compareGroups 4.0: Descriptives by groups'

library(knitr)
knitr::opts_chunk$set(comment="", message=FALSE, warning=FALSE,fig.align="center",fig.height=10,fig.width=10,tidy=TRUE,tidy.opts=list(blank=FALSE, width.cutoff=1200))
options(width=150)
library(kableExtra)

News

Introduction

The compareGroups package [@Subirana2014] allows users to create tables displaying results of univariate analyses, stratified or not by categorical variable groupings.

Tables can easily be exported to CSV, LaTeX, HTML, PDF, Word or Excel, or inserted in R-markdown files to generate reports automatically.

This package can be used from the R prompt or from a user-friendly graphical user interface for non-R familiarized users.

The compareGroups package is available on CRAN repository. To load the package using the R prompt, enter:

library(compareGroups)

This document provides an overview of the usage of the compareGroups package with a real examples, both using the R syntax and the graphical user interface. It is structure as follows:

Package structure: classes and methods {#package}

The compareGroups package has three functions:

Figure 1 shows how the package is structured in terms of functions, classes and methods.

Figure 1. Diagram of package structure.

Since version 4.0, a new function called descrTable has been implemented which is a shortcut of compareGroupsand createTable, i.e. step 1 and step 2 in a single step (see section 4.2.5).

Data used as example {#data}

To illustrate how this package works we sampled 85% data from the participants in the PREDIMED study [@PREDIMED]. PREDIMED is a multicenter trial in Spain, were randomly assigned participants who were at high cardiovascular risk, but with no cardiovascular disease at enrolment, to one of three diets: a Mediterranean diet supplemented with extra-virgin olive oil (MedDiet+VOO), a Mediterranean diet supplemented with mixed nuts (MedDiet+Nuts), or a control diet (advice to reduce dietary fat). Participants received quarterly individual and group educational sessions and, depending on group assignment, free provision of extra-virgin olive oil, mixed nuts, or small non-food gifts. The primary end point was the rate of major cardiovascular events (myocardial infarction, stroke, or death from cardiovascular causes.

First of all, load PREDIMED data typing:

data(predimed)

Variables and labels in this data frame are:

dicc <- data.frame(
"Name"=I(names(predimed)),
"Label"=I(unlist(lapply(predimed, attr, which="label", exact=TRUE))),
"Codes"=I(unlist(lapply(predimed, function(x) paste(levels(x),collapse="; "))))
)
dicc$Codes <- sub(">=","$\\\\geq$",dicc$Codes)
kable(dicc, align=rep("l",4), row.names=FALSE, format = "html")

OBSERVATIONS:

  1. It is important to note that compareGroups is not aimed to perform quality control of the data. Other useful packages such as 2lh [@r2lh] are available for this purpose.

  2. It is strongly recommended that the data.frame contain only the variables to be analyzed; the ones not needed in the present analysis should be removed from the list.

  3. The nature of variables to be analyzed should be known, or at least which variables are to be used as categorical. It is important to code categorical variables as factors and the order of their levels is meaningful in this package.

  4. To label the variables set the "label" attributes from each of them. The tables of results will contain the variable labels (by default).

Time-to-event variables

A variable of class Surv must be created to deal with time-to-event variables (i.e., time to Cardiovascular event/censored in our example):

predimed$tmain <- with(predimed, Surv(toevent, event == 'Yes'))
attr(predimed$tmain,"label") <- "AMI, stroke, or CV Death"

Note that variables tmain and tcv are created as time-to-death and time-to-cardiovascular event, respectively, both taking into account censoring (i.e. they are of class Surv).

Using syntax {#syntax}

Computing descriptives

compareGroups is the main function which does all the calculus. It is needed to store results in an object. Later, applying the function createTable (Section 4.2) to this object will create tables of the analysis results.

For example, to perform a univariate analysis with the predimed data between group ("response" variable) and all other variables ("explanatory" variables), this formula is required:

compareGroups(group ~ . , data=predimed)

Selecting response variables

If only a dot occurs on the right side of the ~ all variables in the data frame will be used.

To remove the variable toevent and event from the analysis:

compareGroups(group ~ . -toevent - event, data=predimed)

To select some explanatory variables (e.g., age, sex and waist) and store results in an object of class compareGroups:

res<-compareGroups(group ~ age + sex + smoke + waist + hormo, data=predimed)
res

Note: Although we have full data (n= r nrow(predimed)) for Age, Sex and Waist circumference, there are some missing data in Hormone-replacement therapy (probably male participants).

Diet groups have some differences in Smoking and Hormone-replacement therapy although those don't reach statistical significance (p-value=0.714 and 0.859, repectively); although Age, Sex and Waist circumference are clearly different.

Age & Waist circumference has been used as continuous and normal distributed. Sex, Smoking & Hormone-replacement therapy as categorical.

No filters have been used (e.g., selecting only treated patients); therefore, the selection column lists "ALL" (for all variables).

Subsetting

To perform the analysis in a subset of participants (e.g., "female" participants):

compareGroups(group ~ age + smoke + waist + hormo, data=predimed, subset = sex=='Female')

Note that only results for female participants are shown.

To subset specific variable/s (e.g., hormo and waist):

compareGroups(group ~ age + sex + smoke + waist + hormo, data=predimed, selec = list(hormo= sex=="Female", waist = waist>20 ))

Combinations are also allowed, e.g.:

compareGroups(group ~ age + smoke + waist + hormo, data=predimed, selec = list(waist= !is.na(hormo)), subset = sex=="Female")

A variable can appear twice in the formula, e.g.:

compareGroups(group ~ age + sex + bmi + bmi + waist + hormo, data=predimed, selec = list(bmi.1=!is.na(hormo)))

In this case results for bmi will be reported for all participants (n= 6324) and also for only those with no missing in Hormone-replacement therapy (!is.na(hormo)). Note that "bmi.1" in the selec argument refers to the second time that bmi appears in the formula.

Methods for continuous variables

By default continuous variables are analyzed as normal-distributed. When a table is built (see createTable function, Section 4.2), continuous variables will be described with mean and standard deviation. To change default options, e.g., "waist" used as non-normal distributed:

compareGroups(group ~ age + smoke + waist + hormo, data=predimed, method = c(waist=2))

Note that "continuous non-normal" is shown in the method column for the variable Hormone-replacement therapy.

Possible values in methods statement are:

If the method argument is stated as NA for a variable, then a Shapiro-Wilk test for normality is used to decide if the variable is normal or non-normal distributed. To change the significance threshold:

compareGroups(group ~ age + smoke + waist + hormo, data=predimed, method = c(waist=NA), alpha= 0.01)

According to Shapiro-Wilk test, stating the cutpoint at 0.01 level, "Hormone-replacement therapy" departed significantly from the normal distribution and therefore the method for this variable will be "continuous non-normal".

All non factor variables are considered as continuous. Exception is made (by default) for those that have fewer than 5 different values. This threshold can be changed in the min.dis statement:

predimed$age7gr<-as.integer(cut(predimed$age, breaks=c(-Inf,55,60,65,70,75,80,Inf), right=TRUE))
compareGroups(group ~ age7gr, data=predimed, method = c(age7gr=NA))
compareGroups(group ~ age7gr, data=predimed, method = c(age7gr=NA), min.dis=8)

To avoid errors the maximum categories for the response variable is set at 5 in this example (default value). If this variable has more than 5 different values, the function compareGroups returns an error message. For example:

compareGroups(age7gr ~ sex + bmi + waist , data=predimed)
wzxhzdk:16

Defaults setting can be changed with the max.ylev statement:

compareGroups(age7gr ~ sex + bmi + waist, data=predimed, max.ylev=7)

Similarly, by default there is a limit for the maximum number of levels for an explanatory variable. If this level is exceeded, the variable is removed from the analysis and a warning message is printed:

compareGroups(group ~ sex + age7gr, method= (age7gr=3), data=predimed, max.xlev=5)
wzxhzdk:19

Dressing up the output

Although the options described in this section correspond to compareGroups function, results of changing/setting them won't be visible until the table is created with the createTable function (explained later).

compareGroups(group ~ age + smoke + waist + hormo, data=predimed, include.label= FALSE)
resu1<-compareGroups(group ~ age + waist, data=predimed, method = c(waist=2))
createTable(resu1)

Note: percentiles 25 and 75 are calculated for "Waist circumference".

To get instead percentile 2.5% and 97.5%:

resu2<-compareGroups(group ~ age + smoke + waist + hormo, data=predimed, method = c(waist=2), Q1=0.025, Q3=0.975)
createTable(resu2)

Note: percentiles 2.5% and 97.5% are calculated for Follow-up.

To get minimum and maximum:

compareGroups(group ~ age + smoke + waist + hormo, data=predimed, method = c(waist=2), Q1=0, Q3=1)
predimed$smk<-predimed$smoke
levels(predimed$smk)<- c("Never smoker", "Current or former < 1y", "Never or former >= 1y", "Unknown")
attr(predimed$smk,"label")<-"Smoking 4 cat."
cbind(table(predimed$smk))

Note that this new category ("unknown") has no individuals:

compareGroups(group ~ age + smk + waist + hormo, data=predimed)
wzxhzdk:26

Note that a "Warning" message is printed related to the problem with smk.

To avoid using empty categories, simplify must be stated as TRUE (Default value).

compareGroups(group ~ age + smk, data=predimed, simplify=FALSE)
wzxhzdk:28

Nota that a "warning" message is shown and no p-values are calculated for "Smoking".

Summary

Applying the summary function to an object of class createTable will obtain a more detailed output:

res<-compareGroups(group ~ age + sex + smoke + waist + hormo, method = c(waist=2), data=predimed)
summary(res[c(1, 2, 4)])

Note that because only variables 1, 3 and 4 are selected, only results for Age, Sex and Waist circumference are shown. Age is summarized by the mean and the standard deviation, Sex by frequencies and percentage, and Waist circumference (method =2) by the median and quartiles.

Plotting

Variables can be plotted to see their distribution. Plots differ according to whether the variable is continuous or categorical. Plots can be seen on-screen or saved in different formats (BMP, JPG', PNG, TIF or PDF). To specify the format use the argument `type'.

plot(res[c(1,2)], file="./figures/univar/", type="png")

alt text alt text

Plots also can be done according to grouping variable. In this case only a boxplot is shown for continuous variables:

plot(res[c(1,2)], bivar=TRUE, file="./figures/bivar/", type="png")

alt text alt text

Updating

The object from compareGroups can later be updated. For example:

res<-compareGroups(group ~ age + sex + smoke + waist + hormo, data=predimed)
res

The object res is updated using:

res<-update(res, . ~. - sex +  bmi + toevent, subset = sex=='Female', method = c(waist=2, tovent=2), selec = list(bmi=!is.na(hormo)))
res

Note that "Sex" is removed as an explanatory variable but used as a filter, subsetting only "Female" participants. Variable "Waist circumference" has been changed to "continuous non-normal". Two new variables have been added: Body mass index and Follow-up (stated continuous non-normal). For Body mass index is stated to show only data of participants with non-missing values in Hormone-replacement therapy.

Substracting results

Since version 3.0, there is a new function called getResults to retrieve some specific results computed by compareGroups, such as p-values, descriptives (means, proportions, ...), etc.

For example, it may be interesting to recover the p-values for each variable as a vector to further manipulate it in R, like adjusting for multiple comparison with p.adjust. For example, lets take the example data from SNPassoc package that contains information of dozens of SNPs (genetic variants) from a sample of cases and controls. In this case we analize five of them:

library(SNPassoc)
data(SNPs)
tab <- createTable(compareGroups(casco ~ snp10001 + snp10002 + snp10005 + snp10008 + snp10009, SNPs))
pvals <- getResults(tab, "p.overall")
p.adjust(pvals, method = "BH")

Odds Ratios and Hazard Ratios

When the response variable is binary, the Odds Ratio (OR) can be printed in the final table. If the response variable is time-to-event (see Section 3.1), the Hazard Ratio (HR) can be printed instead.

res1<-compareGroups(htn ~ age + sex + bmi + smoke, data=predimed, ref=1)
createTable(res1, show.ratio=TRUE)

Note that for categorical response variables the reference category is the first one in the statement:

res2<-compareGroups(htn ~ age + sex + bmi + smoke, data=predimed, ref=c(smoke=1, sex=2))
createTable(res2, show.ratio=TRUE)

Note that the reference category for Smoking status is the first and for Sex the second.

res<-compareGroups(htn ~ age + sex + bmi + hormo + hyperchol, data=predimed, ref.no='NO')
createTable(res, show.ratio=TRUE)

Note: "no", "No" or "NO" will produce the same results; the coding is not case sensitive.

res<-compareGroups(htn ~ age + bmi, data=predimed)
createTable(res, show.ratio=TRUE)

Here the OR is for the increase of one unit for Age and Systolic blood pressure.

res<-compareGroups(htn ~ age + bmi, data=predimed, fact.ratio= c(age=10, bmi=2))
createTable(res, show.ratio=TRUE)

Here the OR is for the increase of 10 years for Age and 2 units for "Body mass index".

res<-compareGroups(htn ~ age + sex + bmi + hyperchol, data=predimed)
createTable(res, show.ratio=TRUE)

Note: This output shows the OR of having hypertension. Therefore, "Non-hypertension" is the reference category.

res<-compareGroups(htn ~ age + sex + bmi + hyperchol, data=predimed, ref.y=2)
createTable(res, show.ratio=TRUE)

Note: This output shows the OR of having No hypertension, and 'Hypertension' is now the reference category.

When the response variable is of class Surv, the bivariate plot function returns a Kaplan-Meier figure if the explanatory variable is categorical. For continuous variables the function returns a line for each individual, ending with a circle for censored and with a plus sign for uncensored.

plot(compareGroups(tmain ~ sex, data=predimed), bivar=TRUE, file="./figures/bivarsurv/", type="png")
plot(compareGroups(tmain ~ age, data=predimed), bivar=TRUE, file="./figures/bivarsurv/", type="png")

alt text alt text

Time-to-event explanatory variables

When a variable of class Surv (see Section 3.1) is used as explanatory it will be described with the probability of event, computed by Kaplan-Meier, up to a stated time.

res<-compareGroups(sex ~  age + tmain, timemax=c(tmain=3), data=predimed)
res

Note that tmain is calculated at 3 years (see section 3.1).

The plot function applied to a variable of class Surv returns a Kaplan-Meier figure. The figure can be stratified by the grouping variable.

plot(res[2], file="./figures/univar/", type="png")
plot(res[2], bivar=TRUE, file="./figures/bivar/", type="png")

alt text alt text

Performing the descritive table

The createTable function, applied to an object of compareGroups class, returns tables with descriptives that can be displayed on-screen or exported to CSV, LaTeX, HTML, Word or Excel.

res<-compareGroups(group ~ age + sex + smoke + waist + hormo, data=predimed, selec = list(hormo=sex=="Female"))
restab<-createTable(res)

Two tables are created with the createTable function: one with the descriptives and the other with the available data. The print method applied to an object of class createTable prints one or both tables on the R console:

print(restab,which.table='descr')

Note that the option "descr" prints descriptive tables.

print(restab,which.table='avail')
print(restab,which.table='avail')

While the option "avail" prints the available data, as well as methods and selections.

By default, only the descriptives table is shown. Stating "both" in which.table argument prints both tables.

Dressing up tables

update(restab, hide = c(sex="Male"))

Note that the percentage of males is hidden.

res<-compareGroups(group ~ age + sex + htn + diab, data=predimed)
createTable(res, hide.no='no', hide = c(sex="Male"))

Note: "no", "No" or "NO" will produce the same results; the coding is not case sensitive.

createTable(res, digits= c(age=2, sex = 3))

Note that mean and standard deviation has two decimal places for age, while percentage in sex has been set to three decimal places.

createTable(res, type=1)

Note that only percentages are displayed.

createTable(res, type=3)

Note that only frequencies are displayed.

Value 2 or NA return the same results, i.e., the default option.

createTable(res, show.n=TRUE)
createTable(res, show.descr=FALSE)
createTable(res, show.all=TRUE)
createTable(res, show.p.overall=FALSE)
createTable(res, show.p.trend=TRUE)

Note: The p-value for trend is computed from the Pearson test when row-variable is normal and from the Spearman test when it is continuous non-normal. If row-variable is of class Surv, the test score is computed from a Cox model where the grouping variable is introduced as an integer variable predictor. If the row-variable is categorical, the p-value for trend is computed as 1-pchisq(cor(as.integer(x),as.integer(y))^2*(length(x)-1),1)

createTable(res, show.p.mul=TRUE)
wzxhzdk:60

Note: Tukey method is used when explanatory variable is normal-distributed and Benjamini & Hochberg [@BH] method otherwise.

createTable(update(res, subset= group!="Control"), show.ratio=TRUE)

Note that category "Control diet" of the response variable has been omitted in order to have only two categories (i.e., a dichotomous variable). No Odds Ratios would be calculated if response variable has more than two categories.

Note that when response variable is of class Surv, Hazard Ratios are calculated instead of Odds Ratios.

createTable(compareGroups(tmain ~  group + age + sex, data=predimed), show.ratio=TRUE)
createTable(compareGroups(tmain ~  group + age + sex, data=predimed), show.ratio=TRUE, digits.ratio= 3)
tab<-createTable(compareGroups(tmain ~  group + age + sex, data=predimed), show.all = TRUE)
print(tab, header.labels = c("p.overall" = "p-value", "all" = "All"))

Combining tables by row (groups of variable)

Tables made with the same response variable can be combined by row:

restab1 <- createTable(compareGroups(group ~ age + sex, data=predimed))
restab2 <- createTable(compareGroups(group ~ bmi + smoke, data=predimed))
rbind("Non-modifiable risk factors"=restab1, "Modifiable risk factors"=restab2)

Note how variables are grouped under "Non-modifiable" and "Modifiable"" risk factors because of an epigraph defined in the rbind command in the example.

The resulting object is of class rbind.createTable, which can be subset but not updated. It inherits the class createTable. Therefore, columns and other arguments from the createTable function cannot be modified:

To select only Age and Smoking:

x <- rbind("Non-modifiable"=restab1,"Modifiable"=restab2)
rbind("Non-modifiable"=restab1,"Modifiable"=restab2)[c(1,4)]

To change the order:

rbind("Modifiable"=restab1,"Non-modifiable"=restab2)[c(4,3,2,1)]

Combining tables by column (strata)

Columns from tables built with the same explanatory and response variables but done with a different subset (i.e. "ALL", "Male" and "Female", strata) can be combined:

res<-compareGroups(group ~ age +  smoke + bmi + htn , data=predimed)
alltab <- createTable(res,  show.p.overall = FALSE)
femaletab <- createTable(update(res,subset=sex=='Female'), show.p.overall = FALSE)
maletab <- createTable(update(res,subset=sex=='Male'), show.p.overall = FALSE)
cbind("ALL"=alltab,"FEMALE"=femaletab,"MALE"=maletab)
wzxhzdk:70

With the argument caption set to NULL no name is displayed for columns.

cbind(alltab,femaletab,maletab,caption=NULL)
wzxhzdk:72

By default the name of the table is displayed for each set of columns.

cbind(alltab,femaletab,maletab)
wzxhzdk:74

NOTE: The resulting object is of class cbind.createTable and inherits also the class createTable. This cannot be updated. It can be nicely printed on the R console and also exported to LaTeX but it cannot be exported to CSV or HTML.


Since version 4.0, it exists the function strataTable to build tables within stratas defined by the values or levels defined of a variable. Notice that the syntax is much simpler than using cbind method. For example, to perform descriptives by groups, and stratified per gender:

  1. first build the table with descriptives by groups:
res <- compareGroups(group ~ . -sex, predimed)
restab <- createTable(res, hide.no="no")
  1. and then apply the strataTable function on the table:
strataTable(restab, "sex")
wzxhzdk:77

Miscellaneous

In this section some other createTable options and methods are discussed:

print(createTable(compareGroups(group ~ age + sex + smoke + waist + hormo, data=predimed)), which.table='both')

With the print method setting nmax argument to FALSE, the total maximum "n" in the available data is omitted in the first row.

print(createTable(compareGroups(group ~ age + sex + smoke + waist + hormo, data=predimed)),  nmax=FALSE)
summary(createTable(compareGroups(group ~ age + sex + smoke + waist + hormo, data=predimed)))
res<-compareGroups(group ~ age + sex + smoke + waist + hormo, data=predimed)
restab<-createTable(res, type=1, show.ratio=TRUE )
restab
update(restab, show.n=TRUE)

In just one statement it is possible to update an object of class compareGroups and createTable:

update(restab, x = update(res, subset=c(sex=='Female')), show.n=TRUE)

Note that the compareGroups object (res) is updated, selecting only "Female"" participants, and the createTable class object (restab) is updated to add a column with the maximum available data for each explanatory variable.

createTable(compareGroups(group ~ age + sex + smoke + waist + hormo, data=predimed))
createTable(compareGroups(group ~ age + sex + bmi, data=predimed))[1:2, ]

Building tables in one step {#descrTable}

Making use of descrTable the user can build descriptive table in one step. This function takes all the arguments of compareGroups function plus the ones from createTable function. The result is the same as if the user first call compareGroups and then createTable. Therefore one can use the same methods and functions avaiable for createTable objects (subsetting, ploting, printing, exporting, etc.)

To describe all varaible from predimed, just type:

descrTable(predimed)

To describe some varaibles, make use of formula argument (just as using compareGroups function)

descrTable(~ age + sex, predimed)

To hide "no" category from yes-no variables, use the hide.no argument from createTable:

descrTable(predimed, hide.no="no")

To report descriptives by group as well as the descriptives of the entire cohort:

descrTable(group ~ ., predimed, hide.no="no", show.all=TRUE)

Or you can select individuals using the subset argument.

descrTable(predimed, subset=age>65)

Exporting tables

Tables can be exported to CSV, HTML, LaTeX, PDF, Markdown, Word or Excel

Note that, since version 3.0, it is necessary write the extension of the file.

General exporting options

Exporting to LaTeX

A special case of exporting is when tables are exported to LaTeX. The function export2latex returns an object with the tex code as a character that can be changed in the R session.

restab<-createTable(compareGroups(group ~ age + sex + smoke + waist + hormo, data=predimed))
export2latex(restab)

Exporting to Rmarkdown

Since 4.0 version, export2md supports cbind.createTable class objects, i.e. when exporting stratified descriptive tables. Also, nicer and more costumizable tables can be reported making use of kableExtra package (such as size, strip rows, etc.).

Following, there are some examples when exporting to HTML.

First create the descriptive table of PREDIMED variables by intervention groups.

res <- compareGroups(group ~ ., predimed)
restab <- createTable(res, hide.no="no")
export2md(restab)
export2md(restab, strip=TRUE, first.strip=TRUE)
export2md(restab, size=6)
export2md(restab, width="400px")
resMales <- compareGroups(group ~ . -sex, predimed, subset=sex=="Male", simplify=FALSE)
resFemales <- compareGroups(group ~ . -sex, predimed, subset=sex=="Female", simplify=FALSE)
restabMales <- createTable(resMales, hide.no="no")
restabFemales <- createTable(resFemales, hide.no="no")
restab <- cbind("Males"=restabMales, "Females"=restabFemales)
export2md(restab, size=8)

Generating an exhaustive report

Since version 2.0 of compareGroups package, there is a function called report which automatically generates a PDF document with the "descriptive" table as well as the corresponding "available"" table. In addition, plots of all analysed variables are shown.

In order to make easier to navigate throught the document, an index with hyperlinks is inserted in the document.

See the help file of this function where you can find an example with the REGICOR data (the other example data set contained in the compareGroups package)

# to know more about report function
?report

# info about REGICOR data set
?regicor

Also, you can use the function radiograph that dumps the raw values on a plain text file. This may be usefull to identify possible wrong codes or non-valid values in the data set.

Dealing with missing values

Many times, it is important to be aware of the missingness contained in each variable, possibly by groups. Althought "available" table shows the number of the non-missing values for each row-variable and in each group, it would be desirable to test whether the frequency of non-available data is different between groups. For this porpose, a new function has been implemented in the compareGroups package, which is called missingTable. This function applies to both compareGroups and createTable class objects. This last option is useful when the table is already created. To illustrate it, we will use the REGICOR data set, comparing missing rates of all variables by year:

# from a compareGroups object
data(regicor)
res <- compareGroups(year ~ .-id, regicor)
missingTable(res)
# or from createTable objects
restab <- createTable(res, hide.no = 'no')
missingTable(restab)

Perhaps a NA value of a categorical variable may mean something different from just non available. For example, patients admitted for "Coronary Acute Syndrome" with NA in "ST elevation" may have a higher risk of in-hospital death than the ones with available data, i.e. "ST elevation" yes or not. If these kind of variables are introduced in the data set as NA, they are removed from the analysis. To avoid the user having to recode NA as a new category for all categorical variables, new argument called include.miss in compareGroups function has been implemented which does it automatically. Let's see an example with all variables from REGICOR data set by cardiovascular event.

# first create time-to-cardiovascular event
regicor$tcv<-with(regicor,Surv(tocv,cv=='Yes'))
# create the table
res <- compareGroups(tcv ~ . -id-tocv-cv-todeath-death, regicor, include.miss = TRUE)
restab <- createTable(res, hide.no = 'no')
restab

Analysis of genetic data

In the version 2.0 of compareGroups, it is possible to analyse genetic data, more concretely Single Nucleotic Polymorphisms (SNPs), using the function compareSNPs. This function takes advantage of SNPassoc [@SNPassoc] and HardyWeinberg [@HW] packages to perform quality control of genetic data displaying the Minor Allele Frequencies, Missingness, Hardy Weinberg Equilibrium, etc. of the whole data set or by groups. When groups are considered, it also performs a test to check whether missingness rates is the same among groups.

Following, we illustrate this by an example taking a data set from SNPassoc package.

First of all, load the SNPs data from SNPassoc, and visualize the first rows. Notice how are the SNPs coded, i.e. by the alleles. The alleles separator can be any character. If so, this must be specified in the sep argument of compareSNPs function (type ?compareSNPs for more details).

data(SNPs)
head(SNPs)

In this data frame there are some genetic and non-genetic data. Genetic variables are those whose names begin with "snp". If we want to summarize the first three SNPs by case control status:

res<-compareSNPs(casco ~ snp10001 + snp10002 + snp10003, data=SNPs)
res

Note that all variables specified in the right hand side of the formula must be SNPs, i.e. variables whose levels or codes can be interpreted as genotypes (see setupSNPs function from SNPassoc package for more information). Separated summary tables by groups of cases and controls are displayed, and the last table corresponds to missingness test comparing non-available rates among groups.

If summarizing SNPs in the whole data set is desired, without separating by groups, leave the left side of formula in blank, as in compareGroups function. In this case, a single table is displayed and no missingness test is performed.

res<-compareSNPs(~ snp10001 + snp10002 + snp10003, data=SNPs)
res

Using the Graphical User Inferface (GUI) - tcltk {#gui}

Once the compareGroups package is loaded, a Graphical User Interface (GUI) is displayed in response to typing cGroupsGUI(predimed). The GUI is meant to make it feasible for users who are unfamiliar with R to construct bivariate tables. Note that, since version 3.0, it is necessary to specifiy an existing data.frame as input. So, for example, you can load the PREDIMED data by typing data(predimed) before calling cGroupsGUI function.

In this section we illustrate, step by step, how to construct a bivariate table containing descriptives by groups from the predimed data using the GUI:

export2md(createTable(compareGroups(group ~  age + sex + smoke + bmi + waist + wth + htn + diab + hyperchol + famhist + hormo + p14 + toevent + event, data=predimed), hide.no="No",hide = c(sex="Male")))

Computing Odds Ratio

For a case-control study, it may be necessary to report the Odds Ratio between cases and controls for each variable. The table below contains Odds Ratios for each row-variable by hypertension status.

export2md(createTable(compareGroups(htn ~  age + sex + smoke + bmi + waist + wth + diab + hyperchol + famhist + hormo + p14 + toevent + event, data=predimed), hide.no="No",hide = c(sex="Male"), show.ratio=TRUE, show.descr=FALSE))

To build this table, as illustrated in the screens below, you would select htn variable (Hypertension status) as the factor variable, indicate "no" category on the "reference" pull-down menu, and mark "Show odds/hazard ratio" in the "Report Options" menu before exporting the table.

Computing Hazard Ratio

In a cohort study, it may be more informative to compute hazard ratio taking into account time-to-event.

export2md(createTable(compareGroups(tmain ~  group + age + sex, data=predimed), show.ratio=TRUE))

To generate this table, select toevent variable and event, indicating the time-to-event and the status, respectively, and select the event category for the status variable. Finally, as for Odds Ratios, mark 'Show odds/hazard ratio' in the 'Report Options' menu before exporting the table.

To return to the R console, just close the GUI window.

Using Web User Inferface (WUI) - Shiny {#wui}

Since version 2.1, compareGropus package incorporates a Web User Interface (WUI) based on shiny R available on CRAN repository [@Shiny] shiny website to facilitate the use of the package for non R users.

This application includes almost all the options existing in "type on" version. Also, thanks to the power of shiny package, the user can see the results when setting the included variable, the groups, number of decimals, etc almost instantaneously ("reactivity" -see shiny manual and examples-). This is very useful to modify and customize the descriptive table before saving it in the desired format saving a lot of time.

In the following subsections, we list and describe all the options available in the Shiny-compareGroups application, and we illustrate how it works with a real example.

Menus and options

Once the WUI is called, a web browser is launched with two main parts: in the left hand side there are all the control menu options to read the data frame from different format files, select the variables list to be analysed, specify the number of digits of the descriptives or p-values displayed in the bivariate table, and much more other aspects; while the right hand side contains the results, i.e, a table containing a basic summary of all variables, the bivariate table or some plots.

Following, an exhaustive list describing all the options are presented:

Control menu (left side)

START

A panel with all the options to select the data base and to select the analysed variables. Different file formats are allowed; in the present version these are: SPSS, plain text (txt or csv), EXCEL (2000 or 2007) or R (.rda or RData). Also, it is possible to select example data already present in the compareGroups package (REGICOR and PREDIMED) and SNPS from SNPassoc package.

SETTINGS

A panel with different issues to control for different aspects of the bivariate table:

Note that to indicate the category, you must type the number that appears in the "VALUES" table on right side of the application instead of its >>name. For example, for gender, type 1 instead of "male" and 2 instead of "female", if "male" is the first category and "female" is the second.

DISPLAY

In this panel, the user can select which information must be displayed in the resulting bivariate table, and in which format, etc.

SAVE

A panel with the options to save the bivariate table in different formats: PDF, CSV, TXT, HTML, Word (.docx) or Excel (.xlsx).

If PDF format is selected, size can be specified as well as landscape format. These options may be useful when the table is big or contains lots of columns.

Results panels (right side)}

Example

In this section we illustrate how to analyse a data set.

Launch the WUI application

To use the WUI locally (and not on a remote server), first load the compareGroups package and call the cGroupsWUI function:

library(compareGroups)
cGroupsWUI()

Load the data

In this example we will load the PREDIMED data set from the PREDIMED study [@PREDIMED]. This data is already available in the compareGroups package.

Select the variables to be analysed

After the data is loaded satisfactory, the "Step 2. Select variable" panel is opened automatically. Using this panel, we will select all variables except the "event" and "toevent" (the time-to-event) variables. They will represent the response and must be removed from the row-variables.

Select the grouping variable

Since PREDIMED is a longitudinal study where the main goal is to check whether the mediterranean diet is related to a less incidence of cardiovascular disease, we will take the response as the time to cardiovascular event. To do so, we will select "Survival" response type, setting the "toevent" as time variable and "event" as indicator variable (taking "yes" as case code).

Take a look at the plots

To see which continuous variables should be treated as normal and which not. By default, compareGroups performs a Shapiro-Wilks normality test to decide which variables are normal. But we may want to check the normality assumption graphically.

Set the options.

What to be displayed and how.

Visualize the table

To see the output, click on TABLE tab on the right panel.

By pressing "View options" button a slider to customize the font-size of the bivariate table appears, as well as an "info" button. Pressing the "info" button a table containg information about the number of available data by variable and group, type of variable (normal, non-normal or categorical, etc) is displayed on a modal.

Saving the table

Finally, once the table contains all the desired figures and in the appropriate format, it can be downloaded and stored in different formats:

a. PDF: a LaTeX compiler such as MikTex must be installed to build the PDF document with the descriptive table. The user can select the font size and whether table must be placed vertically or horizontally (landscape option).

b. CSV: a plain text file with columns separated by commas or semicolons. For Windows users, this format is useful since it can be opened by Excel.

c. HTML: a web browser is opened and the table can be easily copied and pasted to Word, for instance.

d. TXT: a plain text file which can be opened by any text editor program and which contains the table as in R console with a nice format. Once the "download" button is pressed, the file is automatically stored in your PC/Mac.

e. Word: A Word document file (either 2000, 2003 or 2010 version) is created.

f. Excel: An Excel sheet file (either 2000, 2003 or 2010 version) is created.

References



Try the compareGroups package in your browser

Any scripts or data that you put into this service are public.

compareGroups documentation built on May 27, 2019, 5:01 p.m.