HAZARD

Parametric hazard analysis for time-to-event data.

HAZARD is a statistical computing package developed at Cleveland Clinic for fitting flexible, multi-phase parametric hazard models to survival data. It implements the Conservation of Events theorem (Turner et al.) to produce models with early, constant, and late hazard phases — commonly applied to cardiac surgery outcomes and other clinical research.

Originally developed at the University of Alabama Birmingham. Currently developed and maintained at The Cleveland Clinic Foundation.

---

Project Metadata & Links

• Authors
• License
• Installation Guide
• Changelog
• Acknowledgments

---

Overview

HAZARD fits three-phase parametric hazard models of the form:

h(t | X) = μ₁(X) · SG₁(t)   [early phase]
          + μ₂(X)             [constant phase]
          + μ₃(X) · SG₃(t)   [late phase]

where each scale factor μⱼ(X) = exp(intercept_j + βⱼᵀ X) and the shaping functions SG₁, SG₃ are flexible parametric forms supporting a wide range of hazard shapes.

Key capabilities:

• Multi-phase hazard modeling (early, constant, late phases)
• Right censoring, interval censoring, and repeating events
• Time-varying covariates and weighted events
• Automatic stepwise covariate selection (forward, backward, stepwise)
• Conservation of Events theorem for numerically stable parameter estimation
• Covariance and correlation matrix estimation
• SAS integration via XPORT format and PROC HAZARD / PROC HAZPRED syntax

Tested SAS versions: v8.x through v9.3.2

---

Prerequisites

Requirement	Purpose
SAS (v8 or later)	Data preparation and result post-processing
GCC or compatible C compiler	Building from source
Flex / GNU Bison	Lexer and parser generation (build only)
GNU Make	Build automation
Autoconf / Automake	Configure script generation (build only)

SAS is required at runtime for data I/O. The C executables handle all numerical computation.

---

Installation

From Source (Unix/Linux/macOS)

# 1. Clone or download the source
git clone https://github.com/ehrlinger/hazard
cd hazard

# 2. Configure (default installs to /usr/local/hazard)
./configure

# If building from a fresh git checkout after autotools changes:
# autoreconf -fi

# Optional developer flags:
./configure --enable-warnings
./configure --enable-sanitizers

# On macOS, if your shell exports a broken Homebrew gcc-14 as CC,
# configure now falls back to clang automatically.
# Set HAZARD_KEEP_CC=1 to keep the exported compiler.

# To specify a custom install directory:
./configure --prefix=/path/to/install

# 3. Compile
make

# Or rebuild the configured tree with ASan + UBSan:
make asan

# 4. Smoke-test the built executables with staged local fixtures
./tests/run_local_example.sh hazard
./tests/run_local_example.sh hazpred

# Keep staged temp files/logs for inspection:
./tests/run_local_example.sh hazard --keep

# Notes:
# - The helper validates success by checking expected output artifacts.
# - hazard fixture runs may return non-zero while still producing output;
#   this is treated as a successful smoke test when output is present.

# 5. Install
make install

If you are building on a platform not previously tested, please report your experience to hazard@bio.ri.ccf.org with the platform details.

From Source (Windows via MSYS2)

Windows source builds are supported via MSYS2 UCRT64 (same toolchain used in CI releases).

1. Install MSYS2 and open the MSYS2 UCRT64 shell.
2. Install build dependencies:

pacman -S --needed --noconfirm \
  autoconf automake bison flex make \
  mingw-w64-ucrt-x86_64-gcc

3. Build and install:

# Run this when building from git source after autotools edits
autoreconf -fi

./configure --prefix="$PWD/dist/hazard"
make -j"$(nproc)"
make install

To build with assertions disabled (define NDEBUG) on Windows:

./configure --prefix="$PWD/dist/hazard" CPPFLAGS="-DNDEBUG"
make -j"$(nproc)"
make install

From Binary Distribution

Unix/Linux/macOS:

cd $HOME   # or your preferred install parent
gzip -d < /path/to/hazard.PLATFORM.tar.gz | tar -xvf -

This creates a hazard/ directory with the standard layout.

Windows:

Extract the .zip distribution with your preferred archive tool to a root directory such as C:\. The distribution must be extracted to a root-level directory due to path length constraints.

Official release archives (including Windows .zip) are published in GitHub Releases for this repository. Legacy downloads may also exist at: https://www.lerner.ccf.org/quantitative-health/software/

Verify Release Checksums

Each GitHub release includes a SHA256SUMS.txt file. Verify downloaded archives before extracting.

Unix/Linux/macOS:

# Linux (sha256sum)
sha256sum -c SHA256SUMS.txt

# macOS fallback if sha256sum is unavailable
shasum -a 256 -c SHA256SUMS.txt

Windows (PowerShell):

# Example: verify one archive against SHA256SUMS.txt
$file = "hazard-vX.Y.Z-windows-x64.zip"
$actual = (Get-FileHash -Algorithm SHA256 $file).Hash.ToLower()
$expected = (Select-String -Path SHA256SUMS.txt -Pattern [regex]::Escape($file)).Line.Split(" ")[0].ToLower()
if ($actual -eq $expected) { "Checksum OK" } else { "Checksum MISMATCH" }

---

Environment Setup

HAZARD requires four environment variables pointing to its installation directories.

Unix/Linux/macOS

Add to your shell startup file (.profile, .bash_profile, .zshrc, or .cshrc):

# bash / zsh / sh / ksh
export HAZARD=/usr/local/hazard        # or $HOME/hazard
export HAZAPPS=$HAZARD/bin
export HZEXAMPLES=$HAZARD/examples
export MACROS=$HAZARD/macros

# csh / tcsh
setenv HAZARD /usr/local/hazard
setenv HAZAPPS $HAZARD/bin
setenv HZEXAMPLES $HAZARD/examples
setenv MACROS $HAZARD/macros

Windows

Set the following User or System environment variables via Control Panel → System → Advanced → Environment Variables:

Variable	Recommended Value	Description
TMPDIR	%TEMP%	Temporary file storage
HAZARD	C:\hazard	Root installation directory
HAZAPPS	%HAZARD%\bin	Executables
HZEXAMPLES	%HAZARD%\examples	Example SAS programs and data
MACROS	%HAZARD%\macros	SAS helper macros

A system restart may be required for the variables to take effect.

SAS Configuration

HAZARD integrates with SAS via its autocall macro library. You must add HAZAPPS to the sasautos path in your SAS configuration file (sasv8.cfg, sasv9.cfg, or similar).

Locate your SAS config file:

• Unix: ~/sasv?.cfg (in your home directory)
• Windows: C:\Program Files\SAS Institute\SAS\V8\ or My Documents\My SAS Files\V8\. Search for sas*.cfg if unsure.

Unix — edit sasautos to add !HAZAPPS:

-sasautos ( '!SASROOT8/sasautos' '!HAZAPPS' )

Windows — edit sasautos and add -noxwait:

-SET SASAUTOS (
  "!sasroot\core\sasmacro"
  "!sasext0\graph\sasmacro"
  "!sasext0\stat\sasmacro"
  "!HAZAPPS"
)

/* Allow external commands to close automatically */
-noxwait

The -noxwait option prevents SAS from leaving a DOS window open after each external call.

---

Build Verification

After building from source, you can run the test suite runner directly:

# Unit tests
make -C tests/unit clean all check

# Integration-only run (skip memory/concurrency suites)
HAZARD_BIN=$PWD/src/hazard/hazard.exe \
  bash tests/run_all_tests.sh --integration-only --skip-memory --skip-concurrent

If your shell exports a missing compiler path (for example CC=/opt/homebrew/bin/gcc-14), override test builds with CC=cc.

---

Quick Start

Run one of the included example programs to verify your installation:

/* From SAS, with HZEXAMPLES on your path */
%include '!HZEXAMPLES/hm.death.AVC.sas';

This fits a three-phase hazard model to the AVC (atrioventricular canal) surgical outcomes dataset (310 observations, 70 events). Expected output:

Log likelihood = -181.109
Events conserved = 70.000
Convergence attained after 6 iterations, 9 function evaluations

See the examples/ directory for additional programs demonstrating different model configurations.

---

Project Structure

hazard/
├── src/                    C source code
│   ├── common/             Global data structures and shared utilities
│   ├── llike/              Log-likelihood engine (setlik.c, setcoe.c, setobj.c)
│   ├── optim/              BFGS/quasi-Newton optimizer
│   ├── model/              Early and late phase shaping functions
│   ├── vars/               Stepwise variable selection
│   ├── vcov/               Variance-covariance estimation (Cholesky)
│   ├── stat/               Statistical distributions
│   ├── utils/              Error handling and I/O utilities
│   ├── hazard/             PROC HAZARD executable
│   └── hazpred/            PROC HAZPRED executable
├── docs/                   Documentation
│   ├── vignettes/                  User-facing guides (Markdown)
│   │   ├── introduction.md         Hazard function analysis primer
│   │   ├── installation.md         Build, install, SAS config, testing
│   │   ├── examples.md             Annotated walkthrough of included examples
│   │   ├── hazard.md               PROC HAZARD reference
│   │   ├── hazpred.md              PROC HAZPRED reference
│   │   └── overview.md             Parametric hazard overview
│   ├── Claude_CODEBASE_ANALYSIS.md     Architectural overview (internal)
│   ├── Claude_MODERNIZATION_GUIDE.md   C modernization roadmap (internal)
│   └── Claude_R_MIGRATION_GUIDE.md     R migration roadmap (internal)
├── examples/               Example SAS programs and datasets
│   ├── hm.death.AVC.sas    Three-phase model: AV-canal repair mortality
│   ├── hp.death.AVC.sas    Hazard prediction example
│   └── data/               Example XPORT datasets
├── macros/                 SAS helper macros
└── tests/                  Reference test outputs

---

Documentation

User Vignettes

Vignette	Description
Introduction	What is hazard function analysis; multi-phase model overview
Downloads	Binary distributions and source code
Installation	Build, install, configure for Unix/Windows/SAS, run the test suite
Examples	Annotated walkthrough of all included example programs
SAS Utility Macros	Reference for all helper macros

Procedure References

Document	Description
Overview of Procedures	Parametric-hazard system overview and mathematical foundation
The HAZARD Procedure	PROC HAZARD — complete syntax, options, and computational method
The HAZPRED Procedure	PROC HAZPRED — patient-level prediction from fitted parameter estimates

Rendered HTML and PDF builds of the procedure references are not currently shipped. The Markdown files above carry the authoritative content.

Developer Documentation

Document	Description
docs/Claude_CODEBASE_ANALYSIS.md	Architectural overview for contributors
docs/Claude_MODERNIZATION_GUIDE.md	C modernization roadmap
docs/Claude_R_MIGRATION_GUIDE.md	R migration roadmap

---

Citation

If you use HAZARD in published research, please cite:

Blackstone EH, Naftel DC, Turner ME Jr. The decomposition of time-varying hazard into phases, each incorporating a separate stream of concomitant information. J Am Stat Assoc. 1986;81(395):615–624.

Turner ME Jr, Restrepo C, McCarville C, Hauck WW. The continuity correction for the logistic disease risk model. Stat Med. 1992.

---

License

HAZARD is free software distributed under the GNU General Public License v2. See COPYING for the full license text.

Copyright © 2000 The Cleveland Clinic Foundation.

---

Contact

General inquiries	hazard@bio.ri.ccf.org
Clinical/statistical questions	Dr. Eugene H. Blackstone — blackse@ccf.org
Bug reports	hazard@bio.ri.ccf.org
Phone	216.444.6712

Contributions and platform compatibility reports are welcome. If you successfully build HAZARD on a new platform, please write to let us know.