Biostatistics and publication of clinical research studies
Research & Innovation > Research management > Biomedical Data Science > Biostatistics and publication of clinical research studies

Team Biomedical Data Science and publication of clinical research studies


PD MMag. Dr. Wolfgang Hitzl
Head of Team Biomedical Data Science and Publication of Medical Studies; Evaluation of medical studies/Machine learning

Paracelsus Medical Private University Salzburg
Strubergasse 16
A-5020 Salzburg, Austria

Phone: +43 699 14420032

Biostatistical Collaboration and Support

The biostatistical collaboration and support includes the following steps: 

First meeting/discussion

The purpose of the meeting is first to clarify the next steps in your study. They can be – depending on the progress of the study – of different nature: If the study is in the planning phase, subjects such as study design, ethics commission application, clean data entry by means of a database, planning of the required number of cases or study design (sample size computation, open, blind, randomised, controlled, mono- or multi-centric, pilot study, sequential) are discussed. If data have already been collected, all important aims of your study are discussed. At the first meeting, all existing project documents (e.g. project description, list of the subjects to be researched, data collection instruments) should be brought along.

Conceptualisation and study design

In the case of studies in the planning phase, depending on the needs, the statistical part of the ethics commission application is filled out, the study design is determined or the calculation of the number of cases is completed. The list of the subjects/hypotheses to be researched is now refined and fixed. Depending on the data situation (e.g. scale level, distribution type), it is researched which statistical hypothesis tests/methods/models can be used in order to respond optimally to your set of problems.

Evaluation – presentation – publication of your study

We use and apply modern modern mathematical-statistical methods. 

Why is this important for you and your publication?

The reason for this is due to the fact that modern statistical methods have more power to detect signficant differences, changes, relations, effect, etc. - better than any t-test or Chi-Square test ever can. These methods are venerable, but modern and more effective methods are available.

Happily, modern mathematical statistics went and goes through a rapid evolution and so modern methods  are available. You and your publication/study definitely benefit of it.
For example, evalute your data by using a simple t-test and compare the results with modern methods. Suddenly, the results are significant!

Everyone who has ever conducted a study knows: an enourmous amount of  preliminary work is needed, time and efforts are necessary (application forms, discussion, organization, time, money, planning, examination of patients, etc) to sample data which are need to answer the research question properly. So, it is only logical to apply modern statistical methods.


The results from the evaluation are sent to you in writing as a report by e-mail.

Final aim is the publication of your work in leading internation medical journals.

Statistical Methods

Here is an overview which modern statistical methods which are applied in daily practise.

1.  "Resampling/Monte Carlo/bootstrap /permutation methods"

The main advantage of these methods is that they provide higher power to detect a signficiant effect,  so it is more likely to detect the true treatment effect or relation. For example, if blood loss is compared between two types of surgery, 5-year incidence rates for breast cancer, primary patency, superiority of theraphy A over B, etc.). These methods have higher power - no question - as compared to traditional parametric or non-parametric tests.

2. The "Adaptive Group-Sequential Designs"

These methods are likely to reduce the expected random sample size, which is reasonable and necessary not only on time-related or financial grounds, but often from ethical aspects viewpoint.

We use especially so called "two stage designs", "internal pilot study designs" and "Bauer-Köhne designs", etc. These means data analysis is conducted before data collection has been completed. For example, if a clincial trial is particularly beneficial compared to concurrent placebo group, the investigators can early stop the trial and make a decision based on a much lower sample size as originally intended.

3. "Machine learning/artifical intelligence (AI)"

No question: machine learning/AI is rising. Definitely ... and this is a good idea.

Artificial neural networks: 

  • Feedforward networks
  • Convolutional neural networks (CNN)
  • Recurrent neural networks (RNN)
  • Gated recurrent units, long short-term memory networks 
  • Support vector machines
  • Bayes classifier
  • Classification tree analyses

and many more... These methods are usually applied to so-called "classification problems", "sequence to class", "sequence to regression", sequence to sequence predicition.

Here are some examples how they are applied in medicine:

3.1 Ophthalmology: Convolutional neural networks (CNN) are used to detect primary open glaucoma - based on images.
3.2 Internal medicine: prognosis of pulmonary embolism, ventricular hypertrophy and myocardial infarction.
3.3 Pneumology: early detection of lung cancer.
3.4 Urologie: early detection of prostate.
3.5 Dermatology:  detection of melanom.
3.6. Bioinformatics/molecular biology (CNN and RNN): 

  • 1.    Convolutional LSTM Networks for Subcellular Localization of Proteins
  • 2.    Protein Secondary Structure Prediction with Long Short Term Memory
  • 3.    Prediction of DNA-binding residue from Protein sequence by Combining Long Short-Term Memory and Ensemble Learning.
  • 4.    A sequence-based, deep learning model accurately predicts RNA splicing branchpointsDNA-level splice junction prediction using deep recurrent networks

Please note: We are especially interested in studies/scientific research projects/etc which use neuronal networks of any type, especially in so-called "recurrent neural networks (RNN, e.g using long short-term memory units (LSTM), gated recurrent layer (GRU))


4. The "Generalized Estimation Equation Models (GEE)", "Mixed models" or "Generalized linear/nonlinear Models (GL/NM)"

these models have some very important advantages with respect to the traditional methods such as t-test, ANOVA, etc. This is because in the practice there appear often not normally distributed characteristics, which then follow other distributions: e.g. inverse normal, gamma, Poisson, binomial, Tweedie distributions and others). This information is often used by these models and that is why these methods have important advantages compared to the non-parametric methods.

5. Markov chains and homogenous/inhomogenous Markov processes with discrete state space 

Kaplan-Meier methods are often criticised – justifiably – by reviewers. To this purpose, "Markov chains" and "in/homogeneous Markov processes with discrete state space and continuous time" models are available, in particular the "illness-death models" and 'competing risk models'.

Here is an example of a "Dynamic treatment regime" in case of HIV in which Markov process models are successfully applied.

Statistical Software

The Complete evaluation of your study data is performed by using modern statistical methods and software:

  • STATISTICA 13. Hill, T. & Lewicki, P. (2017). STATISTICS: Methods and Applications. StatSoft, Tulsa, OK.
  • MATHEMATICA 12: Wolfram Research, Inc., Mathematica, Version 12.0, Champaign, IL (2020).
  • StatXact 10 (Cytel Software. For Windows User Manual, Cambridge MA, USA).
  • PASS 13 Hintze, J. (2019). NCSS 8. NCSS, LLC. Kaysville, Utah, USA.
  • NCSS 10: Hintze, J. (2019). NCSS 8. NCSS, LLC. Kaysville, Utah, USA.
  • SPSS 27. IBM Corp. Released 2019. IBM SPSS Statistics for Windows, Version 26.0. Armonk, NY: IBM Corp.

Standard Operating Procedure

The "SOP for the cooperation within the framework of the medical studies" regulates the cooperation between the biostatistic service of the Research Support Office of Paracelsus University and the enquirer. The goal is the standardisation of the approach and the determination of a quality standard for the cooperation. The SOP consists of two parts: the main part and an appendix.

With regard to the cooperation, we ask you to read the SOP and fill out the appendix "Documentation of the statistical consulting" (pages 3 and 4), to initialise or sign them and send them in printed form by mail to:

Paracelsus Medical Private University
Research Office (Biostatistics)
Strubergasse 21
5020 Salzburg

Checklists, Flowcharts and Links

Useful check-lists, flowcharts and interesting links: