Detecting Heterogeneity of Intervention Effects Using Analysis and Meta-analysis of Differences in Variance Between Trial Arms.

BACKGROUND: Randomized controlled trials (RCTs) with continuous outcomes usually only examine mean differences in response between trial arms. If the intervention has heterogeneous effects, then outcome variances will also differ between arms. Power of an individual trial to assess heterogeneity is lower than the power to detect the same size of main effect. METHODS: We describe several methods for assessing differences in variance in trial arms and apply them to a single trial with individual patient data and to meta-analyses using summary data. Where individual data are available, we use regression-based methods to examine the effects of covariates on variation. We present an additional method to meta-analyze differences in variances with summary data. RESULTS: In the single trial, there was agreement between methods, and the difference in variance was largely due to differences in prevalence of depression at baseline. In two meta-analyses, most individual trials did not show strong evidence of a difference in variance between arms, with wide confidence intervals. However, both meta-analyses showed evidence of greater variance in the control arm, and in one example, this was perhaps because mean outcome in the control arm was higher. CONCLUSIONS: Using meta-analysis, we overcame low power of individual trials to examine differences in variance using meta-analysis. Evidence of differences in variance should be followed up to identify potential effect modifiers and explore other possible causes such as varying compliance.

The role of rapid diagnostics in managing Ebola epidemics.

Ebola emerged in West Africa around December 2013 and swept through Guinea, Sierra Leone and Liberia, giving rise to 27,748 confirmed, probable and suspected cases reported by 29 July 2015. Case diagnoses during the epidemic have relied on polymerase chain reaction-based tests. Owing to limited laboratory capacity and local transport infrastructure, the delays from sample collection to test results being available have often been 2 days or more. Point-of-care rapid diagnostic tests offer the potential to substantially reduce these delays. We review Ebola rapid diagnostic tests approved by the World Health Organization and those currently in development. Such rapid diagnostic tests could allow early triaging of patients, thereby reducing the potential for nosocomial transmission. In addition, despite the lower test accuracy, rapid diagnostic test-based diagnosis may be beneficial in some contexts because of the reduced time spent by uninfected individuals in health-care settings where they may be at increased risk of infection; this also frees up hospital beds. We use mathematical modelling to explore the potential benefits of diagnostic testing strategies involving rapid diagnostic tests alone and in combination with polymerase chain reaction testing. Our analysis indicates that the use of rapid diagnostic tests with sensitivity and specificity comparable with those currently under development always enhances control, whether evaluated at a health-care-unit or population level. If such tests had been available throughout the recent epidemic, we estimate, for Sierra Leone, that their use in combination with confirmatory polymerase chain-reaction testing might have reduced the scale of the epidemic by over a third.

Methods for Dealing With Missing Covariate Data in Epigenome-Wide Association Studies.

Multiple imputation (MI) is a well-established method for dealing with missing data. MI is computationally intensive when imputing missing covariates with high-dimensional outcome data (e.g., DNA methylation data in epigenome-wide association studies (EWAS)), because every outcome variable must be included in the imputation model to avoid biasing associations towards the null. Instead, EWAS analyses are reduced to only complete cases, limiting statistical power and potentially causing bias. We used simulations to compare 5 MI methods for high-dimensional data under 2 missingness mechanisms. All imputation methods had increased power over complete-case (C-C) analyses. Imputing missing values separately for each variable was computationally inefficient, but dividing sites at random into evenly sized bins improved efficiency and gave low bias. Methods imputing solely using subsets of sites identified by the C-C analysis suffered from bias towards the null. However, if these subsets were added into random bins of sites, this bias was reduced. The optimal methods were applied to an EWAS with missingness in covariates. All methods identified additional sites over the C-C analysis, and many of these sites had been replicated in other studies. These methods are also applicable to other high-dimensional data sets, including the rapidly expanding area of "-omics" studies.

The effect of a lifestyle intervention in obese pregnant women on gestational metabolic profiles: findings from the UK Pregnancies Better Eating and Activity Trial (UPBEAT) randomised controlled trial.

BACKGROUND: Pregnancy is associated with widespread change in metabolism, which may be more marked in obese women. Whether lifestyle interventions in obese pregnant women improve pregnancy metabolic profiles remains unknown. Our objectives were to determine the magnitude of change in metabolic measures during obese pregnancy, to indirectly compare these to similar profiles in a general pregnant population, and to determine the impact of a lifestyle intervention on change in metabolic measures in obese pregnant women. METHODS: Data from a randomised controlled trial of 1158 obese (BMI ≥ 30 kg/m2) pregnant women recruited from six UK inner-city obstetric departments were used. Women were randomised to either the UPBEAT intervention, a tailored complex lifestyle intervention focused on improving diet and physical activity, or standard antenatal care (control group). UPBEAT has been shown to improve diet and physical activity during pregnancy and up to 6-months postnatally in obese women and to reduce offspring adiposity at 6-months; it did not affect risk of gestational diabetes (the primary outcome). Change in the concentrations of 158 metabolic measures (129 lipids, 9 glycerides and phospholipids, and 20 low-molecular weight metabolites), quantified three times during pregnancy, were compared using multilevel models. The role of chance was assessed with a false discovery rate of 5% adjusted p values. RESULTS: All very low-density lipoprotein (VLDL) particles increased by 1.5-3 standard deviation units (SD) whereas intermediate density lipoprotein and specific (large, medium and small) LDL particles increased by 1-2 SD, between 16 and 36 weeks' gestation. Triglycerides increased by 2-3 SD, with more modest changes in other metabolites. Indirect comparisons suggest that the magnitudes of change across pregnancy in these obese women were 2- to 3-fold larger than in unselected women (n = 4260 in cross-sectional and 583 in longitudinal analyses) from an independent, previously published, study. The intervention reduced the rate of increase in extremely large, very large, large and medium VLDL particles, particularly those containing triglycerides. CONCLUSION: There are marked changes in lipids and lipoproteins and more modest changes in other metabolites across pregnancy in obese women, with some evidence that this is more marked than in unselected pregnant women. The UPBEAT lifestyle intervention may contribute to a healthier metabolic profile in obese pregnant women, but our results require replication. TRIAL REGISTRATION: UPBEAT was registered with Current Controlled Trials, ISRCTN89971375 , on July 23, 2008 (prior to recruitment).

Exposure Patterns Driving Ebola Transmission in West Africa: A Retrospective Observational Study.

BACKGROUND: The ongoing West African Ebola epidemic began in December 2013 in Guinea, probably from a single zoonotic introduction. As a result of ineffective initial control efforts, an Ebola outbreak of unprecedented scale emerged. As of 4 May 2015, it had resulted in more than 19,000 probable and confirmed Ebola cases, mainly in Guinea (3,529), Liberia (5,343), and Sierra Leone (10,746). Here, we present analyses of data collected during the outbreak identifying drivers of transmission and highlighting areas where control could be improved. METHODS AND FINDINGS: Over 19,000 confirmed and probable Ebola cases were reported in West Africa by 4 May 2015. Individuals with confirmed or probable Ebola ("cases") were asked if they had exposure to other potential Ebola cases ("potential source contacts") in a funeral or non-funeral context prior to becoming ill. We performed retrospective analyses of a case line-list, collated from national databases of case investigation forms that have been reported to WHO. These analyses were initially performed to assist WHO's response during the epidemic, and have been updated for publication. We analysed data from 3,529 cases in Guinea, 5,343 in Liberia, and 10,746 in Sierra Leone; exposures were reported by 33% of cases. The proportion of cases reporting a funeral exposure decreased over time. We found a positive correlation (r = 0.35, p < 0.001) between this proportion in a given district for a given month and the within-district transmission intensity, quantified by the estimated reproduction number (R). We also found a negative correlation (r = -0.37, p < 0.001) between R and the district proportion of hospitalised cases admitted within ≤4 days of symptom onset. These two proportions were not correlated, suggesting that reduced funeral attendance and faster hospitalisation independently influenced local transmission intensity. We were able to identify 14% of potential source contacts as cases in the case line-list. Linking cases to the contacts who potentially infected them provided information on the transmission network. This revealed a high degree of heterogeneity in inferred transmissions, with only 20% of cases accounting for at least 73% of new infections, a phenomenon often called super-spreading. Multivariable regression models allowed us to identify predictors of being named as a potential source contact. These were similar for funeral and non-funeral contacts: severe symptoms, death, non-hospitalisation, older age, and travelling prior to symptom onset. Non-funeral exposures were strongly peaked around the death of the contact. There was evidence that hospitalisation reduced but did not eliminate onward exposures. We found that Ebola treatment units were better than other health care facilities at preventing exposure from hospitalised and deceased individuals. The principal limitation of our analysis is limited data quality, with cases not being entered into the database, cases not reporting exposures, or data being entered incorrectly (especially dates, and possible misclassifications). CONCLUSIONS: Achieving elimination of Ebola is challenging, partly because of super-spreading. Safe funeral practices and fast hospitalisation contributed to the containment of this Ebola epidemic. Continued real-time data capture, reporting, and analysis are vital to track transmission patterns, inform resource deployment, and thus hasten and maintain elimination of the virus from the human population.

Transmission Potential of Influenza A(H7N9) Virus, China, 2013-2014.

To determine transmission potential of influenza A(H7N9) virus, we used symptom onset data to compare 2 waves of infection in China during 2013-2014. We found evidence of increased transmission potential in the second wave and showed that live bird market closure was significantly less effective in Guangdong than in other regions.

The spatial resolution of epidemic peaks.

The emergence of novel respiratory pathogens can challenge the capacity of key health care resources, such as intensive care units, that are constrained to serve only specific geographical populations. An ability to predict the magnitude and timing of peak incidence at the scale of a single large population would help to accurately assess the value of interventions designed to reduce that peak. However, current disease-dynamic theory does not provide a clear understanding of the relationship between: epidemic trajectories at the scale of interest (e.g. city); population mobility; and higher resolution spatial effects (e.g. transmission within small neighbourhoods). Here, we used a spatially-explicit stochastic meta-population model of arbitrary spatial resolution to determine the effect of resolution on model-derived epidemic trajectories. We simulated an influenza-like pathogen spreading across theoretical and actual population densities and varied our assumptions about mobility using Latin-Hypercube sampling. Even though, by design, cumulative attack rates were the same for all resolutions and mobilities, peak incidences were different. Clear thresholds existed for all tested populations, such that models with resolutions lower than the threshold substantially overestimated population-wide peak incidence. The effect of resolution was most important in populations which were of lower density and lower mobility. With the expectation of accurate spatial incidence datasets in the near future, our objective was to provide a framework for how to use these data correctly in a spatial meta-population model. Our results suggest that there is a fundamental spatial resolution for any pathogen-population pair. If underlying interactions between pathogens and spatially heterogeneous populations are represented at this resolution or higher, accurate predictions of peak incidence for city-scale epidemics are feasible.

Adiposity and cardiovascular outcomes in three-year-old children of participants in UPBEAT, an RCT of a complex intervention in pregnant women with obesity.

BACKGROUND: Maternal obesity is associated with offspring cardiometabolic risk. UPBEAT was a randomised controlled trial of an antenatal diet and physical activity intervention in 1555 women with obesity. The intervention was associated with lower gestational weight gain, healthier diet and metabolic profile in pregnancy, and reduced infant adiposity at six months. OBJECTIVE: We have investigated whether the UPBEAT intervention influenced childhood cardiometabolic outcomes or was associated with sustained improvements in maternal lifestyle 3-years after delivery. METHODS: In UPBEAT mother-child dyads at the 3-year follow-up, we assessed childhood blood pressure, resting pulse rate, and adiposity (body mass index, skinfold thicknesses, body fat, waist and arm circumferences) and maternal diet, physical activity, and anthropometry. RESULTS: 514 three-year-old children attended the appointment (49% intervention, 51% standard care). There was no difference in the main outcome of interest, subscapular skinfold thickness, between the trial arms (-0.30 mm, 95% confidence interval: -0.92, 0.31). However, the intervention was associated with a lower resting pulse rate (-5 bpm [-8.41, -1.07]). There was also a non-significant lower odds of overweight/obesity (OR 0.73; 0.50, 1.08). Maternal dietary improvements observed in the UPBEAT trial, including glycaemic load and saturated fat were maintained 3-years postpartum. CONCLUSION: This study has demonstrated that an antenatal dietary and physical activity intervention in women with obesity is associated with lower offspring pulse rate and sustained improvement in maternal diet. Whilst larger than previous cohorts, there remains potential for bias from attrition and these findings require validation in future cohorts.

Storage of prescription veterinary medicines on UK dairy farms: a cross-sectional study.

Prescription veterinary medicine (PVM) use in the UK is an area of increasing focus for the veterinary profession. While many studies measure antimicrobial use on dairy farms, none report the quantity of antimicrobials stored on farms, nor the ways in which they are stored. The majority of PVM treatments occur in the absence of the prescribing veterinarian, yet there is an identifiable knowledge gap surrounding PVM use and farmer decision making. To provide an evidence base for future work on PVM use, data were collected from 27 dairy farms in England and Wales in Autumn 2016. The number of different PVMs stored on farms ranged from 9 to 35, with antimicrobials being the most common therapeutic group stored. Injectable antimicrobials comprised the greatest weight of active ingredient found, while intramammary antimicrobials were the most frequent unit of medicine stored. Antimicrobials classed by the European Medicines Agency as critically important to human health were present on most farms, and the presence of expired medicines and medicines not licensed for use in dairy cattle was also common. The medicine resources available to farmers are likely to influence their treatment decisions; therefore, evidence of the PVM stored on farms can help inform understanding of medicine use.

Evaluation of metrics for benchmarking antimicrobial use in the UK dairy industry.

The issue of antimicrobial resistance is of global concern across human and animal health. In 2016, the UK government committed to new targets for reducing antimicrobial use (AMU) in livestock. Although a number of metrics for quantifying AMU are defined in the literature, all give slightly different interpretations. This paper evaluates a selection of metrics for AMU in the dairy industry: total mg, total mg/kg, daily dose and daily course metrics. Although the focus is on their application to the dairy industry, the metrics and issues discussed are relevant across livestock sectors. In order to be used widely, a metric should be understandable and relevant to the veterinarians and farmers who are prescribing and using antimicrobials. This means that clear methods, assumptions (and possible biases), standardised values and exceptions should be published for all metrics. Particularly relevant are assumptions around the number and weight of cattle at risk of treatment and definitions of dose rates and course lengths; incorrect assumptions can mean metrics over-represent or under-represent AMU. The authors recommend that the UK dairy industry work towards the UK-specific metrics using the UK-specific medicine dose and course regimens as well as cattle weights in order to monitor trends nationally.

A simple approach to measure transmissibility and forecast incidence.

Outbreaks of novel pathogens such as SARS, pandemic influenza and Ebola require substantial investments in reactive interventions, with consequent implementation plans sometimes revised on a weekly basis. Therefore, short-term forecasts of incidence are often of high priority. In light of the recent Ebola epidemic in West Africa, a forecasting exercise was convened by a network of infectious disease modellers. The challenge was to forecast unseen "future" simulated data for four different scenarios at five different time points. In a similar method to that used during the recent Ebola epidemic, we estimated current levels of transmissibility, over variable time-windows chosen in an ad hoc way. Current estimated transmissibility was then used to forecast near-future incidence. We performed well within the challenge and often produced accurate forecasts. A retrospective analysis showed that our subjective method for deciding on the window of time with which to estimate transmissibility often resulted in the optimal choice. However, when near-future trends deviated substantially from exponential patterns, the accuracy of our forecasts was reduced. This exercise highlights the urgent need for infectious disease modellers to develop more robust descriptions of processes - other than the widespread depletion of susceptible individuals - that produce non-exponential patterns of incidence.

Key data for outbreak evaluation: building on the Ebola experience.

Following the detection of an infectious disease outbreak, rapid epidemiological assessment is critical for guiding an effective public health response. To understand the transmission dynamics and potential impact of an outbreak, several types of data are necessary. Here we build on experience gained in the West African Ebola epidemic and prior emerging infectious disease outbreaks to set out a checklist of data needed to: (1) quantify severity and transmissibility; (2) characterize heterogeneities in transmission and their determinants; and (3) assess the effectiveness of different interventions. We differentiate data needs into individual-level data (e.g. a detailed list of reported cases), exposure data (e.g. identifying where/how cases may have been infected) and population-level data (e.g. size/demographics of the population(s) affected and when/where interventions were implemented). A remarkable amount of individual-level and exposure data was collected during the West African Ebola epidemic, which allowed the assessment of (1) and (2). However, gaps in population-level data (particularly around which interventions were applied when and where) posed challenges to the assessment of (3). Here we highlight recurrent data issues, give practical suggestions for addressing these issues and discuss priorities for improvements in data collection in future outbreaks.This article is part of the themed issue 'The 2013-2016 West African Ebola epidemic: data, decision-making and disease control'.

Heterogeneities in the case fatality ratio in the West African Ebola outbreak 2013-2016.

The 2013-2016 Ebola outbreak in West Africa is the largest on record with 28 616 confirmed, probable and suspected cases and 11 310 deaths officially recorded by 10 June 2016, the true burden probably considerably higher. The case fatality ratio (CFR: proportion of cases that are fatal) is a key indicator of disease severity useful for gauging the appropriate public health response and for evaluating treatment benefits, if estimated accurately. We analysed individual-level clinical outcome data from Guinea, Liberia and Sierra Leone officially reported to the World Health Organization. The overall mean CFR was 62.9% (95% CI: 61.9% to 64.0%) among confirmed cases with recorded clinical outcomes. Age was the most important modifier of survival probabilities, but country, stage of the epidemic and whether patients were hospitalized also played roles. We developed a statistical analysis to detect outliers in CFR between districts of residence and treatment centres (TCs), adjusting for known factors influencing survival and identified eight districts and three TCs with a CFR significantly different from the average. From the current dataset, we cannot determine whether the observed variation in CFR seen by district or treatment centre reflects real differences in survival, related to the quality of care or other factors or was caused by differences in reporting practices or case ascertainment.This article is part of the themed issue 'The 2013-2016 West African Ebola epidemic: data, decision-making and disease control'.

A review of epidemiological parameters from Ebola outbreaks to inform early public health decision-making.

The unprecedented scale of the Ebola outbreak in West Africa has, as of 29 April 2015, resulted in more than 10,884 deaths among 26,277 cases. Prior to the ongoing outbreak, Ebola virus disease (EVD) caused relatively small outbreaks (maximum outbreak size 425 in Gulu, Uganda) in isolated populations in central Africa. Here, we have compiled a comprehensive database of estimates of epidemiological parameters based on data from past outbreaks, including the incubation period distribution, case fatality rate, basic reproduction number (R 0 ), effective reproduction number (R t ) and delay distributions. We have compared these to parameter estimates from the ongoing outbreak in West Africa. The ongoing outbreak, because of its size, provides a unique opportunity to better understand transmission patterns of EVD. We have not performed a meta-analysis of the data, but rather summarize the estimates by virus from comprehensive investigations of EVD and Marburg outbreaks over the past 40 years. These estimates can be used to parameterize transmission models to improve understanding of initial spread of EVD outbreaks and to inform surveillance and control guidelines.

Potential Biases in Estimating Absolute and Relative Case-Fatality Risks during Outbreaks.

Estimating the case-fatality risk (CFR)-the probability that a person dies from an infection given that they are a case-is a high priority in epidemiologic investigation of newly emerging infectious diseases and sometimes in new outbreaks of known infectious diseases. The data available to estimate the overall CFR are often gathered for other purposes (e.g., surveillance) in challenging circumstances. We describe two forms of bias that may affect the estimation of the overall CFR-preferential ascertainment of severe cases and bias from reporting delays-and review solutions that have been proposed and implemented in past epidemics. Also of interest is the estimation of the causal impact of specific interventions (e.g., hospitalization, or hospitalization at a particular hospital) on survival, which can be estimated as a relative CFR for two or more groups. When observational data are used for this purpose, three more sources of bias may arise: confounding, survivorship bias, and selection due to preferential inclusion in surveillance datasets of those who are hospitalized and/or die. We illustrate these biases and caution against causal interpretation of differential CFR among those receiving different interventions in observational datasets. Again, we discuss ways to reduce these biases, particularly by estimating outcomes in smaller but more systematically defined cohorts ascertained before the onset of symptoms, such as those identified by forward contact tracing. Finally, we discuss the circumstances in which these biases may affect non-causal interpretation of risk factors for death among cases.

Errors in reported degrees and respondent driven sampling: implications for bias.

BACKGROUND: Respondent Driven Sampling (RDS) is a network or chain sampling method designed to access individuals from hard-to-reach populations such as people who inject drugs (PWID). RDS surveys are used to monitor behaviour and infection occurence over time; these estimations require adjusting to account for over-sampling of individuals with many contacts. Adjustment is done based on individuals' reported total number of contacts, assuming these are correct. METHODS: Data on the number of contacts (degrees) of individuals sampled in two RDS surveys in Bristol, UK, show larger numbers of individuals reporting numbers of contacts in multiples of 5 and 10 than would be expected at random. To mimic these patterns we generate contact networks and explore different methods of mis-reporting degrees. We simulate RDS surveys and explore the sensitivity of adjusted estimates to these different methods. RESULTS: We find that inaccurate reporting of degrees can cause large and variable bias in estimates of prevalence or incidence. Our simulations imply that paired RDS surveys could over- or under-estimate any change in prevalence by as much as 25%. These are particularly sensitive to inaccuracies in the degree estimates of individuals with who have low degree. CONCLUSIONS: There is a substantial risk of bias in estimates from RDS if degrees are not correctly reported. This is particularly important when analysing consecutive RDS samples to assess trends in population prevalence and behaviour. RDS questionnaires should be refined to obtain high resolution degree information, particularly from low-degree individuals. Additionally, larger sample sizes can reduce uncertainty in estimates.