# Paper Summary

Monotone Smoothing with Application to
Dose-Response Curves and
the Assessment of Synergism
Colleen Kelly and John Rice, Biometrics,
Vol. 46, No. 4, (Dec., 1990), pp. 1071-1085
Outline
Introduction
Monotone Smoothing
Assessing Synergism
Bootstrap Method
Procedure
Application
Result
Monotone Smoothing
0 1 1
1
To monotone smoothing is based on spline
function. Given a partition
... of an interval [ , ]
a spline function is a polynomial on each
interval [ , ).
We shall refer to the as "kno
p
j j
i
a b a b t t t
t t
t
+
+
· < < < ·
ts".
Monotone Smoothing

Given data points ( , ) ( 1,..., ), a smoothing
spline is the function that minimizes
i i
x y i n
µ
·
1
2 2
1
( ) ( ( )) [ ( )]
n
n
x
i i
x
i
Q Y x x dx
i
µ µ i µ
·
´´
· +
¿
l
The minimizing is a cubic spline with
a knot at every data point.
µ
Monotone Smoothing
as 0, tends to an interpolant of the data. i µ ÷
The smaller is, the quicker the running time,
especially when 0.
p
i ·
as , tends to the LS line fit to the data. i µ ÷×
Monotone Smoothing
Knot placement should depend on the density
of data points or on local behavior of curve.
Thus we control smoothing only with . i
1. Chosen the amount of smoothing to apply .
2. Leave-one-out cross-validation.
3. Doing a preliminary QR decomposition.
i
Assessing Synergism
Synergistic if its effect is more than
would be expected form what is known
about the effect of the individuak
constituent.
Antagonistic if the effect is less.
Asynergistic if effect is the same.
Assessing Synergism
Definition of synergism:
Let the repsone to dosages and of two
drug be denote by ( , ).
Assume 0 0, and ( , 0) (0, ) are
monotone.
x y
f x y
x y f x f y > >
Assessing Synergism
For given response value let and be
those dosages such that ( , 0) (0, ) .
if ( , ) , 1.
Synergistic if left side is less than 1.
Antagonistic if it is greater than 1.
z z
z z
z z
z x y
f x f y z
x y
f x y z
x y
· ·
· + ·
Bootstrap Method
Parametric bootstrap samples
Model could be binomial or poisson.
{ ¦
{ ¦
2
*( )
i
*(b) *( )
, ~(0, ), independent.
bootstrap residuals
ˆ
: 1... : 1...
ˆ
bootstrap sample for Y is Y .
b
b
Y X
i n b B
X
þ r r o
r r
þ r
· +
· = ·
· +
Procedure
Monotone spline
Parametric bootstrap sample
Check synergism
Application I
Data form Le Pelley and Sullivan
(1936).
Alcoholic solutions of rotenone,
pyrethrins, and a mixture of the two.
The mixture series was 1:5 ratio and
1:15 ratio.
Application I
2
2
Calculations
for the 1:5 ratio, (3) 81.19 was synergism.
for the 1:15 ratio, (3) 11.72 which is
not significant at 0.05.
)
)
o
·
·
·
Application I
Bliss (1939);
Finney, using probit analysis (1952);
Chou and Talalay with median effect
analysis (1983)
All investigators concluded there was
synergism in the 1:5 ratio and slight
synergism in the 1:15 ratio.
Application II
Data form Dr Stephen Howell of the UCSD
Cancer Center.
Ovarian cancer may be treated by VP-16 and
DPM.
There are 6 concentration levels for each
individual dose-reponse curve, with 18
observations.
There are 4 combination doses ratio 1:150 of
VP-16/DPM.
Application II
Parametric bootstrap based on Poisson
model for the random variability under
asynergy.
Responses are to increasing
concentrations of a conmbination of both
drugs in a 1:150 ratio.
Reference
Howell, S. B., Sanga, R., Vick, J., and Abramson, I. (1989a).
Comparison of the Synergistic Potentiation of Etoposide,
Doxorubicin, and Vinblastine Cytotoxicity by Dipyridamole.
Cancer Research 49, 4147-4153.