clinical-trial-ipd-sim

Name: clinical-trial-ipd-sim
Author: RConsortium

88stars20forks

How to install clinical-trial-ipd-sim?

rconsortium/pharma-skills/clinical-trial-ipd-sim

$npx -y skills add RConsortium/pharma-skills --skill clinical-trial-ipd-sim

Installs into the current project.

Summary

End-to-end R/pharmaverse workflow to simulate individual patient data (IPD) for a registered clinical trial using a g-formula causal-DAG simulator. Given an NCT ID with posted protocol + results, derives SDTM-style CRFs, builds evidence-based structural causal models, parameteriz

Files · 1

View on GitHub

SKILL.md

1	# Clinical Trial IPD Causal-DAG Simulator
2
3	End-to-end R workflow for generating individual-patient-level CRF data for a
4	registered clinical trial. The output is a set of CSV CRFs whose marginal
5	statistics match the published trial results AND whose joint distribution
6	follows an explicit causal DAG, so the data are usable for downstream
7	causal-inference methodology.
8
9	The implementation target is R-first and pharmaverse-compatible. The Python
10	FLAURA2 build used during development remains a worked example of the causal
11	logic, but new work should create an R package-style project unless the user
12	explicitly asks for Python. See [r_implementation.md](r_implementation.md)
13	for the R module layout and package roles.
14
15	This skill generalizes the FLAURA2 causal-DAG workflow into an
16	R/pharmaverse implementation target.
17
18	## When to use
19
20	The user provides an NCT ID and the trial has both a posted protocol and
21	posted results. Examples:
22
23	- "Simulate IPD for NCT04035486" → use this skill
24	- "Generate CRFs for the AEGEAN trial" → look up NCT, then use this skill
25	- "Create a digital-twin cohort for trial X" → use this skill
26	- "I need synthetic patients matching the published KM curves of …" → use this skill
27
28	Do not use when the user only wants summary-level reconstruction (Cox HR,
29	KM medians, AE proportions). For that, reach for tabular reconstruction
30	methods (e.g., reconstructed IPD from KM digitization).
31
32	## Required environment
33
34	- R 4.3+ with `renv` for dependency locking.
35	- Core simulation/analysis: `dplyr`, `tidyr`, `purrr`, `readr`, `tibble`,
36	`lubridate`, `stringr`, `rlang`, `survival`, `flexsurv`, `broom`, `jsonlite`.
37	- Pharmaverse/CDISC stack:
38	- `sdtm.oak` for SDTM-oriented CRF/domain construction from generated source records.
39	- `admiral` for ADaM derivations such as ADSL, ADAE, ADLB, ADTTE.
40	- `haven` for reading/writing SAS transport-adjacent inputs and SAS datasets when needed.
41	- `tidytlg` for TLG generation from ADaM outputs.
42	- `xportr` for XPT metadata handling and transport export.
43	- `datasetjson` for Dataset-JSON export when requested.
44	- Network access for the ClinicalTrials.gov API (`https://clinicaltrials.gov/api/v2/studies/{nct}`)
45	- Citation-capable literature and standards lookup for causal-parent
46	evidence, natural-history rates, priors, CTCAE, and RECIST references.
47	Record stable source identifiers for all non-CTGov evidence; do not rely
48	on model memory for cited claims.
49	- `WebFetch` or equivalent only for fetching specific known URLs
50	(the ClinicalTrials.gov API, a protocol PDF, a publication supplement)
51
52	## Workflow — six steps
53
54	### Step 1 · Intake from ClinicalTrials.gov
55
56	1. Fetch the trial JSON record:
57	`https://clinicaltrials.gov/api/v2/studies/{NCT_ID}?format=json`
58	2. Extract:
59	- Design: arms, randomization ratio, stratification factors, primary/secondary endpoints
60	- Eligibility: inclusion/exclusion criteria → baseline population priors
61	- Outcomes table: per-arm event counts, medians, hazard ratios with 95% CIs
62	- Adverse events table: per-arm SOC/PT counts at any-grade, Gr ≥3, SAE
63	- Protocol document URL (under `ProtocolSection.IPDSharingStatementModule` or `LargeDocumentsModule`)
64	3. Locate the schedule of activities (SoA) in the protocol PDF/HTML.
65	If the SoA cannot be parsed automatically, ask the user to point at the
66	relevant pages, or scaffold a default visit grid based on the dosing schedule.
67	If ClinicalTrials.gov does not provide a protocol, search for the protocol
68	or supplementary appendix in the New England Journal of Medicine (NEJM).
69	If neither source has a protocol, ask the user to provide one and do not
70	proceed until it is available.
71	4. Persist the intake to `intake/{NCT_ID}.json` so later steps don't re-fetch.
72
73	Output of step 1: structured trial summary with design, endpoint targets,
74	AE targets, and SoA visit grid.
75
76	### Step 2 · CRF derivation from the protocol
77
78	Read the protocol's SoA and generate a CDISC-style CRF schema:
79
80	\| Form \| Visits collected \| Variables \| Source \|
81	\|---\|---\|---\|---\|
82	\| Demographics \| Screening \| AGE, SEX, RACE, COUNTRY, ARM \| Protocol §X \|
83	\| Cancer/Disease History \| Screening \| Diagnosis, stage,

People who liked this also viewed

6 picks

Type

clinical-trial-ipd-sim

88stars20forks

View on GitHub

How to install clinical-trial-ipd-sim?

rconsortium/pharma-skills/clinical-trial-ipd-sim

$npx -y skills add RConsortium/pharma-skills --skill clinical-trial-ipd-sim

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Clinical Trial IPD Causal-DAG Simulator
2
3	End-to-end R workflow for generating individual-patient-level CRF data for a
4	registered clinical trial. The output is a set of CSV CRFs whose marginal
5	statistics match the published trial results AND whose joint distribution
6	follows an explicit causal DAG, so the data are usable for downstream
7	causal-inference methodology.
8
9	The implementation target is R-first and pharmaverse-compatible. The Python
10	FLAURA2 build used during development remains a worked example of the causal
11	logic, but new work should create an R package-style project unless the user
12	explicitly asks for Python. See [r_implementation.md](r_implementation.md)
13	for the R module layout and package roles.
14
15	This skill generalizes the FLAURA2 causal-DAG workflow into an
16	R/pharmaverse implementation target.
17
18	## When to use
19
20	The user provides an NCT ID and the trial has both a posted protocol and
21	posted results. Examples:
22
23	- "Simulate IPD for NCT04035486" → use this skill
24	- "Generate CRFs for the AEGEAN trial" → look up NCT, then use this skill
25	- "Create a digital-twin cohort for trial X" → use this skill
26	- "I need synthetic patients matching the published KM curves of …" → use this skill
27
28	Do not use when the user only wants summary-level reconstruction (Cox HR,
29	KM medians, AE proportions). For that, reach for tabular reconstruction
30	methods (e.g., reconstructed IPD from KM digitization).
31
32	## Required environment
33
34	- R 4.3+ with `renv` for dependency locking.
35	- Core simulation/analysis: `dplyr`, `tidyr`, `purrr`, `readr`, `tibble`,
36	`lubridate`, `stringr`, `rlang`, `survival`, `flexsurv`, `broom`, `jsonlite`.
37	- Pharmaverse/CDISC stack:
38	- `sdtm.oak` for SDTM-oriented CRF/domain construction from generated source records.
39	- `admiral` for ADaM derivations such as ADSL, ADAE, ADLB, ADTTE.
40	- `haven` for reading/writing SAS transport-adjacent inputs and SAS datasets when needed.
41	- `tidytlg` for TLG generation from ADaM outputs.
42	- `xportr` for XPT metadata handling and transport export.
43	- `datasetjson` for Dataset-JSON export when requested.
44	- Network access for the ClinicalTrials.gov API (`https://clinicaltrials.gov/api/v2/studies/{nct}`)
45	- Citation-capable literature and standards lookup for causal-parent
46	evidence, natural-history rates, priors, CTCAE, and RECIST references.
47	Record stable source identifiers for all non-CTGov evidence; do not rely
48	on model memory for cited claims.
49	- `WebFetch` or equivalent only for fetching specific known URLs
50	(the ClinicalTrials.gov API, a protocol PDF, a publication supplement)
51
52	## Workflow — six steps
53
54	### Step 1 · Intake from ClinicalTrials.gov
55
56	1. Fetch the trial JSON record:
57	`https://clinicaltrials.gov/api/v2/studies/{NCT_ID}?format=json`
58	2. Extract:
59	- Design: arms, randomization ratio, stratification factors, primary/secondary endpoints
60	- Eligibility: inclusion/exclusion criteria → baseline population priors
61	- Outcomes table: per-arm event counts, medians, hazard ratios with 95% CIs
62	- Adverse events table: per-arm SOC/PT counts at any-grade, Gr ≥3, SAE
63	- Protocol document URL (under `ProtocolSection.IPDSharingStatementModule` or `LargeDocumentsModule`)
64	3. Locate the schedule of activities (SoA) in the protocol PDF/HTML.
65	If the SoA cannot be parsed automatically, ask the user to point at the
66	relevant pages, or scaffold a default visit grid based on the dosing schedule.
67	If ClinicalTrials.gov does not provide a protocol, search for the protocol
68	or supplementary appendix in the New England Journal of Medicine (NEJM).
69	If neither source has a protocol, ask the user to provide one and do not
70	proceed until it is available.
71	4. Persist the intake to `intake/{NCT_ID}.json` so later steps don't re-fetch.
72
73	Output of step 1: structured trial summary with design, endpoint targets,
74	AE targets, and SoA visit grid.
75
76	### Step 2 · CRF derivation from the protocol
77
78	Read the protocol's SoA and generate a CDISC-style CRF schema:
79
80	\| Form \| Visits collected \| Variables \| Source \|
81	\|---\|---\|---\|---\|
82	\| Demographics \| Screening \| AGE, SEX, RACE, COUNTRY, ARM \| Protocol §X \|
83	\| Cancer/Disease History \| Screening \| Diagnosis, stage,

People who liked this also viewed

6 picks

Type