Handbook of Statistics in Clinical Oncology, Second Edition

dk3170_chfm.pdf......Page 1
Handbook of Statistics in Clinical Oncology SECOND EDITION......Page 2
Dedication......Page 4
Preface......Page 5
Editors......Page 6
Contributors......Page 7
Contents......Page 13
COLOR FIGURE 25.2......Page 17
Table of Contents......Page 0
COLOR FIGURE 27.4......Page 18
Handbook of Statistics in Clinical Oncology, Second Edition......Page 19
CONTENTS......Page 20
1.1 INTRODUCTION......Page 21
1.2.1 CLINICAL ISSUES AND STATISTICAL TASKS......Page 22
1.2.3 DEFINITIONS......Page 23
1.2.4 MAXIMUM TOLERATED DOSE......Page 24
1.2.5 DOSE-TOXICITY MODELING......Page 26
1.3.1 CHOICE OF THE DOSE LEVELS: ACTION SPACE......Page 27
1.3.1.1 Modified Fibonacci Dose Escalation......Page 28
1.3.2 DOSE ESCALATION SCHEMES......Page 30
1.3.2.1 Traditional Escalation Rule......Page 31
1.3.2.2 Random Walk (RW) Designs......Page 33
1.3.2.3 Bayesian Designs......Page 34
1.3.2.4 Continual Reassessment Method......Page 35
1.3.2.5 Modifications of the CRM......Page 36
1.3.4 VALIDATION AND COMPARISON OF PHASE I DESIGNS......Page 37
1.4.3 MTD ESTIMATION......Page 38
1.4.4 PHARMACOKINETIC PHASE I DATA ANALYSIS......Page 39
1.4.6 REGULATIONS BY GOOD CLINICAL PRACTICE AND INTERNATIONAL CONFERENCE ON HARMONIZATION (ICH)......Page 40
1.5 OPEN PROBLEMS......Page 41
ACKNOWLEDGMENTS......Page 42
REFERENCES......Page 43
CONTENTS......Page 47
2.1.1 MOTIVATION......Page 48
2.1.2 OPERATING CHARACTERISTICS......Page 51
2.2 TECHNICAL ASPECTS......Page 52
2.2.1 MODEL REQUIREMENTS......Page 53
2.3 IMPLEMENTATION......Page 54
2.3.1 MAXIMUM LIKELIHOOD IMPLEMENTATION......Page 55
2.3.2 TWO-STAGE DESIGNS......Page 56
2.3.4 ILLUSTRATION......Page 58
2.4.1 CONVERGENCE......Page 59
2.4.2 EFFICIENCY......Page 60
2.4.3 SAFETY......Page 62
2.5.1 INCLUSION OF A STOPPING RULE......Page 63
2.5.2 PATIENT HETEROGENEITY......Page 64
2.5.3 PHARMACOKINETIC STUDIES......Page 66
2.5.4 GRADED TOXICITIES......Page 67
2.6 COMBINED TOXICITY-EFFICACY STUDIES......Page 69
2.7 BAYESIAN APPROACHES......Page 71
REFERENCES......Page 73
3.1 INTRODUCTION AND BACKGROUND......Page 75
3.2.2 TRADITIONAL ESCALATION ALGORITHMS......Page 77
3.2.3 A BAYESIAN APPROACH: THE CONTINUAL REASSESSMENT METHOD......Page 79
3.2.4 STORER’S TWO-STAGE DESIGN......Page 80
3.2.5 OTHER APPROACHES......Page 82
3.3.1 SPECIFYING THE DOSE–RESPONSE CURVE......Page 83
3.3.2.4 Non-Bayesian CRM Design......Page 84
3.3.3 SIMULATION AND RESULTS......Page 85
3.4 SUMMARY AND CONCLUSION......Page 88
REFERENCES......Page 90
4.1 INTRODUCTION......Page 92
4.2 PHARMACOKINETICS AND PHARMACODYNAMICS OF ANTICANCER AGENTS......Page 94
4.2.1 EXAMPLE......Page 97
4.2.2 NONPARAMETRIC MODEL......Page 100
4.2.5 DOSE INDIVIDUALIZATION......Page 102
4.2.6 DOSE INDIVIDUALIZATION EXAMPLE......Page 103
4.2.8 DESIGN EXAMPLE......Page 105
4.3 SUMMARY......Page 106
REFERENCES......Page 107
5.1 INTRODUCTION......Page 111
5.2 EXAMPLE OF A PHASE I TRIAL......Page 112
5.3 ELICITATION OF THE DOSE-RESPONSE MODEL AND PRIOR DISTRIBUTION......Page 113
5.4 DOSE ALLOCATION......Page 116
5.5 SAMPLE SIZE AND THE STOPPING RULE......Page 119
5.6 CONCLUSION......Page 122
REFERENCES......Page 123
APPENDIX 5.1 SAS CODES FOR LOGISTIC DOSE-RESPONSE MODEL......Page 124
APPENDIX 5.2 SAS CODES FOR DENSITY FUNCTION OF MTD......Page 125
APPENDIX 5.3 SAS CODES FOR CRM CALCULATION AND DENSITY FUNCTION OF DLT OCCURRENCE RATE......Page 127
DK3170_ch06.pdf......Page 131
6.1 DESIGN......Page 132
6.2 ANALYSIS OF STANDARD PHASE II DESIGNS......Page 134
6.3.2 PHASE II DESIGNS WITH MULTIPLE ENDPOINTS......Page 138
6.3.3 BAYESIAN PHASE II DESIGNS......Page 140
REFERENCES......Page 141
7.1 INTRODUCTION......Page 143
7.2 DESIGNS FOR RESPONSE AND TOXICITY......Page 144
7.3 DESIGNS THAT ALLOW A TRADE-OFF BETWEEN RESPONSE AND TOXICITY......Page 149
7.4 SUMMARY......Page 153
REFERENCES......Page 154
8.1 INTRODUCTION......Page 155
8.2 RESPONSE IS NOT STRONGLY CORRELATED WITH SURVIVAL......Page 156
8.4 PHASE II DESIGNS USING TIME TO EVENT ENDPOINTS......Page 158
8.4.1.1 Comparisons with Historical Experience......Page 159
8.4.1.2 Each Patient as His Own Control......Page 160
8.4.2.3 Randomization Discontinuation Design......Page 162
REFERENCES......Page 165
9.1 BASIC CONCEPT......Page 167
9.2.1 BINARY OUTCOMES......Page 168
9.3.1 DESIGNS WITH MINIMUM ACTIVITY REQUIREMENTS......Page 169
9.3.1.1 Binary Outcomes......Page 171
9.3.1.2 Survival Outcomes......Page 173
9.3.3 DESIGNS FOR ORDERED TREATMENTS......Page 174
REFERENCES......Page 175
Handbook of Statistics in Clinical Oncology, SECOND EDITION......Page 177
10.1 INTRODUCTION......Page 178
10.2.1 PERMUTED BLOCKS RANDOMIZATION......Page 179
10.2.2 POCOCK–SIMON, OR MINIMIZATION, APPROACH......Page 180
10.3 BASIS FOR INFERENCE......Page 181
10.4 ANALYTIC APPROACHES......Page 182
10.5 DESIGN OF SIMULATION STUDIES......Page 183
10.6 RESULTS......Page 184
ACKNOWLEDGMENT......Page 188
REFERENCES......Page 189
11.1 INTRODUCTION......Page 191
11.1.1 APPROACH 1......Page 192
11.2.2 APPROACH 2......Page 193
11.2 SIMULATION STUDY......Page 194
11.3 EXAMPLE: SOUTHWEST ONCOLOGY GROUP STUDY 8300......Page 196
11.4 OTHER APPROACHES TO MULTIARM STUDIES......Page 197
11.5 CONCLUSION......Page 198
REFERENCES......Page 199
12.1 INTRODUCTION......Page 200
12.2 HYPOTHESIS TESTS AND CONFIDENCE INTERVALS......Page 201
12.3.1 NONINFERIORITY TRIAL WITH BINARY OUTCOME......Page 202
12.3.2 NONINFERIORITY TRIAL WITH TIME-TO-EVENT OUTCOME......Page 204
12.4 EXAMPLE......Page 206
12.5 DETERMINING THE NONINFERIORITY MARGIN......Page 207
12.6 BENEFIT OF ECOMPARED TO NO TREATMENT......Page 208
12.7 EXAMPLE (CONTINUED)......Page 211
12.9 DISCUSSION......Page 212
REFERENCES......Page 213
CONTENTS......Page 215
13.1 CLINICAL SUMMARY......Page 216
13.2 INTRODUCTION......Page 217
13.3.1 BINOMIAL EXAMPLE (ONE-SIDED TEST)......Page 218
13.4 EXPONENTIAL SURVIVAL......Page 219
13.5 APPLICATIONS TO THE LOGRANK TEST AND COX REGRESSION......Page 220
13.6.1.2 Output Parameters......Page 221
13.6.2 KEY OBSERVATIONS......Page 222
13.7 SEQUENTIAL MONITORING BY THE O’BRIEN–FLEMING METHOD......Page 223
13.8 EVALUATION OF THE INFORMATION FRACTION......Page 224
13.9 MULTIPLE TREATMENTS AND STRATIFICATION......Page 225
13.11 PRACTICAL CONCLUSIONS......Page 226
REFERENCES......Page 227
APPENDIX I: LOGRANK VS. EXPONENTIAL ESTIMATION–LOCAL OPTIMALITY......Page 228
APPENDIX II: SAS MACRO FOR SAMPLE SIZE AND ACCRUAL DURATION......Page 229
APPENDIX III: SAS MACRO FOR LOGRANK AND STRATIFIED LOGRANK TESTS......Page 230
14.1 INTRODUCTION......Page 235
14.2.1 GENERAL PRINCIPLES......Page 236
14.2.2 SYMMETRIC TWO-SIDED GROUP SEQUENTIAL STOPPING BOUNDARIES......Page 237
14.2.3 ASYMMETRIC STOPPING BOUNDARIES......Page 238
14.3.1 STOCHASTIC CURTAILMENT......Page 240
14.3.2 PREDICTIVE POWER AND CURRENT DATA METHODS......Page 241
14.4 A BAYESIAN APPROACH TO ASSESSING EARLY TERMINATION FOR NO BENEFIT......Page 242
14.5.1 A TRIAL STOPPED EARLY FOR NO BENEFIT......Page 243
14.5.2 THE EFFECT OF TWO EASILY APPLIED RULES IN THE ADJUVANT AND ADVANCED DISEASE SETTING......Page 246
14.5.2.1 Scenario 1......Page 248
14.6 SUMMARY......Page 249
REFERENCES......Page 252
CONTENTS......Page 255
15.1.1 WHAT IS QOL AND HOW IS IT MEASURED?......Page 256
15.1.2 CHAPTER ORGANIZATION......Page 257
15.2.2 CLINICAL SIGNIFICANCE AND POWER......Page 258
15.3 TYPES OF MISSING DATA......Page 261
15.3.1 GRAPHICAL DESCRIPTION OF MISSING DATA MECHANISMS......Page 262
15.4 LONGITUDINAL ANALYSIS OPTIONS FOR “COMPLETE” DATA SETS......Page 264
15.4.2 GENERALIZED LINEAR MODELS......Page 265
15.5.1 JOINT MODELING OF MEASUREMENT AND MISSINGNESS PROCESSES......Page 266
15.5.2 WEIGHTED GEES......Page 268
15.6.1 SUBSTITUTION METHODS......Page 269
15.6.3 ADJUSTED SURVIVAL ANALYSES......Page 270
REFERENCES......Page 271
16.1 INTRODUCTION AND RATIONALE......Page 278
16.2.2 STATEMENT OF STUDY OBJECTIVES AND HYPOTHESES......Page 280
16.2.3 SAMPLE SIZE AND POWER ESTIMATION FOR JOINT CLINICAL/ ECONOMIC TRIALS......Page 281
16.3.1 SELECTING OUTCOMES ENDPOINTS......Page 282
16.3.3 MEASURING COSTS......Page 283
16.3.3.2 Protocol-Induced Costs......Page 284
16.4.1 ANALYSIS OF COSTS......Page 285
16.4.3 MISSING DATA......Page 286
16.4.4 ADJUSTMENT TO CONSTANT DOLLARS AND DISCOUNTING OF FUTURE COSTS AND OUTCOMES......Page 287
16.4.5 SUMMARY MEASURES......Page 288
16.4.6 COST-EFFECTIVENESS ACCEPTABILITY CURVES......Page 289
16.4.7 NET HEALTH BENEFIT APPROACH......Page 290
REFERENCES......Page 291
Handbook of Statistics in Clinical Oncology, SECOND EDITION......Page 294
17.1 INTRODUCTION......Page 295
17.2 DESIGN OF PROGNOSTIC FACTOR STUDIES......Page 297
17.3.1 FREIBURG DNA STUDY......Page 298
17.3.2 GBSG-2 STUDY......Page 299
17.4 THE CUTPOINT MODEL......Page 301
17.5.1 ASSUMING A LOG-LINEAR RELATIONSHIP FOR CONTINUOUS COVARIATES......Page 305
17.5.2 CATEGORIZING CONTINUOUS COVARIATES......Page 306
17.5.3 DETERMINING FUNCTIONAL RELATIONSHIPS FOR CONTINUOUS COVARIATES......Page 308
17.5.4 FURTHER ISSUES......Page 311
17.6 CLASSIFICATION AND REGRESSION TREES......Page 312
17.7 FORMATION AND VALIDATION OF RISK GROUPS......Page 315
17.8 ARTIFICIAL NEURAL NETWORKS......Page 320
17.9 ASSESSMENT OF PROGNOSTIC CLASSIFICATION SCHEMES......Page 325
17.10 SAMPLE SIZE CONSIDERATIONS......Page 328
17.11 CONCLUDING REMARKS......Page 331
REFERENCES......Page 332
18.1 INTRODUCTION......Page 340
18.2.2 FRACTIONAL POLYNOMIALS......Page 342
18.2.3.1 Our Proposal of the Approach......Page 343
18.3.1 DESCRIPTION OF THE DATA......Page 344
18.3.2 CONSTRUCTION OF THE PROGNOSTIC MODEL......Page 345
18.3.3 BUMP HUNTING (IDENTIFICATION OF PREDICTIVE FACTORS)......Page 347
18.4 FINAL REMARKS......Page 348
REFERENCES......Page 349
19.1.1 MOTIVATION......Page 351
19.1.2 A BRIEF OVERVIEW......Page 352
19.2.1 MODEL AND NOTATION......Page 354
19.2.2.1 The Measure R2......Page 355
19.2.2.2 Population Parameter 2......Page 356
19.3.1 PROPERTIES OF R2 AND 2......Page 358
19.3.2 INTERPRETATION......Page 359
19.4.2 STRATIFIED MODEL......Page 360
19.4.3 OTHER RELATIVE RISK MODELS......Page 361
19.5.2 EXAMPLE 2......Page 362
19.5.3 EXAMPLE 3......Page 363
19.5.4 EXAMPLE 4......Page 364
REFERENCES......Page 366
CONTENTS......Page 368
20.1 INTRODUCTION......Page 369
20.2.1 FUNCTION APPROXIMATION......Page 370
20.3.1 CONSTRUCTING A TREE RULE......Page 371
20.3.2 SPLITTING......Page 374
20.3.3.1 Methods Based on Within-Node Deviance......Page 375
20.3.3.2 Methods Based on Between-Node Separation......Page 376
20.4 DATA PEELING METHODS......Page 377
20.4.1 REGIONS OF INTEREST......Page 378
20.4.2 REGION REFINEMENT......Page 379
20.4.3 VARIABLE SELECTION AND PEELING FRACTION......Page 380
20.4.4 SELECTING RULES BY CROSS-VALIDATION......Page 381
20.5 DISCUSSION......Page 383
REFERENCES......Page 384
21.1 INTRODUCTION......Page 386
21.3 JOINT MODEL......Page 387
21.4 APPLICATION......Page 390
21.5 FINAL REMARKS......Page 393
PRIOR DISTRIBUTION FOR PSA MODEL......Page 394
MCMC SAMPLER FOR PSA MODEL......Page 395
REFERENCES......Page 396
Handbook of Statistics in Clinical Oncology, SECOND EDITION......Page 398
22.1 INTRODUCTION......Page 399
22.1.2 NORMALIZATION......Page 400
22.1.4 MULTIPLE TESTING......Page 401
22.2 DATA PREPROCESSING OR NORMALIZATION......Page 402
22.2.1 BACKGROUND INTENSITY SUBTRACTION......Page 403
22.2.2 LOCATION ADJUSTMENT......Page 405
22.2.3 WITHIN- AND INTERARRAY SCALE ADJUSTMENT......Page 407
22.3.1 ANOVA FOR RATIOS......Page 411
22.3.2 ANOVA FOR INTENSITIES......Page 412
22.3.3 COMPARISON OF ANOVA METHODS......Page 413
22.4 MULTIPLE COMPARISONS......Page 414
REFERENCES......Page 418
23.1 INTRODUCTION......Page 421
23.2 GENE EXPRESSION PROFILING FOR BLADDER CANCER......Page 422
23.3.1 CLUSTERING......Page 423
23.3.3 ILLUSTRATION......Page 425
23.4.1 MULTIPLE TESTING......Page 426
23.4.3 ILLUSTRATION......Page 427
23.5.1 SELECTION OF GENES......Page 429
23.5.2 DEVELOPMENT OF PREDICTION MODELS......Page 430
23.5.4 ILLUSTRATION......Page 431
23.6 SOFTWARE NOTE......Page 433
REFERENCES......Page 434
CONTENTS......Page 439
24.2.1 MALDI-TOF MS TECHNOLOGY......Page 440
24.2.2 STUDY DESIGN CONSIDERATIONS......Page 442
24.3 PRE-ANALYSIS PROCESSING OF MALDI-TOF MASS SPECTRA......Page 443
24.3.3 IMPRECISION IN M/ ZMEASUREMENTS......Page 444
24.3.5 NORMALIZATION/STANDARDIZATION OF INTENSITY......Page 446
24.4 EVALUATING PEAKS INDIVIDUALLY AS POTENTIAL MARKERS......Page 448
24.5.1 COMBINATION OF SINGLE-PEAK CLASSIFICATION RULES......Page 449
24.5.2 COMBINING MULTIPLE PEAKS INTO A CLASSIFIER......Page 450
24.5.3 BUILDING A MULTI-PEAK SINGLE CLASSIFIER VIA REAL ADABOOST......Page 452
24.5.4 EXTENSION TO A SETTING WHERE CLASS LABELS MAY CONTAIN ERROR......Page 453
24.6 DISCUSSION......Page 455
REFERENCES......Page 456
25.1 INTRODUCTION......Page 459
25.2.1 INFORMATION-THEORETIC SCORE......Page 462
25.2.3 SIGNIFICANCE ANALYSIS OF MICROARRAYS......Page 463
25.2.5 FUNCTIONAL DATA ANALYSIS (FDA)......Page 464
25.2.6 INCONSISTENCY OF VARIABLE SELECTION......Page 465
25.3.1 COMPOUND COVARIATE METHOD (CCM)......Page 466
25.3.2 WEIGHTED FLEXIBLE COMPOUND COVARIATE METHOD......Page 467
25.3.3 LEAVE-ONE-OUT CROSS-VALIDATED CLASS PREDICTION MODEL......Page 468
25.4 FINAL REMARKS......Page 470
REFERENCES......Page 471
26.1 INTRODUCTION......Page 473
26.2 GENOTYPE-BASED AND HAPLOTYPE-BASED PENETRANCE MODELS......Page 474
26.3 LIKELIHOODS AND CASE-CONTROL ASCERTAINMENT......Page 476
26.4 MULTIPLE COMPARISONS, SPARSE DATA, AND BAYESIAN SMOOTHING OR CLUSTERING......Page 477
26.5 RELATIONSHIP TO COALESCENT METHODS......Page 479
26.6 HAPLOTYPE SHARING METHODS......Page 480
26.7 OTHER ISSUES......Page 481
REFERENCES......Page 484
27.1 INTRODUCTION......Page 490
27.2 GETTING STARTED......Page 492
27.2.1 DATA RESOURCES......Page 494
27.3.1 DNA MICROARRAYS......Page 495
27.3.1.1 Image Processing......Page 496
27.3.1.2 Normalization......Page 497
27.3.1.3 Quantification of Expression......Page 498
27.3.1.4 Differential Expression......Page 499
27.3.1.5 Software for Expression Analysis......Page 500
27.3.2.1 Protein Mass Spectrometry......Page 501
27.3.2.2 Array-CGH......Page 502
27.4 INTEGRATED DATABASES......Page 504
27.5 MACHINE LEARNING......Page 505
27.5.2 UNSUPERVISED MACHINE LEARNING......Page 506
27.5.4 DATA RE-USE......Page 509
27.6 SOFTWARE FOR GRAPHS AND NETWORKS......Page 510
27.8 SOFTWARE DEVELOPMENT......Page 512
27.8.1 DATA TECHNOLOGIES......Page 514
GLOSSARY......Page 515
REFERENCES......Page 517
Handbook of Statistics in Clinical Oncology, SECOND EDITION......Page 521
28.1 INTRODUCTION......Page 522
28.2.1 COMPARING LEUKEMIA TREATMENTS......Page 525
28.2.3 SENSITIVITY ANALYSES OF THE LEUKEMIA DATA......Page 526
28.3.1 DATA AND PROBLEM DEFINITION......Page 532
28.3.2 PROBABILITY MODEL......Page 534
28.3.3 SENSITIVITY ANALYSES OF THE TRANSPLANT DATA......Page 535
REFERENCES......Page 538
29.1 INTRODUCTION......Page 540
29.2 WHY DO MISSING DATA MATTER?......Page 541
29.3 WHAT IS THE QUESTION?......Page 542
29.5 ANALYSIS OF STUDIES WITH NONIGNORABLE DROPOUT......Page 543
29.5.2 ANALYSIS OF COMPLETE CASES......Page 545
29.5.4 MAXIMUM LIKELIHOOD ANALYSIS AND MULTIPLE IMPUTATION OF ALL AVAILABLE DATA......Page 546
29.5.5 SHARED-PARAMETER MODEL (JOINT MIXED-EFFECTS MODEL)......Page 548
29.5.6 PATTERN MIXTURE MODELS......Page 549
29.5.7 SELECTION MODELS......Page 550
29.6 SENSITIVITY ANALYSES......Page 551
REFERENCES......Page 553
30.1 INTRODUCTION......Page 555
30.2 ESTIMATION IN THE ABSENCE OF COMPETING RISKS: KAPLAN–MEIER ESTIMATE......Page 557
30.3 ESTIMATION IN THE PRESENCE OF COMPETING RISKS: CUMULATIVE INCIDENCE VS. KAPLAN–MEIER......Page 559
30.4 EXAMPLE FROM REAL DATA: SQUAMOUS CELL CARCINOMA......Page 561
REFERENCES......Page 562
CONTENTS......Page 564
31.2 STATISTICS......Page 565
31.3.2 WHAT IS THE CONTROL TREATMENT?......Page 567
31.3.3 ARE THERE PREDEFINED HYPOTHESES FOR ALL KEY END POINTS? ARE THE NUMBER OF STATISTICAL TESTS LIMITED, AND IF NOT, HAVE THE SIGNIFICANCE LEVELS BEEN ADJUSTED ACCORDINGLY?......Page 568
31.3.4 DO THE ELIGIBILITY CRITERIA COVER ALL THE PATIENTS WHO ARE LIKELY TO BE TREATED THIS WAY OUTSIDE OF THE TRIAL?......Page 569
31.3.6 ARE THE DETAILS OF THE INTERIM ANALYSES AND STOPPING RULES CLEARLY LAID OUT?......Page 570
31.3.7 ARE ALL RANDOMIZED PATIENTS INCLUDED AND ACCOUNTED FOR IN ALL ANALYSES?......Page 571
31.3.8 WHAT DO THE SURVIVAL COMPARISONS REALLY SHOW?......Page 572
31.3.9 HAVE THE HAZARD RATIO AND ESPECIALLY THE 95% CONFIDENCE INTERVALS BEEN GIVEN?......Page 573
REFERENCES......Page 574
32.1 INTRODUCTION......Page 577
32.2.1 PROBLEMS IN ANALYZING RESULTS IN PATIENTS WHO STOPPED TREATMENT BEFORE ITS INTENDED CONCLUSION......Page 578
32.2.2 CONFOUNDING......Page 579
32.3 ANALYSES COMBINING DATA FROM MULTIPLE STUDIES......Page 580
32.4 DESIGNING TRIALS TO TEST THE DOSE INTENSITY HYPOTHESIS......Page 581
32.4.2 DESIGN OF INDIVIDUAL TRIALS......Page 582
REFERENCES......Page 583
33.1 INTRODUCTION......Page 585
33.2.1.3 Comment......Page 587
33.2.2.3 Comments......Page 588
33.2.4.2 Results......Page 589
33.3 DISCUSSION AND RECOMMENDATIONS......Page 590
REFERENCES......Page 592