dmhct

{dmhct} extracts and prepares data for the MLinHCT project. The main functionality is wrapped into a single function, dm_hct(), that produces a dm object (from the {dm} package) containing cleaned data tables from an SQL Server database.

Installation and Setup

You can install {dmhct} from Github with:

# install.packages("remotes")
# remotes::install_github("jesse-smith/dmhct")
install.packages("dmhct", repos = c('https://jesse-smith.r-universe.dev', 'https://cloud.r-project.org'))

To use the package, you will need access to the SQL Server instance containing the data. This is provided by St. Jude IT; contact Jesse Smith to learn how to gain access.

Introduction

{dmhct} is designed to get data from the SQL server database to your local machine in a standardized fashion. As noted above, the main entry point to this functionality is the dm_hct() function, which may be all you ever need to use. It wraps the entire pipeline into a single call:

library(dmhct)
dm <- dm_hct()
#> Creating remote `dm` object
#> Extracting database tables (1/4)
#> Extracting `adverse_events_v2`
#> Extracting `adverse_events_v3`
#> Extracting `adverse_events_v4_1`
#> Extracting `adverse_events_v4_2`
#> Extracting `cerner_obs`
#> Extracting `cerner1`
#> Extracting `cerner2`
#> Extracting `cerner3`
#> Extracting `chimerism`
#> Extracting `death`
#> Extracting `disease_status`
#> Extracting `encounters`
#> Extracting `engraftment`
#> Extracting `gvhd`
#> Extracting `hla_donor`
#> Extracting `hla_patient`
#> Extracting `master`
#> Extracting `mrd`
#> Extracting `prior_transplant`
#> Extracting `relapse`
#> Done.
#> Standardizing table columns (2/4)
#> Standardizing `adverse_events_v2`
#> Standardizing `adverse_events_v3`
#> Standardizing `adverse_events_v4_1`
#> Standardizing `adverse_events_v4_2`
#> Standardizing `cerner_obs`
#> Standardizing `cerner1`
#> Standardizing `cerner2`
#> Standardizing `cerner3`
#> Standardizing `chimerism`
#> Standardizing `death`
#> Standardizing `disease_status`
#> Standardizing `encounters`
#> Standardizing `engraftment`
#> Standardizing `gvhd`
#> Standardizing `hla_donor`
#> Standardizing `hla_patient`
#> Standardizing `master`
#> Standardizing `mrd`
#> Standardizing `prior_transplant`
#> Standardizing `relapse`
#> Combining similar tables (3/4)
#> Combining `cerner1`, `cerner2`, `cerner3`
#> Combining hla_donor, hla_patient
#> Done.
#> Pivoting E-A-V tables (4/4)
#> Pivoting `hla`
#> Done.
dm
#> ── Metadata ────────────────────────────────────────────────────────────────────
#> Tables: `adverse_events_v2`, `adverse_events_v3`, `adverse_events_v4_1`, `adverse_events_v4_2`, `cerner`, … (17 total)
#> Columns: 332
#> Primary keys: 0
#> Foreign keys: 0

The pipeline can be decompsed into 4 steps, which are each responsible for a set of tasks:

1. dm_extract() pulls the database from the SQL server onto your local machine
2. dm_standardize() ensures that column names, types, and contents are consistent and in a standard format
3. dm_combine() combines certain sets of tables into single tables
4. dm_pivot() pivots long-format datasets that can be widened without losing information

Running these functions in order will produce the same results as running dm_hct():

dm2 <- dm_extract() %>%
  dm_standardize() %>%
  dm_combine() %>%
  dm_pivot()
#> Creating remote `dm` object
#> Extracting `adverse_events_v2`
#> Extracting `adverse_events_v3`
#> Extracting `adverse_events_v4_1`
#> Extracting `adverse_events_v4_2`
#> Extracting `cerner_obs`
#> Extracting `cerner1`
#> Extracting `cerner2`
#> Extracting `cerner3`
#> Extracting `chimerism`
#> Extracting `death`
#> Extracting `disease_status`
#> Extracting `encounters`
#> Extracting `engraftment`
#> Extracting `gvhd`
#> Extracting `hla_donor`
#> Extracting `hla_patient`
#> Extracting `master`
#> Extracting `mrd`
#> Extracting `prior_transplant`
#> Extracting `relapse`
#> Standardizing `adverse_events_v2`
#> Standardizing `adverse_events_v3`
#> Standardizing `adverse_events_v4_1`
#> Standardizing `adverse_events_v4_2`
#> Standardizing `cerner_obs`
#> Standardizing `cerner1`
#> Standardizing `cerner2`
#> Standardizing `cerner3`
#> Standardizing `chimerism`
#> Standardizing `death`
#> Standardizing `disease_status`
#> Standardizing `encounters`
#> Standardizing `engraftment`
#> Standardizing `gvhd`
#> Standardizing `hla_donor`
#> Standardizing `hla_patient`
#> Standardizing `master`
#> Standardizing `mrd`
#> Standardizing `prior_transplant`
#> Standardizing `relapse`
#> Combining `cerner1`, `cerner2`, `cerner3`
#> Combining hla_donor, hla_patient
#> Pivoting `hla`

identical(dm, dm2)
#> [1] FALSE

Changes from previous versions

The previous iteration of this package used the dm_elt() function as the wrapper and attempted to thoroughly clean all datasets. Unfortunately, this requires many ad hoc decisions, hard-coded transformations, and frequent revisions; the resulting codebase was very difficult to maintain and frequently made decisions that would be better left to the user. The updated version of this package attempts to do much less; its goal is to provide the database in a standardized format, with all information from the original data preserved. This passes the task of interpreting the data to the user.

Note that dm_elt() and its components are still exported; dm_extract(.legacy = TRUE) and dm_transform() are the components of this pipeline. However, these functions (and the .legacy argument of dm_extract()) are deprecated, and their use is discouraged.

Variable Typing System

Variables names are snakecased, with a prefix describing the content of the variable. This mostly maps to familiar R classes: num_ and pct_ variables are numeric (with pct_ taking values between 0 and 100), lgl_ variables are logical, and chr_ variables are character. There are exceptions, though; these come in two forms. The first are variable prefixes that could map to an R type, but do not because the underlying data may need additional cleaning. This is currently true for cat_ variable, which are conceptually factors but are stored as character, and dt_ variables, which may be stored as either Date or POSIXct class, depending on whether time information is included. The second form of exception are variables that have no equivalent base R type; this is currently true for mcat_ variables, which represent categorical variables that can take multiple values simultaneously, and intvl_ variables, which are interval-valued numerics. Both are stored as character, with mcat_ values separated by a comma and intvl_ represented in standard interval notation.

Helper Functions

{dmhct} exports several sets of helper functions. These are used internally to make data extraction and cleaning a bit easier. The functions are listed below; see each function’s documentation for more information.

Data cleaning

• intvl_to_matrix(): Converts the interval representation in intvl_ variables to a 4-column numeric matrix (these variables are standardized by std_intvl() - see below). Columns represent open or closed bounds and the location of those bounds.
• non_numeric(): Helps quickly determine what values in a vector cannot be converted to numeric (either directly or by std_num() - see below).
• std_chr(): Standardizes character vectors to ASCII text with no unnecessary whitespace and a given case. By default, it will retain newlines inside a text, though it will condense consecutive newlines and any carriage returns into a single newline.
• std_date(): Standardizes a date vector and returns a vector in Date or POSIXct format, depending on whether there is sub-daily information available in the data.
• std_intvl(): standardizes the various interval representations found in the ML in HCT dataset. These intervals are assumed to be in percentage values and this lie between 0 and 100. Explicit intervals with upper and lower bounds, as well as implicit intervals using < and >, are handled (<= and >= are currently not supported). The return value simplifies to </>/<=/>= or a single numeric value if possible and uses standard interval notation if not.
• std_lgl(): Converts other classes to logical vectors. All but character use as.logical(); character (and factor) vectors are converted by first (optionally) standardizing with std_chr() and then assigning logical value based on the regular expression in true, false, and na.
• std_num(): Converts all base classes, as well as int64, factor, Date, and POSIXt vectors to the simplest numeric form possible. character vectors (and factors) are standardized using std_chr() by default, then converted. double and int64 vectors will be converted to integer if this does not cause overflow or loss of precision. Date is converted to integer, and POSIXt is converted to integer if the range allows, otherwise double.

Database Interaction

• con_sql_server(): Creates an ODBC connection to the SQL server storing MLinHCT data
• dm_sql_server(): Creates a {dm} object representing data on the SQL server
• dm_disconnect(): Closes the remote server connection used by a dm object, if present
• dm_is_remote(): Determines whether a dm object has a remote connection
• dm_compute(): Forces computation of database queries on all tables in a dm object
• dm_collect(): Computes all database queries (as in dm_compute()), and additionally loads tables onto local machine