Choi 2018

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Received: 17 June 2017 Revised: 7 May 2018 Accepted: 10 May 2018

DOI: 10.1002/bimj.201700104

R E S E A R C H PA P E R

Smoothed quantile regression analysis of competing risks

Sangbum Choi1 Sangwook Kang2 Xuelin Huang3

1 Department of Statistics, Korea University,


Censored quantile regression models, which offer great flexibility in assessing covari-
Seoul, South Korea
2 Department of Applied Statistics, Yonsei ate effects on event times, have attracted considerable research interest. In this study,
University, Seoul, South Korea we consider flexible estimation and inference procedures for competing risks quan-
3 Department of Biostatistics, The University tile regression, which not only provides meaningful interpretations by using cumu-
of Texas, MD Anderson Cancer Center, lative incidence quantiles but also extends the conventional accelerated failure time
Houston, Texas
model by relaxing some of the stringent model assumptions, such as global linearity
Correspondence
Sangbum Choi, Department of Statistics, and unconditional independence. Current method for censored quantile regressions
Korea University, Seoul 02841, South Korea. often involves the minimization of the 𝐿1 -type convex function or solving the nons-
Email: choisang@korea.ac.kr
moothed estimating equations. This approach could lead to multiple roots in practical
Funding information
National Research Foundation of Korea,
settings, particularly with multiple covariates. Moreover, variance estimation involves
Grant/Award Numbers: 2017R1A2B4005818, an unknown error distribution and most methods rely on computationally intensive
2017R1C1B1004817; National Institutes resampling techniques such as bootstrapping. We consider the induced smoothing pro-
of Health, Grant/Award Numbers: 5P50
CA100632, U01 CA152958, U54 CA096300; cedure for censored quantile regressions to the competing risks setting. The proposed
Korea University, Grant/Award Number: procedure permits the fast and accurate computation of quantile regression parame-
K1607341; National Science Foundation,
ter estimates and standard variances by using conventional numerical methods such
Grant/Award Number: DMS 1612965
as the Newton–Raphson algorithm. Numerical studies show that the proposed esti-
mators perform well and the resulting inference is reliable in practical settings. The
method is finally applied to data from a soft tissue sarcoma study.

KEYWORDS
censored quantile regression, cumulative incidence function, induced smoothing, variance estimation,
weighted estimating equation

1 I N T RO D U C T I O N

In statistical and econometric research, linear quantile regression models (Koenker & Bassett, 1978) have been extensively
studied as a significant extension of traditional linear models. The regression parameters are often estimated by minimizing the
𝐿1 -type convex objective function or solving quantile-based estimating equations using linear programming or interior point
methods. Interest has also been growing in modeling regression quantiles on time-to-event data (Bang & Tsiatis, 2002; Pang,
Lu, & Wang, 2012; Peng & Huang, 2008; Portnoy, 2003; Powell, 1984, 1986; Ying, Jung, & Wei, 1995; Wang & Wang, 2009).
When an earlier or later stage of the follow-up is the primary focus, conventional unconditional mean-based methods such
as Cox's proportional hazards model or the accelerated failure time (AFT) model may be unsuitable. Alternatively, quantile
regressions can directly assess the lower or higher quantiles of interest by modeling regression quantiles to transform survival
times. In addition, quantile regressions can eliminate the difficulty of going from a conditional mean model to the entire survival
function, thereby fully assessing the covariate effects across different quantiles. These distinctive features make the quantile-
based approach attractive for modeling typically right-skewed failure time data.
In this study, we consider linear censored quantile regression models for competing risks data, a scenario that involves mul-
tiple but mutually disjoint censoring events. The standard approach for competing risks analysis often models the cumulative
Biometrical Journal. 2018;1–13. www.biometrical-journal.com © 2018 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim 1
2 CHOI ET AL.

incidence (or subdistribution) function (Fine & Gray, 1999; Gray, 1988; Lin, 1997), which, coupled with the cause-specific haz-
ard function (Prentice et al., 1978), has been widely used in practice to summarize the marginal risk of failure from a particular
cause over time. By extending the work of Bang and Tsiatis (2002) on median regression with censored medical cost data, Peng
and Fine (2007, 2009) proposed non-parametric and regression quantile methods for competing risks, respectively, by formu-
lating the model based on conditional quantiles adapted to the cumulative incidence function for a specific risk. They derived
an unbiased monotone estimating equation for quantile regression parameters and found the uniform consistency and weak con-
vergence of the resulting estimators. Sun, Wang, and Gilbert (2012) studied competing risks quantile regression when parts
of the failure cause are missing. Recently, Li and Peng (2011, 2015) studied quantile-based inferences to adjust for dependent
censoring from semi-competing risks data.
Although censored quantile regressions have attractive theoretical and practical properties, making associated inferences has
been difficult because of the lack of efficient and reliable computational algorithms. The implementation of inferential proce-
dures usually amounts to minimizing discontinuous functions that generally have multiple local minima. Moreover, the direct
estimation of the variance associated with the regression parameter generally involves estimating the unspecified conditional
error density function for the residual term. Nonparametric methods used to estimate the unknown density function such as
kernel smoothing may require moderate to large sample sizes and the selection of tuning parameters. To estimate the variance,
most existing methods for survival data (e.g., Peng & Huang, 2008; Portnoy, 2003; Wang & Wang, 2009) rely on computation-
ally intensive resampling techniques. Alternatively, Peng and Fine (2009) employed Huang (2002)'s approach that estimates the
slope matrix of the estimating equations based on their asymptotic linearity. Nevertheless, this approach still requires solving a
set of nonsmoothed estimating equations whose convergence may be problematic in a practical setting.
A more computationally efficient and practical approach to this situation is an induced kernel smoothing procedure (Brown
& Wang, 2005, 2006), which has proven useful for accommodating rank-based inferences for the semiparametric AFT model
(Chiou, Kang, & Yan, 2015a, b; Johnson & Strawderman, 2009) and the accelerated hazards model (Li, Zhang, & Tang, 2012).
The induced smoothing technique approximates a discontinuous estimating function by a continuously differentiable function
that is asymptotically equivalent to the original approach, facilitating rapid numerical solutions. In this study, we employ the
induced smoothing technique to enable a fast and reliable inference procedure for semiparametric censored regression quantiles
with competing risks by using the smoothed version of the weighted estimating equation of Peng and Fine (2009). The estimator
from the smoothed estimating function is shown to be consistent and have the same asymptotic distribution as that from the
nonsmoothed estimating equation. A useful consequence of developing smoothed quantile regressions is an easy-to-compute
sandwich variance estimator that avoids the need for resampling methods.
The rest of this article is organized as follows. In Section 2, we present the smoothed estimating functions for competing
risks quantile regressions and establish the asymptotic properties of the proposed estimators. Variance estimation procedures
are also provided. In Section 3, we examine the finite-sample properties by using simulation studies. In Section 4, we illustrate
the application of the proposed method with reference to a sarcoma cancer study. We offer concluding remarks in Section 5 and
present the proofs in the Appendix.

2 MODELS AND METHODS

2.1 Notation and assumptions


Consider a random sample of 𝑛 individuals from competing risks data subject to random right censoring, {𝑇̃𝑖 ≡ 𝑇𝑖 ∧ 𝐶𝑖 , 𝛿𝑖 ≡
𝐼(𝑇𝑖 ≤ 𝐶𝑖 ), 𝛿𝑖 𝜀𝑖 , 𝑍𝑖 ; 𝑖 = 1, … , 𝑛}, where 𝑇𝑖 and 𝐶𝑖 denote the failure and censoring times, respectively, 𝜀𝑖 ∈ {1, … , 𝐾} is the
cause of failure for which the 𝐾 causes are assumed to be known, and 𝑍𝑖 is a 𝑝 × 1 bounded time-independent covariate vector.
Here, 𝑎 ∧ 𝑏 = min(𝑎, 𝑏) and 𝐼(⋅) is the indicator function. It is assumed that 𝑇𝑖 and 𝐶𝑖 are independent variables and the first
entry of 𝑍𝑖 is set to be 1, corresponding to the intercept. Without loss of generality, we focus on the occurrence of the cause-1
failure (i.e., 𝜀𝑖 = 1) in the presence of the other types of events (i.e., 𝜀𝑖 ≠ 1). Then, the implied latent time variable 𝑇𝑖∗ for cause-1
failure is related to the usual survival time 𝑇𝑖 through

{
𝑇𝑖 , 𝜀𝑖 = 1,
𝑇𝑖∗ =
∞, 𝜀𝑖 ≠ 1.

With right-censoring, the failure from cause 1 cannot be fully identified. Hereafter, we write Δ𝑖 = 𝛿𝑖 𝐼(𝜀𝑖 = 1) to denote the
observed status of the failure from cause 1.
CHOI ET AL. 3

Our main objective is to estimate the 𝑝 × 1 vector of the quantile coefficient 𝜷 0 (𝜏) at a fixed quantile level 𝜏 ∈ [𝜏𝐿 , 𝜏𝑈 ] in the
linear quantile regression model,

log 𝑇𝑖∗ = 𝜷 0 (𝜏)′ 𝑍𝑖 + 𝑒𝑖 (𝜏), 𝑖 = 1, … , 𝑛, (1)

where 𝑒𝑖 (𝜏) ∈ {−∞, ∞} is the random error variable whose 𝜏-th conditional quantile given 𝑍𝑖 equals 0. Given the quantile level,
model (1) is equivalent to the AFT model for event time from the first failure cause. Note that 𝑇𝑖∗ has a distribution function
equal to the cause-1 cumulative incidence function:

𝐹1 (𝑡|𝑍𝑖 ) = P(𝑇𝑖∗ ≤ 𝑡|𝑍𝑖 ) = P(𝑇𝑖 ≤ 𝑡, 𝜀𝑖 = 1|𝑍𝑖 ),

which represents the probability of observing the cause-1 failure in the presence of the other types of events given the covariates.
The corresponding conditional quantile given 𝑍𝑖 = 𝑧 is 1 (𝜏|𝑧) = inf {𝑡; 𝐹1 (𝑡|𝑧) ≥ 𝜏} and the 𝜏-th linear quantile under model
(1) is equivalent to

1 (𝜏|𝑧) = exp{𝜷(𝜏)′ 𝑧}. (2)

To ensure identifiability, we let 0 < 𝜏𝐿 ≤ 𝜏 ≤ 𝜏𝑈 < inf 𝑧 𝐹1 (𝐿|𝑍 = 𝑧), where 𝐿 is the maximum follow-up time, satisfying the
conditions given in Section 2.3. Hereafter, we may use 𝜷 in place of 𝜷(𝜏) for ease of presentation if the context is clear.

2.2 Competing risks quantile regressions


If complete competing risks data without censoring are available for each patient, the quantile regression parameters can be

consistently obtained by using the least absolute deviations estimator for 𝜷(𝜏) by minimizing 𝑛−1 𝑛𝑖=1 |𝐼(𝑋𝑖 − 𝜷 ′ 𝑍𝑖 < 0, 𝜀𝑖 =
1) − 𝜏|, where 𝑋𝑖 = log 𝑇̃𝑖 . A minimizer is also a solution of the estimating equation
𝑛

−1
𝑛 𝑍𝑖 {𝐼(𝑋𝑖 − 𝜷 ′ 𝑍𝑖 < 0, 𝜀𝑖 = 1) − 𝜏} ≈ 0.
𝑖=1

The approximation is used here because the estimating equation is a discontinuous function of 𝜷. In the presence of random
right-censoring, Peng and Fine (2009) proposed using the following weighted estimating equation:

𝑛
{ }
∑ Δ𝑖
−1
𝑆𝑛 (𝜷, 𝜏) = 𝑛 𝑍𝑖 𝐼(𝑋𝑖 − 𝜷 ′ 𝑍𝑖 < 0) − 𝜏 ≈ 0, (3)
𝑖=1
̂ 𝑇̃𝑖 )
𝐺(

̂ is the Kaplan–Meier estimator for the censoring survivor function 𝐺(𝑡) = P(𝐶 ≥ 𝑡), based on the data {(𝑇̃𝑖 , 1 − 𝛿𝑖 ); 𝑖 =
where 𝐺(𝑡)
̂ = ∏ {1 − 𝑑𝑁 𝑐 (𝑢)∕𝑌 (𝑢)} with 𝑁 𝑐 (𝑢) = ∑𝑛 𝑁 𝑐 (𝑢) = ∑𝑛 𝐼(𝑇̃𝑖 ≤ 𝑢, 𝛿𝑖 = 0) and 𝑌 (𝑢) = ∑𝑛 𝑌𝑖 (𝑢) =
1, … , 𝑛}, that is 𝐺(𝑡) 𝑢<𝑡 𝑖=1 𝑖 𝑖=1 𝑖=1
∑𝑛 ̃𝑖 ≥ 𝑢). Let the associated censoring martingale process 𝑀 𝑐 (𝑡) = 𝑁 𝑐 (𝑡) − ∫ 𝑡 𝜆𝑐 (𝑢)𝑌𝑖 (𝑢)𝑑𝑡, where 𝜆𝑐 (𝑢) is the hazard
𝑖=1 𝐼( 𝑇 𝑖 𝑖 0
function of the censoring distribution. The unbiasedness of this estimating equation follows easily using a conditioning argument
(Bang & Tsiatis, 2002; Peng & Fine, 2009).
Because of the discontinuity of the estimating function (3), its exact solution may not exist. Instead, one can minimize the
Euclidean norm of the estimating function, but this is also discontinuous and has no derivatives. Peng and Fine (2009) showed
that the implementation of the estimation procedure amounts to minimizing the 𝐿1 -type convex function:

∑𝑛 | ∑𝑛
Δ𝑗 𝑍𝑗 || || ∑𝑛 |
Δ𝑖 | ′ |
| |
𝐿𝑛 (𝜷, 𝜏) = | ′ | | ′
𝑍𝑗 || ,
̂ ̃ |𝑋𝑖 − 𝛽 𝑍𝑖 | + |𝑀0 + 𝛽 ̂ ̃ | + |𝑀0 − 2𝜏𝛽
𝑖=1 𝐺(𝑇𝑖 ) | 𝑗=1 𝐺(𝑇𝑗 ) || | 𝑗=1 |
| | |

where 𝑀0 > 0 is an extremely large constant. They proposed using 𝜷̂ 𝑛 = argmin𝜷∈𝔹 𝐿𝑛 (𝜷, 𝜏). However, the solution to this
minimization problem may not be unique. The convexity of 𝐿𝑛 (𝜷, 𝜏) implies that the set of minimizers on 𝔹 is convex. The lack
of smoothness still makes the minimization of 𝐿𝑛 (𝜷, 𝜏) computationally challenging, particularly with multiple covariates.
Under regularity conditions, the results of Peng and Fine (2009) imply that there exists a sequence of solutions that is strongly
consistent for 𝜷 0 and that 𝑛1∕2 (𝜷̂ 𝑛 − 𝜷 0 ) converges in distribution to an 𝑁(𝟎, Ψ) random vector, where Ψ = 𝐴−1 Γ(𝐴−1 )′ , Γ =
lim𝑛→∞ var{𝑛1∕2 𝑆𝑛 (𝜷 0 , 𝜏)}, and 𝐴 = (𝜕∕𝜕𝜷)𝑆0 (𝜷 0 , 𝜏) for 𝑆0 (𝜷, 𝜏) = lim𝑛→∞ 𝑆𝑛 (𝜷, 𝜏). In addition to the numerical challenges
4 CHOI ET AL.

that arise in computing 𝜷̂ 𝑛 , the variance estimation is complicated by the presence of 𝐴 that involves the unknown distribution
of 𝑒𝑖 (𝜏), and the fact that 𝑆𝑛 (𝜷, 𝜏) is not differentiable in 𝜷. Note that Γ can be approximated by

𝑛
{ }2
∑ Δ𝑖 𝑅𝑖 (𝜷 0 ) 𝐿
𝑑𝑁 𝑐 (𝑢) ̂
Γ̂ 𝑛 (𝜷 0 ) = 𝑛−1 𝑍𝑖⊗2 −𝜏 − 𝑛−1 {𝐵(𝜷 0 , 𝑢)}⊗2 , (4)
𝐺(̂ 𝑇̃𝑖 ) ∫0 𝑌 2 (𝑢)
𝑖=1

where 𝑎⊗2 = 𝑎𝑎′ , 𝑅𝑖 (𝜷 0 ) = 𝐼(𝑋𝑖 − 𝜷 ′0 𝑍𝑖 < 0), and 𝐵(𝜷̂ 0 , 𝑢) = ∑𝑛 {Δ𝑖 𝑍𝑖 𝑌𝑖 (𝑢)𝑅𝑖 (𝜷 0 )}∕𝐺( ̂ 𝑇̃𝑖 ). The Γ̂ 𝑛 can be directly obtained
𝑖=1
from (4) by replacing 𝜷 0 with 𝜷̂ 𝑛 .
Since the formula in (4) has a complicated expression, it may be more convenient and reliable to bootstrap from the data.
In our implementation, we apply an efficient resampling method by Zeng and Lin (2008) to approximate Γ, in which a random
perturbation whose mean and variance both equal one is assigned to estimating function (3) for multiple times and compute
their variance to have Γ̂ 𝑛 . This approach does not require solving estimating equations and is thus much faster than the existing
resampling procedures. Once Γ̂ 𝑛 is obtained, Peng and Fine (2009) estimated the variance of 𝜷̂ 𝑛 by generalizing the technique
of Huang (2002). Specifically, they first decomposed the empirical estimate Γ̂ 𝑛 as 𝜸𝜸 ′ , where 𝜸 = (𝛾1 , … , 𝛾𝑝 )′ , and computed
the solutions 𝒃̃ = (𝑏̃ 1 , … , 𝑏̃ 𝑝 ) by solving Huang (2002)'s local linearization 𝑆𝑛 (𝛽, 𝜏) − 𝛾𝑘 = 0 for 𝑘 = 1, … , 𝑝. The variance
estimator of 𝑛1∕2 (𝛽̂𝑛 − 𝛽0 ) can then be approximated by 𝑛(𝒃̃ − 𝜷̂ 𝑛 )(𝒃̃ − 𝜷̂ 𝑛 )′ . As illustrated in our simulation studies, however,
this method would unsatisfactorily overestimate empirical variances, resulting in broader confidence intervals.

2.3 Induced smoothing for competing risks data


Brown and Wang (2005) proposed an induced smoothing method for approximating discontinuous but monotone estimating
functions using continuously differentiable functions. Assuming independent failure time observations, Brown and Wang (2006)
applied this smoothing method to the problem of estimating the regression parameter in the AFT model. Based on a similar idea,
we seek the smoothed version of the estimating function 𝑆𝑛 (𝜷, 𝜏).
Let 𝑉 be an 𝑁(𝟎, 𝐼𝑝 ) random vector independent of the data, where 𝐼𝑝 denotes the 𝑝 × 𝑝 identity matrix. Let Σ be a 𝑝 × 𝑝
symmetric, positive definite matrix such that ‖Σ‖ = 𝑂(𝑛−1 ). According to the asymptotic normality of 𝜷̂ 𝑛 , we can write 𝜷̂ 𝑛 =
𝜷 0 + Σ1∕2 𝑉 with Σ = 𝑛−1 Ψ, which implies that 𝜷̂ 𝑛 can also be regarded as a random perturbation of 𝜷 0 . By adding the random
perturbation Σ1∕2 𝑉 to the argument 𝜷 of the score function 𝑆𝑛 (𝜷, 𝜏) and taking the expectation with respect to 𝑉 , we obtain the
smoothed version of estimating equation (3) as

𝑆̃𝑛 (𝜷, 𝜏) ≡ 𝐸𝑉 {𝑆𝑛 (𝜷 + Σ1∕2 𝑉 , 𝜏)} =

𝑛
⎧ ⎛ ⎞ ⎫
∑ ⎪ Δ𝑖 ⎜ 𝑋𝑖 − 𝜷 ′ 𝑍𝑖 ⎟ ⎪
−1
=𝑛 𝑍𝑖 ⎨ Φ ⎜− √ ⎟ − 𝜏 ⎬ = 0, (5)
̂ 𝑇̃𝑖 ) ⎜
⎪ 𝐺(
𝑖=1 𝑍𝑖′ Σ𝑍𝑖 ⎟ ⎪
⎩ ⎝ ⎠ ⎭

where Φ(⋅) is the standard normal cumulative distribution function. Let 𝜙(⋅) denote the standard normal density function. The
smoothed Equation (5) is a monotone function with respect to each element of 𝜷.
Moreover, using standard results from normal random variables and integration by parts, we have

𝑛
[ ⎧ ⎛ ⎞⎫
∑ ⎪ Δ𝑖 ⎜ 𝑋𝑖 − 𝜷 ′ 𝑍𝑖 ⎟ ⎪
𝐿̃ 𝑛 (𝜷, 𝜏) = 𝑛−1 ′
(𝑋𝑖 − 𝜷 𝑍𝑖 ) ⎨𝜏 − Φ −√ ⎟⎬ +
⎪ ̂ 𝑇̃𝑖 ) ⎜⎜
𝐺(
𝑖=1 𝑍𝑖′ Σ𝑍𝑖 ⎟ ⎪
⎩ ⎝ ⎠⎭

⎛ ⎞ ]
Δ𝑖 ⎜ 𝑋𝑖 − 𝜷 ′ 𝑍𝑖 ⎟ √ ′
+ 𝜙 −√ ⎟ 𝑍𝑖 Σ𝑍𝑖 .
̂ 𝑇̃𝑖 ) ⎜⎜
(6)
𝐺( 𝑍𝑖′ Σ𝑍𝑖 ⎟
⎝ ⎠

A straightforward calculation shows that 𝜕 𝐿̃ 𝑛 (𝜷, 𝜏)∕(𝜕𝜷) = 𝑆̃𝑛 (𝜷, 𝜏). If Σ is given, the smoothed quantile regression parameter
may be obtained from 𝜷̃ 𝑛 = argmin𝜷∈𝔹 𝐿̃ 𝑛 (𝜷, 𝜏). The smoothed objective function, 𝐿̃ 𝑛 (𝜷, 𝜏), is convex and continuously differ-
entiable and thus standard numerical routines, such as the Newton–Raphson algorithm, can be used to effectively compute 𝜷̃ 𝑛 .
Alternatively, the corresponding coefficient estimator 𝜷̃ 𝑛 can be found as the multivariate root of 𝑆̃𝑛 (𝜷, 𝜏) = 0.
CHOI ET AL. 5

In the case that Σ is unknown and has to be estimated, we propose using a positive definite 𝑝 × 𝑝 matrix Σ̃ (0) = 𝑂(𝑛−1 ) as
an initial value and updating through iterations to estimate Σ as described below. According to Brown and Wang (2005, 2006),
the estimation of the asymptotic variance is mainly based on the sandwich formula of the covariance matrix of the estimated
parameters. The partial derivatives of the smoothed estimating function (5) can be explicitly expressed as

𝑛
⎛ ⎞
𝜕 𝑆̃𝑛 (𝜷, 𝜏) ∑ Δ𝑖 ⎜ 𝑋𝑖 − 𝜷 ′ 𝑍𝑖 ⎟ 𝑍𝑖 𝑍𝑖′
𝐴̃ 𝑛 (𝜷) = =𝑛−1
𝜙 ⎜− √ ⎟√ . (7)
𝜕𝜷 ̂ ̃
𝑖=1 𝐺(𝑇𝑖 ) ⎜ 𝑍𝑖′ Σ𝑍𝑖 ⎟ 𝑍𝑖′ Σ𝑍𝑖
⎝ ⎠

As suggested by Pang et al. (2012) for censored regression quantiles, an iterative procedure can be constructed to simultaneously
estimate the regression parameter 𝜷 and its covariance matrix Σ. The estimating procedure consists of the following steps:
(0)
Step 1 Let 𝜷̃ 𝑛 = 𝜷̂ 𝑛 , from the Peng and Fine (2009) estimator from solving (3), and Σ̃ (0) −1
𝑛 = 𝑛 𝐼𝑝 .
(𝑘−1) (𝑘)
Step 2 Given 𝜷̃ 𝑛 and Σ̃ (𝑘−1)
𝑛 from the (𝑘 − 1)-th step, update 𝜷̃ 𝑛 and Σ̃ (𝑘)
𝑛 as

(𝑘) (𝑘−1) { ( (𝑘−1) )}−1 ( (𝑘−1) )


𝜷̃ 𝑛 = 𝜷̃ 𝑛 − 𝐴̃ 𝑛 𝜷̃ 𝑛 𝑆̃𝑛 𝜷̃ 𝑛 , 𝜏 ,
{ (
−1 ̃ ̃ (𝑘−1)
)}−1 ( (𝑘−1) ){ ( (𝑘−1) )}−1
Σ̃ (𝑘)
𝑛 = 𝑛 𝐴𝑛 𝜷 𝑛 Γ̂ 𝑛 𝜷̃ 𝑛 𝐴̃ 𝑛 𝜷̃ 𝑛 .

Step 3 Repeat Step 2 until convergence. Denote the coefficient estimate and covariance estimate at convergence as 𝜷̃ 𝑛 and Σ̃ 𝑛 ,
respectively.

To establish the asymptotic properties of the proposed method, we state the regularity conditions:

(A1) There exists 𝐿 > 0 such that P(𝐶 = 𝐿) > 𝜈 and P(𝐶 > 𝐿) = 0 for a constant 𝜈 ∈ (0, 1).
(A2) 𝜷 0 (𝜏) is Lipschitz continuous for 𝜏 ∈ [𝜏𝐿 , 𝜏𝑈 ] and the parameter space 𝔹 containing 𝜷 0 (𝜏) is a compact subset of ℝ𝑝 . 𝑍
is uniformly bounded, i.e. sup𝑖 ‖𝑍𝑖 ‖ < ∞.
(A3) Let 𝑓1 (⋅) ≡ 𝑓1 (⋅|𝑍1 ) denote the marginal density associated with the model error term 𝑒1 . Then, 𝑓1 (⋅) and 𝑓1′ (⋅) are bounded
{ 𝑓 ′ (𝑡) }2
functions on ℝ with ∫ℝ 𝑓1 (𝑡) 𝑓1 (𝑡)𝑑𝑡 < ∞.
1

(A4) The matrix 𝐴 = lim𝑛→∞ 𝑛−1 𝑛𝑖=1 𝑍𝑖 𝑍𝑖′ 𝑓𝑖 (0) exists and is nonsingular.

In the Appendix, we show that under the regularity conditions (A1)–(A4),

(i) lim sup ‖𝜷̃ 𝑛 − 𝜷 0 ‖ →𝑝 0,


𝑛→∞ 𝜏∈[𝜏 ,𝜏 ]
𝐿 𝑈
(𝑘)
(ii) 𝑛1∕2 (𝜷̃ 𝑛 − 𝜷 0 ) →𝑑 𝑁(𝟎, Ψ) for any 𝑘 ≥ 1.

This fact implies that the induced smoothing estimator 𝜷̃ 𝑛 has the same asymptotic distribution as the original estima-
tor 𝜷̂ 𝑛 from the nonsmoothed estimating equation. At convergence, the variance estimator for Ψ can be approximated by
Ψ̃ 𝑛 = {𝐴̃ 𝑛 (𝜷̃ 𝑛 )}−1 Γ̂ 𝑛 (𝜷̃ 𝑛 ){𝐴̃ 𝑛 (𝜷̃ 𝑛 )}−1 , which is equivalent to 𝑛Σ̃ 𝑛 . Note that the variance estimator is obtained as a byproduct
while performing the proposed iterative procedure. In addition, the proposed variance estimator is shown to be consistent. This
suggests that Ψ̃ 𝑛 may be used as a variance estimation procedure for the regression parameter. Our simulation study in the next
section confirms that the proposed algorithm converges quickly, usually within five iterations, and that the variance estimates
are fairly accurate.

3 SIMULAT ION STUDIES

In this section, we present the results of the simulation studies carried out to assess the performance of our smoothed estimating
equations with finite samples for competing risks regression quantiles. We also compare the proposed induced smoothing method
with Peng and Fine (2009)'s nonsmoothed estimating equation method. The simulation study involves two covariates with
𝑍𝑖 = (1, 𝑍1𝑖 , 𝑍2𝑖 )′ , where 𝑍1𝑖 is a uniform (−1, 1) variable and 𝑍2𝑖 is a Bernoulli (0.5) variable. Two competing failure causes,
6 CHOI ET AL.

TABLE 1 Simulation results with 𝑛 = 300 from the smoothed and nonsmoothed estimating equations for competing risks regression quantilesa
Smoothed equation Nonsmoothed equation
Dist 𝝉 Par True Bias ASE ESE Cov Bias ASE ESE Cov RE
Normal 0.2 𝛽1 −1.566 −0.002 0.130 0.133 0.928 0.004 0.165 0.139 0.942 1.099
𝛽2 1.000 −0.001 0.156 0.158 0.931 0.001 0.187 0.167 0.936 1.115
𝛽3 0.891 −0.006 0.182 0.183 0.940 −0.005 0.225 0.193 0.942 1.112
0.4 𝛽1 −0.820 0.012 0.144 0.148 0.935 0.008 0.173 0.153 0.938 1.062
𝛽2 1.000 0.010 0.173 0.178 0.936 0.009 0.207 0.186 0.931 1.090
𝛽3 0.820 −0.008 0.198 0.201 0.937 −0.003 0.232 0.209 0.937 1.076
Logistic 0.2 𝛽1 −1.916 −0.012 0.216 0.220 0.935 0.001 0.263 0.228 0.934 1.073
𝛽2 1.000 −0.007 0.257 0.261 0.933 −0.008 0.308 0.273 0.938 1.089
𝛽3 0.817 −0.011 0.302 0.305 0.940 −0.007 0.356 0.315 0.941 1.065
0.4 𝛽1 −0.712 0.018 0.229 0.236 0.940 0.014 0.266 0.242 0.940 1.052
𝛽2 1.000 0.007 0.263 0.270 0.936 0.004 0.310 0.283 0.924 1.098
𝛽3 0.712 −0.019 0.307 0.312 0.936 −0.014 0.343 0.320 0.921 1.048
Cauchy 0.2 𝛽1 −1.797 −0.041 0.235 0.248 0.928 −0.025 0.289 0.244 0.931 0.934
𝛽2 1.000 −0.010 0.291 0.306 0.930 −0.013 0.351 0.310 0.946 1.022
𝛽3 0.797 −0.017 0.345 0.361 0.942 −0.010 0.407 0.355 0.946 0.964
0.4 𝛽1 −0.772 0.023 0.195 0.203 0.952 0.018 0.229 0.202 0.945 0.979
𝛽2 1.000 0.005 0.220 0.228 0.939 0.004 0.254 0.230 0.936 1.016
𝛽3 0.772 −0.023 0.257 0.262 0.942 −0.016 0.290 0.262 0.945 0.995
𝑡(3) 0.2 𝛽1 −1.633 −0.014 0.155 0.159 0.932 −0.003 0.192 0.164 0.943 1.056
𝛽2 1.000 −0.005 0.187 0.192 0.931 −0.005 0.226 0.201 0.933 1.087
𝛽3 0.869 −0.009 0.220 0.222 0.940 −0.006 0.263 0.228 0.940 1.051
0.4 𝛽1 −0.804 0.016 0.159 0.157 0.959 0.014 0.184 0.162 0.933 1.050
𝛽2 1.000 0.015 0.184 0.177 0.936 0.016 0.218 0.179 0.944 1.027
𝛽3 0.804 −0.007 0.214 0.211 0.951 −0.001 0.242 0.216 0.936 1.051
ASE, average of standard error estimates; ESE, empirical standard error; Cov, coverage probability of the 95% Wald-type confidence intervals; RE, relative efficiency of
the mean squared errors (MSEs) for the nonsmoothed estimator over the induced smoothing estimator. 𝑡(3) represents 𝑡-distribution with 3 degrees of freedom.
a The marginal residual error for the failure from cause 1 follows the (i) normal, (ii) logistic, (iii) Cauchy, and (iv) 𝑡
(3) distributions, respectively.

𝜀𝑖 ∈ {1, 2}, are assumed and generated according to P(𝜀𝑖 = 1|𝑍𝑖 = 𝑧𝑖 ) = 𝑝1 𝐼(𝑧2𝑖 = 1) + 𝑝0 𝐼(𝑧2𝑖 = 0) with (𝑝0 , 𝑝1 ) = (0.7, 0.8).
We consider two quantile levels, 𝜏 = 0.2 and 𝜏 = 0.4, for which the respective survival times follow conditional distributions:

P(𝑇𝑖 ≤ 𝑡|𝜀𝑖 = 1, 𝑍𝑖 = 𝑧𝑖 ) = 𝐻0 (log 𝑡 − 𝜃0′ 𝑧𝑖 ), P(𝑇𝑖 ≤ 𝑡|𝜀𝑖 = 2, 𝑍𝑖 = 𝑧𝑖 ) = 𝐻0 (log 𝑡 − 𝜗′0 𝑧𝑖 ).

For the quantile residual subdistribution 𝐻0 (⋅), we consider the following four distributions: (i) standard normal 𝑁(0, 1),
(ii) Logistic(0, 1), (iii) Cauchy(0, 1), and (iv) 𝑡(3) , that is 𝑡-distribution with 3 degrees of freedom, respectively. Under this setting,
the 𝜏-th conditional quantile of 𝑇𝑖 from cause 1 is

1𝑖 (𝜏|𝑍𝑖 = 𝑧𝑖 ) = inf{𝑡 > 0 ∶ P(𝑇𝑖 ≤ 𝑡, 𝜀𝑖 = 1|𝑍𝑖 = 𝑧𝑖 ) ≥ 𝜏} =

= exp{𝛽1 (𝜏) + 𝛽2 (𝜏)𝑧1𝑖 + 𝛽3 (𝜏)𝑧2𝑖 },

where 𝛽1 (𝜏) = 𝜃1 + 𝐻0−1 (𝜏∕𝑝0 ), 𝛽2 (𝜏) = 𝜃2 , and 𝛽3 (𝜏) = 𝜃3 + 𝐻0−1 (𝜏∕𝑝1 ) − 𝐻0−1 (𝜏∕𝑝0 ). We let 𝜽0 = (𝜃1 , 𝜃2 , 𝜃3 )′ = (−1, 1, 1)′
and 𝝑0 = (𝜗1 , 𝜗2 , 𝜗3 )′ = (−1, 1, −1)′ , which leads to the occurrence of events 1 and 2 in a nearly 3:1 ratio. The censoring times
𝐶𝑖 are generated from Uniform(0, 𝐿), where 𝐿 is chosen to yield approximately 20% censoring rates in all cases. On average,
this simulation setup finds that 55, 25, and 20% of subjects experience event 1, event 2, or are censored.
Tables 1 and 2 summarize the simulation results from 1000 replications with sample sizes 𝑛 = 300 and 500, respectively.
(𝑘) (𝑘−1)
All computing work here was done in R (version 3.4.2). The iteration procedure is stopped when either ‖𝜷̃ 𝑛 − 𝜷̃ 𝑛 ‖ < 10−5
or the maximum number of iteration, specified as 100, is first achieved. If starting values are properly selected, our algorithm
CHOI ET AL. 7

TABLE 2 Simulation results with 𝑛 = 500 from the smoothed and nonsmoothed estimating equations for competing risks regression quantilesa
Smoothed equation Nonsmoothed equation
Dist 𝝉 Par True Bias ASE ESE Cov Bias ASE ESE Cov RE
Normal 0.2 𝛽1 −1.566 −0.004 0.101 0.105 0.928 0.001 0.123 0.110 0.932 1.089
𝛽2 1.000 0.005 0.123 0.119 0.943 0.006 0.147 0.126 0.940 1.108
𝛽3 0.891 0.003 0.143 0.148 0.926 0.004 0.172 0.154 0.929 1.088
0.4 𝛽1 −0.820 0.002 0.112 0.111 0.941 0.001 0.131 0.115 0.934 1.064
𝛽2 1.000 0.010 0.139 0.145 0.934 0.010 0.163 0.151 0.936 1.081
𝛽3 0.820 0.006 0.158 0.158 0.943 0.007 0.181 0.164 0.931 1.075
Logistic 0.2 𝛽1 −1.916 −0.008 0.167 0.173 0.932 0.001 0.193 0.179 0.927 1.069
𝛽2 1.000 0.009 0.200 0.194 0.942 0.009 0.233 0.204 0.933 1.110
𝛽3 0.818 0.001 0.234 0.242 0.934 0.001 0.266 0.250 0.928 1.067
0.4 𝛽1 −0.712 0.003 0.175 0.174 0.944 −0.002 0.199 0.177 0.942 1.045
𝛽2 1.000 0.008 0.202 0.211 0.940 0.009 0.231 0.218 0.923 1.068
𝛽3 0.712 −0.002 0.236 0.232 0.947 0.004 0.262 0.239 0.942 1.053
Cauchy 0.2 𝛽1 −1.797 −0.042 0.181 0.188 0.934 −0.019 0.214 0.190 0.943 0.985
𝛽2 1.000 0.011 0.226 0.221 0.950 0.009 0.267 0.231 0.947 1.091
𝛽3 0.797 0.001 0.266 0.278 0.942 0.005 0.304 0.279 0.938 1.008
0.4 𝛽1 −0.772 0.012 0.146 0.139 0.956 0.007 0.165 0.140 0.953 1.011
𝛽2 1.000 0.010 0.166 0.165 0.956 0.010 0.187 0.170 0.939 1.063
𝛽3 0.772 −0.009 0.194 0.191 0.943 −0.004 0.214 0.193 0.929 1.011
𝑡(3) 0.2 𝛽1 −1.633 −0.010 0.120 0.124 0.938 −0.003 0.141 0.128 0.934 1.070
𝛽2 1.000 0.007 0.147 0.140 0.946 0.007 0.171 0.148 0.937 1.112
𝛽3 0.869 −0.002 0.171 0.175 0.934 0.001 0.196 0.181 0.930 1.075
0.4 𝛽1 −0.804 0.006 0.121 0.114 0.955 0.002 0.136 0.115 0.952 1.026
𝛽2 1.000 0.012 0.140 0.139 0.942 0.010 0.164 0.144 0.939 1.067
𝛽3 0.804 −0.001 0.164 0.160 0.938 0.003 0.181 0.166 0.932 1.071
See the note in Table 1.
a
The marginal residual error for the failure from cause 1 follows the (i) normal, (ii) logistic, (iii) Cauchy, and (iv) 𝑡(3) distributions, respectively.

converges after 4–8 iterations. For example, mean (sd) time to convergence and mean iteration per dataset are 0.12 (0.05)
seconds and 5.12 times, respectively, when 𝑛 = 300, and 0.24 (0.07) seconds and 4.56 times when 𝑛 = 500 (case (i), 𝜏 = 0.2).
For each simulation run, we report the mean bias (Bias), averages of the standard error estimates (ASE), empirical standard
deviations (ESE), and empirical coverage probabilities (Cov) of the cause-1 quantile regression coefficients 𝛽𝑘 (𝜏), 𝑘 = 1, 2, 3.
We also compute the MSEs for each method and present the relative efficiency, defined as RE = MSE1 ∕MSE2 , where MSE1 and
MSE2 correspond to the nonsmoothed estimator and proposed induced smoothing estimator, respectively. We used the 𝚍𝚏𝚜𝚊𝚗𝚎
function in 𝚕𝚒𝚋𝚛𝚊𝚛𝚢(𝙱𝙱) to solve the nonsmoothed estimating function (3). To approximate Γ from those provided by Peng and
Fine (2009)'s approach, we used the resampling method by Zeng and Lin (2008) with 300 bootstrap samples. Since this approach
does not require solving estimating functions for bootstrapped samples, it is fairly fast and compatible to the proposed method
in computing time. In our simulation experiment, for example, the induced smoothing method took about 190 seconds and the
nonsmoothed method requires 270 seconds to complete 1000 simulation runs (case (i), 𝜏 = 0.2, and 𝑛 = 300) on our machine
(Intel i5-6200U CPU, 2.30 GHz).
In all scenarios, both the induced smoothing and the nonsmoothed estimators are virtually unbiased, the standard error esti-
mators appear to reasonably reflect the true variations, and the confidence intervals have reasonable coverage probabilities. In
theory, the nonsmoothed estimator is asymptotically equivalent to the proposed induced smoothing estimator. With finite sam-
ple sizes, however, the proposed estimators are slightly more efficient than the nonsmoothed estimators in nearly all scenarios,
even though there are some fluctuations in relative efficiency between the two methods. It is also observed that ASEs from the
proposed smoothing method are fairly close to ESEs, while ASEs from the nonsmoothed method are 10–30% higher than ESEs,
resulting in noticeably inflated standard errors. Nonetheless, Peng and Fine's approach often leads to lower coverage probabili-
ties than the nominal level especially when 𝑛 = 300. This is partly because their variance estimation method has relatively large
8 CHOI ET AL.

Sarcomas Other causes


(A) (B)

0.6

0.6
Chemotherapy
No chemotherapy

0.5

0.5
Cumulative incidence

Cumulative incidence
0.4

0.4
0.3

0.3
0.2

0.2
0.1

0.1
Chemotherapy
No chemotherapy
0.0

0.0
0 5 10 15 0 5 10 15

Time (Years) Time (Years)

FIGURE 1 Nonparametric cumulative incidence (subdistribution) functions for (A) sarcoma-related death and (B) death from other competing
causes of patients who received local therapy only compared with local therapy plus adjuvant chemotherapy

̂
variation. For example, se(ASE) for 𝛽̂2 is about 0.03 for the smoothed method, while it is 0.08 for the nonsmoothed method,
which might cause a higher chance to miss true parameters in computing coverage probabilities for the latter.
Our finding from the simulation study also reveals that the induced smoothing method is relatively more sensitive to the initial
values, especially when 𝜏 is extreme (too small or too large). In that case, the nonsmoothed estimator may be used for the initial
input for the induced smoothing method, as we suggested in Section 2.3.

4 ANALYSIS O F SO F T T I SSU E SARCOM A DATA

We illustrate the application of the proposed method by using data from a study of the effects of chemotherapy on the survival
of soft tissue sarcoma patients. For these patients, the primary treatment is surgical resection. Chemotherapy is often used as
an adjuvant therapy for treating sarcomas. However, the utility of adjuvant chemotherapy remains uncertain in the treatment of
sarcoma patients and the results have been conflicting in several studies. To assess its effect, Cormier et al. (2004) retrospectively
identified and analyzed a cohort of 679 patients from two major cancer centers in the United States. In their initial treatments,
all patients received definitive surgical resection. Among the 679 patients, 228 received adjuvant radiation alone, 109 received
adjuvant chemotherapy alone, 207 received both, and 135 received none of these treatments. Of the 316 patients treated with
adjuvant chemotherapy, 148 (46.8%) died from sarcomas, and of the 363 patients not treated with adjuvant chemotherapy, 140
(38.6%) died from sarcomas. Medical records were examined retrospectively to verify the known prognostic factors, including
tumor size and location treatment sequences, pathologic margin status, and survival outcomes.
The primary objective of this analysis was to evaluate the impact of chemotherapy while accounting for the known prognostic
variables. However, the analysis of this dataset is complicated by the existence of competing events and a considerable proportion
of survivors: 288 (42.4%) patients died from the sarcoma, 65 (9.6%) died from other competing causes, and 326 (48.0%) patients
survived throughout the follow-up period of the study. Furthermore, the nonparametric subdistributions, as shown in Figure 1,
for the sarcoma-related death of patients who received adjuvant chemotherapy compared with those who did not crossed at about
two years after surgery, indicating that the effect of adjuvant chemotherapy changes over time. In this situation, analysis using
the proportional subdistribution hazards model (Fine & Gray, 1999) could be misleading, as it cannot facilitate the crossing of
the subdistribution functions. The mixture approach (Choi & Huang, 2015) can fit such a data pattern but has some complexity
in interpreting the results. Alternatively, we applied the quantile regression model that can evaluate the conditional treatment
effects orderly over a set of different quantiles.
We begin with the univariate quantile analysis. Figure 2 plots the results from the smoothed quantile regression for the sarcoma
endpoint, estimating the effects of chemotherapy (yes or no), radiation (yes or no), tumor size (≥10 or not), and age (≥60 or not),
respectively. The estimated regression coefficients are plotted in bold solid lines and the 95% pointwise confidence intervals are
CHOI ET AL. 9

Chemotherapy Radiation

1.5

1.5
1.0
Quantile coefficients

Quantile coefficients

1.0
0.5

0.5
0.0

0.0
−1.0 −0.5

−0.5
0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4

Tau Tau

Tumor size Age


0.5

0.5
Quantile coefficients

Quantile coefficients
0.0

0.0
−0.5

−0.5
−1.0

−1.0
−1.5

−1.5

0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4

Tau Tau

FIGURE 2 Estimated quantile regression coefficients for death from sarcoma data; the regression coefficient estimates from the smoothed
estimating function (solid line), along with the 95% pointwise confidence intervals (dotted line), and the the regression coefficient estimates from the
nonsmoothed estimating function (dashed line)

in dotted lines at an equally spaced 𝜏-grid for sarcoma-specific cumulative incidence. In Figure 2, we also provide the regression
estimates from the nonsmoothed estimating function in dashed lines. It can be seen that the effects of chemotherapy are positive
roughly for 𝜏 < 0.2, which is statistically significant, while 𝜏 ≤ 0.1, and negative for 𝜏 ≥ 0.2 in reducing sarcoma-related death.
This observation, coupled with the 95% confidence intervals, indicates that the effect of chemotherapy is only significant for
patients at lower quantiles but becomes insignificant for higher quantiles. On the contrary, patients who received radiation died
later than those who did not. The associated coefficient estimates are positive for all the observed quantiles. The effect of tumor
size has an adverse effect on patients' survival times, and this adverse effect tends to be larger as 𝜏 grows. It also shows that
Age ≥ 60 seems to have a negative effect on survival and the associated pointwise confidence intervals are under zero for most
of the observed quantiles.
Next, we apply multiple quantile regression models for competing risks. Table 3 summarizes the quantile coefficient estimates,
corresponding estimated standard errors, 95% confidence intervals, and p-values obtained from the proposed induced smoothing
model and Peng and Fine (2009)'s nonsmoothed model. For Peng and Fine (2009)'s method, we make inferences with 500
bootstrap replicates. We include chemotherapy, radiation, tumor size, age, upper (vs. lower) limb, and proximal site (yes or no)
as the covariates, and let 𝜏 ∈ {0.1, 0.2, 0.3, 0.4}. The estimation and inferences are challenging for higher quantiles, for which
the quantile functions may not be identifiable in the presence of competing risks. Overall, these two methods provide similar
coefficient estimates, while the estimators of the induced smoothing method have slightly larger estimated standard errors and
consequently inflated p-values. By studying different quantiles, our analysis therefore reveals a richer picture. Likewise, for
the univariate case, the effect of chemotherapy is initially effective on patients' survival and quickly becomes adverse, but not
significantly, for higher quantiles. Radiation is consistently effective for different quantiles and tumor size is another important
factor when treating sarcoma patients. Proximal site appears to be significant in some cases. This additional information may not
be detected by the proportional subdistribution hazards model. Hence, quantile-based varying coefficients may provide useful
insights into the association between the progression of sarcoma cancer and potential risk factors.
10 CHOI ET AL.

TABLE 3 Results from the soft tissue sarcoma data, showing the estimated quantile coefficients (Est), associated standard errors (SE), lower
bound (LB), and upper bound (UB) of the 95% Wald-type confidence intervals, and 𝑝-values for the quantiles of 𝜏 = 0.1, 0.2, 0.3, and 0.4a
Induced smoothing Peng and Fine (2009)
𝝉 Effect Est SE LB UB 𝒑-value Est SE LB UB 𝒑-value
0.1 Intercept 0.076 0.245 −0.405 0.557 0.756 0.058 0.104 −0.146 0.261 0.579
Chemotherapy 0.249 0.138 −0.022 0.520 0.072 0.280 0.108 0.067 0.492 0.010
Radiation 0.444 0.142 0.166 0.723 0.002 0.326 0.113 0.104 0.547 0.004
Tumor size −0.318 0.125 −0.564 −0.073 0.011 −0.231 0.096 −0.418 −0.043 0.016
Age −0.319 0.139 −0.591 −0.046 0.022 −0.208 0.112 −0.428 0.011 0.063
Upper limb 0.091 0.174 −0.251 0.432 0.603 0.098 0.101 −0.100 0.296 0.332
Proximal site −0.072 0.185 −0.435 0.291 0.698 −0.070 0.102 −0.269 0.130 0.494
0.2 Intercept 1.138 0.225 0.697 1.578 <0.001 1.042 0.119 0.809 1.276 <0.001
Chemotherapy −0.076 0.147 −0.365 0.212 0.604 −0.091 0.124 −0.334 0.152 0.464
Radiation 0.418 0.136 0.152 0.684 0.002 0.383 0.112 0.164 0.603 0.001
Tumor size −0.550 0.159 −0.862 −0.238 0.001 −0.497 0.112 −0.717 −0.277 <0.001
Age −0.311 0.146 −0.597 −0.025 0.033 −0.309 0.116 −0.537 −0.081 0.008
Upper limb 0.003 0.217 −0.423 0.429 0.989 −0.004 0.147 −0.292 0.284 0.977
Proximal site −0.295 0.214 −0.715 0.124 0.167 −0.213 0.127 −0.462 0.037 0.094
0.3 Intercept 1.927 0.315 1.310 2.544 <0.001 1.785 0.216 1.361 2.208 <0.001
Chemotherapy −0.242 0.212 −0.658 0.173 0.253 −0.183 0.173 −0.523 0.157 0.292
Radiation 0.578 0.214 0.158 0.997 0.007 0.466 0.196 0.081 0.851 0.018
Tumor size −0.722 0.218 −1.149 −0.294 0.001 −0.678 0.194 −1.059 −0.297 <0.001
Age −0.247 0.198 −0.636 0.142 0.214 −0.170 0.167 −0.498 0.157 0.307
Upper limb 0.415 0.537 −0.638 1.467 0.440 0.173 0.563 −0.931 1.277 0.759
Proximal site −0.548 0.332 −1.198 0.102 0.098 −0.426 0.214 −0.845 −0.007 0.046
0.4 Intercept 2.448 0.505 1.459 3.437 <0.001 2.388 0.317 1.866 3.110 <0.001
Chemotherapy −0.411 0.243 −0.888 0.065 0.091 −0.391 0.175 −0.733 −0.049 0.025
Radiation 0.719 0.236 0.256 1.182 0.002 0.698 0.223 0.262 1.134 0.002
Tumor size −0.868 0.249 −1.356 −0.380 <0.001 −0.811 0.226 −1.254 −0.368 <0.001
Age −0.306 0.208 −0.715 0.102 0.142 −0.256 0.160 −0.568 0.057 0.109
Upper limb 0.922 0.838 −0.720 2.565 0.271 0.578 0.716 −0.825 1.980 0.420
Proximal site −0.753 0.480 −1.693 0.187 0.116 −0.506 0.303 −1.100 0.087 0.095
a
Covariates include chemotherapy (yes or no), radiation (yes or no), tumor size (≥10 or not), age (≥60 or not), upper limb (upper vs. lower limb), and proximal site (yes
or no).

5 CONC LU SI O N A N D R E MA R K S

In this study, we consider an induced kernel smoothing technique for making inferences on competing risks regression quantiles.
The proposed method can be easily and effectively implemented by using standard algorithms such as Newton–Raphson, which
do not rely on statistical software. The inference procedure developed in this work is particularly useful for estimating the
variance of quantile regression parameters. Our numerical studies demonstrate that the smoothed estimating approach performs
well with practical sample sizes and variance estimation follows from basic asymptotic normality arguments. On the other hand,
variance estimation for the Peng and Fine (2009)'s method does not seem to be justified by asymptotic theory unless sample
size is large enough, necessitating extra inferential steps, such as bootstrapping. Our approach has the potential to provide more
computing-effective approach than Peng and Fine (2009)'s method. However, our experience reveals that the induced smoothing
estimators seem relatively more sensitive to the initial values if 𝜏 is close to boundary, in which case it is helpful to take Peng
and Fine (2009)'s estimator as an initial value.
The conditional competing risks quantile regression approach not only provides meaningful interpretations by using cumula-
tive incidence quantiles but also extends the AFT model for competing risks by relaxing some of the stringent model assumptions
(e.g., global linearity and unconditional independence) required by many existing procedures. By using the data example of the
CHOI ET AL. 11

sarcoma study, quantile regression is proven useful to describe the time-varying effect of adjuvant chemotherapy on the cause-
specific probability of death from sarcoma and subsequent crossing survival rates. Conventional Cox's or AFT models yield an
overall mean-based assessment of the covariate effects on survival for the entire follow-up period, and thus cannot distinguish
the survival difference that may occur at a specific stage of disease progression. Further, these approaches provide low statistical
power when crossing survival rates are concerned. This observation forced Cormier et al. (2004) to divide patients into two
groups, namely before and after survival crossing, to fit the Cox models. However, splitting the dataset in this manner could
bring about additional selection biases and is not recommended. In this setting, quantile regression analysis offers substantial
flexibility in accommodating the varying covariate effects, as it can directly model a certain quantile of interest and provide a
global picture of the relationship between the covariates and patients' survival.

CONFLICTS OF I NTEREST

The authors declare no conflicts of interest.

ACKNOWLEDGMENTS

Dr. Choi was supported by grants from Korea University (K1607341) and National Research Foundation (NRF) of Korea
(2017R1C1B1004817). Dr. Kang was supported by grant from NRF of Korea (2017R1A2B4005818). The research of Dr.
Huang was supported in part by USA NSF grant (DMS 1612965) and NIH grants (U54 CA096300, U01 CA152958, and 5P50
CA100632).

ORC ID
Sangbum Choi http://orcid.org/0000-0001-6983-5821

REFERENCES
Bang, H., & Tsiatis, A. A. (2002). Median regression with censored cost data. Biometrics, 58, 643–649.
Brown, B. M., & Wang, Y. G. (2005). Standard errors and covariance matrices for smoothed rank estimators. Biometrika, 92, 149–158.
Brown, B. M., & Wang, Y. G. (2006). Induced smoothing for rank regression with censored survival times. Statistics in Medicine, 26, 828–836.
Chiou, S. H., Kang, S., & Yan, J. (2015a). Rank-based estimating equations with general weight for accelerated failure time models: An induced
smoothing approach. Statistics in Medicine, 34, 1495–1510.
Chiou, S. H., Kang, S., & Yan, J. (2015b). Semiparametric accelerated failure time modeling for clustered failure times from stratified sampling.
Journal of American Statistical Association, 110, 621–629.
Choi, S., & Huang, X. (2015). Efficient semiparametric mixture inferences on cure rate models for competing risks. Canadian Journal of Statistics,
43, 420–435.
Cormier, J. N., Huang, X., Xing, Y., Thall, P. F., Wang, X., Benjamin, R. S., … Pisters, P. W. T. (2004). Cohort analysis of patients with localized
high-risk extremity soft tissue sarcoma treated at two cancer centers: Chemotherapy-associated outcomes. Journal of Clinical Oncology, 22, 4567–
4574.
Fine, J. P., & Gray, R. J. (1999). A proportional hazards model for the subdistribution of a competing risk. Journal of American Statistical Association,
94, 496–509.
Fleming, R. T., & Harrington, P. D. (1991). Counting processes and survival analysis. New Jersey: Wiley.
Gray, R. J. (1988). A class of 𝐾-sampling tests for comparing the cumulative incidence of a competing risk. The Annals of Statistics, 16, 1141–1154.
Huang, Y. (2002). Calibration regression of censored lifetime medical cost. Journal of American Statistical Association, 97, 318–327.
Johnson, L. M., & Strawderman, R. L. (2009). Induced smoothing for the semiparametric accelerated failure time model: Asymptotics and extensions
to clustered data. Biometrika, 96, 577–590.
Koenker, R., & Bassett, G. (1978). Regression quantiles. Econometrica, 46, 33–50.
Li, R., & Peng, L. (2011). Quantile regression for left-truncated semicompeting risks data. Biometrics, 67, 701–710.
Li, R., & Peng, L. (2015). Quantile regression adjusting for dependent censoring from semicompeting risks. Journal of Royal Statistical Society, Series
B, 77, 107–130.
12 CHOI ET AL.

Li, H., Zhang, J., & Tang, Y. (2012). Induced smoothing for the semiparametric accelerated hazards model. Computational Statistics and Data Analysis,
56, 4312–4319.
Lin, D. Y. (1997). Non-parametric inference for cumulative incidence functions in competing risks studies. Statistics in Medicine, 16, 901–910.
Pang, L., Lu, W., & Wang, H. J. (2012). Variance estimation in censored quantile regression via induced smoothing. Computational Statistics and
Data Analysis, 56, 785–796.
Peng, L., & Fine, J. P. (2007). Nonparametric quantile inference with competing risks. Biometrika, 94, 735–744.
Peng, L., & Fine, J. P. (2009). Competing risks quantile regression. Journal of American Statistical Association, 104, 1440–1453.
Peng, L., & Huang, Y. (2008). Survival analysis with quantile regression models. Journal of American Statistical Association, 103, 637–649.
Portnoy, S. (2003). Censored regression quantiles. Journal of American Statistical Association, 98, 1001–1012.
Powell, J. L. (1984). Least absolute deviations estimation for the censored regression model. Journal of Econometrics, 25, 303–325.
Powell, J. L. (1986). Censored regression quantiles. Journal of Econometrics, 32, 143–155.
Prentice, R. L., Kalbfleisch, J. D., Peterson, A. V., Flournoy, N., Farewell, V. T., & Breslow, N. E. (1978). The analysis of failure times in the presence
of competing risks. Biometrics, 34, 541–554.
Sun, Y., Wang, H. J., & Gilbert, P. B. (2012). Quantile regression for competing risks data with missing cause of failure. Statistica Sinica, 22, 703–728.
Wang, H., & Wang, L. (2009). Locally weighted censored quantile regression. Journal of American Statistical Association, 104, 1117–1128.
Ying, Z., Jung, S. H., & Wei, L. J. (1995). Survival analysis with median regression models. Journal of American Statistical Association, 90, 178–184.
Zeng, D., & Lin, D. Y. (2008). Efficient resampling methods for nonsmooth estimating functions. Biostatistics, 9, 355–363.

SUP PORTING IN FO R MAT I O N


Additional Supporting Information including source code to reproduce the results may be found online in the supporting infor-
mation tab for this article.

How to cite this article: Choi S, Kang S, Huang X. Smoothed quantile regression analysis of competing risks. Biometrical
Journal. 2018;1–13. https://doi.org/10.1002/bimj.201700104

APPE ND IX
In this section, we provide the proof for the consistency and asymptotic normality of the proposed estimator 𝜷̃ 𝑛 and the con-
sistency of the variance estimation. By using the consistency results of Peng and Fine (2009), 𝑆𝑛 (𝜷, 𝜏) converges uniformly in
probability to a continuous and deterministic function of 𝜷 that has a unique zero at 𝜷 0 ∈ ℝ𝑝 and is bounded in the neighbor-
hood of 𝜷 0 . Thus, it is sufficient to prove that 𝑆̃𝑛 (𝜷, 𝜏) − 𝑆𝑛 (𝜷, 𝜏) uniformly converges to 0 in probability for 𝜷 in the compact
neighborhood of 𝜷 0 as 𝑛 → ∞. √

Define 𝜉𝑖 (𝜷) = Φ{− 𝑛𝜂𝑖 (𝜷)∕𝜎𝑖 } − 𝐼{𝜂𝑖 (𝜷) < 0}, where 𝜂𝑖 (𝜷) = 𝑋𝑖 − 𝜷 ′ 𝑍𝑖 and 𝜎𝑖 = 𝑍𝑖′ Σ𝑍𝑖 . By manipulating the mar-
tingale representation of the Kaplan–Meier estimator (Fleming & Harrington, 1991), we have

𝑛 { }
̂ − 𝐺(𝑡)
𝐺(𝑡) ∑ 𝑡 ̂
𝐺(𝑢−) 𝑑𝑀𝑖𝑐 (𝑢)
=− ,
𝐺(𝑡) ∫ 𝐺(𝑢) 𝑌 (𝑢)
𝑖=1 0

̂ to 𝐺(⋅), shows that


which, along with the uniform convergence of 𝐺(⋅)

𝑛
∑ 𝑍𝑖 Δ𝑖 𝜉𝑖 (𝜷)
𝑛1∕2 {𝑆̃𝑛 (𝜷, 𝜏) − 𝑆𝑛 (𝜷, 𝜏)} = 𝑛−1∕2 =
𝑖=1
̂ 𝑇̃𝑖 )
𝐺(
𝑛 𝑛
∑ 𝑍𝑖 Δ𝑖 𝜉𝑖 (𝜷) ∑ 𝐿 𝑑𝑀𝑖𝑐 (𝑢)
= 𝑛−1∕2 − 𝑛−1∕2 𝐵(𝜷, 𝑢) + 𝑜𝑝 (1), (A.1)
𝑖=1 𝐺(𝑇̃𝑖 ) 𝑖=1
∫0 𝑌 (𝑢)
CHOI ET AL. 13


where 𝐵(𝜷, 𝑢) = lim𝑛→∞ 𝑛−1 𝑛𝑖=1 {Δ𝑖 𝑍𝑖 𝑌𝑖 (𝑢)𝜉𝑖 (𝜷)}∕𝐺(𝑇̃𝑖 ). According to the martingale central limit theorem, it can be checked
that the second term on the right-hand side of (A.1) is dominated by a bounded function and will vanish uniformly in probability
for 𝜷 in the neighborhood of 𝜷 0 . Let us note that for 𝑢 ∈ ℝ,
√ √
|𝑢{Φ(− 𝑛𝑢) − 𝐼(𝑢 < 0)}| = sign(𝑢){𝑢Φ(− 𝑛|𝑢|)},

where sign(𝑢) = 2𝐼(𝑢 ≥ 0) − 1, and hence



lim sup |𝑢{Φ(− 𝑛𝑢) − 𝐼(𝑢 < 0)}| = 0.
𝑛→∞ 𝑢∈ℝ

By applying this result to (A.1) and using the boundedness of 𝑍𝑖 and ‖Σ‖ = 𝑂(𝑛−1 ), it follows that the first term of (A.1) will
also converge in probability to 0 uniformly for 𝜷. Therefore, ‖𝑆̃𝑛 (𝜷, 𝜏) − 𝑆𝑛 (𝜷, 𝜏)‖ → 0 as 𝑛 → ∞.
To show that 𝑛1∕2 (𝜷̂ 𝑛 − 𝜷 0 ) and 𝑛1∕2 (𝜷̃ 𝑛 − 𝜷 0 ) converge to the same asymptotic distribution, it suffices to establish the fol-
lowing two convergence results: as 𝑛 → ∞, (i) 𝑛1∕2 ‖𝑆̃𝑛 (𝜷 0 , 𝜏) − 𝑆𝑛 (𝜷 0 , 𝜏)‖ → 0 and (ii) ‖∇𝜷 𝑆̃𝑛 (𝜷 0 , 𝜏) − 𝐴‖ → 0. Note that (i)
is implied from the previous argument and we show (ii) in the following. Let us write
𝑛
∑ Δ𝑖
𝐴̃ 𝑛 (𝜷 0 ) = ∇𝜷 𝑆̃𝑛 (𝜷 0 , 𝜏) = 𝑛−1 𝑍𝑖 𝑍𝑖′ 𝜑 (𝜷 ),
𝑖=1
̂ 𝑇̃𝑖 ) 𝑖 0
𝐺(
√ √
where 𝜑𝑖 (𝜷 0 ) = ( 𝑛∕𝜎𝑖 )𝜙{− 𝑛𝜂𝑖 (𝜷 0 )∕𝜎𝑖 }. For any vectors 𝒂, 𝒃 ∈ ℝ𝑝 ,
[ 𝑛
] [ 𝑛
]
[ ′ ] ∑ Δ𝑖 ∑
𝐸 𝒂 𝐴̃ 𝑛 (𝜷 0 )𝒃 = 𝒂 𝐸 𝑛
′ −1 ′
𝑍𝑖 𝑍𝑖 ′
𝜑𝑖 (𝜷 0 ) 𝒃 = 𝒂 𝑛 −1 ′
𝑍𝑖 𝑍𝑖 𝐸{𝜑𝑖 (𝜷 0 )} 𝒃,
𝑖=1 𝐺(𝑇̃𝑖 ) 𝑖=1

and by integration by parts


√ ( √ )
∞ 𝑛 𝑛𝜂𝑖 ∞√ √
𝐸{𝜑𝑖 (𝜷 0 )} = 𝜙 − 𝑓1𝑖 (𝜂𝑖 )𝑑𝜂𝑖 = 𝑓1𝑖 (0) + 𝑛𝜂𝑖 𝜙(− 𝑛𝜂𝑖 )𝑓1𝑖′ (𝜎𝑖 𝜂𝑖∗ )𝑑𝜂𝑖 ,
∫−∞ 𝜎𝑖 𝜎𝑖 ∫ −∞


where 𝜂𝑖 ≡ 𝜂𝑖 (𝜷 0 ) for brevity and 𝜂𝑖∗ is some point between 0 and 𝜂𝑖 . Since lim𝑛→∞ sup𝑢∈ℝ |𝑢𝜙(− 𝑛𝑢)| = 0 and from Assump-
tion (A3), we find that 𝐸{𝜑𝑖 (𝜷 0 )} → 𝑓𝑖 (0) as 𝑛 → ∞, and thus
{ 𝑛
}

′ ̃ ′ −1
lim 𝒂 𝐴𝑛 (𝜷 0 )𝒃 = 𝒂 lim 𝑛 𝑍𝑖 𝑍𝑖 𝑓𝑖 (0) 𝒃 = 𝒂′ 𝐴𝒃.

(A.2)
𝑛→∞ 𝑛→∞
𝑖=1

Following similar arguments in Pang et al. (2012), we can also show that
[ { 𝑛
} ]
∑ Δ 𝑖
var 𝒂′ 𝑛−1 𝑍𝑖 𝑍𝑖′ 𝜑𝑖 (𝜷 0 ) 𝒃 → 0.
𝑖=1 𝐺(𝑇̃𝑖 )

This result, coupled with (A.2), implies 𝐴̃ 𝑛 (𝜷 0 ) →𝑝 𝐴, and hence 𝐴̃ 𝑛 (𝜷̃ 𝑛 ) →𝑝 𝐴 according to the consistency result of 𝜷̃ 𝑛 to
𝜷 0 . It then follows from the asymptotic normality of 𝜷̂ 𝑛 (Peng & Fine, 2009) that 𝑛1∕2 (𝜷̃ 𝑛 − 𝜷 0 ) →𝑑 𝑁(𝟎, Ψ). In addition,
̃ 𝑛 (𝜷̃ 𝑛 ) = {𝐴̃ 𝑛 (𝜷̃ 𝑛 )}−1 Γ̂ 𝑛 (𝜷̃ 𝑛 ){𝐴̃ 𝑛 (𝜷̃ 𝑛 )}−1 converges in probability to Ψ = 𝐴−1 Γ𝐴−1 , since Γ̂ 𝑛 (𝜷̃ 𝑛 ) →𝑝 Γ as 𝑛 → ∞.
Ψ

You might also like