TY - UNPB
T1 - Exploring the use of synthetic placebo populations in ALS randomized clinical trials
AU - Bowles, H.
AU - Opie-Martina, S.
AU - Shojaiea, A.
AU - Libonatib, L.
AU - Iacoangelia, A.
AU - Al Khleifat, A.
PY - 2022
Y1 - 2022
N2 - ObjectivesThe use of synthetic data to supplement clinical trial placebo groups or for trial planning is rapidly gaining interest. However, there is not yet an established framework for generating synthetic data for these purposes. In this work we test two approaches to generating synthetic placebo arms for ALS trials with survival being the primary outcome variable.
MethodsFor the first approach, we extracted sample subsets from the UK MND register (n = 308) using an evolutionary algorithm such that the subset baseline variables matched a target trials group, either people enrolled in LiCALS (n = 106) or people included in the PRO-ACT database (n = 171). We also applied trial specific exclusion criteria where possible or alternatively we applied a custom time filter. For the second approach, survival was predicted for LiCALS participants using the ENCALS model. Survival probabilities from each method were compared to real placebo participants using Kaplan-Meier analysis and the log rank test.
ResultsWe found that the synthetic placebo groups derived from the MND register matched the target trials outcomes very well. The ENCALS model produces synthetic placebo groups that are significantly different to the real placebo groups. However, when participants are censored at 6 month intervals, the ENCALS synthetic group matches the target group very well between 24 and 48 months, indicating a possible timeframe that this method could be utilised.
ConclusionBoth the register based approach and the ENCALS prognostic model generated synthetic placebo groups that matched placebo groups from historical trials. These methods need to be validated in prospective trials.
AB - ObjectivesThe use of synthetic data to supplement clinical trial placebo groups or for trial planning is rapidly gaining interest. However, there is not yet an established framework for generating synthetic data for these purposes. In this work we test two approaches to generating synthetic placebo arms for ALS trials with survival being the primary outcome variable.
MethodsFor the first approach, we extracted sample subsets from the UK MND register (n = 308) using an evolutionary algorithm such that the subset baseline variables matched a target trials group, either people enrolled in LiCALS (n = 106) or people included in the PRO-ACT database (n = 171). We also applied trial specific exclusion criteria where possible or alternatively we applied a custom time filter. For the second approach, survival was predicted for LiCALS participants using the ENCALS model. Survival probabilities from each method were compared to real placebo participants using Kaplan-Meier analysis and the log rank test.
ResultsWe found that the synthetic placebo groups derived from the MND register matched the target trials outcomes very well. The ENCALS model produces synthetic placebo groups that are significantly different to the real placebo groups. However, when participants are censored at 6 month intervals, the ENCALS synthetic group matches the target group very well between 24 and 48 months, indicating a possible timeframe that this method could be utilised.
ConclusionBoth the register based approach and the ENCALS prognostic model generated synthetic placebo groups that matched placebo groups from historical trials. These methods need to be validated in prospective trials.
KW - neurology
U2 - 10.1101/2022.12.09.22283281
DO - 10.1101/2022.12.09.22283281
M3 - Preprint
BT - Exploring the use of synthetic placebo populations in ALS randomized clinical trials
ER -