betaFeedback

WorksMathematics

Causality and order restricted inference in the Aalen additive hasard model

Abstract

We study a graphical model (DAG) with time to event as a response variable. We use the Aalen additive hazard model for modelling the effect on the response variable of covariates, combined with a multiple linear regression for earlier layers in a DAG. This enables modelling covariates affecting the event process both directly and indirectly. The indirect component effect is represented by a product θ1(t)β2(t)θ_1(t)β_2(t), where θ1(t)θ_1(t) describes the effect of the independent covariate on the dependent covariate and β2(t)β_2(t) is the effect of the dependent covariate on the final time to response variable, in the additive hazard model. We estimate β2(t)β_2(t) by the order-restricted estimator recently introduced in Anevski and Merai (arXiv:2606.23882v2), thereby enabling disease models with early onset or late onset covariate effects. Combining our previously obtained results with the DAG allows for modelling early onset or late onset covariate effects together with direct and indirect covariate effects, which should be of particular interest for studying causality for complex diseases. We present a nonparametric plug-in estimator of the indirect effect. We derive the limit distribution of the plug-in estimator of the indirect component, and show that it has n1/3n^{1/3} pointwise asymptotic rate and a limit distribution which is proportional to the Chernoff distribution.

Introduction

We consider a survival analysis model, in which we treat right censored data, combined with a DAG. Thus we let T≥0T \ge0 be a positive continuous random variable with an unknown distribution function FF, where TT models the time to an event of interest. We assume that we observe the minimum ti=min⁡(Ti,Ci)t_i=\min(T_i,C_i) of the time TiT_i and a censoring time CiC_i, together with an indicator for the time being exact, δi=1{Ti≤Ci}\delta_i=1\{T_i\le C_i\}.

Introduce the individual counting processes Ni(t)=1{ti≤t,δi=1}N_i(t)=1\{t_i\le t,\delta_i=1\} for which one has the stochastic differential equation

dNi(t)=Yi(t)h(t) dt+dMi(t),(1)dN_i(t)=Y_i(t)h(t)\,dt+dM_i(t), \tag*{(1)}

for i=1,…,ni=1,\ldots,n. Here h(t)h(t) is the individual hasard function, which exists when the distribution of TT is absolutely continuous, and is then given by h(t)=−ddtlog⁡(1−F(t))h(t)=-\frac{d}{dt}\log(1-F(t)), where Yi(t)=1{ti≥t}Y_i(t)=1\{t_i\ge t\} is the left-continuous indicator process for the individual being at risk at time t−t-.

Also MiM_i is the individual martingale, i.e. satisfying E(dMi(t)∣Ft)=0E(dM_i(t)\mid\mathcal{F}_t)=0, and where the sigma algebras {Ft,t≥0}\{\mathcal{F}_t,t\ge0\} is a filtration storing the information available at times tt.

The σ\sigma-algebra generated by the information depends on the amount and type of information about the observed times that is available and the amount of information about the covariates that is available. Starting with the model (1) satisfying E(dMi(t)∣Ft)=0E(dM_i(t)\mid\mathcal{F}_t)=0 one needs to establish that E(dMi(t)∣Gt)=0E(dM_i(t)\mid\mathcal{G}_t)=0 where Gt\mathcal{G}_t is the σ\sigma-algebra generated by the observed and available information at time tt. The concept of noninformative and independent censoring as well as the innovation theorem are used to establish this link. We do not make this assumptions explicit here, since they are of no relevance to us, and refer to reader to a standard reference such as [3], cf. also [5]. We will in the sequel assume that Ft\mathcal{F}_t is a filtration, containing the information available at time tt, and that E(dMi(t)∣Ft)=0E(dM_i(t)\mid\mathcal{F}_t)=0 holds so that E(dNi(t)∣Ft)=Yi(t)h(t) dtE(dN_i(t)\mid\mathcal{F}_t)=Y_i(t)h(t)\,dt.

A basic inference problem in survival analysis is assessing the effect of group indicators or continuous covariate measurements on the distribution of the time to an event. It is then necessary to model the distribution function depending on covariates x1,…,xpx_1,\ldots,x_p. This can be done using various models. One standard model which is particularly convenient in our framework is the Aalen additive hasard model

h(t)=β0(t)+β1(t)x1+…+βp(t)xp,(2)h(t)=\beta_0(t)+\beta_1(t)x_1+\ldots+\beta_p(t)x_p, \tag*{(2)}

where β0,…,βp\beta_0,\ldots,\beta_p are unknown functions. Thus β=(β0,…,βp)\beta=(\beta_0,\ldots,\beta_p) is the unknown parameter vector of functions and Ep+1E^{p+1} is the parameter space, where E={g:[0,∞)→R,∫g<∞}E=\{g:[0,\infty)\to\mathbb{R},\int g<\infty\} is the set of integrable functions on [0,∞)[0,\infty).

The Aalen model is immediately generalisable to time dependent covariates xk(t)x_k(t), which we will allow for in the sequel, and they may even be stochastic processes; the only necessary demand is that Xk(t)X_k(t) is Ft\mathcal{F}_t measurable, for every kk.

A principal difficulty in estimation in the Aalen model is that the coefficient functions are local functions and can not be estimated directly by nonparametric procedures. Instead, one turns to estimating the cumulative coefficient functions

Bk(t)=∫0tβk(u) du.(3)B_k(t) = \int_0^t \beta_k(u)\,\mathrm{d}u. \tag*{(3)}

The ordinary Aalen estimator, defined below in (10), provides an estimator B^k\widehat{B}_k of BkB_k.

In [5] we proposed an order-restricted estimator for the coefficient functions βk\beta_k under a monotonicity assumption. The estimator is obtained by projecting the local increments of the Aalen estimator B^k\widehat{B}_k, defined in (10), onto the cone of non-increasing sequences. Under some regularity conditions we showed in [5] that the resulting pointwise estimator has a nonstandard n1/3n^{1/3} asymptotic rate and a Chernoff-type limit distribution.

One motivation for introduction order restricted inference of the βk\beta_k functions is in order to model early onset covariate effects. In fact, we note that βk\beta_k being decreasing means that the effect of (an increase in) the covariate value xkx_k decreases with time or age.

A further motivation of the work in [5], was as a first step in the use of a graphical model in a survival analysis setting, thereby allowing for a categorising of the covariate effects as direct versus indirect effect. This was first suggested in [8], see also [9].

In this paper we incorporate our previously derived limit distribution results into such a graphical model setting and present estimators of momentanous direct and indirect effects, as well as limit distribution properties for the estimated effects. This has to our knowledge not been done before.

Thus the purpose of the present paper is to study a natural extension of a standard Aalen model, in which one of the covariates acts through an intermediate, possibly time dependent, covariate. This leads to a decomposition of the effect of a covariate into a direct component and an indirect component. For simplicity we will treat the simplest possible model, with two covariates. The generalisation to more complex models is straightforward.

More precisely, let X1(t)X_1(t) denote an (independent) covariate and let X2(t)X_2(t) denote another (dependent) covariate. We consider the structural representation

dNi(t)=Yi(t){β0(t)+β1(t)X1(t)+β2(t)X2(t)}dt+dMi(t),(4)dN_i(t) = Y_i(t)\{\beta_0(t) + \beta_1(t)X_1(t) + \beta_2(t)X_2(t)\}\mathrm{d}t + \mathrm{d}M_i(t), \tag*{(4)}
X2(t)=θ0(t)+θ1(t)X1(t)+ε(t),(5)X_2(t) = \theta_0(t) + \theta_1(t)X_1(t) + \varepsilon(t), \tag*{(5)}

where NiN_i is the event counting process, YiY_i is the at-risk process, Xi(t)X_i(t) are Ft\mathcal{F}_t-adapted processes and MiM_i is a martingale with respect to the filtration Ft\mathcal{F}_t.

The hazard model is

h(t)=β0(t)+β1(t)X1(t)+β2(t)X2(t).(6)h(t) = \beta_0(t) + \beta_1(t)X_1(t) + \beta_2(t)X_2(t). \tag*{(6)}

Substitution of the covariate equation (5) into (6) gives

h(t)={β0(t)+β2(t)θ0(t)}+{β1(t)+θ1(t)β2(t)}X1(t)+β2(t)ε(t).h(t) = \{\beta_0(t) + \beta_2(t)\theta_0(t)\} + \{\beta_1(t) + \theta_1(t)\beta_2(t)\}X_1(t) + \beta_2(t)\varepsilon(t).

Thus, at the level of this structural decomposition, the coefficient multiplying X1(t)X_1(t) is

β1(t)⏟direct component+θ1(t)β2(t)⏟indirect component.(7)\underbrace{\beta_1(t)}_{\text{direct component}} + \underbrace{\theta_1(t)\beta_2(t)}_{\text{indirect component}}. \tag*{(7)}

The two pathways represented in (7) are illustrated in Figure 1.

Figure 1. Dynamic path diagram illustrating the direct and indirect pathways from X1(t)X_1(t) to the event process N(t)N(t).

We thus define the instantaneous indirect effect of covariate X1(t)X_1(t) as

I(t)=θ1(t)β2(t).I(t) = \theta_1(t)\beta_2(t).

Our main question is the point wise asymptotic behaviour of the plug-in estimator

I^n(t)=θ^1(t)β~2(t)\widehat{I}_n(t) = \widehat{\theta}_1(t)\widetilde{\beta}_2(t)

of the instantaneous indirect effect I(t)I(t), where β~2\widetilde{\beta}_2 is the order restricted estimator defined in [5], and where θ^1\widehat{\theta}_1 is a least squares estimator, to be defined in the sequel.

We will show below that the two factors θ^1\widehat{\theta}_1 and β~2\widetilde{\beta}_2 have different asymptotic rates. In fact, for each fixed tt, the estimator θ^1(t)\widehat{\theta}_1(t) is obtained by ordinary least squares regression and satisfies a standard n1/2n^{1/2} distributional limit, whereas β~2(t)\widetilde{\beta}_2(t) has a nonstandard n1/3n^{1/3} limit distribution behaviour that is common in order restricted inference. Since n−1/2=o(n−1/3)n^{-1/2}=o(n^{-1/3}), estimation of θ1(t)\theta_1(t) does not contribute to the first order asymptotic distribution of the estimated indirect effect I^n(t)\widehat{I}_n(t).

The main result therefore transfers the Chernoff-type limit of β~2(t)\widetilde{\beta}_2(t) directly to the indirect component, with the limit multiplied by the fixed value θ1(t)\theta_1(t). The proof is straightforward once the pointwise result for the order restricted estimator is available, but the formulation is useful because it identifies precisely which part of the estimation procedure determines the asymptotic behaviour.

The paper is organised as follows. Section 2 introduces the model and the estimators. Section 3 derives the limit distribution for the new estimator. Section 4 discusses the interpretation and limitations of the result.

Model and estimation

Assume we have data (ti,δi,x1i(t),x2i(t))(t_i,\delta_i,x_{1i}(t),x_{2i}(t)) for i=1,…,ni=1,\ldots,n for nn individuals, which are seen as observations in the counting process Aalen model (4), which satisfies the structural equations (5). Note that we allow the observed covariates x1i,x2ix_{1i},x_{2i} to be time-dependent and arbitrary. The only requirement needed is that they are observations of Ft\mathcal{F}_t-adapted processes, so that the martingale structure is maintained under stochastic integrations. We assume we have observations on a fixed compact time interval [0,τ][0,\tau].

Now for each time point t∈[0,τ]t\in[0,\tau], let J(t)={i:Yi(t)=1}J(t)=\{i:Y_i(t)=1\} be the risk set, i.e. the index set of those individuals that have not yet experienced the event. Since we do not have left-truncated data, it is clear that J(t)J(t) is a set of indices that is decreasing with time, so that J(t)⊇J(s)J(t)\supseteq J(s) if t<st<s. Furthermore the cardinality of J(t)J(t) decreases with one at each observation time tit_i, so that ∣J(ti+)∣=∣J(ti)∣−1\lvert J(t_i+)\rvert=\lvert J(t_i)\rvert-1, assuming that we have no ties, which we hereby assume.

Therefore, for each fixed observed time point tit_i, for i=1,…,ni=1,\ldots,n, the structural equations (5) give rise to a set of linear equations

X2j(ti)=θ0(ti)+θ1(ti)X1j(ti)+εj(ti),(8)X_{2j}(t_i)=\theta_0(t_i)+\theta_1(t_i)X_{1j}(t_i)+\varepsilon_j(t_i), \tag*{(8)}

for j∈J(ti)j \in J(t_i), and we note that there are ni=∣J(ti)∣n_i = \lvert J(t_i)\rvert many such equations. Therefore, for each fixed tit_i, we can estimate the parameters θ0(ti),θ1(ti)\theta_0(t_i), \theta_1(t_i) by ordinary least squares (ols) regression. We denote the ols estimates by (θ^0(ti),θ^1(ti))(\hat{\theta}_0(t_i), \hat{\theta}_1(t_i)). Since we can solve the system of equations (8) for every time point tit_i, we will obtain nonparametric estimates of the full parameter function θ0(t)\theta_0(t) and θ1(t)\theta_1(t), and these are

θ^0:=(θ^0(t(1)),…,θ^0(t(n))),\hat{\theta}_0 := \left(\hat{\theta}_0(t_{(1)}), \ldots, \hat{\theta}_0(t_{(n)})\right),
θ^1:=(θ^1(t(1)),…,θ^1(t(n))).\hat{\theta}_1 := \left(\hat{\theta}_1(t_{(1)}), \ldots, \hat{\theta}_1(t_{(n)})\right).

In fact θ^0\hat{\theta}_0 and θ^1\hat{\theta}_1 are the ols estimate of the vectors θ0:=(θ0(t(1)),…,θ0(t(n)))\theta_0 := \left(\theta_0(t_{(1)}), \ldots, \theta_0(t_{(n)})\right) and θ1:=(θ1(t(1)),…,θ1(t(n)))\theta_1 := \left(\theta_1(t_{(1)}), \ldots, \theta_1(t_{(n)})\right).

Note that even if the covariates would be time-constant, the ols estimates (θ^0(ti),θ^1(ti))(\hat{\theta}_0(t_i), \hat{\theta}_1(t_i)) will typically be different for different tit_i, since the risk set J(ti)J(t_i) changes with tit_i, and thus the set of linear equations (8) will change as well (the number of equations will change).

Note also that if the covariates would be time-dependent then we could formulate system of equation (8) for every fixed t∈(0,τ)t \in(0,\tau), with tit_i replaced by tt and the index set J(ti)J(t_i) replaced by J(t)J(t). We would then get ols estimates of θ^0(t)\hat{\theta}_0(t) and θ^1(t)\hat{\theta}_1(t) of θ0(t)\theta_0(t) and θ1(t)\theta_1(t). Since we can do this for any fixed tt we actually get nonparametric estimates of the parameter functions, under very mild regularity conditions on the covariates, e.g. that they are piecewise continuous.

Then a multivariate central limit theorem will give a limit distribution result for the 2n2n-length vector (θ^0,θ^1)(\hat{\theta}_0,\hat{\theta}_1) as

n1/2((θ^0,θ^1)−(θ0,θ1))→dU,n^{1/2}\left((\hat{\theta}_0,\hat{\theta}_1)-(\theta_0,\theta_1)\right) \xrightarrow{d} U,

on R2n\mathbb{R}^{2n} where UU is a mean zero 2n2n-dimensional Gaussian random vector, with a certain covariance structure. It is also straightforward to show that we have a process result

n1/2((θ^0(⋅),θ^1(⋅))−(θ0(⋅),θ1(⋅)))→dU(⋅),n^{1/2}\left((\hat{\theta}_0(\cdot),\hat{\theta}_1(\cdot))-(\theta_0(\cdot),\theta_1(\cdot))\right) \xrightarrow{d} U(\cdot),

on D[0,τ]2D[0,\tau]^2, where U(⋅)U(\cdot) is a mean zero Gaussian process in D[0,τ]2D[0,\tau]^2, with a certain covariance function, under mild regularity conditions on the covariates.

For the purpose of the results presented in this paper, only the limit distribution of θ^1(t0)\hat{\theta}_1(t_0) at a fixed point interior t0∈(0,τ)t_0 \in(0,\tau) is of interest. We state that limit distribution as an assumption, to allow for general statements.

Assumption 2.1. For the fixed interior point t0∈(0,τ)t_0 \in(0,\tau), the ols estimator θ^1(t0)\widehat{\theta}_1(t_0) satisfies

n1/2(θ^1(t0)−θ1(t0))⟶dG(t0),(9)n^{1/2}\left(\widehat{\theta}_1(t_0)-\theta_1(t_0)\right) \mathrel{\overset{d}{\longrightarrow}} G(t_0), \tag*{(9)}

where G(t0)G(t_0) is a mean zero random variable.

Note that this assumption is satisfied for the case that we treat above of independent data, with G(t0)G(t_0) a mean zero Gaussian random variable, under mild regularity conditions.

Now, recall the definition of the cumulative functions BkB_k, for k=0,1,2k=0,1,2, in (3). Then a standard result, cf. e.g. [3], is that the least squares estimator of B=(B0,B1,B2)TB=(B_0,B_1,B_2)^T is

B^(t)=∫0tIn(u)Y−(u) dN(u),(10)\widehat{B}(t)=\int_0^t I_n(u)Y^{-}(u)\,dN(u), \tag*{(10)}

where In(u)I_n(u) is a indicator for a certain design matrix in a matrix formulation of (1) and (2) has full rank, and Y−(u)Y^{-}(u) denotes a generalized inverse of the corresponding design matrix; see [2] and [3] for more details and [5] for a review.

Recently, in [5], this estimator was used as an input into an order-restricted scheme, to obtain a monotone estimator of the βk\beta_k functions themselves. Thus if we assume that β2\beta_2 is non-increasing, and B^2\widehat{B}_2 denote the second coordinate of B^\widehat{B}, then [5] defined an order-restricted estimator β~2\widetilde{\beta}_2, namely the isotonic regression of the local increment estimates of B^2\widehat{B}_2 onto the cone of non-increasing sequences. Furthermore, in [5] it was shown that a characterisation of the solution is

β~2(t)=ddtSB^2(t),(11)\widetilde{\beta}_2(t)=\frac{d}{dt}S\widehat{B}_2(t), \tag*{(11)}

where the derivative is understood to be the appropriate one sided derivative. Furthermore, algorithms for calculating the isotonic regression are well known, and e.g. the minimum lower set algorithm works well, cf. e.g. [6].

Now, since we have obtained estimators of the factors β2\beta_2 and θ1\theta_1 of the indirect effect I(t)I(t), we can define the plug-in estimator of the indirect effect by

I^n(t)=θ^1(t)β~2(t).(12)\widehat{I}_n(t)=\widehat{\theta}_1(t)\widetilde{\beta}_2(t). \tag*{(12)}

Limit distributions for the indirect effect estimator

We first recall and state the assumptions for the survival analysis setting that enable the limit distribution results of [5] for the order restricted estimator β~2(t)\widetilde{\beta}_2(t). In particular, these conditions ensure local regularity of the design and risk processes, nonsingularity of the limiting design matrix, and the smoothness and strict local monotonicity conditions on the coefficient function at the point under consideration. We emphasise that they are equivalent to the standard assumptions for deriving limit distributions of the Aalen estimator (10), cf. [3], and thus we do not need more than what is regularly assumed in a standard survival analysis setting.

Assumption 3.1. Assume that Assumptions 1,2 and 3 in [5] hold and that β2\beta_2 is differentiable with β2′<0\beta'_2 < 0 in a neighbourhood around the fixed interior point t0∈(0,τ)t_0 \in(0,\tau).

Under Assumption 3.1, Theorem 3 in [5] gives the limit distribution

n1/3c(t0)(β~2(t0)−β2(t0))→dS(−s2+w(s))′(0)(13)n^{1/3}c(t_0)\left(\widetilde{\beta}_2(t_0)-\beta_2(t_0)\right) \mathrel{\overset{d}{\to}} S\left(-s^2+w(s)\right)'(0) \tag*{(13)}

where ww is a standard two-sided Brownian motion, and where

c(t0)=2−1/3∣β2′(t0)∣1/3(σ22)−4/3,c(t_0) = 2^{-1/3}|\beta'_2(t_0)|^{1/3}(\sigma_2^2)^{-4/3},

and, σ22\sigma_2^2 is specified in [5].

Next we state the main result on the limit distribution for the plug estimator I^n(t)\widehat{I}_n(t), defined in (12), of the indirect effect I(t)I(t).

Theorem 3.2. Suppose that Assumptions 2.1 and 3.1 hold. Then

n1/3c(t0)(I^n(t0)−I(t0))→dθ1(t0)S(−s2+w(s))′(0).n^{1/3}c(t_0)\left(\widehat{I}_n(t_0)-I(t_0)\right) \mathrel{\overset{d}{\to}} \theta_1(t_0)S\left(-s^2+w(s)\right)'(0).

We note that the limit distribution is proportional to the Chernoff random variable S(−s2+w(s))′(0)S\left(-s^2+w(s)\right)'(0), common in order restricted inference.

Proof. We have that

n1/3(I^n(t0)−I(t0))=θ^1(t0)n1/3(β~2(t0)−β2(t0))n^{1/3}\left(\widehat{I}_n(t_0)-I(t_0)\right) = \widehat{\theta}_1(t_0)n^{1/3}\left(\widetilde{\beta}_2(t_0)-\beta_2(t_0)\right)
+β2(t0)n1/3(θ^1(t0)−θ1(t0)).\qquad{}+\beta_2(t_0)n^{1/3}\left(\widehat{\theta}_1(t_0)-\theta_1(t_0)\right).

We first see that the second term is

n1/3(θ^1(t0)−θ1(t0))=n−1/6n(θ^1(t0)−θ1(t0)).n^{1/3}\left(\hat{\theta}_1(t_0)-\theta_1(t_0)\right)=n^{-1/6}\sqrt{n}\left(\hat{\theta}_1(t_0)-\theta_1(t_0)\right).

and by Assumption 2.1, since n−1/6→0n^{-1/6} \to0 and the second factor is bounded in probability, the product converges to zero in probability, and therefore the second term is oP(1)o_P(1).

Next we treat the first term: Since θ^1(t0)−θ1(t0)→P0\hat{\theta}_1(t_0)-\theta_1(t_0)\xrightarrow{P}0, and since by (13)

n1/3(β~2(t0)−β2(t0))=OP(1),n^{1/3}\left(\widetilde{\beta}_2(t_0)-\beta_2(t_0)\right)=O_P(1),

we have that

(θ^1(t0)−θ1(t0))n1/3(β~2(t0)−β2(t0))=oP(1).\left(\hat{\theta}_1(t_0)-\theta_1(t_0)\right)n^{1/3}\left(\widetilde{\beta}_2(t_0)-\beta_2(t_0)\right)=o_P(1).

Therefore

θ^1(t0)n1/3(β~2(t0)−β2(t0))=θ1(t0)n1/3(β~2(t0)−β2(t0))+(θ^1(t0)−θ1(t0))n1/3(β~2(t0)−β2(t0))=θ1(t0)n1/3(β~2(t0)−β2(t0))+oP(1).\begin{aligned} \hat{\theta}_1(t_0)n^{1/3}\left(\widetilde{\beta}_2(t_0)-\beta_2(t_0)\right) &=\theta_1(t_0)n^{1/3}\left(\widetilde{\beta}_2(t_0)-\beta_2(t_0)\right) \\ &\quad+\left(\hat{\theta}_1(t_0)-\theta_1(t_0)\right)n^{1/3}\left(\widetilde{\beta}_2(t_0)-\beta_2(t_0)\right) \\ &=\theta_1(t_0)n^{1/3}\left(\widetilde{\beta}_2(t_0)-\beta_2(t_0)\right)+o_P(1). \end{aligned}

Thus

n1/3(I^(t0)−I(t0))=θ1(t0)n1/3(β~2(t0)−β2(t0))+oP(1).n^{1/3}\left(\hat{I}(t_0)-I(t_0)\right)=\theta_1(t_0)n^{1/3}\left(\widetilde{\beta}_2(t_0)-\beta_2(t_0)\right)+o_P(1).

Multiplying by the fixed constant c(t0)c(t_0) and applying the limit distribution result (13) gives

n1/3c(t0)(I^n(t0)−I(t0))→dθ1(t0)S(−s2+w(s))′(0).n^{1/3}c(t_0)\left(\hat{I}_n(t_0)-I(t_0)\right)\xrightarrow{d}\theta_1(t_0)S\left(-s^2+w(s)\right)'(0).

□\square

Conclusion and discussion

We have established the pointwise asymptotic distribution of a plug-in estimator for the indirect component in an DAG with the final stage modeled by an additive hazard model. When the coefficient θ1\theta_1 modelling the independent to dependent covariate effect is estimated at the regular n1/2n^{1/2} rate and the covariate coefficient in the final Aalen model is estimated by the newly developed order restricted Aalen procedure in [5], we show that the latter determines the asymptotic behaviour of the plug-in estimator of the indirect effect. In particular the resulting plug-in estimator has the same n1/3n^{1/3} pointwise scale as the order-restricted coefficient, with a Chernoff-type limit multiplied by the exposure to covariate coefficient.

The result provides a theoretical basis for inference on an indirect component constructed from an order-restricted additive hazard estimator.

The suggested model and estimator is novel, and allows for studying both early onset covariate effects and indirect covariate effects, and should be highly interesting for modelling complex diseases with time-to-event as response variable.

References

  1. [2]O. O. Aalen. A model for nonparametric regression analysis of counting processes. In Mathematical Statistics and Probability Theory, Lecture Notes in Statistics, volume 2, pages 1–25. Springer, New York, 1980.DOI
  2. [3]P. K. Andersen, Ø. Borgan, R. D. Gill, and N. Keiding. Statistical Models Based on Counting Processes. Springer-Verlag, New York, 1993.
  3. [5]D. Anevski and E. M. Merai. Order restricted estimation of the parameter functions in an additive hazard model. arXiv preprint arXiv:2606.23882, 2026.arxiv.org/abs/2606.23882
  4. [6]T. Robertson, F. T. Wright, and R. L. Dykstra. Order Restricted Statistical Inference. Wiley, Chichester, 1988.
  5. [8]FOSEN, J., Ø. BORGAN, H. WEEDON-FEKJÆR, AND O. O. AALEN (2006). Dynamic analysis of recurrent event data using the additive hazard model. Biometrical Journal 48, 381–398.pubmed.ncbi.nlm.nih.gov/16845903
  6. [9]ODD AALEN, ORNULF BORGAN AND HAKON GJESSING (2008). Survival and Event History Analysis. Springer Verlag.DOI

Paper details

vlnc:W_mxtzvs5w
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents