using geud to model volume effects for pneumonitis in prospective data from 4 non-small cell lung...

Using gEUD to model volume effects forpneumonitis in prospective data from 4 Non-SmallCell Lung Cancer (NSCLC) dose-escalation trials

E. Williams1, J. Belderbos2, W.R. Bosch3, F. Kong4, J.V. Lebesque2,F. Liu5, K.E. Rosenzweig6, W.L. Straube3, R.K. Ten Haken4, A. Jackson1

1Memorial Sloan-Kettering Cancer Center, New York, NY2The Netherlands Cancer Institute, Amsterdam, Netherlands3Washington University School of Medicine, St. Louis, MO

4University of Michigan, Ann Arbor, MI5New York Medical College, Valhalla, NY

6Mount Sinai School of Medicine, New York, NY

April 11, 2013

Overview

Goal: To model the dependence of radiation pneumonitis (RP) ongeneralized equivalent uniform dose (gEUD), in prospective data fromfour dose-escalation trials

Radiation PneumonitisWhat is it?

From DVH to gEUD to NTCPA short history of acronyms

The dataWho, what and where

MethodsDataset combination and analysis

Results and Conclusions

E. Williams (MSKCC) Pooled RP gEUD Analysis April 11, 2013 1 / 12

Radiation Pneumonitis (RP)

Lung damage from ionizing radiation is generally divided into:Early (acute) inflammatory damage: radiation pneumonitisLate chronic scarring: radiation fibrosis

In the treatment of NSCLC, RP is often dose limitingcomplicationRP in this study was defined as requiring steroids or worsetoxicity before 6 months from end of treatment

Toxicity grades: ≥ RTOG RP grade 3 or ≥ SWOG RP grade 2

Incidence of RP for Cancer of the Lung

RP Def. ofStudy Incidence RP

Byhardt 1998 1 27/461 (6%) RTOG≥Gd 3Inoue 2001 2 25/191 (13%) RTOG≥Gd 3Rancati 2003 3 14/84 (17%) SWOG≥Gd 2Seppenwoolde 2004 4 17/106 (16%) SWOG≥Gd 2Yorke 2005 5 10/78 (13%) RTOG≥Gd 3

Chest radiographs before (L) and after (R) treatment ofNSCLC in the left lung (arrow). Post-tx shows faint areasof increased opacity (RP) within field


Overview








From DVH to NTCP

Normal Tissue Complication Proability (NTCP) models attempt to reducecomplicated dosimetric and anatomic information into a single risk measure(e.g. probability of complication)

DVH-reduction models estimate complication probability under uniformirradition from nonuniform dose distributions

→ ?? →

??


From DVH to NTCP



→ MeanDose

→

DmeanMarks, IJROBP 76:2010

QUANTEC: Radiation Dose-Volume Effects in the Lung


From DVH to NTCP



��

→ VD →

VDMarks, IJROBP 76:2010

QUANTEC: Radiation Dose-Volume Effects in the Lung


From DVH to NTCP



→ gEUD →

gEUD


From DVH to NTCP



→ gEUD →

gEUD

Generalized Equivalent Uniform Dose:

gEUD(a) = (∑i(di)

aνi)1/a (Lyman Model with n = 1/a)


From DVH to NTCP



→ gEUD →

gEUD

Generalized Equivalent Uniform Dose:

gEUD(a) = (∑i(di)

aνi)1/a (Lyman Model with n = 1/a)

Lyman-Kutcher-Burman (LKB) model describes dose-response for uniformirradiation (gEUD) with a two-parameter (TD50,m) probit function:

P = 1√2π

∫ t−∞ e−x

2/2dx, t = gEUD(a)−TD50

m×TD50


Generalized Equivalent Uniform Dose

gEUD(a) = (∑

i(di)aνi)

1/a

Reduces DVH to a single biologically relevant index, accounting fordose heterogeneity (partial organ uniform irradiation)Represents the uniform dose which yields the same complication rateas the delivered dose distribution’Volume parameter’ a reflects volumetric dose-response oftissue

Value of a gEUD Description

High (a→∞) ∼ Max dose Series organsLow (a→ 0) ∼ Min dose Parallel organs

a = 1 = Mean dose

Marks, IJROBP 76:2010


Overview








The data

RP data pooled from four prospective dose-escalation trials:

→ Small number of events in individual datasets, combiningmulti-institutional data increases statistical power

# RP RPInstitute Patients Incidence Grade Protocol

Memorial Sloan-KetteringCancer Center (MSK)

78 13% (10/78) RTOG ≥ 3 Cancer 103;2005 6

Netherlands CancerInstitute (NKI)

86 16% (14/86) SWOG ≥ 2 IJROBP 66;2006 7

Radiation Therapy Onc.Group 93-11 (RTOG)

113 8% (9/113) RTOG ≥ 3 IJROBP 61;2005 8

University ofMichigan (UMich)

80 20% (16/80) SWOG ≥ 2 IJROBP 65;2006 9

Total 357 14% (49/357)

RP prospectively scored and primary end point in each trial


Overview








Methods

Radiation pneumonitis defined as toxicity scored ≥ RTOG Grade 3or ≥ SWOG Grade 2, within 6 months of RT

→ Treatment with steroids and/or oxygen

Same lung definition for DVHs from each study: excluding GTV

Institute Treated Doses [Gy] Fraction Size [Gy]

MSK 57.6− 90.0 1.8− 2.0NKI 60.7− 94.5 2.25

RTOG 70.9− 90.3 2.15UMich 63.0− 103 2.10

Linear quadratic correction to doses in 2 Gy fractions using α/β = 3 Gy

LKB maximum likelihood fits for −1 ≤ log10(a) ≤ 1 in 0.1 steps

→ Primary interest: volume paramater (a) wrt lung tissue complicationarchitecture (parallel/serial)


Overview









← parallel serial →E. Williams (MSKCC) Pooled RP gEUD Analysis April 11, 2013 10 / 12


LKB model fits: Combined data



LKB model best fit: Combined data

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1

−0.42

−0.4

−0.38

−0.36

−0.34

log10

(a)

log

lik

elih

oo

d p

er

de

gre

e o

f fr

ee

do

m

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 110

−10

10−8

10−6

10−4

10−2

100

log10

(a)

p−

va

lue

0 10 20 30 40 50 60 70 80 900

0.2

0.4

0.6

0.8

1

gEUD [Gy]

RP

ra

te o

bse

rve

d

log10

(a) = −0.2

p−val: 1.7e−09

MSK + NKI + RTOG + UMich

TD50 [95% CI] m [95% CI] a [95% CI]

19 Gy [14.7− 24.3 Gy] 0.4 [0.30− 0.52] 0.63 [0.32− 1.02]



Represents the largest gEUD study on prospective pneumonitisdata to date

The LKB model ’volume parameter’, a, was determined to be0.63 (n = 1.6) with a 95% confidence interval between 0.32 and1.02

Values of a < 1 suggest parallel tissue architecture; mean lungdose (a = 1) has not been excludedObserved heterogeneity between datasets in fit results motivatesfuture work→ e.g. incidental irradiation of the heart10


Backup

References I

[1] Byhardt RW, Scott C, Sause WT, et al. Response, toxicity, failure patterns, and survival in five radiation therapy oncologygroup (rtog) trials of sequential and/or concurrent chemotherapy and radiotherapy for locally advancednon-small-cell-carcinoma of the lung. Int J Radiat Oncol Biol Phys, 42:469–78, 1998.

[2] Inoue A, Kunitoh H, Saijo N, et al. Radiation pneumonitis in lung cancer patients: A retrospective study of risk factors andthe long-term prognosis. Int J Radiat Oncol Biol Phys, 49:649–55, 2001.

[3] Rancati T, Ceresoli GL, Cattaneo GM, et al. Factors predicting radiation pneumonitis in lung cancer patients: Aretrospective study. Radiother Oncol, 67:275–83, 2003.

[4] Seppenwoolde Y, De JK, Lebesque JV, et al. Regional differences in lung radiosensitivity after radiotherapy fornon-small-cell-lung cancer. Int J Radiat Oncol Biol Phys, 60:748–58, 2004.

[5] Yorke ED, Jackson A, Ling C, et al. Correlation of dosimetric factors and radiation pneumonitis for non-small-cell lungcancer patients in a recently completed dose escalation study. Int J Radiat Oncol Biol Phys, 63:672–682, 2005.

[6] Rosenzweig KE, Fox J, et al. Results of a phase i dose-escalation study using three-dimensional conformal radiotherapy inthe treatment of inoperable nonsmall cell lung carcinoma. Cancer, 103:2118–27, 2005.

[7] Belderbos JS, Heemsbergen W, Lebesque JV, et al. Final results of a phase i/ii dose escalation trial in non-small-cell lungcancer using three-dimensional conformal radiotherapy. Int J Radiat Oncol Biol Phys, 66:126–34, 2006.

[8] Bradley J, Graham MV, and Emami B. Toxicity and outcome results of rtog 9311: A phase i/ii dose-escalation study usingthree-dimensional conformal radiotherapy in patients with inoperable non-small-cell lung carcinoma. Int J Radiat OncolBiol Phys, 61:318–28, 2005.

[9] Kong FM, Hayman J, Ten Haken RK, et al. Final toxicity results of a radiation-dose escalation study in patients withnon-small-cell lung cancer (nsclc): Predictors for radiation pneumonitis and fibrosis. Int J Radiat Oncol Biol Phys,65:1075–86, 2006.

[10] Huang EX, Hope AJ, Bradley JD, and Deasy J. Heart irradiation as a risk factor for radiation pneumonitis. ActaOncologica, 50:51–60, 2011.


LKB Results

MSKCC

0 10 20 30 40 50 60 70 801

0.5

0

−0.5

−1

1

0.5

0

−0.5

−1

1

Average gEUDs

gEUD [Gy]

log

10(a

)

Avg. with comp

Central 68%

Avg. no comp

Central 68%

UMich

0 10 20 30 40 50 60 70 801

0.5

0

−0.5

−1

1

0.5

0

−0.5

−1

1

Average gEUDs

gEUD [Gy]

log

10(a

)

Avg. with comp

Central 68%

Avg. no comp

Central 68%

RTOG

0 10 20 30 40 50 60 70 801

0.5

0

−0.5

−1

1

0.5

0

−0.5

−1

1

Average gEUDs

gEUD [Gy]

log

10(a

)

Avg. with comp

Central 68%

Avg. no comp

Central 68%

NKI

0 10 20 30 40 50 60 70 801

0.5

0

−0.5

−1

1

0.5

0

−0.5

−1

1

Average gEUDs

gEUD [Gy]

log

10(a

)

Avg. with comp

Central 68%

Avg. no comp

Central 68%


LKB Results

MSKCC

gEUD [Gy]

log

10(a

)

Probability that RPS rate ≥ 20%

0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 751

0.90.80.70.60.50.40.30.20.1−0

−0.1−0.2−0.3−0.4−0.5−0.6−0.7−0.8−0.9

−1

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

UMich

gEUD [Gy]

log

10(a

)


0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 801

0.90.80.70.60.50.40.30.20.1−0

−0.1−0.2−0.3−0.4−0.5−0.6−0.7−0.8−0.9

−1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

RTOG

gEUD [Gy]

log

10(a

)


0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 951

0.90.80.70.60.50.40.30.20.1−0

−0.1−0.2−0.3−0.4−0.5−0.6−0.7−0.8−0.9

−1

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

NKI

gEUD [Gy]

log

10(a

)


0 5 10 15 20 25 30 35 40 45 50 55 60 65 701

0.90.80.70.60.50.40.30.20.1−0

−0.1−0.2−0.3−0.4−0.5−0.6−0.7−0.8−0.9

−1

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9


LKB Results

Combined: MSK + NKI + RTOG + UMich

gEUD [Gy]

log

10(a

)


0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 951

0.90.80.70.60.50.40.30.20.1−0

−0.1−0.2−0.3−0.4−0.5−0.6−0.7−0.8−0.9

−1

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9


LKB Results

MSKCC

gEUD [Gy]

log

10(a

)

Low 68% CL on RP rate

0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 751

0.90.80.70.60.50.40.30.20.1−0

−0.1−0.2−0.3−0.4−0.5−0.6−0.7−0.8−0.9

−1

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

UMich

gEUD [Gy]

log

10(a

)


0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 801

0.90.80.70.60.50.40.30.20.1−0

−0.1−0.2−0.3−0.4−0.5−0.6−0.7−0.8−0.9

−1

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

RTOG

gEUD [Gy]

log

10(a

)


0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 951

0.90.80.70.60.50.40.30.20.1−0

−0.1−0.2−0.3−0.4−0.5−0.6−0.7−0.8−0.9

−1

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

NKI

gEUD [Gy]

log

10(a

)


0 5 10 15 20 25 30 35 40 45 50 55 60 65 701

0.90.80.70.60.50.40.30.20.1−0

−0.1−0.2−0.3−0.4−0.5−0.6−0.7−0.8−0.9

−1

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5


LKB Results

Combined: MSK + NKI + RTOG + UMich

gEUD [Gy]

log

10(a

)


0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 951

0.90.80.70.60.50.40.30.20.1−0

−0.1−0.2−0.3−0.4−0.5−0.6−0.7−0.8−0.9

−1

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5


LKB Results

LKB model fits (zoom)


LKB Results

LKB model fits


LKB Results

LKB Log-likelihoods

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1−0.55

−0.5

−0.45

−0.4

−0.35

−0.3

−0.25

−0.2

log10

(a)

log lik

elih

ood p

er

degre

e o

f fr

eedom

MSK

LKB p-values

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 110

−10

10−9

10−8

10−7

10−6

10−5

10−4

10−3

10−2

10−1

100

log10

(a)

p−

valu

e

MSK

MSK

TD50 m a(Q p-value, I2 [95%CIs]) (Q p-value, I2 [95%CIs]) (Q p-value, I2 [95%CIs])

(0.75, 0 [0− 0.01]) (0.13, 0.57 [0.34− 0.79]) (0.33, 0 [0− 0.01])(0.08, 0.60 [0.35− 0.83]) (< 0.01, 0.89 [0.40− 0.1]) (0.06, 0.65 [0.3− 1])(< 0.01, 0.98 [0.24− 1]) (< 0.01, 0.85 [0.33− 1]) (0.02, 0.70 [0.46− 0.93])


LKB Results

LKB Log-likelihoods

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1−0.55

−0.5

−0.45

−0.4

−0.35

−0.3

−0.25

−0.2

log10

(a)

log lik

elih

ood p

er

degre

e o

f fr

eedom

MSK

NKI

LKB p-values

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 110

−10

10−9

10−8

10−7

10−6

10−5

10−4

10−3

10−2

10−1

100

log10

(a)

p−

valu

e

MSK

NKI

MSK+NKI


(0.75, 0 [0− 0.01]) (0.13, 0.57 [0.34− 0.79]) (0.33, 0 [0− 0.01])(0.08, 0.60 [0.35− 0.83]) (< 0.01, 0.89 [0.40− 0.1]) (0.06, 0.65 [0.3− 1])(< 0.01, 0.98 [0.24− 1]) (< 0.01, 0.85 [0.33− 1]) (0.02, 0.70 [0.46− 0.93])


LKB Results

LKB Log-likelihoods

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1−0.55

−0.5

−0.45

−0.4

−0.35

−0.3

−0.25

−0.2

log10

(a)

log lik

elih

ood p

er

degre

e o

f fr

eedom

MSKNKIUMich

LKB p-values

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 110

−10

10−9

10−8

10−7

10−6

10−5

10−4

10−3

10−2

10−1

100

log10

(a)

p−

valu

e

MSKNKIUMich

MSK+NKI+UMich


(0.75, 0 [0− 0.01]) (0.13, 0.57 [0.34− 0.79]) (0.33, 0 [0− 0.01])(0.08, 0.60 [0.35− 0.83]) (< 0.01, 0.89 [0.40− 0.1]) (0.06, 0.65 [0.3− 1])(< 0.01, 0.98 [0.24− 1]) (< 0.01, 0.85 [0.33− 1]) (0.02, 0.70 [0.46− 0.93])


LKB Results

LKB Log-likelihoods

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1−0.55

−0.5

−0.45

−0.4

−0.35

−0.3

−0.25

−0.2

log10

(a)

log lik

elih

ood p

er

degre

e o

f fr

eedom

MSKNKIUMichRTOG

LKB p-values

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 110

−10

10−9

10−8

10−7

10−6

10−5

10−4

10−3

10−2

10−1

100

log10

(a)

p−

valu

e

MSKNKIUMichRTOG

MSK+NKI+UMich+RTOG


(0.75, 0 [0− 0.01]) (0.13, 0.57 [0.34− 0.79]) (0.33, 0 [0− 0.01])(0.08, 0.60 [0.35− 0.83]) (< 0.01, 0.89 [0.40− 0.1]) (0.06, 0.65 [0.3− 1])(< 0.01, 0.98 [0.24− 1]) (< 0.01, 0.85 [0.33− 1]) (0.02, 0.70 [0.46− 0.93])


LKB Results

LKB Log-likelihoods

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1−0.55

−0.5

−0.45

−0.4

−0.35

−0.3

−0.25

−0.2

log10

(a)

log lik

elih

ood p

er

degre

e o

f fr

eedom

MSKNKIUMichRTOGCOMB

LKB p-values

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 110

−10

10−9

10−8

10−7

10−6

10−5

10−4

10−3

10−2

10−1

100

log10

(a)

p−

valu

e

MSKNKIUMichRTOGCOMB

MSK+NKI+UMich+RTOG


(0.75, 0 [0− 0.01]) (0.13, 0.57 [0.34− 0.79]) (0.33, 0 [0− 0.01])(0.08, 0.60 [0.35− 0.83]) (< 0.01, 0.89 [0.40− 0.1]) (0.06, 0.65 [0.3− 1])(< 0.01, 0.98 [0.24− 1]) (< 0.01, 0.85 [0.33− 1]) (0.02, 0.70 [0.46− 0.93])


History of gEUDWhat to expect?

Fitting institutional datasets to find best values of three parameters:

Volume parameter, a (= 1/n)50% complication tolerance dose, TD50

Cohort radiosensitivity slope parameter, m

Consensus based lung tolerance parameters (Emami et al., 1991)combined with a fit to clinical data (Burman et al., 1991) resultedLKB parameter values:

n = 1/a = 0.87, TD50 = 24.5 Gy, m = 0.18

Since then:

Marks, IJROBP 76:2010QUANTEC: Radiation Dose-Volume Effects in the Lung


log(a) log-likelihood

Logistic Regression LKB Model

Dashed line: 68% CLSolid line: 95% CL


log(a) p-values

Logistic Regression LKB


gEUD Rank-Sum p-values


LKB Model consistency

Data Q Q p-value I2 [95% CL]

MSK + NKITD50 0.1024 0.749 0 [0 - 0]

m 2.305 0.129 0.566 [0.343 - 0.789]a 0.9412 0.332 0 [0 - 0.008]

MSK + NKI + RTOGTD50 88.79 < 0.001 0.977 [0.728 - 1]

m 20.09 < 0.001 0.900 [0.328 - 1]a 9.756 0.008 0.795 [0.69 - 0.9]

MSK + NKI + UMichTD50 4.951 0.084 0.596 [0.362 - 0.831]

m 18.39 < 0.001 0.891 [0.402 - 1]a 5.762 0.056 0.653 [0.3 - 1]

MSK + NKI +UMich + RTOG

TD50 130.8 < 0.001 0.977 [0.239 - 1]m 20.23 < 0.001 0.852 [0.325 - 1]a 9.85 0.020 0.695 [0.457 -0.934]


Logistic Regression ROC Curves

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

False positive rate (1−Specificity)

Tru

e p

ositiv

e r

ate

(S

en

sitiv

ity)

ROC for classification by logistic regression

MSKAUC: 0.715

NKIAUC: 0.658

RTOGAUC: 0.797

UMichAUC: 0.914

MSK+NKI+RTOG+UMichAUC: 0.753


gEUDs


Combined gEUDs


Combined gEUD response (zoom)


Combined gEUD response


using geud to model volume effects for pneumonitis in prospective data from 4 non-small cell lung...

Health & Medicine

equivalent uniform dose

mi5new york medical

ann arbor

mo4university of michigan