Safety Analyzer

Predict which clinical-stage drugs are most likely to have catastrophic safety failures — based on modality, target, and 30 years of historical patterns.

A pure-Python analysis tool built on a curated dataset of 63 notable drug safety events across 8 therapeutic modalities (1993–2024). No ML, no dependencies — just historical rates and risk multipliers that surface real patterns in drug development failure.

Built by Kyle Stringfellow — environmental consultant turned biotech tools developer.

---

Why This Exists

Drug safety failures destroy shareholder value and harm patients. Certain modalities — gene therapy, CAR-T, ADCs — have well-documented patterns of catastrophic events that repeat across programs. This tool quantifies those patterns so you can assess risk before it materializes.

Key findings from the dataset:

• CAR-T has the highest patient death rate (37.5% of recorded events)
• Gene therapy leads in clinical holds (33.3%) and has the highest base risk score (72/100)
• Cell therapy and small molecules tie for termination rate (60%)
• Safety terminations in cell therapy average -47% stock impact — worst of any modality

Features

• Curated Dataset — 63 safety events with drug name, company, ticker, modality, indication, phase, event type, stock impact, and source citations
• Risk Scoring Engine — Composite 0–100 score from modality base risk, target modifiers, indication category, and clinical phase
• Statistical Analysis — Termination rates, clinical hold frequency, death rates by modality/phase, and stock impact by event type
• Time Trends — Tracks whether newer modalities are getting safer over time (spoiler: it's complicated)
• Company Lookup — Compare any public biotech by ticker against the safety event database
• Web Dashboard — Static HTML visualization of key findings, deployable to GitHub Pages

Quick Start

# No external dependencies — pure Python 3.6+
git clone https://github.com/kstring99/safety-analyzer.git
cd safety-analyzer

View safety statistics by modality

python3 main.py stats

======================================================================
DRUG SAFETY EVENT ANALYSIS
Dataset: 63 events across 8 modalities
======================================================================

--- Safety Termination Rate by Modality ---
Modality                    Events   Terminated     Rate
-------------------------------------------------------
Cell Therapy                     5            3    60.0%
Small Molecule                  10            6    60.0%
mRNA/LNP                         8            3    37.5%
ADC                              7            2    28.6%
Gene Therapy                     9            2    22.2%
Monoclonal Antibody             10            2    20.0%
Bispecific                       6            0     0.0%
CAR-T                            8            0     0.0%

--- Patient Death Events by Modality ---
Modality                    Events   Deaths     Rate
---------------------------------------------------
CAR-T                            8        3    37.5%
Gene Therapy                     9        3    33.3%
Cell Therapy                     5        1    20.0%
Small Molecule                  10        2    20.0%
Monoclonal Antibody             10        1    10.0%
...

Score a drug's safety risk

python3 main.py score --modality gene_therapy --target AAV9 --indication SMA --phase "Phase 1"

======================================================================
SAFETY RISK ASSESSMENT
======================================================================

  Modality:    Gene Therapy
  Target:      AAV9
  Indication:  SMA
  Ind. Class:  rare_disease
  Phase:       Phase 1

--- Risk Components ---
  Modality base risk:       72/100
  Target risk modifier:     1.3x
  Indication risk modifier: 1.1x
  Phase risk modifier:      1.3x

  OVERALL SAFETY RISK SCORE: 64/100 [HIGH]
  [################################------------------]

--- Comparable Historical Events (5) ---

  [1999] OTC transgene (adenovirus vector) — University of Pennsylvania
    Patient Death
    Jesse Gelsinger died from a massive immune reaction to the adenoviral vector...

  [2021] Zynteglo (betibeglogene autotemcel) — bluebird bio
    Clinical Hold (stock: -37.5%)
    Clinical hold placed after two patients developed myelodysplastic syndrome...

Filter and browse safety events

python3 main.py events --modality car_t
python3 main.py events --event-type patient_death
python3 main.py events --year 2023

Look up a company's safety profile

python3 main.py compare BLUE
python3 main.py compare MRNA

Web Dashboard

Open index.html in a browser or visit the live GitHub Pages site to see an interactive visualization of:

• Risk scores by modality (bar chart)
• Patient death rates by modality
• Safety termination rates
• Stock impact by event type
• Full event timeline

Project Structure

safety-analyzer/
├── main.py              # CLI interface (4 commands: stats, score, events, compare)
├── safety_events.json   # Curated dataset — 63 events, 8 modalities, 1993–2024
├── safety_stats.py      # Statistical analysis (termination, holds, deaths, stock impact, trends)
├── risk_scorer.py       # Risk scoring engine (base risk × target × indication × phase)
├── index.html           # Web dashboard — dark theme, static HTML/JS
└── README.md

How the Risk Score Works

The scoring algorithm combines four factors into a 0–100 composite score:

Factor	Range	Example
Modality base risk	30–72	Gene Therapy = 72, Small Molecule = 30
Target modifier	1.0–1.5x	Adenovirus = 1.5x, CD20 = 1.0x
Indication modifier	0.85–1.5x	Healthy volunteers = 1.5x, Heme malignancy = 0.85x
Phase modifier	0.8–1.3x	Phase 1 = 1.3x, Post-market = 0.8x

score = min(100, base × target × indication × phase × 100 / 210)

Modalities Covered

Modality	Base Risk	Key Risks	Notable Events
Gene Therapy	72	Immunogenicity, hepatotoxicity, insertional oncogenesis	Gelsinger (1999), Zolgensma liver failure, bluebird MDS
CAR-T	68	CRS, ICANS, T-cell malignancy	JCAR015 cerebral edema deaths, secondary cancers
Cell Therapy	55	GVHD, contamination, manufacturing failures	Allogeneic transplant mortality, Provenge (-95% stock)
Bispecific	52	CRS, ICANS, infections from B-cell depletion	High CRS rates (39–72%) across approved products
ADC	48	Payload toxicity (ILD, VOD, skin), off-target effects	Mylotarg withdrawal, Enhertu ILD, Blenrep keratopathy
mRNA/LNP	38	Myocarditis, reactogenicity, immunogenicity	COVID vaccine myocarditis, CureVac failure (-52%)
Monoclonal Antibody	35	PML, immune-mediated AEs, cytokine storm	TGN1412 catastrophe, Tysabri PML (-43%)
Small Molecule	30	Organ toxicity, cardiovascular risk	Vioxx (88K+ excess CV events), Fen-Phen ($21B settlements)

Data Sources

All events sourced from public records:

• FDA safety communications, clinical holds, and prescribing information
• Published clinical trial results (NEJM, Lancet, JAMA, Blood)
• SEC filings and company press releases
• MHRA and EMA regulatory actions
• CDC VAERS data

Design Decisions

• No ML — Historical rates and risk multipliers are more interpretable and auditable than black-box models for this dataset size
• No dependencies — Pure Python standard library only. Runs anywhere Python 3.6+ exists
• Public data only — Every event includes a source citation
• Extensible — JSON dataset and modular scoring make it straightforward to add new events or modalities

Author

Kyle Stringfellow — Environmental consultant building biotech and data analysis tools. Learning to code by solving real problems in drug safety, environmental compliance, and scientific data management.

• GitHub: @kstring99