Skip to main content


Georg Heinze
Georg HeinzeHead of Institute of Clinical Biometrics

Center for Medical Statistics, Informatics and Intelligent Systems (Institute of Clinical Biometrics)
Position: Associate Professor

ORCID: 0000-0003-1147-8491
T +43 1 40400 66890

Further Information


Confounding Factors (Epidemiology); Models, Statistical; Prognosis

Research group(s)

Research interests

My work is devoted to prognosis research, in particular to all aspects of developing, translating and applying statistical methodology in prognosis research. Prognosis research in medicine aims at understanding and improving future outcomes of patients (Hemingway BMJ 2013). I develop, translate and apply biostatistical methods for prognosis research with observational data.

I have developed methods for statistical modeling with rare outcome events and sparse data (e.g. refs. 2, 5). 
Our review of variable selection (ref. 1) has been classified by Clarivate Analytics Web of Science as ‘Hot Paper’, placed in the top 0.1% of papers by citation intensity in the academic field of Mathematics.  

The STRATOS initiative (, in which I serve as member of the steering group and chair of a topic group, aims at developing guidance documents that translate statistical knowledge  to the research community at large. Moreover, I have published a series of educational papers for the medical research community.

I also apply state-of-the-art methodology in clinical epidemiology and health services research and I have published several observational studies as first author (e.g. refs. 3, 4). I contributed to more than 200 collaborative papers.

Techniques, methods & infrastructure

In prognosis research, we focus on penalized likelihood techniques for prediction and effect estimation. In several research projects, we have in particular investigated the Firth penalty (Jeffreys prior) as a solution to the problem of non-existence of regression coefficients estimated by maximum likelihood in various risk models. Other research has focused on algorithmic variable selection methods, on analyses of high-dimensional (omics) data or on the optimization of prediction models.

In evaluating new methodology, we use simulation studies, by which we can learn how methods perform under various conditions. In our simulation studies, we always try to define scenarios that are likely to be encountered in real-life data analysis. While in single-data set analyses the underlying population is usually unknown, simulation studies have the advantage that the population properties are defined by the experimentator, enabling the generalizability of results.


Selected publications

  1. Wynants L, Van Calster B, Collins G S, Riley R D, Heinze G, Schuit E et al, 2020. Prediction models for diagnosis and prognosis of covid-19: systematic review and critical appraisal. BMJ 369 :m1328. Available at:
  2. Heinze G, Wallisch C, Dunkler D, 2018. Variable selection - A review and recommendations for the practicing statistician. Biometrical Journal 60(3):431-449. Available at:
  3. Mansournia MA, Geroldinger A, Greenland S, Heinze G, 2018. Separation in Logistic Regression - Causes, Consequences, and Control. American Journal of Epidemiology 187(4):864-870. Available at:
  4. Eichinger S, Heinze G, Jandeck LM, Kyrle PA, 2010. Risk Assessment of Recurrence in Patients With Unprovoked Deep Vein Thrombosis or Pulmonary Embolism: The Vienna Prediction Model. Circulation, 121(14):1630-1636. Available at:
  5. Heinze G, Schemper M, 2002. A solution to the problem of separation in logistic regression. Statistics in Medicine, 21(16):2409-2419. Available at: