Modelling the effects of drug dose in hierarchical network ...€¦ · (e.g., Low, Medium, and High...
Transcript of Modelling the effects of drug dose in hierarchical network ...€¦ · (e.g., Low, Medium, and High...
-
Modelling the effects of drug
dose in hierarchical network
meta-analysis models to account
for variability in treatment
classifications Areti Angeliki Veroniki, MSc, PhD
Prepared for: 2016 CADTH Symposium
April 12, 2016
C. Del Giovane, D. Jackson, S. E. Straus, S. Thomas, E. Blondal, K. Thavorn, A. C. Tricco
Knowledge Translation Program,
Li Ka Shing Knowledge Institute,
St. Michael's Hospital,
Toronto, Canada
E-mail: [email protected]
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
I have no actual or potential conflict of interest in relation to this presentation
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
of the presentation
• To present approaches to model the effects of drug dosages in hierarchical network meta-analysis (NMA)
• How can we model dose-effects in NMA while accounting for the drug-dose relationship?
• To present hierarchical NMA models accounting for effects of: 1) treatment, 2) strength (e.g., low, medium, high) of dose, and 3) specific dose
• How do initial decisions on the network structure impact heterogeneity and inconsistency?
• To illustrate examples of the available approaches
• Does the effect size vary across different dose levels?
• Is there a pattern in dose-effects?
3
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Network Meta-analysis (NMA)
NMA has become increasingly popular over the last two decades with ~500 publications
Temporal distribution of publications (n=494)
*2015 sample is not complete
0.2%
0.2%
0.4%
0.2%
0.2%
0.4%
1.0%
2.8%
6.3%
6.7%
11.7%
12.5%
18.3%
23.8%
10.5%
0 25 50 75 100 125
1997
1999
2000
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015*
2.4%
2.4%
• In many conditions there are multiple
treatments that could be considered
• When policy makers are considering
what interventions to cover through
health plans or what safety labels to
put on medications, they need
evidence from an NMA because this
method uses all available RCTs for a
specific clinical topic
• Clinicians and patients need to know
the safest and most effective
treatment dose for a particular
situation
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Combining placebo-controlled trials to learn about toothpaste vs. rinse may yield erroneous results!
Salanti et al JCE 2009
Common Dilemma in NMA…
Lumping or splitting nodes?
Typical approach: Choice between ‘lumping’ and ‘splitting’ methods
Splitting:
Allows comparison between different treatment doses
⊠ Ignores the relationship between doses that belong to the same intervention ⊠ May lead to a relatively small number of studies across comparisons resulting in
considerable uncertainty around the overall treatment effects
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Common Dilemma in NMA…
How do NMA authors deal with treatment doses?
219 (44%) of 494 NMAs published up to April 14, 2015 included different treatment doses in the network
Only 11 NMAs accounted
properly for the treatment-dose
relationship!
Guidelines on modeling different dose-effects in NMA are imperative
Lumping:
Increases the amount of evidence across comparisons and hence precision in treatment effects
⊠ Ignores the different doses included in each intervention ⊠ Cannot model studies that compare the same treatment at different doses
• Researchers have to select a specific dose for the underlying treatment
• If no other treatment is compared in the particular study, the study has to be excluded.
5%
0
20
4
0
60
8
0
Fre
qu
en
cy
NR Lump (L) Split (S) Recomm. Most Prevalent
Model
3%
51%
34%
0.5% 0.5%
S&L
6%
100
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Variance components added to each new level
A B
RCTs comparing 2 treatment
dosages
within-study variance
Meta-analysis
𝜇𝐴𝐵
between-study variance Usually we assume common between-study variance across dose comparisons
Network Meta-analysis
Categorizing doses into groups (e.g., Low, Medium, and High doses)
between-dose-category variance within treatment*
*here the between dose variance is within dose-category (and not treatment)
between-dose variance within treatment
within dose category if doses are categorized
into groups
Categorizing doses into groups of treatments
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Modelling dose-effects in NMA
Decision-makers are often interested in a treatment’s effectiveness at lower and higher doses
Hierarchical models can be used to address dose-effects in NMA and to facilitate the identification of the most effective treatment and optimal dose
B
A
E
D C
The potential dependence of treatment-effects of drug dose is an important issue for effectiveness and safety and is challenging in NMA because comparisons between interventions may vary in doses
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Modelling dose-effects in NMA Independent dose-effects
Random dose-effects within treatments
Fixed dose-effects
Random dose-effects within different dose categories
Del Giovane et al Stat Med 2013
Dose-effects are related and exchangeable*
All dose-effects are unrelated
All dose-effects are assumed equal within the same treatment
Dose-effects are related and exchangeable accounting for the dose-category they belong to
a) within-study and b) between-study variance across doses
a) within-study and b) between-study variance within dose level
a) within-study, b) between-study within dose level, and c) between-dose variance within-treatment level
a) within-study, b) between-study within dose level, c) between-dose within-dose-category level, and d) between-dose-category variance within-treatment level
* Within the treatment they belong to
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Illustrative Examples
1. Serotonin 5-hydroxytryptamine 3 (5HT3) receptor antagonists to treat patients undergoing surgery
2. Anti-vascular endothelial growth factor (anti-VEGF) drugs to treat patients with wet age-related macular degeneration (AMD)
* All analyses have been performed in a Bayesian framework using OpenBUGS
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Illustrative Example 1
Aim
To estimate the relative safety of
5HT3 receptor antagonists at
different dosages
Serotonin 5-hydroxytryptamine 3 (5HT3) receptor antagonists to treat patients undergoing surgery
• 5-HT3 receptor antagonists are commonly used to decrease nausea and
vomiting for surgery patients
• Evidence shows that these agents may be harmful, but it is unclear whether there
are certain doses that increase the risk of arrhythmia
• Clinicians and patients are interested in the safest treatment dose
• We used data from a published systematic review and NMA
Treatment Dose (mg) Placebo Ondansetron 1mg, 2mg, 3mg, 4mg, 8mg, 16mg, 24mg Granisetron 0.1mg, 1mg, 3mg Dolasetron 12.5mg, 25mg, 50mg, 100mg, 200mg Tropisetron 0.5mg, 2mg, 5mg Ramosetron 0.3mg, 0.9mg
Tricco et al BMC Medicine 2015
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Illustrative Example 1
• Patients of any age undergoing any type of surgery who were given a 5-HT3
receptor antagonist for nausea and/or vomiting
• Databases searched: MEDLINE, Embase, the Cochrane Central Register of
Controlled Trials; protocol registries and conference proceedings – up to January
2013
• One dichotomous outcome was considered to assess safety: arrhythmia –
We will also assess efficacy using the vomiting outcome
• 29 RCTs were eligible after the screening process
• Exclusion:
Compared a combination of treatments against a single treatment (1 study).
Did not report the treatment dosages used (1 study).
• In total, 27 studies were included in the analysis
5HT3 receptor antagonists to treat patients undergoing surgery
Tricco et al BMC Medicine 2015
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
5-HT3 receptor antagonists for surgery
Tricco et al BMC Medicine 2015
Arrhythmia : 27 studies, 8871 patients, 6 treatments, 21 doses
Circles from outside in refer to: 1st: Fixed Effects model 2nd: Random Effects (with dose consistency) 3rd: Random Effects (no dose consistency) 4th: Independent Effects
1 : Placebo 2 : Ondansetron-Fixed 3 : Ondansetron-1mg 4 : Ondansetron-2mg 5 : Ondansetron-3mg 6 : Ondansetron-4mg 7 : Ondansetron-8mg 8 : Ondansetron-16mg 9 : Ondansetron-24mg 10 : Granisetron-Fixed 11 : Granisetron-0.1mg 12 : Granisetron-1mg 13 : Granisetron-3mg
14 : Dolasetron-Fixed 15 : Dolasetron-12.5mg 16 : Dolasetron-25mg 17 : Dolasetron-50mg 18 : Dolasetron-100mg 19 : Dolasetron-200mg 20 : Tropisteron-Fixed 21 : Tropisteron-0.5mg 22 : Tropisteron-2mg 23 : Tropisteron-5mg 24 : Ramosetron-Fixed 25 : Ramosetron-0.3mg 26 : Ramosetron-0.9mg
safest
Least safe
Veroniki et al JCE 2016
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
5-HT3 receptor antagonists for surgery
1 .0225 1 44.5
OR (95% CrI) Common comparator: Placebo
Dolasetron
0.75 (0.52, 1.08)
0.76 (0.51, 1.11)
0.79 (0.47, 1.31)
0.74 (0.54, 1.03)
0.76 (0.40, 1.44)
0.76 (0.50, 1.11)
0.79 (0.46, 1.31)
0.76 (0.53, 1.09)
0.76 (0.53, 1.09)
0.75 (0.51, 1.10)
0.61 (0.32, 1.11)
0.73 (0.50, 1.05)
0.76 (0.50, 1.10)
0.53 (0.24, 1.10)
0.73 (0.48, 1.05)
0.75 (0.51, 1.11)
Fixed
12.5mg
25mg
50mg
100mg
200mg
Placebo better Active better
0.88 (0.35, 2.23)
0.84 (0.35, 2.33)
1.21 (0.36, 4.34) 0.84 (0.34, 2.28)
0.84 (0.35, 2.33) 0.62 (0.14, 2.48) 0.81 (0.33, 2.19) 0.88 (0.35, 2.23)
0.88 (0.34, 2.24) 0.76 (0.18, 2.80) 0.82 (0.33, 2.22)
Fixed
0.1mg
1mg
3mg
Granisetron
1 .0225 1 44.5
OR (95% CrI) Common comparator: Placebo
Ondansetron
Placebo better Active better
0.91 (0.62, 1.30) 0.77 (0.42, 1.33) 0.89 (0.64, 1.20)
1.01 (0.46, 2.00)
0.90 (0.69, 1.18) 1.11 (0.54, 2.30) 0.92 (0.66, 1.30) 0.91 (0.62, 1.30) 1.37 (0.44, 4.51) 0.91 (0.65, 1.32) 0.91 (0.62, 1.30) 0.72 (0.19, 2.72) 0.90 (0.62, 1.27)
0.77 (0.42, 1.33)
0.91 (0.62, 1.30)
0.91 (0.65, 1.28) 0.91 (0.62, 1.30) 0.95 (0.50, 1.75) 0.90 (0.65, 1.24) 0.91 (0.62, 1.30) 0.65 (0.18, 2.29) 0.90 (0.62, 1.26) 0.91 (0.62, 1.30)
Fixed
1mg
2mg
3mg
4mg
8mg
16mg
24mg
Ramosetron
1.20 (0.70, 2.11) 1.27 (0.53, 2.95) 1.20 (0.67, 2.18) 1.24 (0.67, 2.38) 1.16 (0.48, 2.78) 1.19 (0.66, 2.16) 1.24 (0.66, 2.40)
Fixed 0.3mg
0.9mg
Tropisetron
0.85 (0.43, 1.76)
0.83 (0.33, 2.06) 0.85 (0.43, 1.63)
0.86 (0.44, 1.67) 0.75 (0.02, 7.20) 0.86 (0.42, 1.73) 0.85 (0.43, 1.76) 0.86 (0.25, 2.59) 0.86 (0.43, 1.67) 0.85 (0.43, 1.75)
Fixed
0.5mg
2mg
5mg
Fixed Effects
Random Effects (with dose consistency)
Independent Effects
Random Effects (no dose consistency)
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
5-HT3 receptor antagonists for surgery
FE model RE model with no dose consistency
RE model with dose consistency
Independent dose-effects with no dose consistency
𝜏2 (95% CrI)
0.01 (0.00, 0.10)
0.01 (0.00, 0.10)
0.01 (0.00, 0.12)
0.04 (0.00, 0.47)
𝜎2 (95% CrI)
0.01 (0.00, 0.08)
0.00 (0.00, 0.07)
DIC 109.44 110.86 111.02 133.68
D 90.29 90.42 89.91 97.77
pD 19.15 20.44 21.11 35.91
Data points
82 82 82 82
The design by treatment interaction model suggested consistency on both treatment (χ2=2.46, P-value= 0.7829, τ2=0.00) and dose (χ2=14.35, P-value=0.6423, τ2=0.00) levels
5HT3 surgery – Arrhythmia
Spiegelhalter et al J R Stat Soc Ser B Stat Methodol 2002
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Illustrative Example 2
Aim
To estimate the relative efficacy of anti-VEGF agents at different dosages
Anti-vascular endothelial growth factor (anti-VEGF) drugs to treat patients with wet age-related macular degeneration (AMD)
• Anti-VEGF drugs are injected into the eye where they inhibit the
abnormal angiogenesis that underlies many retinal conditions
associated with vision loss
• It is unclear whether these agents have equivalent efficacy, and specifically whether
there are certain doses that patients would benefit from when comparing these
different drugs CADTH Therapeutic Review 2015
Treatment Abbreviation Dose (mg) Placebo PLAC Ranibizumab RANI 0.3mg, 0.5mg, 2mg,
Bevacizumab BEVA 1.25mg, 2.5mg Aflibercept AFLI 0.5mg, 2mg, 4mg Triamcinolone acetonide TRIAM 4mg
Photodynamic therapy with verteporfin PDT Standard fluence, reduced fluence
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Illustrative Example 2
Anti-VEGF drugs to treat patients with wet AMD
CADTH Therapeutic Review 2015
• Adults (≥ 18 years) with wet AMD
• Databases searched: MEDLINE, Embase, Cochrane Central Register of Controlled
Trials; Grey literature (trial protocols) – up to October 2015
• Two dichotomous outcomes were considered: vision gain & vision loss
• 35 RCTs were eligible after the screening process
• Exclusion:
compared the same treatment at the same dose (8 studies).
did not report vision gain outcome (3 studies) or vision loss outcome (3
studies).
• In total, 24 studies were included in the analysis for both outcomes
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Anti-vascular endothelial growth factor
Vision gain: 24 studies, 9444 patients, 5 treatments, 11 doses
Vision loss: 24 studies, 8311 patients, 6 treatments, 12 doses
RANI – 0.5 mg
RANI – 0.3 mg
RANI – 2 mg BEVA – 1.25 mg
BEVA – 2.5 mg
AFLI – 0.5 mg
AFLI – 2 mg
AFLI – 4 mg
PLAC PDT - standard
PDT - reduced
RANI
BEVA
AFLI
PLAC
PDT
RANI
BEVA AFLI
PLAC
PDT TRIAM
RANI – 0.5 mg
RANI – 0.3 mg
RANI – 2 mg BEVA – 1.25 mg
BEVA – 2.5 mg
AFLI – 0.5 mg
AFLI – 2 mg
AFLI – 4 mg
PLAC PDT- standard
PDT - reduced
TRIAM – 4 mg
Treatment Level
Dose Level
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Anti-vascular endothelial growth factor Vision gain: 24 studies, 9444 patients, 5 treatments, 11 doses
Treatment Level
Fixed Effects
Independent Effects
Random Effects (no dose consistency)
Random Effects (with dose consistency)
Dose Level
RANI vs PLAC
BEVA vs PLAC
AFLI vs PLAC
PDT vs PLAC
8.46 (5.04,14.09)
8.45 (4.93,14.13)
7.93 (4.57,13.47)
7.72 (4.54,12.76)
6.80 (3.63,12.26)
6.70 (3.48,12.31)
6.66 (3.24,13.08)
6.33 (2.88,13.03)
7.74 (3.65,15.58)
7.65 (3.43,15.87)
7.87 (3.74,15.25)
5.38 (2.22,11.85)
3.09 (1.42,7.59)
3.15 (1.35,8.05)
3.13 (1.32,8.19)
3.54 (1.23,11.00)
OR (95%CrI) Treatment Comparison
2.2 1 10 20
Odds Ratio
Active better Placebo better
RANI-0.3mg vs PLAC
RANI-0.5mg vs PLAC
RANI-2mg vs PLAC
BEVA-1.25mg vs PLAC
BEVA-2.5mg vs PLAC
AFLI-0.5mg vs PLAC
AFLI-2mg vs PLAC
AFLI-4mg vs PLAC
PDT-reduced vs PLAC
PDT-standard vs PLAC
8.44 (4.08,17.76) 7.50 (4.69,12.03) 7.18 (4.41,11.64) 8.47 (4.07,17.64) 8.50 (5.20,13.52) 8.76 (5.27,14.17) 8.42 (4.05,17.55) 7.79 (4.35,13.49) 7.30 (3.14,16.09) 6.70 (2.89,14.62) 6.87 (3.71,11.84) 7.07 (3.74,12.41) 6.69 (2.83,14.75) 6.50 (3.03,13.03) 5.62 (1.80,16.49) 7.65 (3.00,18.79) 7.82 (4.01,14.30) 7.77 (3.76,14.98) 7.67 (2.97,18.82) 8.61 (4.39,15.93) 9.20 (4.50,17.78) 7.63 (2.98,18.47) 7.35 (2.82,14.94) 2.22 (0.33,11.29)
3.13 (1.19,9.26) 3.19 (1.28,8.72)
4.20 (0.74,25.26) 3.14 (1.20,9.16) 3.07 (1.39,7.12) 2.99 (1.35,7.28)
OR (95%CrI) Treatment Comparison
.3 1 9 27
Active better
Placebo better
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Anti-vascular endothelial growth factor - SUCRA
Vision gain: 24 studies, 9444 patients, 5 treatments, 11 doses
Treatment Level
PD
T
Treatment hierarchy according to the SUrface under the Cumulative RAnking curve
Salanti JCE 2009; Veroniki et al JCE 2016
Dose Level
Circles from outside in refer to: 1st: Fixed Effects model 2nd: Random Effects (no dose consistency) 3rd: Random Effects (with dose consistency) 4th: Independent Effects
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Anti-VEGF - Vision Gain
FE model RE model with no dose consistency
RE model with dose consistency
Independent dose-effects with no dose consistency
𝜏2 (95% CrI)
0.09 (0.02, 0.29)
0.04 (0.00, 0.23)
0.06 (0.00, 0.24)
0.07 (0.00, 0.29)
𝜎2 (95% CrI)
0.05 (0.00, 0.23)
0.02 (0.00, 0.33)
DIC 99.64 99.03 99.18 102.43
D 59.1 57.58 58.86 59.06
pD 40.54 41.45 40.32 43.37
Data points
55 55 55 55
The design by treatment interaction model suggested consistency on the dose level: (χ2=1.94, P-value=0.585, τ2=0.08) – Consistency cannot be assessed on the treatment level
Spiegelhalter et al J R Stat Soc Ser B Stat Methodol 2002
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Anti-vascular endothelial growth factor Vision loss: 24 studies, 8311 patients, 6 treatments, 12 doses
Treatment Level
Fixed Effects
Independent Effects
Random Effects (no dose consistency)
Random Effects (with dose consistency)
Dose Level
RANI vs PLAC
BEVA vs PLAC
AFLI vs PLAC
PDT vs PLAC
TRIAM vs PLAC
0.11 (0.08,0.16) 0.11 (0.07,0.17) 0.12 (0.07,0.25) 0.15 (0.09,0.25) 0.10 (0.06,0.17) 0.10 (0.06,0.18) 0.11 (0.05,0.30) 0.15 (0.06,0.33) 0.10 (0.05,0.18) 0.10 (0.05,0.20) 0.09 (0.04,0.20) 0.06 (0.02,0.18) 0.45 (0.33,0.60) 0.44 (0.30,0.63) 0.92 (0.28,2.76) 0.90 (0.40,2.01) 0.91 (0.42,2.03) 0.91 (0.38,2.21) 0.40 (0.16,0.71) 0.29 (0.16,0.49)
OR (95%CrI) Treatment Comparison
0 .1 1 .8 3
Odds Ratio
Placebo better
Active better
RANI-0.3mg vs PLAC
RANI-0.5mg vs PLAC
RANI-2mg vs PLAC
BEVA-1.25mg vs PLAC
BEVA-2.5mg vs PLAC
AFLI-0.5mg vs PLAC
AFLI-2mg vs PLAC
AFLI-4mg vs PLAC
PDT-reduced vs PLAC
PDT-standard vs PLAC
TRIAM-4mg vs PLAC
0.11 (0.07,0.20) 0.11 (0.07,0.16) 0.10 (0.07,0.16) 0.11 (0.07,0.20) 0.11 (0.07,0.16) 0.11 (0.07,0.17) 0.11 (0.06,0.20) 0.14 (0.08,0.40) 0.30 (0.10,0.96) 0.10 (0.05,0.20) 0.10 (0.06,0.17) 0.10 (0.06,0.18) 0.10 (0.05,0.20) 0.12 (0.06,0.38) 0.21 (0.06,0.80) 0.10 (0.04,0.22) 0.09 (0.05,0.18) 0.09 (0.05,0.19) 0.10 (0.04,0.22) 0.10 (0.05,0.19) 0.10 (0.05,0.20) 0.10 (0.04,0.22) 0.08 (0.02,0.21) 0.02 (0.00,0.33) 0.91 (0.36,2.35) 0.33 (0.12,0.59) 0.18 (0.06,0.45) 0.91 (0.36,2.37) 0.46 (0.34,0.62) 0.47 (0.35,0.64) 0.44 (0.26,0.73) 0.92 (0.41,2.00) 0.90 (0.40,2.01)
OR (95%CrI) Treatment Comparison
1 .3 2.5
Odds Ratio
Placebo better Active better
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Anti-vascular endothelial growth factor - SUCRA
Vision loss: 24 studies, 8311 patients, 6 treatments, 12 doses
Treatment Level
Treatment hierarchy according to the SUrface under the Cumulative RAnking curve
Salanti JCE 2009; Veroniki et al JCE 2016
Dose Level
Circles from outside in refer to: 1st: Fixed Effects model 2nd: Random Effects (no dose consistency) 3rd: Random Effects (with dose consistency) 4th: Independent Effects
BEVA
PDT
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Anti-VEGF - Vision loss
FE model RE model with no dose consistency
RE model with dose consistency
Independent dose-effects with no dose consistency
𝜏2 (95% CrI)
0.02 (0.00, 0.16)
0.02 (0.00, 0.17)
0.02 (0.00, 0.16)
0.02 (0.00, 0.17)
𝜎2 (95% CrI)
0.01 (0.00, 0.17)
0.08 (0.00, 0.93)
DIC 97.57 98.70 96.93 98.44
D 61.32 60.13 58.04 56.85
pD 36.25 38.57 38.9 41.59
Data points
54 54 54 54
The design by treatment interaction model suggested consistency on the dose level: (χ2=1.17, P-value=0.761, τ2=0.00) – Consistency cannot be assessed on the treatment level
Spiegelhalter et al J R Stat Soc Ser B Stat Methodol 2002
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Summary…
Modelling dose-effects in NMA and accounting for the intervention-dose relationship:
• Adds to borrow strength in estimating dose-effects within treatment classes • Overcomes problems with sparse data in the treatment networks • Can incorporate studies that compare the same treatment at different doses • Allows the identification of not only the best treatment in a network, but
also the most effective dose • Increases power compared to carrying out several independent subgroup
analyses, lumping or extreme splitting approaches • May provide additional insight on heterogeneity, inconsistency, intervention
ranking, and hence decision-making
Why is it important to model dose-effects in NMA?
Different approaches used to classify treatments or model dose-effects in a network may result in important variations in interpretations drawn from NMA
But….
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Summary…
The available approaches cannot model studies that compare combinations of treatments in a single node (e.g., A+B vs. C+D)
Dose-effect models do not always explain inconsistency. Inconsistency may still be evident due to an imbalance in the distribution of effect modifiers across comparisons NMA models require consistency in at least one (e.g., treatment) level.
Consistency should be evaluated at each level it is assumed
An IPD-NMA is the optimal approach to explore these factors. Hence, it
is imperative to advance NMA models addressing dose-effects using IPD, improving interpretability of NMA results
But….
Studies are needed to establish and advance IPD-NMA
addressing dose-effects to make results more useful for decision-making
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Future Work Develop dose-response NMA models allowing to impose structural
relationships between the basic parameters Use reduced forms of standard NMA models where all doses and treatments are
conceptualized as separate treatments (accounting for between study within dose level variance)
Potentially introduce fractional polynomials in these models
Develop hierarchical models that relax the consistency assumption in dose and
treatment levels Use higher dimension design specific inconsistency parameters than the standard
NMA models, so that all studies receive appropriate inconsistency parameters related to their treatment-doses
Extend all models including the between-dose-category level (e.g., low,
recommended, high category of doses) We are currently working on the statistical code for the random dose-effects within
different dose categories model
Extend dose-response NMA models including IPD.
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
1. Del Giovane C, Vacchi L, Mavridis D, Filippini G, Salanti G. Network meta-analysis models to account for variability in treatment definitions: application to dose effects. Statistics in Medicine 2013; 32:25–39
2. Donegan S, Williamson P, D'Alessandro U, et al. Assessing the consistency assumption by exploring treatment by covariate interactions in mixed treatment comparison meta-analysis: individual patient-level covariates versus aggregate trial-level covariates. Statistics in medicine 2012;31(29):3840-57.
3. Nikolakopoulou A, Chaimani A, Veroniki AA, Vasiliadis HS, Schmid CH, Salanti G. PLoS One. 2014 22;9(1):e86754. doi: 10.1371/journal.pone.0086754.
4. Owen RK, Tincello DG, Keith RA. Network meta-analysis: development of a three-level hierarchical modeling approach incorporating dose-related constraints. Value Health 2015 18(1):116-26. doi: 10.1016/ j.jval.2014.10.006.
5. Salanti G, Marinho V, Higgins JP. A case study of multiple-treatments meta-analysis demonstrates that covariates should be considered. J Clin Epidemiol. 2009;62(8):857-64
6. Spiegelhalter DJ BN, Carlin BP, Van Der Linde A. Bayesian measures of model complexity and fit. J R Stat Soc Ser B Stat Methodol 2002;64(64):57
7. Tricco AC, Soobiah C, Blondal E, Veroniki AA, Khan PA, Vafaei A, et al. Comparative safety of serotonin (5-HT3) receptor antagonists in patients undergoing surgery: A systematic review and network meta-analysis, BMC Medicine. 2015; 13:136
8. Thomas S, Lillie E, Adams J, et al. Anti–vascular endothelial growth factor (VEGF) agents for the treatment of retinal conditions. CADTH Therapeutic Review, 2015
9. Veroniki AA, Mavridis D, Higgins JP, Salanti G. Statistical evaluation of inconsistency in a loop of evidence: a simulation study informed by empirical evidence. BMC Medical Research Methodology 2014; 14:106
10. Veroniki AA, Straus SE, Fyraridis A, Tricco AC. The rank-heat plot is a novel way to present the results from a network meta-analysis including multiple outcomes. J. Clin. Epidemiol. 2016; pii: S0895-4356(16)00153-0
11. Warren FC,AbramsKR,SuttonAJ.Hierarchical Bayesian networkmeta-analysis models to address sparsity of events and differing treatment classifications with regard to adverse outcomes. Stat Med 2014;33:2449–66.
12. White IR, Barret JK, Jackson D, Higgins JPT. Consistency and inconsistency in multiple treatments meta-analysis: model estimation using multivariate meta-regression. Research Synthesis Methods 2012;3(2):111-25.
References…
-
Knowledge Translation, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Canada
Special thanks to:
• My supervisor and mentors:
Dr. Sharon Straus, Dr. Andrea Tricco
• Co-authors:
Dr. Cinzia Del Giovane, Dr. Daniel Jackson, Ms. Sonia Thomas, Mr. Erik Blondal, Dr. Kednapa Thavorn
• CIHR Banting Postdoctoral Fellowship Program
E-mail: [email protected]