Independent population-PK analysis — run R-20261008-141300-3341

sensitivity template pk_2cmt_oral_v1 v1.0.3 · child of R-20261008-134356-4ad9 · status completed · 2026-10-08T14:13:14Z

Sensitivity analysis. This run departs from the applicant's declared methodology on: blq_method.
This report shows differences and generic possible sources of difference. It does not draw conclusions about the applicant's analysis; interpretation remains with the assessor.

1. Pre-comparison checks

Check	Status	Detail
convergence	pass	Optimizer message: Normal exit from bobyqa
covariance step	pass	Covariance matrix obtained (method \|r\|,\|s\| (full))
RSE feasibility	pass	All RSE below 50%
parameter bounds	pass	All estimates and variances within plausible bounds
estimation diagnostics	warn	nlmixr2 reported: gradient problems with covariance; see $scaleInfo; since sandwich matrix is corrected, you may compare to $covR or $covS if you wish; S matrix non-positive definite but corrected by S = sqrtm(S%%S); R matrix non-positive definite but corrected by R = sqrtm(R%%R); last objective function was not at minimum, possible problems in optimization; ETAs were reset to zero during optimization; (Can control by foceiControl(resetEtaP=.))

2. Comparison with the applicant's reported values

Relative difference = (independent − applicant) / applicant. Flag when |Δ| > 20%.

Parameter	Applicant	Independent	95% CI	Δ	Applicant value in CI
CL/F (L/h)	4.2	4.14	3.85 – 4.46	-1.4%	yes
V2/F (L)	35	35.2	31.1 – 39.8	+0.6%	yes
Q/F (L/h)	8.1	7.97	7.55 – 8.42	-1.6%	yes
V3/F (L)	120	115	108 – 122	-4.4%	yes
KA (1/h)	1.1	1.13	0.948 – 1.36	+3.1%	yes
IIV CL (CV%)	30	27.4	–	-8.8%	–

Possible sources of difference

No parameter differs from the applicant's value by more than the tolerance. Generic factors are listed for completeness.

Facts about this run

BLQ method used in this run: M3; 7 censored observations used; BLQ fraction 1.7% overall, 6.2% of samples with NTIME ≥ 24 h.
Actual sampling times used; 9 records deviate from nominal time beyond the QC tolerance.
3 records excluded and 11 records corrected during data preparation (see QC log).
Estimation: FOCEi, template pk_2cmt_oral_v1.
Software: R 4.6.1, nlmixr2 7.0.1, nlmixr2est 7.1.0, rxode2 5.1.7.1.

Generic factors

Handling of BLQ observations — M1 (discarding) and M3 (likelihood-based censoring) use different information from late samples, which mostly informs terminal-phase parameters (V3/F, Q/F) and CL/F.
Time variables and rounding — Use of actual vs nominal times, and rounding of sampling times, changes the observation design seen by the model.
Data exclusions and corrections — Records excluded or corrected during data preparation (pre-dose records, unit conversions, outliers) may differ from the applicant's data handling.
Estimation settings — Estimation algorithm, interaction term, initial estimates, parameterisation and optimiser tolerances can lead to different local optima.
Software and versions — Different software (e.g. NONMEM vs nlmixr2) or versions may give numerically different estimates for the same model.

3. Parameter estimates

Parameter	Estimate	RSE %	95% CI	IIV CV%	η-shrinkage SD%	Fixed
CL/F (L/h)	4.143	3.73	3.85 – 4.46	27.4	4.77	no
V2/F (L)	35.2	6.33	31.1 – 39.8	25.6	20.2	no
Q/F (L/h)	7.973	2.79	7.55 – 8.42	–	–	no
V3/F (L)	114.7	2.95	108 – 122	–	–	no
KA (1/h)	1.134	9.16	0.948 – 1.36	29.1	35.1	no
Residual: add.sd	8.09	34	–	–	–	no
Residual: prop.sd	0.1437	3.74	–	–	–	no

RSE on the natural scale ≈ 100 × SE(log θ); 95% CI = exp(log θ ± 1.96 SE). OFV = 4397.2982 · AIC = 5189.2065 · 60 subjects, 420 observations (7 censored) · covariance |r|,|s| (full) · estimation time 11.3 s.

4. Goodness-of-fit

Individual profiles (first six subjects)

5. Data QC and preparation

Every exclusion or correction is listed. Raw records: 483 → prepared: 480 (420 observations, 60 subjects).

Check	Outcome	Records	Detail	Subjects
required_columns	pass	0	All 11 required columns present
units	fix	4	Converted DV and LLOQ from ug/mL to ng/mL	SYN-052
duplicates	pass	0	No duplicate records
chronology	pass	0	Every subject has a dose record (60 subjects)
records_before_dose	exclude	3	Observation records before the first dose excluded (3 with a quantified concentration > 0, which is not expected before dosing)	SYN-007, SYN-023, SYN-041
ntime_vs_time	warn	9	Actual time deviates from nominal by more than max(0.5 h, 10% of NTIME); records kept, actual TIME used	SYN-005, SYN-007, SYN-012, SYN-014, SYN-019, SYN-025, SYN-035, SYN-046
missing_covariates	pass	0	WT available for all subjects
blq_proportion	pass	7	7 BLQ records = 1.7% of observations (6.2% of samples with NTIME >= 24 h)
blq_handling	fix	7	BLQ method M3: BLQ records kept as left-censored (DV set to LLOQ, CENS = 1)

6. Analysis spec (confirmed by the assessor)

Confirmed 2026-10-08T14:13:00Z by djindetoss (GitHub) (/confirm on issue #1).

{
  "compartments": 2,
  "absorption": "first_order",
  "route": "oral",
  "error_model": "combined",
  "iiv": [
    "CL",
    "V2",
    "KA"
  ],
  "estimation": "FOCEi",
  "reference_values": {
    "CL": 4.2,
    "V2": 35.0,
    "Q": 8.1,
    "V3": 120.0,
    "KA": 1.1,
    "IIV_CL": 30.0
  },
  "analysis_type": "sensitivity",
  "blq_method": "M3",
  "template": "pk_2cmt_oral_v1"
}

7. Traceability manifest

Run ID / parent	`R-20261008-141300-3341` / `R-20261008-134356-4ad9` (root `R-20261008-134033-5040`)
Template	`pk_2cmt_oral_v1` v1.0.3, catalogue 1.0.3, template hash `589f254e36326ab2…`
Script SHA-256	`fdffd00a137b56c3daa89e23c5ac8b89b3918ed8e8ca2f41756e74f8b8f8b03b`
Raw data SHA-256	`1cb7adff311dcd25660f2353d892964c98ad1932a6b69b6ed28c6ebfcaa48c6b`
Prepared data SHA-256	`38d30404d452db3275fe44a8f1a96a4e78d3d35377ba67e53bfba0a645a67ff2`
Seed	20240607
Extractor	anthropic (claude-opus-5-5, prompt system_v2) jobs J-71710b716b
R packages	nlmixr2 7.0.1, nlmixr2est 7.1.0, rxode2 5.1.7.1, lotri 1.0.5, ggplot2 4.0.3, jsonlite 2.0.0, R 4.6.1, BLAS /usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3, LAPACK /usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.26.so
Python packages	python 3.12.3, jinja2 3.1.4, pandas 2.2.3, numpy 2.1.3, jsonschema 4.23.0, typer 0.12.5
Environment	Linux-6.17.0-1022-azure-x86_64-with-glibc2.39 · image ghcr.io/djindetoss/pk-analysis-poc@sha256:dd39ca7673a48f1e3f7b6df134be2b4f4ae00ef4e0a726894a72cbdbdf295b4b · commit n/a

Full manifest.json

{
  "run_id": "R-20261008-141300-3341",
  "parent_run_id": "R-20261008-134356-4ad9",
  "root_run_id": "R-20261008-134033-5040",
  "purpose": "analysis",
  "analysis_type": "sensitivity",
  "sensitivity_fields": [
    "blq_method"
  ],
  "status": "completed",
  "created_at": "2026-10-08T14:13:00Z",
  "completed_at": "2026-10-08T14:13:14Z",
  "spec": {
    "compartments": 2,
    "absorption": "first_order",
    "route": "oral",
    "error_model": "combined",
    "iiv": [
      "CL",
      "V2",
      "KA"
    ],
    "estimation": "FOCEi",
    "reference_values": {
      "CL": 4.2,
      "V2": 35.0,
      "Q": 8.1,
      "V3": 120.0,
      "KA": 1.1,
      "IIV_CL": 30.0
    },
    "analysis_type": "sensitivity",
    "blq_method": "M3",
    "template": "pk_2cmt_oral_v1"
  },
  "spec_hash": "0030756b6dc1154888296838e116220220efa71e79063db429abd47172f527db",
  "methodology_hash": "15de490dc0bac871c1176633475cba04a3be6b1269daf85223fe38a62eed9bde",
  "template": {
    "id": "pk_2cmt_oral_v1",
    "version": "1.0.3",
    "hash": "589f254e36326ab2ae18dce251949169b318f7e59c79210d33428f972dc38600",
    "catalogue_version": "1.0.3"
  },
  "script_hash": "fdffd00a137b56c3daa89e23c5ac8b89b3918ed8e8ca2f41756e74f8b8f8b03b",
  "seed": 20240607,
  "input_data": {
    "raw_file": "data/raw/adpc_synthetic.csv",
    "raw_hash": "1cb7adff311dcd25660f2353d892964c98ad1932a6b69b6ed28c6ebfcaa48c6b",
    "prepared_hash": "38d30404d452db3275fe44a8f1a96a4e78d3d35377ba67e53bfba0a645a67ff2"
  },
  "extractor": {
    "name": "anthropic",
    "model": "claude-opus-5-5",
    "prompt_version": "system_v2",
    "job_ids": [
      "J-71710b716b"
    ]
  },
  "confirmation": {
    "at": "2026-10-08T14:13:00Z",
    "by": "djindetoss (GitHub)",
    "mode": "/confirm on issue #1"
  },
  "packages": {
    "python": {
      "python": "3.12.3",
      "jinja2": "3.1.4",
      "pandas": "2.2.3",
      "numpy": "2.1.3",
      "jsonschema": "4.23.0",
      "typer": "0.12.5"
    },
    "r": {
      "nlmixr2": "7.0.1",
      "nlmixr2est": "7.1.0",
      "rxode2": "5.1.7.1",
      "lotri": "1.0.5",
      "ggplot2": "4.0.3",
      "jsonlite": "2.0.0",
      "R": "4.6.1",
      "BLAS": "/usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3",
      "LAPACK": "/usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.26.so"
    }
  },
  "environment": {
    "platform": "Linux-6.17.0-1022-azure-x86_64-with-glibc2.39",
    "machine": "x86_64",
    "cpu": "Intel(R) Xeon(R) Platinum 8370C CPU @ 2.80GHz",
    "math_env": {
      "OPENBLAS_CORETYPE": "Prescott",
      "OPENBLAS_NUM_THREADS": "1",
      "GLIBC_TUNABLES": "glibc.cpu.hwcaps=-AVX2_Usable,-AVX512F_Usable,-FMA_Usable,-FMA4_Usable,-AVX2,-AVX512F,-FMA,-FMA4"
    },
    "container_image": "ghcr.io/djindetoss/pk-analysis-poc@sha256:dd39ca7673a48f1e3f7b6df134be2b4f4ae00ef4e0a726894a72cbdbdf295b4b",
    "git_commit": null,
    "python_executable": "/opt/venv/bin/python"
  },
  "tool_version": "0.1.0",
  "settings": {
    "tolerance": 0.2,
    "rse_threshold": 50.0,
    "qc": {
      "ntime_abs_tol_h": 0.5,
      "ntime_rel_tol": 0.1,
      "blq_warn_fraction": 0.1
    }
  },
  "engine": {
    "returncode": 0,
    "elapsed_s": 14.2,
    "resumed_from_checkpoint": false
  },
  "outputs": {},
  "comparable": true
}

Generated 2026-10-08T14:13:14Z by pk-analysis-poc 0.1.0 — synthetic data, demonstrator only.