Concept Architecture
Concept
Theoretically, a Unimodal Hazard is a hazard function characterised by a single maximum over time, such that the hazard increases to one peak before subsequently declining or, less commonly, declines to a minimum before increasing no further. In survival analysis, the term most commonly refers to hazard functions exhibiting a single peak in the instantaneous event rate. Unimodal hazards arise naturally in many biological and clinical processes where risk initially accumulates before diminishing because susceptible individuals experience the event or treatment effects stabilise.
Mathematically, a unimodal hazard is represented by a hazard function possessing a single local maximum over the observed time domain. Numerous parametric survival distributions, including the log-normal and log-logistic distributions, can produce unimodal hazard functions depending on their parameter values. The defining mathematical property is the existence of one turning point where the first derivative of the hazard function changes from positive to negative.
In practice, unimodal hazards are identified through exploratory hazard estimation, kernel-smoothed hazard plots or fitted parametric survival models. In health economic modelling they are important because selecting a survival distribution capable of representing a unimodal hazard may substantially improve long-term extrapolation compared with models that assume monotonic hazards.
Purpose
Used to represent survival processes in which the instantaneous event risk rises to a single peak before declining, allowing appropriate selection and validation of parametric survival models for health economic evaluation.
Mathematical Formulae
Primary Formula
There is no universally recognised canonical mathematical formula.
Supporting Formulae
Hazard function:
h(t) = f(t) / S(t)
Turning point defining a unimodal hazard:
dh(t) / dt = 0
with:
d�h(t) / dt� < 0
at the unique maximum.
Related Mathematical Methods
- Parametric Survival Analysis
- Hazard Function Estimation
- Kernel Hazard Estimation
- Log-Normal Survival Model
- Log-Logistic Survival Model
- Survival Extrapolation
Example
A health technology assessment evaluates overall survival after treatment for aggressive lymphoma.
Estimated hazard rates increase from 0.012 per month at treatment initiation to a peak of 0.054 per month after 14 months before declining to 0.026 per month by month 36.
A log-normal survival model produces a unimodal hazard that closely matches the observed clinical data, whereas an exponential model fails to capture the peak in mortality risk.
Excel Implementation
| Function | Example Formula | Health Economics Application |
|---|---|---|
| MAX | =MAX(B2:B61) | Identify the peak estimated hazard |
| MATCH | =MATCH(MAX(B2:B61),B2:B61,0) | Locate the time of maximum hazard |
| FORECAST.LINEAR | =FORECAST.LINEAR(A62,B2:B61,A2:A61) | Explore hazard trends during model assessment |
| INDEX | =INDEX(A2:A61,MATCH(MAX(B2:B61),B2:B61,0)) | Return the time corresponding to the peak hazard |
| Scatter Chart | Hazard versus time | Visualise whether the estimated hazard is unimodal |
VBA (Optional)
A VBA procedure can automatically identify the peak hazard, classify hazard shape and compare fitted parametric survival models against observed hazard estimates.
Sources
- Collett D. Modelling Survival Data in Medical Research.
- Klein JP, Moeschberger ML. Survival Analysis: Techniques for Censored and Truncated Data.
- Royston P, Parmar MKB. Flexible Parametric Proportional-Hazards and Proportional-Odds Models for Censored Survival Data.
- Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation.
- NICE. Health Technology Evaluation Manual.
Related Concepts (2)
Library
Publications
1
NICE DSU Technical Support Document 15: Cost-effectiveness modelling using patient-level simulation — Davis, Stevenson, Tappenden & Wailoo, TSD 15 ed., 2014 (NICE Decision Support Unit (University of Sheffield))
Guidance on individual patient-level (microsimulation) cost-effectiveness modelling — when to use it in preference to cohort models, how to structure it, and how to handle the associated computational and uncertainty challenges.
Frequently Asked Questions (6)
What is an unimodal hazard?
A hazard function that rises to a single peak and then declines over follow-up, rather than a monotonic or bathtub pattern.
Source: Collett 2015
What is a unimodal hazard?
A unimodal hazard is a hazard function that rises to a single peak and then declines over follow-up, rather than following a monotonic pattern that only rises or falls, or a bathtub pattern with high risk at both ends. The risk increases initially, reaches a maximum at some time, and then decreases. Unimodal hazards arise where risk builds, peaks, and then falls, and they are represented by distributions such as the log-logistic and log-normal, which can capture this turning shape.
Source: Collett 2015
What distributions have a unimodal hazard?
Distributions with a unimodal hazard, rising to a peak then declining, include the log-logistic distribution, when its shape parameter exceeds one, and the log-normal distribution, both of which can represent a hazard that turns. These contrast with distributions having monotonic hazards, such as the exponential, Weibull, and Gompertz, which only rise or fall. Where the hazard peaks and then declines, the log-logistic or log-normal is appropriate, since monotonic distributions cannot capture the turning shape of a unimodal hazard.
Source: Collett 2015
When does a unimodal hazard arise?
A unimodal hazard arises where the risk of an event increases at first, reaches a peak, and then decreases, which occurs in various clinical situations, such as recovery processes where risk builds then subsides, or conditions where mortality is highest at an intermediate time. It differs from monotonic risk that only rises or falls and from a bathtub pattern high at both ends. Recognising a unimodal hazard in the data indicates that a distribution able to represent a turning hazard is needed.
Source: Kalbfleisch & Prentice 2002
How does a unimodal hazard differ from a monotonic hazard?
A unimodal hazard rises to a single peak and then declines, changing direction once, whereas a monotonic hazard either only rises or only falls throughout, without turning. Monotonic hazards are represented by distributions like the exponential, Weibull, and Gompertz, while unimodal hazards require distributions such as the log-logistic or log-normal that can peak and decline. The distinction matters for choosing a survival distribution, since a monotonic model cannot capture a hazard that turns, and vice versa.
Source: Collett 2015
Why does identifying a unimodal hazard matter?
Identifying a unimodal hazard matters because it determines which survival distributions can represent the data: a hazard that rises then falls requires a distribution such as the log-logistic or log-normal, whereas monotonic distributions would fit poorly and misrepresent the risk, biasing extrapolation. Choosing a distribution whose hazard shape matches the unimodal pattern is important for both fit and projected survival. Recognising the unimodal shape from the estimated hazard therefore guides the selection of an appropriate model.
Source: Collett 2015
Trust Record
Verified by Dr Darrin Baines
British health economist
Professional identity: darrinbaines.org
Verification date: 24 Oct 2025
Content version: 1.0.0
Canonical Identity
- Persistent URI
- https://healtheconomics.wiki/concept/unimodal-hazard
- Term code
- HE-EM-SM-089
Stable URI · Machine-readable · Resolvable · CC BY 4.0