included in the local databases from 01/01/2007 to 31/12/2019. All patients received treatment with rituximab plus CHOP21 (n=98), CHOP14 (n=181), megaCHOP (n=31), MACOPB (225), VACOPB (n=179), DAEPOCH (n=179). Patients were excluded from this analysis if they received suboptimal treatment (22) or ASCT as consolidation of first-line therapy (12) or were censored within 365 days of treatment initiation (22). We defined early failure (EF) as any relapse or progression within 365 days of treatment initiation. Candidate predictors were clinical factors at the time of diagnosis, including: age, ECOG PS (a 4-level scale), LDH higher than Upper Limit of Normal, Ann Arbor Stage (a 4-level scale), systemic B symptoms, bulky mediastinum &amp;gt; 10 cm, number of Extra-Nodal sites (a 4-level scale: 0, 1, 2, or &amp;gt;=3), and pericardial or pleural effusion. Missing data in the evaluated predictors were multiple-imputed, and combined estimates were obtained from 20 imputed data sets. We used a multivariable logistic model with Backward Selection (level to stay p&amp;lt;0.1). Estimates were adjusted for front-line therapy. Standard errors were adjusted using the clustered sandwich estimator according to the patient's center. The model was internally validated with 1000 bootstrap samples from the original data. The optimism-corrected AUROC was estimated based on the results of the bootstrap process. We constructed a risk score by assigning each factor a weight proportional to the smallest coefficient, rounded to the nearest integer. The discrimination ability of the prognostic score was also evaluated considering the time to progression or relapse (Early Failure Survival) using the Kaplan-Meier method. Results Among the 869 patients included in the present study, 109 (12.5%) were identified as early failures (EF). Four factors were selected after the backward selection process to construct the prognostic score (AUROC 0.744, optimism-corrected 0.710). These factors included Systemic B-symptoms (OR=1.76, p=0.007, weight=1), Bulky mediastinum &amp;gt; 10 cm (OR=2.12, p=0.017, weight=1), AA Stage II (OR=2.20, p=0.018, weight=1), AA Stage III-IV (OR=2.73, p=0.017, weight=2), and Number of Extra-Nodal sites &amp;gt;=2 (OR=2.88, p&amp;lt;0.001, weight=2) (Table 1). Based on the incidence of EF, the score was classified into 3 classes: low (0-1) with 216 patients and an EF incidence of 3.7% (95% CI: 1.6% - 7.2%), intermediate (2-3) with 408 patients and an EF incidence of 11.8% (95% CI: 9.0% - 15.0%), and high (&amp;gt; 4) with 122 patients and an EF incidence of 30.8% (95% CI: 22.3% - 39.3%). (Figure 1A). At 24 months, Early-Failure-Survival (EFS) rates were as follows: High - 63.7% (95% CI: 54.5% - 71.6%), Intermediate - 87.0% (95% CI: 83.6% - 89.7%), and Low - 95.4% (95% CI: 91.5% - 97.5%). Conclusions Based on a large PMBL retrospective case series, we developed a simple and practical prognostic score subdividing patients into three clear-cut groups with different likelihoods of experiencing early failure. We acknowledge that the study has potential biases inherent in retrospective analysis, and further efforts aim to validate these results in external series of patients. If the predictive ability is confirmed, the score could be a valuable tool for treatment planning, interpretation, and comparison of clinical studies.

Practical Considerations for Sandwich Variance Estimation in 2-Stage Regression Settings.

In this paper, we present a practical approach for computing the sandwich variance estimator in 2-stage regression model settings. As a motivating example for 2-stage regression, we consider regression calibration, a popular approach for addressing covariate measurement error. The sandwich variance approach has rarely been applied in regression calibration, despite its requiring less computation time than popular resampling approaches for variance estimation, specifically the bootstrap. This is probably because it requires specialized statistical coding. Here we first outline the steps needed to compute the sandwich variance estimator. We then develop a convenient method of computation in R for sandwich variance estimation, which leverages standard regression model outputs and existing R functions and can be applied in the case of a simple random sample or complex survey design. We use a simulation study to compare the sandwich estimator to a resampling variance approach for both settings. Finally, we further compare these 2 variance estimation approaches in data examples from the Women's Health Initiative (1993-2005) and the Hispanic Community Health Study/Study of Latinos (2008-2011). In our simulations, the sandwich variance estimator typically had good numerical performance, but simple Wald bootstrap confidence intervals were unstable or overcovered in certain settings, particularly when there was high correlation between covariates or large measurement error.

Impact of augmented renal clearance on anticoagulant therapy in critically ill patients with coronavirus disease 2019: A retrospective cohort study

IntroductionThis study aimed to determine the impact of augmented renal clearance (ARC) on anticoagulation therapy in critically ill patients with coronavirus disease 2019 (COVID-19). MethodsThis retrospective cohort study included adult patients with severe COVID-19 with ARC who had been treated at our hospital between 2020 and 2021. We measured the estimated glomerular filtration rate calculated by the Chronic Kidney Disease Epidemiology Collaboration formula (eGFRCKD-EPI) every morning, and ARC condition was defined as eGFRCKD-EPI ≥ 130 mL/min/1.73 m2. Multivariate regression analysis with Huber–White sandwich estimator was performed to examine the association of unfractionated heparin (UH) dosage between blood test timings with activated partial thromboplastin time (APTT) compared with and without ARC. ResultsWe identified 38 enrolled patients: seven and 31 in the ARC and non-ARC groups, respectively. In the ARC coexisting condition, a higher dose of UH, which corresponded to the total dose in 24 h from the previous day, was required to achieve the same APTT prolongation, with a significant difference (p < 0.001). ConclusionsOur study suggests that careful monitoring and consideration of higher UH doses in critically ill patients with COVID-19 is necessary because anticoagulation failure can occur during ARC.

Standard error estimates in hierarchical generalized linear models

Hierarchical generalized linear models are often used to fit random effects models. However, attention is mostly paid to the estimation of fixed unknown parameters and inference for latent random effects. In contrast, standard error estimators receive less attention than they should be. Currently, the standard error estimators are based on various approximations, even when the mean parameters may be estimated from a higher-order approximation of the likelihood and the dispersion parameters are estimated by restricted maximum likelihood. Existing standard error estimation procedures are reviewed. A numerical illustration shows that the current standard errors are not necessarily accurate. Alternative standard errors are also proposed. In particular, a sandwich estimator that accounts for the dependence between the mean parameters and the dispersion parameters greatly improve the current standard errors.