Web-based Supplementary Materials for A Modified Partial Likelihood Score Method for Cox Regression with Covariate Error Under the Internal

Similar documents
Convergence of random variables. (telegram style notes) P.J.C. Spreij

Chapter 3. Strong convergence. 3.1 Definition of almost sure convergence

7.1 Convergence of sequences of random variables

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.436J/15.085J Fall 2008 Lecture 19 11/17/2008 LAWS OF LARGE NUMBERS II THE STRONG LAW OF LARGE NUMBERS

Linear regression. Daniel Hsu (COMS 4771) (y i x T i β)2 2πσ. 2 2σ 2. 1 n. (x T i β y i ) 2. 1 ˆβ arg min. β R n d

17. Joint distributions of extreme order statistics Lehmann 5.1; Ferguson 15

STAT331. Example of Martingale CLT with Cox s Model

Lecture 19: Convergence

Lecture 7: Properties of Random Samples

Direction: This test is worth 250 points. You are required to complete this test within 50 minutes.


32 estimating the cumulative distribution function

Estimation of the Mean and the ACVF

7.1 Convergence of sequences of random variables

An Introduction to Randomized Algorithms

STA Object Data Analysis - A List of Projects. January 18, 2018

Random Matrices with Blocks of Intermediate Scale Strongly Correlated Band Matrices

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 3 9/11/2013. Large deviations Theory. Cramér s Theorem

Random Variables, Sampling and Estimation

Statistical Inference Based on Extremum Estimators

Estimation for Complete Data

Let us give one more example of MLE. Example 3. The uniform distribution U[0, θ] on the interval [0, θ] has p.d.f.

Solution to Chapter 2 Analytical Exercises

6.3 Testing Series With Positive Terms

Sequences and Series of Functions

FACULTY OF MATHEMATICAL STUDIES MATHEMATICS FOR PART I ENGINEERING. Lectures

January 25, 2017 INTRODUCTION TO MATHEMATICAL STATISTICS

Resampling Methods. X (1/2), i.e., Pr (X i m) = 1/2. We order the data: X (1) X (2) X (n). Define the sample median: ( n.

Lecture 2: Concentration Bounds

Probability 2 - Notes 10. Lemma. If X is a random variable and g(x) 0 for all x in the support of f X, then P(g(X) 1) E[g(X)].

ACCESS TO SCIENCE, ENGINEERING AND AGRICULTURE: MATHEMATICS 1 MATH00030 SEMESTER / Statistics

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 21 11/27/2013

Lecture 15: Learning Theory: Concentration Inequalities

Generalized Semi- Markov Processes (GSMP)

Since X n /n P p, we know that X n (n. Xn (n X n ) Using the asymptotic result above to obtain an approximation for fixed n, we obtain

REAL ANALYSIS II: PROBLEM SET 1 - SOLUTIONS

Product measures, Tonelli s and Fubini s theorems For use in MAT3400/4400, autumn 2014 Nadia S. Larsen. Version of 13 October 2014.

ECE 901 Lecture 14: Maximum Likelihood Estimation and Complexity Regularization

LECTURE 8: ASYMPTOTICS I

Estimation of the essential supremum of a regression function

Regression with an Evaporating Logarithmic Trend

Stochastic Simulation

ECE 901 Lecture 12: Complexity Regularization and the Squared Loss

Topic 9: Sampling Distributions of Estimators

Topic 9: Sampling Distributions of Estimators

Chapter 6 Infinite Series

Detailed proofs of Propositions 3.1 and 3.2

Section 14. Simple linear regression.

Frequentist Inference

ECONOMETRIC THEORY. MODULE XIII Lecture - 34 Asymptotic Theory and Stochastic Regressors

Statistical Inference (Chapter 10) Statistical inference = learn about a population based on the information provided by a sample.

Glivenko-Cantelli Classes

Appendix to: Hypothesis Testing for Multiple Mean and Correlation Curves with Functional Data

MATH/STAT 352: Lecture 15

The standard deviation of the mean

MATH 413 FINAL EXAM. f(x) f(y) M x y. x + 1 n

Entropy Rates and Asymptotic Equipartition

Lecture 3 The Lebesgue Integral

1 Convergence in Probability and the Weak Law of Large Numbers

Lecture Chapter 6: Convergence of Random Sequences

MAS111 Convergence and Continuity

LECTURE 14 NOTES. A sequence of α-level tests {ϕ n (x)} is consistent if

Application to Random Graphs

62. Power series Definition 16. (Power series) Given a sequence {c n }, the series. c n x n = c 0 + c 1 x + c 2 x 2 + c 3 x 3 +

Statistics 511 Additional Materials

Topic 9: Sampling Distributions of Estimators

CHAPTER 5. Theory and Solution Using Matrix Techniques

b i u x i U a i j u x i u x j

STAT Homework 1 - Solutions

Expectation and Variance of a random variable

2.2. Central limit theorem.

The picture in figure 1.1 helps us to see that the area represents the distance traveled. Figure 1: Area represents distance travelled

Ma 530 Introduction to Power Series

Simulation. Two Rule For Inverting A Distribution Function

4. Partial Sums and the Central Limit Theorem

Lecture 3 : Random variables and their distributions

Slide Set 13 Linear Model with Endogenous Regressors and the GMM estimator

Introduction to Probability. Ariel Yadin

EECS564 Estimation, Filtering, and Detection Hwk 2 Solns. Winter p θ (z) = (2θz + 1 θ), 0 z 1

Mathematical Statistics - MS

DS 100: Principles and Techniques of Data Science Date: April 13, Discussion #10

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 6 9/23/2013. Brownian motion. Introduction

Summary and Discussion on Simultaneous Analysis of Lasso and Dantzig Selector

Integrable Functions. { f n } is called a determining sequence for f. If f is integrable with respect to, then f d does exist as a finite real number

Direction: This test is worth 150 points. You are required to complete this test within 55 minutes.

Solutions to HW Assignment 1

Randomized Algorithms I, Spring 2018, Department of Computer Science, University of Helsinki Homework 1: Solutions (Discussed January 25, 2018)

Asymptotic Results for the Linear Regression Model

STA6938-Logistic Regression Model

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 2 9/9/2013. Large Deviations for i.i.d. Random Variables

1 Introduction to reducing variance in Monte Carlo simulations

Sieve Estimators: Consistency and Rates of Convergence

COMPUTING SUMS AND THE AVERAGE VALUE OF THE DIVISOR FUNCTION (x 1) + x = n = n.

EFFECTIVE WLLN, SLLN, AND CLT IN STATISTICAL MODELS

Bull. Korean Math. Soc. 36 (1999), No. 3, pp. 451{457 THE STRONG CONSISTENCY OF NONLINEAR REGRESSION QUANTILES ESTIMATORS Seung Hoe Choi and Hae Kyung

Economics 241B Relation to Method of Moments and Maximum Likelihood OLSE as a Maximum Likelihood Estimator

EXAMINATIONS OF THE ROYAL STATISTICAL SOCIETY

1 = δ2 (0, ), Y Y n nδ. , T n = Y Y n n. ( U n,k + X ) ( f U n,k + Y ) n 2n f U n,k + θ Y ) 2 E X1 2 X1

This exam contains 19 pages (including this cover page) and 10 questions. A Formulae sheet is provided with the exam.

Lecture 5: Parametric Hypothesis Testing: Comparing Means. GENOME 560, Spring 2016 Doug Fowler, GS

Transcription:

Web-based Supplemetary Materials for A Modified Partial Likelihood Score Method for Cox Regressio with Covariate Error Uder the Iteral Validatio Desig by David M. Zucker, Xi Zhou, Xiaomei Liao, Yi Li, ad Doa Spiegelma

Itroductio, Notatio, ad Prelimiaries I this documet we preset the details of the asymptotic theory for the modified score estimator. At the ed of the documet we preset tables with the simulatio results for the commo disease case that we referred to i the mai paper. Deote the true value of θ by θ. We assume that θ lies i a suitably large compact set Q whose iterior cotais θ. Two additioal assumptios will be itroduced later, with clear sigpostig. Let τ deote the maximum follow-up time. All assertios below of uiform covergece of fuctios of θ ad/or t refer to θ Q ad t [, τ]. We use the otatio used i the mai paper. From ow o, i the various S s defied i the mai paper we will write θ istead of β, α. We ow defie some additioal otatio. S e t, θ = S 2a t, θ = S 2b t, θ = S 2c t, θ = S 2d t, θ = ω j Y j t X j α expβ T X j t ω j Y j tx j X T j expβ T X j t ω j Y j t Xα Xα T expβ T Xαt ω j Y j t Xα Xα T expβ T Xαt ω j Y j t XαX T expβ T Xt E W t, θ = E[Y t expβ T Xα] E W W t, θ = E[Y t Xα expβ T Xα] XX t, θ = E[Y txx T expβ T X] E W W W t, θ = E[Y t Xα Xα T expβ T Xα] E W XX t, θ = E[Y t XαX T expβ T X] I additio, we defie Nt = i= N it ad N t = E[Nt]. We ow itroduce the followig assumptio: Assumptio : t, θ is bouded below over t ad θ.

This assumptio implies that there is a positive probability of reachig the maximum follow-up time τ with either a evet or a cesorig i the ope iterval, τ. I view of the fuctioal strog law of large umbers Aderse ad Gill, 982, Appedix III, the followig covergece relatios hold uiformly i t ad θ almost surely: S a t, θ π t, θ, S b t, θ πe W t, θ, S c t, θ πe W t, θ S d t, θ t, θ, S a t, θ πx t, θ, S b t, θ πe W W t, θ 2 S c t, θ πe W W t, θ, S d t, θ πe W X t, θ, S e t, θ πe W X t, θ 3 S 2a t, θ πxx t, θ, S 2b t, θ πe W W W t, θ 4 S 2c t, θ πe W W W t, θ, S 2d t, θ πe W XX t, θ 5 It follows that the followig covergece relatios hold uiformly i t ad θ almost surely ote that φπ = π: S t, β t, θ, S t, θ s t, θ 6 Also, it is clear that α α almost surely. Recall the coutig process represetatio of the score fuctio U MS θ: U MS θ = i= [ω i X i + ω i X i α]dn i t S t, θ dnt 7 S t, θ We ca write U MS θ = i= [ω i X i + ω i X i α]dn i t s t, θ t, θ dnt + R θ with R θ = S t, θ S t, θ s t, θ dnt t, θ Give the uiform covergece of S t, θ ad S t, θ, the assumptio that t, θ is bouded below, ad the fact that Nτ, we obtai the result that R θ uiformly i θ. Recall ow the defiitio dm i t = dn i t Y i t expβ T X i tλ tdt. Accordig to coutig 2

process theory Gill 984, M i t is a mea-zero martigale process w.r.t. the history defied by F t = σx i ; Y i s, N i s, s [, t], i =,...,. Give the oiformative measuremet error assumptio ad the fact that ω i is idepedet of all the other basic radom variables associated with idividual i, M i t is also a martigale w.r.t. the history defied by G t = σx i, W i, ω i, Y i s, N i s, s [, t], i =,...,. Cosistecy We ca write i= ω i X i dn i t = ω i X i dm i t + i= S a β, αλ t dt ad i= ω i X i α dn i t = ω i X i α dm i t + i= S e β, αλ t dt Similarly, s t, θ t, θ dnt = i= s t, θ τ t, θ dm it + s t, θ t, θ S dt, θ λ tdt Hece, U MS θ = S a β, αλ t dt + +R θ + Mθ S e β, αλ t dt s t, θ t, θ S dt, θ λ tdt where Mθ = i= ω i X i + ω i W i s t, θ dm i t t, θ Now, Mθ is the sample mea of i.i.d. mea-zero r.v. s, ad therefore coverges a.s. to, ad sice Q is compact ad t, θ ad s t, θ are cotiuous i θ uiformly over t the covergece is uiform i θ. Usig this result, alog with the covergece results -5 ad 3

the fact that α α a.s., we fid that U MS β, α coverges uiformly almost surely to uβ = s t, θ s t, β, α t, β, α t, θ λ t dt. We have that times the matrix of derivatives of U MS θ with respect to β is give by D ββ θ = Ṡt, θ S t, θ S t, θ S t, θ T Ṡ t, θ dnt 8 S t, θ where Sa t, θ Ṡ t, θ = S a t, θ + S b t, θ S c t, θ Sb t, θ Ṡ t, θ = S 2a t, θ + S 2b t, θ + S c t, θ + S d t, θ S c t, θ [ Sa t, θ + S b t, θ S c t, θ S 2d t, θ S 2c t, θ [ Sb t, θ S c t, θ Sb t, θ S c t, θ Sa t, θ S c t, θ ] T Sc t, θ S c t, θ ] Sc t, θ S c t, θ The limitig value of D ββ θ is give by [ τ XX t, θ d ββ θ = π t, θ + π EXX t, θ t, θ [ E W XX t, θ t, θ ] T EXX t, θ t, θ λ tdt t, θ EW X t, θ t, θ ] T EXX t, θ t, θ λ tdt t, θ which is times the matrix of derivatives of uθ with respect to β. The first term i the above expressio is a positive defiite matrix for all θ. At this poit, we itroduce a additioal assumptio: Assumptio 2: d ββ θ is osigular. By ispectio of uθ, we see that uβ, α =. Give this fact, the assumed osigularity of d ββ θ, the covergece of α to α, ad the uiform covergece of U MS θ to uθ ad D ββ θ to d ββ θ, it follows from the argumets of Foutz 977 that there exists a uique root of the score equatio U MS β, α = that coverges to β. 4

Asymptotic Distributio of βms The backgroud for the derivatio of the asymptotic distributio of β MS is preseted i Sectio 2.2 of the mai paper. We start here with the derivatio of U θ. As i Zucker ad Spiegelma 28, we use the argumet of Li ad Wei 989. I what follows, the symbol. = will deote equality up to egligible terms. Let us recall the coutig process represetatio of U θ: U θ = i= ω i X i + ω i X i αdn i t S t, θ S t, θ dnt We ow proceed to aalyze this quatity. For brevity, we will omit the argumets t ad θ. We ca write U θ = U a θ U θ with U a θ = U b θ = b i= S S ŝ ω i X i + ω i X i α ŝ dn =. S ŝ S dn i dn where ŝ t, β, α = πx t, β, α + πe W X t, β, α. Regardig U a θ, after some maipulatio we obtai U a θ = ω i i= + X i X ω i i= dn i π X i E W X dn i + π E W X X dn i dn E W X X dn i dn We ow tur to U b θ. We eed to work o the term S t, θ/s t, θ. We ca write S ŝ = S S ŝ S S E X. = E X = E X S ŝ S E X S ŝ ŝ S 5

Now, Sa S = S a + S b S c. = S a π + S b πe W + E φs a π φ S c πe W W E W = π ω j Y j e βt X j + ω j Y j e βt Xj E W E W φ ω j Y j e βt Xj E W By a similar argumet, S ŝ. = E W ω j Y j X j e βt X j X + + φ + E W E W X E W W ω j Y j Xj e βt Xj E W W ω j Y j Xj e βt X j E W X φ ω j Y j Xj e βt Xj E W W ω j Y j e βt Xj E W φ ω j Y j e βt Xj E W As a result, we ca write U θ. = U θ with where Z i = U θ = π m X i X φ φ E X τ i= ω i Z i dn i π Y i Xi e βt X i E W X E X dn + φ + π m i= E W X X dn i dn ω i Z 2 i Y i Xi e βt Xi E W W E X dn E W E W X E W W Y i e βt Xi E W E X dn + φŝ E W Y i e βt Xi E W E X dn ŝ π Y i X i e βt X i X E X dn Y i e βt X i E X dn 6

ad Z 2 i = X i E W X dn i + π E W X X dn i dn + E W E W X E W W Y i e βt Xi E W E X dn + Y i Xi e βt Xi E W W E X dn ŝ E W Y i e βt Xi E W E X dn I computig Ĉ as described i the mai paper, we substitute θ for θ, dnu for dn u, S for, S for ŝ, S b + S c for E W, π S a for X, π S d for E W X, ad S b + S c for E W W. We ow tur to the derivatio of the relevat compoets of Jacobia matrix Dθ. derivative of U θ with respect to β is give by 8. Let etry equals X ir / α t. The matrix ad whose remaiig rows are filled with s. We the have The X i deote the matrix whose r, t X i is a p p+p matrix whose first p rows are I p w T i U θ D βα = α = i= δ i ω i X i + i= δ i H2 S S H S S where ad H = S α = H 2 = S α = Sa i= S c i= ω i Y i β T X i e βt Xi ω i Y i X i e βt Xi + + S c S d S c + Sb S c Sb S 2 c i= S d S c i= i= Sa S b S 2 c i= ω i Y i Xi β T X i e βt Xi ω i Y i β T X i e βt Xi ω i Y i X i e βt X i i= i= ω i Y i X i e βt Xi ω i Y i β T X i e βt Xi ω i Y i β T X i e βt Xi i= ω i Y i Xi β T X i e βt Xi Fially, U2 θ D αα = α = ω i I p w i w T i i= 7

Table S: Simulatio results for the sigle-covariate commo disease case with idepedet measuremet error. β is the true value of β. Bias% is the relative bias, i.e. Bias%= ˆβ β /β. IQR is.74 times the iterquartile rage of the ˆβ values. SE is the mea of the estimated stadard error of ˆβ. SD is the empirical stadard deviatio of the ˆβ values. CR is the empirical coverage rate of the asymptotic 95% cofidece iterval. Methods cosidered: MS = modified score, CH = Che, RC = regressio calibratio, CC = complete case, NA = aive. Mea Media CorrX, W expβ β Method ˆβ Bias% ˆβ Bias% IQR SE SD CR.9.5.455 MS.436 -.5.423 -.8.5.982..938 CH.442 -.3.3999 -.4.6.45.89.938 RC.42 -.9.47 -.9.998.966.986.934 CC.3927-3..3847-5..394.472.53.945 NA.354-2.7.353-3.6.849.858.875.92.7.5.455 MS.453 -..3997 -.4.25.33.47.926 CH.428 -.7.3995 -.5.97.225.278.938 RC.399 -.6.3969-2..87.88.2.93 CC.3927-3..3847-5..394.472.53.945 NA.2484-38.7.2458-39.4.659.73.723.46.5.5.455 MS.442 -.3.3944-2.7.333.28.299.945 CH.3988 -.7.3936-2.9.295.334.393.93 RC.393-3..3923-3.2.227.22.26.942 CC.3927-3..3847-5..394.472.53.945 NA.445-64.4.445-64.4.53.538.545..9 2.5.963 MS.975..9 -.7.2.29.47.934 CH.984.2.96 -..7.78.26.922 RC.8965-2.2.8855-3.4.9.32.5.93 CC.938 2.4.9259..389.65.576.953 NA.7847-4.4.7759-5.3.854.9.934.664.7 2.5.963 MS.9256..9229.7.55.36.42.949 CH.923.5.978 -.9.49.37.469.938 RC.8695-5..8659-5.5.224.32.63.94 CC.938 2.4.9259..389.65.576.953 NA.522-43..526-43.2.76.722.75..5 2.5.963 MS.9295.4.928.6.683.546.635.96 CH.997.4.98 -.9.474.462.555.922 RC.8469-7.6.838-8.5.295.98.233.895 CC.938 2.4.9259..389.65.576.953 NA.292-68.3.2924-68..537.53.548..9 4..3863 MS.3883..3739 -.9.345.396.44.949 CH.389 -.3.3669 -.4.47.44.472.934 RC.359-5..368-5.7.92.49.23.887 CC.426 2.5.42..77.896.82.94 NA.42-7.7.3-8.4.4.986.57.324.7 4..3863 MS.498.7.397.8.78.662.757.938 CH.3858 -..3684 -.3.62.65.78.94 RC.2394 -.6.2362 -.8.249.28.272.75 CC.426 2.5.42..77.896.82.94 NA.7-48.8.797-48.8.777.736.79..5 4..3863 MS.48 2.3.4..268.97.283.953 CH.3865..378 -..78.679.787.98 RC.926-4..892-4.2.259.24.3.633 CC.426 2.5.42..77.896.82.94 NA.387-72.5.38-72.5.543.524.534. 8

Table S2: Simulatio results for the sigle-covariate commo disease case with depedet measuremet error. β is the true value of β. Bias% is the relative bias, i.e. Bias%= ˆβ β /β. IQR is.74 times the iterquartile rage of the ˆβ values. SE is the mea of the estimated stadard error of ˆβ. SD is the empirical stadard deviatio of the ˆβ values. CR is the empirical coverage rate of the asymptotic 95% cofidece iterval. Methods cosidered: MS = modified score, CH = Che, RC = regressio calibratio, CC = complete case, NA = aive. Mea Media CorrX, W expβ β Method ˆβ Bias% ˆβ Bias% IQR SE SD CR.9.5.455 MS.433 -.5.45 -.2.957.98.3.938 CH.438 -.4.396-2.3.2.4.9.934 RC.48 -.2.4 -.3.968.958.982.93 CC.3927-3..3847-5..394.472.53.945 NA.3556-2.3.352-3.4.847.857.877.92.7.5.455 MS.45 -..42 -.3.68.32.53.93 CH.424 -.8.3965-2.2.58.224.284.934 RC.3984 -.8.396-2.3.9.82.5.934 CC.3927-3..3847-5..394.472.53.945 NA.247-39..2466-39.2.659.76.725.387.5.5.455 MS.44 -.4.399-3.4.359.28.38.945 CH.398 -.8.3937-2.9.33.332.393.922 RC.393-3..3873-4.5.284.2.222.938 CC.3927-3..3847-5..394.472.53.945 NA.447-64.3.439-64.5.58.537.55..9 2.5.963 MS.967..99 -.6.36.6.75.938 CH.967..979 -.9.47.8.285.922 RC.8875-3..8845-3.5.96.5.44.922 CC.938 2.4.9259..389.65.576.953 NA.785-4.8.7763-5.3.84.893.942.64.7 2.5.963 MS.9237.8.957 -..57.378.42.949 CH.925.5.96 -.6.444.365.477.934 RC.8652-5.6.865-6..239.6.6.9 CC.938 2.4.9259..389.65.576.953 NA.535-44..546-43.8.768.7.765..5 2.5.963 MS.928.3.9248.9.76.558.658.938 CH.988.3.93 -.5.42.457.556.926 RC.8473-7.5.844-7.9.38.2.244.92 CC.938 2.4.9259..389.65.576.953 NA.292-68.2.2938-67.9.565.525.56..9 4..3863 MS.386 -..375 -..47.575.6.953 CH.3778 -.6.3625 -.7.49.4.486.98 RC.2942-6.6.2843-7.4.22.26.27.824 CC.426 2.5.42..77.896.82.94 NA.26-8.8.8-9.4.58.953.92.25.7 4..3863 MS.424.2.377 -..896.8.86.94 CH.3846 -..3658 -.5.594.597.694.94 RC.2325 -..2298 -.3.239.98.279.75 CC.426 2.5.42..77.896.82.94 NA.6987-49.6.72-49.4.825.75.838..5 4..3863 MS.477 2.3.3872..224.277.224.94 CH.3853 -..375 -..725.673.783.922 RC.963-3.7.97-3.7.274.247.3.66 CC.426 2.5.42..77.896.82.94 NA.3856-72.2.3868-72..552.59.559. 9

Table S3: Simulatio results for the multiple-covariate commo disease case with idepedet covariates ad idepedet measuremet error. β is the true value of β. Bias% is the relative bias, i.e. Bias%= ˆβ β /β. IQR is.74 times the iterquartile rage of the ˆβ values. SE is the mea of the estimated stadard error of ˆβ. SD is the empirical stadard deviatio of the ˆβ values. CR is the empirical coverage rate of the asymptotic 95% cofidece iterval. Methods cosidered: MS = modified score, CH = Che, RC = regressio calibratio, CC = complete case, NA = aive. Mea Media CorrX, W expβ β Method ˆβ Bias% ˆβ Bias% IQR SE SD CR.9.5.455 MS.486 3.3.424.7.25.6.99.96 CH.4244 4.7.4225 4.2.82.84.2.94 RC.448 2.3.482.7.3.986.972.949 CC.438 8..4545 2..439.538.69.93 NA.363 -.5.3538-2.7.875.875.849.933.7.5.455 MS.4264 5.2.496 3.5.22.88.24.949 CH.4324 6.6.4438 9.4.355.266.397.926 RC.438 2..484.7.63..42.942 CC.438 8..4545 2..439.538.69.93 NA.258-38..2478-38.9.782.727.728.4.5.5.455 MS.4342 7..4337 7..365.357.426.942 CH.4357 7.5.4458 9.9.4.372.549.898 RC.427.8.454 2.5.24.228.29.938 CC.438 8..4545 2..439.538.69.93 NA.432-64.7.426-64.8.576.55.562..9 2.5.963 MS.9463 3.3.9359 2..267.28.284.938 CH.948 3.5.9377 2.3.233.243.339.922 RC.95 -..996 -.7.27.67.25.938 CC.977 6..9652 5.3.74.78.793.938 NA.7982-2.9.798-2.9.936.938.967.734.7 2.5.963 MS.9679 5.6.95 3.8.598.536.642.926 CH.9639 5.2.959 3.9.55.442.668.895 RC.8887-3..883-3.9.93.76.27.94 CC.977 6..9652 5.3.74.78.793.938 NA.5265-42.5.5254-42.7.73.745.79..5 2.5.963 MS.983 7.3.9738 6.3.244.792.888.933 CH.976 6..9559 4.3.66.534.786.895 RC.8697-5..8677-5.3.382.252.338.96 CC.977 6..9652 5.3.74.78.793.938 NA.297-68.3.2896-68.4.554.547.582..9 4..3863 MS.446 4.3.428 2.6.72.734.863.957 CH.428 3..44.8.43.498.722.98 RC.3443-3..34-3.3.254.93.36.92 CC.465 5.4.44 3.9.277.979.249.957 NA.643-6..63-6..2.28.3.398.7 4..3863 MS.499 8..4768 6.5.2355.2573.268.933 CH.4524 4.8.4352 3.5.78.75.22.94 RC.2666-8.6.2594-9.2.334.28.392.82 CC.465 5.4.44 3.9.277.979.249.957 NA.7236-47.8.722-48..8.768.875..5 4..3863 MS.5228 9.8.54 8.3.258.2963.2777.965 CH.4645 5.6.4493 4.5.984.793.28.9 RC.227 -.9.2226 -.8.386.39.4.79 CC.465 5.4.44 3.9.277.979.249.957 NA.3865-72..3822-72.4.65.546.65.

Table S4: Simulatio results for the multiple-covariate commo disease case with idepedet covariates ad depedet measuremet error. β is the true value of β. Bias% is the relative bias, i.e. Bias%= ˆβ β /β. IQR is.74 times the iterquartile rage of the ˆβ values. SE is the mea of the estimated stadard error of ˆβ. SD is the empirical stadard deviatio of the ˆβ values. CR is the empirical coverage rate of the asymptotic 95% cofidece iterval. Methods cosidered: MS = modified score, CH = Che, RC = regressio calibratio, CC = complete case, NA = aive. Mea Media CorrX, W expβ β Method ˆβ Bias% ˆβ Bias% IQR SE SD CR.9.5.455 MS.487 3.3.47.5.987.6.99.96 CH.4244 4.7.427 4..53.78.86.98 RC.44 2..478.6.5.979.959.949 CC.438 8..4545 2..439.538.69.93 NA.3655-9.9.3558-2.2.86.875.844.94.7.5.455 MS.4259 5..49 3.4.296.83.99.945 CH.4327 6.7.443 8.8.36.263.382.9 RC.436 2..48.3.29.6.28.942 CC.438 8..4545 2..439.538.69.93 NA.255-38.2.245-39.6.745.723.76.46.5.5.455 MS.4339 7..4386 8.2.33.358.42.945 CH.4356 7.4.4484.6.47.369.537.92 RC.43.9.463 2.7.6.229.282.942 CC.438 8..4545 2..439.538.69.93 NA.44-64.5.435-64.6.59.55.555..9 2.5.963 MS.953 3.7.9345 2..3.328.473.945 CH.9463 3.3.9366 2.2.226.24.34.94 RC.984 -.9.9 -.7.8.5.4.938 CC.977 6..9652 5.3.74.78.793.938 NA.7973-3..797-3..97.924.962.722.7 2.5.963 MS.9655 5.4.9485 3.5.592.564.685.93 CH.9626 5..9445 3..473.435.659.887 RC.8868-3.2.8823-3.7.79.6.242.926 CC.977 6..9652 5.3.74.78.793.938 NA.5222-43..522-43.2.699.728.775..5 2.5.963 MS.988 7..9628 5..85.838.895.949 CH.9694 5.8.9496 3.6.57.528.782.92 RC.877-4.9.8723-4.8.39.255.33.96 CC.977 6..9652 5.3.74.78.793.938 NA.2933-68..29-68.3.575.544.574..9 4..3863 MS.454 4.9.44.7.767.85.73.96 CH.423 2.7.446 2..57.495.74.98 RC.3277-4.2.327-4.7.24.66.299.878 CC.465 5.4.44 3.9.277.979.249.957 NA.553-6.7.552-6.7.97..46.34.7 4..3863 MS.4954 7.9.4553 5..2283.2383.2426.945 CH.4496 4.6.4369 3.6.84.75.29.9 RC.2648-8.8.258-9.2.354.257.38.84 CC.465 5.4.44 3.9.277.979.249.957 NA.774-48.3.747-48.4.858.742.883..5 4..3863 MS.55 9.3.4847 7..2765.287.2953.96 CH.46 5.4.4452 4.3.229.786.293.94 RC.2282 -.4.2285 -.4.39.35.42.75 CC.465 5.4.44 3.9.277.979.249.957 NA.3935-7.6.3962-7.4.666.542.63.

Table S5: Simulatio results for the multiple-covariate commo disease case with depedet covariates ad idepedet measuremet error. β is the true value of β. Bias% is the relative bias, i.e. Bias%= ˆβ β /β. IQR is.74 times the iterquartile rage of the ˆβ values. SE is the mea of the estimated stadard error of ˆβ. SD is the empirical stadard deviatio of the ˆβ values. CR is the empirical coverage rate of the asymptotic 95% cofidece iterval. Methods cosidered: MS = modified score, CH = Che, RC = regressio calibratio, CC = complete case, NA = aive. Mea Media CorrX, W expβ β Method ˆβ Bias% ˆβ Bias% IQR SE SD CR.9.5.455 MS.422 -.8.46 -.2.87.39.9.945 CH.46..3987 -.7.267.8.264.926 RC.395-3.4.389-4..6.984.4.94 CC.47.4.3852-5..86.56.7.93 NA.3435-5.3.344-5..877.878.9.894.7.5.455 MS.495..49 -.9.353.234.353.922 CH.495..398 -.8.56.29.58.94 RC.3789-6.5.3747-7.6.64.9.93.98 CC.47.4.3852-5..86.56.7.93 NA.236-42.9.2295-43.4.737.75.765.296.5.5.455 MS.456 2.5.42 -..677.422.587.94 CH.4..3875-4.4.682.394.72.895 RC.3685-9..3563-2..332.92.32.895 CC.47.4.3852-5..86.56.7.93 NA.32-67.9.277-68.5.495.532.583..9 2.5.963 MS.9387 2.4.9467 3.3.396.258.36.945 CH.9396 2.5.945 3..455.254.432.96 RC.8892-3..99 -.7.74.59..926 CC.949 3.6.933.8.795.76.775.949 NA.7758-5.3.778-5..9.93.953.652.7 2.5.963 MS.9678 5.6.9463 3.3.76.75.848.949 CH.9493 3.6.9445 3..72.45.72.94 RC.8432-8..852-7..349.54.262.852 CC.949 3.6.933.8.795.76.775.949 NA.498-45.6.55-45.4.885.727.88..5 2.5.963 MS.983 7.3.968 5..28.92.25.96 CH.958 3.9.9454 3.2.87.539.78.94 RC.84 -.2.8263-9.8.456.22.353.89 CC.949 3.6.933.8.795.76.775.949 NA.272-7.5.275-7.4.628.527.63..9 4..3863 MS.439 3.8.4287 3..483.76.747.937 CH.422 2.4.486 2.3.7.473.8.94 RC.354-5..32-4.7.338.76.273.848 CC.4249 2.8.42.7.95.956.243.94 NA.4-7.8.446-7.4.45.6.9.347.7 4..3863 MS.4924 7.7.46 5.4.2357.2558.244.964 CH.4272 3..47 2.2.994.68.264.94 RC.238-2.4.23-2.5.398.259.424.73 CC.4249 2.8.42.7.95.956.243.94 NA.693-5.2.6886-5.3..75.928..5 4..3863 MS.5228 9.9.4622 5.5.2639.2956.2898.972 CH.429 3..499 2.4.894.757.24.98 RC.6-6.3.596-6.4.368.292.469.562 CC.4249 2.8.42.7.95.956.243.94 NA.368-73.9.367-73.9.647.529.65. 2

Table S6: Simulatio results for the multiple-covariate commo disease case with depedet covariates ad depedet measuremet error. β is the true value of β. Bias% is the relative bias, i.e. Bias%= ˆβ β /β. IQR is.74 times the iterquartile rage of the ˆβ values. SE is the mea of the estimated stadard error of ˆβ. SD is the empirical stadard deviatio of the ˆβ values. CR is the empirical coverage rate of the asymptotic 95% cofidece iterval. Methods cosidered: MS = modified score, CH = Che, RC = regressio calibratio, CC = complete case, NA = aive. Mea Media CorrX, W expβ β Method ˆβ Bias% ˆβ Bias% IQR SE SD CR.9.5.455 MS.426 -.7.43 -..9.4.93.945 CH.44.5.489.8.26.2.55.93 RC.399-3.6.3876-4.4.5.975.36.937 CC.47.4.3852-5..86.56.7.93 NA.3455-4.8.346-4.6.856.876.94.92.7.5.455 MS.493.9.3996 -.5.34.233.35.924 CH.45 2.4.423.7.365.37.382.926 RC.3787-6.6.3728-8..32.85.88.98 CC.47.4.3852-5..86.56.7.93 NA.238-43..2329-42.6.68.78.765.277.5.5.455 MS.453 2.4.437 -.4.535.424.575.94 CH.488 3.3.464 2.7.455.42.496.922 RC.3694-8.9.3598 -.3.327.92.34.898 CC.47.4.3852-5..86.56.7.93 NA.32-67.7.33-67.2.474.53.586..9 2.5.963 MS.9388 2.5.9463 3.3.355.297.353.96 CH.92.5.96 -.6.253.243.33.926 RC.8828-3.7.8892-3...4.94.94 CC.949 3.6.933.8.795.76.775.949 NA.7744-5.5.7734-5.6.984.95.959.636.7 2.5.963 MS.9588 4.6.933.6.864.62.765.953 CH.9253..95 -.5.627.439.567.922 RC.8427-8..8345-8.9.344.39.247.84 CC.949 3.6.933.8.795.76.775.949 NA.4944-46..4922-46.3.849.78.8..5 2.5.963 MS.9773 6.7.9525 3.9.29.879.25.965 CH.9278.3.966..665.535.653.926 RC.886 -.7.825 -.3.376.222.344.89 CC.949 3.6.933.8.795.76.775.949 NA.2746-7..2748-7..59.524.6..9 4..3863 MS.447 4.4.4253 2.8.696.225.2.953 CH.3876..3663 -.4.654.468.648.922 RC.36-6.2.327-6..277.47.257.836 CC.4249 2.8.42.7.95.956.243.94 NA.322-8.3.365-8..52.989.24.34.7 4..3863 MS.4932 7.7.448 4.5.22.2822.2633.964 CH.397.8.387..97.662.876.9 RC.249-2.4.224 -.9.342.236.388.69 CC.4249 2.8.42.7.95.956.243.94 NA.6872-5.4.6886-5.3.5.727.97..5 4..3863 MS.55 9.3.4676 5.9.2638.3234.35.98 CH.45..383 -.2.983.74.942.926 RC.7-5.5.77-5.5.34.298.458.59 CC.4249 2.8.42.7.95.956.243.94 NA.3723-73..375-73.2.688.527.638. 3

Table S7: Simulatio results for the multiple-covariate itermediate disease rate case with idepedet covariates ad idepedet measuremet error. β is the true value of β. Bias% is the relative bias, i.e. Bias%= ˆβ β /β. IQR is.74 times the iterquartile rage of the ˆβ values. SE is the mea of the estimated stadard error of ˆβ. SD is the empirical stadard deviatio of the ˆβ values. CR is the empirical coverage rate of the asymptotic 95% cofidece iterval. Mea Media CorrX, W expβ β Method ˆβ Bias% ˆβ Bias% IQR SE SD CR.9 4..3863 MS.4639 5.6.4283 3..22.286.2399.949 CH.4547 4.9.435 3.5.279.894.286.9.7 4..3863 MS.56 9.4.4683 5.9.2856.3223.346.957 CH.482 6.9.4833 7..2436.262.2528.938.5 4..3863 MS.5397..4899 7.5.37.3626.335.965 CH.4956 7.9.4875 7.3.276.2273.2633.933 Table S8: Simulatio results for various sample sizes for the multiple-covariate commo disease case with idepedet measuremet error. β is the true value of β. NN is the sample size of the mai study ad NV is the sample size of the iteral validatio sample. Bias% is the relative bias, i.e. Bias%= ˆβ β /β. IQR is.74 times the iterquartile rage of the ˆβ values. SE is the mea of the estimated stadard error of ˆβ. SD is the empirical stadard deviatio of the ˆβ values. CR is the empirical coverage rate of the asymptotic 95% cofidece iterval. Methods cosidered: MS = modified score, CH = Che, RC = regressio calibratio, CC = complete case, NA = aive. Mea Media CorrX, W expβ β Sample Size ˆβ Bias% ˆβ Bias% IQR SE SD CR.9.5.455 NN=5, NV=2.486 3.3.424.7.25.6.99.96 NN=25, NV=.484.7.456..466.442.468.938 NN=5, NV=2.47.4.462.2.36.34.34.949.7.5.455 NN=5, NV=2.4264 5.2.496 3.5.22.88.24.949 NN=25, NV=.47.3.492.9.54.52.543.93 NN=5, NV=2.496..48.6.332.362.368.949.5.5.455 NN=5, NV=2.4342 7..4337 7..365.357.426.942 NN=25, NV=.43.9.4.4.584.578.68.922 NN=5, NV=2.422.7.42.6.379.49.45.949.9 2.5.963 NN=5, NV=2.9463 3.3.9359 2..267.28.284.938 NN=25, NV=.9229.7.994.3.57.58.52.945 NN=5, NV=2.9226.7.923.7.393.367.379.938.7 2.5.963 NN=5, NV=2.9679 5.6.95 3.8.598.536.642.926 NN=25, NV=.927.2.925.5.62.637.657.926 NN=5, NV=2.9253..9243.9.448.447.472.942.5 2.5.963 NN=5, NV=2.983 7.3.9738 6.3.244.792.888.933 NN=25, NV=.934.5.926..74.73.74.98 NN=5, NV=2.9276.2.9267..483.52.527.93.9 4..3863 NN=5, NV=2.446 4.3.428 2.6.72.734.863.957 NN=25, NV=.3993.9.3962.7.622.677.679.945 NN=5, NV=2.3995..3999..442.472.444.977.7 4..3863 NN=5, NV=2.499 8..4768 6.5.2355.2573.268.933 NN=25, NV=.485.6.398.9.758.954.963.938 NN=5, NV=2.445.3.3993.9.547.69.59.957.5 4..3863 NN=5, NV=2.5228 9.8.54 8.3.258.2963.2777.965 NN=25, NV=.496.7.4..878.944.972.957 NN=5, NV=2.47.5.47..667.662.669.942 4

Table S9: Simulatio results for various sample sizes for the multiple-covariate commo disease case with idepedet measuremet error. β is the true value of β. NN is the sample size of the mai study ad NV is the sample size of the iteral validatio sample. Bias% is the relative bias, i.e. Bias%= ˆβ β /β. IQR is.74 times the iterquartile rage of the ˆβ values. SE is the mea of the estimated stadard error of ˆβ. SD is the empirical stadard deviatio of the ˆβ values. CR is the empirical coverage rate of the asymptotic 95% cofidece iterval. Methods cosidered: MS = modified score, CH = Che, RC = regressio calibratio, CC = complete case, NA = aive. Mea Media CorrX, W expβ β Sample Size ˆβ Bias% ˆβ Bias% IQR SE SD CR.9.5.455 NN=, NV=2.48.6.4..52.684.573.949 NN=, NV=4.453 -..424 -.8.47.526.464.965 NN=5, NV=.47.4.47.4.26.246.243.96 NN=, NV=2.459..469.3.73.75.63.959.7.5.455 NN=, NV=2.475 3..467.3.76.86.84.957 NN=, NV=4.486.8.436 -.5.679.726.639.977 NN=5, NV=.484.7.453 -..392.356.352.96 NN=, NV=2.456..447 -.2.244.252.233.968.5.5.455 NN=, NV=2.43 6..448 2.3.5.42.34.952 NN=, NV=4.45 2.4.48 -.2.96.78.9.98 NN=5, NV=.49.3.43.2.543.526.54.976 NN=, NV=2.462.2.457..36.369.354.953.9 2.5.963 NN=, NV=2.9429 2.9.9289.4.76.272.23.937 NN=, NV=4.9247.9.99.3.655.698.644.969 NN=5, NV=.924.6.9226.7.399.38.389.938 NN=, NV=2.993.3.99.3.26.27.259.96.7 2.5.963 NN=, NV=2.9657 5.4.9398 2.6.9.25.246.955 NN=, NV=4.9447 3..9286.3.33.367.25.968 NN=5, NV=.9292.4.9247.9.68.674.696.953 NN=, NV=2.9223.7.92.5.394.45.45.968.5 2.5.963 NN=, NV=2.264 2..9356 2..34.33.3227.938 NN=, NV=4.9689 5.7.939.7.828.228.273.96 NN=5, NV=.9396 2.5.929.4.3.39.77.97 NN=, NV=2.9278.3.9237.8.553.686.645.965.9 4..3863 NN=, NV=2.4395 3.8.475 2.3.245.333.49.943 NN=, NV=4.4273 3..467.5.53.447.65.952 NN=5, NV=.3985.9.3935.5.663.68.686.949 NN=, NV=2.48..3945.6.449.556.465.966.7 4..3863 NN=, NV=2.5253..44 3.9.38.3329.3486.936 NN=, NV=4.4776 6.6.445 4..277.272.223.972 NN=5, NV=.472 2.2.476.5.95.98.253.944 NN=, NV=2.473.5.42..76.84.78.957.5 4..3863 NN=, NV=2.547..4472 4.4.4363.45.444.925 NN=, NV=4.58 9..4439 4.2.2793.384.3242.966 NN=5, NV=.4499 4.6.425 2.8.724.964.248.957 NN=, NV=2.424 2.5.437 2..28.87.22.964 5