We help you to develop and implement a clear plan of action from the outset of your project to ensure you remain on your optimal trajectory for the project duration.
In the initial stages we refine you research questions, fine tune your hypotheses for precise testing and enable a statistical analysis plan that optimizes your study results. Study power and required sample size can vary wildly as a function of statistical test and data type. Determining a statistical analysis plan up front is key as it enables us to calculate the most appropriate sample size and power for your study.
We advise on data collection and inputting techniques that reduce the need for data cleaning and preparation at the analysis stage. This saves you time and money and allows you to more efficiently answer your research questions.
We use a rigorous approach to ensure the most appropriate statistical methodology has been implemented at the final analysis stage. We will fine tune your statistical analysis plan as needed in line with any unexpected developments arising from the nuances of real-world data. We follow protocol to ensure your final analysis is as accurate and free from bias to every possible extent. We then write up the results of your study and guide your interpretation to fully ensure that any conclusions are correct and concise.
Execute your clinical research project with greater Clarity, Efficiency & Accuracy!
Biostatistics & Bioinformatics Services
Biostatistical Expert Witness
- Biostatistics & Clinical Data Analysis
- Biomedical Complex Systems & Network Analysis
- Bayesian Analysis & Clinical Trial Design
- Clinical Survey Design & Analysis
Our biostatistical consulting services are individually tailored to suit the project at hand and can cover a broad range of approaches. We can theoretically enter into collaboration at any stage of your research, whether it be the design stage, data collection, statistical analysis, reporting and preparing for presentations. Most commonly a statistical consulting project will cover one or more of the following areas:
Study design and planning
Prior to the collection of data, a suitable study design should be determined, including calculation of the appropriate sample size for the desired power or vice versa. Having a clearly outlined plan for your research can make the later stages of analysis much easier. We can also prepare randomisation schedules in preparation for patient allocation to treatment or study groups. Common study designs include pilot studies, randomized controlled trials, epidemiological studies such as retrospective and prospective cohort, case-control and cross-sectional studies.
Data collection and cleaning
We can assist you to determine, in advance, the most beneficial format to collect your data for the purposes of addressing your research questions. Having research questions clearly set out in advance of data collection allows the data collection process to be that much more efficient, as data is in an optimal format from the outset. This reduces the amount of cleaning and transformation of the data file, reducing costs and making our job a little less tedious.
Statistical programming and data analysis
Once experimental data has been collected, data analysis can be performed and/or procedures demonstrated to those interested in performing the analysis themselves. Commonly used software includes SPSS, SAS, R and Stata. Other similar statistical software can be adopted in line with individual preferences. For individuals performing their own analysis a troubleshooting service is also available. The following are some statistical approaches that are commonly encountered.
-Standard hypothesis testing:
t tests, ANOVA, ANCOVA, Mann-Whitney U test, Wilcoxon signed-rank test, chi squared analysis and survey analysis using various methods.
-Predictive models:
Linear regression, logistic regression, polynomial regression, poisson regression, negative binomial ridge regression, lasso regression, ecologic regression, logic regression, probit, jacknife regression
-Latent variable models:
PCA, Factor analysis, Common factor models, item response theory models, latent class model, structural equation models with latent variables.
-Partitioning methods:
Heirachical clustering, k-means clustering, medoid clustering, , fuzzy clustering, and regression clustering.
-Generative models:
Bayesian generative models, dynamic causal models, Markov models.
-Epidemiological analysis:
Cox PH regression, Kaplan-Meier survival curves, age-period-cohort analysis, sequence symmetry analysis.
-Meta-analysis:
Meta analysis of means, proportions, correlated proportions or hazard ratios, Bayesian meta-analysis.
-Data Visualisation
Graphs and charts can be created using SPSS, SAS, Stata, Tableau, Qlikview or R Shiny.
Statistical reporting
Once all analysis has been completed, a summary interpreting the results in plain English will be provided as well as a complete statistical report, including results and conclusions, in line with conventional (APA) standards, or other preferred standards. Typically we will first send through to you a standard statistical report with results and any visuals but without the bells and whistles. Once you are happy with the contents it will then be written up to publication standards or even made into a power point presentation as per your requirements.
The finished product will typically include a statistical report outlining the results of the analyses and any tables, graphs and other visualisations. We are also happy to include any programming code used to perform the analyses and all output produced, as required. As often is the case many clients prefer to remain liberated of these finer details, therefore include these components upon your request.
We assist with many aspects of grant proposals including sample size and power calculations, clinical study designs, cost estimations for the biostatistical consulting component of the research, and in the development and analysis of pilot studies.
Complex Systems Analysis
Network Analysis
Branching networks
Bayesian meta-analysis
Meta-analysis or evidence synthesis is a necessary part of clinical trial design which serves to optimise the insights and accuracy of the present trial. This is achieved by combining evidence from previous trials. Accurately combining evidence from multiple sources requires a hierarchical framework. This can be more reliably achieved by use of Bayesian rather-than frequentist meta-analysis methods. Bayesian meta-analysis is beneficial on a stand-alone basis as well as in the planning of any clinical trial and is an integral initial planning stage of any adaptive clinical trial design.
Bayesian adaptive designs for phase 1-3 clinical trials
Adaptive clinical trial design entails the prospective planning and detailed specification of methodological adaptation to be implemented in clinical trials in cases where the need arises due to changes in trial environment and updates in knowledge such as drug effectiveness data. As well as avoiding clinical trial failure by managing uncertainty and optimising resources, thereby increasing time and cost efficiency, as well as the flexibility to end an unproductive trial early or update treatment doses without running subsequent dose-finding trials.
The initial protocol stage of a phase one trial is the most ideal stage to incorporate the adaptive design, the use of adaptive design can however be initiated at any phase of clinical trial provided the data for that phase remains blinded to all parties. Adaptive trial design can be of benefit within any individual phase of clinical trial as well as highly beneficial in optimizing the entire clinical research process between phases 1 to 3 of any clinical trial sequence.
Bayesian predictive models/ Bayesian generative models
Statistical models are a simplification of reality that aim to capture primary factors underlying a system under investigation based on observed data. The aim of model fitting is to increase understanding of specific factors making up an underlying system. Statistical models can vary from quite simplistic to increasingly complex as the situation necessitates. Model fitting involves estimating model parameters in line with observed data.
Bayesian model development consists of:
- Determining a prior distribution based on previous data and subjecting to a prior predictive checking process
- Determining the likelihood function
- Combining the likelihood function with the prior distribution to form a posterior distribution
- Summarising resulting probabilities with associated point estimates
Markov Chain Monte Carlo (MCMC) based methods:
Metropolis hastings
Reversible jump MCMC
Hamiltonial Monte Carlo
Gibbs Sampler
Particles MCMC
Evolutionary Monte Carlo
Other methods:
Sequential Monte Carlo
Approximate Bayesian Computation
Integrated nested Laplace approximations
Variational Bayes.
Survey design and development:
We ensure that a survey addresses your overarching research goals and evaluates the constructs it intends to measure by fine tuning every detail.
Development of your survey may involve the collection of qualitative responses (such as open-ended or free response style questions) or quantitative responses (such as nominal or Likert scale questions), or a degree of combining the two. Once the response style has been established we will assist in determining the right questions to include and how to ask them in order to optimise data collection in service of evaluating your constructs.
Sampling method determination:
This decision with be based on feasibility criteria in line with what best serves your research parameters such as distribution capabilities, budget, timeline and study questions. Randomised probability or stratified sampling, convenience sampling, or more targeted selective sampling can be used dependent on circumstances.
Online platform navigation:
We can help set up and implement your survey on the platform of your choice such as Qualtrics, survey monkey or Google Forms.
Pilot testing:
This involves running a pilot test of the survey and analysing results for reliability and construct validity using statistics such as test-retest reliability and Chronbach’s alpha. From here any tweaks to the survey content can be made and confirmatory factor analysis can establish the statistical power of your final survey, which enables the appropriate sample size to be calculated.
Final survey analysis:
statistical analysis of your final survey data can involve factor analysis, principle components analysis, latent variable models, predictive models or many other options depending on the data and research questions involved.
