Extended MAIC to Validate Results in Cases of Poor Population Overlap: An Application to Compare Prophylactic Treatments in Pediatric Patients With Hemophilia A
Author(s)
Mikolaj Parkitny1, Samuel Aballea, MSc, PhD2, Piotr Wojciechowski, Msc3, Mondher Toumi, Sr., MSc, PhD, MD4.
1Student, Aix-Marseille University, Kraków, Poland, 2Inovintell, Paris, France, 3Clever-Access, Kraków, Poland, 4Aix-Marseille University, Aix-Marseille, France.
1Student, Aix-Marseille University, Kraków, Poland, 2Inovintell, Paris, France, 3Clever-Access, Kraków, Poland, 4Aix-Marseille University, Aix-Marseille, France.
OBJECTIVES: Poor overlap in population characteristics limits the feasibility and credibility of Matching-Adjusted Indirect Comparisons (MAIC). In this analysis, we apply a novel method to quantify the risk of bias when comparing samples with significantly different characteristics.
METHODS: Patient-level data from a single-arm trial evaluating a novel antihemophilic therapy (Int_A) were compared with aggregated data from the LEOPOLD Kids trial assessing octocog alfa (OctA). The comparison focused on annualized FVIII consumption and employed unanchored MAIC, adjusting for baseline variables: age, BMI, race, and presence of target joints (TJ) (Full MAIC). The proportion of patients with TJ was highly imbalanced (2.7% vs. 27.5%). To address this issue, a Reduced MAIC model excluding target joints was used. In addition, an Extended MAIC was performed, whereby FVIII consumption was simulated across a range of potential impact values of target joints on FVIII consumption for a proportion of patients from the Int-A trial equal to the proportion in LEOPOLD Kids.
RESULTS: In 74 patients treated with Int_A, unadjusted mean annual FVIII consumption was lower than in OctA (2,934IU/kg vs. 4,884IU/kg). Full MAIC demonstrated a significant reduction in FVIII consumption with Int_A compared to OctA (difference: -1,962IU/kg [95%CI: -2,350; -1,574]). However, Effective Sample Size (ESS) was low (11%) due to poor population overlap. Reduced MAIC results were consistent, with a higher ESS (38%). Extended MAIC indicated that for the difference to become non-significant, patients with TJ would need to consume on average 3,565IU/kg of FVIII more compared to those without TJ. Given the average FVIII consumption of 2,934IU/kg, such a consumption was considered highly unlikely.
CONCLUSIONS: The proposed method quantifies the risk of bias in MAIC when populations poorly overlap. By addressing imbalances and testing plausible scenarios, this approach enhances the robustness and credibility of MAIC results, reducing the risk of misleading conclusions.
METHODS: Patient-level data from a single-arm trial evaluating a novel antihemophilic therapy (Int_A) were compared with aggregated data from the LEOPOLD Kids trial assessing octocog alfa (OctA). The comparison focused on annualized FVIII consumption and employed unanchored MAIC, adjusting for baseline variables: age, BMI, race, and presence of target joints (TJ) (Full MAIC). The proportion of patients with TJ was highly imbalanced (2.7% vs. 27.5%). To address this issue, a Reduced MAIC model excluding target joints was used. In addition, an Extended MAIC was performed, whereby FVIII consumption was simulated across a range of potential impact values of target joints on FVIII consumption for a proportion of patients from the Int-A trial equal to the proportion in LEOPOLD Kids.
RESULTS: In 74 patients treated with Int_A, unadjusted mean annual FVIII consumption was lower than in OctA (2,934IU/kg vs. 4,884IU/kg). Full MAIC demonstrated a significant reduction in FVIII consumption with Int_A compared to OctA (difference: -1,962IU/kg [95%CI: -2,350; -1,574]). However, Effective Sample Size (ESS) was low (11%) due to poor population overlap. Reduced MAIC results were consistent, with a higher ESS (38%). Extended MAIC indicated that for the difference to become non-significant, patients with TJ would need to consume on average 3,565IU/kg of FVIII more compared to those without TJ. Given the average FVIII consumption of 2,934IU/kg, such a consumption was considered highly unlikely.
CONCLUSIONS: The proposed method quantifies the risk of bias in MAIC when populations poorly overlap. By addressing imbalances and testing plausible scenarios, this approach enhances the robustness and credibility of MAIC results, reducing the risk of misleading conclusions.
Conference/Value in Health Info
2025-05, ISPOR 2025, Montréal, Quebec, CA
Value in Health, Volume 28, Issue S1
Code
MSR57
Topic
Methodological & Statistical Research
Disease
SDC: Systemic Disorders/Conditions (Anesthesia, Auto-Immune Disorders (n.e.c.), Hematological Disorders (non-oncologic), Pain)