tableone and its Basic Statistics

Baseline Characteristics in Biomedical Research

Slides

Make slides full screen

Research Questions

Potential outcomes could include

Changes in inflammation levels, as measured by the elisa_cytokines and flow_cytometry tables. Changes in clinical measurements such as Nugent Score, C-reactive protein blood test (CRP), and vaginal pH, as recorded in the visit_clinical_measurements table.

Potential Research Question, Objective and Statistical Approach

Efficacy of the Treatment

• Question: How effective is the drug in reducing inflammation at different time points (week 1 and week 7) compared to the baseline? Is there a significant difference between the treatment and placebo groups?
• Objective: Determine the effectiveness of the drug in reducing inflammation.
• Statistical Analysis: Use a repeated measures ANOVA or mixed-effects model to compare inflammation levels at different time points between the treatment and placebo groups.

Impact of Smoking

• Question: Does smoking status affect the efficacy of the drug? Is there a significant difference in outcomes between smokers and non-smokers?
• Objective: Investigate the influence of smoking on the efficacy of the drug.
• Statistical Analysis: Perform subgroup analysis or interaction analysis to compare outcomes between smokers and non-smokers.

Age and Drug Efficacy

• Question: Does age influence the effectiveness of the drug? Are there differences in outcomes among different age groups?
• Objective: Explore the influence of age on the effectiveness of the drug.
• Statistical Analysis: Use regression analysis to examine the relationship between age and treatment outcomes.

Education Level and Treatment Outcome

• Question: Is there a correlation between the level of education and the treatment outcome? Do individuals with higher education levels have better outcomes?
• Objective: Investigate whether education level correlates with treatment outcomes.
• Statistical Analysis: Perform a chi-square test or Fisher’s exact test to examine the association between education level and treatment outcomes.

Cytokine Concentration Analysis

• Question: How do cytokine concentrations change over time in response to the treatment? Are certain cytokines more responsive to the treatment than others?
• Objective: Analyze changes in cytokine concentrations over time in response to treatment.
• Statistical Analysis: Use a repeated measures ANOVA or mixed-effects model to compare cytokine concentrations at different time points.

Cell Count Analysis

• Question: How do cell counts (from flow cytometry data) change over time in response to the treatment? Are certain cell types more affected by the treatment than others?
• Objective: Analyze changes in cell counts over time in response to treatment.
• Statistical Analysis: Use a repeated measures ANOVA or mixed-effects model to compare cell counts at different time points.

Correlation between Clinical Measurements and Outcomes

• Question: Is there a correlation between Nugent Score, C-reactive protein blood test (CRP), vaginal pH, and treatment outcomes?
• Objective: Examine correlations between clinical measurements (Nugent Score, CRP, vaginal pH) and treatment outcomes.
• Statistical Analysis: Use correlation analysis or multiple regression analysis to examine these relationships.

Long-term Efficacy of the Drug

Question: Does the drug’s efficacy persist over time (from week 1 to week 7), or does it diminish? Objective: Assess whether the drug’s efficacy persists over time. Statistical Analysis: Use a repeated measures ANOVA or mixed-effects model to compare treatment outcomes at week 1 and week 7.

Refresher

Notes

Introduction

In biomedical research, it is often necessary to compare the effects of different interventions or exposures on the health outcomes of interest. However, before making any causal inference, it is important to evaluate the baseline characteristics of the patients who participate in the study. Baseline characteristics are the demographic and clinical features of the participants at the start of a trial or a study, such as age, sex, disease severity, comorbidities, etc. They provide information about the population that is being studied and the context of the research question (Schulz, Altman, & Moher, 2010).

Evaluating baseline characteristics of patients in different treatment groups is important for several reasons:

• It allows readers to assess the external validity of the trial results, which is the extent to which the results can be generalised to other settings and populations. For example, if the trial participants are very different from the target population in terms of age, sex, or other factors that may affect the outcome, then the results may not be applicable to the target population (Altman, 1990).
• It allows researchers to check the internal validity of the trial results, which is the extent to which the results are free from bias and confounding. Bias is a systematic error that leads to a deviation from the true effect of the intervention or exposure. Confounding is a situation where a third variable is associated with both the intervention or exposure and the outcome, and may distort the true effect of the intervention or exposure. For example, if the treatment groups are not balanced in terms of baseline characteristics that may influence the outcome, then there may be a bias or a confounding effect that needs to be adjusted for in the statistical analysis (Matthews, 2006).
• It allows researchers to explore the heterogeneity of the treatment effect, which is the variation in the effect across different subgroups of participants. Heterogeneity can be due to biological, clinical, or methodological factors that may modify or interact with the intervention or exposure. For example, if the treatment effect differs by age, sex, or disease severity, then it may be useful to perform subgroup analyses to identify which subgroups benefit more or less from the intervention or exposure (Matthews, 2006).

To evaluate baseline characteristics, researchers need to use appropriate methods of summarisation, comparison, and adjustment, depending on the type and distribution of the variables. Summarisation involves describing the distribution and characteristics of the variables using descriptive statistics such as means, standard deviations, medians, interquartile ranges, counts, and percentages. Comparison involves testing whether there are significant differences or associations between the groups or variables using statistical tests such as t-tests, chi-squared tests, Fisher’s exact tests, and rank sum tests. Adjustment involves controlling for the potential confounding factors using methods such as stratification, matching, covariate adjustment, or weighting (Altman, 1990).

Types of variables

Continuous

Continuous variables are variables that can take any numeric value within a range, such as age, weight, height, blood pressure, etc. They are usually measured or calculated using a scale or a device.

Categorical

Categorical variables are variables that can take only a limited number of values, such as sex, race, diagnosis, treatment group, etc. They are usually assigned or observed based on some criteria or classification.

Descriptive statistics

Descriptive statistics are numerical summaries that describe the distribution and characteristics of a variable. They include measures of central tendency (such as mean, median, mode), measures of variability (such as standard deviation, interquartile range, range), and measures of frequency (such as count, percentage, proportion).

Mean

The mean is the sum of all the values in a data set divided by the number of values. The formula is:

\[\bar{x} = \frac{\sum x}{n}\]

where \(\bar{x}\) is the mean, \(\sum x\) is the sum of all the values, and \(n\) is the number of values.

Median

The median is the middle value in an ordered data set. To find the median, arrange the values from smallest to largest and then locate the middle one. If there are an even number of values, the median is the average of the two middle ones. The formula is:

\[\tilde{x} = \begin{cases} x_{(n+1)/2}, & \text{if } n \text{ is odd} \\ \frac{x_{n/2} + x_{(n/2)+1}}{2}, & \text{if } n \text{ is even} \end{cases}\]

where

• \(\tilde{x}\) is the median,
• \(x_i\) is the \(i\)th value in the ordered data set, and
• \(n\) is the number of values.

Mode

The mode is the most frequent value in a data set. There can be more than one mode if there are multiple values with the same frequency. There is no formula for the mode, but it can be found by counting how many times each value or category occurs and then selecting the one(s) with the highest frequency.

Standard deviation

The variance is a measure of how much the values vary from the mean while the standard deviation is a measure of how spread out the values are from the mean. It is calculated by taking the square root of the variance. The formula is:

\[s = \sqrt{\frac{\sum (x - \bar{x})^2}{n-1}}\]

where

• \(s\) is the standard deviation,
• \(x\) is a value in the data set,
• \(\bar{x}\) is the mean, and
• \(n\) is the number of values.

Count

The count is the number of times a value or a category occurs in a data set. There is no formula for the count, but it can be found by counting how many times each value or category appears in the data set.

Percentage

The proportion is the ratio of a part to the whole expressed as a fraction between 0 and 1 whereas the percentage is the ratio of a part to the whole expressed as a fraction of 100. The formula is:

\[p = \frac{n}{N} \times 100\]

where

• \(p\) is the percentage,
• \(n\) is the count of a value or a category, and
• \(N\) is the total count of all values or categories.

These are some of the basic formulas for descriptive statistics. You can find more information and examples on this website.

Types of statistical tests

One way to expand the sentence on baseline characteristics of patients in biomedical research papers is:

Statistical tests are methods to evaluate whether there is a significant difference or association between two or more groups or variables. They are often used to compare the baseline characteristics of patients in different treatment groups, such as intervention and control groups, in biomedical research papers (Altman, 1990). They usually involve calculating a test statistic, which is a numerical value that summarizes the strength and direction of the relationship between the groups or variables. Then they compare the test statistic with a critical value or a p-value, which are thresholds that indicate the level of significance of the test. The level of significance is the probability of obtaining a test statistic as extreme or more extreme than the observed one, if the null hypothesis of no difference or no association is true. If the test statistic exceeds the critical value or is lower than the p-value, then the null hypothesis can be rejected and the alternative hypothesis of a difference or an association can be accepted. Otherwise, the null hypothesis cannot be rejected and the alternative hypothesis cannot be accepted.

Depending on the type and distribution of the variables, different statistical tests can be performed. For continuous variables, such as age, weight, height, blood pressure, etc., t-tests (for normally distributed data) or rank sum tests (for non-normally distributed data) can be performed. T-tests compare the means of two groups and assume that the data are normally distributed and have equal variances. Rank sum tests compare the medians of two groups and do not assume any distribution or variance equality (Altman, 1990). For categorical variables, such as sex, race, diagnosis, treatment group, etc., chi-squared tests (for large sample sizes) or Fisher’s exact tests (for small sample sizes) can be performed. Chi-squared tests compare the frequencies or proportions of two or more groups and assume that the expected frequencies are sufficiently large. Fisher’s exact tests compare the frequencies or proportions of two groups and do not assume any frequency requirement (Altman, 1990).

Another aspect that can be considered when comparing baseline characteristics is the standardized mean difference (SMD), It is a unitless measure that can be used to compare the magnitude of group differences across different variables. It can also be used to combine results from different studies that measure the same outcome but use different scales (Schulz, Altman, & Moher, 2010).

tableone package

Introduction

R package tableone is a package by Yoshida & Victorina (2021) that eases the construction of “Table 1: Baseline demographics and clinical characteristics”. The package can handle both continuous and categorical variables, and provide descriptive statistics such as means, standard deviations, medians, interquartile ranges, counts, and percentages. It can also perform statistical tests to compare groups, such as t-tests, chi-squared tests, Fisher’s exact tests, and rank sum tests. It can also calculate standardized mean differences to measure the effect size of group differences. tableone can handle weighted data using the survey package, which allows researchers to account for complex sampling designs and adjust for confounding factors. tableone has a simple and flexible syntax, and can produce nice-looking tables using the kableone function.

How tableone works

• To use tableone, you need to install it from CRAN or GitHub (Yoshida & Victorina, 2021).
• You need to load the package using library(tableone).
• You need to specify the variables you want to summarize using the vars argument. You can also specify which variables are categorical using the factorVars argument.
• You need to provide the data frame that contains the variables using the data argument. You can also provide a grouping variable using the strata argument.
• You need to create a tableone object using the Createtableone`` function. This object contains all the summary statistics and test results for each variable and group.
• You can print or export the tableone object using the print or kableone functions.

Other packages for creating tables

Besides tableone, there are many other R packages that can be used to create tables in various output formats, such as PDF, HTML and Word. Some of them are: flextable (Gohel & Skintzos, 2023) and huxtable (Hugh-Jones, 2022).

References

Altman, D. G. (1990). Practical statistics for medical research. Chapman & Hall/CRC.

Matthews, J. N. (2006). Introduction to randomized controlled clinical trials. CRC Press.

Schulz, K. F., Altman, D. G., & Moher, D.; CONSORT Group. (2010). CONSORT 2010 statement: updated guidelines for reporting parallel group randomised trials. BMJ, 340, c332.

Therneau, T. M., & Grambsch, P. M. (2000). Modeling survival data: Extending the Cox model. Springer.

Yoshida, K., & Victorina, L. K. (2021). tableone: An R package for creating ‘Table 1’. R package version 0.12.0.