[Appropriate application of statistical methods in physiological research].

OBJECTIVE: To offer a series of efficient methods to physiologists in appropriate selection, and application of statistical techniques. METHODS: We bring about two questions as follows:What's the role of statistics in the process of a physiological research? How to make sure the results produced in a physiological research can be repeatable in practice in the long run. From the answers to these two questions, we highlight the importance of the discipline of statistics to research work, explain why it is difficult, how to choose a suitable statistical method in a specific situation, and offer the critical methods to use statistics accurately and appropriately. RESULTS: We abstract three core sections from the discipline of statistics:how to make the design of a study impeccable; How to strictly follow the protocol of a study, and How to draw conclusions well reasoned and strongly supported by evidence. By elaborating these sections, it is feasible to correctly use statistical methods for data analysis and results interpretation. CONCLUSIONS: In physiological research, conclusion can stand with time and repeatable in practice only when researchers strictly and rigorously follow the rule of scientific research.

[How to scientifically estimate sample size in physiological research].

OBJECTIVE: To bring about physiological researchers' attention of the importance of sample size estimation. METHODS: The significance as well as the current problems of sample size estimation were illustrated and the commonly-used sample size estimation methods were introduced. RESULTS: The basic concepts and necessary premises of sample size estimation were stated. The estimation processes and results under two different circumstances were elaborated in detail via examples. CONCLUSIONS: To attain the proper estimated sample sizes, the computation must satisfy the necessary premises which included the appropriate statistical analysis methods to be used.

Clinical progression and predictors of death in patients with severe fever with thrombocytopenia syndrome in China.

BACKGROUND: Severe fever with thrombocytopenia syndrome (SFTS) is an emerging infectious disease of which the clinical progression and factors related to death are still unclear. OBJECTIVE: To identify the clinical progression of SFTS and explore predictors of fatal outcome throughout the disease progress. STUDY DESIGN: A prospective study was performed in a general hospital located in Xinyang city during 2011-2013. Confirmed SFTS patients were recruited and laboratory parameters that were commonly evaluated in clinical practice were collected. The clinical progression was determined based on analysis of dynamic profiles and Friedman's test. At each clinical stage, the laboratory features that could be used to predict fatal outcome of SFTS patients were identified by stepwise discriminant analysis. RESULTS: Totally 257 survivors and 54 deceased SFTS patients were recruited and the data of 11 clinical and laboratory parameters along their entire disease course were consecutively collected. Three clinical stages (day 1-5 post onset, day 6-11 post onset and day 12 to hospital discharge) were determined based on distinct clinical parameters evaluations. Multivariate discriminant analysis at each clinical stage disclosed the indicators of the fatal outcome as decreased platelet counts at early stage, older age and increased AST level at middle stage, and decreased lymphocyte percentage and increased LDH level at late stage. CONCLUSIONS: The significant indicators at three clinical stages could be used to assist identifying the patients with high risk of death. This knowledge might help to perform supportive treatment and avoid fatality.

Multifactor designs able to examine the interactions.

Multifactor designs that are able to examine the interactions include factorial design, factorial design with a block factor, repeated measurement design; orthogonal design, split-block design, etc. Among all the above design types that are able to examine the interactions, the factorial design is the most commonly used. It is also called the full-factor experimental design, which means that the levels of all the experimental factors involved in the research are completely combined, and k independent repeated experiments are conducted under each experimental condition. The factorial design with a block factor can also examine the influence of a block factor formed by one or more important non experimental factors based on the factorial design. This article introduces the factorial design and the factorial design with a block factor by examples.

Three-factor designs unable to examine the interactions (part 2).

Three-factor designs that are unable to examine the interactions include crossover design and Latin square design, which can examine three factors: an experimental factor and two block factors. Although the two design types are not quite frequently used in practical research, an unexpected research effect will be achieved if they are correctly adopted on appropriate occasions. This article introduced the 3×3 crossover design and the Latin square design by examples.

Three-factor designs unable to examine the interactions (part 1).

Three-factor designs that are unable to examine the interactions include crossover design and Latin square design, which can examine three factors, namely, an experimental factor and two block factors. Although the two design types are not quite frequently used in practical research, an unexpected research effect will be achieved if they are correctly adopted on appropriate occasions. Due to the limit of space, this article introduces two forms of crossover design.

Two-factor designs unable to examine the interactions (Part 2).

Two-factor designs are very commonly used in scientific research. If the two factors have interactions, research designs like the factorial design and the orthogonal design can be adopted; however, these designs usually require many experiments. If the two factors have no interaction or the interaction is not statistically significant on result in theory and in specialty, and the measuring error of experimental data under a certain condition (usually one of the experimental conditions that are formed by the complete combination of the levels of the two factors) is allowed in specialty, researchers can use random block design without repeated experiments, balanced incomplete random block design without repeated experiments, single factor design with a repeatedly measured factor, two-factor design without repeated experiments and two-factor nested design. This article introduces the last two design types by examples.

Two-factor designs unable to examine the interactions (part 1).

Two-factor designs are quite commonly used in scientific research. If the two factors have interactions, research designs like the factorial design and the orthogonal design can be adopted; however, these designs usually require many experiments. If the two factors have no interaction or the interaction is not statistically significant on result in theory and in specialty, and the measuring error of the experimental data under a certain condition (usually it is one of the experimental conditions which is formed by the complete combination of the levels of two factors) is allowed in specialty, researchers can use random block design without repeated experiments, balanced non-complete random block design without repeated experiments, single factor design with a repeatedly measured factor, two-factor design without repeated experiments and two-factor nested design. This article introduced the first three design types with examples.

How to choose an appropriate experimental design type (Part 2).

How to choose an appropriate design type to arrange research factors and their levels is an important issue in scientific research. Choosing an appropriate design type is directly related to the accuracy, scientificness and credibility of a research result. When facing a practical issue, how can researchers choose the most appropriate experimental design type to arrange an experiment based on the research objective and the practical situation? This article mainly introduces the related contents of the design of one factor with two levels and the design of one factor with k (k≥3) levels by analyzing some examples.

How to choose an appropriate experimental design type (Part 1).

How to choose an appropriate experimental design type to arrange research factors and their levels is an important issue in experimental research. Choosing an appropriate design type is directly related to the accuracy and reliability of the research result. When confronting a practical issue, how can researchers choose the most appropriate design type to arrange the experiment based on research objective and specified situation? This article mainly introduces the related contents of the single-group design and the paired design through practical examples.

The balance principle in scientific research.

The principles of balance, randomization, control and repetition, which are closely related, constitute the four principles of scientific research. The balance principle is the kernel of the four principles which runs through the other three. However, in scientific research, the balance principle is always overlooked. If the balance principle is not well performed, the research conclusion is easy to be denied, which may lead to the failure of the whole research. Therefore, it is essential to have a good command of the balance principle in scientific research. This article stresses the definition and function of the balance principle, the strategies and detailed measures to improve balance in scientific research, and the analysis of the common mistakes involving the use of the balance principle in scientific research.

Estimation of sample size and testing power (part 7).

Two-factor factorial design refers to the research involving two experimental factors and the number of the experimental groups equals to the product of the levels of the two experimental factors. In other words, it is the complete combination of the levels of the two experimental factors. The research subjects are randomly divided into the experimental groups. The two experimental factors are performed on the subjects at the same time, meaning that there is no order. The two experimental factors are equal during statistical analysis, that is to say, there is no primary or secondary distinction, nor nested relation. This article introduces estimation of sample size and testing power of quantitative data with two-factor factorial design.

Performance comparison between Logistic regression, decision trees, and multilayer perceptron in predicting peripheral neuropathy in type 2 diabetes mellitus.

BACKGROUND: Various methods can be applied to build predictive models for the clinical data with binary outcome variable. This research aims to explore the process of constructing common predictive models, Logistic regression (LR), decision tree (DT) and multilayer perceptron (MLP), as well as focus on specific details when applying the methods mentioned above: what preconditions should be satisfied, how to set parameters of the model, how to screen variables and build accuracy models quickly and efficiently, and how to assess the generalization ability (that is, prediction performance) reliably by Monte Carlo method in the case of small sample size. METHODS: All the 274 patients (include 137 type 2 diabetes mellitus with diabetic peripheral neuropathy and 137 type 2 diabetes mellitus without diabetic peripheral neuropathy) from the Metabolic Disease Hospital in Tianjin participated in the study. There were 30 variables such as sex, age, glycosylated hemoglobin, etc. On account of small sample size, the classification and regression tree (CART) with the chi-squared automatic interaction detector tree (CHAID) were combined by means of the 100 times 5-7 fold stratified cross-validation to build DT. The MLP was constructed by Schwarz Bayes Criterion to choose the number of hidden layers and hidden layer units, alone with levenberg-marquardt (L-M) optimization algorithm, weight decay and preliminary training method. Subsequently, LR was applied by the best subset method with the Akaike Information Criterion (AIC) to make the best used of information and avoid overfitting. Eventually, a 10 to 100 times 3-10 fold stratified cross-validation method was used to compare the generalization ability of DT, MLP and LR in view of the areas under the receiver operating characteristic (ROC) curves (AUC). RESULTS: The AUC of DT, MLP and LR were 0.8863, 0.8536 and 0.8802, respectively. As the larger the AUC of a specific prediction model is, the higher diagnostic ability presents, MLP performed optimally, and then followed by LR and DT in terms of 10-100 times 2-10 fold stratified cross-validation in our study. Neural network model is a preferred option for the data. However, the best subset of multiple LR would be a better choice in view of efficiency and accuracy. CONCLUSION: When dealing with data from small size sample, multiple independent variables and a dichotomous outcome variable, more strategies and statistical techniques (such as AIC criteria, L-M optimization algorithm, the best subset, etc.) should be considered to build a forecast model and some available methods (such as cross-validation, AUC, etc.) could be used for evaluation.

Estimation of sample size and testing power (part 6).

The design of one factor with k levels (k ≥ 3) refers to the research that only involves one experimental factor with k levels (k ≥ 3), and there is no arrangement for other important non-experimental factors. This paper introduces the estimation of sample size and testing power for quantitative data and qualitative data having a binary response variable with the design of one factor with k levels (k ≥ 3).

Estimation of sample size and testing power (part 5).

Estimation of sample size and testing power is an important component of research design. This article introduced methods for sample size and testing power estimation of difference test for quantitative and qualitative data with the single-group design, the paired design or the crossover design. To be specific, this article introduced formulas for sample size and testing power estimation of difference test for quantitative and qualitative data with the above three designs, the realization based on the formulas and the POWER procedure of SAS software and elaborated it with examples, which will benefit researchers for implementing the repetition principle.

Estimation of sample size and testing power (Part 4).

Sample size estimation is necessary for any experimental or survey research. An appropriate estimation of sample size based on known information and statistical knowledge is of great significance. This article introduces methods of sample size estimation of difference test for data with the design of one factor with two levels, including sample size estimation formulas and realization based on the formulas and the POWER procedure of SAS software for quantitative data and qualitative data with the design of one factor with two levels. In addition, this article presents examples for analysis, which will play a leading role for researchers to implement the repetition principle during the research design phase.

Estimation of sample size and testing power (Part 3).

This article introduces the definition and sample size estimation of three special tests (namely, non-inferiority test, equivalence test and superiority test) for qualitative data with the design of one factor with two levels having a binary response variable. Non-inferiority test refers to the research design of which the objective is to verify that the efficacy of the experimental drug is not clinically inferior to that of the positive control drug. Equivalence test refers to the research design of which the objective is to verify that the experimental drug and the control drug have clinically equivalent efficacy. Superiority test refers to the research design of which the objective is to verify that the efficacy of the experimental drug is clinically superior to that of the control drug. By specific examples, this article introduces formulas of sample size estimation for the three special tests, and their SAS realization in detail.

Estimation of sample size and testing power (part 2).

This article introduces definitions of three special tests, namely, non-inferiority test (to verify that the efficacy of the experimental drug is clinically not inferior to that of the positive control drug), equivalence test (to verify that the efficacy of the experimental drug is equivalent to that of the control drug) and superiority test (to verify that the efficacy of the experimental drug is superior to that of the control drug), and methods of sample size estimation under the three different conditions. By specific examples, the article introduces formulas of sample size estimation for the three special tests, and their SAS realization in detail.

Estimation of sample size and testing power (Part 1).

This article introduces the general concepts and methods of sample size estimation and testing power analysis. It focuses on parametric methods of sample size estimation, including sample size estimation of estimating the population mean and the population probability. It also provides estimation formulas and introduces how to realize sample size estimation manually and by SAS software.

The repetition principle in scientific research.

The repetition principle is important in scientific research, because the observational indexes are random variables, which require a certain amount of samples to reveal their changing regularity. The repetition principle stabilizes the mean and the standard variation, so that statistics of the sample can well represent the parameters of the population. Thus, the statistical inference will be reliable. This article discussed the repetition principle from the perspective of common sense and specialty with examples.