Abstract:In this work we propose a stochastic differential equation (SDE) for modelling health related quality of life (HRQoL) over a lifespan. HRQoL is assumed to be bounded between 0 and 1, equivalent to death and perfect health, respectively. Drift and diffusion parameters of the SDE are chosen to mimic decreasing HRQoL over life and ensuring epidemiological meaningfulness. The Euler-Maruyama method is used to simulate trajectories of individuals in a population of n = 1000 people. Age of death of an individual is simulated as a stopping time with Weibull distribution conditioning the current value of HRQoL as time-varying covariate. The life expectancy and health adjusted life years are compared to the corresponding values for German women.
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is to describe the health - related quality of life (HRQoL) throughout an individual's life by establishing a stochastic differential equation (SDE) model. Specifically, the paper focuses on the following aspects:
1. **Modeling of HRQoL**: The paper proposes to use stochastic differential equations to model HRQoL, which is different from traditional deterministic models. HRQoL is assumed to vary between 0 and 1, corresponding to the states of death and perfect health respectively.
2. **Parameter selection**: To ensure the rationality and biological significance of the model, the paper selects specific drift term \(b(t, X_t)\) and diffusion term \(\sigma(t, X_t)\). These parameters are designed to simulate the trend of HRQoL decreasing with age and ensure that the HRQoL value is always between 0 and 1.
3. **Simulation of death time**: The death time of an individual is simulated as a stopping time \(\tau\), which follows a Weibull distribution and is conditional on the current HRQoL value \(X_t\) as a time - dependent covariate.
4. **Population - level analysis**: By simulating the trajectories of 1,000 individuals, the paper calculates life expectancy and health - adjusted life years (HALY) and compares them with the actual data of German women to verify the validity of the model.
5. **Theory and practical application**: The paper not only proposes a theoretical model but also discusses the problem of how to estimate SDE parameters from actual data, although this part has not been elaborated in detail.
### Specific form of the model
The Ito stochastic differential equation proposed in the paper is as follows:
\[dX_t = b(t, X_t)dt+\sigma(t, X_t)dB_t\]
where:
- \(X_t\) represents the HRQoL of an individual at age \(t\).
- \(b(t, X_t)\) is the drift term, defined as \(b(t, X_t)=X_t\delta(t)\), where \(\delta(t)\) is a function that changes with age.
- \(\sigma(t, X_t)\) is the diffusion term, defined as \(\sigma(t, X_t) = 0.05\sqrt{(1 - X_t)X_t}\).
- \(B_t\) is a standard Brownian motion.
### Selection of drift term and diffusion term
- **Drift term \(b(t, X_t)\)**: The drift term \(b(t, X_t)=X_t\delta(t)\), where \(\delta(t)\) is a function that changes with age, as shown in Figure 1. This function gradually decreases before the age of 85 and then decreases rapidly.
- **Diffusion term \(\sigma(t, X_t)\)**: The diffusion term \(\sigma(t, X_t)=0.05\sqrt{(1 - X_t)X_t}\), which ensures that \(X_t\) is always between 0 and 1.
### Simulation of death time
The death time \(\tau\) of an individual is simulated as a stopping time, and its hazard rate \(h(t)\) is defined as:
\[h(t)=\frac{h_0(t)}{\sqrt{X_t}}\]
where \(h_0(t)=\exp(\alpha+\beta t)\), and the parameters \(\alpha\) and \(\beta\) are determined by the general mortality data of Germany.
### Simulation results
By simulating the trajectories of 1,000 individuals, the paper obtains the following results:
- The median lifespan is 83.19 years, and the interquartile range (IQR) is 74.62 to 88.63 years.
- The median HALY is 72.23 years, and the interquartile range (IQR) is 66.43 to 76.80 years.
These results are quite close to the actual data of German women, indicating that the model has certain practical value.
### Discussion