import polars as pl
from csrlite import load_plan
pl.Config.set_tbl_rows(-1)polars.config.Config3 Planning
Objective
Create a minimal visible product (MVP) to enable the A&R plan.
We created two YAML files link for an MVP. We uses Python (source code), the same idea can be built in R.
organization.yaml: organization YAML file.plan_xyz123.yaml: study specific YAML file.
The structure of the YAML files are discussed in the inheritance section.
3.1 Loading Plan
We load the analysis plan that leverage information in pre-specified templates.
study_plan = load_plan('studies/xyz123/yaml/plan_xyz123.yaml')2026-02-03 15:25:46,399 - csrlite.common.plan - INFO - Successfully loaded dataset 'adsl' from 'studies/xyz123/yaml/../../../data/adsl.parquet'
2026-02-03 15:25:46,402 - csrlite.common.plan - INFO - Successfully loaded dataset 'adae' from 'studies/xyz123/yaml/../../../data/adae.parquet'
2026-02-03 15:25:46,404 - csrlite.common.plan - INFO - Successfully loaded dataset 'adie' from 'studies/xyz123/yaml/../../../data/adie.parquet'
2026-02-03 15:25:46,405 - csrlite.common.plan - INFO - Successfully loaded dataset 'adpd' from 'studies/xyz123/yaml/../../../data/adpd.parquet'study_plan.get_plan_df()
shape: (18, 5)
| analysis | population | observation | parameter | group |
|---|---|---|---|---|
| str | str | str | str | str |
| "disposition_summary" | "enrolled" | null | null | "trt01a" |
| "demographics" | "itt" | null | null | "trt01a" |
| "ae_summary" | "apat" | "week12" | "any;rel;ser" | "trt01a" |
| "ae_summary" | "apat" | "week24" | "any;rel;ser" | "trt01a" |
| "ae_specific" | "apat" | "week12" | "any" | "trt01a" |
| "ae_specific" | "apat" | "week12" | "rel" | "trt01a" |
| "ae_specific" | "apat" | "week12" | "ser" | "trt01a" |
| "ae_specific" | "apat" | "week24" | "any" | "trt01a" |
| "ae_specific" | "apat" | "week24" | "rel" | "trt01a" |
| "ae_specific" | "apat" | "week24" | "ser" | "trt01a" |
| "ae_listing" | "apat" | "week12" | "ser" | "trt01a" |
| "ae_listing" | "apat" | "week24" | "ser" | "trt01a" |
| "ae_listing" | "apat" | "week12" | "ser" | "trt01a" |
| "ae_listing" | "apat" | "week24" | "ser" | "trt01a" |
| "ie_summary" | "all" | null | null | null |
| "ie_listing" | "discon" | null | null | null |
| "pd_listing" | "itt" | "pd_major" | null | "trt01a" |
| "pd_listing" | "apat" | null | null | "trt01a" |
For each row, developer can read it as a function call to generate production ready TLF. e.g.
ae_summary(population = "itt", observation = "week12", parameter = "any;rel;ser")
3.1.1 Source Data
study_plan.get_dataset_df()
shape: (4, 3)
| name | path | loaded |
|---|---|---|
| str | str | bool |
| "adsl" | "../../../data/adsl.parquet" | true |
| "adae" | "../../../data/adae.parquet" | true |
| "adie" | "../../../data/adie.parquet" | true |
| "adpd" | "../../../data/adpd.parquet" | true |
study_plan.datasets["adsl"].head()
shape: (5, 50)
| STUDYID | USUBJID | SUBJID | SITEID | SITEGR1 | ARM | TRT01P | TRT01PN | TRT01A | TRT01AN | TRTSDT | TRTEDT | TRTDUR | AVGDD | CUMDOSE | AGE | AGEGR1 | AGEGR1N | AGEU | RACE | RACEN | SEX | ETHNIC | SAFFL | ITTFL | EFFFL | COMP8FL | COMP16FL | COMP24FL | DISCONFL | DSRAEFL | DTHFL | BMIBL | BMIBLGR1 | HEIGHTBL | WEIGHTBL | EDUCLVL | DISONSDT | DURDIS | DURDSGR1 | VISIT1DT | RFSTDTC | RFENDTC | VISNUMEN | RFENDT | DCDECOD | DCSREAS | MMSETOT | EOSSTT | DCSREASP |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | cat | str |
| "CDISCPILOT01" | "01-701-1015" | "1015" | "701" | "701" | "Placebo" | "Placebo" | "5.397605346934028e-79" | "Placebo" | "5.397605346934028e-79" | "19725.0" | "19906.0" | "182.0" | "5.397605346934028e-79" | "5.397605346934028e-79" | "63.0" | "<65" | "1.0" | "YEARS" | "WHITE" | "1.0" | "F" | "HISPANIC OR LATINO" | "Y" | "Y" | "Y" | "Y" | "Y" | "Y" | "" | "" | "" | "25.1" | "25-<30" | "147.3" | "54.4" | "16.0" | "18382.0" | "43.9" | ">=12" | "19718.0" | "2014-01-02" | "2014-07-02" | "12.0" | "19906.0" | "COMPLETED" | "Completed" | "23.0" | "Completed" | "BP too high" |
| "CDISCPILOT01" | "01-701-1023" | "1023" | "701" | "701" | "Placebo" | "Placebo" | "5.397605346934028e-79" | "Placebo" | "5.397605346934028e-79" | "19210.0" | "19237.0" | "28.0" | "5.397605346934028e-79" | "5.397605346934028e-79" | "64.0" | "<65" | "1.0" | "YEARS" | "WHITE" | "1.0" | "M" | "HISPANIC OR LATINO" | "Y" | "Y" | "Y" | "N" | "N" | "N" | "Y" | "Y" | "" | "30.4" | ">=30" | "162.6" | "80.3" | "14.0" | "16871.0" | "76.4" | ">=12" | "19196.0" | "2012-08-05" | "2012-09-02" | "5.0" | "19238.0" | "ADVERSE EVENT" | "Adverse Event" | "23.0" | "Discontinued" | "disposition event" |
| "CDISCPILOT01" | "01-701-1028" | "1028" | "701" | "701" | "Xanomeline High Dose" | "Xanomeline High Dose" | "81.0" | "Xanomeline High Dose" | "81.0" | "19558.0" | "19737.0" | "180.0" | "77.7" | "13986.0" | "71.0" | "65-80" | "2.0" | "YEARS" | "WHITE" | "1.0" | "M" | "NOT HISPANIC OR LATINO" | "Y" | "Y" | "Y" | "Y" | "Y" | "Y" | "" | "" | "" | "31.4" | ">=30" | "177.8" | "99.3" | "16.0" | "18247.0" | "42.8" | ">=12" | "19550.0" | "2013-07-19" | "2014-01-14" | "12.0" | "19737.0" | "COMPLETED" | "Completed" | "23.0" | "Completed" | "BP too low" |
| "CDISCPILOT01" | "01-701-1033" | "1033" | "701" | "701" | "Xanomeline Low Dose" | "Xanomeline Low Dose" | "54.0" | "Xanomeline Low Dose" | "54.0" | "19800.0" | "19813.0" | "14.0" | "54.0" | "756.0" | "74.0" | "65-80" | "2.0" | "YEARS" | "WHITE" | "1.0" | "M" | "NOT HISPANIC OR LATINO" | "Y" | "Y" | "Y" | "N" | "N" | "N" | "Y" | "" | "" | "28.8" | "25-<30" | "175.3" | "88.5" | "12.0" | "18111.0" | "55.3" | ">=12" | "19792.0" | "2014-03-18" | "2014-04-14" | "5.0" | "19827.0" | "STUDY TERMINATED BY SPONSOR" | "Sponsor Decision" | "23.0" | "Discontinued" | null |
| "CDISCPILOT01" | "01-701-1034" | "1034" | "701" | "701" | "Xanomeline High Dose" | "Xanomeline High Dose" | "81.0" | "Xanomeline High Dose" | "81.0" | "19905.0" | "20087.0" | "183.0" | "76.9" | "14067.0" | "77.0" | "65-80" | "2.0" | "YEARS" | "WHITE" | "1.0" | "F" | "NOT HISPANIC OR LATINO" | "Y" | "Y" | "Y" | "Y" | "Y" | "Y" | "" | "" | "" | "26.1" | "25-<30" | "154.9" | "62.6" | "9.0" | "18899.0" | "32.9" | ">=12" | "19898.0" | "2014-07-01" | "2014-12-30" | "12.0" | "20087.0" | "COMPLETED" | "Completed" | "21.0" | "Completed" | null |
study_plan.datasets["adae"].head()
shape: (5, 55)
| STUDYID | SITEID | USUBJID | TRTA | TRTAN | AGE | AGEGR1 | AGEGR1N | RACE | RACEN | SEX | SAFFL | TRTSDT | TRTEDT | ASTDT | ASTDTF | ASTDY | AENDT | AENDY | ADURN | ADURU | AETERM | AELLT | AELLTCD | AEDECOD | AEPTCD | AEHLT | AEHLTCD | AEHLGT | AEHLGTCD | AEBODSYS | AESOC | AESOCCD | AESEV | AESER | AESCAN | AESCONG | AESDISAB | AESDTH | AESHOSP | AESLIFE | AESOD | AEREL | AEACN | AEOUT | AESEQ | TRTEMFL | AOCCFL | AOCCSFL | AOCCPFL | AOCC02FL | AOCC03FL | AOCC04FL | CQ01NAM | AOCC01FL |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | f64 | str | f64 | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str | str |
| "CDISCPILOT01" | "701" | "01-701-1015" | "Placebo" | "5.397605346934028e-79" | "63.0" | "<65" | "1.0" | "WHITE" | "1.0" | "F" | "Y" | "19725.0" | "19906.0" | "19726.0" | "" | 2.0 | "" | null | "" | "" | "APPLICATION SITE ERYTHEMA" | "APPLICATION SITE REDNESS" | "" | "APPLICATION SITE ERYTHEMA" | "" | "HLT_0617" | "" | "HLGT_0152" | "" | "GENERAL DISORDERS AND ADMINIST… | "GENERAL DISORDERS AND ADMINIST… | "" | "MILD" | "N" | "N" | "N" | "N" | "N" | "N" | "N" | "N" | "PROBABLE" | "" | "NOT RECOVERED/NOT RESOLVED" | "1.0" | "Y" | "Y" | "Y" | "Y" | "" | "" | "" | "DERMATOLOGIC EVENTS" | "Y" |
| "CDISCPILOT01" | "701" | "01-701-1015" | "Placebo" | "5.397605346934028e-79" | "63.0" | "<65" | "1.0" | "WHITE" | "1.0" | "F" | "Y" | "19725.0" | "19906.0" | "19726.0" | "" | 2.0 | "" | null | "" | "" | "APPLICATION SITE PRURITUS" | "APPLICATION SITE ITCHING" | "" | "APPLICATION SITE PRURITUS" | "" | "HLT_0317" | "" | "HLGT_0338" | "" | "GENERAL DISORDERS AND ADMINIST… | "GENERAL DISORDERS AND ADMINIST… | "" | "MILD" | "N" | "N" | "N" | "N" | "N" | "N" | "N" | "N" | "PROBABLE" | "" | "NOT RECOVERED/NOT RESOLVED" | "2.0" | "Y" | "" | "" | "Y" | "" | "" | "" | "DERMATOLOGIC EVENTS" | "" |
| "CDISCPILOT01" | "701" | "01-701-1015" | "Placebo" | "5.397605346934028e-79" | "63.0" | "<65" | "1.0" | "WHITE" | "1.0" | "F" | "Y" | "19725.0" | "19906.0" | "19732.0" | "" | 8.0 | "19734.0" | 10.0 | "3.0" | "DAY" | "DIARRHOEA" | "DIARRHEA" | "" | "DIARRHOEA" | "" | "HLT_0148" | "" | "HLGT_0588" | "" | "GASTROINTESTINAL DISORDERS" | "GASTROINTESTINAL DISORDERS" | "" | "MILD" | "N" | "N" | "N" | "N" | "N" | "N" | "N" | "N" | "REMOTE" | "" | "RECOVERED/RESOLVED" | "3.0" | "Y" | "" | "Y" | "Y" | "" | "" | "" | "" | "" |
| "CDISCPILOT01" | "701" | "01-701-1023" | "Placebo" | "5.397605346934028e-79" | "64.0" | "<65" | "1.0" | "WHITE" | "1.0" | "M" | "Y" | "19210.0" | "19237.0" | "19212.0" | "" | 3.0 | "19235.0" | 26.0 | "24.0" | "DAY" | "ERYTHEMA" | "ERYTHEMA" | "" | "ERYTHEMA" | "" | "HLT_0284" | "" | "HLGT_0192" | "" | "SKIN AND SUBCUTANEOUS TISSUE D… | "SKIN AND SUBCUTANEOUS TISSUE D… | "" | "MILD" | "N" | "N" | "N" | "N" | "N" | "N" | "N" | "N" | "POSSIBLE" | "" | "NOT RECOVERED/NOT RESOLVED" | "1.0" | "Y" | "Y" | "Y" | "Y" | "" | "" | "" | "DERMATOLOGIC EVENTS" | "Y" |
| "CDISCPILOT01" | "701" | "01-701-1023" | "Placebo" | "5.397605346934028e-79" | "64.0" | "<65" | "1.0" | "WHITE" | "1.0" | "M" | "Y" | "19210.0" | "19237.0" | "19212.0" | "" | 3.0 | "" | null | "" | "" | "ERYTHEMA" | "LOCALIZED ERYTHEMA" | "" | "ERYTHEMA" | "" | "HLT_0284" | "" | "HLGT_0192" | "" | "SKIN AND SUBCUTANEOUS TISSUE D… | "SKIN AND SUBCUTANEOUS TISSUE D… | "" | "MODERATE" | "N" | "N" | "N" | "N" | "N" | "N" | "N" | "N" | "PROBABLE" | "" | "NOT RECOVERED/NOT RESOLVED" | "2.0" | "Y" | "" | "" | "" | "" | "" | "" | "DERMATOLOGIC EVENTS" | "" |
3.2 Population
study_plan.get_population_df()
shape: (5, 3)
| name | label | filter |
|---|---|---|
| str | str | str |
| "enrolled" | "All Enrolled Participants" | "" |
| "apat" | "All Participants as Treated" | "adsl:saffl == 'Y'" |
| "itt" | "Intention-to-Treat" | "adsl:ittfl == 'Y'" |
| "all" | null | "" |
| "discon" | null | "adsl:disconfl == 'Y'" |
3.3 Observation
study_plan.get_observation_df()
shape: (4, 3)
| name | label | filter |
|---|---|---|
| str | str | str |
| "week12" | "Week 0 to 12" | "adae:aendy <= 12*7" |
| "week24" | "Week 0 to 24" | "adae:aendy <= 24*7" |
| "pd_major" | "Major Protocol Deviations" | "adpd:DVCAT == 'MAJOR'" |
| "pd_minor" | "Minor Protocol Deviations" | "adpd:DVCAT == 'MINOR'" |
3.4 Parameter
study_plan.get_parameter_df()
shape: (5, 3)
| name | label | filter |
|---|---|---|
| str | str | str |
| "any" | "with one or more adverse event… | "adae:trtemfl == 'Y'" |
| "rel" | "with drug-related adverse even… | "adae:trtemfl == 'Y' and adae:a… |
| "ser" | "with serious adverse events" | "adae:trtemfl == 'Y' and adae:a… |
| "death" | "with adverse events resulting … | "adae:trtemfl == 'Y' and adae:a… |
| "ser_death" | "with serious adverse events re… | "adae:trtemfl == 'Y' and adae:a… |
3.5 Group
study_plan.get_group_df()
shape: (1, 4)
| name | variable | levels | labels |
|---|---|---|---|
| str | str | str | str |
| "trt01a" | "adsl:trt01a" | "[]" | "[]" |
4 Human-in-loop
The YAML framework creates a structured workflow where humans maintain control over critical decisions while leveraging AI for automation:
Human review (paired with a review agent) responsibility can be like.
4.0.1 1. Scientific Validation
- Verify analysis plans align with study protocol and statistical analysis plan (SAP)
- Confirm population definitions match regulatory requirements (e.g., safety population vs. ITT)
- Validate parameter specifications capture the intended clinical outcomes
4.0.2 2. Template Governance
- Approve organization and TA templates through version control.
etc…
The framework ensures that while AI accelerates implementation, humans remain in control of all critical scientific and regulatory decisions.
5 GenAI
GenAI transforms the development process by working from structured YAML specifications rather than free-form requirements:
5.1 AI-Powered Workflow
5.1.1 1. YAML Generation from Natural Language
# Human: "Create AE analysis plan for oncology study with standard safety endpoints"
# AI Output:
plans:
- analysis: "ae_summary"
population: "apat"
observation: ["week12", "week24"]
parameter: "any;rel;ser"
group: "trt01a"
- analysis: "ae_specific"
population: "apat"
observation: ["week12", "week24"]
parameter: ["any", "rel", "ser"] # Cartesian expansion
group: "trt01a"5.1.2 2. Code Generation from YAML
AI converts approved YAML specifications into production-ready code:
# AI generates function implementations
def ae_summary(population="apat", observation="week12", parameter="any;rel;ser"):
"""Generate AE summary table from YAML specification."""
# Extract filters from keywords
pop_filter = keywords.get_population(population).filter
obs_filter = keywords.get_observation(observation).filter
# Apply combined parameter logic
param_filters = [keywords.get_parameter(p).filter for p in parameter.split(';')]
# Generate table
return create_summary_table(pop_filter, obs_filter, param_filters)5.1.3 3. Documentation and Validation
- Auto-generate table shells with proper headers and formatting
- Create validation scripts based on filter specifications
- Generate regulatory documentation (e.g., analysis specifications)
5.2 AI Advantages in this Framework
Structured Approach Benefits
- Reduced Hallucinations: AI works from precise specifications, not ambiguous requirements
- Consistency: Same YAML always produces identical code implementations
- Auditability: Clear traceability from YAML specification to generated output
5.3 Quality Safeguards
5.3.1 1. Specification-First Development
- AI cannot generate code without approved YAML specifications
- All changes tracked through version control
- Clear separation between “what” (YAML) and “how” (code)
5.3.2 2. Template-Driven Consistency
- AI leverages organization templates for standardized approaches
- Reduced variability across studies and therapeutic areas
- Built-in compliance with regulatory requirements
5.3.3 3. Automated Validation
# AI generates validation checks from YAML
def validate_ae_analysis(yaml_spec, output_data):
"""Validate generated analysis against YAML specification."""
# Check population filters
expected_n = apply_filter(source_data, yaml_spec['population'])
assert len(output_data) == len(expected_n)
# Verify parameter logic
# ... additional validation based on YAMLThis GenAI integration ensures that artificial intelligence amplifies human expertise rather than replacing human judgment, creating a reliable and scalable framework for clinical analysis and reporting.