ADDITIVE VERSUS MULTIPLICATIVE MODELS FOR CONTINUOUS OUTCOMES IN BAYESIAN META-ANALYSIS: IMPLICATIONS FOR EVIDENCE SYNTHESIS IN POSTOPERATIVE PAIN
Author(s)
Suzanne Freeman, PhD1, Nicola Cooper, PhD1, Brett Doleman, PhD2, Nicky Welton, PhD3, Alex Sutton, PhD1.
1University of Leicester, Leicester, United Kingdom, 2University of Nottingham, Nottingham, United Kingdom, 3University of Bristol, Bristol, United Kingdom.
1University of Leicester, Leicester, United Kingdom, 2University of Nottingham, Nottingham, United Kingdom, 3University of Bristol, Bristol, United Kingdom.
OBJECTIVES: Meta-analysis of continuous outcomes is typically conducted on an additive scale using mean differences (MD). However, heterogeneity in postoperative pain studies may be partly explained by severity of outcome in the control group, suggesting that treatment effects may not be additive. This study evaluates additive and multiplicative (ratio of means, RoM) Bayesian meta-analysis models focusing on their implications for model fit and heterogeneity in evidence synthesis relevant to healthcare decision-making.
METHODS: A systematic search of the Cochrane Database of Systematic Reviews was conducted (January 2026) to identify intervention reviews in ‘pain & anaesthesia’, focusing on acute pain and regional anaesthesia. Eligible reviews included ≥3 trials and reported at least one MD meta-analysis. Bayesian fixed- and random-effects pairwise meta-analyses were fitted using both MD and RoM parameterizations, with and without adjustment for pain scores on the control arm. Model fit was assessed using deviance information criteria (DIC), and between-study heterogeneity was evaluated using percentage shrinkage.
RESULTS: A total of 47 eligible reviews (range 3-90 trials; median 6) were analyzed. Switching from MD to RoM reduced DIC in 30 (64%) and 26 (55%) of reviews for fixed- and random-effects models, respectively. Adjustment for pain scores on the control arm in random-effects RoM and MD models improved model fit in 17 (36%) and 13 (28%) reviews, respectively. The random-effects RoM model showed lower heterogeneity compared with the MD model (% shrinkage 33% vs 27%).
CONCLUSIONS: Multiplicative models, which allow treatment effects to vary with initial severity of pain, improved model fit and reduced heterogeneity compared with conventional additive approaches in postoperative pain meta-analyses. These findings highlight the importance of model choice for continuous outcomes and support routine consideration of RoM models, as model choice may influence treatment effect estimates and downstream decision-making in health technology assessment and economic evaluation.
METHODS: A systematic search of the Cochrane Database of Systematic Reviews was conducted (January 2026) to identify intervention reviews in ‘pain & anaesthesia’, focusing on acute pain and regional anaesthesia. Eligible reviews included ≥3 trials and reported at least one MD meta-analysis. Bayesian fixed- and random-effects pairwise meta-analyses were fitted using both MD and RoM parameterizations, with and without adjustment for pain scores on the control arm. Model fit was assessed using deviance information criteria (DIC), and between-study heterogeneity was evaluated using percentage shrinkage.
RESULTS: A total of 47 eligible reviews (range 3-90 trials; median 6) were analyzed. Switching from MD to RoM reduced DIC in 30 (64%) and 26 (55%) of reviews for fixed- and random-effects models, respectively. Adjustment for pain scores on the control arm in random-effects RoM and MD models improved model fit in 17 (36%) and 13 (28%) reviews, respectively. The random-effects RoM model showed lower heterogeneity compared with the MD model (% shrinkage 33% vs 27%).
CONCLUSIONS: Multiplicative models, which allow treatment effects to vary with initial severity of pain, improved model fit and reduced heterogeneity compared with conventional additive approaches in postoperative pain meta-analyses. These findings highlight the importance of model choice for continuous outcomes and support routine consideration of RoM models, as model choice may influence treatment effect estimates and downstream decision-making in health technology assessment and economic evaluation.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
SA14
Topic
Methodological & Statistical Research, Study Approaches
Topic Subcategory
Meta-Analysis & Indirect Comparisons
Disease
No Additional Disease & Conditions/Specialized Treatment Areas, Systemic Disorders/Conditions (Anesthesia, Auto-Immune Disorders (n.e.c.), Hematological Disorders (non-oncologic), Pain)