NEURAL NETWORKS AND BOOTSTRAP SIMULATION IN NEURAL NETWORKS AND BOOTSTRAP SIMULATION IN PREDICTION OF OUTCOME OF NON-SMALL CELL LUNG PREDICTION OF OUTCOME OF NON-SMALL CELL LUNG CANCER PATIENTS AFTER COMPLETE LOBECTOMIES AND CANCER PATIENTS AFTER COMPLETE LOBECTOMIES AND PNEUMONECTOMIES PNEUMONECTOMIES Oleg Kshivets, MD, PhD Oleg Kshivets, MD, PhD Surgery Department,Siauliai Cancer Center, Surgery Department,Siauliai Cancer Center, Lithuania Lithuania The Society of Cardiothoracic Surgeons of Great Britain The Society of Cardiothoracic Surgeons of Great Britain and Ireland Annual and Ireland Annual Scientific Scientific Meeting Meeting , , London London , the UK, , the UK, March March 5-8, 2005. 5-8, 2005.
NEURAL NETWORKS AND BOOTSTRAP SIMULATION IN PREDICTION OF OUTCOME OF NON-SMALL CELL LUNG CANCER PATIENTS AFTER COMPLETE LOBECTOMIES AND PNEUMONECTOMIES
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
NEURAL NETWORKS AND BOOTSTRAP SIMULATION IN NEURAL NETWORKS AND BOOTSTRAP SIMULATION IN PREDICTION OF OUTCOME OF NON-SMALL CELL LUNG PREDICTION OF OUTCOME OF NON-SMALL CELL LUNG
CANCER PATIENTS AFTER COMPLETE LOBECTOMIES AND CANCER PATIENTS AFTER COMPLETE LOBECTOMIES AND PNEUMONECTOMIES PNEUMONECTOMIES
Oleg Kshivets, MD, PhDOleg Kshivets, MD, PhD
Surgery Department,Siauliai Cancer Center, LithuaniaSurgery Department,Siauliai Cancer Center, LithuaniaThe Society of Cardiothoracic Surgeons of Great Britain and Ireland AnnualThe Society of Cardiothoracic Surgeons of Great Britain and Ireland Annual
ScientificScientific Meeting Meeting,, London London, the UK, March, the UK, March 5-8, 2005. 5-8, 2005.
AbstractAbstractNEURAL NETWORKS AND BOOTSTRAP SIMULATION IN PREDICTION OF OUTCOME OFNON-SMALL CELL LUNG CANCER PATIENTS AFTER COMPLETE LOBECTOMIES AND PNEUMONECTOMIES Oleg Kshivets Surgery Department, Siauliai Cancer Center, LithuaniaOBJECTIVE: The potential prognostic clinicomorphological factors for outcome of non-small lung cancer (LC) patients (LCP) after surgery were investigated.METHODS: In trial (1985-2004) the data of consecutive 511 LCP after complete resections R0 (age=57.1±0.4 years; male=460, female=51; tumor diameter: D=4.6±0.1 cm; pneumonectomy=212, upper lobectomy=173, lower lobectomy=93, middle lobectomy=7, bilobectomy=26, combined procedures with resection of pericardium, left atrium, aorta, v. cava superior, carina, diaphragm, ribs=143; only surgery-S=310, adjuvant chemoimmunoradiotherapy-AT=99: CAV/gemzar + cisplatin + thymalin/taktivin + radiotherapy 45-50Gy, postoperative radiotherapy 45-50Gy-RT=102) with stage II-III (squamous cell=329, adenocarcinoma=144, large cell=38; stage II=171, stage III=340; T1=143, T2=225, T3=112, T4=31; N0=297, N1=116, N2=98; G1=122, G2=144, G3=245) was reviewed. Variables selected for 5YS study were input levels of blood, biochemic and hemostatic factors, sex, age, TNMG, D. Survival curves were estimated by Kaplan-Meier method. Differences in curves between groups were evaluated using a log-rank test. Neural networks computing, Cox regression, clustering, discriminant analysis, structural equation modeling, Monte Carlo and bootstrap simulation were used to determine any significant regularity. RESULTS: For total of 511 LCP overall life span (LS) was 57.7±1.9 months and 5-year (5Y) survival (5YS) reached 57.9%. 296 LCP (age=56.1±0.5 years; LS=86.1±2.0 months; D=4.3±0.1 cm) lived more than 5Y without LC progressing. 185 LCP (age=57.2±0.6 years; LS=18.7±0.9 months; D=5.0±0.2 cm) died because of LC during first 5Y after surgery. . Cox modeling displayed that 5YS of LCP significantly depended on: N0-2 (P=0.000), AT (P=0.000), histology (P=0.001), T1-4 (P=0.024), age (P=0.006), weight (P=0.000), 16 blood factors (P=0.000-0.041). Neural networks computing, genetic algorithm selection and bootstrap simulation revealed relationships between 5YS of LCP and N0-2 (rank=1), LC growth (2), S (3), T1-4 (4), procedure type (5), G1-3 (6), histology (7), RT (8), AT (9), ESS (10), blood protein (11), prothrombin index (12), gender (13), percent of segmented neutrophils (14), D (15), percent of lymphocytes (16), ratio of monocytes/LC cells (LCC) (17), thrombocytes/LCC (18), eosinophils/LCC (19), healthy cells/LCC (20), leucocytes/LCC (21), blood glucose (22), lymphocytes/LCC (23), blood bilirubin (24). Correct prediction of LCP survival after surgery was 76.6% by logistic regression, 81.3% by discriminant analysis and 99.8% by neural networks computing (error=0.0456; urea under ROC curve=0.996).
Factors:1) Antropometric Factors…………...42) Blood Analysis…………………...263) Hemostasis Factors……………….84) Cell Ratio Factors………………...9 5) Lung Cancer Characteristics…….86) Biochemic Factors………………...57) Treatment Characteristics……….58) Survival Data……………………...3 In All……………………………….68
Main Problem of Analysis of Alive Supersystems Main Problem of Analysis of Alive Supersystems (e.g. Lung Cancer Patient Homeostasis):(e.g. Lung Cancer Patient Homeostasis):
Phenomenon of «Combinatorial Explosion» Phenomenon of «Combinatorial Explosion»
Number of Clinicomorphological Factors:……...….. 68Number of Possible Combination for Random Search:……………..………………….. n!=68!=2.48e+96 Operation Time of IBM Blue Gene/L Supercomputer (70.72TFLOPS) ………………………….1.11e+75 YearsThe Age of Our Universe……….....1.3e+10Years
Basis:Basis:
NP RP P n! n*n*2(e+n) or n log n n AI CSA+S+B SM
Antropometric Factors:
Male………….…………..460Female………..……………51Age……..…….57.1±0.4 yearsWeight………...……70±05 kgHeight…………168.5 ±0.3 cm
Survival Rate of Lung Cancer Patients after Survival Rate of Lung Cancer Patients after Lobectomies and Pneumonectomies (R0) (n=511):Lobectomies and Pneumonectomies (R0) (n=511):Surgery alone………………………………..310 (60.7%)P/o Radiotherapy…………………………....102 (20%)Adjuvant Chemoimmunoradiotherapy……..99 (19.3%) Alive………………………………………….304 (59.5%)5-Year Survivors…………………………….296 (57.9%) Losses from Lung Cancer…………………..185 (36.2%)Life Span………………………………..57.7±1.9 months 5-Year Survivors after Surgery alone……...194 (62.6%)5-Year Survivors after P/o Radiotherapy.…..48 (47.1%)5-Year Survivors after Adjuvant Chemoimmunoradiotherapy…………………54 (54.5%)
Adjuvant Therapy after Lobectomies and Pneumonectomies:Adjuvant Therapy after Lobectomies and Pneumonectomies:Adjuvant Chemoimmunoradiotherapy (n=99): 1 cycle of bolus chemotherapy (CAVT) was initiated 10-14 days after resections and consisted of Cyclophosphamid 500 mg/m2 IV on day 1, Doxorubicin 50 mg/m2 IV on day 1, Vincristin 1.4 mg/m2 IV on day 1. Immunotherapy consisted Thymalin or Taktivin 20 mg IM on days 1, 2, 3, 4 and 5. Chest radiotherapy (45-50 Gy) was administered since 7 day after 1 cycle chemoimmunotherapy at a daily dose of 1.8-2 Gy. No prophylactic cranial irradiation was used. From 2 to 3 weeks after completion of radiotherapy 3-4 courses of CAVT were repeated every 21-28 day. Chemotherapy by gemzar 1250 mg/m2 IV on day 1, 8, 15 and cisplatin 75 mg/m2 on day 1 was initiated on 14 day after surgery and was repeated every 14 day (5-6 courses). P/o Radiotherapy (n=102): Radiotherapy (60CO; ROKUS, Russia) with a total tumor dose 45-50 Gy (2-4 weeks after surgery) consisted of single daily fractions of 180-200 cGy 5 days weekly. The treatment volume included the ipsilateral hilus, the supraclavicular fossa and the mediastinum from the incisura jugularis to 5-7 cm below the carina. The lower mediastinum was included in cases of primary tumors in the lower lobes. The resected tumor bed was included in all patients. Parallel-opposed AP-PA fields were used. All fields were checked using the treatment planning program COSPO. Doses were specified at middepth for parallel-opposed technique or at the intersection of central axes for oblique technique. No prophylactic cranial irradiation was used.
Significant Factors between Lung Cancer Losses & 5-Year Survivors (n=481)
Significant Factors between Lung Cancer Losses & 5-Year Survivors (n=481)
Factors Log-Rank Test P
O(I) vs. A(II) 0.03687G1 vs. G3 0.00061T1 vs. T2 0.00460T1 vs. T3 0.01848T1 vs. T4 0.00041T2 vs. T4 0.02097T3 vs. T4 0.03976N0 vs. N1 0.00000 N0 vs. N2 0.00000 N1 vs. N2 0.00001Stage II vs. Stage III 0.00000Surgery alone vs. P/o Radiotherapy 0.00046Ad.Chemioimmunoradiotherapy vs. P/o Radiotherapy 0.00025
Product-Limit (Kaplan-Maier) Analysis Results in Prediction of Lung Cancer Patients Survival after Lobectomies and Pneumonectomies (n=511)
Graph of Survival Times vs. Cum. Proportion Surviving
Survival FunctionComplete Censored
Survival Rate of Lung Cancer Patients after Lobectomies and Pneumonectomiesn=511
Survival Time (Years after Lobectomies and Pneumonectomies)
Results of Multifactor Analysis in Prediction of Lung Cancer Results of Multifactor Analysis in Prediction of Lung Cancer
Patients Survival after Lobectomies and Pneumonectomies (n=511)Patients Survival after Lobectomies and Pneumonectomies (n=511) Factor Loadings, Factor 1 vs. Factor 2
Rotation: Varimax normalizedExtraction: Principal components
Prediction of Lung Cancer Patients after Lobectomies and Pneumonectomies, n=511
Factor 1
Fact
or 2
SEX
AGE
MASSA
GROUTHHIST
G
T
N12_0
D
ERHB
CP
THR L
E
P
S
LYM M EABS
PABSSABS
LYMABS MABS
SOE
COAG_B
T_HEMGLU
PI
BILIR TIMSULPROT T_RECTHR_T
FIBR_BFIBR
TOL_HEP
OPER
GAMMACHT_IT_R
SURGERY
ER_CC
THR__CCL_CC
E_CC
P_CC
S_CCLYM_CC
M_CC
MASS_CC
ER_TOT
THR_TOT L_TOT
E_TOT
P_TOTS_TOT
LYM_TOTM_TOTSU5 LS
-1.0
-0.6
-0.2
0.2
0.6
1.0
1.4
-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
Results of Discriminant Analysis in Prediction of Lung Cancer Patients Survival after Lobectomies and Pneumonectomies (n=481)
Results of Multifactor Clustering of Clinicomorphologic Factors in Prediction of Results of Multifactor Clustering of Clinicomorphologic Factors in Prediction of Lung Cancer Patients Survival after Lobectomies and Pneumonectomies (n=481)Lung Cancer Patients Survival after Lobectomies and Pneumonectomies (n=481)
Results of Results of CorrespondenceCorrespondence Analysis in Prediction of Analysis in Prediction of LungLung Cancer Cancer Patients Survival after Lobectomies and Pneumonectomies (n=Patients Survival after Lobectomies and Pneumonectomies (n=481481))
Results of Results of CorrespondenceCorrespondence Analysis in Prediction of Analysis in Prediction of LungLung Cancer Cancer Patients Survival after Lobectomies and Pneumonectomies (n=Patients Survival after Lobectomies and Pneumonectomies (n=481481))
Results of Results of CorrespondenceCorrespondence Analysis in Prediction of Analysis in Prediction of LungLung Cancer Cancer Patients Survival after Lobectomies and Pneumonectomies (n=Patients Survival after Lobectomies and Pneumonectomies (n=481481))
Results of Results of CorrespondenceCorrespondence Analysis in Prediction of Analysis in Prediction of LungLung Cancer Cancer Patients Survival after Lobectomies and Pneumonectomies (n=Patients Survival after Lobectomies and Pneumonectomies (n=481481))
Classification Tree in Prediction of Lung Cancer Patients Survival after Lobectomies and Pneumonectomies (n=481)
Rankings on scale from 0=low importance to 100=high importancePrediction of Lung Cancer Patients Survival after Lobectomies & Pneumonectomies
Predictor variable
Ran
king
0
20
40
60
80
100
SEX
GR
OU
THH
IST G T
N12
_0TR
EAT
AG
EM
ASS
A D ER HB
THR L E P S
LYM M
SOE
CO
AG
_BT_
HEM
GLU PI
BILI
RPR
OT
T_R
ECTH
R_T
FIBR
_BFI
BRTO
L_H
EPER
_CC
THR
__C
CL_
CC
E_C
CP_
CC
S_C
CLY
M_C
CM
_CC
MA
SS_C
CER
_TO
TTH
R_T
OT
L_TO
TE_
TOT
P_TO
TS_
TOT
LYM
_TO
TM
_TO
T
Classification Tree in Prediction of Lung Cancer Patients Survival after Lobectomies and Pneumonectomies (n=481)
Classification Tree for Lung Cancer Patients SurvivalPrediction of Lung Cancer Patients Survival after Lobectomies and Pneumonectomie
Number of splits = 11; Number of terminal nodes = 12
1
2 3
4 5
6 7 8 9
10 11 12 13 14 15
16 17 18 19
20 21
22 23
N12_0=N2
PI107.64
PI94.369 AGE 58.021
FIBR_B 1.8848 N12_0=N1 LYM 14.523
N12_0=N1,N2 TOL_HEP 244.95
MASSA 67.48
G=G3
92 389
346 43
178 168 23 20
150 28 51 117 5 15
8 20 36 15
16 20
9 11
l>5
d<5 l>5
l>5 l>5
l>5 l>5 l>5 d<5
l>5 l>5 d<5 l>5 l>5 d<5
d<5 l>5 d<5 d<5
d<5 l>5
d<5 l>5
d<5l>5
Neural Networks in Prediction of Lung Cancer Patients Survival after Neural Networks in Prediction of Lung Cancer Patients Survival after Lobectomies and Pneumonectomies (n=481)Lobectomies and Pneumonectomies (n=481)
Losses 5-year survivors Baseline Errors=0.0456;Total 185 296 Area under ROC curve=0.996;
Ratio of Erythrocytes, Leucocytes, Eosinophils, Healthy Cells and Cancer Cell Ratio of Erythrocytes, Leucocytes, Eosinophils, Healthy Cells and Cancer Cell Populations Populations & & Blood Glucose Level Blood Glucose Level in in Prediction Prediction of Lung Cancer of Lung Cancer Patients Patients
Survival Survival (n= (n=481481))
Ratio of Erythrocytes, Leucocytes, Eosinophils, Healthy Cells and Cancer Cell Ratio of Erythrocytes, Leucocytes, Eosinophils, Healthy Cells and Cancer Cell Populations Populations & & Blood Glucose Level Blood Glucose Level in in Prediction Prediction of Lung Cancer of Lung Cancer Patients Patients
Survival Survival (n= (n=481481))
Ratio of Erythrocytes, Leucocytes, Eosinophils, Healthy Cells and Cancer Cell Ratio of Erythrocytes, Leucocytes, Eosinophils, Healthy Cells and Cancer Cell Populations Populations & & Blood Glucose Level Blood Glucose Level in in Prediction Prediction of Lung Cancer of Lung Cancer Patients Patients
Survival Survival (n= (n=481481))
Ratio of Erythrocytes, Leucocytes, Eosinophils, Healthy Cells and Cancer Cell Ratio of Erythrocytes, Leucocytes, Eosinophils, Healthy Cells and Cancer Cell Populations Populations & & Blood Glucose Level Blood Glucose Level in in Prediction Prediction of Lung Cancer of Lung Cancer Patients Patients
Survival Survival (n= (n=481481))
Networks between Clinicopathologic, Biochemic, Hemostasis & Hematologic Data and 5-Networks between Clinicopathologic, Biochemic, Hemostasis & Hematologic Data and 5-Year Survival of Lung cancer Patients after Lobectomies and Pneumonectomies (n=481)Year Survival of Lung cancer Patients after Lobectomies and Pneumonectomies (n=481)
Results of Monte Carlo Simulation in Prediction of Lung Cancer Results of Monte Carlo Simulation in Prediction of Lung Cancer Patients Survival after Lobectomies and Pneumonectomies (n=481)Patients Survival after Lobectomies and Pneumonectomies (n=481)
From: Monte Carlo DataPrediction of Lung Cancer Patients Survival after Lobectomies & Pneumonectomies
Chi2=64.411; df=35; P=0.00178; n=481
Results of Monte Carlo Simulation in Prediction of Lung Cancer Results of Monte Carlo Simulation in Prediction of Lung Cancer Patients Survival after Lobectomies and Pneumonectomies (n=481)Patients Survival after Lobectomies and Pneumonectomies (n=481)
From: Monte Carlo DataPrediction of Lung Cancer Patients Survival after Lobectomies & Pneumonectomies
Chi2=28.09; df=14; P=0.0139; n=481
Results of Monte Carlo Simulation in Prediction of Lung Cancer Results of Monte Carlo Simulation in Prediction of Lung Cancer Patients Survival after Lobectomies and Pneumonectomies (n=481)Patients Survival after Lobectomies and Pneumonectomies (n=481)
From: Monte Carlo DataPrediction of Lung Cancer Patients Survival after Lobectomies and Pneumonectomie
Chi2=41.0; df=27; P=0.0412; n=481
SEPATH Networks in Prediction of Lung Cancer Patients SEPATH Networks in Prediction of Lung Cancer Patients Survival after Lobectomies and Pneumonectomies (n=481)Survival after Lobectomies and Pneumonectomies (n=481)
Holling-Tenner Models of Alive Supersystem “Lung Cancer-Holling-Tenner Models of Alive Supersystem “Lung Cancer-Cytotoxic Cell PopulationCytotoxic Cell Population”
Holling-Tenner Models of Alive Supersystem “Lung Cancer-Holling-Tenner Models of Alive Supersystem “Lung Cancer-Cytotoxic Cell PopulationCytotoxic Cell Population”
Holling-Tenner Models of Alive Supersystem “Lung Cancer-Holling-Tenner Models of Alive Supersystem “Lung Cancer-Cytotoxic Cell PopulationCytotoxic Cell Population”
Holling-Tenner Models of Alive Supersystem “Lung Cancer-Holling-Tenner Models of Alive Supersystem “Lung Cancer-Cytotoxic Cell PopulationCytotoxic Cell Population”
0 100 200 300 400
0.1
10
100
Cytotoxic CellsLung Cancer Cells
Model "Lung Cancer---Cytotoxic Cells"
Time
Lung
Cel
l Pop
ulat
ion
Dyn
amic
s11
0.185
X1 2
X1 3
3000 X1 1
Lung Cancer Dynamics
SUPERONCOPROGNOSIS-1.0
PROGNOSIS SURVIVAL-2
PROG-1 PROG-2 PROG-3 E
SURVIVAL LESS 5 YEARS SURVIVAL MORE 5 YEARS
SURVIVAL-1
A B
C
Conclusions:Conclusions:It was revealed that 5-year survival and life span of lung cancer patients after complete lobectomies and pneumonectomies significantly depended on: 1) lung cancer characteristics;2) level of blood cell subpopulations circuit; 3) cell ratio factors (ratio of total lung cancer cell population to blood cell subpopulations; 4) hemostasis system; 5) biochemic homeostasis; 6) adjuvant treatment.
Patents:1. Kshivets O.M. Method of Prognosis of
Survival Rate of Radically Operated Patients with Malignant Neoplasms. Patent from 10.02.94; N2101704: 24pp.
2. Kshivets O.M. Method of Prognosis of Survival Rate of Non-Radically Operated Patients with Malignant Neoplasms. Patent from 14.03.94; N2104536: 10pp.