JAMA retraction after miscoding – new Finalfit function to check recoding

This post was originally published here

Riinu and I are sitting in Frankfurt airport discussing the paper retracted in JAMA this week.

During analysis, the treatment variable coded [1,2] was recoded in error to [1,0]. The results of the analysis were therefore reversed. The lung-disease self-management program actually resulted in more attendances at hospital, rather than fewer as had been originally reported.  

Recode check

Checking of recoding is such an important part of data cleaning – we emphasise this a lot in HealthyR courses – but of course mistakes happen.

Our standard approach is this:

library(finalfit)
colon_s %>%
  mutate(
    sex.factor2 = forcats::fct_recode(sex.factor,
      "F" = "Male",
      "M" = "Female")
  ) %>%
  count(sex.factor, sex.factor2)
# A tibble: 2 x 3
  sex.factor sex.factor2     n
  <fct>      <fct>       <int>
1 Female     M             445
2 Male       F             484

The miscode should be obvious.

check_recode()

However, mistakes may still happen and be missed. So we’ve bashed out a useful function that can be applied to your whole dataset. This is not to replace careful checking, but may catch something that has been missed. 

The function takes a data frame or tibble and fuzzy matches variable names. It produces crosstables similar to above for all matched variables. 

So if you have coded something from sex to sex.factor it will be matched. The match is hungry so it is more likely to match unrelated variables than to miss similar variables. But if you recode death to mortality it won’t be matched. 

Here’s a walk through.

# Install
devtools::install_github('ewenharrison/finalfit')
library(finalfit)
library(dplyr)
# Recode example
colon_s_small = colon_s %>%
  select(-id, -rx, -rx.factor) %>%
  mutate(
    age.factor2 = forcats::fct_collapse(age.factor,
      "<60 years" = c("<40 years", "40-59 years")),
    sex.factor2 = forcats::fct_recode(sex.factor,
    # Intentional miscode
      "F" = "Male",
      "M" = "Female")
  )
# Check
colon_s_small %>%
  check_recode()
$index
# A tibble: 3 x 2
  var1        var2       
  <chr>       <chr>      
1 sex.factor  sex.factor2
2 age.factor  age.factor2
3 sex.factor2 age.factor2
$counts
$counts[[1]]
# A tibble: 2 x 3
  sex.factor sex.factor2     n
  <fct>      <fct>       <int>
1 Female     M             445
2 Male       F             484
$counts[[2]]
# A tibble: 3 x 3
  age.factor  age.factor2     n
  <fct>       <fct>       <int>
1 <40 years   <60 years      70
2 40-59 years <60 years     344
3 60+ years   60+ years     515
$counts[[3]]
# A tibble: 4 x 3
  sex.factor2 age.factor2     n
  <fct>       <fct>       <int>
1 M           <60 years     204
2 M           60+ years     241
3 F           <60 years     210
4 F           60+ years     274

As can be seen, the output takes the form of a list length 2. The first is an index of matched variables. The second is crosstables as tibbles for each variable combination. sex.factor2 can be seen as being miscoded. sex.factor2 and age.factor2 have been matched, but should be ignored.

Numerics are not included by default. To do so:

out = colon_s_small %>%
  select(-extent, -extent.factor,-time, -time.years) %>% # choose to exclude variables
  check_recode(include_numerics = TRUE)
out
# Output not printed for space

Miscoding in survival::colon dataset?

When doing this just today, we noticed something strange in our example dataset, survival::colon.

The variable node4 should be a binary recode of nodes greater than 4. But as can be seen, something is not right!

We’re interested in any explanations those working with this dataset might have.

# Select a tibble and expand
out$counts[[9]] %>%
  print(n = Inf)
# Compressed output shown
# A tibble: 32 x 3
   nodes node4     n
   <dbl> <dbl> <int>
 1     0     0     2
 2     1     0   269
 3     1     1     5
 4     2     0   194
 5     3     0   124
 6     3     1     1
 7     4     0    81
 8     4     1     3
 9     5     0     1
10     5     1    45
# … with 22 more rows

There we are then, a function that may be useful in detecting miscoding. So useful in fact, that we have immediately found probable miscoding in a standard R dataset.

Fun with Regression

“All models are wrong, but some are useful”

George Box

This quote by statistician George Box feels like a good starting point from which to consider some of the challenges of regression modelling.  If we start with the idea that all models are wrong, it follows that one of the main skills in carrying out regression modelling is working out where the weaknesses are and how to minimise these to produce as close an approximation as possible to the data you are working with – to make the model useful.

The idea that producing high-quality regression models is often more of an art than a science appeals to me.  Understanding the underlying data, what you want to explore, and the tools you have at hand are essential parts of this process.

After attending the excellent HealthyR+: Practical Logistic Regression course a few weeks ago, my head was buzzing with probabilities, odds ratios and confounding.  It was not just the data which was confounded.  As someone fairly new to logistic regression, I thought it might be useful to jot down some of the areas I found particularly interesting and concepts which made me want to find out more.  In this first blog post we take a brief look at:

  • Probability and odds
    • The difference between probability and odds
    • Why use log(odds) and not just odds?
    • Famous probability problems
  • Collinearity and correlation
    • What is collinearity?
    • How do we detect collinearity?
    • Is collinearity a problem?

Probability and odds

The difference between probability and odds

Odds and probability are both measures of how likely it is that a certain outcome might occur in a series of events.  Probability is perhaps more intuitive to understand, but its properties make it less useful in statistical models and so odds, odds ratios, and log(odds) are used instead, more on this in the next section.

Interestingly, when the probability of an event occurring is small – <0.1 (or less than 10%) – the odds are quite similar.  However, as probability increases, the odds also increase but at a greater rate, see the following figure:

Here we can also see that whilst probabilities range from 0 to 1, odds can take on any value between 0 and infinity.

Why use log(odds) and not just odds?

Asymmetry of the odds scale makes it difficult to compare binary outcomes, but by using log(odds) we can produce a symmetrical scale, see figure below:

In logistic regression, the odds ratio concerning a particular variable represents the change in odds with each unit increase, whilst holding all other variables constant.

Famous probability problems

I find probability problems fascinating, particularly those which seem counter-intuitive. Below are links to explanations of two intriguing probability problems:

Collinearity and correlation

What is collinearity?

The term collinearity (also referred to as multicollinearity) is used to describe a high correlation between two explanatory variables.  This can cause problems in regression modelling because the explanatory variables are assumed to be independent (and indeed are sometimes called independent variables, see word clouds below). 

The inclusion of variables which are collinear (highly correlated) in a regression model, can lead to the false impression for example, that neither variable is associated with the outcome, when in fact, individually each variable does have a strong association.  The figure below might help to visualise the relationships between the variables:

In this image, y represents the control variable, and x1 and x2 are the highly correlated, collinear explanatory variables.  As you can see, there is a large area of (light grey) overlap between the x variables, whereas there are only two very small areas of independent overlap between each x and y variable.  These small areas represent the limited information available to the regression model when trying to carry out analysis.

How do we detect collinearity?

A regression coefficient can be thought of as the rate of change, or as the slope of the regression line.  The slope describes the mean change in the outcome variable for every unit of change in the explanatory variable.  It is important to note that regression coefficients are calculated based on the assumption that all other variables (apart from the variables of interest) are kept constant. 

When two variables are highly correlated, this creates problems. The model will try to predict the outcome but finds it hard to disentangle the influence of either of the explanatory variables due to their strong correlation. As a result, coefficient estimates may change erratically in response to small changes in the model.

Various terms are used to describe these x and y variables depending on context.  There are slight differences in the meanings, but here are a few terms that you might encounter:

The information I used to generate these word clouds was based on a crude estimate of the number of mentions in Google Scholar within the context of medical statistics.

Is collinearity a problem?

Collinearity is a problem if the purpose of your analysis is to explain the interactions between the data, however it has little effect on the overall predictive properties of your model, i.e. the model will provide accurate predictions based on all variables as one big bundle, but will not be able to tell you about the interactions of isolated variables.

If you are concerned with exploring specific interactions and you encounter collinearity, there are two main approaches you can take:

  • Drop one of the variables if it is not vital to your analysis
  • Combine the variables (e.g. weight and height can be combined to produce BMI)

An example of a publication where missed collinearity led to potentially erroneous conclusions, concerns analyses carried out on data relating to the World Trade Organisation (WTO). Here is a related article which attempts to unpick some of the problems with previous WTO research.

Finishing on an example of a problematic attempt at regression analysis may perhaps seem slightly gloomy, but on the contrary, I hope that this might provide comfort if your own analysis throws up challenges or problems – you are in good company!  It also brings us back to the quote by George Box at the beginning of this blog post, where we started with the premise that all models are wrong.  They are at best a close approximation, and we must always be alert to their weaknesses.

What next?

Look out for the next HealthyR+: Practical Logistic Regression course and sign up.  What areas of medical statistics do you find fun, puzzling, tricky, surprising? Let us know below.

Multiple imputation support in Finalfit

This post was originally published here

We are using multiple imputation more frequently to “fill in” missing data in clinical datasets. Multiple datasets are created, models run, and results pooled so conclusions can be drawn.

We’ve put some improvements into Finalfit on GitHub to make it easier to use with the mice package. These will go to CRAN soon but not immediately.

See finalfit.org/missing.html for more on handling missing data.

Let’s get straight to it by imputing smoking status in a cancer dataset.

Install

devtools::install_github("ewenharrison/finalfit")
library(finalfit)
library(dplyr)

Create missing data for example

# Smoking missing completely at random

set.seed(1)

colon_s = colon_s %>% 
  mutate(
    smoking_mcar = sample(c("Smoker", "Non-smoker", NA), 
      dim(colon_s)[1], replace=TRUE, 
      prob = c(0.2, 0.7, 0.1)) %>% 
    factor() %>% 
    ff_label("Smoking (MCAR)")
    )

# Smoking missing conditional on patient sex
colon_s$smoking_mar[colon_s$sex.factor == "Female"] = 
  sample(c("Smoker", "Non-smoker", NA), 
    sum(colon_s$sex.factor == "Female"), 
    replace = TRUE,
    prob = c(0.1, 0.5, 0.4)
  )

colon_s$smoking_mar[colon_s$sex.factor == "Male"] = 
  sample(c("Smoker", "Non-smoker", NA), 
    sum(colon_s$sex.factor == "Male"), 
    replace=TRUE, prob = c(0.15, 0.75, 0.1)
  )
 
colon_s = colon_s %>% 
  mutate(
    smoking_mar = factor(smoking_mar) %>% 
    ff_label("Smoking (MAR)")
  )

Check data

explanatory = c("age", "sex.factor", 
  "nodes", "obstruct.factor",  
  "smoking_mcar", "smoking_mar")
 dependent = "mort_5yr"
 colon_s %>% 
  ff_glimpse(dependent, explanatory)

 Continuous
            label var_type   n missing_n missing_percent mean   sd  min quartile_25 median quartile_75  max
age   Age (years)    <dbl> 929         0             0.0 59.8 11.9 18.0        53.0   61.0        69.0 85.0
nodes       nodes    <dbl> 911        18             1.9  3.7  3.6  0.0         1.0    2.0         5.0 33.0

Categorical
                           label var_type   n missing_n missing_percent levels_n
sex.factor                   Sex    <fct> 929         0             0.0        2
obstruct.factor      Obstruction    <fct> 908        21             2.3        2
mort_5yr        Mortality 5 year    <fct> 915        14             1.5        2
smoking_mcar      Smoking (MCAR)    <fct> 828       101            10.9        2
smoking_mar        Smoking (MAR)    <fct> 719       210            22.6        2
                                             levels  levels_count   levels_percent
sex.factor                         "Female", "Male"      445, 484           48, 52
obstruct.factor            "No", "Yes", "(Missing)"  732, 176, 21 78.8, 18.9,  2.3
mort_5yr               "Alive", "Died", "(Missing)"  511, 404, 14 55.0, 43.5,  1.5
smoking_mcar    "Non-smoker", "Smoker", "(Missing)" 645, 183, 101       69, 20, 11
smoking_mar     "Non-smoker", "Smoker", "(Missing)" 591, 128, 210       64, 14, 23

Multivariate Imputation by Chained Equations (mice)

miceis a great package and contains lots of useful functions for diagnosing and working with missing data. The purpose here is to demonstrate how mice can be integrated into the Finalfit workflow with inclusion of model from imputed datasets in tables and plots.

Choose variables to impute and variables to impute from

finalfit::missing_predictorMatrix()makes it easy to specify which variables do what. For instance, we often do not want to impute our outcome or explanatory variable of interest (exposure), but do want to use them to impute other variables.

This is straightforward to code using the arguments drop_from_imputed and drop_from_imputer.

library(mice)

# Specify model
explanatory = c("age", "sex.factor", "nodes", 
  "obstruct.factor", "smoking_mar")
dependent = "mort_5yr"

# Choose not to impute missing values
# for explanatory variable of interest and
# outcome variable. 
# But include in algorithm for imputation.
predM = colon_s %>% 
	select(dependent, explanatory) %>% 
	missing_predictorMatrix(
		drop_from_imputed = c("obstruct.factor", "mort_5yr")
	)

Create imputed datasets

A set of multiple imputed datasets (mids) can be created as below. Various checks should be performed to ensure you understand the data that has been created. See here.

mids = colon_s %>% 
  select(dependent, explanatory) %>%
  mice(m = 4, predictorMatrix = predM)    # Usually m = 10

Run models

Here we sill use a logistic regression model. The with.mids() function takes a model with a formula object, so use base R functions rather than Finalfit wrappers.

fits = mids %>% 
  with(glm(formula(ff_formula(dependent, explanatory)), 
    family="binomial"))

We now have multiple models run with each of the imputed datasets. We haven’t found good methods for combining common model metrics like AIC and c-statistic. I’d be interested to hear from anyone working on methods for this. Metrics can be extracted for each individual model to give an idea of goodness-of-fit and discrimination. We’re not suggesting you use these to compare imputed datasets, but could use them to compare models containing different variables created using the imputed datasets, e.g.

fits %>% 
  getfit() %>% 
  purrr::map(AIC)
[[1]]
[1] 1192.57

[[2]]
[1] 1191.09

[[3]]
[1] 1195.49

[[4]]
[1] 1193.729

# C-statistic
fits %>% 
  getfit() %>% 
  purrr::map(~ pROC::roc(.x$y, .x$fitted)$auc)
[[1]]
Area under the curve: 0.6839

[[2]]
Area under the curve: 0.6818

[[3]]
Area under the curve: 0.6789

[[4]]
Area under the curve: 0.6836

Pool results

Rubin’s rules are used to combine results of multiple models.

# Pool  results
fits_pool = fits %>% 
  pool()

Plot results

Pooled results can be passed directly to Finalfit plotting functions.

# Can be passed to or_plot
colon_s %>% 
  or_plot(dependent, explanatory, glmfit = fits_pool, table_text_size=4)

Put results in table

The pooled result can be passed directly to fit2df() as can many common models such as lm(), glm(), lmer(), glmer(), coxph(), crr(), etc.

# Summarise and put in table
fit_imputed = fits_pool %>%                                  
  fit2df(estimate_name = "OR (multiple imputation)", exp = TRUE)
fit_imputed

         explanatory  OR (multiple imputation)
1                age 1.01 (1.00-1.02, p=0.212)
2     sex.factorMale 1.01 (0.77-1.34, p=0.917)
3              nodes 1.24 (1.18-1.31, p<0.001)
4 obstruct.factorYes 1.34 (0.94-1.91, p=0.105)
5  smoking_marSmoker 1.28 (0.88-1.85, p=0.192)

Combine results with summary data

Any model passed through fit2df() can be combined with a summary table generated with summary_factorlist() and any number of other models.

# Imputed data alone
## Include missing data in summary table
colon_s %>% 
  summary_factorlist(dependent, explanatory, na_include = TRUE, fit_id = TRUE) %>% 
  ff_merge(fit_imputed, last_merge = TRUE) 

           label     levels       Alive        Died  OR (multiple imputation)
1    Age (years)  Mean (SD) 59.8 (11.4) 59.9 (12.5) 1.01 (1.00-1.02, p=0.212)
6            Sex     Female  243 (55.6)  194 (44.4)                         -
7                      Male  268 (56.1)  210 (43.9) 1.01 (0.77-1.34, p=0.917)
2          nodes  Mean (SD)   2.7 (2.4)   4.9 (4.4) 1.24 (1.18-1.31, p<0.001)
4    Obstruction         No  408 (56.7)  312 (43.3)                         -
5                       Yes   89 (51.1)   85 (48.9) 1.34 (0.94-1.91, p=0.105)
3                   Missing   14 (66.7)    7 (33.3)                         -
9  Smoking (MAR) Non-smoker  328 (56.4)  254 (43.6)                         -
10                   Smoker   68 (53.5)   59 (46.5) 1.28 (0.88-1.85, p=0.192)
8                   Missing  115 (55.8)   91 (44.2)                         -

Combine results with other models

Models can be run separately, or using the finalfit()wrapper including the argument keep_fit_it = TRUE.

colon_s %>% 
  finalfit(dependent, explanatory, keep_fit_id = TRUE) %>% 
  ff_merge(fit_imputed, last_merge = TRUE) 

  Dependent: Mortality 5 year                  Alive        Died          OR (univariable)        OR (multivariable)  OR (multiple imputation)
1                 Age (years)  Mean (SD) 59.8 (11.4) 59.9 (12.5) 1.00 (0.99-1.01, p=0.986) 1.02 (1.00-1.03, p=0.010) 1.01 (1.00-1.02, p=0.212)
5                         Sex     Female  243 (47.6)  194 (48.0)                         -                         -                         -
6                                   Male  268 (52.4)  210 (52.0) 0.98 (0.76-1.27, p=0.889) 0.88 (0.64-1.23, p=0.461) 1.01 (0.77-1.34, p=0.917)
2                       nodes  Mean (SD)   2.7 (2.4)   4.9 (4.4) 1.24 (1.18-1.30, p<0.001) 1.25 (1.18-1.33, p<0.001) 1.24 (1.18-1.31, p<0.001)
3                 Obstruction         No  408 (82.1)  312 (78.6)                         -                         -                         -
4                                    Yes   89 (17.9)   85 (21.4) 1.25 (0.90-1.74, p=0.189) 1.26 (0.85-1.88, p=0.252) 1.34 (0.94-1.91, p=0.105)
7               Smoking (MAR) Non-smoker  328 (82.8)  254 (81.2)                         -                         -                         -
8                                 Smoker   68 (17.2)   59 (18.8) 1.12 (0.76-1.65, p=0.563) 1.25 (0.82-1.89, p=0.300) 1.28 (0.88-1.85, p=0.192)

Model missing explicitly in complete case models

A straightforward method of modelling missing cases is to make them explicit using the forcats function fct_explicit_na().

library(forcats)
colon_s %>% 
  mutate(
    smoking_mar = fct_explicit_na(smoking_mar)
  ) %>% 
  finalfit(dependent, explanatory, keep_fit_id = TRUE) %>% 
  ff_merge(fit_imputed, last_merge = TRUE)

  Dependent: Mortality 5 year                  Alive        Died          OR (univariable)        OR (multivariable)  OR (multiple imputation)
1                 Age (years)  Mean (SD) 59.8 (11.4) 59.9 (12.5) 1.00 (0.99-1.01, p=0.986) 1.01 (1.00-1.02, p=0.119) 1.01 (1.00-1.02, p=0.212)
5                         Sex     Female  243 (47.6)  194 (48.0)                         -                         -                         -
6                                   Male  268 (52.4)  210 (52.0) 0.98 (0.76-1.27, p=0.889) 0.96 (0.72-1.30, p=0.809) 1.01 (0.77-1.34, p=0.917)
2                       nodes  Mean (SD)   2.7 (2.4)   4.9 (4.4) 1.24 (1.18-1.30, p<0.001) 1.25 (1.19-1.32, p<0.001) 1.24 (1.18-1.31, p<0.001)
3                 Obstruction         No  408 (82.1)  312 (78.6)                         -                         -                         -
4                                    Yes   89 (17.9)   85 (21.4) 1.25 (0.90-1.74, p=0.189) 1.34 (0.94-1.91, p=0.102) 1.34 (0.94-1.91, p=0.105)
8               Smoking (MAR) Non-smoker  328 (64.2)  254 (62.9)                         -                         -                         -
9                                 Smoker   68 (13.3)   59 (14.6) 1.12 (0.76-1.65, p=0.563) 1.24 (0.82-1.88, p=0.308) 1.28 (0.88-1.85, p=0.192)
7                              (Missing)  115 (22.5)   91 (22.5) 1.02 (0.74-1.41, p=0.895) 0.99 (0.69-1.41, p=0.943)                         -

Export tables to PDF and Word

As described elsewhere, knitr::kable() can be used to export good looking tables.

rmedicine2019 – some quick thoughts and good packages

Kenny McLean and I recently attended rmedicine 2019 in Boston MA. The conference is aimed at clinicians and non-clinicians who use R for day-to-day research and monitoring of clinical processes.

Day 1 covered two parallel workshops: R Markdown for Medicine and Wrangling Survival Data 

I attended R Markdown for Medicine run by Alison Hill from RStudio. Using .rmd files has become the default for the Surgical Informatics Group and, so it seems, a great number of others who attended rmedicine. Around a third of the presentations at rmedicine covered workflows involving sharing of data via either .rmd files or through shiny, an R package for creating deploy-able dashboards for data visualisation and interactive exploration.

R Markdown for Medicine

An Overview of Useful Tips and Tricks

R markdown is an extension of R which allows you to combine narrative text and R code within one document. This means your notes, code, results and plots are all in one place. Code is contained in between three backticks with an {r} after the first set. Inline code can also be used between single backticks followed by r without the curly brackets and then the code. This means that results can be changed automatically so that for a trial when you describe the results of numbers included / excluded, this only needs changed in one place so that the rest of the text (and / or flowcharts) updates automatically. It is also possible to mix-and-match other chunks of code from other languages.

Use Params!

Parameters are set in the YAML header at the top of the .rmd document. If you set a parameter of data to a default .csv or .rda file then this can be changed for other similar files without creating a new document. A really useful example would be when you have multiple hospitals or multiple diseases each with a separate data file, a report can then be generated for each file. If you use rmarkdown::render() along with purrr::pwalk you can generate a separate output file for any number of hospitals / diseases / countries / individuals etc. in just a couple of lines of code.

Use Helper Packages

There are some greater .rmd helper packages to improve the workflow, improve the rendering of documents and generally make life easier.

bookdown allows several .rmd documents to be combined to a book but also has some general usefulness for single documents as well. Using bookdown::word_document2 or bookdown::html_document2 in the YAML header under the output field is designed to improve cross-referencing of tables and figures compared to the default versions.

wordcountaddin allows an accurate word count to be performed which will not count YAML or code etc. without knitting the document. This is much easier than knitting the document and then performing a word count!

citr allows automated insertion of markdown citations to assist with referencing. Check out my earlier blog on referencing to get an idea of how to set up .bib files. I may add another blog on this topic, watch this space!

xaringan is a useful package for creating HTML presentations with high levels of customisation. It is possible to use an additional .css file for even greater customisation and styling of your slides but xaringan offers a great deal of user-friendly options.

distill appears to be good at supporting mobile-friendly web publishing for scientific communication with flexible figure layouts, table pagination, LaTeX math support and incorporation of javascript.

There are countless other helper packages and more likely to be on their way. Many allow additional aesthetic modification of the output documents and may allow you to run R code rather than modifying a .css file.

List Numbering the Lazy Way

List numbering in .rmd works without needing to manually enter the correct numbers. Just make a list where every element begins with 1. and .rmd will transform it into an appropriately-numbered list. Great if you need to add in a new element to the middle of the list later!

Multiple lots in a Grid

I’ve previously come across patchwork as a way to plot several plots into a grid which could be 1×2, 2×2, 3 in one column and one in the other etc. There are also two other packages cowplot and egg. I haven’t explored the differences between them but if you find that one doesn’t give you the exact customisation or alignment you need then possibly try another one. cowplot looks as if it might perform better at overlaying plots on top of another and at exact axis line matching.

Use the here package to help with file paths

here is a great package for swapping between Windows and Mac file paths (no more swapping backslashes and forward slashes!). Using here::here() will default to looking for a file in the .Rproj directory rather than the .rmd directory which is the default otherwise – great if you want to have multiple .rmd documents each in their own sub-directory with a shared data file in the parent directory.

Customise Code Outputs

R markdown allows customisation of appearance of code. Some of this can be done through modifying a .css file but there are some simpler ways to make basic changes. Try adding comment = "#>" to knitr::opst_chunk$set()to customise how comments appear in your document.

Word document creations tips

R markdown is generally great for HTML and PDF formats. The options for knitting to Word are not as well developed but there are some good options. The bookdown package is useful as discussed. The redoc package has been used to facilitate conversion to and from word – not tried it personally but if it can print out to word and then handle tracked changes back into markdown then it could be very useful.

For converting more complex tables and figures to word an option is to knit to rtf (rich text format) and then open the rtf file in word. This tends to be very good at keeping the desired formatting.

Future updates – hopefully!

R markdown is a great resource although there are a handful of minor issues which are currently difficult to resolve. One of the main problems I find it with tables and cross-referencing. I really like the syntax and customisation of the gt package but at present it appears cross-referencing in a way which works across HTML, PDF and Word outputs is not supported – a great opportunity to submit a pull request if you think you can get this to work.

Other Useful rmedicine Packages and Ideas

survival Package Update

The latest version (version 3.0) of the survival package was presented by Terry Therneau and is now available on github. This package is used by over 650 additional downstream dependencies. The latest version allows for multiple observations per subject, multiple endpoints per subject and multiple types of end-point. This will be particularly useful for competing risks analyses e.g. outcomes for liver transplant patients (transplanted, still on list, removed from list as no longer eligible or died).

Keep an eye-out for Kenny McLean’s blog where he plans to cover the survival package and many other useful packages presented at rmedicine 2019.

hreport Automated Trial Reporting

hreport by Frank Harrel (currently available on github) is for automated reporting of trials and studies with generation of interactive html graphs based in plotly. Several aspects of a study can be rendered easily into plots demonstrating accrual, exclusions, descriptive statistics, adverse events and time-to-event data. Another key theme of rmedicine 2019 appears to have been the use of plotly or similar packages to enable interaction with data.

timevis – interactive timelines

timevis allows generation of highly interactive timeline plots which allow zooming, adding or removal of events, resizing, etc.

Holepunch package

For working with projects that require a number of packages that then need shared with a colleague, holepunch provides a quick method for generating a list of dependencies and a Dockerfile. The package creates a link for another user to open a free RStudio server with all of the required packages installed. This may be useful for trouble-shooting in a department and showing code examples.

Summary

rmedicine 2019 has shown that clinical researchers are moving increasingly towards literate programming, interactive visualisations and automated workflows using R and Rmarkdown.

The conference was a great mix of methods presentations and data presentations from R users. You definitely don’t need any in-depth knowledge of R to benefit from it and I’d highly recommend booking for rmedicine 2020.

Survival analysis with strata, clusters, frailties and competing risks in in Finalfit

This post was originally published here

Background

In healthcare, we deal with a lot of binary outcomes. Death yes/no, disease recurrence yes/no, for instance. These outcomes are often easily analysed using binary logistic regression via finalfit().

When the time taken for the outcome to occur is important, we need a different approach. For instance, in patients with cancer, the time taken until recurrence of the cancer is often just as important as the fact it has recurred.

Finalfit wraps a number of functions to make these analyses easy to perform and output into PDFs and Word documents.

Installation

# Make sure finalfit is up-to-date 
install.packages("finalfit")

Dataset

We’ll use the classic “Survival from Malignant Melanoma” dataset from the boot package to illustrate. The data consist of measurements made on patients with malignant melanoma. Each patient had their tumour removed by surgery at the Department of Plastic Surgery, University Hospital of Odense, Denmark during the period 1962 to 1977.

For the purposes of demonstration, we are interested in the association between tumour ulceration and survival after surgery.

Get data and check

library(finalfit)
melanoma = boot::melanoma #F1 here for help page with data dictionary
ff_glimpse(melanoma)
#> Continuous
#>               label var_type   n missing_n missing_percent   mean     sd
#> time           time    <dbl> 205         0             0.0 2152.8 1122.1
#> status       status    <dbl> 205         0             0.0    1.8    0.6
#> sex             sex    <dbl> 205         0             0.0    0.4    0.5
#> age             age    <dbl> 205         0             0.0   52.5   16.7
#> year           year    <dbl> 205         0             0.0 1969.9    2.6
#> thickness thickness    <dbl> 205         0             0.0    2.9    3.0
#> ulcer         ulcer    <dbl> 205         0             0.0    0.4    0.5
#>              min quartile_25 median quartile_75    max
#> time        10.0      1525.0 2005.0      3042.0 5565.0
#> status       1.0         1.0    2.0         2.0    3.0
#> sex          0.0         0.0    0.0         1.0    1.0
#> age          4.0        42.0   54.0        65.0   95.0
#> year      1962.0      1968.0 1970.0      1972.0 1977.0
#> thickness    0.1         1.0    1.9         3.6   17.4
#> ulcer        0.0         0.0    0.0         1.0    1.0
#> 
#> Categorical
#> data frame with 0 columns and 205 rows

As can be seen, all variables are coded as numeric and some need recoding to factors.

Death status

status is the the patients status at the end of the study.

  • 1 indicates that they had died from melanoma;
  • 2 indicates that they were still alive and;
  • 3 indicates that they had died from causes unrelated to their melanoma.

There are three options for coding this.

  • Overall survival: considering all-cause mortality, comparing 2 (alive) with 1 (died melanoma)/3 (died other);
  • Cause-specific survival: considering disease-specific mortality comparing 2 (alive)/3 (died other) with 1 (died melanoma);
  • Competing risks: comparing 2 (alive) with 1 (died melanoma) accounting for 3 (died other); see more below.

Time and censoring

time is the number of days from surgery until either the occurrence of the event (death) or the last time the patient was known to be alive. For instance, if a patient had surgery and was seen to be well in a clinic 30 days later, but there had been no contact since, then the patient’s status would be considered 30 days. This patient is censored from the analysis at day 30, an important feature of time-to-event analyses.

Recode

library(dplyr)
library(forcats)
melanoma = melanoma %>%
  mutate(
    # Overall survival
    status_os = case_when(
      status == 2 ~ 0, # "still alive"
      TRUE ~ 1), # "died melanoma" or "died other causes"
    
    # Diease-specific survival
    status_dss = case_when(
      status == 2 ~ 0,  # "still alive"
      status == 1 ~ 1,  # "died of melanoma"
      status == 3 ~ 0), # "died of other causes is censored"

    # Competing risks regression
    status_crr = case_when(
    	status == 2 ~ 0,  # "still alive"
        status == 1 ~ 1,  # "died of melanoma"
        status == 3 ~ 2), # "died of other causes"

    # Label and recode other variables
    age = ff_label(age, "Age (years)"), # table friendly labels
    thickness = ff_label(thickness, "Tumour thickness (mm)"),
    sex = factor(sex) %>% 
      fct_recode("Male" = "1", 
                 "Female" = "0") %>% 
      ff_label("Sex"),
    ulcer = factor(ulcer) %>% 
      fct_recode("No" = "0",
                 "Yes" = "1") %>% 
      ff_label("Ulcerated tumour")
  )

Kaplan-Meier survival estimator

We can use the excellent survival package to produce the Kaplan-Meier (KM) survival estimator. This is a non-parametric statistic used to estimate the survival function from time-to-event data. Note use of %$% to expose left-side of pipe to older-style R functions on right-hand side.

library(survival)

survival_object = melanoma %$% 
  Surv(time, status_os)

# Explore:
head(survival_object) # + marks censoring, in this case "Alive"
#> [1]  10   30   35+  99  185  204

# Expressing time in years
survival_object = melanoma %$% 
  Surv(time/365, status_os)

KM analysis for whole cohort

Model

The survival object is the first step to performing univariable and multivariable survival analyses.

If you want to plot survival stratified by a single grouping variable, you can substitute “survival_object ~ 1” by “survival_object ~ factor”

# Overall survival in whole cohort
my_survfit = survfit(survival_object ~ 1, data = melanoma)
my_survfit # 205 patients, 71 events
#> Call: survfit(formula = survival_object ~ 1, data = melanoma)
#> 
#>       n  events  median 0.95LCL 0.95UCL 
#>  205.00   71.00      NA    9.15      NA

Life table

A life table is the tabular form of a KM plot, which you may be familiar with. It shows survival as a proportion, together with confidence limits. The whole table is shown with summary(my_survfit).

summary(my_survfit, times = c(0, 1, 2, 3, 4, 5))
#> Call: survfit(formula = survival_object ~ 1, data = melanoma)
#> 
#>  time n.risk n.event survival std.err lower 95% CI upper 95% CI
#>     0    205       0    1.000  0.0000        1.000        1.000
#>     1    193      11    0.946  0.0158        0.916        0.978
#>     2    183      10    0.897  0.0213        0.856        0.940
#>     3    167      16    0.819  0.0270        0.767        0.873
#>     4    160       7    0.784  0.0288        0.730        0.843
#>     5    122      10    0.732  0.0313        0.673        0.796
# 5 year overall survival is 73%

Kaplan Meier plot

We can plot survival curves using the finalfit wrapper for the package excellent package survminer. There are numerous options available on the help page. You should always include a number-at-risk table under these plots as it is essential for interpretation.

As can be seen, the probability of dying is much greater if the tumour was ulcerated, compared to those that were not ulcerated.

dependent_os = "Surv(time/365, status_os)"
explanatory = "ulcer"

melanoma %>% 
  surv_plot(dependent_os, explanatory, pval = TRUE)

Cox-proportional hazards regression

CPH regression can be performed using the all-in-one finalfit() function. It produces a table containing counts (proportions) for factors, mean (SD) for continuous variables and a univariable and multivariable CPH regression.

A hazard is the term given to the rate at which events happen.
The probability that an event will happen over a period of time is the hazard multiplied by the time interval.
An assumption of CPH is that hazards are constant over time (see below).

It produces a table containing counts (proportions) for factors, mean (SD) for continuous variables and a univariable and multivariable CPH regression.

Univariable and multivariable models

dependent_os = "Surv(time, status_os)"
dependent_dss = "Surv(time, status_dss)"
dependent_crr = "Surv(time, status_crr)"
explanatory = c("age", "sex", "thickness", "ulcer")

melanoma %>% 
    finalfit(dependent_os, explanatory)

The labelling of the final table can be easily adjusted as desired.

melanoma %>% 
    finalfit(dependent_os, explanatory, add_dependent_label = FALSE) %>% 
    rename("Overall survival" = label) %>% 
    rename(" " = levels) %>% 
    rename(" " = all)

Reduced model

If you are using a backwards selection approach or similar, a reduced model can be directly specified and compared. The full model can be kept or dropped.

explanatory_multi = c("age", "thickness", "ulcer")
melanoma %>% 
    finalfit(dependent_os, explanatory, explanatory_multi, 
      keep_models = TRUE)

Testing for proportional hazards

An assumption of CPH regression is that the hazard associated with a particular variable does not change over time. For example, is the magnitude of the increase in risk of death associated with tumour ulceration the same in the early post-operative period as it is in later years.

The cox.zph() function from the survival package allows us to test this assumption for each variable. The plot of scaled Schoenfeld residuals should be a horizontal line. The included hypothesis test identifies whether the gradient differs from zero for each variable. No variable significantly differs from zero at the 5% significance level.

explanatory = c("age", "sex", "thickness", "ulcer", "year")
melanoma %>% 
    coxphmulti(dependent_os, explanatory) %>% 
    cox.zph() %>% 
    {zph_result <<- .} %>% 
    plot(var=5)
zph_result
#>               rho  chisq      p
#> age        0.1633 2.4544 0.1172
#> sexMale   -0.0781 0.4473 0.5036
#> thickness -0.1493 1.3492 0.2454
#> ulcerYes  -0.2044 2.8256 0.0928
#> year       0.0195 0.0284 0.8663
#> GLOBAL         NA 8.4695 0.1322

Stratified models

One approach to dealing with a violation of the proportional hazards assumption is to stratify by that variable. Including a strata() term will result in a separate baseline hazard function being fit for each level in the stratification variable. It will be no longer possible to make direct inference on the effect associated with that variable.

This can be incorporated directly into the explanatory variable list.

explanatory= c("age", "sex", "ulcer", "thickness", "strata(year)")
melanoma %>% 
    finalfit(dependent_os, explanatory)

Correlated groups of observations

As a general rule, you should always try to account for any higher structure in the data within the model. For instance, patients may be clustered within particular hospitals.

There are two broad approaches to dealing with correlated groups of observations.

Including a cluster() term is akin to using generalised estimating equations (GEE). Here, a standard CPH model is fitted but the standard errors of the estimated hazard ratios are adjusted to account for correlations.

Including a frailty() term is akin to using a mixed effects model, where specific random effects term(s) are directly incorporated into the model.

Both approaches achieve the same goal in different ways. Volumes have been written on GEE vs mixed effects models. We favour the latter approach because of its flexibility and our preference for mixed effects modelling in generalised linear modelling. Note cluster() and frailty() terms cannot be combined in the same model.

# Simulate random hospital identifier
melanoma = melanoma %>% 
  mutate(hospital_id = c(rep(1:10, 20), rep(11, 5)))

# Cluster model
explanatory = c("age", "sex", "thickness", "ulcer", "cluster(hospital_id)")
melanoma %>% 
  finalfit(dependent_os, explanatory)
# Frailty model
explanatory = c("age", "sex", "thickness", "ulcer", "frailty(hospital_id)")
melanoma %>% 
  finalfit(dependent_os, explanatory)

The frailty() method here is being superseded by the coxme package, and we’ll incorporate this soon.

Hazard ratio plot

A plot of any of the above models can be produced by passing the terms to hr_plot().

melanoma %>% 
    hr_plot(dependent_os, explanatory)

Competing risks regression

Competing-risks regression is an alternative to CPH regression. It can be useful if the outcome of interest may not be able to occur because something else (like death) has happened first. For instance, in our example it is obviously not possible for a patient to die from melanoma if they have died from another disease first. By simply looking at cause-specific mortality (deaths from melanoma) and considering other deaths as censored, bias may result in estimates of the influence of predictors.

The approach by Fine and Gray is one option for dealing with this. It is implemented in the package cmprsk. The crr() syntax differs from survival::coxph() but finalfit brings these together.

It uses the finalfit::ff_merge() function, which can join any number of models together.

explanatory = c("age", "sex", "thickness", "ulcer")
dependent_dss = "Surv(time, status_dss)"
dependent_crr = "Surv(time, status_crr)"

melanoma %>%

  # Summary table
  summary_factorlist(dependent_dss, explanatory, 
    column = TRUE, fit_id = TRUE) %>%

  # CPH univariable
  ff_merge(
    melanoma %>%
      coxphmulti(dependent_dss, explanatory) %>%
      fit2df(estimate_suffix = " (DSS CPH univariable)")
    ) %>%
    
# CPH multivariable
  ff_merge(
    melanoma %>%
      coxphmulti(dependent_dss, explanatory) %>%
      fit2df(estimate_suffix = " (DSS CPH multivariable)")
    ) %>%
    
# Fine and Gray competing risks regression
  ff_merge(
    melanoma %>%
      crrmulti(dependent_crr, explanatory) %>%
      fit2df(estimate_suffix = " (competing risks multivariable)")
    ) %>%

  select(-fit_id, -index) %>%
  dependent_label(melanoma, "Survival")

Summary

So here we have various aspects of time-to-event analysis commonly used when looking at survival. There are many other applications, some which may not be obvious: for instance we use CPH for modelling length of stay in in hospital.

Stratification can be used to deal with non-proportional hazards in a particular variable.

Hierarchical structure in your data can be accommodated with cluster or frailty (random effects) terms.

Competing risks regression may be useful if your outcome is in competition with another, such as all-cause death, but is currently limited in its ability to accommodate hierarchical structures.

HealthyR Notebooks Estonia Day 2

Everybody came back for Day 2 of HealthyR Notebooks in Estonia!

Today focussed on modelling kicking off with linear regression in detail from Riinu – if you understand this you understand the majority of statistical tests!

Factors were introduced by Cameron, which led nicely into logistic regression. By the end of this the whole room was comfortably building regression models as if they had been doing it for years!

Notebooks are a really powerful tool for teaching this sort of material allowing seamless output into PDF and Word format. Some of the delegates commented on how they had struggled with this aspect of R in the past.

After all the intense work it was a great relief to break early for some fantastic team building including stuff like this!

Some even tried there hand at archery and felt pretty smug about their performance 😂

Finally, we are so grateful Julius Juurmaa for all of the organisation. He even started a company to administer the course! Thank you Julius.

HealthyR Notebooks Estonia Day 1

The Surgical Informatics team arrived in beautiful Estonia yesterday.

We are here as part of our Wellcome Trust Open Research Fund grant – “HealthyR Notebooks: Democratising open and reproducible data analysis in resource-poor environments”.

It’s a mouthful, but important! We have adapted our popular HealthyR training course to be easily delivered on small laptop screens allowing state-of-the-art data analysis to be performed using RStudio anywhere by anyone. We’re testing this in Estonia and running it again in Ghana in November.

Also look out for HealthyR the book coming soon.

The setting is fantastic:

The delegates are already amazing at using R, but we’re teaching Tidyverse and Finalfit to bring everyone up to date with all the great new modern packages.

There are threats of pedalo racing later 😳.

Encryptr now makes it easy to encrypt and decrypt files

This post was originally published here

Data security is paramount and encryptr was written to make this easier for non-experts. Columns of data can be encrypted with a couple of lines of R code, and single cells decrypted as required.

But what was missing was an easy way to encrypt the file source of that data.

Now files can be encrypted with a couple of lines of R code.

Encryption and decryption with asymmetric keys is computationally expensive. This is how encrypt for data columns works. This makes it easy for each piece of data in a data frame to be decrypted without compromise of the whole data frame. This works on the presumption that each cell contains less than 245 bytes of data.

File encryption requires a different approach as files are larger in size. encrypt_file encrypts a file using a symmetric “session” key and the AES-256 cipher. This key is itself then encrypted using a public key generated using genkeys. In OpenSSL this combination is referred to as an envelope.

It should work with any type of single file but not folders.

Documentation is maintained at encrypt-r.org

Generate keys

genkeys()
#> Private key written with name 'id_rsa'
#> Public key written with name 'id_rsa.pub'

Encrypt file

To demonstrate, the included dataset is written as a .csv file.

write.csv(gp, "gp.csv")
encrypt_file("gp.csv")
#> Encrypted file written with name 'gp.csv.encryptr.bin'

Important: check that the file can be decrypted prior to removing the original file from your system.

Warning: it is strongly suggested that the original unencrypted data file is securely stored else where as a back-up in case unencryption is not possible, e.g., the private key file or password is lost

Decrypt file

The decrypt_file function will not allow the original file to be overwritten, therefore if it is still present, use the option to specify a new name for the unencrypted file.

decrypt_file("gp.csv.encryptr.bin", file_name = "gp2.csv")
#> Decrypted file written with name 'gp2.csv'

Support / bugs

The new version 0.1.3 is on its way to CRAN today or you can install from github:

github.com/SurgicalInformatics/encryptr

Encryptr package: easily encrypt and decrypt columns of sensitive data

This post was originally published here

A number of existing R packages support data encryption. However, we haven’t found one that easily suits our needs: to encrypt one or many columns of a data frame or tibble using a private/public key pair in tidyversefunctions. The emphasis is on the easily.

Encrypting and decrypting data securely is important when it comes to healthcare and sociodemographic data. We have developed a simple and secure package encryptyr which allows non-experts to encrypt and decrypt columns of data.

There is a simple and easy-to-follow vignette available on our GitHub page which guides you through the process of using encryptr:

https://github.com/SurgicalInformatics/encryptr.

Confidential data – security challenges

Data containing columns of disclosive or confidential information such as a postcode or a patient ID (CHI in Scotland) require extreme care. Storing sensitive information as raw values leaves the data vulnerable to confidentiality breaches.

It is best to just remove confidential information from the records whenever possible. However, this can mean the data can never be re-associated with an individual. This may be a problem if, for example, auditors of a clinical trial need to re-identify an individual from the trial data.

One potential solution currently in common use is to generate a study number which is linked to the confidential data in a separate lookup table, but this still leaves the confidential data available in another file.

Encryptr package solution – storing encrypted data

The encryptr package allows users to store confidential data in a pseudoanonymised form, which is far less likely to result in re-identification.

The package allows users to create a public key and a private key to enable RSA encryption and decryption of the data. The public key allows encryption of the data. The private key is required to decrypt the data. The data cannot be decrypted with the public key. This is the basis of many modern encryption systems.

When creating keys, the user sets a password for the private key using a dialogue box. This means that the password is included in an R script. We recommend creating a secure password with a variety of alphanumeric characters and symbols.

As the password is not stored, it is important that you are able to remember it if you need to decrypt the data later.

Once the keys are created it is possible to encrypt one or more columns of data in a data frame or tibble using the public key. Every time RSA encryption is used it will generate a unique output. Even if the same information is encrypted more than once, the output will always be different. It is not possible therefore to match two encrypted values.

These outputs are also secure from decryption without the private key. This may allow sharing of data within or between research teams without sharing confidential data.

Caution: data often remains potentially disclosive (or only pseudoanomymised) even after encryption of identifiable variables and all of the required permissions for usage and sharing of data must still be in place.

Encryptr package – decrypting the data

Sometimes decrypting data is necessary. For example, participants in a clinical trial may need to be contacted to explain a change or early termination of the trial.

The encryptr package allows users to securely and reliably decrypt the data. The decrypt function will use the private key to decrypt one or more columns. The user will be required to enter the password created when the keys were generated.

As the private key is able to decrypt all of the data, we do not recommend sharing this key.

Blinding and unblinding clinical trials – another encryptr package use

Often when working with clinical trial data, the participants are randomised to one or more treatment groups. Often teams working on the trial are unaware of the group to which patients were randomised (blinded).

Using the same method of encryption, it is possible to encrypt the participant allocation group, allowing the sharing of data without compromising blinding. If other members of the trial team are permitted to see treatment allocation (unblinded), then the decryption process can be followed to reveal the group allocation.

What this is not

This is a simple set of wrappers of openssl aimed at non-experts. It does not seek to replace the many excellent encryption packages available in R, such as PKI, sodium and safer. We believe however that it makes things much easier. Comments and forks welcome.