Model selection for prognostic time-to-event gene signature discovery with applications in early breast cancer data

Model selection between competing models is a key consideration in the discovery of prognostic multigene signatures. The use of appropriate statistical performance measures as well as verification of biological significance of the signatures is imperative to maximise the chance of external validation of the generated signatures. Current approaches in time-to-event studies often use only a single measure of performance in model selection, such as logrank test p-values, or dichotomise the follow-up times at some phase of the study to facilitate signature discovery. In this study we improve the prognostic signature discovery process through the application of the multivariate partial Cox model combined with the concordance index, hazard ratio of predictions, independence from available clinical covariates and biological enrichment as measures of signature performance. The proposed framework was applied to discover prognostic multigene signatures from early breast cancer data. The partial Cox model combined with the multiple performance measures were used in both guiding the selection of the optimal panel of prognostic genes and prediction of risk within cross validation without dichotomising the follow-up times at any stage. The signatures were successfully externally cross validated in independent breast cancer datasets, yielding a hazard ratio of 2.55 [1.44, 4.51] for the top ranking signature.

The relationship between breastfeeding and postpartum weight change-a systematic review and critical evaluation

Pregnancy and the postpartum period is a time of increased vulnerability for retention of excess body fat in women. Breastfeeding (BF) has been shown to have many health benefits for both mother and baby; however, its role in postpartum weight management is unclear. Our aim was to systematically review and critically appraise the literature published to date in relation to the impact of BF on postpartum weight change, weight retention and maternal body composition. Electronic literature searches were carried out using MEDLINE, EMBASE, PubMed, Web of Science, BIOSIS, CINAHL and British Nursing Index. The search covered publications up to 12 June 2012 and included observational studies (prospective and retrospective) carried out in BF mothers (either exclusively or as a subgroup), who were 2 years postpartum and with a body mass index (BMI) >18.5 kg m(-2), with an outcome measure of change in weight (including weight retention) and/or body composition. Thirty-seven prospective studies and eight retrospective studies were identified that met the selection criteria; studies were stratified according to study design and outcome measure. Overall, studies were heterogeneous, particularly in relation to sample size, measurement time points and in the classification of BF and postpartum weight change. The majority of studies reported little or no association between BF and weight change (n=27, 63%) or change in body composition (n=16, 89%), although this seemed to depend on the measurement time points and BF intensity. However, of the five studies that were considered to be of high methodological quality, four studies demonstrated a positive association between BF and weight change. This systematic review highlights the difficulties of examining the association between BF and weight management in observational research. Although the available evidence challenges the widely held belief that BF promotes weight loss, more robust studies are needed to reliably assess the impact of BF on postpartum weight management.International Journal of Obesity advance online publication, 20 August 2013; doi:10.1038/ijo.2013.132.

A randomized trial of paired tutoring in elementary schools:effects on self-esteem

This paper reports data on self-esteem collected during a large-scale randomized trial of paired reading and Duolog Maths with primary school children (The Fife Peer-Learning Project). Self-report measures were analysed to assess the effect of a range of organisational variables on measured self-esteem. The presentation reports findings from a sample of reading-only schools in the first year of the project. Statistically significant gains in self-esteem were found across a range of conditions. Patterns of scores pointed to differences between same-age and cross-age conditions, and highlighted the influence of role played (tutor v tutee). In addition to the findings summarized here, it is hoped that preliminary findings from year two of the project - with an increased sample size - will also be shared. © 2009.

Seasonality of cardiovascular risk factors: an analysis including over 230 000 participants in 15 countries

Objective: To assess the seasonality of cardiovascular risk factors (CVRF) in a large set of population-based studies.

Methods: Cross-sectional data from 24 population-based studies from 15 countries, with a total sample size of 237 979 subjects. CVRFs included Body Mass Index (BMI) and waist circumference; systolic (SBP) and diastolic (DBP) blood pressure; total, high (HDL) and low (LDL) density lipoprotein cholesterol; triglycerides and glucose levels. Within each study, all data were adjusted for age, gender and current smoking. For blood pressure, lipids and glucose levels, further adjustments on BMI and drug treatment were performed.

Results: In the Northern and Southern Hemispheres, CVRFs levels tended to be higher in winter and lower in summer months. These patterns were observed for most studies. In the Northern Hemisphere, the estimated seasonal variations were 0.26 kg/m2 for BMI, 0.6 cm for waist circumference, 2.9 mm Hg for SBP, 1.4 mm Hg for DBP, 0.02 mmol/L for triglycerides, 0.10 mmol/L for total cholesterol, 0.01 mmol/L for HDL cholesterol, 0.11 mmol/L for LDL cholesterol, and 0.07 mmol/L for glycaemia. Similar results were obtained when the analysis was restricted to studies collecting fasting blood samples. Similar seasonal variations were found for most CVRFs in the Southern Hemisphere, with the exception of waist circumference, HDL, and LDL cholesterol.

Conclusions: CVRFs show a seasonal pattern characterised by higher levels in winter, and lower levels in summer. This pattern could contribute to the seasonality of CV mortality.

How much do you need: a randomised experiment of whether readers can understand the key messages from summaries of Cochrane Reviews, without reading the full review?

Objective: We explored whether readers can understand key messages without having to read the full review, and if there were differences in understanding between various types of summary.
Design: A randomised experiment of review summaries which compared understanding of a key outcome.
Participants: Members of university staff (n = 36).
Setting: Universities on the island of Ireland.
Method: The Cochrane Review chosen examines the health impacts of the use of electric fans during heat waves. Participants were asked their expectation of the effect these would have on mortality. They were then randomly assigned a summary of the review (i.e. abstract, plain language summary, podcast or podcast transcription) and asked to spend a short time reading/listening to the summary. After this they were again asked about the effects of electric fans on mortality and to indicate if they would want to read the full Review.
Main outcome measure: Correct identification of a key review outcome.
Results: Just over half (53%) of the participants identified its key message on mortality after engaging with their summary. The figures were 33% for the abstract group, 50% for both the plain language and transcript groups and 78% for the podcast group.
Conclusions: The differences between the groups were not statistically significant but suggest that the audio summary might improve knowledge transfer compared to written summaries. These findings should be explored further using a larger sample size and with other reviews.

The statistical significance of randomized controlled trial results is frequently fragile: A case for a Fragility Index

Objectives

A P-value <0.05 is one metric used to evaluate the results of a randomized controlled trial (RCT). We wondered how often statistically significant results in RCTs may be lost with small changes in the numbers of outcomes.

Study Design and Setting

A review of RCTs in high-impact medical journals that reported a statistically significant result for at least one dichotomous or time-to-event outcome in the abstract. In the group with the smallest number of events, we changed the status of patients without an event to an event until the P-value exceeded 0.05. We labeled this number the Fragility Index; smaller numbers indicated a more fragile result.

Results

The 399 eligible trials had a median sample size of 682 patients (range: 15-112,604) and a median of 112 events (range: 8-5,142); 53% reported a P-value <0.01. The median Fragility Index was 8 (range: 0-109); 25% had a Fragility Index of 3 or less. In 53% of trials, the Fragility Index was less than the number of patients lost to follow-up.

Conclusion

The statistically significant results of many RCTs hinge on small numbers of events. The Fragility Index complements the P-value and helps identify less robust results. 

Milling plant and soil material in plastic tubes over-estimates carbon and under-estimates nitrogen concentrations

Milling of plant and soil material in plastic tubes, such as microcentrifuge tubes, over-estimates carbon (C) and under-estimates nitrogen (N) concentrations due to the introduction of polypropylene into milled samples, as identified using Fourier-transform infra-red spectroscopy.

This study compares C and N concentrations of roots and soil milled in microcentrifuge tubes versus stainless steel containers, demonstrating that a longer milling time, greater milling intensity, smaller sample size and inclusion of abrasive sample material all increase polypropylene contamination from plastic tubes leading to overestimation of C concentrations by up to 8 % (0.08 g g(-1)).

Erroneous estimations of C and N, and other analytes, must be assumed after milling in plastic tubes and milling methods should be adapted to minimise such error.

'Leptorapide' - a one-step assay for rapid diagnosis of human leptospirosis

Leptospirosis is a globally important zoonotic infection caused by spirochaetes of the genus Leptospira. It is transmitted to humans by direct contact with infected animals or indirectly via contaminated water. It is mainly a problem of the resource-poor developing countries of the tropical and sub-tropical regions of the world but outbreaks due to an increase in travel and recreational activities have been reported in developed and more industrialized areas of the world. Current methods of diagnosis are costly, time-consuming and require the use of specialized laboratory equipment and personnel. The purpose of this paper is to report the validation of the 'Leptorapide®' test (Linnodee Ltd, Northern Ireland) for the diagnosis of human leptospirosis. It is a simple one-step latex agglutination assay performed using equal volumes of serum sample and antigen-bound latex beads. Evidence of leptospiral antibodies is determined within minutes. Agglutination is scored on a scale of 1-5 and the results interpreted using a score card provided with the kit. Validation has been performed with a large sample size obtained from individuals originating from various parts of the world including Brazil and India. The test has shown sensitivity and specificity values of 97·1% and 94·0%, respectively, relative to the microscopic agglutination test. The results demonstrate that Leptorapide offers a cost-effective and accurate alternative to the more historical methods of antibody detection.

Min-BDeu and Max-BDeu Scores for Learning Bayesian Networks

This work presents two new score functions based on the Bayesian Dirichlet equivalent uniform (BDeu) score for learning Bayesian network structures. They consider the sensitivity of BDeu to varying parameters of the Dirichlet prior. The scores take on the most adversary and the most beneficial priors among those within a contamination set around the symmetric one. We build these scores in such way that they are decomposable and can be computed efficiently. Because of that, they can be integrated into any state-of-the-art structure learning method that explores the space of directed acyclic graphs and allows decomposable scores. Empirical results suggest that our scores outperform the standard BDeu score in terms of the likelihood of unseen data and in terms of edge discovery with respect to the true network, at least when the training sample size is small. We discuss the relation between these new scores and the accuracy of inferred models. Moreover, our new criteria can be used to identify the amount of data after which learning is saturated, that is, additional data are of little help to improve the resulting model.

Bayesian network data imputation with application to survival tree analysis

Retrospective clinical datasets are often characterized by a relatively small sample size and many missing data. In this case, a common way for handling the missingness consists in discarding from the analysis patients with missing covariates, further reducing the sample size. Alternatively, if the mechanism that generated the missing allows, incomplete data can be imputed on the basis of the observed data, avoiding the reduction of the sample size and allowing methods to deal with complete data later on. Moreover, methodologies for data imputation might depend on the particular purpose and might achieve better results by considering specific characteristics of the domain. The problem of missing data treatment is studied in the context of survival tree analysis for the estimation of a prognostic patient stratification. Survival tree methods usually address this problem by using surrogate splits, that is, splitting rules that use other variables yielding similar results to the original ones. Instead, our methodology consists in modeling the dependencies among the clinical variables with a Bayesian network, which is then used to perform data imputation, thus allowing the survival tree to be applied on the completed dataset. The Bayesian network is directly learned from the incomplete data using a structural expectation–maximization (EM) procedure in which the maximization step is performed with an exact anytime method, so that the only source of approximation is due to the EM formulation itself. On both simulated and real data, our proposed methodology usually outperformed several existing methods for data imputation and the imputation so obtained improved the stratification estimated by the survival tree (especially with respect to using surrogate splits).

Zooming In on the Progenitors of Superluminous Supernovae With the HST

We present Hubble Space Telescope (HST) rest-frame ultraviolet imaging of the host galaxies of 16 hydrogen-poor superluminous supernovae (SLSNe), including 11 events from the Pan-STARRS Medium Deep Survey. Taking advantage of the superb angular resolution of HST, we characterize the galaxies' morphological properties, sizes, and star formation rate (SFR) densities. We determine the supernova (SN) locations within the host galaxies through precise astrometric matching and measure physical and host-normalized offsets as well as the SN positions within the cumulative distribution of UV light pixel brightness. We find that the host galaxies of H-poor SLSNe are irregular, compact dwarf galaxies, with a median half-light radius of just 0.9 kpc. The UV-derived SFR densities are high ([Sigma(SFR)] similar or equal to 0.1M(circle dot) yr(-1) kpc(-1)), suggesting that SLSNe form in overdense environments. Their locations trace the UV light of their host galaxies, with a distribution intermediate between that of long-duration gamma-ray bursts (LGRBs; which are strongly clustered on the brightest regions of their hosts) and a uniform distribution (characteristic of normal core-collapse SNe), though cannot be statistically distinguished from either with the current sample size. Taken together, this strengthens the picture that SLSN progenitors require different conditions than those of ordinary core-collapse SNe to form and that they explode in broadly similar galaxies as do LGRBs. If the tendency for SLSNe to be less clustered on the brightest regions than are LGRBs is confirmed by a larger sample, this would indicate a different, potentially lower-mass progenitor for SLSNe than LRGBs.

Rapidly Evolving and Luminous Transients from Pan-STARRS1

In the past decade, several rapidly evolving transients have been discovered whose timescales and luminosities are not easily explained by traditional supernovae (SNe) models. The sample size of these objects has remained small due, at least in part, to the challenges of detecting short timescale transients with traditional survey cadences. Here we present the results from a search within the Pan-STARRS1 Medium Deep Survey (PS1-MDS) for rapidly evolving and luminous transients. We identify 10 new transients with a time above half-maximum (t_1/2) of less than 12 days and -16.5 > M > -20 mag. This increases the number of known events in this region of SN phase space by roughly a factor of three. The median redshift of the PS1-MDS sample is z = 0.275 and they all exploded in star-forming galaxies. In general, the transients possess faster rise than decline timescale and blue colors at maximum light (g_P1-rP1 ≲ -0.2). Best-fit blackbodies reveal photospheric temperatures/radii that expand/cool with time and explosion spectra taken near maximum light are dominated by a blue continuum, consistent with a hot, optically thick, ejecta. We find it difficult to reconcile the short timescale, high peak luminosity (L > 10⁴³erg s^-1), and lack of UV line blanketing observed in many of these transients with an explosion powered mainly by the radioactive decay of ⁵⁶Ni. Rather, we find that many are consistent with either (1) cooling envelope emission from the explosion of a star with a low-mass extended envelope that ejected very little (<0.03 M) radioactive material, or (2) a shock breakout within a dense, optically thick, wind surrounding the progenitor star. After calculating the detection efficiency for objects with rapid timescales in the PS1-MDS we find a volumetric rate of 4800-8000 events yr^-1Gpc^-3(4%-7% of the core-collapse SN rate at z = 0.2).

A feasibility study of auricular therapy and self-administered acupressure for insomnia following cancer treatment

Introduction: Many cancer patients experience sleeping difficulties which can persist several years after the completion of cancer treatment. Previous research suggests that acupuncture, and variants of acupuncture (acupressure, auricular therapy) may be effective treatment options for sleep disturbance. However, current evidence is limited for cancer patients.

Methods: Feasibility study with 3 arms. Seven cancer patients with insomnia randomised to receive either auricular therapy (attaching semen vaccariae seeds to ear acupoints) (n=4), self-acupressure (n=1) or no treatment (n=2). Participants assigned to receive auricular therapy or self-acupressure stimulated the acupoints each night an hour before retiring to bed. The duration of participant involvement was 5 weeks. Subjective sleep quality was measured at baseline and post-treatment using the Pittsburgh Sleep Quality Index (PSQI). The impact of treatment on concerns of importance to the participants themselves was measured using the Measure Yourself Concerns and Wellbeing (MYCaW). Each participant also completed a treatment log book.

Results: All participants completed their treatment. All auricular therapy and self-acupressure participants recorded clinically significant improvements in global PSQI scores. In the auricular therapy arm mean global PSQI reduced from 12.5 at baseline to 8 following completion of treatment. In the self-acupressure arm PSQI reduced from 15 to 11. While in the no treatment arm the mean PSQI score was 14.5 at both baseline and follow up.

Conclusions: Despite the limited sample size, both auricular therapy and self-acupressure may represent potentially effective treatments for cancer patients with insomnia. The positive findings suggest further research is warranted into both treatment modalities.

How individual participant data meta-analyses have influenced trial design, conduct, and analysis

OBJECTIVES: To demonstrate how individual participant data (IPD) meta-analyses have impacted directly on the design and conduct of trials and highlight other advantages IPD might offer.

STUDY DESIGN AND SETTING: Potential examples of the impact of IPD meta-analyses on trials were identified at an international workshop, attended by individuals with experience in the conduct of IPD meta-analyses and knowledge of trials in their respective clinical areas. Experts in the field who did not attend were asked to provide any further examples. We then examined relevant trial protocols, publications, and Web sites to verify the impacts of the IPD meta-analyses. A subgroup of workshop attendees sought further examples and identified other aspects of trial design and conduct that may inform IPD meta-analyses.

RESULTS: We identified 52 examples of IPD meta-analyses thought to have had a direct impact on the design or conduct of trials. After screening relevant trial protocols and publications, we identified 28 instances where IPD meta-analyses had clearly impacted on trials. They have influenced the selection of comparators and participants, sample size calculations, analysis and interpretation of subsequent trials, and the conduct and analysis of ongoing trials, sometimes in ways that would not possible with systematic reviews of aggregate data. We identified additional potential ways that IPD meta-analyses could be used to influence trials.

CONCLUSIONS: IPD meta-analysis could be better used to inform the design, conduct, analysis, and interpretation of trials.

A supernova distance to the anchor galaxy NGC 4258

The fortuitous occurrence of a type II-Plateau (IIP) supernova, SN 2014bc, in a galaxy for which distance estimates from a number of primary distance indicators are available provides a means with which to cross-calibrate the standardised candle method (SCM) for type IIP SNe. By applying calibrations from the literature we find distance estimates in line with the most precise measurement to NGC 4258 based on the Keplerian motion of masers (7:6 ± 0:23 Mpc), albeit with significant scatter. We provide an alternative local SCM calibration by only considering type IIP SNe that have occurred in galaxies for which a Cepheid distance estimate is available. We find a considerable reduction in scatter (σ<inf>I</inf> = 0:16 mag), but note that the current sample size is limited. Applying this calibration, we estimate a distance to NGC 4258 of 7:08 ± 0:86 Mpc.