Bayesian meta-analysis, with application to studies of ETS and lung cancer

Meta-Analysis and Systematic Review in Environmental Tobacco Smoke Risk of Female Lung Cancer by Research Type

More than 50% of women worldwide are exposed to Environmental Tobacco Smoke (ETS). The impact of ETS on lung cancer remains unclear. Cohort studies since the late 1990s have provided new evidence of female lung cancer risk due to ETS. The objective of this meta-analysis and systematic review was to analyze the association of ETS with female lung cancer risk from 1997 to 2017, organised based on research design. According to our applied inclusion and exclusion criteria, 41 published studies were included. The relative risk (RR) from the cohort studies or odds ratio (OR) from case-control studies were extracted to calculate the pooled risks based on the type of study. The summary risks of ETS were further explored with the modulators of ETS exposure sources and doses. The pooled risks of lung cancer in non-smoking women exposed to ETS were 1.35 (95% CI: 1.17⁻1.56), 1.17 (95% CI: 0.94⁻1.44), and 1.33 (95% CI: 1.17⁻1.51) for case-control studies, cohort studies, and both types of studies, respectively. The summary RR estimate of the cohort studies was not statistically significant, but the RR increased with increasing doses of ETS exposure (p trend < 0.05). Based on the results of this study, ETS might be an important risk factor of female lung cancer in non-smokers.

The choice of prior distribution for a covariance matrix in multivariate meta-analysis: a simulation study

Bayesian meta-analysis is an increasingly important component of clinical research, with multivariate meta-analysis a promising tool for studies with multiple endpoints. Model assumptions, including the choice of priors, are crucial aspects of multivariate Bayesian meta-analysis (MBMA) models. In a given model, two different prior distributions can lead to different inferences about a particular parameter. A simulation study was performed in which the impact of families of prior distributions for the covariance matrix of a multivariate normal random effects MBMA model was analyzed. Inferences about effect sizes were not particularly sensitive to prior choice, but the related covariance estimates were. A few families of prior distributions with small relative biases, tight mean squared errors, and close to nominal coverage for the effect size estimates were identified. Our results demonstrate the need for sensitivity analysis and suggest some guidelines for choosing prior distributions in this class of problems. The MBMA models proposed here are illustrated in a small meta-analysis example from the periodontal field and a medium meta-analysis from the study of stroke. Copyright © 2015 John Wiley & Sons, Ltd.

Foliar interception of radionuclides in dry conditions: a meta-analysis using a Bayesian modeling approach

Uncertainty on the parameters that describe the transfer of radioactive materials into the (terrestrial) environment may be characterized thanks to datasets such as those compiled within International Atomic Energy Agency (IAEA) documents. Nevertheless, the information included in these documents is too poor to derive a relevant and informative uncertainty distribution regarding dry interception of radionuclides by the pasture grass and the leaves of vegetables. In this paper, 145 sets of dry interception measurements by the aboveground biomass of specific plants were collected from published scientific papers. A Bayesian meta-analysis was performed to derive the posterior probability distributions of the parameters that reflect their uncertainty given the collected data. Four competing models were compared in terms of both fitting performances and predictive abilities to reproduce plausible dry interception data. The asymptotic interception factor, applicable whatever the species and radionuclide to the highest aboveground biomass values (e.g. mature leafy vegetables), was estimated with the best model, to be 0.87 with a 95% credible interval (0.85, 0.89).

Estimating the probability of IQ impairment from blood phenylalanine for phenylketonuria patients: a hierarchical meta-analysis

Though the control of blood phenylalanine (Phe) levels is essential for minimizing impairment in individuals with phenylketonuria (PKU), the empirical basis for the selection of specific blood Phe levels as targets has not been evaluated. We evaluated the current evidence that particular Phe levels are optimal for minimizing or avoiding cognitive impairment in individuals with PKU. This work uses meta-estimates of blood Phe-IQ correlation to predict the probability of low IQ for a range of Phe levels. We believe this metric is easily interpretable by clinicians, and hence useful in making recommendations for Phe intake. The median baseline association of Phe with IQ was estimated to be negative, both in the context of historical (median = -0.026, 95 % BCI = [-0.040, -0.013]) and concurrent (-0.007, [-0.014, 0.000]) measurement of Phe relative to IQ. The estimated additive fixed effect of critical period Phe measurement was also nominally negative for historical measurement (-0.010, [-0.022, 0.003]) and positive for concurrent measurement (0.007, [-0.018, 0.035]). Probabilities corresponding to historical measures of blood Phe demonstrated an increasing chance of low IQ with increasing Phe, with a stronger association seen between blood Phe measured during the critical period than later. In contrast, concurrently-measured Phe was more weakly correlated with the probability of low IQ, though the correlation is still positive, irrespective of whether Phe was measured during the critical or non-critical period. This meta-analysis illustrates the utility of a Bayesian hierarchical approach for not only combining information from a set of candidate studies, but also for combining different types of data to estimate parameters of interest.

Quantifying, displaying and accounting for heterogeneity in the meta-analysis of RCTs using standard and generalised Q statistics

BACKGROUND

Clinical researchers have often preferred to use a fixed effects model for the primary interpretation of a meta-analysis. Heterogeneity is usually assessed via the well known Q and I2 statistics, along with the random effects estimate they imply. In recent years, alternative methods for quantifying heterogeneity have been proposed, that are based on a 'generalised' Q statistic.

METHODS

We review 18 IPD meta-analyses of RCTs into treatments for cancer, in order to quantify the amount of heterogeneity present and also to discuss practical methods for explaining heterogeneity.

RESULTS

Differing results were obtained when the standard Q and I2 statistics were used to test for the presence of heterogeneity. The two meta-analyses with the largest amount of heterogeneity were investigated further, and on inspection the straightforward application of a random effects model was not deemed appropriate. Compared to the standard Q statistic, the generalised Q statistic provided a more accurate platform for estimating the amount of heterogeneity in the 18 meta-analyses.

CONCLUSIONS

Explaining heterogeneity via the pre-specification of trial subgroups, graphical diagnostic tools and sensitivity analyses produced a more desirable outcome than an automatic application of the random effects model. Generalised Q statistic methods for quantifying and adjusting for heterogeneity should be incorporated as standard into statistical software. Software is provided to help achieve this aim.

Alternative approaches to endoscopic ablation for benign enlargement of the prostate: systematic review of randomised controlled trials

OBJECTIVE

To compare the effectiveness and risk profile of newer methods for endoscopic ablation of the prostate against the current standard of transurethral resection.

DESIGN

Systematic review and meta-analysis.

DATA SOURCES

Electronic and paper records in subject area up to March 2006.

REVIEW METHODS

We searched for randomised controlled trials of endoscopic ablative interventions that included transurethral resection of prostate as one of the treatment arms. Two reviewers independently extracted data and assessed quality. Meta-analyses of prespecified outcomes were done using fixed and random effects models and reported using relative risk or weighted mean difference.

RESULTS

We identified 45 randomised controlled trials meeting the inclusion criteria and reporting on 3970 participants. The reports were of moderate to poor quality, with small sample sizes. None of the newer technologies resulted in significantly greater improvement in symptoms than transurethral resection at 12 months, although a trend suggested a better outcome with holmium laser enucleation (random effects weighted mean difference -0.82, 95% confidence interval 1.76 to 0.12) and worse outcome with laser vaporisation (1.49, -0.40 to 3.39). Improvements in secondary measures, such as peak urine flow rate, were consistent with change in symptoms. Blood transfusion rates were higher for transurethral resection than for the newer methods (4.8% v 0.7%) and men undergoing laser vaporisation or diathermy vaporisation were more likely to experience urinary retention (6.7% v 2.3% and 3.6% v 1.1%). Hospital stay was up to one day shorter for the newer technologies.

CONCLUSIONS

Although men undergoing more modern methods of removing benign prostatic enlargement have similar outcomes to standard transurethral resection of prostate along with fewer requirements for blood transfusion and shorter hospital stay, the quality of current evidence is poor. The lack of any clearly more effective procedure suggests that transurethral resection should remain the standard approach.

A Bayesian mixture model for metaanalysis of microarray studies

The increased availability of microarray data has been calling for statistical methods to integrate findings across studies. A common goal of microarray analysis is to determine differentially expressed genes between two conditions, such as treatment vs control. A recent Bayesian metaanalysis model used a prior distribution for the mean log-expression ratios that was a mixture of two normal distributions. This model centered the prior distribution of differential expression at zero, and separated genes into two groups only: expressed and nonexpressed. Here, we introduce a Bayesian three-component truncated normal mixture prior model that more flexibly assigns prior distributions to the differentially expressed genes and produces three groups of genes: up and downregulated, and nonexpressed. We found in simulations of two and five studies that the three-component model outperformed the two-component model using three comparison measures. When analyzing biological data of Bacillus subtilis, we found that the three-component model discovered more genes and omitted fewer genes for the same levels of posterior probability of differential expression than the two-component model, and discovered more genes for fixed thresholds of Bayesian false discovery. We assumed that the data sets were produced from the same microarray platform and were prescaled.

Meta-analysis of studies of passive smoking and lung cancer: effects of study type and continent

BACKGROUND

To calculate a pooled estimate of relative risk (RR) of lung cancer associated with exposure to passive smoking in never smoking women exposed to smoking spouses. This study is an updated meta-analysis that also assesses the differences between estimated risks according to continent and study type using meta-regression.

METHODS

From a total of 101 primary studies, 55 studies are included in this meta-analysis, of which, 7 are cohort studies, 25 population-based case-control and 23 non-population-based case-control studies. Twenty previously published meta-analyses are also reviewed. Fixed and random effect models and meta-regression are used to obtain pooled estimates of RR and P-value functions are used to demonstrate consistency of results.

RESULTS

The pooled RR for never-smoking women exposed to passive smoking from spouses is 1.27 (95% CI 1.17-1.37). The RR for North America is 1.15 (95% CI 1.03-1.28), Asia, 1.31 (95% CI 1.16-1.48) and Europe, 1.31 (1.24-1.52). Sequential cumulative meta-analysis shows no trend. There is no strong evidence of publication bias.

CONCLUSIONS

The abundance of evidence, consistency of finding across continent and study type, dose-response relationship and biological plausibility, overwhelmingly support the existence of a causal relationship between passive smoking and lung cancer.

Assessing the combined effect of asbestos exposure and smoking on lung cancer: a Bayesian approach

We review the literature on the combined association between lung cancer and two environmental exposures, asbestos exposure and smoking, and explore a Bayesian approach to assess evidence of interaction between the exposures. The meta-analysis combines separate indices of additive and multiplicative relationships and multivariate relative risk estimates. By making inferences on posterior probabilities we can explore both the form and strength of interaction. This analysis may be more informative than providing evidence to support one relation over another on the basis of statistical significance. Overall, we find evidence for a more than additive and less than multiplicative relation.

Bayesian meta-analysis models for microarray data: a comparative study

Background

With the growing abundance of microarray data, statistical methods are increasingly needed to integrate results across studies. Two common approaches for meta-analysis of microarrays include either combining gene expression measures across studies or combining summaries such as p-values, probabilities or ranks. Here, we compare two Bayesian meta-analysis models that are analogous to these methods.

Results

Two Bayesian meta-analysis models for microarray data have recently been introduced. The first model combines standardized gene expression measures across studies into an overall mean, accounting for inter-study variability, while the second combines probabilities of differential expression without combining expression values. Both models produce the gene-specific posterior probability of differential expression, which is the basis for inference. Since the standardized expression integration model includes inter-study variability, it may improve accuracy of results versus the probability integration model. However, due to the small number of studies typical in microarray meta-analyses, the variability between studies is challenging to estimate. The probability integration model eliminates the need to model variability between studies, and thus its implementation is more straightforward. We found in simulations of two and five studies that combining probabilities outperformed combining standardized gene expression measures for three comparison values: the percent of true discovered genes in meta-analysis versus individual studies; the percent of true genes omitted in meta-analysis versus separate studies, and the number of true discovered genes for fixed levels of Bayesian false discovery. We identified similar results when pooling two independent studies of Bacillus subtilis. We assumed that each study was produced from the same microarray platform with only two conditions: a treatment and control, and that the data sets were pre-scaled.

Conclusion

The Bayesian meta-analysis model that combines probabilities across studies does not aggregate gene expression measures, thus an inter-study variability parameter is not included in the model. This results in a simpler modeling approach than aggregating expression measures, which accounts for variability across studies. The probability integration model identified more true discovered genes and fewer true omitted genes than combining expression measures, for our data sets.

Never smokers and lung cancer risk: a case-control study of epidemiological factors

We performed an analysis of potential epidemiological risk factors for lung cancer using data from 280 cases and 242 hospital-based controls, all lifetime never smokers (those who had smoked <100 cigarettes in their lifetimes) and frequency matched on age, gender and ethnicity. The data on demographic characteristics, medical history of respiratory diseases (asthma, emphysema, pneumonia and hay fever), weight and height, family history, female characteristics and environmental tobacco smoke (ETS) and dust exposure were derived from personal interviews. We performed a logistic regression analysis of these variables adjusting for age, gender, ethnicity, income and years of education. Exposure to ETS (OR = 2.08, 95% CI [1.25-3.43]) and dusts (OR = 2.43, 95% CI [1.53-3.88]) were associated with significantly increased risk. In the analysis for joint effects, exposure to both ETS and dusts conferred a higher risk (OR = 3.25, 95% CI [1.58-6.70]) than exposure to either alone. Family history of any cancer with onset before age 50 in at least 1 first degree relative was a significant risk predictor (OR = 1.70, 95% CI [1.10-2.64]). Individuals with a self-reported physician-diagnosed history of hay fever, but not asthma, had a decreased lung cancer risk (OR = 0.57, 95% CI [0.35-0.92]). In the multivariate analysis, exposure to ETS and dusts, and family history of cancer with onset before age 50 were significant risk factors, while a history of hay fever (occurring without asthma) was significantly protective.

Comparison of the dose-response relationships of 2 lipid-lowering agents: a Bayesian meta-analysis

BACKGROUND

Comparing the dose-response of a new drug to that of a previously studied drug can aid in understanding their relative potencies. Two dose-finding studies addressed the effect of a new drug, rosuvastatin, on its ability to decrease low-density lipoprotein cholesterol (LDL-C) levels. One of these studies included 2 doses of atorvastatin, and substantial additional information is available in the literature about the effect of atorvastatin on LDL-C level lowering.

METHODS

The 2 dose-finding studies of rosuvastatin considered otherwise healthy patients who had hypercholesterolemia. Comparable studies of atorvastatin were identified via a MEDLINE search in December 1999. Multiple reviewer consensus identified 15 of 41 studies on atorvastatin published since 1996 that met these selection criteria: reporting of LDL-C level change from baseline at least 6 weeks after treatment initiation, doses administered, and treatment group sizes. Eligible populations had clinical evidence of hypercholesterolemia. We excluded studies with patients who had severe illness or a previous history of transplantation. Data extraction of the mean, sample sizes, and SDs (or CIs) by dose was carried out independently by multiple reviewers. We combined the results from the various studies with Bayesian hierarchical modeling and analyzed them with Markov chain Monte Carlo techniques.

RESULTS

Combining this study and literature results substantially increased the power to compare the dose-response relationships of rosuvastatin and atorvastatin. Rosuvastatin reduced LDL-C level by an estimated 10 to 17 percentage points more than atorvastatin when both were given at the same dose. Approximately one quarter of the dose of rosuvastatin achieved about the same magnitude of LDL-C level reduction as atorvastatin at dosages as high as 80 mg. This finding does not imply a 4-fold difference in efficacy overall and specifically does not describe the results at higher dosage levels.

CONCLUSIONS

Bayesian meta-analysis of results from related studies allows the comparison of the dose-response relationships of 2 drugs, better estimates of a particular dose-response relationship within an individual study, and the expression of relative benefits (of dose and drug) in terms of probabilities. Explicitly comparing a study's results with historical data using Bayesian meta-analysis allows clinicians to view the study in the larger context of medical research.

Noninvasive ventilation in acute respiratory failure--a meta-analysis update

OBJECTIVE

To present a meta-analytic update on the effects of noninvasive ventilation (NIV) in the management of acute respiratory failure.

DESIGN

Meta-analysis of randomized controlled clinical trials in acute respiratory failure comparing NIV with standard medical therapy.

PATIENTS

Randomized controlled trials of NIV in acute respiratory failure were identified by search of i) MEDLINE (1966-2000), ii) published abstracts from scientific meetings, and iii) bibliographies of relevant articles.

MEASUREMENTS AND MAIN RESULTS

Of the 15 randomized controlled trials that were identified (13 published and 2 in abstract form), 8 studies were on exacerbations of chronic obstructive pulmonary disease (COPD) and 7 on diverse disease processes in both COPD and non-COPD groups ("mixed-group"). Because of underlying heterogeneity of treatment effects, only the DerSimonian-Laird random effects estimator was used and reported. The effects of NIV vs. standard therapy on mortality and subsequent invasive mechanical ventilation (MV) was assessed as risk difference, and hospital length of stay as mean weighted difference (days). NIV was associated with reduction in mortality (8%, p = .03), reduced need for MV (19%, p = .001) and shortened hospital length of stay (2.74 days, p = .004). In the COPD cohort, significant reductions in mortality (13%, p = .001), need for MV (18%, p = .02), and hospital length of stay (5.66 days, p = .01) were observed in the group treated with NIV. In contrast, in the mixed-group, there was no demonstrable reduction in mortality (0%, p = .98). However, there was significant reduction in the need for MV (22%, p = .001). Publication bias was not evident on analysis. Treatment effect i) as mortality or need for mechanical ventilation was not modified by enrollment pH, PaCO2, nor age and ii) was not related (as log odds ratio) to underlying risk (control arm log odds). Cumulative meta-analysis did not demonstrate any substantial variation in the point estimates with the addition of the recently published studies. However a contraction in the confidence intervals was observed in the COPD subgroup. Complication rates were not significantly different in the standard medical therapy group and the NIV treated patients.

CONCLUSION

Substantial reductions in mortality and the need for subsequent MV were associated with NIV in acute respiratory failure, especially in the COPD subgroup. Hospital length of stay was variably affected. Heterogeneity of treatment effects was observed.

Predictive modeling techniques in prostate cancer

A number of new predictive modeling techniques have emerged in the past several years. These methods can be used independently or in combination with traditional modeling techniques to produce useful tools for the management of prostate cancer. Investigators should be aware of these techniques and avail themselves of their potentially useful properties. This review outlines selected predictive methods that can be used to develop models that may be useful to patients and clinicians for prostate cancer management.

Passive smoking and lung cancer: a cumulative meta-analysis

OBJECTIVE

To review the epidemiological evidence for the association between passive smoking and lung cancer.

METHOD

Primary studies and meta-analyses examining the relationship between passive smoking and lung cancer were identified through a computerised literature search of Medline and Embase, secondary references, and experts in the field of passive smoking. Primary studies meeting the inclusion criteria were meta-analysed.

RESULTS

From 1981 to the end of 1999 there have been 76 primary epidemiological studies of passive smoking and lung cancer, and 20 meta-analyses. There were 43 primary studies that met the inclusion criteria for this meta-analysis; more studies than previous assessments. The pooled relative risk (RR) for never-smoking women exposed to environmental tobacco smoke (ETS) from spouses, compared with unexposed never-smoking women was 1.29 (95% CI 1.17-1.43). Sequential cumulative meta-analysed results for each year from 1981 were calculated: since 1992 the RR has been greater than 1.25. For Western industrialised countries the RR for never-smoking women exposed to ETS compared with unexposed never-smoking women, was 1.21 (95% CI 1.10-1.33). Previously published international spousal meta-analyses have all produced statistically significant RRs greater than 1.17.

CONCLUSIONS

The abundance of evidence in this paper, and the consistency of findings across domestic and workplace primary studies, dosimetric extrapolations and meta-analyses, clearly indicates that non-smokers exposed to ETS are at increased risk of lung cancer.

IMPLICATIONS

The recommended public health policy is for a total ban on smoking in enclosed public places and work sites.

Systematic reviews and meta-analysis: a structured review of the methodological literature

OBJECTIVE

To systematically review methods for systematic review/meta-analysis in order to identify the different methodological and statistical methods that have been proposed. A summary of the main findings is presented here, with emphasis given to health services research topics.

METHODS

A thorough systematic search for methodological papers was carried out using a variety of methods, including the use of electronic databases. Approximately 1000 potentially relevant references were identified, a number of them from education, psychology and sociology.

RESULTS

After briefly reviewing the procedural methods required to carry out a review, and the basic statistical methods used to combine study estimates, less established methods are discussed. These include methods for dealing with publication bias, meta-regression, meta-analysis of individual patient data, the synthesis of non-randomized evidence alone and in combination with randomized studies. Bayesian modelling and economic evaluation through meta-analysis. Recommendations for meta-analytical practice are given; these are either distilled from previous guidelines, or constructed where there appears to be a broad consensus across the literature.

CONCLUSIONS

It is hoped that this review will provide a consistent and comprehensive, but concise, description of the methods available for synthesizing evidence, that it will promote better quality reviews of the results of health services research and identify specific areas which require methodological development.

Reliability of league tables of in vitro fertilisation clinics: retrospective analysis of live birth rates

OBJECTIVE

To determine to what extent institutions carrying out in vitro fertilisation can reasonably be ranked according to their live birth rates.

DESIGN

Retrospective analysis of prospectively collected data on live birth rate after in vitro fertilisation.

SETTING

52 clinics in the United Kingdom carrying out in vitro fertilisation over the period April 1994 to March 1995.

MAIN OUTCOME MEASURE

Estimated adjusted live birth rate for each clinic; their rank and its associated uncertainty.

RESULTS

There were substantial and significant differences between the live birth rates of the clinics. There was great uncertainty, however, concerning the true ranks, particularly for the smaller clinics. Only one clinic could be confidently ranked in the bottom quarter according to this measure of performance. Many centres had substantial changes in rank between years, even though their live birth rate did not change significantly.

CONCLUSIONS

Even when there are substantial differences between institutions, ranks are extremely unreliable statistical summaries of performance and change in performance, particularly for smaller institutions. Any performance indicator should always be associated with a measure of sampling variability.