Package 'TreeMineR'

Title: Tree-Based Scan Statistics
Description: Implementation of unconditional Bernoulli Scan Statistic developed by Kulldorff et al. (2003) <doi:10.1111/1541-0420.00039> for hierarchical tree structures. Tree-based Scan Statistics are an exploratory method to identify event clusters across the space of a hierarchical tree.
Authors: Joshua P. Entrop [aut, cre, cph] , Viktor Wintzell [aut]
Maintainer: Joshua P. Entrop <[email protected]>
License: GPL (>= 3)
Version: 1.0.2
Built: 2025-03-11 05:47:22 UTC
Source: https://github.com/entjos/treeminer

Help Index

Hierarchical tree of the ATC system for classifying drugs

Description

A dataset including the following column:

pathString

A string identifying all the parents of a node. Each parent is separated by a /.

Usage

data(atc_codes)

Creating a tree file for further use in TreeMineR().

Description

Creating a tree file for further use in TreeMineR().

Usage

create_tree(x)

Arguments

x

A data frame that includes two or three columns:

node

A string defining a node

parent

A string defining the partent of the node

Value

A data.frame with one variable pathString that describes the full path for each leaf included in the hierarchical tree.

Test dataset of ICD diagnoses

Description

A simulated dataset of hospital diagnoses created with the help of the comorbidity package including the following columns:

id

Individual identifier,

case

Indicator for case status,

diag

An ICD-10 diagnosis code.

Usage

data(diagnoses)

Format

A data frame with 23,144 rows and 3 columns

Remove cuts from your tree. This is, e.g., useful if you would like to remove certain chapters from the ICD-10 tree used for the analysis as some chapters might be a prior deemed irrelevant for the exposure of interest, e.g., chapter 20 (external causes of death) might not be of interest when comparing two drug exposures.

Description

Remove cuts from your tree. This is, e.g., useful if you would like to remove certain chapters from the ICD-10 tree used for the analysis as some chapters might be a prior deemed irrelevant for the exposure of interest, e.g., chapter 20 (external causes of death) might not be of interest when comparing two drug exposures.

Usage

drop_cuts(tree, cuts, delimiter = "/", return_removed = FALSE)

Arguments

tree

A dataset with one variable pathString defining the tree structure that you would like to use. This dataset can, e.g., be created using create_tree.
cuts

A character vector of cuts to remove. Please make sure that your string uniquely identifies the cut that should be removed. Each string is passed to base::gsub() to identify the cuts that should be removed. Hence, strings can include regular expressions for identifying cuts. If you would like to remove a cut on the top level of the hierarchy, it might be helpful to use the regular expression operator ^.

Regular expression are composed as follows: paste0(cuts, delimiter, "?(.*)")
delimiter

A character defining the delimiter of different tree levels within your pathString. The default is /.
return_removed

A logical value for indicating whether you would like to get a list of removed cuts returned by the function.

Value

If return_removed = FALSE a data.frame with a single variable named pathString is returned, which includes the updated tree. If return_removed = TRUE a list with two elements is return:

tree

The updated tree file

removed

A list of character vectors including the paths that have been removed from the supplied tree. The list is named using the cuts supplied to cut.

Examples

drop_cuts(icd_10_se, c("B35-B49", "F41")) |>
   head()

Swedish version of the ICD-10 diagnoses code tree

Description

A dataset including the following column:

pathString

A string identifying all the parents of a node. Each parent is separated by a /.

Usage

data(icd_10_se)

Dictionary for the Swedish version of the ICD-10 diagnoses code tree

Description

A dataset including the following column:

node

A string identidying a node

title

A label for the node

Usage

data(icd_10_se_dict)

Unconditional Bernoulli Tree-Based Scan Statistics for R

Description

Unconditional Bernoulli Tree-Based Scan Statistics for R

Usage

TreeMineR(
  data,
  tree,
  p = NULL,
  n_exposed = NULL,
  n_unexposed = NULL,
  dictionary = NULL,
  delimiter = "/",
  n_monte_carlo_sim = 9999,
  random_seed = FALSE,
  return_test_dist = FALSE,
  future_control = list(strategy = "sequential")
)

Arguments

data

The dataset used for the computation. The dataset needs to include the following columns:

id

An integer that is unique to every individual.

leaf

A string identifying the unique diagnoses or leafs for each individual.

exposed

A 0/1 indicator of the individual's exposure status.

See below for the first and last rows included in the example dataset. 

   id leaf exposed
    1 K251       0
    2 Q702       0
    3  G96       0
    3 S949       0
    4 S951       0
 ---
  999 V539       1
  999 V625       1
  999 G823       1
 1000  L42       1
 1000 T524       1
tree

A dataset with one variable pathString defining the tree structure that you would like to use. This dataset can, e.g., be created using create_tree.
p

The proportion of exposed individuals in the dataset. Will be calculated based on n_exposed, and n_unexposed if both are supplied.
n_exposed

Number of exposed individuals (Optional).
n_unexposed

Number of unexposed individuals (Optional).
dictionary

A data.frame that includes one node column and a title column, which are used for labeling the cuts in the output of TreeMineR.
delimiter

A character defining the delimiter of different tree levels within your pathString. The default is /.
n_monte_carlo_sim

The number of Monte-Carlo simulations to be used for calculating P-values.
random_seed

Random seed used for the Monte-Carlo simulations.
return_test_dist

If true, a data.frame of the maximum log-likelihood ratios in each Monte Carlo simulation will be returned. This distribution of the maximum log-likelihood ratios is used for estimating the P-value reported in the result table.
future_control

A list of arguments passed future::plan. This is useful if one would like to parallelise the Monte-Carlo simulations to decrease the computation time. The default is a sequential run of the Monte-Carlo simulations.

Value

A data.frame with the following columns:

cut

The name of the cut G.

n1

The number of exposed events belonging to cut G.

n1

The number of inexposed events belonging to cut G.

risk1

The absolute risk of getting an event belonging to cut G among the exposed.

risk0

The absolute risk of getting an event belonging to cut G among the unexposed.

RR

The risk ratio of the absolute risk among the exposed over the absolute risk among the unexposed

llr

The log-likelihood ratio comparing the observed and expected number of exposed events belonging to cut G.

p

The P-value that cut G is a cluster of events.

If return_test_dist is true the function returns a list of two data.frame.

result_table

A data.frame including the results as described above.

test_dist

A data.frame with two columns: iteration the number of the Monte Carlo iteration. Note that iteration is the calculation based on the original data and is, hence, not included in this data.fame. max_llr: the highest observed log-likelihood ratio for each Monte Carlo simulation

References

Kulldorff et al. (2003) A tree-based scan statistic for database disease surveillance. Biometrics 56(2): 323-331. DOI: 10.1111/1541-0420.00039.

Examples

TreeMineR(data = diagnoses,
          tree  = icd_10_se,
          p = 1/11,
          n_monte_carlo_sim = 99,
          random_seed = 1234) |>
  head()