Contact

Daria Pašalić
Editor-in-Chief
Department of Medical Chemistry, Biochemistry and Clinical Chemistry
Zagreb University School of Medicine
Šalata ul 2.
10 000 Zagreb, Croatia
Phone +385 (1) 4590 205; +385 (1) 4566 940
E-mail: dariapasalic [at] gmail [dot] com

Useful links

S8-1

Sample size and power analysis: how large sample do you really need?

 

Mary McHugh. Sample size and power analysis: how large sample do you really need? Biochemia Medica 2015;25(Suppl 1):S46.

Dean of Nursing, Angeles College, Los Angeles, California, USA

 

Determination of sample size is key to successful research. Many researchers, however, dread performing the task of calculating the sample size needed in their research. Even more do not understand exactly how sample size relates to the probability that their research will be successful. Sample size is one of the four keys to obtaining a significant result in research, the other three being effect size, homogeneity of the population, and the power of the statistic used to test the research hypothesis.  The effect size is, at the core, what the researcher is trying to find. Therefore the researcher does not (and cannot) have any influence or control over the effect size.  Homogeneity of the population is one of the essential characteristics of the population studied and thus not subject to researcher control in medical research. The power of the statistic relates to selection of parametric versus non-parametric (distribution free) statistics and that selection is dictated by the type of data and the research hypotheses. The only pillar of significance under the researcher’s control is therefore sample size, and manipulation of the sample size is the researcher’s most powerful tool to obtain statistical significance in research. A power analysis is a statistical procedure performed to determine the minimum sample size needed for the researcher to find a significant difference between experimental and control groups, if indeed the independent variable does produce a real difference in the population.  The sample must be large enough to allow statistical significance to be achieved, but samples larger than necessary simply add to the cost of research and are therefore unnecessary. Statistics used to test hypotheses all assume random sampling and random assignment to study group in order to reliably estimate the likelihood that the sample is a good reflection of the population.  When random procedures are fair, and the power analysis has been correctly performed, the researcher may have great confidence that real differences in the population will be detected by the study.

e-mail: mchugh8688 [at] gmail [dot] com

 

S8-2

Method comparison and evaluation statistics

 

Mladen Petrovečki. Method comparison and evaluation statistics. Biochemia Medica 2015;25(Suppl 1):S47.

Department of Laboratory Medicine, Dubrava University Hospital, Zagreb, Croatia

Department of Medical Informatics, Rijeka University School of Medicine, Rijeka, Croatia

 

Procedures to determine the quantity, quality, composition, or concentration of the same ingredient or substances in a clinical laboratory, or generally the same variable in scientific research, might be compared statistically using special procedures that relate the same indicator of two (or more) measurements of the same group of patients, or measurements obtained using different measurement systems (devices, instruments, tests, or questionnaires). Instead of usual correlation coefficients (Pearson, Spearman, Cramer) as measures of association, Lin coefficient of concordance (W), interclass correlation coefficient (ICC, interclass correlation coefficient), coefficient of repeatability (BSI, British Standards Institute of repeatability coefficient), and Cohen’s kappa (k) are used as a typical statistical measures of agreement. Measuring procedures with results expressed using numerical values ​​(numerical measurement scale) can be compared graphically by Bland-Altman’s diagram and Krouwer-Monti’s chart (mountain plot), or compared numerically by Passing-Bablok’s and Deming’s regression analyzes.

e-mail: mladenp [at] kbd [dot] hr

 

S8-3

The most common errors in biostatistics and ways to avoid them

 

Lidija Bilić-Zulle. The most common errors in biostatistics and ways to avoid them.Biochemia Medica 2015;25(Suppl 1):S48-S49.

Clinical Department of Laboratory Diagnostics, Clinical Hospital Centre Rijeka, Rijeka, Croatia

Department of Medical Informatics, Rijeka University School of Medicine, Rijeka, Croatia

 

Scientific paper is a communication tool in scientific and academic community. Once published it becomes subject of continued judgment of scientific community and possible errors in paper become everlasting. Sometimes errors stay even unrecognized for long time. In biomedicine researches often investigate data on biological systems and it is mandatory to process data by statistical data analysis. Statistics procedures summarized and present data and compare them i.e. reveal relationship among them. Interpretation of results of statistical analysis allows conclusions that will reject or accept previously established hypothesis. That is why errors in statistical analysis can completely devaluate whole research, cause false conclusions and thus lead to misconceptions instead to knowledge. In biomedicine such misconceptions can be threat to human health and life.

The most frequent errors that can be found in scientific papers can be classified as: errors in sampling, data entering and sorting, data presentation (measures od central tendency and dispersion), selecting statistical tests and interpreting and presenting results of statistical testing. Investigated sample needs to be representative to population. Sample size depends on research type, power of the study and should be calculated. Data entry is a crucial step in research. Errors made once cannot be easily discovered and can significantly influence final results. Data should be presented by proper measure of central tendency and dispersion, depending on data type and distribution. Each measure of central tendency has corresponding measurement of dispersion and it is erroneous to use others. It is not correct to use confidence interval or standard error as measure of dispersion. Selection of statistical test is an important step and it is mandatory to clearly identify statistical hypothesis e.g. if there is difference is investigate or for example, correlation. It is completely wrong to compare everything and then find out if there was statistically significant difference between some variables. Selection of statistical test that will support or decline hypothesis depends on research design, number of samples and their size and data distribution. For the most of statistical tests specific criterions of data have to be fulfilled in order to obtain results that will point out appropriate conclusions. It is important to remember that conclusions based on statistical analysis contain certain level of error that has to be established previous data analysis. Conclusions have to be commented in statistical and clinical manner and not all statistical differences are clinically important.

Unfortunately, errors in statistics are not rare even in published papers and they are real challenge for editors and peer-reviewers in scientific journals. Scientists need to be familiar with statistical procedures and have to include statistician in research teams from the very beginning of the research, not at the end when already made errors are difficult to resolve.

 

e-mail: lidija [dot] bilic [dot] zulle [at] medri [dot] uniri [dot] hr