Concept Architecture
Concept
Theoretically, a Flexible Parametric Model is a survival modelling approach that represents the baseline hazard or cumulative hazard using smooth spline functions rather than assuming a fixed parametric distribution. It was developed to provide greater flexibility than conventional parametric survival models while retaining the ability to extrapolate beyond observed follow-up. In health economics, flexible parametric models are widely used to estimate long-term survival for cost-effectiveness analyses and health technology assessments when standard distributions provide an inadequate fit.
Mathematically, flexible parametric models express the logarithm of the cumulative hazard or cumulative odds as a restricted cubic spline function of log time. The spline coefficients are estimated using maximum likelihood, allowing the hazard function to assume a wide range of shapes while remaining smooth. The mathematical framework accommodates proportional and non-proportional hazards and supports prediction of survival, hazards and restricted mean survival.
In practice, flexible parametric models are fitted to individual patient survival data using statistical software. The number and placement of spline knots are selected to balance model fit and complexity, with competing models evaluated using statistical criteria, visual inspection and clinical plausibility. The fitted model is then used to estimate survival probabilities, extrapolate long-term outcomes and generate inputs for economic models.
Purpose
Used to model complex survival patterns, improve goodness-of-fit over conventional parametric survival models and generate clinically plausible long-term survival projections for health economic evaluation.
Mathematical Formulae
Primary Formula
ln(H(t)) = s(ln(t); ?)
where:
- H(t) = cumulative hazard function
- s(ln(t); ?) = restricted cubic spline function of log time
- ? = estimated spline coefficients
Supporting Formulae
S(t) = exp(?H(t))
h(t) = dH(t) / dt
Model parameters are estimated by maximising the log-likelihood:
?? = arg max L(?)
Related Mathematical Methods
- Restricted cubic splines
- Maximum likelihood estimation
- Survival analysis
- Parametric survival modelling
- Royston-Parmar modelling
- Model selection using AIC and BIC
Example
A clinical trial of an immunotherapy follows patients for four years, but the economic evaluation requires lifetime survival estimates. Standard Weibull and Gompertz models fail to capture the changing hazard over time. A flexible parametric model with four spline knots is fitted, providing a substantially better fit to the observed data and more clinically plausible long-term survival projections for estimating lifetime QALYs and costs.
Excel Implementation
| Function | Example Formula | Health Economics Application |
|---|---|---|
| LN | =LN(A2) | Calculate log time for spline-based survival modelling. |
| IF | =IF(A2>Knot1,(A2-Knot1)^3,0) | Construct truncated cubic spline basis functions. |
| EXP | =EXP(-B2) | Convert cumulative hazard estimates into survival probabilities. |
| LINEST | =LINEST(Y_range,X_range,TRUE,TRUE) | Estimate simplified regression coefficients for exploratory modelling. |
VBA (Optional)
Automate the construction of spline basis functions, estimation of survival predictions and comparison of alternative knot configurations for economic modelling.
Sources
- Royston P, Parmar MKB. Flexible parametric proportional-hazards and proportional-odds models for censored survival data, with application to prognostic modelling and estimation of treatment effects.
- Lambert PC, Royston P. Further development of flexible parametric models for survival analysis.
- Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation.
- NICE. Health Technology Evaluation Manual.
- ISPOR-SMDM Modeling Good Research Practices Task Force Reports.
Related Concepts (3)
Library
Publications
1
NICE DSU Technical Support Document 14: Survival analysis for economic evaluations alongside clinical trials – extrapolation with patient-level data — Nicholas R. Latimer, TSD 14 ed., 2013 (NICE Decision Support Unit (University of Sheffield))
The reference guidance on survival analysis for economic evaluation: fitting standard parametric models (exponential, Weibull, Gompertz, log-logistic, log-normal) to censored trial data and extrapolating to estimate lifetime survival benefit, with a process guide for model selection and justification.
Frequently Asked Questions (6)
What is a flexible parametric model?
A class of survival models using flexible functions, such as restricted cubic splines, to represent the hazard or survival function beyond standard distributions.
Source: Royston & Parmar 2002
Why can standard distributions be too rigid for some survival data?
A standard survival distribution imposes one fixed shape on the hazard over time, such as always rising or always falling, which may not match data whose risk changes direction or has more than one turning point. Flexible parametric models relax this by using adjustable functions, typically splines, that can bend to follow complex hazard patterns while remaining fully parametric and so usable for extrapolation. They occupy a middle ground between rigid standard forms and the non-parametric Kaplan-Meier estimate. Royston and Parmar (2002) developed this approach.
Source: Royston & Parmar 2002
How do flexible parametric models work?
Flexible parametric models work by modelling a transformation of the survival function, commonly the log cumulative hazard, as a flexible function of log time using restricted cubic splines, with knots placed along the time scale. The splines allow the hazard to take complex shapes, while the parametric form gives a smooth, fully specified model that can be extrapolated. Covariates can be included, and the number and placement of knots control the flexibility, so the model adapts to the data's hazard pattern.
Source: Royston & Parmar 2002
What are the advantages of flexible parametric models?
Flexible parametric models can represent complex hazard shapes, such as those that rise then fall, that standard distributions cannot, giving better fit where the hazard is not simple. Being fully parametric, they provide smooth survival and hazard functions and can be extrapolated, unlike non-parametric methods. They accommodate covariates and time-varying effects. This combination of flexibility and parametric structure makes them useful in survival analysis and extrapolation where simple distributions fit poorly but a smooth, extendable model is wanted.
Source: Latimer 2013
What is the role of splines in flexible parametric models?
Splines, typically restricted cubic splines, are the means by which flexible parametric models achieve their flexibility: they represent the log cumulative hazard as a smooth, piecewise function of log time, able to bend at specified knots to follow complex hazard shapes. The knots determine where the function can change, and their number controls the flexibility, more knots allowing more complex shapes but risking overfitting. Splines thus let the model depart from the fixed forms of standard distributions while remaining smooth and parametric.
Source: Royston & Parmar 2002
What are the limitations of flexible parametric models?
Flexible parametric models require choices about the number and placement of knots, which affect the fit and the extrapolation, and too many knots risk overfitting the observed data and producing implausible projections beyond it. Their flexibility within the data does not guarantee sensible extrapolation, since the fitted shape may not hold in the tail. They are also more complex than simple distributions. These limitations mean knot choices are made carefully and the extrapolated behaviour scrutinised for plausibility, as with other survival models.
Source: Latimer 2013
Trust Record
Verified by Dr Darrin Baines
British health economist
Professional identity: darrinbaines.org
Verification date: 20 Oct 2025
Content version: 1.0.0
Canonical Identity
- Term code
- HE-EM-SM-023
Stable URI · Machine-readable · Resolvable · CC BY 4.0