what is the best statistical test to use to determine risk levels

Today statistics provides the footing for inference in most medical research. Yet, for want of exposure to statistical theory and practise, it continues to be regarded every bit the Achilles heel by all concerned in the loop of enquiry and publication – the researchers (authors), reviewers, editors and readers.

Well-nigh of us are familiar to some caste with descriptive statistical measures such as those of central trend and those of dispersion. However, we falter at inferential statistics. This need non be the example, especially with the widespread availability of powerful and at the aforementioned time convenient statistical software. As we have outlined below, a few fundamental considerations will pb ane to select the appropriate statistical exam for hypothesis testing. Yet, information technology is important that the appropriate statistical assay is decided earlier starting the written report, at the phase of planning itself, and the sample size chosen is optimum. These cannot be decided arbitrarily subsequently the report is over and data have already been collected.

The swell bulk of studies can be tackled through a basket of some 30 tests from over a 100 that are in use. The test to be used depends upon the blazon of the research question existence asked. The other determining factors are the type of information being analyzed and the number of groups or data sets involved in the study. The following schemes, based on five generic research questions, should assist.[1]

Question 1: Is there a difference betwixt groups that are unpaired? Groups or information sets are regarded as unpaired if there is no possibility of the values in one data set being related to or existence influenced by the values in the other information sets. Different tests are required for quantitative or numerical information and qualitative or categorical data every bit shown in Fig. ane. For numerical data, information technology is important to decide if they follow the parameters of the normal distribution curve (Gaussian curve), in which instance parametric tests are applied. If distribution of the data is not normal or if ane is not sure about the distribution, information technology is safer to use non-parametric tests. When comparing more 2 sets of numerical data, a multiple group comparison examination such as i-way analysis of variance (ANOVA) or Kruskal-Wallis exam should be used first. If they return a statistically significant p value (usually meaning p < 0.05) then but they should be followed by a post hoc test to make up one's mind between exactly which ii data sets the difference lies. Repeatedly applying the t examination or its not-parametric counterpart, the Isle of man-Whitney U test, to a multiple grouping situation increases the possibility of incorrectly rejecting the null hypothesis.

Tests to address the question: Is there a departure between groups – unpaired (parallel and independent groups) state of affairs?

Question two: Is at that place a difference betwixt groups which are paired? Pairing signifies that information sets are derived by repeated measurements (e.g. before-subsequently measurements or multiple measurements across time) on the same fix of subjects. Pairing volition also occur if subject groups are dissimilar but values in one group are in some way linked or related to values in the other grouping (e.g. twin studies, sibling studies, parent-offspring studies). A crossover written report pattern also calls for the awarding of paired group tests for comparing the effects of different interventions on the same subjects. Sometimes subjects are deliberately paired to match baseline characteristics such as age, sex, severity or duration of disease. A scheme like to Fig. 1is followed in paired data set testing, as outlined in Fig. ii. In one case again, multiple data set comparison should be done through appropriate multiple group tests followed by post hoc tests.

Tests to address the question: Is at that place a difference between groups – paired situation?

Question iii: Is there whatever association between variables? The diverse tests applicative are outlined in Fig. iii. It should be noted that the tests meant for numerical data are for testing the association betwixt ii variables. These are correlation tests and they express the force of the association every bit a correlation coefficient. An inverse correlation between 2 variables is depicted by a minus sign. All correlation coefficients vary in magnitude from 0 (no correlation at all) to one (perfect correlation). A perfect correlation may indicate just does not necessarily hateful causality. When 2 numerical variables are linearly related to each other, a linear regression analysis can generate a mathematical equation, which can predict the dependent variable based on a given value of the independent variable.[2] Odds ratios and relative risks are the staple of epidemiologic studies and express the association betwixt categorical information that can be summarized as a 2 × 2 contingency table. Logistic regression is actually a multivariate analysis method that expresses the strength of the clan betwixt a binary dependent variable and two or more independent variables as adjusted odds ratios.

Tests to accost the question: Is there an association between variables?

Question iv: Is there agreement between information sets? This tin can be a comparison between a new screening technique confronting the standard test, new diagnostic test confronting the bachelor gold standard or agreement betwixt the ratings or scores given by different observers. Equally seen from Fig. 4, agreement between numerical variables may be expressed quantitatively by the intraclass correlation coefficient or graphically by amalgam a Bland-Altman plot in which the difference between two variables x and y is plotted against the mean of x and y. In instance of categorical information, the Cohen's Kappa statistic is oft used, with kappa (which varies from 0 for no understanding at all to 1 for perfect understanding) indicating strong agreement when it is > 0.7. Information technology is inappropriate to infer agreement by showing that in that location is no statistically significant departure betwixt means or by calculating a correlation coefficient.

Tests to accost the question: Is there an agreement between assessment (screening / rating / diagnostic) techniques?

Question 5: Is there a difference betwixt time-to-event trends or survival plots? This question is specific to survival analysis[3](the endpoint for such analysis could be decease or whatsoever consequence that can occur after a menstruation of time) which is characterized by censoring of data, meaning that a sizeable proportion of the original study subjects may not reach the endpoint in question by the time the study ends. Data sets for survival trends are e'er considered to exist not-parametric. If there are ii groups then the applicable tests are Cox-Mantel test, Gehan's (generalized Wilcoxon) exam or log-rank test. In case of more than two groups Peto and Peto'due south exam or log-rank test tin be applied to expect for significant difference between time-to-event trends.

Information technology can be appreciated from the to a higher place outline that distinguishing between parametric and non-parametric data is of import. Tests of normality (due east.thousand. Kolmogorov-Smirnov test or Shapiro-Wilk goodness of fit test) may be applied rather than making assumptions. Some of the other prerequisites of parametric tests are that samples accept the same variance i.eastward. fatigued from the same population, observations within a grouping are contained and that the samples have been drawn randomly from the population.

A ane-tailed test calculates the possibility of departure from the null hypothesis in a specific management, whereas a ii-tailed exam calculates the possibility of divergence from the cipher hypothesis in either direction. When Intervention A is compared with Intervention B in a clinical trail, the zip hypothesis assumes there is no difference between the two interventions. Divergence from this hypothesis can occur in favor of either intervention in a ii-tailed test but in a one-tailed test it is presumed that only ane intervention can show superiority over the other. Although for a given data gear up, a one-tailed exam will render a smaller p value than a two-tailed examination, the latter is usually preferred unless at that place is a watertight case for one-tailed testing.

It is obvious that we cannot refer to all statistical tests in one editorial. However, the schemes outlined will embrace the hypothesis testing demands of the majority of observational equally well every bit interventional studies. Finally one must recollect that, in that location is no substitute to actually working hands-on with dummy or real data sets, and to seek the communication of a statistician, in order to acquire the nuances of statistical hypothesis testing.

References

one. Parikh MN, Hazra A, Mukherjee J, Gogtay N, editors. Research methodology simplified: Every clinician a researcher. New Delhi: Jaypee Brothers; 2010. Hypothesis testing and pick of statistical tests; pp. 121–viii. [Google Scholar]

two. Petrie A, Sabin C, editors. Medical statistics at a glance. ii nd. London: Blackwell Publishing; 2005. The theory of linear regression and performing a linear regression analysis; pp. 70–3. [Google Scholar]

iii. Wang D, Clayton T, Bakhai A. Analysis of survival information. In: Wang D, Bakhai A, editors. Clinical trials: A practical guide to design, analysis and reporting. London: Remedica; 2006. pp. 235–52. [Google Scholar]

peaseheathence.blogspot.com

Source: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3116565/

Share this post