Malcolm J Price

Multiparameter evidence synthesis in epidemiology and medical decision-making:

Meta-analysis has been well-established for many years, but has been largely confined to pooling evidence on pair-wise contrasts. Broader forms of synthesis have also been described, apparently re-invented in disparate fields, each time taking different computational approaches. The potential value of Bayesian estimation of a joint posterior parameter distribution and simultaneously sampling from it for decision analysis has also been appreciated. However, applications have been relatively few in number, sometimes stylized, and presented mainly to a statistical methods audience. As a result, the potential for multiparameter evidence synthesis in both epidemiology and health technology assessment has remained largely unrecognized. The advent of flexible software for Bayesian Markov chain Monte Carlo in the shape of WinBUGS has the made these earlier strands of work more widely available. Researchers can now carry out synthesis at a realistic level of complexity. The Bristol programme has not only contributed to a growing body of literature on how to synthesize different evidence structures, but also on how to check the consistency of multiple information sources and how to use the resulting models to prioritize future research.

Multivariate meta‐analysis using individual participant data

When combining results across related studies, a multivariate meta-analysis allows the joint synthesis of correlated effect estimates from multiple outcomes. Joint synthesis can improve efficiency over separate univariate syntheses, may reduce selective outcome reporting biases, and enables joint inferences across the outcomes. A common issue is that within-study correlations needed to fit the multivariate model are unknown from published reports. However, provision of individual participant data (IPD) allows them to be calculated directly. Here, we illustrate how to use IPD to estimate within-study correlations, using a joint linear regression for multiple continuous outcomes and bootstrapping methods for binary, survival and mixed outcomes. In a meta-analysis of 10 hypertension trials, we then show how these methods enable multivariate meta-analysis to address novel clinical questions about continuous, survival and binary outcomes; treatment–covariate interactions; adjusted risk/prognostic factor effects; longitudinal data; prognostic and multiparameter models; and multiple treatment comparisons. Both frequentist and Bayesian approaches are applied, with example software code provided to derive within-study correlations and to fit the models.

Supported self-management for patients with moderate to severe chronic obstructive pulmonary disease (COPD): an evidence synthesis and economic analysis

BACKGROUND Self-management (SM) support for patients with chronic obstructive pulmonary disease (COPD) is variable in its coverage, content, method and timing of delivery. There is insufficient evidence for which SM interventions are the most effective and cost-effective. OBJECTIVES To undertake (1) a systematic review of the evidence for the effectiveness of SM interventions commencing within 6 weeks of hospital discharge for an exacerbation for COPD (review 1); (2) a systematic review of the qualitative evidence about patient satisfaction, acceptance and barriers to SM interventions (review 2); (3) a systematic review of the cost-effectiveness of SM support interventions within 6 weeks of hospital discharge for an exacerbation of COPD (review 3); (4) a cost-effectiveness analysis and economic model of post-exacerbation SM support compared with usual care (UC) (economic model); and (5) a wider systematic review of the evidence of the effectiveness of SM support, including interventions (such as pulmonary rehabilitation) in which there are significant components of SM, to identify which components are the most important in reducing exacerbations, hospital admissions/readmissions and improving quality of life (review 4). METHODS The following electronic databases were searched from inception to May 2012: MEDLINE, MEDLINE In-Process and Other Non-Indexed Citations, EMBASE, Cochrane Central Register of Controlled Trials (CENTRAL), and Science Citation Index [Institute of Scientific Information (ISI)]. Subject-specific databases were also searched: PEDro physiotherapy evidence database, PsycINFO and the Cochrane Airways Group Register of Trials. Ongoing studies were sourced through the metaRegister of Current Controlled Trials, International Standard Randomised Controlled Trial Number database, World Health Organization International Clinical Trials Registry Platform Portal and ClinicalTrials.gov. Specialist abstract and conference proceedings were sourced through ISIs Conference Proceedings Citation Index and British Librarys Electronic Table of Contents (Zetoc). Hand-searching through European Respiratory Society, the American Thoracic Society and British Thoracic Society conference proceedings from 2010 to 2012 was also undertaken, and selected websites were also examined. Title, abstracts and full texts of potentially relevant studies were scanned by two independent reviewers. Primary studies were included if ≈90% of the population had COPD, the majority were of at least moderate severity and reported on any intervention that included a SM component or package. Accepted study designs and outcomes differed between the reviews. Risk of bias for randomised controlled trials (RCTs) was assessed using the Cochrane tool. Random-effects meta-analysis was used to combine studies where appropriate. A Markov model, taking a 30-year time horizon, compared a SM intervention immediately following a hospital admission for an acute exacerbation with UC. Incremental costs and quality-adjusted life-years were calculated, with sensitivity analyses. RESULTS From 13,355 abstracts, 10 RCTs were included for review 1, one study each for reviews 2 and 3, and 174 RCTs for review 4. Available studies were heterogeneous and many were of poor quality. Meta-analysis identified no evidence of benefit of post-discharge SM support on admissions [hazard ratio (HR) 0.78, 95% confidence interval (CI) 0.52 to 1.17], mortality (HR 1.07, 95% CI 0.74 to 1.54) and most other health outcomes. A modest improvement in health-related quality of life (HRQoL) was identified but this was possibly biased due to high loss to follow-up. The economic model was speculative due to uncertainty in impact on readmissions. Compared with UC, post-discharge SM support (delivered within 6 weeks of discharge) was more costly and resulted in better outcomes (£683 cost difference and 0.0831 QALY gain). Studies assessing the effect of individual components were few but only exercise significantly improved HRQoL (3-month St Georges Respiratory Questionnaire 4.87, 95% CI 3.96 to 5.79). Multicomponent interventions produced an improved HRQoL compared with UC (mean difference 6.50, 95% CI 3.62 to 9.39, at 3 months). Results were consistent with a potential reduction in admissions. Interventions with more enhanced care from health-care professionals improved HRQoL and reduced admissions at 1-year follow-up. Interventions that included supervised or unsupervised structured exercise resulted in significant and clinically important improvements in HRQoL up to 6 months. LIMITATIONS This review was based on a comprehensive search strategy that should have identified most of the relevant studies. The main limitations result from the heterogeneity of studies available and widespread problems with their design and reporting. CONCLUSIONS There was little evidence of benefit of providing SM support to patients shortly after discharge from hospital, although effects observed were consistent with possible improvement in HRQoL and reduction in hospital admissions. It was not easy to tease out the most effective components of SM support packages, although interventions containing exercise seemed the most effective. Future work should include qualitative studies to explore barriers and facilitators to SM post exacerbation and novel approaches to affect behaviour change, tailored to the individual and their circumstances. Any new trials should be properly designed and conducted, with special attention to reducing loss to follow-up. Individual participant data meta-analysis may help to identify the most effective components of SM interventions. STUDY REGISTRATION This study is registered as PROSPERO CRD42011001588. FUNDING The National Institute for Health Research Health Technology Assessment programme.

Self-management of health care behaviors for COPD: a systematic review and meta-analysis

Purpose This systematic review aimed to identify the most effective components of interventions to facilitate self-management of health care behaviors for patients with COPD. PROSPERO registration number CRD42011001588. Methods We used standard review methods with a systematic search to May 2012 for randomized controlled trials of self-management interventions reporting hospital admissions or health-related quality of life (HRQoL). Mean differences (MD), hazard ratios, and 95% confidence intervals (CIs) were calculated and pooled using random-effects meta-analyses. Effects among different subgroups of interventions were explored including single/multiple components and multicomponent interventions with/without exercise. Results One hundred and seventy-three randomized controlled trials were identified. Self-management interventions had a minimal effect on hospital admission rates. Multicomponent interventions improved HRQoL (studies with follow-up >6 months St George’s Respiratory Questionnaire (MD 2.40, 95% CI 0.75–4.04, I2 57.9). Exercise was an effective individual component (St George’s Respiratory Questionnaire at 3 months MD 4.87, 95% CI 3.96–5.79, I2 0%). Conclusion While many self-management interventions increased HRQoL, little effect was seen on hospital admissions. More trials should report admissions and follow-up participants beyond the end of the intervention.

Supported self-management for patients with COPD who have recently been discharged from hospital: a systematic review and meta-analysis

Purpose Although many hospitals promote self-management to chronic obstructive pulmonary disease (COPD) patients post discharge from hospital, the clinical effectiveness of this is unknown. We undertook a systematic review of the evidence as part of a Health Technology Assessment review. Methods A comprehensive search strategy with no language restrictions was conducted across relevant databases from inception to May 2012. Randomized controlled trials of patients with COPD, recently discharged from hospital after an acute exacerbation and comparing a self-management intervention with control, usual care or other intervention were included. Study selection, data extraction, and risk of bias assessment were undertaken by two reviewers independently. Results Of 13,559 citations, 836 full texts were reviewed with nine randomized controlled trials finally included in quantitative syntheses. Interventions were heterogeneous. Five trials assessed highly supported multi-component interventions and four trials were less supported with fewer contacts with health care professionals and mainly home-based interventions. Total sample size was 1,466 (range 33–464 per trial) with length of follow-up 2–12 months. Trials varied in quality; poor patient follow-up and poor reporting was common. No evidence of effect in favor of self-management support was observed for all-cause mortality (pooled hazard ratio =1.07; 95% confidence interval [0.74 to 1.55]; I2=0.0%, [n=5 trials]). No clear evidence of effect on all-cause hospital admissions was observed (hazard ratio 0.88 [0.61, 1.27] I2=66.0%). Improvements in St George’s Respiratory Questionnaire score were seen in favor of self-management interventions (mean difference =3.84 [1.29 to 6.40]; I2=14.6%), although patient follow-up rates were low. Conclusion There is insufficient evidence to support self-management interventions post-discharge. There is a need for good quality primary research to identify effective approaches.

Multivariate and network meta-analysis of multiple outcomes and multiple treatments: rationale, concepts, and examples

Organisations such as the National Institute for Health and Care Excellence require the synthesis of evidence from existing studies to inform their decisions—for example, about the best available treatments with respect to multiple efficacy and safety outcomes. However, relevant studies may not provide direct evidence about all the treatments or outcomes of interest. Multivariate and network meta-analysis methods provide a framework to address this, using correlated or indirect evidence from such studies alongside any direct evidence. In this article, the authors describe the key concepts and assumptions of these methods, outline how correlated and indirect evidence arises, and illustrate the contribution of such evidence in real clinical examples involving multiple outcomes and multiple treatments

The effect of domiciliary noninvasive ventilation on clinical outcomes in stable and recently hospitalized patients with COPD: a systematic review and meta-analysis

Introduction Noninvasive ventilation (NIV) improves survival among patients with hypercapnic respiratory failure in hospital, but evidence for its use in domiciliary settings is limited. A patient’s underlying risk of having an exacerbation may affect any potential benefit that can be gained from domiciliary NIV. This is the first comprehensive systematic review to stratify patients based on a proxy for exacerbation risk: patients in a stable state and those immediately post-exacerbation hospitalization. Methods A systematic review of nonrandomized and randomized controlled trials (RCTs) was undertaken in order to compare the relative effectiveness of different types of domiciliary NIV and usual care on hospital admissions, mortality, and health-related quality of life. Standard systematic review methods were used for identifying studies (until September 2014), quality appraisal, and synthesis. Data were presented in forest plots and pooled where appropriate using random-effects meta-analysis. Results Thirty-one studies were included. For stable patients, there was no evidence of a survival benefit from NIV (relative risk [RR] 0.88 [0.55, 1.43], I2=60.4%, n=7 RCTs), but there was a possible trend toward fewer hospitalizations (weighted mean difference −0.46 [−1.02, 0.09], I2=59.2%, n=5 RCTs) and improved health-related quality of life. For posthospital patients, survival benefit could not be demonstrated within the three RCTs (RR 0.89 [0.53, 1.49], I2=25.1%), although there was evidence of benefit from four non-RCTs (RR 0.45 [0.32, 0.65], I2=0%). Effects on hospitalizations were inconsistent. Post hoc analyses suggested that NIV-related improvements in hypercapnia were associated with reduced hospital admissions across both populations. Little data were available comparing different types of NIV. Conclusion The effectiveness of domiciliary NIV remains uncertain; however, some patients may benefit. Further research is required to identify these patients and to explore the relevance of improvements in hypercapnia in influencing clinical outcomes. Optimum time points for commencing domiciliary NIV and equipment settings need to be established.

Proportion of Pelvic Inflammatory Disease Cases Caused by Chlamydia trachomatis: Consistent Picture From Different Methods

Background. Pelvic inflammatory disease (PID) is a leading cause of both tubal factor infertility and ectopic pregnancy. Chlamydia trachomatis is an important risk factor for PID, but the proportion of PID cases caused by C. trachomatis is unclear. Estimates of this are required to evaluate control measures. Methods. We consider 5 separate methods of estimating age-group-specific population excess fractions (PEFs) of PID due to C. trachomatis, using routine data, surveys, case-control studies, and randomized controlled trials, and apply these to data from the United Kingdom before introduction of the National Chlamydia Screening Programme. Results. As they are informed by randomized comparisons and national exposure and outcome estimates, our preferred estimates of the proportion of PID cases caused by C. trachomatis are 35% (95% credible interval [CrI], 11%–69%) in women aged 16–24 years and 20% (95% CrI, 6%–38%) in women aged 16–44 years in the United Kingdom. There is a fair degree of consistency between adjusted estimates of PEF, but all have wide 95% CrIs. The PEF decreases from 53.5% (95% CrI, 15.6%–100%) in women aged 16–19 years to 11.5% (95% CrI, 3.0%–25.7%) in women aged 35–44 years. Conclusions. The PEFs of PID due to C. trachomatis decline steeply with age by a factor of around 5-fold between younger and older women. Further studies of the etiology of PID in different age groups are required.

Genital chlamydia trachomatis infections clear more slowly in men Than women, but are less likely to become established

Summary Rigorous estimates for untreated chlamydia clearance rates are important for understanding and controlling infection. We synthesized literature data to investigate clearance in men. Established infections clear more slowly in men than women, and more slowly than had previously been assumed.

Proportion of Tubal Factor Infertility due to Chlamydia: Finite Mixture Modeling of Serum Antibody Titers

In this study, we examined whether the proportion of tubal factor infertility (TFI) that is attributable to Chlamydia trachomatis, the population excess fraction (PEF), can be estimated from serological data using finite mixture modeling. Whole-cell inclusion immunofluorescence serum antibody titers were recorded among infertile women seen at St. Michaels Hospital in Bristol, United Kingdom, during the period 1985–1995. Women were classified as TFI cases or controls based on laparoscopic examination. Finite mixture models were used to identify the number of component titer distributions and the proportion of serum samples in each, from which estimates of PEF were derived. Four titer distributions were identified. The component at the highest titer was found only in samples from women with TFI, but there was also an excess of the second-highest titer component in TFI cases. Minimum and maximum estimates of the PEF were 28.0% (95% credible interval: 6.9, 50.0) and 46.8% (95% credible interval: 23.2, 64.1). Equivalent estimates based on the standard PEF formula from case-control studies were 0% and over 65%. Finite mixture modeling can be applied to serological data to obtain estimates of the proportion of reproductive damage attributable to C. trachomatis. Further studies using modern assays in contemporary, representative populations should be undertaken.