Frequentist Model Research Articles

P642 Network meta-analysis: efficacy and safety of treatments for fistulising Crohn’s disease

Abstract Background Fistulas are a debilitating complication of Crohn’s disease (CD) and treatment options remain limited, with sub-optimal efficacy. It is essential that other treatment options are considered, and there is a lack of head-to-head comparisons between them making positioning treatments challenging. To our knowledge this is the first network meta-analysis to help position all treatments for fistulising CD. Methods MEDLINE, EMBASE, EMBASE classic, the Cochrane central registry and Cochrane Specialised Trial Register were searched between 1978 and 2022 for randomised controlled trials (RCTs) reporting on treatments for fistulising CD. The primary outcomes were the efficacy of medical therapies at achieving fistula remission and maintenance of fistula closure. A network meta-analysis was performed using the frequentist model with pooled relative risks and P scores used to rank treatments, in accordance to PRISMA guidelines [Figure 1]. Results 964 studies were identified with the search criteria, with 28 RCTs after screening included for analysis with a total of 2,239 patients. At the 8 – 12 week end-point, none of the interventions were statistically significant, with tacrolimus ranked first at 4 times better than placebo (RR, 4.56; 95% CI, 0.76 – 27.30) at inducing remission of fistulas [Figure 2]. At the 16 – 24 week time point, infliximab produced the only statistically significant result with the 5mg/kg dose proving most effective (RR, 2.30; 95% CI, 1.40 – 3.77) [Figure 3]. At 44 weeks ustekinumab was found to be most superior with it being 2.38 times (RR, 2.38; 95% CI, 1.24 – 4.56) more superior to placebo, with adalimumab (RR, 2.06; 95% CI, 1.06 – 3.99) and infliximab 5mg/kg (RR, 1.68; 95% CI, 1.03 – 2.75) also producing a statistically significant result [Figure 4]. A limitation of this network meta-analysis is the significant heterogeneity between studies and variations in end-point definition, with certain small numbers in certain intervention groups limiting generalisability. Conclusion Despite infliximab being favoured in international guidelines for the treatment of perianal fistulising CD, other biologics such as ustekinumab and adalimumab may have a role. Future RCTs with standardised end-points are needed to further investigate biologics, especially ustekinumab and adalimumab, head-to-head against infliximab in the treatment of fistulising CD.

The beneficial effect on cognition of noninvasive brain stimulation intervention in patients with dementia: a network meta-analysis of randomized controlled trials

BackgroundDementia [i.e., Alzheimer disease (AD)], the most common neurodegenerative disease, causes profound negative impacts on executive function and quality of life. Available pharmacological treatments often fail to achieve satisfactory outcomes. Noninvasive brain stimulation (NIBS) techniques, which focally modify cortical function and enhance synaptic long-term potentiation, are potentially beneficial for the cognition in patients with AD. The aim of the current network meta-analysis (NMA) was to evaluate the efficacy and safety of different NIBS interventions in patients with AD through NMA.MethodsOnly randomized controlled trials (RCTs) examining NIBS interventions in patients with AD had been included. All NMA procedures were performed under the frequentist model. The primary and secondary outcomes were changes in cognitive function and quality of life, respectively.ResultsNineteen RCTs (639 participants) were included. The mean treatment and follow-up durations were 5.7 and 10.5 weeks, respectively. The combination of cathodal tDCS of the left dorsolateral prefrontal cortex and anodal tDCS over the right supraorbital region (c-tDCS-F3 + a-tDCS-Fp2) was associated with a significant beneficial effect on cognition compared with sham controls (standardized mean difference=2.43, 95% confidence interval=0.61–4.26, n=12 and 11). It was also associated with the greatest beneficial effect on cognition among all the investigated NIBS approaches. All the methods were well tolerated with regard to the safety profile, as reflected in the rates of adverse events or local discomfort, as well as acceptability, as indicated by dropout rate.ConclusionsThe present findings provide evidence of the benefits of NIBS, especially tDCS, for beneficial effect on cognition in patients with AD. However, because of few studies included, this effect was not replicated yet in the other studies. Therefore, future larger-scale and longer follow-up duration RCTs should be warranted.Trial registrationPROSPERO CRD42020209516. The current study had been approved by the Institutional Review Board of the Tri-Service General Hospital, National Defense Medical Center (TSGHIRB No. B-109-29).

An introductory examination on the differences between frequentist and Bayesian multiple regression using real-world data on bias-based victimization among Latinx adults.

The aim of the current paper is to provide an applied introduction and overview of Bayesian methodology, how it compares from commonly used frequentist methods, and to elaborate on the utility of Bayesian methods in trauma and mental health research. Using data from the second wave of the Longitudinal Examination of Victimization Experiences of Latinos (LEVEL) study (N = 323) we ran frequentist modeling using OLS regression to test the effects of lifetime victimization, hate crime, and noncriminal bias events on anxiety, depression, anger, and dissociation. For the Bayesian analyses, we replicate these regressions using both weakly informative and highly informative priors, as well as a likelihood function that addresses data skew. Results across the 3 analyses present some key differences. In the frequentist models we find that lifetime victimization, hate crime, and noncriminal bias events had significant and positive relationship with anxiety, depression, and anger. Only hate crimes were significantly related to dissociation. The Bayesian results change based on which priors were implemented into the models. Ultimately, the results differ both across methodologies and within the Bayesian methodology depending on type of prior used. Several meaningful differences between the approaches emerge resulting in different interpretations of these results. Bayesian analyses serve as an additional tool for researchers that can be used to answer new and unique research questions that may be inaccessible by frequentist methods. (PsycInfo Database Record (c) 2022 APA, all rights reserved).

Diagnostic accuracy and added value of blood-based protein biomarkers for pancreatic cancer: A meta-analysis of aggregate and individual participant data.

Novel blood-based protein biomarkers may be of value for efficient, accurate, and non-invasive diagnosis of pancreatic cancer. This study assesses the diagnostic accuracy of newly recognized blood-based protein biomarkers for detecting pancreatic cancer, and investigates their added value to CA19-9, the common blood-based biomarker in clinical use for pancreatic cancer. PubMed, Embase, Web of Science, and the Wiley/Cochrane Library were systematically searched from inception until June 2022. A meta-analysis of aggregate and individual participant data was conducted using frequentist and Bayesian hierarchical random-effects models. The added clinical utility of protein biomarkers was investigated using bootstrap bias-corrected decision curve analyses. Aggregate data from 28 primary studies (6127 participants) were included, of which 8 studies (1790 participants) provided individual participant data. CA19-9 was significantly more accurate than MIC-1 for distinguishing pancreatic cancer from benign disease (AUC, 0.83 vs 0.74; relative diagnostic odds ratio [rDOR], 2.10 [95% CI, 0.98-4.48]; p=0.002), THBS2 (AUC, 0.87 vs 0.69; rDOR, 4.53 [2.16-9.39]; p<0.0001), TIMP-1 (AUC, 0.91 vs 0.70; rDOR, 8.00 [3.81-16.9]; p<0.0001), OPN (AUC, 0.89 vs 0.74; rDOR, 4.22 [1.13-15.6]; p<0.0001), ICAM-1 (AUC, 0.91 vs 0.68; rDOR 9.30 [0.87-99.5]; p<0.0001), and IGFBP2 (AUC, 0.91 vs 0.68; rDOR, 4.48 [0.78-24.3]; p<0.0001). The addition of these novel protein biomarkers to CA19-9 did not significantly improve the AUC, and resulted in minor increases or limited decreases in clinical utility. Novel protein biomarkers have moderate diagnostic accuracy, do not outperform CA19-9 in differentiating pancreatic cancer from benign disease, and show limited added clinical value to CA19-9. We propose recommendations to aid the development of minimally invasive diagnostic tests with sufficient clinical utility to improve the management of patients with suspected pancreatic cancer. Bennink Foundation, Dutch Cancer Foundation (KWF Kankerbestrijding), and AIRC.

Frequentist Model Averaging for the Nonparametric Additive Model

Frequentist Model Averaging for the Nonparametric Additive Model

Model averaging for generalized linear models with missing at random covariates

We consider a frequentist model averaging approach for prediction issue of generalized linear model with covariates involving missing at random. To avoid the uncertainty problem from imputation on the missing covariates data, this work adjusts model averaging criterion by selecting a batch of candidate models with the available covariate patterns rather than all possible combinations of covariates. The weights over the batch of candidate models are estimated based on the leave-one-out cross-validation criterion of the fully observed data. The asymptotic optimality of the proposed method is established under some regular conditions. Simulation studies are carried out to demonstrate the finite sample performance of our proposed approach in comparison to other methods. A real data set from NHANES 2017–2018 is also applied to check the effectiveness of the proposed approach.

Reproducibility in Small-N Treatment Research: A Tutorial Using Examples From Aphasiology.

Small-N studies are the dominant study design supporting evidence-based interventions in communication science and disorders, including treatments for aphasia and related disorders. However, there is little guidance for conducting reproducible analyses or selecting appropriate effect sizes in small-N studies, which has implications for scientific review, rigor, and replication. This tutorial aims to (a) demonstrate how to conduct reproducible analyses using effect sizes common to research in aphasia and related disorders and (b) provide a conceptual discussion to improve the reader's understanding of these effect sizes. We provide a tutorial on reproducible analyses of small-N designs in the statistical programming language R using published data from Wambaugh et al. (2017). In addition, we discuss the strengths, weaknesses, reporting requirements, and impact of experimental design decisions on effect sizes common to this body of research. Reproducible code demonstrates implementation and comparison of within-case standardized mean difference, proportion of maximal gain, tau-U, and frequentist and Bayesian mixed-effects models. Data, code, and an interactive web application are available as a resource for researchers, clinicians, and students. Pursuing reproducible research is key to promoting transparency in small-N treatment research. Researchers and clinicians must understand the properties of common effect size measures to make informed decisions in order to select ideal effect size measures and act as informed consumers of small-N studies. Together, a commitment to reproducibility and a keen understanding of effect sizes can improve the scientific rigor and synthesis of the evidence supporting clinical services in aphasiology and in communication sciences and disorders more broadly. Supplemental Material and Open Science Form: https://doi.org/10.23641/asha.21699476.

Comparative Effectiveness of Biologics Across Subgroups of Patients with Moderate-to-Severe Plaque Psoriasis: Results at Week12 from the PSoHO Study in a Real-World Setting.

In routine clinical care, important treatment outcomes among patients with moderate-to-severe plaque psoriasis (PsO) have been shown to vary according to patient demographics and disease characteristics. This study aimed to provide direct comparative effectiveness data at week12 between anti-interleukin (IL)-17A biologics relative to other approved biologics for the treatment of PsO across seven clinically relevant patient subgroups in the real-world setting. From the international, non-interventional Psoriasis Study of Health Outcomes (PSoHO), 1981 patients with moderate-to-severe PsO were grouped a priori according to seven clinically relevant demographic and disease variables with binary categories, which were sex (male or female), age (< 65 or ≥ 65years), body mass index (≤ 30or > 30kg/m2), race (White or Asian), PsO disease duration (< 15 or ≥ 15years), psoriatic arthritis (PsA) comorbidity (present or absent), and prior biologic use (never or ≥ 1). Across these subgroups, effectiveness was compared between the anti-IL-17A cohort (ixekizumab, secukinumab) versus all other approved biologics and ixekizumab versus five individual biologics. The proportion of patients in each subgroup who achieved 90% improvement in Psoriasis Area and Severity Index (PASI90) and/or static Physician Global Assessment (sPGA) 0/1, PASI100, or PASI90 at week12 were assessed. Comparative analyses were conducted using frequentist model averaging (FMA). Missing data were imputed using non-responder imputation. Patients in each of the seven subgroups achieved similar response rates to those of the overall treatment cohort, apart from patients with PsA treated with other biologics who had 7-10% lower response rates. Consequently, patients with comorbid PsA had significantly higher odds of achieving skin clearance at week12 with anti-IL-17A biologics compared to other biologics. Patients in all subgroups had significantly higher odds of achieving PASI90 and/or sPGA (0,1), PASI100, and PASI90 in the anti-IL-17A cohort relative to the other biologics cohort, except for the Asian subgroup. No sex- or age-specific differences in treatment effectiveness after 12weeks were identified, neither between the treatment cohorts nor between the individual treatment comparisons. Despite relative consistency of comparative treatment effectiveness across subgroups, the presence of comorbid PsA may affect a patient's clinical response to some treatments.

Frequentist model averaging in the generalized multinomial logit model

Frequentist model averaging in the generalized multinomial logit model

PP88 Bayesian Joint Models For Cost-Effectiveness Analyses Based On Clustered Participant Data, With Implementation In Stan

IntroductionCost-effectiveness analyses of empirical participant data are frequently complicated by irregularly distributed and correlated observations, which are not well approximated by normal distributions. Things get even more difficult when observations are clustered within higher level units (for example, hospitals) or the participant (that is, multiple measurements at different timepoints). Therefore, we developed a flexible Bayesian approach to jointly model costs and effects of two competing interventions with a multilevel structure.MethodsOur new model is presented in mathematical form and discussed in detail. We model costs and Quality-Adjusted Life-Years effects through Gamma and Beta distributions, and account for the dependency between costs and effects by adding the effects as a predictor for the costs. We further include hurdle models to account for costs of for the presence of zero costs and perfect health scores.The full model is implemented in the probabilistic programming language Stan. To compare the performance of our Bayesian model to a frequentist approach (linear mixed model combined with bootstrapping), we simulate 1000 datasets consisting of 400 participants and 20 clusters. Performance of both models is assessed in terms of variance, bias and coverage probability with respect to the costs and effects defined in the simulation.ResultsWe ran a preliminary simulation with high intraclass correlation, strong negative correlation for patient-level costs and effects, and positive correlation of cluster effects on both outcomes. The analysis shows that the Bayesian model exhibits a slightly larger bias for estimated costs, but smaller errors and higher coverage probability compared to the frequentist alternative. We will explore different scenarios where we vary the parameters of the simulations and assess whether the results are robust to change.ConclusionsIt is very important that economic evaluations in health care produce precise and reliable results. Our Bayesian approach is able to handle multiple statistical complexities at once and performs better than a comparable frequentist model. Whether this conclusion holds for different simulation scenarios will be explored in further stages of this study.

Estimates of fine fuel litter biomass in the northern Great Basin reveal increases during short fire-free intervals associated with invasive annual grasses

Exotic annual grasses invasion across northern Great Basin rangelands has promoted a grass-fire cycle that threatens the sagebrush (Artemisia spp.) steppe ecosystem. In this sense, high accumulation rates and persistence of litter from annual species largely increase the amount and continuity of fine fuels. Here, we highlight the potential use and transferability of remote sensing-derived products to estimate litter biomass on sagebrush rangelands in southeastern Oregon, and link fire regime attributes (fire-free period) with litter biomass spatial patterns at the landscape scale. Every June, from 2018 to 2021, we measured litter biomass in 24 field plots (60 m × 60 m). Two remote sensing-derived datasets were used to predict litter biomass measured in the field plots. The first dataset used was the 30-m annual net primary production (NPP) product partitioned into plant functional traits (annual grass, perennial grass, shrub, and tree) from the Rangeland Analysis Platform (RAP). The second dataset included topographic variables (heat load index -HLI- and site exposure index -SEI-) computed from the USGS 30-m National Elevation Dataset. Through a frequentist model averaging approach (FMA), we determined that the NPP of annual and perennial grasses, as well as HLI and SEI, were important predictors of field-measured litter biomass in 2018, with the model featuring a high overall fit (R2 = 0.61). Model transferability based on extrapolating the FMA predictive relationships from 2018 to the following years provided similar overall fits (R2 ≈ 0.5). The fire-free period had a significant effect on the litter biomass accumulation on rangelands within the study site, with greater litter biomass in areas where the fire-free period was <10 years. Our findings suggest that the proposed remote sensing-derived products could be a key instrument to equip rangeland managers with additional information towards fuel management, fire management, and restoration efforts.

Systemic Antibiotics as an Adjunct to Subgingival Debridement: A Network Meta-Analysis.

This review aimed to evaluate the effectiveness of systemic antibiotics as adjunctive treatment to subgingival debridement in patients with periodontitis. Randomized controlled trials were included that assessed the effectiveness of systemic antibiotics in improving periodontal status, indicated by clinical attachment gain level, probable pocket depth reduction, and bleeding on probing reduction of patients with any form of periodontitis at any follow-up time. Network meta-analyses with a frequentist model using random effects was employed to synthesize the data. The relative effects were reported as mean difference with a 95% confidence interval. Subsequently, all treatments were ranked based on their P-scores. A total of 30 randomized controlled trials were included in this network meta-analyses. Minimally important clinical differences were observed following the adjunctive use of satranidazole, metronidazole, and clindamycin for clinical attachment gain level and probable pocket depth reduction. For bleeding on probing reduction, minimally important clinical differences were observed following the adjunctive use of metronidazole and a combination of amoxycillin and metronidazole. However, the network estimates were supported by evidence with certainty ranging from very low to high. Therefore, the findings of this network meta-analyses should be interpreted with caution. Moreover, the use of these antibiotics adjunct to subgingival debridement should be weighed against possible harm to avoid overuse and inappropriate use of these antibiotics in patients with periodontitis.

A general framework for quantifying aleatoric and epistemic uncertainty in graph neural networks

Graph Neural Networks (GNN) provide a powerful framework that elegantly integrates Graph theory with Machine learning for modeling and analysis of networked data. We consider the problem of quantifying the uncertainty in predictions of GNN stemming from modeling errors and measurement uncertainty. We consider aleatoric uncertainty in the form of probabilistic links and noise in feature vector of nodes, while epistemic uncertainty is incorporated via a probability distribution over the model parameters. We propose a unified approach to treat both sources of uncertainty in a Bayesian framework, where Assumed Density Filtering is used to quantify aleatoric uncertainty and Monte Carlo dropout captures uncertainty in model parameters. Finally, the two sources of uncertainty are aggregated to estimate the total uncertainty in predictions of a GNN. Results in the real-world datasets demonstrate that the Bayesian model performs at par with a frequentist model and provides additional information about predictions uncertainty that are sensitive to uncertainties in the data and model.

Uncertainty-aware mixed-variable machine learning for materials design

Data-driven design shows the promise of accelerating materials discovery but is challenging due to the prohibitive cost of searching the vast design space of chemistry, structure, and synthesis methods. Bayesian optimization (BO) employs uncertainty-aware machine learning models to select promising designs to evaluate, hence reducing the cost. However, BO with mixed numerical and categorical variables, which is of particular interest in materials design, has not been well studied. In this work, we survey frequentist and Bayesian approaches to uncertainty quantification of machine learning with mixed variables. We then conduct a systematic comparative study of their performances in BO using a popular representative model from each group, the random forest-based Lolo model (frequentist) and the latent variable Gaussian process model (Bayesian). We examine the efficacy of the two models in the optimization of mathematical functions, as well as properties of structural and functional materials, where we observe performance differences as related to problem dimensionality and complexity. By investigating the machine learning models’ predictive and uncertainty estimation capabilities, we provide interpretations of the observed performance differences. Our results provide practical guidance on choosing between frequentist and Bayesian uncertainty-aware machine learning models for mixed-variable BO in materials design.

Some Right Ways to Analyze (Psycho)Linguistic Data

Much has been written on the abuse and misuse of statistical methods, including p values, statistical significance, and so forth. I present some of the best practices in statistics using a running example data analysis. Focusing primarily on frequentist and Bayesian linear mixed models, I illustrate some defensible ways in which statistical inference—specifically, hypothesis testing using Bayes factors versus estimation or uncertainty quantification—can be carried out. The key is to not overstate the evidence and to not expect too much from statistics. Along the way, I demonstrate some powerful ideas, including the use of simulation to understand the design properties of one's experiment before running it, visualization of data before carrying out a formal analysis, and simulation of data from the fitted model to understand the model's behavior.

Comprehensive analysis and prediction of long-term durability of factor IX activity following etranacogene dezaparvovec gene therapy in the treatment of hemophilia B

Objective Congenital hemophilia B is a rare bleeding disorder caused by defects in the gene encoding factor IX (FIX) leading to coagulation deficiency. Recurrent bleeds may cause chronic pain, disability, and reduced quality of life. Phase 2 b and 3 single-arm, open-label, single-dose trials assessing etranacogene dezaparvovec gene therapy for hemophilia B have demonstrated sustained FIX activity levels over observed periods, but long-term durability of the treatment effect has not been established. Using statistical modeling, we estimate long-term durability of FIX activity levels after receiving etranacogene dezaparvovec. Methods Participants from Phase 2 b (N = 3; NCT03489291) and 3 studies (N = 52; NCT03569891) were included. Two participants who did not respond to treatment were excluded. FIX activity was assessed by one-stage activated partial thromboplastin time assay. FIX activity levels at Month 6 post-treatment were considered baseline. Bayesian and Frequentist linear mixed models predicted FIX activity levels up to 25.5 years at an individual and population level with pre-treatment adeno-associated virus 5 (AAV5) neutralizing antibody (NAb) status as primary covariate. Results Bayesian and Frequentist linear mixed models predicted no more than 6/55 (10.91%) observed participants would have FIX activity levels <2% up to 25.5 years post-infusion. Bayesian model-based predictions of future participants suggest >80% would be free from prophylactic FIX replacement products 25.5 years post-infusion. Both models predicted FIX activity levels were not significantly influenced by pre-treatment AAV5 NAb status. Conclusions People with hemophilia B receiving etranacogene dezaparvovec would likely achieve durable FIX activity levels and remain free of prophylactic FIX replacement products for up to 25.5 years following single administration. The long-term factor IX durability predictions are based on statistical methods and results in vivo may differ.

Optimal Model Averaging of Mixed-Data Kernel-Weighted Spline Regressions

Model averaging has a rich history dating from its use for combining forecasts from time-series models (Bates and Granger) and presents a compelling alternative to model selection methods. We propose a frequentist model averaging procedure defined over categorical regression splines (Ma, Racine, and Yang) that allows for mixed-data predictors, as well as nonnested and heteroscedastic candidate models. We demonstrate the asymptotic optimality of the proposed model averaging estimator, and develop a post-averaging inference theory for it. Theoretical underpinnings are provided, finite-sample performance is evaluated, and an empirical illustration reveals that the method is capable of outperforming a range of popular model selection criteria in applied settings. An R package is available for practitioners (Racine).

Frequentist model averaging for zero‐inflated Poisson regression models

AbstractThis paper considers frequentist model averaging for estimating the unknown parameters of the zero‐inflated Poisson regression model. Our proposed weight choice procedure is based on the minimization of an unbiased estimator of a conditional quadratic loss function. We prove that the resulting model average estimator enjoys optimal asymptotic property and improves finite sample properties over the two commonly used information‐based model selection estimators and their model average estimators via simulation studies. The proposed method is illustrated by a real data example.

The Dose and Duration-dependent Association between Melatonin Treatment and Overall Cognition in Alzheimer's Dementia: A Network Meta- Analysis of Randomized Placebo-Controlled Trials.

While Alzheimer's dementia (AD) has a prevalence as high as 3-32% and is associated with cognitive dysfunction and the risk of institutionalization, no efficacious and acceptable treatments can modify the course of cognitive decline in AD. Potential benefits of exogenous melatonin for cognition have been divergent across trials. The current network meta-analysis (NMA) was conducted under the frequentist model to evaluate the potential beneficial effects of exogenous melatonin supplementation on overall cognitive function in participants with AD in comparison to other FDA-approved medications (donepezil, galantamine, rivastigmine, memantine, and Namzaric). The primary outcome was the changes in the cognitive function [measured by mini-mental state examination (MMSE)] after treatment in patients with Alzheimer's dementia. The secondary outcomes were changes in the quality of life, behavioral disturbance, and acceptability (i.e., drop-out due to any reason and rate of any adverse event reported). The current NMA of 50 randomized placebo-controlled trials (RCTs) revealed the medium-term lowdose melatonin to be associated with the highest post-treatment MMSE (mean difference = 1.48 in MMSE score, 95% confidence intervals [95% CIs] = 0.51 to 2.46) and quality of life (standardized mean difference = -0.64, 95% CIs = -1.13 to -0.15) among all of the investigated medications in the participants with AD. Finally, all of the investigated exogenous melatonin supplements were associated with similar acceptability as was the placebo. The current NMA provides evidence for the potential benefits of exogenous melatonin supplementation, especially medium-term low-dose melatonin, in participants with AD.

Allergen immunotherapy for atopic dermatitis: Systematic review and meta-analysis of benefits and harms

Atopic dermatitis (AD, eczema) is driven by a combination of skin barrier defects, immune dysregulation, and extrinsic stimuli such as allergens, irritants, and microbes. The role of environmental allergens (aeroallergens) in triggering AD remains unclear. We systematically synthesized evidence regarding the benefits and harms of allergen immunotherapy (AIT) for AD. As part of the 2022 American Academy of Allergy, Asthma & Immunology/American College of Allergy, Asthma and Immunology Joint Task Force on Practice Parameters AD Guideline update, we searched the MEDLINE, EMBASE, CENTRAL, CINAHL, LILACS, Global Resource for Eczema Trials, and Web of Science databases from inception to December 2021 for randomized controlled trials comparing subcutaneous immunotherapy (SCIT), sublingual immunotherapy (SLIT), and/or no AIT (placebo or standard care) for guideline panel-defined patient-important outcomes: AD severity, itch, AD-related quality of life (QoL), flares, and adverse events. Raters independently screened, extracted data, and assessed risk of bias in duplicate. We synthesized intervention effects using frequentist and Bayesian random-effects models. The GRADE approach determined the quality of evidence. Twenty-three randomized controlled trials including 1957 adult and pediatric patients sensitized primarily to house dust mite showed that add-on SCIT and SLIT have similar relative and absolute effects and likely result in important improvements in AD severity, defined as a 50% reduction in SCORing Atopic Dermatitis (risk ratio [95% confidence interval] 1.53 [1.31-1.78]; 26% vs 40%, absolute difference 14%) and QoL, defined as an improvement in Dermatology Life Quality Index by 4 points or more (risk ratio [95% confidence interval] 1.44 [1.03-2.01]; 39% vs 56%, absolute difference 17%; both outcomes moderate certainty). Both routes of AIT increased adverse events (risk ratio [95% confidence interval] 1.61 [1.44-1.79]; 66% with SCIT vs 41% with placebo; 13% with SLIT vs 8% with placebo; high certainty). AIT's effect on sleep disturbance and eczema flares was very uncertain. Subgroup and sensitivity analyses were consistent with the main findings. SCIT and SLIT to aeroallergens, particularly house dust mite, can similarly and importantly improve AD severity and QoL. SCIT increases adverse effects more than SLIT. These findings support a multidisciplinary and shared decision-making approach to optimally managing AD.