Академический Документы
Профессиональный Документы
Культура Документы
10
81 30
Dose
81 96
Concentration
V CL
71
97
68 62 45 58 10 51 79 75 3 65 2 84 24 7
21 51 59 67 9 3 47 60 2 52 10
24 68 84 22 43 20 65 72 34
72
13 8614
50
50
0 0 5 10 15
Time
3
Regression (population)
REAL WORLD MODEL WORLD Dose
V CL
+V ~ N (V , v 2 )
~ N (0, 2 )
CL ~ N (CL , CL 2 )
ESTIMATION
Covariates
Demographics Gender, age, size (weight, height, BSA, BMI), race Lab values Serum creatinine, bilirubin, albumin, pheno/genotype Disease parameters Baseline, etiology, disease duration, general status Therapy related Comedication, pretreatment, dialysis
6
Covariates (contd)
Habits / Environmental factors Smoking, alcohol, food, diet, time of day or year Study related Center, investigator, visit
Covariate types
Continuous Age Bivariate (dichotomous) Sex Ordered categorical Smoking (none, occasional, daily) Non-ordered categorical Race (caucasian, black, asian)
Skewed distributions
Non-linear relationships with covariate, or Linear relationships with log(covariate)
ALT distribution
3000 2500 2000 1500 1000 500 1600 1400 1200 1000 800
log(ALT) distribution
600
400
200
0 2 16 30 44 58 72 86 100 114 128 142 156 170 184 198 212 226 240 254 268 281 295 309 323 337 351 365 0 0.7 1.4 2.0 2.7 3.3 4.0 4.7 5.3
10
Covariate correlations
1.8 1.6 1.4 1.2 1.2 1.4 1.6 1.8
SECR
0.6
0.8
1.0
1.2
100
120
120
100
WT
80
170 160
HT
160
140
50
AGE
40 30 30 40
11
Adults
Males
Females
Children
Males
Females
Nonsmokers
12
Nonsmokers
15
Smokers
18
Smokers
12
Covariate correlations
GENO: 2
40
GENO: 3
30
20
10
Percentage of total
GENO: 0
40
GENO: 1
30
20
10
0 20 25 30 35 40 45 20 25 30 35 40 45
AGE
13
Covariate correlations
Correlated covariates partially carry the same information To simultaneously have correlated covariates in the model will increase model instability and may result in counterintuitive models The relationship between a covariate and a parameter will be different if it is in a model alone or with a correlated covariate To determine which of correlated covariates carry the most predictive value is often hard Problems with correlated covariates increase with increasing correlation Correlation coefficient is often used to capture degree of correlation, but is (alone) not an ideal measure to decide on modeling strategy
14
Clinically available versus experimental Routine use or not for individualisation Of predictive value or not
15
1Bonate
16
Many
Increasing problems with model selection bias, model interpretability, model stability and/or parameter imprecision
Few
17
Statistical significance
Clinical relevance
18
19
Best-guess model
Advantage: unbiased estimates for all Par-Cov relations included in best guess model Like the approaches above it will involve further covariate model building (later lectures)
20
21
22
23
24
25
26
Extra slides
27
28
29
30
31