Lege Artis Medicinae

[Bayesian statistical methods in medicine]

MOLNÁR D. László1

FEBRUARY 05, 2016

Lege Artis Medicinae - 2016;26(01-02)

[Frequentist statistical methods are wide-spread and their use is easy due to the many statistical softwares and books available for the researchers. Despite the growing interest in Bayesian statistical methods teaching statistics is still mainly based on the frequentist theory. Frequentist statistics is based on the definition of probability as a relative frequency, while the Bayesian probability is a quantity that represents a state of knowledge or belief. Bayesian statistics is, in essence, a learning process. In certain situations the bayesian methods are the only possibility for data analysis. Bayesian statistical methods can be applied well in health data analysis. Every re- searcher is, to a certain extent, a Bayesian. The article discusses the application of Bayesian statistical methods in medicine. ]


  1. Szociomed Kft.



Further articles in this publication

Lege Artis Medicinae

[Making the Female Face Ugly, or How Modern Painting was Born? ]


Lege Artis Medicinae

[Results of comprehensive health screening of Hungary (MÁESZ) in 2015]

KISS István, BARNA István, DAIKI Tenno, DANKOVICS Gergely

[In 2010, a novel, complex health screening was launched in Hungary, titled as Comprehensive Health Screening of Hungary. As part of the National Heart & Vascular Program, it operates with the support of 67 professional-scientific societies and companies. In the last six years the special screening truck has visited 1123 places and traveled more than 10 000 km. The screening station has been offering 27 medical screening options. It has accomplished the complex screening of 135 879 residents and informed 314 184 residents about the right lifestyle, the protection of health and prevention of diseases on the ground. 241 753 people have received a Prevention Information Package. In this article, the authors summarize the results of 2015, summary is based on the blood tests, instrumental measurements and on the questionnaires, which were consisting 100 questions. It was again found that the Hungarian population is at high medical risk from multiple perspectives. The authors also outline possible solutions to risk reduction and disease prevention.]

Lege Artis Medicinae

[A Magician in Music: Bernstein]


Lege Artis Medicinae

[When shall we choose fulvestrant treatment? Case report and some considerations]


Lege Artis Medicinae

[The Metamorphoses of Turandot – Thoughts About Puccini’s Opera ]


All articles in the issue

Related contents

Lege Artis Medicinae

[Factors influencing anisocytosis among inpatients with chronic kidney disease]

MOLNÁR D. László, KISS István, SZAKONY Szilvia, AMBRUS Csaba

[The coefficient of variation of RDW (RDW-CV) is a predictor of mortality in several patient cohorts. RDW and other factors were analyzed in an earlier report as potential predictors of inpatient mortality. In this paper, determinants of RDW were examined in a sample of hospitalized patients with chronic kidney disease not yet on dialysis using both frequentist and bayesian ANCOVA models. For the non-informative Bayesian model no prior knowledge about the model parameters was assumed. For the informative Bayesian model prior knowledge from previous experience was applied. Calculations were performed in R with the faraway, car and MCMCPack programme packages. Male gender identity, higher mch and mcv were strongly associated with higher RDW. Blood glucose concentration, white blood cell count, blood hemoglobin concentration, platelet count, age and glomerular fil-tration rate showed inverse relationship with RDW levels. ]

Lege Artis Medicinae

[The medical application of Bayesian T-tests and traditional parametric T-tests]

MOLNÁR D. László

[Bayesian statistical learning is a mixture of the hypothetico-deductive (“top down”) and inductive (“bottom up”) perspective of data analysis with a combination of prior distributions and the data using the Bayes theorem. The resulting posterior distributions are, in a figurative sense, similar to the middle part of a sand-clock. By analogy the upper part of this clock represents the hypothetico-deductive part in terms of prior distributions, while the lower part of that represents the inductive part in terms of the data. The prior distributions will then be continuously replaced with the posterior distributions in a learning cycle. The “top-down” and the “bottom up” methods are simultaneously used both in medicine and management sciences. On the other hand, traditional frequentist statistics can primarily be considered as an inductive method, which is mainly based on the data collected. Comparison of a scale variable measured in two groups is shown with the traditional parametric t-tests, nonparametric permutation test and various Bayesian t-tests. The corresponding R and WinBUGS codes are presented. ]

Lege Artis Medicinae

[Bayesian and frequentist T-test and ANOVA]

MOLNÁR D. László

[T-test and ANOVA are just special cases of linear models. With these models measured values of a continuous variable (e.g. blood pressure) can be compared in two or more groups (e.g. treatment groups). In real life medical application of the frequentist linear models the data often fail the assumptions (e.g. normality and equal variances) originally needed to derive the mathematical equations and distributions for these models. Bayesian linear models are robust, flexible and transparent analogues and alternatives of the frequentist models. Bayesian statistical models can also be useful, when the data do comply with the prescribed conditions for the t-test or ANOVA. An advantage of the bayesian method is the possible incorporation of any prior knowledge into the models. It must be borne in mind, however, that there are many frequentist methods that can be used even if the distributions are not normal, or in the case of heteroscedasciticy or other model inadequacy as well. It is worth mentioning that even the frequentist Lasso (least absolute shrinkage and selection operator) is in near relationship with the Bayesian method, since it is a linear regression method, where the traditional fitting algorithm is modified with a constraint, and some information will be discarded in order to focus on the most relevant coefficients and make the model interpretation easy. Cal­culations were performed in R, primarily with the BayesFactor package developed by Morey and Rouder in 2015. ]

Clinical Oncology

[How to evaluate clinical trials?]


[Data from a clinical study should be always viewed with great care, and is recommended to examine and evaluate the data in-depth. First of all should be determined the evidence level of the trial’s results and what recommendations can be made based on them. The study’s type and the study design, the number of participating patients, statistical power, randomisation, stratifi cation, the balance between study arms, the endpoints and the adequacy of follow-up time should be examined. In addition the importance of the observed statistical signifi cance, the clinical relevance of the data and the real magnitude of the detected differences, and the time-dependent variables are worth to consider because these factors can determine the impact of the given clinical trial on the every-day clinical practice. In the era of the “creative study plans” a critical attitude is necessary to be convinced that the study does not contain any typical bias that could infl uence the real value and actual results of the trial.]