Statistical Methods in Medical Research recent issues

Editorial: Special Issue on Statistical Bioinformatics

Gilks, W. R. — Fri, 25 Sep 2009 02:58:04 PDT

Considerations for the processing and analysis of GoldenGate-based two-colour Illumina platforms

Lynch, A., Dunning, M., Iddawela, M., Barbosa-Morais, N., Ritchie, M. — Fri, 25 Sep 2009 02:58:04 PDT

Illumina’s GoldenGate technology is a two-channel microarray platform that allows for the simultaneous interrogation of about 1 500 locations in the genome. GoldenGate has proved a flexible platform not only in the choice of those 1 500 locations, but also in the choice of the property being measured at them. It retains the desirable properties of Illumina’s BeadArrays in that the probes (in this case ‘beads’) are randomly arranged across the microarray, there are multiple instances of each probe and many samples can be processed simultaneously. As for other Illumina technologies, however, these properties are not exploited as they might be. Here we review the various common adaptations of the GoldenGate platform, review the analysis methods that are associated with each adaptation and then, with the aid of a number of example data sets we illustrate some of the improvements that can be made over the default analysis.

Stochastic models of sequence evolution including insertion--deletion events

Miklos, I., Novak, A., Satija, R., Lyngso, R., Hein, J. — Fri, 25 Sep 2009 02:58:04 PDT

Comparison of sequences that have descended from a common ancestor based on an explicit stochastic model of substitutions, insertions and deletions has risen to prominence in the last decade. Making statements about the positions of insertions-deletions (abbr. indels) is central in sequence and genome analysis and is called alignment. This statistical approach is harder conceptually and computationally, than competing approaches based on choosing an alignment according to some optimality criteria. But it has major practical advantages in terms of testing evolutionary hypotheses and parameter estimation. Basic dynamic approaches can allow the analysis of up to 4—5 sequences. MCMC techniques can bring this to about 10—15 sequences. Beyond this, different or heuristic approaches must be used. Besides the computational challenges, increasing realism in the underlying models is presently being addressed. A recent development that has been especially fruitful is combining statistical alignment with the problem of sequence annotation, making statements about the function of each nucleotide/amino acid. So far gene finding, protein secondary structure prediction and regulatory signal detection has been tackled within this framework. Much progress can be reported, but clearly major challenges remain if this approach is to be central in the analyses of large incoming sequence data sets.

Modelling the evolution of multi-gene families

Nye, T. M. — Fri, 25 Sep 2009 02:58:04 PDT

A number of biological processes can lead to genes being copied within the genome of some given species. Duplicate genes of this form are called paralogs and such genes share a high degree sequence similarity as well as often having closely related functions. Some genes have become widely duplicated to form multigene families in which the copies are distributed both within the genomes of individual species and across different species. Statistical modelling of gene duplication and the evolution of multi-gene families currently lags behind well-established models of DNA sequence evolution despite an increasing volume of available data, but the analysis of multi-gene families is important as part of a wider effort to understand evolution at the genomic level. This article reviews existing approaches to modelling multi-gene families and presents various challenges and possibilities for this exciting area of research.

Probabilistic models and machine learning in structural bioinformatics

Hamelryck, T. — Fri, 25 Sep 2009 02:58:04 PDT

Structural bioinformatics is concerned with the molecular structure of biomacromolecules on a genomic scale, using computational methods. Classic problems in structural bioinformatics include the prediction of protein and RNA structure from sequence, the design of artificial proteins or enzymes, and the automated analysis and comparison of biomacromolecules in atomic detail. The determination of macromolecular structure from experimental data (for example coming from nuclear magnetic resonance, X-ray crystallography or small angle X-ray scattering) has close ties with the field of structural bioinformatics. Recently, probabilistic models and machine learning methods based on Bayesian principles are providing efficient and rigorous solutions to challenging problems that were long regarded as intractable. In this review, I will highlight some important recent developments in the prediction, analysis and experimental determination of macromolecular structure that are based on such methods. These developments include generative models of protein structure, the estimation of the parameters of energy functions that are used in structure prediction, the superposition of macromolecules and structure determination methods that are based on inference. Although this review is not exhaustive, I believe the selected topics give a good impression of the exciting new, probabilistic road the field of structural bioinformatics is taking.

Capture--recapture and anchored prevalence estimation of injecting drug users in England: national and regional estimates

Hay, G., Gannon, M., MacDougall, J., Eastwood, C., Williams, K., Millar, T. — Thu, 30 Jul 2009 04:22:54 PDT

Capture—recapture (C—RC) using four data sources, one of which accounted for 81% of captured injectors, and multiple indicator methods (MIM) were used to obtain national, regional and local estimates of the prevalence of injecting drug use among opiate and/or crack cocaine users in England. Persons aged 15 to 64 years, in contact with health and/or criminal justice services during 2005/2006, and known to be using opiates and/or crack cocaine and injecting drugs were included in the C—RC analysis. The MIM analysis included indicators relating to drug treatment, drug-related deaths, population density and drug offences.

There were an estimated 130,000 opiate and/or crack cocaine users who injected drugs in 2005/06 (95% confidence interval 125,800 to 137,000), corresponding to 3.9 per thousand of the population aged 15 to 64 years (95% confidence interval 3.8—4.1). Regional variation in the prevalence of injecting was evident, ranging from 6.1 per thousand of the population aged 15 to 64 years in Yorkshire and the Humber (95% confidence interval 5.6 to 6.6) to 2.3 per thousand in the East of England (95% confidence interval 1.8 to 2.9). Application of gender and age-group distributions for treated injecting drug users (IDUs) to the prevalence estimates suggested that there were 97,200 male injectors (95% confidence interval 94,000 to 102,500) and 63,600 female injectors aged 25 to 34 years (95% confidence interval 61,500 to 67,000).

The prevalence estimates provide a basis from which numbers of current IDUs infected with hepatitis C virus (HCV) can be approximated.

Estimating current injectors in Scotland and their drug-related death rate by sex, region and age-group via Bayesian capture--recapture methods

King, R., Bird, S. M, Hay, G., Hutchinson, S. J — Thu, 30 Jul 2009 04:22:54 PDT

Using Bayesian capture—recapture methods, we estimate current injectors in Scotland in 2003, and, thereby, injectors' drug-related death rates for the period 2003—2005. Four different data sources are considered [Hepatitis C Virus (HCV) database, hospital admissions, social enquiry reports, and drug misuse database reports by General Practices or Drug Treatment Agencies] which provide covariate information on sex, region (Greater Glasgow versus elsewhere in Scotland) and age group (15—34 years and 35+ years).

We quantified Scotland's current injectors in 2003 at 27,400 (95% highest probability density interval: 20,700—32,100) by incorporating underlying model uncertainty in terms of the possible interactions present between data sources and/or covariates. The posterior probability was 72% that Scotland had more current injectors in 2003 than in 2000. Detailed comparison with 2000 gave evidence of importantly changed numbers of current injectors for different covariate classes.

In addition, and of particular social interest, is the estimation of injectors' drug-related death rates. Expert information was used to construct upper and lower bounds on the number of drug-related deaths pertaining to injectors, which were then used to provide bounds on injectors' drug-related death rates. Failure to incorporate expert information could result in over-estimation of drug-related death rates for subclasses of injectors.

An evidence synthesis approach to estimating Hepatitis C Prevalence in England and Wales

De Angelis, D., Sweeting, M., Ades, A., Hickman, M., Hope, V., Ramsay, M. — Thu, 30 Jul 2009 04:22:54 PDT

In England and Wales, routine monitoring has been consistently showing an increase in mortality and morbidity due to infection with the Hepatitis C Virus (HCV). However, the magnitude of the underlying epidemic is still the subject of debate. In this paper we present estimates of the number of individuals aged 15—59 chronically infected with HCV in 2003, derived from a Bayesian synthesis of information available from multiple sources on the size of the groups at risk for HCV and the risk specific anti-HCV prevalence. Results show that the number of chronic infections is of the order of 142,000 (95% CrI: 90,000, 231,000), with the majority (85%, 95% CrI: 74%, 93%) in injecting drug users and about 80% (95% CrI: 74%, 84%) in the age group 15—44.

Estimating the prevalence of ex-injecting drug use in the population

Sweeting, M., De Angelis, D., Ades, A., Hickman, M. — Thu, 30 Jul 2009 04:22:54 PDT

Injecting drug use is the main route of transmission for the hepatitis C virus in the developed world. Knowledge about the characteristics of the past and current injecting drug user (IDU) population is therefore vital in order to understand the epidemiology of hepatitis C. The IDU population is `hard to reach' and hence most epidemiological studies have concentrated on estimating current IDU prevalence, whilst little is known about the potentially large pool of ex-injectors. We demonstrate a method for estimating the proportion of ex-users in the population, by considering injecting drug use as a time-to-event process. We show how unbiased estimates of injecting duration and historical patterns in injecting initiation can be derived from a sample of ex-IDUs obtained from a population survey, and how such data lead to estimates of the proportion of ex-IDUs in the population. Finally, we show how to obtain estimates of the prevalence of ex-IDUs by using additional information on the prevalence of current IDUs.

Optimal Designs for Empirical Bayes Estimators of Individual Linear and Quadratic Growth Curves in Linear Mixed Models

Candel, M. J. — Thu, 30 Jul 2009 04:22:54 PDT

Many studies on optimal designs for linear mixed model analysis of repeated measures data have focussed on estimating the fixed effects. The present study investigates the optimal number of time points and subjects in case random effects have to be estimated. Linear mixed models with a linear or quadratic trend across equidistant time points are studied. Given a particular cost function, we examine which designs minimise the expected average squared prediction error. Robustness of the optimal design, important when one does not know the underlying model, is also treated.

Meta-analyses of safety data: a comparison of exact versus asymptotic methods

Thu, 30 Jul 2009 04:22:54 PDT

The objectives of this study were to establish and describe a database of Cochrane and non-Cochrane meta-analyses of safety data and to determine under what conditions exact methods differ from asymptotic methods in meta-analyses of safety data. A sample of Cochrane (n = 500) and non-Cochrane (n = 200) systematic reviews was randomly selected and a database of safety meta-analyses established. Point estimates and confidence intervals for each meta-analysis were recalculated using exact methods and compared to the results of asymptotic methods. Cochrane reviews were nearly four times as likely as non-Cochrane reviews to contain meta-analyses of safety data (35% compared to 9%). More than 50% of safety meta-analyses contained an outcome with a rare event rate (<5%) and 30% contained at least one study with no events in one arm of the study. For rare event meta-analyses, exact point estimates differed substantially from asymptotic estimates 46% of the time, compared to 17% for those without rare events. Exact confidence intervals differed substantially from asymptotic ones 67% of the time compared to only 19% for those without rare events. The magnitude of differences was also correlated with the number of studies and the summary statistic used to combine the data. Asymptotic methods will not always be a good approximation for exact methods in safety meta-analyses. Event rates and number of studies should be closely examined when choosing the statistical method for combining rare event data.

Special Issue of Statistical Methods in Medical Research on: hepatitis C virus and injection drug use

Bird, S. M — Wed, 27 May 2009 07:48:41 PDT

Modelling the past, current and future HCV burden in France: detailed analysis and perspectives

Deuffic-Burban, S., Mathurin, P., Valleron, A.-J. — Wed, 27 May 2009 07:48:41 PDT

Two national HCV projections have been published in France which assumed that a part of observed hepatocellular carcinoma (HCC) deaths is a consequence of HCV epidemic. They applied the back-calculation method, in combination with a Markov model, to reconstruct the past history of HCV infection and then to predict HCV-related mortality. A preliminary model was first developed in the absence of effective therapy. It allowed testing many assumptions to model HCV natural history that were compatible with observed incidence of HCV-related HCC deaths. This model was then updated to take into account the availability of treatment and more recent epidemiological data. These two models are described in detail and results are discussed with a view to addressing the models' limitations. The models offered a useful tool to assess public health policy scenarios in planning healthcare responses to the HCV epidemic.

Modelling and calibration of the hepatitis C epidemic in Australia

Razali, K., Amin, J., Dore, G., Law, M., HCV Projections Working Group, — Wed, 27 May 2009 07:48:41 PDT

Hepatitis C virus (HCV) infection in Australia is predominantly transmitted through injecting drug use. A reduction in the heroin supply in Australia in late 2000 and early 2001 may have impacted the number of injecting drug users (IDUs) and the number of new hepatitis C infections. This paper updates estimates of HCV incidence between 1960 and 2005 and models long-term sequelae from infection. Outcomes among those with HCV were also recently assessed in a linkage study assessing cancer and causes of death following HCV diagnosis in New South Wales. Linkage study outcomes have been used here to calibrate modelled outcomes. Mathematical models were used to estimate HCV incidence among IDUs, migrants to Australia from high HCV-prevalence countries, and other HCV exposure groups. Recent trends in numbers of IDUs were based on indicators of injecting drug use. A natural history of HCV model was applied to estimate the prevalence of HCV in the population. Model predicted endpoints that were calibrated against the NSW linkage data over the period 1995—2002 were: (i) incident hepatocellular carcinoma (HCC); (ii) opioid overdose deaths; (iii) liver-related deaths; and (iv) all-cause mortality. Modelled estimates and the linkage data show reasonably good calibration for HCC cases and all-cause mortality. The estimated HCC incidence was increased from 70 cases in 1995 to 100 cases in 2002. All-cause mortality estimated at 1000 in 1995 increased to 1600 in 2002. Comparison of annual opioid deaths shows some agreement. However, the models underestimate the rate of increase observed between 1995 and 1999 and do not entirely capture the rapid decrease in overdose deaths from 2000 onwards. The linkage data showed a peak of overdose deaths at 430 in 1999 compared to 320 estimated by the models. Comparison of observed liver deaths with the modelled numbers showed poor agreement. A good agreement would require an increase in liver deaths from the assumed 2 to 5% per annum following cirrhosis in the models. Mathematical models suggest that HCV incidence decreased from a peak of 14,000 infections in 1999 to 9700 infections in 2005, largely attributable to a reduction in injecting drug use. The poor agreement between projected and linked liver deaths could reflect differing coding of causes of deaths, underestimates of the numbers of people with cirrhosis following HCV, or underestimates of rates of liver death following cirrhosis. The reasonably good agreement between most of the modelled estimates with observed linkage data provides some support for the assumptions used in the models.

A population-based record linkage study of mortality in hepatitis C-diagnosed persons with or without HIV coinfection in Scotland

Wed, 27 May 2009 07:48:41 PDT

Infection with the hepatitis C virus (HCV) is known to increase the risk of death from severe liver disease and, because HCV status is strongly associated with a history of injecting drug use, the effect of a key disease progression cofactor, infection with human immunodeficiency virus (HIV), is of interest. We examined all-cause, liver-related and drug-related mortality and excess risk of death from these causes in a large cohort of HCV-monoinfected and HIV-coinfected persons in Scotland. The study population consisted of 20,163 persons confirmed to be infected with hepatitis C through laboratory testing in Scotland between 1991 and 2005. Records with sufficient identifiers were linked to the General Register Office for Scotland death register to retrieve associated mortality data, and were further linked to a national database of HIV-positive individuals to determine coinfection status. A total of 1715 HCV monoinfected and 305 HIV coinfected persons died of any cause during the follow-up period (mean of 5.4 and 6.4 years, respectively). Significant excess mortality was observed in both HCV monoinfected and HIV coinfected populations from liver-related underlying causes (standardised mortality ratios of 25, 95% CI = 23—27; and 37, 95% CI = 26—52 for the two groups, respectively) and drug-related causes (25, 95% CI = 23—27; 39, 95% CI = 28—53. The risk of death from hepatocellular carcinoma, alcoholic or non-alcoholic liver disease, or from a drug-related cause, was greatly increased compared with the general Scottish population, with the highest standardised mortality ratio observed for hepatocellular carcinoma in the monoinfected group (70, 95% CI = 57—85). This study has revealed considerable excess mortality from liver- and drug-related causes in the Scottish HCV-diagnosed population; these data are crucial to inform on the clinical management, and projected future public health burden, of HCV infection.

Measurement of food and alcohol intake in relation to chronic liver disease

Wrieden, W. L, Anderson, A. S — Wed, 27 May 2009 07:48:41 PDT

It is well established that the consumption of alcohol is implicated in both the cause and progression of chronic liver disease. The quantity of drink that is consumed, the pattern of drinking and type of alcoholic beverages consumed are all possible factors in disease aetiology. The impact of specific dietary components on the cause and progression of chronic liver disease is unclear although it is known that obesity, and hence the over-consumption of energy, is a predictor of fatty liver. Work to elucidate the role of both diet and alcohol in the aetiology of liver disease is hindered by the methods currently available to measure dietary (including alcohol) intake. The validity and reliability of retrospective methods of assessing diet are limited by their reliance on memory and, for the 24 h recall, the short-time period of intake assessed and its inability to assess variability across the week. Prospective methods which measure food and drink intake at the time of consumption, and include weighed or estimated food diaries, are useful for prospective cohort studies but are expensive and have a high respondent burden. For estimation of alcohol intake retrospectively, the Cognitive Lifetime Drinking questionnaire, which prompts responses using a lifetime calendar, is a useful tool but still depends on memory. More work is required to develop valid, reliable and easily administered tools for measurement of both diet and alcohol.

Re-weighted inference about hepatitis C virus-infected communities when analysing diagnosed patients referred to liver clinics

Bo Fu, , Tom, B. D., Bird, S. M — Wed, 27 May 2009 07:48:41 PDT

To project national hepatitis C virus (HCV) burden, unbiased estimation of HCV progression to liver cirrhosis is required for the whole community of HCV-infected individuals. However, widely varying estimates of progression rates to cirrhosis have been produced. This disparity is partly associated with the statistical methods applied, but is mainly due to the differing types of study cohort. We use an inverse probability weighted estimation method to recover the true parameters for the (Weibull regression) model that determines the incubation period from infection to cirrhosis for the community of HCV-infected individuals, when there is cirrhosis-related recruitment bias to the studied cohort. We apply the method to simulated data for a liver clinic which attracts patients from a community of 1000 HCV-infected individuals under different event-biased referral patterns. We investigate how well the method performs in recovering the true community parameters, and then apply it to Edinburgh Royal Infirmary's liver clinic series. The results obtained are compared to those from a Weibull survival analysis which ignores the selection bias.

Recursive estimation method for predicting residual bladder urine volumes to improve accuracy of timed urine collections

Afshartous, D., Preston, R. A — Thu, 19 Mar 2009 06:57:40 PDT

Clinical research studies often collect data via repeated measurements of collected urine. Unfortunately, the accuracy of timed urine collections is limited by the presence of a residual volume of urine remaining in the bladder following each timed void due to incomplete emptying of the bladder. This residual urine volume adds significant imprecision to the urine collection method, rendering an important and fundamental clinical research tool inaccurate. We present an unbiased method to estimate the residual bladder volumes via a mathematical model of the bladder process. Regardless of the substance of primary interest, the model leverages conservation of mass and conservation of concentration principles towards a substance of secondary interest in order to solve a system of recursive equations, resulting in our Recursive Residual Estimation method to predict the residual volumes at each time point. We verify the model on simulated patients and also investigate the sensitivity of the model to initial value specification.

Some notes on parametric link functions in clinical research

Baldi, I., Maule, M., Bigi, R., Cortigiani, L., Bo, S., Gregori, D. — Thu, 19 Mar 2009 06:57:40 PDT

A piecewise-constant Markov model and the effects of study design on the estimation of life expectancies in health and ill health

van den Hout, A., Matthews, F. E — Thu, 19 Mar 2009 06:57:40 PDT

Multi-state models are frequently applied to describe transitions over time between three states: healthy, not healthy and death. The three-state model can be used to estimate life expectancies in health and ill health. In this article, continuous-time Markov models are specified for the transitions between the three states. Transition intensities are regressed on age as a time-dependent covariate. The covariate is handled in a piecewise-constant fashion where the time interval between two consecutive observations is divided into subintervals of fixed and equal lengths. Study design choices such as sample size, length of follow-up, and time intervals between observations are investigated in a simulation study. The effects on parameter estimation are discussed as well as the effects on the estimation of life expectancies. In addition, data taken from the UK Cognitive Functioning and Ageing Study are analysed.

A deterministic model for estimating the reduction in colorectal cancer incidence due to endoscopic surveillance

Cafferty, F. H, Sasieni, P. D, Duffy, S. W — Thu, 19 Mar 2009 06:57:40 PDT

There is evidence that the removal of adenomas, by endoscopy, from the large bowel can prevent the occurrence of colorectal cancer (CRC). However, the reduction in cancer incidence due to endoscopic surveillance is difficult to estimate. Studies of cohorts of adenoma patients typically rely on comparisons with groups of historical controls. We present a model for disease progression which enables estimation of this quantity without direct comparison to a reference group. Models are applied to data from the National Polyp Study. Rates of adenoma recurrence and progression to carcinoma are estimated based on study data and relevant literature. This allows calculation of the number of cancers expected in the absence of surveillance and, thus, the number of cancers prevented. Results are compared with the original analysis. Models estimate that surveillance reduced CRC incidence by at least 97% in this cohort. The majority of the effect was due to the initial removal of adenomas rather than the follow-up surveillance. These results are similar to those produced in the original analysis when using the most appropriate reference groups. They indicate that polypectomy and follow-up surveillance can lead to large reductions in cancer incidence which may have been under-estimated in previous studies.

The choice of sample size: a mixed Bayesian / frequentist approach

Pezeshk, H., Nematollahi, N., Maroufy, V., Gittins, J. — Thu, 19 Mar 2009 06:57:40 PDT

Sample size computations are largely based on frequentist or classical methods. In the Bayesian approach the prior information on the unknown parameters is taken into account. In this work we consider a fully Bayesian approach to the sample size determination problem which was introduced by Grundy et al. and developed by Lindley. This approach treats the problem as a decision problem and employs a utility function to find the optimal sample size of a trial. Furthermore, we assume that a regulatory authority, which is deciding on whether or not to grant a licence to a new treatment, uses a frequentist approach. We then find the optimal sample size for the trial by maximising the expected net benefit, which is the expected benefit of subsequent use of the new treatment minus the cost of the trial.

Multi-state models for the analysis of time-to-event data

Meira-Machado, L., de Una-Alvarez, J., Cadarso-Suarez, C., Andersen, P. K — Thu, 19 Mar 2009 06:57:40 PDT

The experience of a patient in a survival study may be modelled as a process with two states and one possible transition from an "alive" state to a "dead" state. In some studies, however, the "alive" state may be partitioned into two or more intermediate (transient) states, each of which corresponding to a particular stage of the illness. In such studies, multi-state models can be used to model the movement of patients among the various states. In these models issues, of interest include the estimation of progression rates, assessing the effects of individual risk factors, survival rates or prognostic forecasting. In this article, we review modelling approaches for multi-state models, and we focus on the estimation of quantities such as the transition probabilities and survival probabilities. Differences between these approaches are discussed, focussing on possible advantages and disadvantages for each method. We also review the existing software currently available to fit the various models and present new software developed in the form of an R library to analyse such models. Different approaches and software are illustrated using data from the Stanford heart transplant study and data from a study on breast cancer conducted in Galicia, Spain.

Book review: Cook D, Swayne DF 2007: Interactive and dynamic graphics for data analysis, With R and GGobi. New York: Springer. 190 pp. $59.95 (PB). ISBN: 978-0-387-71761-6

Stoimenova, E. — Thu, 19 Mar 2009 06:57:40 PDT

Book review: Lombardo JS, Buckeridge DL (eds) 2007: Disease surveillance -- a public health informatics approach. New Jersey: John Wiley & Sons. 458 pp. ISBN 978-0-470-06812-0

Balram, S. — Thu, 19 Mar 2009 06:57:40 PDT

Erratum

Thu, 19 Mar 2009 06:57:40 PDT