The objective of this simulation study is to compare the accuracy and efficiency of population-averaged (i.e. generalized estimating equations (GEE)) and cluster-specific (i.e. random-effects logistic regression (RELR)) models for analyzing data from cluster randomized trials (CRTs) with missing binary responses.
In this simulation study, clustered responses were generated from a beta-binomial distribution. The number of clusters per trial arm, the number of subjects per cluster, intra-cluster correlation coefficient, and the percentage of missing data were allowed to vary. Under the assumption of covariate dependent missingness, missing outcomes were handled by complete case analysis, standard multiple imputation (MI) and within-cluster MI strategies. Data were analyzed using GEE and RELR. Performance of the methods was assessed using standardized bias, empirical standard error, root mean squared error (RMSE), and coverage probability.
GEE performs well on all four measures — provided the downward bias of the standard error (when the number of clusters per arm is small) is adjusted appropriately — under the following scenarios: complete case analysis for CRTs with a small amount of missing data; standard MI for CRTs with variance inflation factor (VIF) <3; within-cluster MI for CRTs with VIF≥3 and cluster size>50. RELR performs well only when a small amount of data was missing, and complete case analysis was applied.
GEE performs well as long as appropriate missing data strategies are adopted based on the design of CRTs and the percentage of missing data. In contrast, RELR does not perform well when either standard or within-cluster MI strategy is applied prior to the analysis.
Marginal model; Population-averaged model; Cluster-specific model; Multiple imputation; Cluster randomized trial; Covariate dependent missingness; Generalized estimating equations; Random-effects logistic regression
To perform a meta-analysis of gene expression microarray data from animal studies of lung injury, and to identify an injury-specific gene expression signature capable of predicting the development of lung injury in humans.
We performed a microarray meta-analysis using 77 microarray chips across six platforms, two species and different animal lung injury models exposed to lung injury with or/and without mechanical ventilation. Individual gene chips were classified and grouped based on the strategy used to induce lung injury. Effect size (change in gene expression) was calculated between non-injurious and injurious conditions comparing two main strategies to pool chips: (1) one-hit and (2) two-hit lung injury models. A random effects model was used to integrate individual effect sizes calculated from each experiment. Classification models were built using the gene expression signatures generated by the meta-analysis to predict the development of lung injury in human lung transplant recipients.
Two injury-specific lists of differentially expressed genes generated from our meta-analysis of lung injury models were validated using external data sets and prospective data from animal models of ventilator-induced lung injury (VILI). Pathway analysis of gene sets revealed that both new and previously implicated VILI-related pathways are enriched with differentially regulated genes. Classification model based on gene expression signatures identified in animal models of lung injury predicted development of primary graft failure (PGF) in lung transplant recipients with larger than 80% accuracy based upon injury profiles from transplant donors. We also found that better classifier performance can be achieved by using meta-analysis to identify differentially-expressed genes than using single study-based differential analysis.
Taken together, our data suggests that microarray analysis of gene expression data allows for the detection of “injury" gene predictors that can classify lung injury samples and identify patients at risk for clinically relevant lung injury complications.
Glycemic index (GI) and glycemic load (GL) have been associated with coronary heart disease (CHD) risk in some but not all cohort studies. We therefore assessed the association of GI and GL with CHD risk in prospective cohorts.
Methods and Results
We searched MEDLINE, EMBASE, and CINAHL (through April 5, 2012) and identified all prospective cohorts assessing associations of GI and GL with incidence of CHD. Meta-analysis of observational studies in epidemiology (MOOSE) methodologies were used. Relative measures of risk, comparing the group with the highest exposure (mean GI of cohorts=84.4 GI units, range 79.9 to 91; mean GL of cohorts=224.8, range 166 to 270) to the reference group (mean GI=72.3 GI units, range 68.1 to 77; mean GL=135.4, range 83 to 176), were pooled using random-effects models, expressed as relative risk (RR) with heterogeneity assessed by χ2 and quantified by I2. Subgroups included sex and duration of follow-up. Ten studies (n=240 936) were eligible. Pooled analyses showed an increase in CHD risk for the highest GI quantile compared with the lowest, with RR=1.11 (95% confidence interval [CI] 0.99 to 1.24) and for GL, RR=1.27 (95% CI 1.09 to 1.49), both with evidence of heterogeneity (I2>42%, P<0.07). Subgroup analyses revealed only a significant modification by sex, with the female cohorts showing significance for GI RR=1.26 (95% CI 1.12 to 1.41) and for GL RR=1.55 (95% CI 1.18 to 2.03).
High GI and GL diets were significantly associated with CHD events in women but not in men. Further studies are required to determine the relationship between GI and GL with CHD in men.
coronary heart disease; glycemic index and load; meta-analysis; nutrition; prospective cohort
Intraclass correlation coefficients (ICCs) are used in a wide range of applications. However, most commonly used estimators for the ICC are known to be subject to bias.
Using second order Taylor series expansion, we propose a new bias-corrected estimator for one type of intraclass correlation coefficient, for the ICC that arises in the context of the balanced one-way random effects model. A simulation study is performed to assess the performance of the proposed estimator. Data have been generated under normal as well as non-normal scenarios.
Our simulation results show that the new estimator has reduced bias compared to the least square estimator which is often referred to as the conventional or analytical estimator. The results also show marked bias reduction both in normal and non-normal data scenarios. In particular, our estimator outperforms the analytical estimator in a non-normal setting producing estimates that are very close to the true ICC values.
The proposed bias-corrected estimator for the ICC from a one-way random effects analysis of variance model appears to perform well in the scenarios we considered in this paper and can be used as a motivation to construct bias-corrected estimators for other types of ICCs that arise in more complex scenarios. It would also be interesting to investigate the bias-variance trade-off.
While there is some consensus on methods for investigating statistical and methodological heterogeneity, little attention has been paid to clinical aspects of heterogeneity. The objective of this study is to summarize and collate suggested methods for investigating clinical heterogeneity in systematic reviews.
We searched databases (Medline, EMBASE, CINAHL, Cochrane Library, and CONSORT, to December 2010) and reference lists and contacted experts to identify resources providing suggestions for investigating clinical heterogeneity between controlled clinical trials included in systematic reviews. We extracted recommendations, assessed resources for risk of bias, and collated the recommendations.
One hundred and one resources were collected, including narrative reviews, methodological reviews, statistical methods papers, and textbooks. These resources generally had a low risk of bias, but there was minimal consensus among them. Resources suggested that planned investigations of clinical heterogeneity should be made explicit in the protocol of the review; clinical experts should be included on the review team; a set of clinical covariates should be chosen considering variables from the participant level, intervention level, outcome level, research setting, or others unique to the research question; covariates should have a clear scientific rationale; there should be a sufficient number of trials per covariate; and results of any such investigations should be interpreted with caution.
Though the consensus was minimal, there were many recommendations in the literature for investigating clinical heterogeneity in systematic reviews. Formal recommendations for investigating clinical heterogeneity in systematic reviews of controlled trials are required.
Hyperuricemia is linked to gout and features of metabolic syndrome. There is concern that dietary fructose may increase uric acid concentrations. To assess the effects of fructose on serum uric acid concentrations in people with and without diabetes, we conducted a systematic review and meta-analysis of controlled feeding trials. We searched MEDLINE, EMBASE, and the Cochrane Library for relevant trials (through August 19, 2011). Analyses included all controlled feeding trials ≥7 d investigating the effect of fructose feeding on uric acid under isocaloric conditions, where fructose was isocalorically exchanged with other carbohydrate, or hypercaloric conditions, and where a control diet was supplemented with excess energy from fructose. Data were aggregated by the generic inverse variance method using random effects models and expressed as mean difference (MD) with 95% CI. Heterogeneity was assessed by the Q statistic and quantified by I2. A total of 21 trials in 425 participants met the eligibility criteria. Isocaloric exchange of fructose for other carbohydrate did not affect serum uric acid in diabetic and nondiabetic participants [MD = 0.56 μmol/L (95% CI: −6.62, 7.74)], with no evidence of inter-study heterogeneity. Hypercaloric supplementation of control diets with fructose (+35% excess energy) at extreme doses (213–219 g/d) significantly increased serum uric acid compared with the control diets alone in nondiabetic participants [MD = 31.0 mmol/L (95% CI: 15.4, 46.5)] with no evidence of heterogeneity. Confounding from excess energy cannot be ruled out in the hypercaloric trials. These analyses do not support a uric acid-increasing effect of isocaloric fructose intake in nondiabetic and diabetic participants. Hypercaloric fructose intake may, however, increase uric acid concentrations. The effect of the interaction of energy and fructose remains unclear. Larger, well-designed trials of fructose feeding at “real world” doses are needed.
In meta-regression, as the number of trials in the analyses decreases, the risk of false positives or false negatives increases. This is partly due to the assumption of normality that may not hold in small samples. Creation of a distribution from the observed trials using permutation methods to calculate P values may allow for less spurious findings. Permutation has not been empirically tested in meta-regression. The objective of this study was to perform an empirical investigation to explore the differences in results for meta-analyses on a small number of trials using standard large sample approaches verses permutation-based methods for meta-regression.
We isolated a sample of randomized controlled clinical trials (RCTs) for interventions that have a small number of trials (herbal medicine trials). Trials were then grouped by herbal species and condition and assessed for methodological quality using the Jadad scale, and data were extracted for each outcome. Finally, we performed meta-analyses on the primary outcome of each group of trials and meta-regression for methodological quality subgroups within each meta-analysis. We used large sample methods and permutation methods in our meta-regression modeling. We then compared final models and final P values between methods.
We collected 110 trials across 5 intervention/outcome pairings and 5 to 10 trials per covariate. When applying large sample methods and permutation-based methods in our backwards stepwise regression the covariates in the final models were identical in all cases. The P values for the covariates in the final model were larger in 78% (7/9) of the cases for permutation and identical for 22% (2/9) of the cases.
We present empirical evidence that permutation-based resampling may not change final models when using backwards stepwise regression, but may increase P values in meta-regression of multiple covariates for relatively small amount of trials.
Alzheimer's disease (AD) is common among older adults and leads to significant disability. Volatile anesthetic gases administered during general anesthesia (GA) have been hypothesized to be a risk factor for the development of AD. The objective of this study is to systematically review the association between exposure to GA and risk of AD.
We searched electronic databases including MEDLINE, Embase, and Google scholar for observational studies examining the association between exposure to GA and risk of AD. We examined study quality using a modified version of the Newcastle-Ottawa risk of bias assessment for observational studies. We used standard meta-analytic techniques to estimate pooled odds ratios (OR) and 95% confidence intervals (CI). Subgroup and sensitivity analyses were undertaken to evaluate the robustness of the findings.
A total of 15 case-control studies were included in the review. No cohort studies were identified that met inclusion criteria. There was variation in the methodological quality of included studies. There was no significant association between any exposure to GA and risk of AD (pooled OR: 1.05; 95% CI: 0.93 - 1.19, Z = 0.80, p = 0.43). There was also no significant association between GA and risk of AD in several subgroup and sensitivity analyses.
A history of exposure to GA is not associated with an increased risk of AD although there are few high-quality studies in this area. Prospective cohort studies with long-term follow-up or randomized controlled trials are required to further understand the association between GA and AD.
dementia; Alzheimer's disease; anesthesia; surgery; meta-analysis; systematic review
Infection of the CNS is considered to be the major cause of encephalitis and more than 100 different pathogens have been recognized as causative agents. Despite being identified worldwide as an important public health concern, studies on encephalitis are very few and often focus on particular types (with respect to causative agents) of encephalitis (e.g. West Nile, Japanese, etc.). Moreover, a number of other infectious and non-infectious conditions present with similar symptoms, and distinguishing encephalitis from other disguising conditions continues to a challenging task.
We used canonical correlation analysis (CCA) to assess associations between set of exposure variable and set of symptom and diagnostic variables in human encephalitis. Data consists of 208 confirmed cases of encephalitis from a prospective multicenter study conducted in the United Kingdom. We used a covariance matrix based on Gini's measure of similarity and used permutation based approaches to test significance of canonical variates.
Results show that weak pair-wise correlation exists between the risk factor (exposure and demographic) and symptom/laboratory variables. However, the first canonical variate from CCA revealed strong multivariate correlation (ρ = 0.71, se = 0.03, p = 0.013) between the two sets. We found a moderate correlation (ρ = 0.54, se = 0.02) between the variables in the second canonical variate, however, the value is not statistically significant (p = 0.68). Our results also show that a very small amount of the variation in the symptom sets is explained by the exposure variables. This indicates that host factors, rather than environmental factors might be important towards understanding the etiology of encephalitis and facilitate early diagnosis and treatment of encephalitis patients.
There is no standard laboratory diagnostic strategy for investigation of encephalitis and even experienced physicians are often uncertain about the cause, appropriate therapy and prognosis of encephalitis. Exploration of human encephalitis data using advanced multivariate statistical modelling approaches that can capture the inherent complexity in the data is, therefore, crucial in understanding the causes of human encephalitis. Moreover, application of multivariate exploratory techniques will generate clinically important hypotheses and offer useful insight into the number and nature of variables worthy of further consideration in a confirmatory statistical analysis.
The timely provision of critical care to hospitalised patients at risk for cardiopulmonary arrest is contingent upon identification and referral by frontline providers. Current approaches require improvement. In a single-centre study, we developed the Bedside Paediatric Early Warning System (Bedside PEWS) score to identify patients at risk. The objective of this study was to validate the Bedside PEWS score in a large patient population at multiple hospitals.
We performed an international, multicentre, case-control study of children admitted to hospital inpatient units with no limitations on care. Case patients had experienced a clinical deterioration event involving either an immediate call to a resuscitation team or urgent admission to a paediatric intensive care unit. Control patients had no events. The scores ranged from 0 to 26 and were assessed in the 24 hours prior to the clinical deterioration event. Score performance was assessed using the area under the receiver operating characteristic (AUCROC) curve by comparison with the retrospective rating of nurses and the temporal progression of scores in case patients.
A total of 2,074 patients were evaluated at 4 participating hospitals. The median (interquartile range) maximum Bedside PEWS scores for the 12 hours ending 1 hour before the clinical deterioration event were 8 (5 to 12) in case patients and 2 (1 to 4) in control patients (P < 0.0001). The AUCROC curve (95% confidence interval) was 0.87 (0.85 to 0.89). In case patients, mean scores were 5.3 at 20 to 24 hours and 8.4 at 0 to 4 hours before the event (P < 0.0001). The AUCROC curve (95% CI) of the retrospective nurse ratings was 0.83 (0.81 to 0.86). This was significantly lower than that of the Bedside PEWS score (P < 0.0001).
The Bedside PEWS score identified children at risk for cardiopulmonary arrest. Scores were elevated and continued to increase in the 24 hours before the clinical deterioration event. Prospective clinical evaluation is needed to determine whether this score will improve the quality of care and patient outcomes.
Several randomized, controlled trials (RCTs) have tested strategies to prevent sexual acquisition of HIV infection, but their quality has been variable. We aimed to identify, describe, and evaluate the quality of RCTs studying biomedical interventions to prevent HIV acquisition by sexual transmission.
We conducted a systematic review to identify all RCTs evaluating the efficacy of biomedical HIV prevention interventions. We assessed seven generic and content-specific quality components important in HIV prevention trials, factors influencing study power, co-interventions provided, and trial ethics.
We identified 26 eligible RCTs. The median number of quality components judged to be inadequate or unclear was 3 (range, 1-4) in 1992-1998, 3 (range, 1-4) in 1999-2003, and 0 (range 0-2) in 2004-2008 (p < 0.001). Common problems that may have biased results included low retention (median 84%), poor adherence to interventions requiring ongoing use (median ≤78%), and lower HIV incidence than expected a priori (in 8 of 11 trials where evaluable).
Reporting of trials of biomedical HIV prevention interventions has improved over time. However, quality improvement is needed in several key areas that influence study power, including participant retention, adherence to interventions, and estimation of expected HIV incidence.
HIV; primary prevention; clinical trial; systematic review; research methodology; statistical bias
Encephalitis is an acute clinical syndrome of the central nervous system (CNS), often associated with fatal outcome or permanent damage, including cognitive and behavioural impairment, affective disorders and epileptic seizures. Infection of the central nervous system is considered to be a major cause of encephalitis and more than 100 different pathogens have been recognized as causative agents. However, a large proportion of cases have unknown disease etiology.
We perform hierarchical cluster analysis on a multicenter England encephalitis data set with the aim of identifying sub-groups in human encephalitis. We use the simple matching similarity measure which is appropriate for binary data sets and performed variable selection using cluster heatmaps. We also use heatmaps to visually assess underlying patterns in the data, identify the main clinical and laboratory features and identify potential risk factors associated with encephalitis.
Our results identified fever, personality and behavioural change, headache and lethargy as the main characteristics of encephalitis. Diagnostic variables such as brain scan and measurements from cerebrospinal fluids are also identified as main indicators of encephalitis. Our analysis revealed six major clusters in the England encephalitis data set. However, marked within-cluster heterogeneity is observed in some of the big clusters indicating possible sub-groups. Overall, the results show that patients are clustered according to symptom and diagnostic variables rather than causal agents. Exposure variables such as recent infection, sick person contact and animal contact have been identified as potential risk factors.
It is in general assumed and is a common practice to group encephalitis cases according to disease etiology. However, our results indicate that patients are clustered with respect to mainly symptom and diagnostic variables rather than causal agents. These similarities and/or differences with respect to symptom and diagnostic measurements might be attributed to host factors. The idea that characteristics of the host may be more important than the pathogen is also consistent with the observation that for some causes, such as herpes simplex virus (HSV), encephalitis is a rare outcome of a common infection.
Because of blood lipid concerns, diabetes associations discourage fructose at high intakes. To quantify the effect of fructose on blood lipids in diabetes, we conducted a systematic review and meta-analysis of experimental clinical trials investigating the effect of isocaloric fructose exchange for carbohydrate on triglycerides, total cholesterol, LDL cholesterol, and HDL cholesterol in type 1 and 2 diabetes.
RESEARCH DESIGN AND METHODS
We searched MEDLINE, EMBASE, CINAHL, and the Cochrane Library for relevant trials of ≥7 days. Data were pooled by the generic inverse variance method and expressed as standardized mean differences with 95% CI. Heterogeneity was assessed by χ2 tests and quantified by I2. Meta-regression models identified dose threshold and independent predictors of effects.
Sixteen trials (236 subjects) met the eligibility criteria. Isocaloric fructose exchange for carbohydrate raised triglycerides and lowered total cholesterol under specific conditions without affecting LDL cholesterol or HDL cholesterol. A triglyceride-raising effect without heterogeneity was seen only in type 2 diabetes when the reference carbohydrate was starch (mean difference 0.24 [95% CI 0.05–0.44]), dose was >60 g/day (0.18 [0.00–0.37]), or follow-up was ≤4 weeks (0.18 [0.00–0.35]). Piecewise meta-regression confirmed a dose threshold of 60 g/day (R2 = 0.13)/10% energy (R2 = 0.36). A total cholesterol–lowering effect without heterogeneity was seen only in type 2 diabetes under the following conditions: no randomization and poor study quality (−0.19 [−0.34 to −0.05]), dietary fat >30% energy (−0.33 [−0.52 to −0.15]), or crystalline fructose (−0.28 [−0.47 to −0.09]). Multivariate meta-regression analyses were largely in agreement.
Pooled analyses demonstrated conditional triglyceride-raising and total cholesterol–lowering effects of isocaloric fructose exchange for carbohydrate in type 2 diabetes. Recommendations and large-scale future trials need to address the heterogeneity in the data.
With rapid advances in genotyping technologies in recent years and the growing number of available markers, genome-wide association studies are emerging as promising approaches for the study of complex diseases and traits. However, there are several challenges with analysis and interpretation of such data. First, there is a massive multiple testing problem due to the large number of markers that need to be analyzed, leading to an increased risk of false positives and decreased ability for association studies to detect truly associated markers. In particular, the ability to detect modest genetic effects can be severely compromised. Second, a genetic association of a given single-nucleotide polymorphism as determined by univariate statistical analyses does not typically explain biologically interesting features and often requires subsequent interpretation using a higher unit such as a gene or region, for example as defined by haplotype blocks. Third, missing genotypes in the data set and other data quality issues can pose challenges when comparisons across platforms and replications are planned. Finally, depending on the type of univariate analysis, computational burden can arise as the number of markers continues to grow into the millions. One way to deal with these and related challenges is to consider higher units for the analysis such as genes or regions. This paper summarizes analytical methods and strategies that have been proposed and applied by Group 16 to two genome-wide association data sets made available through the Genetic Analysis Workshop 16.
rheumatoid arthritis; case-control data; family-based study
Objective To determine the relation between overweight and obesity in mothers and preterm birth and low birth weight in singleton pregnancies in developed and developing countries.
Design Systematic review and meta-analyses.
Data sources Medline and Embase from their inceptions, and reference lists of identified articles.
Study selection Studies including a reference group of women with normal body mass index that assessed the effect of overweight and obesity on two primary outcomes: preterm birth (before 37 weeks) and low birth weight (<2500 g).
Data extraction Two assessors independently reviewed titles, abstracts, and full articles, extracted data using a piloted data collection form, and assessed quality.
Data synthesis 84 studies (64 cohort and 20 case-control) were included, totalling 1 095 834 women. Although the overall risk of preterm birth was similar in overweight and obese women and women of normal weight, the risk of induced preterm birth was increased in overweight and obese women (relative risk 1.30, 95% confidence interval 1.23 to 1.37). Although overall the risk of having an infant of low birth weight was decreased in overweight and obese women (0.84, 0.75 to 0.95), the decrease was greater in developing countries than in developed countries (0.58, 0.47 to 0.71 v 0.90, 0.79 to 1.01). After accounting for publication bias, the apparent protective effect of overweight and obesity on low birth weight disappeared with the addition of imputed “missing” studies (0.95, 0.85 to 1.07), whereas the risk of preterm birth appeared significantly higher in overweight and obese women (1.24, 1.13 to 1.37).
Conclusions Overweight and obese women have increased risks of preterm birth and induced preterm birth and, after accounting for publication bias, appeared to have increased risks of preterm birth overall. The beneficial effects of maternal overweight and obesity on low birth weight were greater in developing countries and disappeared after accounting for publication bias.
This study was designed to test the hypothesis that fetal exposure to corticosteroids in the antenatal period is an independent risk factor for the development of asthma in early childhood with little or no effect in later childhood. A population-based cohort study of all pregnant women who resided in Nova Scotia, Canada, and gave birth to a singleton fetus between 1989 and 1998
was undertaken. After a priori specified exclusions, 80,448 infants were available for analysis.
Using linked health care utilization records, incident asthma cases developed after 36 months of
age were identified. Extended Cox proportional hazards models were used to estimate hazard
ratios while controlling for confounders. Exposure to corticosteroids during pregnancy was
associated with a risk of asthma in childhood between 3–5 years of age: adjusted hazard ratio of
1.19 (95% confidence interval: 1.03, 1.39), with no association noted after 5 years of age:
adjusted hazard ratio for 5–7 years was 1.06 (95% confidence interval: 0.86, 1.30)
and for 8 or greater years was 0.74 (95% confidence interval: 0.54, 1.03). Antenatal steroid therapy appears to be an independent risk factor for the development of asthma between 3 and 5 years of age.
We propose the use of latent growth curve model to assess the influence of genetic, environmental, demographic, and lifestyle factors on multiple phenotypes related to coronary heart disease. We model four quantitative traits (systolic blood pressure, high-density lipoprotein, low-density lipoprotein, and triglycerides) simultaneously in a multivariate framework that allows us to study their change over time, assess individual variation, and investigate cross-phenotype relationships. Environmental, demographic, and lifestyle covariates are included at different levels of the model as time-varying or time-invariant, as appropriate. To investigate the change over time attributed to genetic factors, we use candidate markers that have previously been shown to be associated with the quantitative traits. We illustrate our approach using independent observations from the offspring cohort of the Framingham Heart Study data.
In high-dimensional studies such as genome-wide association studies, the correction for multiple testing in order to control total type I error results in decreased power to detect modest effects. We present a new analytical approach based on the higher criticism statistic that allows identification of the presence of modest effects. We apply our method to the genome-wide study of rheumatoid arthritis provided in the Genetic Analysis Workshop 16 Problem 1 data set. There is evidence for unknown bias in this study that could be explained by the presence of undetected modest effects. We compared the asymptotic and empirical thresholds for the higher criticism statistic. Using the asymptotic threshold we detected the presence of modest effects genome-wide. We also detected modest effects using 90th percentile of the empirical null distribution as a threshold; however, there is no such evidence when the 95th and 99th percentiles were used. While the higher criticism method suggests that there is some evidence for modest effects, interpreting individual single-nucleotide polymorphisms with significant higher criticism statistics is of undermined value. The goal of higher criticism is to alert the researcher that genetic effects remain to be discovered and to promote the use of more targeted and powerful studies to detect the remaining effects.
Evaluation of the association between single-nucleotide polymorphisms (SNPs) and disease outcomes is widely used to identify genetic risk factors for complex diseases. Although this analysis paradigm has made significant progress in many genetic studies, many challenges remain, such as the requirement of a large sample size to achieve adequate power. Here we use rheumatoid arthritis (RA) as an example and explore a new analysis strategy: pathway-based analysis to search for related genes and SNPs contributing to the disease.
We first propose the application of measure of explained variation to quantify the predictive ability of a given SNP. We then use gene set enrichment analysis to evaluate enrichment of specific pathways, where pathways, are considered enriched if they consist of genes that are associated with the phenotype of interest above and beyond is expected by chance. The results are also compared with score tests for association analysis by adjusting for population stratification.
Our study identified some significantly enriched pathways, such as "cell adhesion molecules," which are known to play a key role in RA. Our results showed that pathway-based analysis may identify other biologically interesting loci (e.g., rs1018361) related to RA: the gene (CTLA4) closest to this marker has previously been shown to be associated with RA and the gene is in the significant pathways we identified, even though the marker has not reached genome-wide significance in univariate single-marker analysis.
Multivariate linear growth curves were used to model high-density lipoprotein (HDL), low-density lipoprotein (LDL), triglycerides (TG), and systolic blood pressure (SBP) measured during four exams from 1659 independent individuals from the Framingham Heart Study. The slopes and intercepts from each of two phenotype models were tested for association with 348,053 autosomal single-nucleotide polymorphisms from the Affymetrix Gene Chip 500 k set. Three regions were associated with LDL intercept, TG slope, and SBP intercept (p < 1.44 × 10-7). We observed results consistent with previously reported associations between rs599839, on chromosome 1p13, and LDL. We note that the association is significant with LDL intercept but not slope. Markers on chromosome 17q25 were associated with TG slope, and a single-nucleotide polymorphism on chromosome 7p11 was associated with SBP intercept. Growth curve models can be used to gain more insight on the relationships between SNPs and traits than traditional association analysis when longitudinal data has been collected. The power to detect association with changes over time may be limited if the subjects are not followed over a long enough time period.
Development of efficient analytic methodologies for combining microarray results is a major challenge in gene expression analysis. The widely used effect size models are thought to provide an efficient modeling framework for this purpose, where the measures of association for each study and each gene are combined, weighted by the standard errors. A significant disadvantage of this strategy is that the quality of different data sets may be highly variable, but this information is usually neglected during the integration. Moreover, it is widely known that the estimated standard deviations are probably unstable in the commonly used effect size measures (such as standardized mean difference) when sample sizes in each group are small.
We propose a re-parameterization of the traditional mean difference based effect measure by using the log ratio of means as an effect size measure for each gene in each study. The estimated effect sizes for all studies were then combined under two modeling frameworks: the quality-unweighted random effects models and the quality-weighted random effects models. We defined the quality measure as a function of the detection p-value, which indicates whether a transcript is reliably detected or not on the Affymetrix gene chip. The new effect size measure is evaluated and compared under the quality-weighted and quality-unweighted data integration frameworks using simulated data sets, and also in several data sets of prostate cancer patients and controls. We focus on identifying differentially expressed biomarkers for prediction of cancer outcomes.
Our results show that the proposed effect size measure (log ratio of means) has better power to identify differentially expressed genes, and that the detected genes have better performance in predicting cancer outcomes than the commonly used effect size measure, the standardized mean difference (SMD), under both quality-weighted and quality-unweighted data integration frameworks. The new effect size measure and the quality-weighted microarray data integration framework provide efficient ways to combine microarray results.
Multiple regression models are used in a wide range of scientific disciplines and automated model selection procedures are frequently used to identify independent predictors. However, determination of relative importance of potential predictors and validating the fitted models for their stability, predictive accuracy and generalizability are often overlooked or not done thoroughly.
Using a case study aimed at predicting children with acute lymphoblastic leukemia (ALL) who are at low risk of Tumor Lysis Syndrome (TLS), we propose and compare two strategies, bootstrapping and random split of data, for ordering potential predictors according to their relative importance with respect to model stability and generalizability. We also propose an approach based on relative increase in percentage of explained variation and area under the Receiver Operating Characteristic (ROC) curve for developing models where variables from our ordered list enter the model according to their importance. An additional data set aimed at identifying predictors of prostate cancer penetration is also used for illustrative purposes.
Age is chosen to be the most important predictor of TLS. It is selected 100% of the time using the bootstrapping approach. Using the random split method, it is selected 99% of the time in the training data and is significant (at 5% level) 98% of the time in the validation data set. This indicates that age is a stable predictor of TLS with good generalizability. The second most important variable is white blood cell count (WBC). Our methods also identified an important predictor of TLS that was otherwise omitted if relying on any of the automated model selection procedures alone. A group at low risk of TLS consists of children younger than 10 years of age, without T-cell immunophenotype, whose baseline WBC is < 20 × 109/L and palpable spleen is < 2 cm. For the prostate cancer data set, the Gleason score and digital rectal exam are identified to be the most important indicators of whether tumor has penetrated the prostate capsule.
Our model selection procedures based on bootstrap re-sampling and repeated random split techniques can be used to assess the strength of evidence that a variable is truly an independent and reproducible predictor. Our methods, therefore, can be used for developing stable and reproducible models with good performances. Moreover, our methods can serve as a good tool for validating a predictive model. Previous biological and clinical studies support the findings based on our selection and validation strategies. However, extensive simulations may be required to assess the performance of our methods under different scenarios as well as check their sensitivity to a random fluctuation in the data.
The primary objective of this meta-analytic study was to determine the impact of RSV-IGIV and palivizumab on risk of respiratory syncytial virus (RSV)-related hospitalization. Secondary objectives were to determine if antibody therapy decreases the risk of RSV infection, intensive care admission, mechanical ventilation, and mortality in high risk infant populations.
We performed searches of electronic data bases from 1966 to April 2009. Inclusion and exclusion criteria were defined a priori. Inclusion criteria were as follows: 1) There was randomization between polyclonal or monoclonal antibodies and placebo or no therapy, and 2) Polyclonal or monoclonal antibodies were given as prophylaxis.
Of the six included studies, three utilized RSV-IGIV (total of 533 randomized to treatment groups) and three utilized palivizumab (total of 1,663 randomized to treatment groups). The absolute risk of hospitalization in the control arms was 12% and overall RR for all 2,196 children who received one of the antibody products was 0.53 (95% CI 0.43, 0.66), P < 0.00001. When looking only at the children who received palivizumab, the RR for hospitalization was 0.50 (95% CI 0.38, 0.66), P < 0.00001. For the children receiving RSV-IGIV, the RR for hospitalization was 0.59 (95% CI 0.42, 0.83, P < 0.002). The use of palivizumab resulted in a significant decrease in admission to the ICU (RR 0.29 (95% CI 0.14, 0.59; P = 0.0007). There was no significant reduction in the risk of mechanical ventilation or mortality with the use of antibody prophylaxis. Infants born at less than 35 weeks gestational age, and those with chronic lung and congenital heart disease all had a significant reduction in the risk of RSV hospitalization with children born under 35 weeks gestational age showing a trend towards the greatest benefit.
Both palivizumab and RSV-IGIV decrease the incidence of RSV hospitalization and ICU admission and their effect appears to be qualitatively similarly. There was neither a statistically significant reduction in the incidence of mechanical ventilation nor in all cause mortality. This meta-analysis separately quantifies the impact of RSV-IGIV and palivizumab on various measures of severe RSV disease and builds upon a previous study that was only able to examine the pooled effect of all antibody products together.