Skip to content

A2-ai/yspec

 
 

Repository files navigation

yspec

Use yspec to document analysis data sets and utilize data attributes in a modeling and simulation workflow.

Installation

You can install the development version of yspec from GitHub with:

# install.packages("devtools")
devtools::install_github("metrumresearchgroup/yspec")

Create a yspec object

Before or while programming an analysis data set, write out data definitions in yaml format:

library(yspec)
library(dplyr)
library(tidyr)

readLines("inst/internal/analysis1.yml")[1:20] %>% writeLines()
. SETUP__:
.   description: Example PopPK analysis data set
.   sponsor: example-project
.   projectnumber: EXAMPK1011F
.   use_internal_db: true
.   glue: 
.     bmiunits: "kg/m$^2$"
.   flags: 
.     covariate: [AGE:SCR, HT, AST:ALT]
.   lookup_file: "look.yml"
.   extend_file: "analysis1-ext.yml"
. C:
. NUM:
. ID:
. SUBJ: !look
. TIME: !look
.   label: time after first dose
.   unit: hour
. SEQ: 
.   label: data type

Now use yspec to read that into a object in R

spec <- ys_load("inst/internal/analysis1.yml")

Query this object to get a sense of what is in the data overall

head(spec)
.    name info         unit              short        source
. 1     C  cd-            .  comment character ysdb_internal
. 2   NUM  ---            .      record number ysdb_internal
. 3    ID  ---            . subject identifier ysdb_internal
. 4  SUBJ  c--            . subject identifier ysdb_internal
. 5  TIME  ---         hour               TIME          look
. 6   SEQ  -d-            .                SEQ             .
. 7   CMT  ---            . compartment number ysdb_internal
. 8  EVID  -d-            .           event ID ysdb_internal
. 9   AMT  ---           mg        dose amount ysdb_internal
. 10   DV  --- micrograms/L dependent variable ysdb_internal

or on a column by column basis for continuous data

spec$WT
.  name  value    
.  col   WT       
.  type  numeric  
.  short weight   
.  unit  kg       
.  range 40 to 100

as well as categorical data

spec$BLQ
.  name  value                        
.  col   BLQ                          
.  type  numeric                      
.  short below limit of quantification
.  value 0 : above QL                 
.        1 : below QL

And we can render a define.pdf file as well

Using yspec

This section illustrates a few examples for how yspec might be used (other than creating define.pdf).

To make it easier to get started with yspec, we’ve included example data and corresponding yspec object in the package

data <- ys_help$data()
spec <- ys_help$spec()

Add factors

When you have discrete data, “decodes” can be provided and used to create factors in the data. We have that for the RF column

spec$RF
.  name  value               
.  col   RF                  
.  type  character           
.  short renal function stage
.  value norm : Normal       
.        mild : Mild         
.        mod  : Moderate     
.        sev  : Severe

Now we’ll have a column called RF_f which is a factor version of RF

ys_add_factors(data, spec, RF) %>% count(RF, RF_f)
.     RF     RF_f    n
. 1 mild     Mild  360
. 2  mod Moderate  360
. 3 norm   Normal 3280
. 4  sev   Severe  360

Recode

Every column can have a “short” name; for WT it is

spec$WT$short
. [1] "weight"

Every continuous data can also have a unit; again for WT

spec$WT$unit
. [1] "kg"

We use the spec to “recode” using this information. First create a data summary in long format

summ <- 
  data %>% 
  select(WT, ALB, AGE) %>% 
  pivot_longer(everything()) %>%
  group_by(name) %>% 
  summarise(Mean = mean(value), Sd = sd(value))

summ  
. # A tibble: 3 × 3
.   name   Mean     Sd
.   <chr> <dbl>  <dbl>
. 1 AGE   33.8   8.60 
. 2 ALB    4.30  0.707
. 3 WT    70.9  12.8

To recode, pull the information from spec

summ %>% 
  mutate(name = ys_recode(name, spec,  unit = TRUE, title_case = TRUE))
. # A tibble: 3 × 3
.   name            Mean     Sd
.   <chr>          <dbl>  <dbl>
. 1 Age (years)    33.8   8.60 
. 2 Albumin (g/dL)  4.30  0.707
. 3 Weight (kg)    70.9  12.8

Manipulating yspec

There are several functions for working on your yspec object

Select some columns

body_size <- ys_select(spec, WT, BMI, HT)

body_size
.  name info unit  short  source       
.  WT   ---  kg    weight ysdb_internal
.  BMI  ---  m2/kg BMI    ysdb_internal
.  HT   ---  cm    height ysdb_internal

Filter based on some flags that were set

ys_filter(spec, covariate)
.  name info unit   short                      source       
.  AGE  ---  years  age                        ysdb_internal
.  WT   ---  kg     weight                     ysdb_internal
.  CRCL ---  ml/min CRCL                       .            
.  ALB  ---  g/dL   albumin                    ysdb_internal
.  BMI  ---  m2/kg  BMI                        ysdb_internal
.  AAG  ---  mg/dL  alpha-1-acid glycoprotein  .            
.  SCR  ---  mg/dL  serum creatinine           .            
.  AST  ---  .      aspartate aminotransferase .            
.  ALT  ---  .      alanine aminotransferase   .            
.  HT   ---  cm     height                     ysdb_internal
.  CP   -d-  .      Child-Pugh score           look

Rename

spec %>% ys_select(BWT = WT, AGE, SCR) 
.  name info unit  short            source       
.  BWT  ---  kg    weight           ysdb_internal
.  AGE  ---  years age              ysdb_internal
.  SCR  ---  mg/dL serum creatinine .

Projects

An analysis project typically has several data sets that can be documented together. We make a project like this

pk   <- ys_load("inst/spec/DEM104101F_PK.yml")
pkpd <- ys_load("inst/spec/DEM104101F_PKPD.yml")
ae   <- ys_load("inst/spec/DEM104101F_AE.yml")
proj <- ys_project(pk, pkpd, ae)
proj
. projectnumber:  ABC101104F 
. sponsor:        ABC-Pharma 
. --------------------------------------------
. datafiles: 
.  name            description                       data_stem       
.  DEM104101F_PK   Population PK analysis data set   DEM104101F_PK   
.  DEM104101F_PKPD Population PKPD analysis data set DEM104101F_PKPD 
.  DEM104101F_AE   AE analysis data set              DEM0104101F_AE_2

This object can be rendered into a single define.pdf document.

Releases

No releases published

Packages

No packages published

Languages

  • R 86.9%
  • TeX 7.5%
  • Jsonnet 5.0%
  • Makefile 0.6%