knitr::opts_chunk$set( collapse = TRUE, comment = "#>" )
install.packages("finalfit")
Two-way tables are used extensively in healthcare research, e.g. a 2x2 table comparing two factors with two levels each, or table 1 from a typical clinical study or trial
The main functions all take a dependent
variable - the outcome - and explanatory
variables - predictors or exposures (any number categorical or continuous variables).
library(finalfit) explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "perfor.factor" colon_s %>% summary_factorlist(dependent, explanatory) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r"))
Note, chi-squared warnings will be generated when the expected count in any cell is less than 5. Fisher's exact test can be used as below, or go straight to a univariable logistic regression, e.g. colon_s %>% finalfit(dependent, explanatory)
library(finalfit) library(dplyr) explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "perfor.factor" colon_s %>% mutate( sex.factor = ff_label(sex.factor, "Gender") ) %>% summary_factorlist(dependent, explanatory) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r"))
Defaults are chi-squared for categorical explanatory variables and an F-test for continuous (aov
, analysis of variance). Alternatives can be specified as per below.
library(finalfit) explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "perfor.factor" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r"))
library(finalfit) explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "perfor.factor" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE, p_cat = "fisher") -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r"))
Summaries for continuous explanatory variables are mean (standard deviation) with aov
statistical test by default.
The statistical test can be changed to the Welch t-test when there are two dependent variable levels if desired.
library(finalfit) explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "perfor.factor" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE, p_cont_para = "t.test") -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r"))
If desired, all continuous explanatory variables can be considered non-parametric. Summaries will be median (interquartile range) and the statistical test is Kruskal-Wallis/Mann-Whitney U. Use cont_range = FALSE
if wish single-digit IQR, i.e. Q3-Q1.
library(finalfit) explanatory = c("age", "nodes", "age.factor", "sex.factor", "obstruct.factor") dependent = "perfor.factor" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE, cont = "median") -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r"))
Many have asked in the past if only particular variables can be considered non-parametric.
The argument cont_nonpara
can take a vector (e.g. c(1, 2, 3, 4)
) of values corresponding to the explanatory variable to specify which should be summarised as a median and be passed to a non-parametric test.
library(finalfit) explanatory = c("age", "nodes", "age.factor", "sex.factor", "obstruct.factor") dependent = "perfor.factor" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE, cont_nonpara = c(2)) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r"))
Always consider summarising missing values when describing your data.
library(finalfit) explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "perfor.factor" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE, na_include = TRUE) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r"))
This is a change from the default behaviour introduced in Finalfit 1.0.0.
Previously, when missing data was presented it was also considered as a level in the statistical test.
This may or may not be desired.
Control this now using na_to_p = TRUE
to include missing data in test.
A message is produced reminding you that you are doing that.
library(finalfit) explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "perfor.factor" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE, na_include = TRUE, na_to_p = TRUE) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r"))
library(finalfit) explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "perfor.factor" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE, na_include = TRUE, column = FALSE) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r"))
The terms total column was introduced before this function summarised continuous variables. It would be better to be "All data" or something similar, as the continous explanatory variables a summary statistic is produced for all data. However, to keep backwards compatibility we have left it unchanged for now. For producing row totals including continous explanatory variables, see add_row_total
below.
library(finalfit) explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "perfor.factor" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE, cont = "median", na_include = TRUE, column = TRUE, total_col = TRUE) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r"))
This was introduced to deal with the problem of summarising missing data for continuous variables. By default, it provides the total N for the variable and includes a column enumerating missing values.
library(finalfit) explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "perfor.factor" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE, na_include = TRUE, total_col = TRUE, add_row_total = TRUE) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r"))
Remove missing column.
library(finalfit) explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "perfor.factor" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE, na_include = TRUE, total_col = TRUE, add_row_total = TRUE, include_row_missing_col = FALSE) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r"))
library(finalfit) explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "perfor.factor" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE, total_col = TRUE, add_row_total = TRUE, row_totals_colname = "N (total)", row_missing_colname = "N (missing)") -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r"))
This is intended for when there is only one explanatory
variable.
library(finalfit) explanatory = c("extent.factor") dependent = "perfor.factor" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE, cont = "median", na_include = TRUE, column = TRUE, total_col = TRUE, orderbytotal = TRUE) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r"))
Column totals can be added, and by default are presented with a row percentage.
explanatory = c("age.factor", "sex.factor") dependent = "rx.factor" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE, add_col_totals = TRUE) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "l", "l", "r", "r", "r"))
explanatory = c("age.factor", "sex.factor") dependent = "rx.factor" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE, add_col_totals = TRUE, include_col_totals_percent = FALSE) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "l", "l", "r", "r", "r"))
explanatory = c("age.factor", "sex.factor") dependent = "rx.factor" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE, add_col_totals = TRUE, include_col_totals_percent = FALSE, col_totals_rowname = "", col_totals_prefix = "N=") -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "l", "l", "r", "r", "r"))
dependent
namelibrary(finalfit) explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "perfor.factor" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE, cont = "median", na_include = TRUE, column = TRUE, total_col = TRUE, add_dependent_label = TRUE) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r"))
The dependent name cannot be passed directly to the table intentionally. This is to avoid errors when code is copied and the name is not updated. Change the dependent label using the following. The prefix ("Dependent: ") and any suffix can be altered.
library(finalfit) explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "perfor.factor" colon_s %>% dplyr::mutate( perfor.factor = ff_label(perfor.factor, "Perforated cancer") ) %>% summary_factorlist(dependent, explanatory, p = TRUE, cont = "median", na_include = TRUE, column = TRUE, total_col = TRUE, add_dependent_label = TRUE, dependent_label_prefix = "") -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r"))
library(finalfit) explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "extent.factor" colon_s %>% dplyr::mutate( perfor.factor = ff_label(perfor.factor, "Perforated cancer") ) %>% summary_factorlist(dependent, explanatory, p = TRUE, cont = "median", na_include = TRUE, column = TRUE, total_col = TRUE, add_dependent_label = TRUE, dependent_label_prefix = "") -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
If you are careful to count totals and you know your data, you should realise when there is data missing from the dependent, e.g.:
explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "mort_5yr" colon_s %>% ff_glimpse(dependent, explanatory)
To make sure, a warning is generated when data are dropped from the dependent:
library(finalfit) explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "mort_5yr" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE, na_include = TRUE, total_col = TRUE, add_col_totals = TRUE, add_row_totals = TRUE) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
You may consider making the missing data explicit.
library(finalfit) explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "mort_5yr" colon_s %>% mutate( mort_5yr = forcats::fct_na_value_to_level(mort_5yr, level = "(Missing)") ) %>% summary_factorlist(dependent, explanatory, p = TRUE, na_include = TRUE, total_col = TRUE, add_col_totals = TRUE, add_row_totals = TRUE) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
Rather than making the data explicit in the dataset, you can use na_include_dependent = TRUE
to do the same in summary_factorlist()
.
library(finalfit) explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "mort_5yr" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE, na_include = TRUE, na_include_dependent = TRUE, total_col = TRUE, add_col_totals = TRUE, add_row_totals = TRUE) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
You may wish to see summaries for complete cases across included variables. Rather than selecting including variables and drop_na()
, you can pass na_complete_cases = TRUE
to summary_factorlist()
to do the same.
library(finalfit) explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "mort_5yr" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE, na_complete_cases = TRUE, total_col = TRUE, add_col_totals = TRUE, add_row_totals = TRUE) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
You may wish to actively remove any rows with missing data, so you are explicit around which data are being used in models.
Unfortunately some tidyverse functions silently remove variable attributes (labels). This is complained about then put right. But here is a workaround if it is happening with a variable you wish to use, such as tidyr::drop_na()
.
library(finalfit) explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "mort_5yr" vlabels = colon_s %>% extract_variable_label() colon_s %>% select(dependent, explanatory) %>% tidyr::drop_na() %>% # Silently removes attributes ff_relabel(vlabels) %>% # Relabel summary_factorlist(dependent, explanatory, p = TRUE, na_include = TRUE, total_col = TRUE, add_col_totals = TRUE, add_row_totals = TRUE) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
library(finalfit) # Here, `extent` is a continuous variable with 4 distinct values. # Any continuous variable with 5 or fewer unique values is converted silently to factor # e.g. explanatory = c("extent") dependent = "mort_5yr" colon_s %>% summary_factorlist(dependent, explanatory) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
library(finalfit) # Here, `extent` is a continuous variable with 4 distinct values. # Any continuous variable with 5 or fewer unique values is converted silently to factor # e.g. explanatory = c("extent") dependent = "mort_5yr" colon_s %>% summary_factorlist(dependent, explanatory, cont_cut = 0) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
I've been meaning to include support for table stratification for a while. I have delayed for a good reason. Perhaps the most straightforward way to implement stratificiation is with dplyr::group_by()
. However, the non-standard evaluation required for multiple strata may confuse as it is not implemented else where in the package.
This translates to whether variable names are passed in quotes or not.
Here is a solution, which while not that pretty, is effective.
Note that tidyverse
functions every so often start stripping labels/attributes. Hence the addition of the help function.
library(dplyr) explanatory = c("age.factor", "sex.factor") dependent = "perfor.factor" # Pick option below split = "rx.factor" split = c("rx.factor", "node4.factor") # Piped function to generate stratified crosstabs table colon_s %>% group_by(!!! syms(split)) %>% # Looks awkward, but avoids unquoted var names group_modify(~ summary_factorlist(.x, dependent, explanatory)) %>% ff_stratify_helper(colon_s) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "l", "l", "r", "r", "r"))
explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "perfor.factor" colon_s %>% summary_factorlist(dependent, explanatory, p = TRUE, digits = c(1,2,3,4,0)) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
A simple weighting can be applied to tables. Explanatory continuous variables are multiplied by weights. Explanatory categorical variables are counted with a frequency weight (sum(weights)
). This could be used with, say, inverse probability of treatment weightings (IPTW). The example uses random weights for demonstration purposes only. Hypothesis tests are not run on weighted data, p is set to FALSE.
explanatory = c("age", "age.factor", "sex.factor", "obstruct.factor") dependent = "perfor.factor" colon_s %>% mutate(my_weights = runif(929, 0, 1)) %>% # Random just to demonstrate summary_factorlist(dependent, explanatory, weights = "my_weights", digits = c(1, 1, 3, 1, 1))-> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
finalfit()
Logistic regression first.
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = "mort_5yr" colon_s %>% finalfit(dependent, explanatory) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
Most appropriate when all explanatory variables are continuous or well-known binary variables, such as sex.
library(finalfit) explanatory = c("age", "sex.factor") dependent = "mort_5yr" colon_s %>% finalfit(dependent, explanatory, add_dependent_label = FALSE) %>% ff_remove_ref() %>% dependent_label(colon_s, dependent)-> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = "mort_5yr" colon_s %>% finalfit(dependent, explanatory, metrics = TRUE) -> t
library(knitr) kable(t[[1]], row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r")) kable(t[[2]], row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"), col.names = "")
library(finalfit) glm(mort_5yr ~ age.factor + sex.factor + obstruct.factor + perfor.factor, data = colon_s, family = "binomial") %>% ff_metrics() -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"), col.names = "")
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") explanatory_multi = c("age.factor", "obstruct.factor") dependent = "mort_5yr" colon_s %>% finalfit(dependent, explanatory, explanatory_multi) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") explanatory_multi = c("age.factor", "obstruct.factor") dependent = "mort_5yr" colon_s %>% finalfit(dependent, explanatory, explanatory_multi, metrics = TRUE, keep_models = TRUE) -> t
library(knitr) kable(t[[1]], row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r")) kable(t[[2]], row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"), col.names = "")
Interactions can be specified in the normal way. Formatting the output is trickier. At the moment, we have left the default model output. This can be adjusted as necessary.
library(finalfit) explanatory = c("age.factor*sex.factor", "obstruct.factor", "perfor.factor") dependent = "mort_5yr" colon_s %>% finalfit(dependent, explanatory) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
library(finalfit) #explanatory = c("age.factor*sex.factor", "obstruct.factor", "perfor.factor") explanatory = c("obstruct.factor", "perfor.factor") dependent = "mort_5yr" colon_s %>% ff_interaction(age.factor, sex.factor) %>% finalfit(dependent, c(explanatory, "age.factor_sex.factor")) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
The dependent name cannot be specified directly intentionally. This is to prevent errors when copying code. Re-label using ff_label()
. The dependent prefix and suffix can also be altered.
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = "mort_5yr" colon_s %>% dplyr::mutate( mort_5yr = ff_label(mort_5yr, "5-year mortality") ) %>% finalfit(dependent, explanatory, dependent_label_prefix = "", dependent_label_suffix = " (full model)") -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = "mort_5yr" colon_s %>% finalfit(dependent, explanatory, estimate_name = "Odds ratio") -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
Number of digits to round to regression results. (1) estimate, (2) confidence interval limits, (3) p-value. Default is c(2,2,3). Trailing zeros are preserved. Number of decimal places for counts and mean (sd) / median (IQR) not currently supported. Defaults are senisble :)
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = "mort_5yr" colon_s %>% finalfit(dependent, explanatory, digits = c(3,3,4)) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
One of c("profile", "default")
for GLM models (confint.glm()
). Note, a little awkwardly, the 'default' setting is profile
, rather than default
. Profile levels are probably a little more accurate. Only go to default if taking a significant length of time for profile, i.e. data is greater than hundreds of thousands of lines.
For glmer/lmer models (confint.merMod()
), c("profile", "Wald", "boot")
. Not implemented for lm()
, coxph()
or coxphlist
, which use default.
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = "mort_5yr" colon_s %>% finalfit(dependent, explanatory, confint_type = "default") -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
Probably never change this :) Note, the p-value is intentionally not included for confidence levels other than 95% to avoid confusion.
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = "mort_5yr" colon_s %>% finalfit(dependent, explanatory, confint_level = 0.90) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
Some like to avoid the hyphen so as not to confuse with minus sign. Obviously not an issue in logistic regression.
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = "mort_5yr" colon_s %>% finalfit(dependent, explanatory, confint_sep = " to ") -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
explanatory = c("age", "sex.factor") dependent = 'mort_5yr' # Standard finalfit regression table t1 = colon_s %>% finalfit(dependent, explanatory, keep_fit_id = TRUE) # GLM with Stata-like robust standard errors t2 = colon_s %>% glmmulti(dependent, explanatory) %>% lmtest::coeftest(., vcov = sandwich::vcovHC(., "HC1")) %>% broom::tidy(conf.int = TRUE) %>% remove_intercept() %>% select(term, estimate, conf.low, conf.high, p.value) %>% mutate(across(c(estimate, conf.low, conf.high), exp)) %>% # or mutate_at(vars()) as.data.frame() %>% condense_fit(estimate_name = "OR (multivariable robust SE)") ff_merge(t1, t2, last_merge = TRUE)
library(finalfit) explanatory = c("age", "sex.factor") dependent = 'mort_5yr' t1 = colon_s %>% finalfit(dependent, explanatory, keep_fit_id = TRUE) t2 = colon_s %>% glmmulti(dependent, explanatory) %>% lmtest::coeftest(., vcov = sandwich::vcovHC(., "HC1")) %>% broom::tidy(conf.int = TRUE) %>% remove_intercept() %>% dplyr::select(term, estimate, conf.low, conf.high, p.value) %>% dplyr::mutate_at(dplyr::vars(estimate, conf.low, conf.high), exp) %>% as.data.frame() %>% condense_fit(estimate_name = "OR (multivariable robust SE)") ff_merge(t1, t2, last_merge = TRUE) %>% knitr::kable(row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = "mort_5yr" colon_s %>% finalfit(dependent, explanatory) %>% ff_remove_p() -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
At its simplest, a random-intercept model can be specified using a single quoted variable, e.g. random_effect = "hospital"
. This is equivalent to random_effect = "(1 | hospital)"
. Alternatively you can provide the full specification including parenthesis, e.g. random_effect = "(1 | hospital) + (1 | country)"
.
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = "mort_5yr" random_effect = "hospital" colon_s %>% finalfit(dependent, explanatory, random_effect = random_effect, dependent_label_suffix = " (random intercept)") -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
Some recently asked about this and it is a good question. Here is an approach.
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = "mort_5yr" random_effect = "hospital" colon_s %>% finalfit(dependent, explanatory, random_effect = random_effect, keep_fit_id = TRUE) %>% ff_merge( explanatory %>% purrr::map_df(~ glmmixed(colon_s, dependent, .x, random_effect = random_effect) %>% fit2df(estimate_suffix = " (univariable with RE)")), last_merge = TRUE ) %>% dplyr::relocate(7, .before = 6) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
In the example below, allow the effect of age on outcome to vary by hospital. Note, this specification must have parentheses included.
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = "mort_5yr" random_effect = "(age.factor | hospital)" colon_s %>% finalfit(dependent, explanatory, random_effect = random_effect, dependent_label_suffix = " (random slope: age)") -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
lme4
Clearly, as models get more complex, parameters such as random effect group variances may require to be extracted directly from model outputs.
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = "mort_5yr" random_effect = "(age.factor | hospital)" colon_s %>% glmmixed(dependent, explanatory, random_effect = random_effect) %>% broom.mixed::tidy() -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
This can be useful if you want the numbers in the final table to match the final multivariable model. However, be careful to include a full explanation of this in the methods and the reason for exluding the missing data.
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = 'mort_5yr' colon_s %>% dplyr::select(explanatory, dependent) %>% tidyr::drop_na() %>% finalfit(dependent, explanatory) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = 'nodes' colon_s %>% finalfit(dependent, explanatory) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = "nodes" random_effect = "hospital" colon_s %>% finalfit(dependent, explanatory, random_effect = random_effect, dependent_label_suffix = " (random intercept)") -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = "nodes" random_effect = "(age.factor | hospital)" colon_s %>% finalfit(dependent, explanatory, random_effect = random_effect, dependent_label_suffix = " (random slope: age)") -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = "Surv(time, status)" colon_s %>% finalfit(dependent, explanatory) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
As above, the dependent label cannot be specfied directly in the model to avoid errors. However, in survival modelling the surivial object specification can be long or awkward. Therefore, here is the work around.
library(finalfit) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = "Surv(time, status)" colon_s %>% finalfit(dependent, explanatory, add_dependent_label = FALSE) %>% dplyr::rename("Overall survival" = label) %>% dplyr::rename(" " = levels) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
ff_merge()
Note summary_factorlist()
needs argument, fit_id = TRUE
.
library(finalfit) library(dplyr) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = "mort_5yr" ## Crosstable colon_s %>% summary_factorlist(dependent, explanatory, fit_id=TRUE) -> table_1 ## Univariable colon_s %>% glmuni(dependent, explanatory) %>% fit2df(estimate_suffix=" (univariable)") -> table_2 ## Merge table_1 %>% ff_merge(table_2) %>% select(-c(fit_id, index)) %>% dependent_label(colon_s, dependent)-> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
library(finalfit) library(dplyr) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") random_effect = "hospital" dependent = "mort_5yr" # All in one pipe colon_s %>% ## Crosstable summary_factorlist(dependent, explanatory, fit_id=TRUE) %>% ## Add univariable ff_merge( glmuni(colon_s, dependent, explanatory) %>% fit2df(estimate_suffix=" (univariable)") ) %>% ## Add multivariable ff_merge( glmmulti(colon_s, dependent, explanatory) %>% fit2df(estimate_suffix=" (multivariable)") ) %>% ## Add mixed effects ff_merge( glmmixed(colon_s, dependent, explanatory, random_effect) %>% fit2df(estimate_suffix=" (multilevel)"), last_merge = TRUE ) %>% dependent_label(colon_s, dependent) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
library(finalfit) library(dplyr) ## Dobson (1990) Page 93: Randomized Controlled Trial : counts = c(18,17,15,20,10,20,25,13,12) outcome = gl(3,1,9) treatment = gl(3,3) d.AD <- data.frame(treatment, outcome, counts) dependent = "counts" explanatory = c("outcome", "treatment") fit_uni = d.AD %>% glmuni(dependent, explanatory, family = poisson) %>% fit2df(estimate_name = "Rate ratio (univariable)") fit_multi = d.AD %>% glmmulti(dependent, explanatory, family = poisson) %>% fit2df(estimate_name = "Rate ratio (multivariable)") # All in one pipe d.AD %>% ## Crosstable summary_factorlist(dependent, explanatory, cont = "median", fit_id=TRUE) %>% ## Add univariable ff_merge(fit_uni, estimate_name = "Rate ratio") %>% ## Add multivariable ff_merge(fit_multi, estimate_name = "Rate ratio", last_merge = TRUE) %>% dependent_label(d.AD, dependent) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
library(finalfit) library(dplyr) # A Gamma example, from McCullagh & Nelder (1989, pp. 300-2) clotting <- data.frame( u = c(5,10,15,20,30,40,60,80,100), lot1 = c(118,58,42,35,27,25,21,19,18), lot2 = c(69,35,26,21,18,16,13,12,12)) dependent = "lot1" explanatory = "log(u)" fit_uni = clotting %>% glmuni(dependent, explanatory, family = Gamma) %>% fit2df(estimate_name = "Coefficient", exp = FALSE, digits = c(3,3,4)) # All in one pipe clotting %>% ## Crosstable summary_factorlist(dependent, explanatory, cont = "median", fit_id=TRUE) %>% ## Add fit ff_merge(fit_uni, last_merge = TRUE) %>% dependent_label(colon_s, dependent) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
Note this is updated August 2022. Weights can specified from data and summary_factorlist()
table is now weighted.
library(finalfit) library(dplyr) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = "mort_5yr" colon_s %>% mutate(myweights = runif(dim(colon_s)[1])) %>% # random just for example finalfit(dependent, explanatory, weights = "myweights") -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
Note ff_formula()
convenience function to make multivariable formula (y ~ x1 + x2 + x3
etc.) from a dependent
and explanatory
vector of names.
library(finalfit) library(dplyr) explanatory = c("age.factor", "sex.factor", "obstruct.factor", "perfor.factor") dependent = "mort_5yr" # All in one pipe colon_s %>% ## Crosstable summary_factorlist(dependent, explanatory, fit_id=TRUE) %>% ## Add univariable ff_merge( glmuni(colon_s, dependent, explanatory) %>% fit2df(estimate_suffix=" (univariable)") ) %>% ## Add multivariable ff_merge( glm( ff_formula(dependent, explanatory), data = colon_s, family = "binomial", weights = NULL ) %>% fit2df(estimate_suffix=" (multivariable)"), last_merge = TRUE ) %>% dependent_label(colon_s, dependent) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
This can be done as any dataframe would be edited.
library(finalfit) library(dplyr) explanatory = c("age.factor*sex.factor", "obstruct.factor", "perfor.factor") dependent = "mort_5yr" # Run model for term test fit <- glm( ff_formula(dependent, explanatory), data=colon_s, family = binomial ) # Not run #term_test <- survey::regTermTest(fit, "age.factor:sex.factor") # Run final table with results of term test colon_s %>% finalfit(dependent, explanatory) %>% rbind(c( "age.factor:sex.factor (overall)", "Interaction", "-", "-", "-", paste0("p = 0.775") ))-> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r"))
This was an email enquiry about how to build on a base model. The example request was in a survival context.
This has been updated August 2019. We have left the original example of building the table from scratch as a comparison.
ff_permute()
allows combinations of variables to be built on a base model. See options on help page to,
library(dplyr) mydata = colon_s explanatory_base = c("age.factor", "sex.factor") explanatory_permute = c("obstruct.factor", "perfor.factor", "node4.factor") dependent = "Surv(time, status)" mydata %>% ff_permute(dependent, explanatory_base, explanatory_permute) %>% rename("Overall survival" = `Dependent: Surv(time, status)`, # optional tidying `n (%)` = "all") -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r", "r", "r"))
library(finalfit) library(dplyr) mydata = colon_s base_explanatory = c("age.factor", "sex.factor") explanatory = c("obstruct.factor", "perfor.factor", "node4.factor") dependent = "Surv(time, status)" mydata %>% # Counts summary_factorlist(dependent, c(base_explanatory, explanatory), column = TRUE, fit_id = TRUE) %>% # Univariable ff_merge( coxphuni(mydata, dependent, c(base_explanatory, explanatory)) %>% fit2df(estimate_suffix = " (Univariable)") ) %>% # Base ff_merge( coxphmulti(mydata, dependent, base_explanatory) %>% fit2df(estimate_suffix = " (Base model)") ) %>% # Model 1 ff_merge( coxphmulti(mydata, dependent, c(base_explanatory, explanatory[1])) %>% fit2df(estimate_suffix = " (Model 1)") ) %>% # Model 2 ff_merge( coxphmulti(mydata, dependent, c(base_explanatory, explanatory[2])) %>% fit2df(estimate_suffix = " (Model 2)") ) %>% # Model 3 ff_merge( coxphmulti(mydata, dependent, c(base_explanatory, explanatory[3])) %>% fit2df(estimate_suffix = " (Model 3)") ) %>% # Full ff_merge( coxphmulti(mydata, dependent, c(base_explanatory, explanatory)) %>% fit2df(estimate_suffix = " (Full)"), last_merge = TRUE ) %>% # Tidy-up rename("Overall survival" = label) %>% rename(" " = levels) %>% rename(`n (%)` = all) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r", "r", "r"))
library(survey)
Examples taken from survey::svyglm()
help page.
library(survey) library(dplyr) data(api) dependent = "api00" explanatory = c("ell", "meals", "mobility") # Label data frame apistrat = apistrat %>% mutate( api00 = ff_label(api00, "API in 2000 (api00)"), ell = ff_label(ell, "English language learners (percent)(ell)"), meals = ff_label(meals, "Meals eligible (percent)(meals)"), mobility = ff_label(mobility, "First year at the school (percent)(mobility)"), sch.wide = ff_label(sch.wide, "School-wide target met (sch.wide)") ) # Linear example dependent = "api00" explanatory = c("ell", "meals", "mobility") # Stratified design dstrat = svydesign(id=~1,strata=~stype, weights=~pw, data=apistrat, fpc=~fpc) # Univariable fit fit_uni = dstrat %>% svyglmuni(dependent, explanatory) %>% fit2df(estimate_suffix = " (univariable)") # Multivariable fit fit_multi = dstrat %>% svyglmmulti(dependent, explanatory) %>% fit2df(estimate_suffix = " (multivariable)") # Pipe together apistrat %>% summary_factorlist(dependent, explanatory, fit_id = TRUE) %>% ff_merge(fit_uni) %>% ff_merge(fit_multi, last_merge = TRUE) %>% dependent_label(apistrat, dependent) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r", "r", "r"))
Note model family needs specified and exponentiation set to TRUE
if desired.
library(survey) library(dplyr) data(api) dependent = "sch.wide" explanatory = c("ell", "meals", "mobility") # Label data frame apistrat = apistrat %>% mutate( api00 = ff_label(api00, "API in 2000 (api00)"), ell = ff_label(ell, "English language learners (percent)(ell)"), meals = ff_label(meals, "Meals eligible (percent)(meals)"), mobility = ff_label(mobility, "First year at the school (percent)(mobility)"), sch.wide = ff_label(sch.wide, "School-wide target met (sch.wide)") ) # Univariable fit fit_uni = dstrat %>% svyglmuni(dependent, explanatory, family = "quasibinomial") %>% fit2df(exp = TRUE, estimate_name = "OR", estimate_suffix = " (univariable)") # Multivariable fit fit_multi = dstrat %>% svyglmmulti(dependent, explanatory, family = "quasibinomial") %>% fit2df(exp = TRUE, estimate_name = "OR", estimate_suffix = " (multivariable)") # Pipe together apistrat %>% summary_factorlist(dependent, explanatory, fit_id = TRUE) %>% ff_merge(fit_uni) %>% ff_merge(fit_multi, last_merge = TRUE) %>% dependent_label(apistrat, dependent) -> t
library(knitr) kable(t, row.names=FALSE, align = c("l", "l", "r", "r", "r", "r", "r", "r", "r", "r"))
Any scripts or data that you put into this service are public.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.