Rotterdam Study → CDF → FHIR Pipeline (Cheat Sheet)

This document summarizes the preliminary workflow used to transform tabular CSV data into CDF (ICDF) files and subsequently into FHIR resources for federated learning and harmonization experiments.

The pipeline mainly consists of:

1. CSV variable extraction
2. Config generation for CDF conversion
3. CSV → CDF conversion
4. CDF → FHIR transformation
5. Dataset‑specific mapping (Rotterdam Study)

---

1. CSV → CDF Conversion (lifelinescsv_to_icdf)

To transform sample or real data files into CDF format:

python -m lifelinescsv_to_icdf.cdfgenerator <file_with_ids> <config_file> <output_folder>

You must first define the location of the assessments of the variables in the configuration file.

Example paths

# Example 1
python -m lifelinescsv_to_icdf.cdfgenerator \
  /home/umcg-hcadavid/temporal-data/csv2csd/ids.csv \
  /home/umcg-hcadavid/temporal-data/csv2csd/csv2csdconfig.json \
  /home/umcg-hcadavid/temporal-data/pheno_lifelines_csd_out

# Example 2
python -m lifelinescsv_to_icdf.cdfgenerator \
  /home/hmo/RS_CSV2CDF/dummy_practice/samplecsv/bigfiles/pseudo_ids.csv \
  /home/hmo/RS_CSV2CDF/dummy_practice/sample-configs/one-csv-15-vars-config.json \
  /home/hmo/RS_CSV2CDF/dummy_practice/dummy_csd_out

---

2. MDT Proof‑of‑Concept Variables

Predictors

• age
• sex
• eGFR
• albumin
• HDL cholesterol
• LDL cholesterol
• total cholesterol
• HbA1c
• hypertension history
• type 2 diabetes history
• creatinine
• systolic blood pressure
• diastolic blood pressure
• smoking history
• smoking quantity

CVD Outcome Definition

Composite outcome:

• stroke
• myocardial infarction (MI)
• heart failure (HF)

---

3. Generate Sample CSV Files

Script location:

/home/hmo/LifelinesCSV2CDF/samplecsv/generate_sample_csv_datafiles.py

Example usage:

python rs_fl_variables_csv_gen.py \
  --in /home/hmo/RS_CSV2CDF/data_csv/RS_ergo_tabular_05032023.csv \
  --out /home/hmo/RS_CSV2CDF/data_csv/RS_ergo_tabular_fl_selected_var_population.csv \
  --codebook /home/hmo/RS_CSV2CDF/data_csv/rs_cvd_variables.xlsx

---

4. Prepare Config File Before CDF Conversion

python rs_cdf_config_gen.py \
  --csv /home/hmo/RS_CSV2CDF/data_csv/RS_ergo_tabular_fl_selected_var_population.csv \
  --id-col ergoid \
  --ids-out /home/hmo/RS_CSV2CDF/data_csv/ids.csv \
  --config-out /home/hmo/RS_CSV2CDF/data_csv/rs_csv_var_config.json \
  --assessment a1 \
  --csv-normalized-out /home/hmo/RS_CSV2CDF/data_csv/RS_ergo_tabular_with_ppid.csv

---

5. Convert CSV → CDF

python -m lifelinescsv_to_icdf.cdfgenerator \
  /home/hmo/RS_CSV2CDF/data_csv/ids.csv \
  /home/hmo/RS_CSV2CDF/data_csv/rs_csv_var_config.json \
  /home/hmo/RS_CSV2CDF/data_cdf

CDF output directory:

/home/hmo/RS_CSV2CDF/data_cdf

---

6. Common Predictor Variables (Lifelines)

sexe
age
sbp
dbp
HDL_mmol
LDL_mmol_centri
TC_mmol
GFR
creat_umol
smoking
prev_DM
prev_HT

---

7. Outcome Variables

stroke_date
inc_hf_2018
enddat_hf
Inc_MI
enddat_MI
CVD_STATUS
CVD_ONSET_DATE

---

8. FHIR Transformation

Example commands:

npm run transform -- ./fhirvalidation/sampleinputs/input-p1234.json -o ./out

npm run transform -- ./fhirvalidation/sampleinputs/input-pa2739b7129c7319d73189273817318973822-follow_up_conditions_w_undefined_onset_date.json -o ./out

npm run transform -- ./fhirvalidation/sampleinputs/input-p675432-hf-diab-followup-missing-date.json -o ./out

npm run transform -- ./fhirvalidation/sampleinputs -o /home/hmo/CDF2Medmij-Mapping-tool/fhirvalidation/temp_out

---

9. CDF → MedMij FHIR Harmonization (Rotterdam Study)

Working directory:

/home/hmo/CDF2Medmij-Mapping-tool/src

Files That Do NOT Need Modification

Reusable generic components:

• functionsCatalog.ts
• inputSingleton.ts
• mapper.ts
• transformationParameters.ts
• unexpectedInputException.ts
• fhir-resource-interfaces/*
• zib-2017-mappings/*.jsonata

These remain dataset‑agnostic.

---

10. Rotterdam‑Specific TypeScript Files

src/
  rotterdam/
    Patient.ts
    BloodPressure.ts
    HDLCholesterol.ts
    LDLCholesterol.ts
    TotalCholesterol.ts
    eGFR.ts
    Diabetes.ts
    TobaccoUse.ts
    HeartFailure.ts
    Stroke.ts
    MyocardialInfarction.ts
    CardioVascularDisease.ts
    HistoryHTN.ts
    HistoryDM.ts
    HistoryCVD.ts
    rsFunctions.ts

---

11. Example RS CDF Variables

IDs / Baseline

PROJECT_PSEUDO_ID.a1
date_int_cen.a1
gebdatum.a1
sex_mapped.a1
age_at_baseline_years_derived.a1

Predictors

sbp.a1
dbp.a1
HDL_mmol.a1
LDL_mmol_chosen.a1
TC_mmol.a1
GFR.a1
smoking_status.a1
prev_HT_bool.a1
prev_DM_bool.a1

Outcomes

incident_mi_bool.a1
incident_mi_date_derived.a1
incident_stroke_bool.a1
incident_stroke_date_derived.a1
incident_hf_bool.a1
incident_hf_date_derived.a1
incident_cvd_composite_bool.a1
incident_cvd_date_derived.a1

---

12. Age Handling

Age should NOT be generated in FHIR mapping.

Instead, compute during feature engineering:

age = (baseline_date - birthdate).days / 365.25

FHIR represents age indirectly using:

• birthDate

Downstream systems compute age when required.

---

13. Rotterdam Transform Command

npm run transform:rotterdam -- \
  /home/hmo/RS_CSV2CDF/data_cdf \
  /home/hmo/RS_CSV2CDF/data_fhir

---

14. Notes

• The Lifelines federated learning pipeline consumes engineered predictors rather than raw FHIR attributes.
• The CDF → FHIR step is primarily for interoperability and harmonization.
• Dataset‑specific logic should be isolated inside the rotterdam/ folder to keep the core framework reusable.

---

15. Quick Pipeline Summary

CSV → Variable Selection → Config Generation → CDF → FHIR → FL Preprocessing → Model

---

End of cheat sheet.