constrained bayes estimates of random effects when data are subject to a limit of detection
DESCRIPTION
Constrained Bayes Estimates of Random Effects when Data are Subject to a Limit of Detection. Reneé H. Moore Department of Biostatistics and Epidemiology University of Pennsylvania Robert H. Lyles, Amita K. Manatunga , Kirk A. Easley - PowerPoint PPT PresentationTRANSCRIPT
Constrained Bayes Estimates of Random Effects
when Data are Subject to a Limit of Detection
Reneé H. MooreDepartment of Biostatistics and Epidemiology
University of Pennsylvania
Robert H. Lyles, Amita K. Manatunga, Kirk A. EasleyDepartment of Biostatistics and Bioinformatics Emory
University
OUTLINE• Motivating Example• Background
• Review the Mixed Linear Model• Bayes predictor• Censoring under the Mixed Model• CB Predictors
• Application of Methodology for CB adjusted for LOD• Motivating Example• Simulation Studies
P2C2 HIV INFECTION STUDY: IS THIS CHILD’S HIV INFECTION AT GREATER RISK OF
RAPID PROGRESSION?
• 1990-1993 HIV transmitted from mother to child in utero
• Children in this dataset enrolled at birth or by 28 days of life
• HIV RNA Data at 3-6mos through 5 years of age
• Rapid Progression is defined as the occurrence of AIDS (Class C) or death before 18 months of age
• One goal of the study was to identify children with RP of disease because they may benefit from early and intense antiretroviral therapy
• One Indicator: high initial and/or steeply increasing HIV RNA levels over time
• Limitation: HIV RNA below a certain threshold not quantifiable
IS THIS CHILD’S HIV INFECTION AT GREATER RISK OF RAPID PROGRESSION?
• Given non-detects, how do we predict each child’s HIV RNA intercept and slope?
• Given non-detects, how do we predict each child’s HIV RNA level at a meaningful time point associated with RP?
THE MIXED LINEAR MODEL
Y: N by 1 outcome variable
X: known N by p fixed effects design matrix
: p by 1 vector of fixed effects
Z: known N by q random effects design matrix
u: q by 1 vector of random effects
e: N by 1 vector of random error terms
eZuXΨY
6
The Mixed Linear Model
Assumptions: E(u)= 0 and E(e)= 0
eZuXΨY
XΨ(Y) E
ZuXΨu)|(Y E
7
The Mixed Linear Model
V ar e( )e 2 V
V ar ( ) 'Y Z D Z Ve2
euDΔσ(u) ,2Var
eZuXΨY
8
BP (best predictor, Searle et.al. 1992): - minimizes - invariant to the choice of A, any pos. symmetric matrix
- holds regardless of joint distribution of (u, Y) - unbiased, i.e. - linear in Y
“Bayes Predictor”
M S E P E ( ~ ) ( ~ )'u u A u u
E(u|Y)
)()|~( uEYuE
eZuXΨY
Censoring under the mixed model* common feature of HIV data is that some values fall below a LOD
Ad hoc approach: substitute the LOD or a fraction of it for all values below the limit (Hornung and Reed, 1990)
Other Approaches:- Likelihood using the EM algorithm (Pettitt 1986, Hughes 1991)- Bayesian Methods (Carriquiry 1987)- Likelihood based approach using algortihms (Jacgmin-Gadda et.al. 2000)
Lyles et. al. (2000) maximize an integrated joint log-likelihood directly to handle informative drop-out and left censoring
10
Left-censoring under the mixed model
Lyles et.al. (2000) work under framework of
Y a b tij i i ij ij ( ) ( ) (i = 1, … , k ; j=1, …, ni)
L f f a b F d a b f a b f b da dbii
k
i
k
ij i ij
ni
Y i i i i i i ij ni
ni( ; ) ( ) [ { ( | , )}{ ( | , )} ( | ) ( ) ] .* Y Y Y
1 1 1
1
1 1
To get estimates of =
- ni1 detectable measurements: f(Yij|ai,bi)
- ni - ni1 non-detectable measurements: FY(d|ai ,bi)
T),,,,( 222
21
11
E(u|Y) can’t be calculated in practice!Why? - knowledge of all parameters in the joint distribution of (u,Y)
What do we do? - develop predictors based on their theoretical properties for known
parameters - evaluate effect of estimating unknown parameters via simulation
studies
eZuXΨY
Bayes Predictor (posterior mean)
• minimizes MSEP s.t.
- Prediction Properties (bias, MSEP) deteriorate for individuals whose random effects put them in tails of distribution
• Motivated research for alternatives to Bayes- Limited translation rules (Efron and Morris, 1971)- Constrained Bayes
tends to overshrink individual ui toward u
E(u|Y))()|~( uEYuE
13
Bayes with LOD
Lyles et al. (2000), using the MLEs from L( ;Y),
)(
)()|(),|()|()( *
*
i
iiiiiiiii
iiBi
Yf
ddffYfYEye
)(ii aor )(ii b
14
Censoring under the mixed model
None of the references cited for
dealing with left-censored longitudinal or repeated measures data
considered alternatives to the Bayes predictors for random effects
We Do!
15
Constrained Bayes Estimation
Louis (1984)• Expectation of sample variance of Bayes estimates is only a
fraction of expected variance of unobserved parameters derived from the prior
Shrinkage of the Bayes estimate
k
i
k
ii
Bi
Bi yEyeye
1 1
22 |)())()((
• Reduces shrinkage by matching first two moments of estimates with corresp. moments from posterior histogram of k normal means
16
Constrained Bayes Estimation
• Ghosh (1992): “recipe” to generalize Louis’ modified Bayes predictor for use with any distribution
• Lyles and Xu (1999): match predictor’s mean and variance with prior mean and variance of random effect
Ghosh (1992)
where
Constrained Bayes Estimation
(1 )CB B Bi iw w
12
1
2
( )( ) 1( )
H yw w yH y
1( ) ( | )1 y
kH y tr Var
1
1
kB B
iik
2
21
( ) ( )k
B B
ii
H y
(1) posterior mean matches sample mean
(2) posterior variance matches sample variance
Recall:
minimizes MSEP =
within the class of predictors of
CB
i 2
1( ) |
k
i ii
E y
(1 )CB B Bi iw w
( ) ( ,..., )T
i i ky
i
s.t.
satisfies (1) but NOT (2)B
i Bayes
Adjust
Con. Bayes
CONSTRAINED BAYES ESTIMATION
Ghosh (1992)
19
Constrained Bayes (CB) Estimation
We Do!Moore, Lyles, Manatunga (2010). Empirical constrained Bayes predictors accounting for non-detects among repeated Measures. Statistics in Medicine.
CB Predictors have been shown to
reduce the shrinkage of the Bayes estimate in an appealing way
BUT none had been adapted to account for censored data
CB Predictors with LOD
• Lyles (2000): adjusted Bayes estimate to accommodate data subject to a LOD but did not consider CB
• Moore (2010): combine Lyles (2000) BayesLOD and Ghosh (1992) CB CBLOD
RANDOM INTERCEPT-SLOPE MODEL
• Yij : Observed HIV RNA measurement at jth time point (tij) for ith child
• ai : ith child’s random intercept deviation
• bi : ith child’s random slope deviation
(i = 1, … , k ; j=1, …, ni)
i ia
i ib
Y a b tij i i ij ij ( ) ( )
i
i
ij
N
~
,
3
1
2
12
12 2
2
20
00
0 0
Intercept:
Slope:
Under random intercept-slope model, Lyles et.al. (2000) get MLEs of =
• ni1 detectable measurements: f(Yij|ai,bi)
• ni - ni1 non-detectable measurements: FY(d|ai
,bi)• d = limit of detection (LOD)
T),,,,( 222
21
*
1
1
1 1 1 1
( ; ) ( )
{ ( | , )}{ ( | , )} ( | ) ( )
Y Yk
ii
n nk i i
ij i i Y i i i i i i ii j j ni
L f
f Y a b F d a b f a b f b da db
BAYES PREDICTOR FOR LOD
*
*
( )( ) ( | )
( )i i i i
B
i i i
i
f Y d de y E Y
f Y
)( ii a or )( ii b
• minimizes MSEP s.t. posterior mean matches sample mean
Prediction properties (bias, MSEP) deteriorate for individuals whose random effects put them in the tail of the
distribution
strongly shrinks predicted βi toward β or αi toward α
CB PREDICTIONS OF αi AND βi
(i = 1, … , k ; j=1, …, ni)
Y a b tij i i ij ij ( ) ( )
1*
1 1 1 1
( ; ) [ { ( | , )}{ ( | , )} ( | ) ( ) ]n nk i i
ij i i Y i i i i i i ii j j ni
L Y f Y F d f f d d
)(
)()|(),|()|(~
*
*
i
iiiiiiiii
iiBi Yf
ddffYfYE
BBi
CBi ww ~)1(~~
COMPARING CONSTRAINED BAYES ESTIMATES
PARAMETER ESTIMATES BASED ON 2 METHODS:& Adjust Likelihood for LODAd Hoc Imputation
*
1
1
1 1 1 1
( ; ) ( )
{ ( | , )}{ ( | , )} ( | ) ( )
Y Yk
ii
n nk i i
ij i i Y i i i i i i ii j j ni
L f
f Y a b F d a b f a b f b da db
1( ; ) (Y ) ( | , ) ( | ) ( )
k
i ij i i i i i i ii
L f f Y a b f a b f b da db
Y
EXAMPLE SIMULATION STUDY
Table IV. (Moore et al. Statistics in Medicine, 2010)
EXAMPLE SIMULATION STUDY
IS THIS INFANT’S HIV INFECTION AT GREATER RISK OF RAPID PROGRESSION?
• Given non-detects, how do we predict each patient’s HIV RNA intercept and slope?
Viable option now available
• Given non-detects, how do we predict each patient’s HIV RNA level at a meaningful time point?
Extending our Stat in Med 2010 work
IS THIS CHILD’S HIV INFECTION AT GREATER RISK OF RAPID PROGRESSION?
P2C2 HIV Data (Chinen, J., Easley, K. et.al., J. Allergy Clin. Immunol. 2001)
• 343 HIV RNA measurements from 59 kids (range: 2-11, median=6)
• detection limit= 2.6 =log(400 copies/mL)• 6% (21 /343) of measurements < LOD• 19% (11 /59) kids have at least one meas. < LOD• 59 unique times (t) reached Class A HIV*
Goal: Predict Yit: HIV RNA level at time reached Class A
PREDICTION OF Yit = αi + t βi
• Goal of Predictor is to Match
• Compare and
• Recall: Yij= (α + ai) + (β + bi)tij + εij
, ,( ) ( )i CB i CB itVar t Var Y
, , ,,
it CB i CB i CBBut Y t BECAUSE
,it BY
,it CBY
2 2 21 12 2[ ] [ ] 2it itE Y t and Var Y t t
, , ,( | )Yit i i B i B it BE Y t Y
itY
, ,( ) ( ) ( ) ( )i CB i i CB iVar Var and Var Var
PREDICTION OF Yit = αi + t βi • Our previous CB predictors set out to match
but did not enforce constraint
, ,cov( , ) cov( , )i CB i CB i i
•We develop a CB predictor for the scalar R.V. Yit
OBJECTIVE 1: PREDICTION OF Yit = αi + t βi
What is new in adapting this extension of Ghosh’s CB?
• calculated for all k subjects at each unique t,it B
Y2
2 , ,1
( ) ( )k
it B t Bi
H y Y Y
11
1( ) (1 ) ( | )Y
k
iti
H y k Var Y
2 2,( | ) ( | )Y Yit it i it BVar Y E Y Y
PREDICTION OF Yit = αi + t βi 2 2
,( | | )Y)= ( Yit it i it BVar Y E Y Y
2 * 2 *
2 *
*
2 *
*
( | ) ( | )
1 ( , , )( )1 ( | , ) ( | ) ( )
( )
Y Y
YY
YY
it i it it i it
it it i i it i
i
it i it i it i i it i
i
E Y y f y dy
y f y dy df
y f y f y f dy df
1 1* *
1( | , ) ( | , ) ( | , )Y
ni n ni ii it i ij it i Y it i
jwhere f y f Y y F d y
P2C2 ALL 59 PREDICTORS OF Yit AT EACH t
0.0 0.2 0.4 0.6 0.8 1.0 1.2
23
45
67
Time Reached Class A
Pre
dict
or o
f Y
it =
alp
ha +
bet
a*t
THE 59 INDIVIDUAL PREDICTORS OF Yit AT EACH CHILD’S UNIQUE t
0.0 0.2 0.4 0.6 0.8 1.0 1.2
23
45
6
Time Reached Class A
Pre
dict
or o
f Y
it =
alp
ha +
bet
a*t • Bayes
o CB
SIMULATION STUDY FOR Yit
• Parameter Assumptions:
• 1500 subjects, each with five HIV RNA values taken every six months for 2 years
• 15% (1,089 /7,500) values < LOD = 2.8• 8 times (t) of interest = (0.03, 0.16, 0.36, 0.66, 0.85, 1.17, 1.32, 1.60)
2 2
1 2
2
12
5, 1 , 1, 0.1,- 0.26, 0.23
-
SIMULATION STUDY FOR Yit
Time Reached Class A HIV
Status (years)Mean Variance
0.03 4.98 0.98 4.99 0.86 0.95
0.16 4.85 0.92 4.86 0.80 0.89
0.36 4.65 0.82 4.66 0.73 0.79
0.66 4.35 0.69 4.36 0.62 0.68
0.85 4.16 0.62 4.17 0.56 0.61
1.17 3.84 0.52 3.85 0.47 0.51
1.32 3.69 0.47 3.70 0.43 0.47
1.60 3.41 0.41 3.41 0.36 0.41
it i iY t Sample Mean
, ,,it B it CBY Y ,it CBYSample Variance
,it BY
Sample Variance
SIMULATION STUDY FOR Yit
Bayes (closed circles) and CB (open circles) estimates of 80 simulated patients. The line plotted is . .
ˆˆ ˆitY t
.
SUMMARY
• Proposed LOD-adjusted CB predictors- Intercepts and Slopes- R.V. (Yit) at a meaningful time point
Relative to ad hoc and Bayes predictors:
“CBs Attenuate the Shrinkage”Better Match True Distribution of Random
Effects
Thank You!!