REDDIE WP4

Analysis plans, Implementation and Results

Swedish group

Outline

  • Statistical Analysis Plans and changes 
  • Implementation
  • Results

Statistical Analysis Plans

Inspired by Comparison Outcome zenodo.org
LEADER GLP1 vs DPP4 Cardiovascular https://doi.org/10.5281/zenodo.15311401
EMPA-REG SGLT2 vs DPP4 Cardiovascular https://doi.org/10.5281/zenodo.15320146
LEAD2-3 GLP1 vs DPP4 Metabolic (HbA1c & Weight) https://doi.org/10.5281/zenodo.15347461

Linked registries:

Swedish National Diabetes Register (NDR)

Swedish National Prescribed Drug Register (NPDR)

Swedish National Patient Register (NPR)

Swedish Cause of Death Register (CODR)

Swedish Longitudinal Integrated Database of Health Insurance and Labour Market Studies (LISA)

Study design / Methodology:

New user design - Persons in NDR with type 2 diabetes are included if/when they redeem relevant prescriptions within a predefined time window, and satisfy eligibility criteria emulating/inspired by the corresponding trial.

Follow-up until earliest occurrence of an outcome, death, emigration or end of trial, split into 6-month intervals.

Baseline covariates determined at inclusion looking back a predetermined period of time (any event within 180 days before inclusion for comorbidities, redemption of prescriptions within 180 days before inclusion for medications).

Time dependent covariates and outcome variables are summarized per interval (any event or prescription within interval, or, mean for continuous variables).

Efficacy is defined through absolute risk differences and risk ratios with bootstrap confidence intervals at prespecified follow-up time using the LTMLE framework with penalized outcome- and propensity score models, with penalty parameter determined by cross validation and without truncation. All-cause mortality is treated as competing event and emigration/end-of-trial as censoring event.

LEADER

EMPA-REG

LEAD 2-3

Primary Outcome:

MACE at 3.5 years

Primary Outcome:

MACE at 4 years

Primary Outcome(s):

Reduction in HbA1c

Reduction in Weight

at 1 year

Criteria within time
before BL
✅No prior type 1 diagnosis
✅HbA1c over 53 mmol/mol 1 year
✅Age over 50 1 year
✅Concomitant cardiovascular, cerebrovascular, peripheral vascular disease or renal disease or chronic heart failure OR age over 60 and other specified risk factors of vascular disease 10 years
❌Prescription of short-acting insulin 180 days
​❌Acute decompensation of glycemic control 90 days
​❌Primary or secondary outcome  14 days
❌Renal replacement therapy 180 days
❌End-stage liver disease 10 years
❌Solid organ transplant  10years
❌Malignant neoplasm 10 years
❌Familial medullary thyroid carcinoma 10 years
​❌Non-familial medullary thyroid carcinoma 10 years
​❌Any SGLT2 10 years
❌Immigration 10 years
​❌Emigration 10 years
❌No measurements for eGFR, Albuminuria, Smoking 2 years 3years
❌No measurements for BMI 1 year
Criteria within time before BL
✅No prior Type 1 diagnosis
✅HbA1c between 53-83 mmol/mol 1 year
✅Age over 18 1 year
✅Established cardiovascular disease 10 years
✅eGFR over 30 2 years
❌Liver disease 10 years
❌eGFR under 30 2 years
❌Bariatric surgery 10 years
❌Blood dyscrasias 20 years
❌Cancer (excluding non-melanoma skin cancer) 10 years
❌Any systemic steroids 90 days
❌Uncontrolled endocrine disorder 10 years
❌Familial medullary thyroid carcinoma 10 years
❌Occurrence of primary or secondary outcomes 60 days
❌Any GLP1 90 days
❌Immigration 10 years
❌Emigration 10 years
❌No measurements for eGFR, Albuminuria, Smoking 2 years
❌No measurements for BMI 1 year
Criteria within time before BL
✅No prior type 1 diagnosis
✅Existing measurement for HbA1c (or weight) 28 days
✅Age 18-90 years 1 year
✅BMI less than 42 kg./m2 90 days
✅HbA1c between 53 and 97 mmol/mol 90 days
❌Insulin prescription redeemed 90 days
❌Prescription of systemic corticosteroids  180 days
❌Liver-disease diagnosis 10 years
❌Prescription of excluded GLP1 180 days
❌Pregnancy 1 year
❌Any SGLT2 10 years
❌Immigration 10 years
❌Emigration 10 years
❌No measurements for eGFR or Albuminuria 2 years

List of changes

  • Confidence intervals: Influence Function 
    → Bootstrap
  • "Non-fatal" (alive after 30 days) MI and Stroke 
    → Any MI and Stroke from NPR
    (results also for non-fatal and including events from CoDR)
  • Require measurements for HbA1c, BMI (within 1 year) and eGFR, Albuminuria (within 2 years) for baseline values. 
    → Only HbA1c and eGFR at baseline for larger cohorts and power.
  • HbA1c, BMI, eGFR, Albuminuria, Smoking as time dependent covariates using LOCF to impute missing values in intervals.
    →  Only comedication and comorbidities as time dependent covaritates
  • LEAD2-3: Continuous, time dependent HbA1c outcome 
    → Categorical improvement, cumulative incidence

Implementation

Trial meta data

Targets - pipeline

Report / Diagnostics

  • Defines variables (register, ICD-codes, etc.)
  • Specifies eligibility criteria
  • Specifies analyses
  • Flowchart
  • Cohort characteristics
  • Overview of results
  • Diagnostic plots
  • Reads in trial meta data
  • Constructs cohorts and applies eligibility criteria
  • Performs specified analyses
  • Outputs a report and log
 # Created by use_targets().
#
# LEADER trial emulation
# Jens Michelsen 2024-12-09

Sys.setenv(OMP_THREAD_LIMIT=2,R_DATATABLE_NUM_PROC_PERCENT=5)
# Load packages required to define the pipeline:
library(targets)
library(tarchetypes) # Load other packages as needed.
library(crew)


# Set target options:
tar_option_set(
  # Packages that your targets need for their tasks.
  packages = c("data.table","dtplyr","tidyverse","purrr","qs",
               "qs2","officer","flextable","survival",
               "rtmle"), 
  # default storage format. qs is fast.
  format = "qs",
  # set number of workers
    controller = crew_controller_local(
      workers = future::availableCores(), 
      options_local = crew_options_local(log_directory="Logs/crew/"),
  #    #options_metrics = crew_options_metrics(path = "/dev/stdout", seconds_interval = 1),
      garbage_collection = TRUE
   ),
   storage = "worker",
   retrieval = "worker"
)

 # if (tar_active()) {
 #   log_start(
 #     path = "main_process.txt",
 #     seconds = 1
 #   )
 # }
# Run the R scripts in the R/ folder with your custom functions:
tar_source("R")

# Target list
list(
  #-----------------------------------------------------------------------------
  # Define variables for study start and study end
  tar_target(
    name    = study_start,
    command = lubridate::as_date("2012-01-01")
  ),
  tar_target(
    name    = study_end,
    command = lubridate::as_date("2022-12-31")
  ),
  #-----------------------------------------------------------------------------
  # Define data sources to be used.
  tar_target(
    name    = ndr_file, # National Diabetes Register
    command = "/safe/data/REDDIE/Data/Clean/ndr_plus.qs",
    format  = "file"
  ),
  tar_target(
    name    = lmed_file, # Prescribed Drug register
    command = "/safe/data/REDDIE/Data/Import/lmed.qs",
    format  = "file"
  ),
  tar_target( 
    name    = par_file, # Patient in-/out care
    command = "/safe/data/REDDIE/Data/Clean/par_plus.qs",
    format  = "file"
  ),
  tar_target( 
    name    = dors_file, # Cause of death register
    command = "/safe/data/REDDIE/Data/Import/dors.qs",
    format  = "file"
  ),
  tar_target( 
    name    = lev_base_file, # Basic population data (Sex/Birth date/etc)
    command = "/safe/data/REDDIE/Data/Import/lev_base.qs",
    format  = "file"
  ),
  tar_target( 
    name    = migrations_file, # Patient in-/out care
    command = "/safe/data/REDDIE/Data/Import/migrations.qs",
    format  = "file"
  ),
  tar_target( 
    name    = lisa_file, # Patient in-/out care
    command = "/safe/data/REDDIE/Data/Clean/lisa_plus.qs",
    format  = "file"
  ),
  #-----------------------------------------------------------------------------
  # load xlsx-file with variable meta data (incl. ICD, KVA and ATC codes)
  tar_target(
    name    = variable_meta_data_xlsx,
    command = "variable_meta_data.xlsx",
    format  = "file",
    cue     = tar_cue(mode = "thorough")
  ),
  tar_target(
    name    = variable_lookup,
    command = openxlsx::read.xlsx(variable_meta_data_xlsx, 
                                  sheet = "lookup",
                                  rowNames=TRUE),
    cue     = tar_cue(mode = "thorough")
  ),
  #-----------------------------------------------------------------------------
  # Extract a data set containing id's of people with type-2 diabetes
  tar_target(
    name    = type2_cohort,
    command = get_type2_cohort(ndr_file, lev_base_file)
  ),
  #-----------------------------------------------------------------------------
  # collect data on GLP1/DPP4, define initial treatment arms,
  # reduce data set to treatment cohort,
  # 
  tar_target(
    name    = treatment_cohort,
    command = get_treatment_cohort(variable_lookup = variable_lookup,
                                   type2_cohort    = type2_cohort,
                                   lmed_file       = lmed_file,
                                   study_start     = study_start,
                                   study_end       = study_end)
  ),
  #-----------------------------------------------------------------------------
  #### Now store reduced, cohort-specific, versions of ndr, lmed, dors, lev
  #### for speed and memory purposes
  tar_target(
    name    = ndr,
    command = left_join(treatment_cohort,qread(ndr_file),by="LopNr")
  ),
  tar_target(
    name    = lmed,
    command = left_join(treatment_cohort,qread(lmed_file),by="LopNr")
  ),
  tar_target(
    name    = par,
    command = left_join(treatment_cohort,qread(par_file),by="LopNr")
  ),
  tar_target(
    name    = dors,
    command = left_join(treatment_cohort,qread(dors_file),by="LopNr")
  ),
  tar_target(
    name    = migrations,
    command = left_join(treatment_cohort,qread(migrations_file),by="LopNr")
  ),
  tar_target(
    name    = lisa,
    command = left_join(treatment_cohort,qread(lisa_file),by="LopNr")
  ),
  #-----------------------------------------------------------------------------
  #### get inclusion and exclusion criteria
  tar_target(
    name    = incl_excl,
    command = openxlsx::read.xlsx(variable_meta_data_xlsx, 
                                  sheet = "inclusion_exclusion"),
    cue     = tar_cue(mode = "thorough")
  ),
  #-----------------------------------------------------------------------------
  #### and create study cohort
  tar_target(
    name    = study_cohort_with_criteria_values, # can be used later for flow chart
    command = get_study_cohort(treatment_cohort,
                               variable_lookup, incl_excl,
                               ndr=ndr, lmed=lmed, par=par, dors=dors, 
                               migrations=migrations, lisa=lisa),
    cue     = tar_cue("thorough")
  ),
  tar_target(
    name    = study_cohort,
    command = filter(study_cohort_with_criteria_values,included) |>
      select(LopNr,index,treat,study_start,study_end),
    cue     = tar_cue("thorough")
  ),
  #-----------------------------------------------------------------------------
  # Collect baseline- and timedependent data for study cohort
  tar_target(
    name    = baseline_data,
    command = get_baseline_data(study_cohort,
                                variable_lookup,
                                ndr=ndr, par=par, lmed=lmed, lisa=lisa)
  ),
   tar_target(
     name    = timevar_data,
     command = get_timevar_data(study_cohort,
                                variable_lookup,
                                ndr=ndr, par=par, lmed=lmed, dors, migrations)
  ),
  #-----------------------------------------------------------------------------
  # Define specifications for the different analyses (from metadata_file) 
  tar_target(
    name     = analysis_specs,
    command  = openxlsx::read.xlsx(variable_meta_data_xlsx, 
                                   sheet = "outcomes"),
    cue      = tar_cue(mode = "thorough")
  ),
  tar_target(
    name      = args,
    command   = split(analysis_specs,seq(nrow(analysis_specs))),
    iteration = "list"
  ),
  #-----------------------------------------------------------------------------
  # Compute LTMLE estimates
  tar_target(
    name    = estimate,
    command = fit_rtmle_args(args, 
                             study_cohort, variable_lookup, 
                             baseline_data, timevar_data,
                             bootstrap  = TRUE,
                             M_boot     = 200,
                             rnd_seed   = 321),
    pattern = map(args),
    iteration = "list"#,
    #error = "null"
  ),
  tar_target(
    name    = estimates,
    command = dplyr::bind_rows(estimate)
  ),
  #-----------------------------------------------------------------------------
  # Create report
  tar_target(
    name     = report,
    command  = get_report(estimates, study_cohort, 
                          study_cohort_with_criteria_values, 
                          variable_lookup, incl_excl, 
                          baseline_data, timevar_data),
    cue      = tar_cue(mode="always")
  ),
  # Create log
  tar_target(
    name     = log,
    command  = print_log(),
    cue      = tar_cue(mode="always")
  )
)

Trial meta data

Targets pipeline

 # Created by use_targets().
#
# LEADER trial emulation
# Jens Michelsen 2024-12-09

Sys.setenv(OMP_THREAD_LIMIT=2,R_DATATABLE_NUM_PROC_PERCENT=5)
# Load packages required to define the pipeline:
library(targets)
library(tarchetypes) # Load other packages as needed.
library(crew)


# Set target options:
tar_option_set(
  # Packages that your targets need for their tasks.
  packages = c("data.table","dtplyr","tidyverse","purrr","qs",
               "qs2","officer","flextable","survival",
               "rtmle"), 
  # default storage format. qs is fast.
  format = "qs",
  # set number of workers
    controller = crew_controller_local(
      workers = future::availableCores(), 
      options_local = crew_options_local(log_directory="Logs/crew/"),
  #    #options_metrics = crew_options_metrics(path = "/dev/stdout", seconds_interval = 1),
      garbage_collection = TRUE
   ),
   storage = "worker",
   retrieval = "worker"
)

 # if (tar_active()) {
 #   log_start(
 #     path = "main_process.txt",
 #     seconds = 1
 #   )
 # }
# Run the R scripts in the R/ folder with your custom functions:
tar_source("R")

# Target list
list(
  #-----------------------------------------------------------------------------
  # Define variables for study start and study end
  tar_target(
    name    = study_start,
    command = lubridate::as_date("2012-01-01")
  ),
  tar_target(
    name    = study_end,
    command = lubridate::as_date("2022-12-31")
  ),
  #-----------------------------------------------------------------------------
  # Define data sources to be used.
  tar_target(
    name    = ndr_file, # National Diabetes Register
    command = "/safe/data/REDDIE/Data/Clean/ndr_plus.qs",
    format  = "file"
  ),
  tar_target(
    name    = lmed_file, # Prescribed Drug register
    command = "/safe/data/REDDIE/Data/Import/lmed.qs",
    format  = "file"
  ),
  tar_target( 
    name    = par_file, # Patient in-/out care
    command = "/safe/data/REDDIE/Data/Clean/par_plus.qs",
    format  = "file"
  ),
  tar_target( 
    name    = dors_file, # Cause of death register
    command = "/safe/data/REDDIE/Data/Import/dors.qs",
    format  = "file"
  ),
  tar_target( 
    name    = lev_base_file, # Basic population data (Sex/Birth date/etc)
    command = "/safe/data/REDDIE/Data/Import/lev_base.qs",
    format  = "file"
  ),
  tar_target( 
    name    = migrations_file, # Patient in-/out care
    command = "/safe/data/REDDIE/Data/Import/migrations.qs",
    format  = "file"
  ),
  tar_target( 
    name    = lisa_file, # Patient in-/out care
    command = "/safe/data/REDDIE/Data/Clean/lisa_plus.qs",
    format  = "file"
  ),
  #-----------------------------------------------------------------------------
  # load xlsx-file with variable meta data (incl. ICD, KVA and ATC codes)
  tar_target(
    name    = variable_meta_data_xlsx,
    command = "variable_meta_data.xlsx",
    format  = "file",
    cue     = tar_cue(mode = "thorough")
  ),
  tar_target(
    name    = variable_lookup,
    command = openxlsx::read.xlsx(variable_meta_data_xlsx, 
                                  sheet = "lookup",
                                  rowNames=TRUE),
    cue     = tar_cue(mode = "thorough")
  ),
  #-----------------------------------------------------------------------------
  # Extract a data set containing id's of people with type-2 diabetes
  tar_target(
    name    = type2_cohort,
    command = get_type2_cohort(ndr_file, lev_base_file)
  ),
  #-----------------------------------------------------------------------------
  # collect data on GLP1/DPP4, define initial treatment arms,
  # reduce data set to treatment cohort,
  # 
  tar_target(
    name    = treatment_cohort,
    command = get_treatment_cohort(variable_lookup = variable_lookup,
                                   type2_cohort    = type2_cohort,
                                   lmed_file       = lmed_file,
                                   study_start     = study_start,
                                   study_end       = study_end)
  ),
  #-----------------------------------------------------------------------------
  #### Now store reduced, cohort-specific, versions of ndr, lmed, dors, lev
  #### for speed and memory purposes
  tar_target(
    name    = ndr,
    command = left_join(treatment_cohort,qread(ndr_file),by="LopNr")
  ),
  tar_target(
    name    = lmed,
    command = left_join(treatment_cohort,qread(lmed_file),by="LopNr")
  ),
  tar_target(
    name    = par,
    command = left_join(treatment_cohort,qread(par_file),by="LopNr")
  ),
  tar_target(
    name    = dors,
    command = left_join(treatment_cohort,qread(dors_file),by="LopNr")
  ),
  tar_target(
    name    = migrations,
    command = left_join(treatment_cohort,qread(migrations_file),by="LopNr")
  ),
  tar_target(
    name    = lisa,
    command = left_join(treatment_cohort,qread(lisa_file),by="LopNr")
  ),
  #-----------------------------------------------------------------------------
  #### get inclusion and exclusion criteria
  tar_target(
    name    = incl_excl,
    command = openxlsx::read.xlsx(variable_meta_data_xlsx, 
                                  sheet = "inclusion_exclusion"),
    cue     = tar_cue(mode = "thorough")
  ),
  #-----------------------------------------------------------------------------
  #### and create study cohort
  tar_target(
    name    = study_cohort_with_criteria_values, # can be used later for flow chart
    command = get_study_cohort(treatment_cohort,
                               variable_lookup, incl_excl,
                               ndr=ndr, lmed=lmed, par=par, dors=dors, 
                               migrations=migrations, lisa=lisa),
    cue     = tar_cue("thorough")
  ),
  tar_target(
    name    = study_cohort,
    command = filter(study_cohort_with_criteria_values,included) |>
      select(LopNr,index,treat,study_start,study_end),
    cue     = tar_cue("thorough")
  ),
  #-----------------------------------------------------------------------------
  # Collect baseline- and timedependent data for study cohort
  tar_target(
    name    = baseline_data,
    command = get_baseline_data(study_cohort,
                                variable_lookup,
                                ndr=ndr, par=par, lmed=lmed, lisa=lisa)
  ),
   tar_target(
     name    = timevar_data,
     command = get_timevar_data(study_cohort,
                                variable_lookup,
                                ndr=ndr, par=par, lmed=lmed, dors, migrations)
  ),
  #-----------------------------------------------------------------------------
  # Define specifications for the different analyses (from metadata_file) 
  tar_target(
    name     = analysis_specs,
    command  = openxlsx::read.xlsx(variable_meta_data_xlsx, 
                                   sheet = "outcomes"),
    cue      = tar_cue(mode = "thorough")
  ),
  tar_target(
    name      = args,
    command   = split(analysis_specs,seq(nrow(analysis_specs))),
    iteration = "list"
  ),
  #-----------------------------------------------------------------------------
  # Compute LTMLE estimates
  tar_target(
    name    = estimate,
    command = fit_rtmle_args(args, 
                             study_cohort, variable_lookup, 
                             baseline_data, timevar_data,
                             bootstrap  = TRUE,
                             M_boot     = 200,
                             rnd_seed   = 321),
    pattern = map(args),
    iteration = "list"#,
    #error = "null"
  ),
  tar_target(
    name    = estimates,
    command = dplyr::bind_rows(estimate)
  ),
  #-----------------------------------------------------------------------------
  # Create report
  tar_target(
    name     = report,
    command  = get_report(estimates, study_cohort, 
                          study_cohort_with_criteria_values, 
                          variable_lookup, incl_excl, 
                          baseline_data, timevar_data),
    cue      = tar_cue(mode="always")
  ),
  # Create log
  tar_target(
    name     = log,
    command  = print_log(),
    cue      = tar_cue(mode="always")
  )
)

Report / Diagnostics

Results

LEADER

Only baseline HbA1c

Time dependent HbA1c

EMPA-REG

LEAD 2-3