Search

INTRODUCTION: Missing age presents a significant challenge when evaluating individual case safety reports (ICSRs) in the FDA Adverse Event Reporting System (FAERS). When age is missing in an ICSR’s structured field, it may be in the report’s free-text narrative. OBJECTIVES: This study aimed to evaluate the performance and assess the potential impact of a rule-based natural language processing (NLP) tool that utilizes a text string search to identify patients’ numerical age from unstructured narratives. METHODS: Using FAERS ICSRs from 2002 to 2018, we evaluated the annual proportion of ICSRs with age missing in the structured field before and after NLP application. Reviewers manually identified patients’ age from ICSR narratives (gold standard) from a random sample of 1500 ICSRs. The gold standard was compared to the NLP-identified age. RESULTS: During the study period, the percentage of ICSRs missing age in the structured field increased from 21.9 to 43.8%. The NLP tool performed well among the random sample: sensitivity 98.5%, specificity 92.9%, positive predictive value (PPV) 94.9%, and F-measure 96.7%. It also performed well for the subset of ICSRs missing age in the structured field; when applied to these cases, NLP identified age for an additional one million ICSRs (10% of the total number of ICSRs from 2002 to 2018) and decreased the percentage of ICSRs missing age to 27% overall. CONCLUSIONS: NLP has potential utility to extract patients’ age from ICSR narratives. Use of this tool would enhance pharmacovigilance and research using FAERS data.

The US Food and Drug Administration Adverse Event Reporting System (FAERS) contains over 24 million individual case safety reports (ICSRs). In this research project, we evaluated a natural language processing (NLP) tool’s ability to extract four demographic variables (gender, weight, ethnicity, race) from ICSR narratives. Specificity of the NLP algorithm was over 94% for all demographics, while sensitivity varied between the demographics: 98.6% (gender), 45.5% (weight), 100% (ethnicity), and 85.3% (race). Among ICSRs missing weight, ethnicity, and race in the structured field, few cases had this information in the narrative (>95% missing); consequently, the positive predictive value (PPV) for these three demographics had wide 95% confidence intervals. After NLP implementation, the total number of ICSRs missing gender was reduced by 33% (i.e., NLP identified 472 thousand reports having a gender value in the narrative that was not in the structured field), while the total number of ICSRs missing weight, ethnicity, or race was reduced by less than 4%. This study demonstrated that the implementation of an NLP tool can provide meaningful improvements in the availability of gender information for pharmacovigilance activities conducted with FAERS data. In contrast, NLP tools targeting the extraction of weight, ethnicity, or race from free-text fields have minimal impact largely because the information was infrequently provided by the reporter. Further gains in completeness of these fields must originate from increases in provision of demographic information from the reporter rather than informatic solutions.

BACKGROUND: A previous FDA study reported a favorable benefit risk for apixaban compared with warfarin for stroke prevention in older non-valvular atrial fibrillation (NVAF) patients (≥ 65 years). However, it remains unclear whether this favorable benefit risk persists in other populations including younger users. We examined if a similar benefit risk was observed in the Sentinel System and if it varied by age group. OBJECTIVE: To examine the risk of ischemic stroke, gastrointestinal (GI) bleeding, and intracranial hemorrhage (ICH) in apixaban users compared with warfarin users in Sentinel Distributed Database (SDD). DESIGN AND PARTICIPANTS: A retrospective new user cohort study was conducted among patients, 21 years and older initiating apixaban and warfarin for NVAF, between December 28, 2012, and June 30, 2018, in the SDD. MAIN MEASURES: Cox proportional hazard regression was used to estimate the hazard ratios (HR) and 95% confidence intervals (95% CI) for each outcome (ischemic stroke, GI bleeding, and ICH) in propensity score matched apixaban users compared with the warfarin users. Subgroup analyses by age (21–64, 65–74, and 75+ years) were conducted. KEY RESULTS: After matching, 55.3% and 58.4% (n = 55,038) of the apixaban and warfarin users were included in the main analysis. GI bleeding was the most common outcome. The HR (95% CI) for GI bleeding, ICH, and ischemic stroke in apixaban users compared with warfarin users were 0.57 (0.50–0.66), 0.53 (0.40–0.70), and 0.56 (0.45–0.71) respectively. The reduced risk of these outcomes in apixaban compared with warfarin users persisted across age groups. CONCLUSION: In NVAF patients of all ages initiating either apixaban or warfarin for stroke prevention in the Sentinel System, apixaban was associated with a decreased risk of GI bleeding, ICH, and ischemic stroke compared with warfarin. Among patients less than 65 years of age, apixaban use was associated with a decreased risk of GI bleeding and ischemic stroke. ELECTRONIC SUPPLEMENTARY MATERIAL: The online version of this article (10.1007/s11606-020-06180-8) contains supplementary material, which is available to authorized users.

Only about half of patients with atrial fibrillation (AF) who are at increased risk for stroke are treated with an oral anticoagulant (OAC), despite guideline recommendations for their use. Educating patients with AF about prevention of stroke with OACs may enable them as agents of change to initiate OAC treatment.To determine whether an educational intervention directed to patients and their clinicians stimulates the use of OACs in patients with AF who are not receiving OACs.The Implementation of a Randomized Controlled Trial to Improve Treatment With Oral Anticoagulants in Patients With Atrial Fibrillation (IMPACT-AFib) trial was a prospective, multicenter, open-label, pragmatic randomized clinical trial conducted from September 25, 2017, to May 1, 2019, embedded in health plans that participate in the US Food and Drug Administration's Sentinel System. It used the distributed database comprising health plan members to identify eligible patients, their clinicians, and outcomes. IMPACT-AFib enrolled patients with AF, a CHA2DS2-VASc (cardiac failure or dysfunction, hypertension, age 65-74 [1 point] or ≥75 years [2 points], diabetes, and stroke, transient ischemic attack or thromboembolism [2 points]-vascular disease, and sex category [female]) score of 2 or more, no evidence of OAC prescription dispensing in the preceding 12 months, and no hospitalization-related bleeding event within the prior 6 months.Randomization to a single mailing of patient and/or clinician educational materials vs control.Analysis was performed on a modified intention-to-treat basis. The primary end point was the proportion of patients with at least 1 OAC prescription dispensed or at least 4 international normalized ratio test results within 1 year of the intervention.Among 47 333 patients, there were 24 909 men (52.6%), the mean (SD) age was 77.9 (9.7) years, mean (SD) CHA2DS2-VASc score was 4.5 (1.7), 22 404 patients (47.3%) had an ATRIA (Anticoagulation and Risk Factors in Atrial Fibrillation) bleeding risk score of 5 or more, and 8890 patients (18.8%) had a history of hospitalization for bleeding. There were 2328 of 23 546 patients (9.9%) in the intervention group with initiation of OAC at 1 year compared with 2330 of 23 787 patients (9.8%) in the control group (adjusted OR, 1.01 [95% CI, 0.95-1.07]; P = .79).Among a large population with AF with a guideline indication for OACs for stroke prevention who were randomized to a mailed educational intervention or to usual care, there was no clinically meaningful, numerical, or statistically significant difference in rates of OAC initiation. More-intensive interventions are needed to try and address the public health issue of underuse of anticoagulation for stroke prevention among patients with AF.ClinicalTrials.gov Identifier: NCT03259373.

Observational studies provide a core resource in assessing post-market drug safety and effectiveness. Propensity scores are a predominant method for confounding adjustment to achieve unbiased estimation of average treatment effects in observational data. However, the use of propensity score methods has been limited to comparing two treatment groups, while medical situations frequently present with multiple treatment options. Inverse probability of treatment weighting (IPTW) is a popular propensity score adjustment method, but its performance degrades with decreased positivity leading to extreme weights, a problem that can be amplified with multiple treatment groups. Meanwhile, regression on a spline of the propensity score has shown favorable performance compared to other propensity score methods in recent studies involving two treatments. This project utilizes a simulation study to compare IPTW and propensity score splines as adjustment methods in a three-treatment setting. We test a variety of spline methods, including natural cubic splines with varying numbers of interior knots, and thin-plate regression splines. We vary several parameters across simulations, including the degree of propensity score overlap among treatment groups, treatment prevalence, outcome prevalence, and true marginal relative risk. We assess methods based on their bias, root mean squared error, and coverage of the true marginal relative risk across simulations. We find that all methods perform similarly well when there is good propensity score distribution overlap. However, with even moderate decrease in overlap or low outcome prevalence, IPTW produces more biased estimates and higher variance than propensity score splines. Low treatment prevalence or unequal treatment prevalences across groups also worsens IPTW performance. Overall, a natural cubic spline with a relatively small number of interior knots provides good performance across a range of simulations.

Background— The comparative safety of dabigatran versus warfarin for treatment of nonvalvular atrial fibrillation in general practice settings has not been established. Methods and Results— We formed new-user cohorts of propensity score–matched elderly patients enrolled in Medicare who initiated dabigatran or warfarin for treatment of nonvalvular atrial fibrillation between October 2010 and December 2012. Among 134 414 patients with 37 587 person-years of follow-up, there were 2715 primary outcome events. The hazard ratios (95% confidence intervals) comparing dabigatran with warfarin (reference) were as follows: ischemic stroke, 0.80 (0.67–0.96); intracranial hemorrhage, 0.34 (0.26–0.46); major gastrointestinal bleeding, 1.28 (1.14–1.44); acute myocardial infarction, 0.92 (0.78–1.08); and death, 0.86 (0.77–0.96). In the subgroup treated with dabigatran 75 mg twice daily, there was no difference in risk compared with warfarin for any outcome except intracranial hemorrhage, in which case dabigatran risk was reduced. Most patients treated with dabigatran 75 mg twice daily appeared not to have severe renal impairment, the intended population for this dose. In the dabigatran 150-mg twice daily subgroup, the magnitude of effect for each outcome was greater than in the combined-dose analysis. Conclusions— In general practice settings, dabigatran was associated with reduced risk of ischemic stroke, intracranial hemorrhage, and death and increased risk of major gastrointestinal hemorrhage compared with warfarin in elderly patients with nonvalvular atrial fibrillation. These associations were most pronounced in patients treated with dabigatran 150 mg twice daily, whereas the association of 75 mg twice daily with study outcomes was indistinguishable from warfarin except for a lower risk of intracranial hemorrhage with dabigatran.

We address several questions relating to the use of standard regression and Structural Nested Mean Model (SNMM) approach (e. g., Ten Have et al. 2007) to analyze post-randomization effect modifiers of the intent-to-treat effect of a randomized intervention on a subsequent outcome, which has not been well examined. We show through simulations that the SNMM performs better with respect to bias of estimates of the intervention and interaction effects than does the corresponding standard interaction approach when the baseline intervention is randomized and the post-randomization factors are subject to confounding, and even when there is no association between the intervention and effect modifier. However, causal inference under the SNMM makes untestable assumptions that the causal contrasts do not vary across observed levels of the intervention and post-randomization factor. In addition, the precision of the SNMM-based estimators depends on the effect of the randomized intervention on the post-randomization factor varying across baseline covariate combinations. These issues and methods are illustrated with the application of the standard and causal methods to a randomized cognitive therapy (CT) trial, for which there is a conceptual model of negative cognitive styles or distortions impacted by CT but then in turn modifying the effect of CT on subsequent suicide ideation and social problem solving outcomes.

OBJECTIVE The cardiovascular safety of saxagliptin, a dipeptidyl-peptidase 4 inhibitor, compared with other antihyperglycemic treatments is not well understood. We prospectively examined the association between saxagliptin use and acute myocardial infarction (AMI). RESEARCH DESIGN AND METHODS We identified patients aged ≥18 years, starting from the approval date of saxagliptin in 2009 and continuing through August 2014, using data from 18 Mini-Sentinel data partners. We conducted seven sequential assessments comparing saxagliptin separately with sitagliptin, pioglitazone, second-generation sulfonylureas, and long-acting insulin, using disease risk score (DRS) stratification and propensity score (PS) matching to adjust for potential confounders. Sequential testing kept the overall chance of a false-positive signal below 0.05 (one-sided) for each pairwise comparison. RESULTS We identified 82,264 saxagliptin users and more than 1.5 times as many users of each comparator. At the end of surveillance, the DRS-stratified hazard ratios (HRs) (95% CI) were 1.08 (0.90–1.28) in the comparison with sitagliptin, 1.11 (0.87–1.42) with pioglitazone, 0.79 (0.64–0.98) with sulfonylureas, and 0.57 (0.46–0.70) with long-acting insulin. The corresponding PS-matched HRs were similar. Only one interim analysis of 168 analyses met criteria for a safety signal: the PS-matched saxagliptin-pioglitazone comparison from the fifth sequential analysis, which yielded an HR of 1.63 (1.12–2.37). This association diminished in subsequent analyses. CONCLUSIONS We did not find a higher AMI risk in saxagliptin users compared with users of other selected antihyperglycemic agents during the first 5 years after U.S. Food and Drug Administration approval of the drug.

A single nucleotide polymorphism, � 160C/A, has been identified in the promoter region of the E-cadherin gene and has been shown to alter its transcriptional activity. To assess susceptibility of � 160A allele carriers to seven types of cancers, the authors conducted a comprehensive meta-analysis, up to November 2006, of 26 case-control studies comprising 7,042 cases and 7,011 controls. Pooled odds ratios and 95% confidence intervals were calculated by using the random-effects model. Publication bias, subgroup, and sensitivity analyses were also performed, which showed that � 160A allele carriers, compared with noncarriers, had about a 17–19% increased risk of several invasive/ metastatic tumors. Analyses of various types of cancers revealed that, in Europeans, the � 160AA homozygote was associated with an increased risk of urothelial cancer, carriers of � 160A were at increased risk of lung and prostate cancers, and carriers of � 160A with gastric cancer were found to suffer a significantly increased risk, whereas their Asian counterparts seemed to be tolerant. No evidence was found that the � 160A allele predisposed its carriers to breast, colorectal, or esophageal cancers. These findings indicate that � 160A of the E-cadherin gene is emerging as a low-penetrance tumor susceptibility allele for the development of gastric, lung, prostate, and urothelial cancers.

Uncovering genetic control of variation in ethanol tolerance in natural populations of yeast Saccharomyces cerevisiae is essential for understanding the evolution of fermentation, the dominant lifestyle of the species, and for improving efficiency of selection for strains with high ethanol tolerance, a character of great economic value for the brewing and biofuel industries. To date, as many as 251 genes have been predicted to be involved in influencing this character. Candidacy of these genes was determined from a tested phenotypic effect following gene knockout, from an induced change in gene function under an ethanol stress condition, or by mutagenesis. This article represents the first genomics approach for dissecting genetic variation in ethanol tolerance between two yeast strains with a highly divergent trait phenotype. We developed a simple but reliable experimental protocol for scoring the phenotype and a set of STR/SNP markers evenly covering the whole genome. We created a mapping population comprising 319 segregants from crossing the parental strains. On the basis of the data sets, we find that the tolerance trait has a high heritability and that additive genetic variance dominates genetic variation of the trait. Segregation at five QTL detected has explained ∼50% of phenotypic variation; in particular, the major QTL mapped on yeast chromosome 9 has accounted for a quarter of the phenotypic variation. We integrated the QTL analysis with the predicted candidacy of ethanol resistance genes and found that only a few of these candidates fall in the QTL regions.

Condor provides high throughput computing by leveraging idle-cycles on off-the-shelf desktop machines. It also supports flocking, a mechanism for sharing resources among Condor pools. Since Condor pools distributed over a wide area can have dynamically changing availability and sharing preferences, the current flocking mechanism based on static configurations can limit the potential of sharing resources across Condor pools. This paper presents a technique for resource discovery in distributed Condor pools using peer-to-peer mechanisms that are self-organizing, fault-tolerant, scalable, and locality-aware. Locality-awareness guarantees that applications are not shipped across long distances when nearby resources are available. Measurements using a synthetic job trace show that self-organized flocking reduces the maximum job wait time in queue for a heavily loaded pool by a factor of 10 compared to without flocking. Simulations of 1000 Condor pools are also presented and the results confirm that our technique discovers and utilizes nearby resources in the physical network.

Allozyme and PCR-based molecular markers have been widely used to investigate genetic diversity and population genetic structure in autotetraploid species. However, an empirical but inaccurate approach was often used to infer marker genotype from the pattern and intensity of gel bands. Obviously, this introduces serious errors in prediction of the marker genotypes and severely biases the data analysis. This article developed a theoretical model to characterize genetic segregation of alleles at genetic marker loci in autotetraploid populations and a novel likelihood-based method to estimate the model parameters. The model properly accounts for segregation complexities due to multiple alleles and double reduction at autotetrasomic loci in natural populations, and the method takes appropriate account of incomplete marker phenotype information with respect to genotype due to multiple-dosage allele segregation at marker loci in tetraploids. The theoretical analyses were validated by making use of a computer simulation study and their utility is demonstrated by analyzing microsatellite marker data collected from two populations of sycamore maple (Acer pseudoplatanus L.), an economically important autotetraploid tree species. Numerical analyses based on simulation data indicate that the model parameters can be adequately estimated and double reduction is detected with good power using reasonable sample size.

An increased risk of depressive symptoms has been associated with the transition to menopause, but the risk of depressive symptoms in the early postmenopausal years has not been well characterized.To identify within-woman changes in depressive symptoms during a 14-year period around menopause, determine associations of a history of depression with the pattern of depressive symptoms, and evaluate the rate of change in reproductive hormones as predictors of depressive symptoms following menopause.A randomly identified, population-based sample in Philadelphia County, Pennsylvania, of 203 late-reproductive-age women who were premenopausal at baseline and reached natural menopause.Center for Epidemiologic Studies Depression Scale.The prevalence of high scores on the Center for Epidemiologic Studies Depression Scale decreased from 10 years before to 8 years after the final menstrual period (FMP), with a decrease of approximately 15% of baseline per year (odds ratio, 0.85; 95% CI, 0.81-0.89; P .001). Relative to the FMP, the risk of depressive symptoms was higher in the years before and lower in the years after the FMP. Among women with a history of depression, the likelihood of depressive symptoms was more than 13 times greater overall and 8 times greater after menopause compared with women with no depression history. Among women who first experienced depressive symptoms approaching menopause, the risk of depressive symptoms declined after the FMP, with a significantly lower risk the second year after menopause. The risk of depressive symptoms after menopause decreased by 35% for each unit (SD) increase before the FMP in the log rate of change of follicle-stimulating hormone (odds ratio, 0.65; 95% CI, 0.46-0.91; P = .01).The FMP was pivotal in the overall pattern of decreasing depressive symptoms in midlife women, with higher risk before and lower risk after the FMP. A history of depression strongly increased the risk both before and after menopause. Women who had no history of depression before the menopause transition had a low risk of depressive symptoms 2 or more years after the FMP.

Internet distance prediction gives pair-wise latency information with limited measurements. Recent studies have revealed that the quality of existing prediction mechanisms from the application perspective is short of satisfactory. In this paper, we explore the root causes and remedies for this problem. Our experience with different landmark selection schemes shows that although selecting nearby landmarks can increase the prediction accuracy for short distances, it can cause the prediction accuracy for longer distances to degrade. Such uneven prediction quality significantly impacts application performance. Instead of trying to select the landmark nodes in some "intelligent" fashion, we propose a hierarchical prediction approach with straightforward landmark selection. Hierarchical prediction utilizes multiple coordinate sets at multiple distance scales, with the "right" scale being chosen for prediction each time. Experiments with Internet measurement datasets show that this hierarchical approach is extremely promising for increasing the accuracy of network distance prediction.

An international consortium has launched the whole-genome sequencing of potato, the fourth most important food crop in the world. Construction of genetic linkage maps is an inevitable step for taking advantage of the genome projects for the development of novel cultivars in the autotetraploid crop species. However, linkage analysis in autopolyploids, the kernel of linkage map construction, is theoretically challenging and methodologically unavailable in the current literature. We present here a theoretical analysis and a statistical method for tetrasomic linkage analysis with dominant and/or codominant molecular markers. The analysis reveals some essential properties of the tetrasomic model. The method accounts properly for double reduction and incomplete information of marker phenotype in regard to the corresponding phenotype in estimating the coefficients of double reduction and recombination frequency and in testing their significance by using the marker phenotype data. Computer simulation was developed to validate the analysis and the method and a case study with 201 AFLP and SSR markers scored on 228 full-sib individuals of autotetraploid potato is used to illustrate the utility of the method in map construction in autotetraploid species.

Publish/Subscribe (pub/sub) is an important paradigm for distributed content delivery. Traditionally, there have been two approaches to supporting pub/sub service: subject-based and content-based. Content-based pub/sub allows fine-grained expressiveness, and thus is a more attractive solution for content dissemination. However, the performance of a content-based pub/sub network is bounded by the expensive matching cost of content messages. In this paper, we propose a hybrid approach capable of minimizing both the matching and forwarding overhead within the pub/sub network and the delay experienced by clients receiving the content. The hybrid approach aims to eliminate redundant matching and forwarding inside the pub/sub network. In particular, it identifies a number of virtual groups by exploring common subscription interests among clients, and messages for each virtual group are only matched once at the group entry point. In addition, for each virtual group, the content delivery tree embedded in the underlying pub/sub network can benefit from shortcutting forwarding-only paths. Simulations have shown that the hybrid approach is highly effective in improving the service efficiency and quality of a content-based pub/sub system

Purpose: Reduced expression of the transforming growth factor β receptor type II (TGFβRII), a key inhibitor of epithelial cell growth and tumor suppressor gene, was reported frequently in many types of tumors including non-small cell lung cancer (NSCLC). This study explored the significance of the TGFβRII gene in NSCLC carcinogenesis. Experimental Design: With 43 independent pairs of tumor and paracarcinoma tissue samples from patients with primary NSCLC, we carried out PCR-denaturing gradient gel electrophoresis screening for DNA variants over the coding sequence of the TGFβRII gene, immunohistochemical assay of TGFβRII expression, methylation-specific PCR analysis, and semiquantitative reverse transcription-PCR. Results: The PCR-denaturing gradient gel electrophoresis did not detect variation in the whole coding sequence of the TGFβRII gene, but the immunohistochemistry experiment revealed reduced or lost expression of the gene in 44% (19 of 43) of the tumor samples. The methylation analysis on the 19 pairs detected the frequent occurrence of methylated TGFβRII promoter in tumor tissues, whereas most of the paracarcinoma tissues were free of methylation. The reduced TGFβRII expression was highly significantly associated with the methylation event (P < 10−4). The reverse transcription-PCR analysis demonstrated a clear agreement between reduced TGFβRII expression and decreased mRNA level of the gene in the tumor tissue samples. Conclusions: TGFβRII plays an important role as a tumor suppressor in NSCLC carcinogenesis. The defective expression may serve as one of most important molecular mechanisms in explaining progression of the disease. In particular, aberrant 5′ CpG methylation of the gene has explained the down-regulation of the gene at a transcriptional level.

CC chemokine receptor 5 (CCR5) is a member of the G-protein-coupled receptor superfamily. It plays an important role in macrophage tropic human immunodeficiency virus-1 entry and in some inflammatory reactions. CCR5-893(-) is a single-nucleotide deletion that results in complete truncation of the C tail of the gene product. We detected CCR5-893(-) in a sample of patients infected with non-tuberculosis mycobacteria and found that it was maintained heterozygously with a frequency of 2%. There is no association between this mutation and any immunodeficiency. Membrane expression of CCR5-893(-) was substantially reduced compared to the wild type, but this defective surface presentation recovered greatly recovered in the presence of 2 mg l(-1) phytohemagglutinin (PHA). However, PHA inducement did not affect the total intracellular expression of CCR5-893(-) or wild-type CCR5. Thus we suggest there exist some PHA-induced factor(s) that could mediate the presentation of truncated CCR5.

This paper advocates implementing anycast in peer-to-peer (p2p) overlay networks. We argue that anycast in p2p overlays (exemplified by Pastry, Tapestry, Chord, CAN) combines the advantages of IP anycast and existing application-layer anycast services. We show that anycast can leverage the locality-awareness embedded in existing p2p overlays. The locality-awareness of the p2p routing substrates can be extended to support anycast and anycast is achieved as the result of the enhanced locality-aware routing. We have implemented anycast in Pastry, and experiments confirm that with high probability, a message addressed to an anycast group is routed to the closest node in the group according to the proximity metric. We also evaluate the performance of anycast using a realistic failure trace and the results show that our implementation of anycast is resilient to node failures.

The average length of genes in a eukaryote is larger than in a prokaryote, implying that evolution of complexity is related to change of gene lengths. Here, we show that although the average lengths of genes in prokaryotes and eukaryotes are much different, the average lengths of genes are highly conserved within either of the two kingdoms. This suggests that natural selection has clearly set a strong limitation on gene elongation within the kingdom. Furthermore, the average gene size adds another distinct characteristic for the discrimination between the two kingdoms of organisms.

In this paper, we consider large-scale high-bandwidth on-demand media streaming in a dynamic and heterogeneous environment. We present MetaStream, a scalable and distributed content discovery protocol that enables clients across the Internet to self-organize into a topology-aware overlay network, in which they can “cache and relay” a stream among nearby peers. We present the design and implementation of MetaStream and show experimental results obtained in a large-scale emulated environment. Our evaluation shows that MetaStream distributes relaying load among the clients in a topology-aware manner, imposing low link cost to the underlying network and low streaming load on the media server. MetaStream is also highly resilient to node or network failures and is capable of quickly recovering the streaming quality at clients even at high failure rates.