Clinical Development Review

Integrated Clinical Development Review — NTX-101

Integrated cross-phase evidence package for NTX-101 in mild-to-moderate Alzheimer's dementia.

Integrated report Program overview Data lineage Model review Phase I Phase II Phase III

Program

NTX-101 / NTX-101

Raw backbone

12 XPT domains + 32 SafetyData exports

Analysis layer

6 ADaM-like datasets

ML result

Phase II ROC AUC 0.868

Executive conclusion

The integrated review presents a coherent development path: Phase I GO based on manageable safety findings, Phase II GO supported by directional efficacy with formal treatment-vs-placebo testing, and Phase III GO based on superiority-style time-to-event evidence with dose-consistent treatment effect.

Importantly, the asset under study is NTX-101, a fictional investigational therapy inspired by public source structure.

Biopharma operating model reflected in the repo

Immutable raw inputs
XPT / RDA

Normalized working layer
interim parquet

Standards-aware harmonization
SDTM-like

Bridged development continuity
synthetic, documented

Analysis-ready datasets
ADaM-like

Phase gates & memos
decision-facing outputs

Source strategy and standards posture

Source	Role	Coverage	Industry relevance
CDISC Pilot XPT	Core public structural backbone	DM, AE, LB, VS, EX, DS plus ADaM-like examples	Sponsor-style submission structure and standards vocabulary
SafetyData RDA	Supplemental safety and standards reference	32 exported parquet datasets	Additional clinical programming realism and R interoperability
Synthetic continuity bridge	Phase continuity and endpoint completion	Subject master, phase assignments, efficacy longitudinal, event outcomes	Provides a continuous clinical program narrative where public source coverage is incomplete
Thin R reference layer	Standards-facing cross-check	Reference ADSL/ADAE derivations and row-count comparisons	Provides an independent cross-check of selected derivations against an alternate standards-facing workflow

Key metrics and formal evidence snapshot

Metric	Value
Phase I gate	GO
Phase II gate	GO
Phase II primary model	logistic_regression
Phase II ROC AUC	0.868
Phase II responder uplift	19.07 pp
Phase II primary endpoint p-value	0.0019
Phase III median delay to worsening	83.58 days
Phase III log-rank p-value	0.0421

The inferential layer includes treatment comparisons, p-values, confidence interval summaries, benchmark model comparison, and a documented primary model choice to support cross-functional development review.

The predictive modeling package remains interpretable, clinically contextualized, and anchored to a prespecified threshold rationale.

Methodology

Stage(s)	Focus	Conclusion
1	Operating model, source inventory, ingestion	Program contract frozen; 12 XPT domains and 32 SafetyData exports normalized to parquet.
2	Contracts, harmonization, continuity bridge	Schema validation added; harmonization preserved row counts across 6 core domains; synthetic bridge created 306 subject-level records and 163 Phase III event records.
3	ADaM-like layer and R reference	ADSL/ADAE/ADLB/ADVS/ADEFF/ADTTE produced; R reference comparisons showed row deltas of 0 for ADSL and 0 for ADAE.
4	Phase I early development review	Gate = GO. Phase I now supports progression with manageable on-treatment safety burden, no treatment-related death imbalance, and explicit statistical comparison versus placebo retained in the report.
5	Phase II endpoint and interpretable ML	Gate = GO. Week 24 responder separation was kept in a more plausible advancement range (difference = 19.07 pp) while the primary logistic model was tuned to a clinically justified threshold and good discrimination.
6	Phase III confirmatory review	Formal Kaplan-Meier, log-rank, and Cox-model analyses support superiority-style interpretation with hazard ratio control below 1.0 and dose-consistent treatment effect.
7	Report and final QC	Static review package, lineage manifest, final validation report, and completed stage tracker delivered; 13 tracked stages marked complete.

Harmonization and QC snapshot

Domain	Input rows	Output rows
dm	306	306
ae	1191	1191
lb	59580	59580
vs	29643	29643
ex	591	591
ds	596	596

Visual snapshots

Phase I safety review

Reframed as a sponsor-style safety graphic with ordered arms and restrained executive palette.

Phase II efficacy and ML

Benchmarking supports conservative model selection while keeping logistic regression primary for interpretability.

Phase III confirmatory view

Left-to-right time-to-event display with arm-specific curves and executive-style visual treatment.

Open the main artifacts

Phase I safety review

Safety thresholds, AE rate comparison, and early development gate.

Phase II endpoint and ML review

Hypothesis testing, responder logic, benchmarks, calibration, and model card.

Phase III confirmatory review

KM view, time-to-worsening test, safety integration, and final recommendation.

Data lineage page

Raw -> interim -> harmonized -> synthetic -> derived traceability.

Model card page

Inputs, exclusions, leakage controls, metrics, and limitations.