Investigating the relationship between predictability and imbalance in minimisation: a simulation study

Gladys C McPherson, Marion K Campbell, Diana R Elbourne

Research output: Contribution to journalArticle

5 Citations (Scopus)
3 Downloads (Pure)

Abstract

Background
The use of restricted randomisation methods such as minimisation is increasing. This paper investigates under what conditions it is preferable to use restricted randomisation in order to achieve balance between treatment groups at baseline with regard to important prognostic factors and whether trialists should be concerned that minimisation may be considered deterministic.

Methods
Using minimisation as the randomisation algorithm, treatment allocation was simulated for hypothetical patients entering a theoretical study having values for prognostic factors randomly assigned with a stipulated probability. The number of times the allocation could have been determined with certainty and the imbalances which might occur following randomisation using minimisation were examined.

Results
Overall treatment balance is relatively unaffected by reducing the probability of allocation to optimal treatment group (P) but within-variable balance can be affected by any P <1. This effect is magnified by increased numbers of prognostic variables, the number of categories within them and the prevalence of these categories within the study population.

Conclusions
In general, for smaller trials, probability of treatment allocation to the treatment group with fewer numbers requires a larger value P to keep treatment and variable groups balanced. For larger trials probability of allocation values from P = 0.5 to P = 0.8 can be used while still maintaining balance. For one prognostic variable there is no significant benefit in terms of predictability in reducing the value of P. However, for more than one prognostic variable, significant reduction in levels of predictability can be achieved with the appropriate choice of P for the given trial design.
Original languageEnglish
Article number86
JournalTrials
Volume14
DOIs
Publication statusPublished - 27 Mar 2013

Fingerprint

Random Allocation
Therapeutics
Theoretical Models
Population

Keywords

  • randomisation
  • minimisation
  • allocation imbalance
  • allocation predictability

Cite this

Investigating the relationship between predictability and imbalance in minimisation : a simulation study. / McPherson, Gladys C; Campbell, Marion K; Elbourne, Diana R.

In: Trials, Vol. 14, 86, 27.03.2013.

Research output: Contribution to journalArticle

@article{50f0c9994ecc41eca696f0cca3a5b669,
title = "Investigating the relationship between predictability and imbalance in minimisation: a simulation study",
abstract = "BackgroundThe use of restricted randomisation methods such as minimisation is increasing. This paper investigates under what conditions it is preferable to use restricted randomisation in order to achieve balance between treatment groups at baseline with regard to important prognostic factors and whether trialists should be concerned that minimisation may be considered deterministic.MethodsUsing minimisation as the randomisation algorithm, treatment allocation was simulated for hypothetical patients entering a theoretical study having values for prognostic factors randomly assigned with a stipulated probability. The number of times the allocation could have been determined with certainty and the imbalances which might occur following randomisation using minimisation were examined.ResultsOverall treatment balance is relatively unaffected by reducing the probability of allocation to optimal treatment group (P) but within-variable balance can be affected by any P <1. This effect is magnified by increased numbers of prognostic variables, the number of categories within them and the prevalence of these categories within the study population.ConclusionsIn general, for smaller trials, probability of treatment allocation to the treatment group with fewer numbers requires a larger value P to keep treatment and variable groups balanced. For larger trials probability of allocation values from P = 0.5 to P = 0.8 can be used while still maintaining balance. For one prognostic variable there is no significant benefit in terms of predictability in reducing the value of P. However, for more than one prognostic variable, significant reduction in levels of predictability can be achieved with the appropriate choice of P for the given trial design.",
keywords = "randomisation, minimisation, allocation imbalance, allocation predictability",
author = "McPherson, {Gladys C} and Campbell, {Marion K} and Elbourne, {Diana R}",
year = "2013",
month = "3",
day = "27",
doi = "10.1186/1745-6215-14-86",
language = "English",
volume = "14",
journal = "Trials",
issn = "1745-6215",
publisher = "BioMed Central",

}

TY - JOUR

T1 - Investigating the relationship between predictability and imbalance in minimisation

T2 - a simulation study

AU - McPherson, Gladys C

AU - Campbell, Marion K

AU - Elbourne, Diana R

PY - 2013/3/27

Y1 - 2013/3/27

N2 - BackgroundThe use of restricted randomisation methods such as minimisation is increasing. This paper investigates under what conditions it is preferable to use restricted randomisation in order to achieve balance between treatment groups at baseline with regard to important prognostic factors and whether trialists should be concerned that minimisation may be considered deterministic.MethodsUsing minimisation as the randomisation algorithm, treatment allocation was simulated for hypothetical patients entering a theoretical study having values for prognostic factors randomly assigned with a stipulated probability. The number of times the allocation could have been determined with certainty and the imbalances which might occur following randomisation using minimisation were examined.ResultsOverall treatment balance is relatively unaffected by reducing the probability of allocation to optimal treatment group (P) but within-variable balance can be affected by any P <1. This effect is magnified by increased numbers of prognostic variables, the number of categories within them and the prevalence of these categories within the study population.ConclusionsIn general, for smaller trials, probability of treatment allocation to the treatment group with fewer numbers requires a larger value P to keep treatment and variable groups balanced. For larger trials probability of allocation values from P = 0.5 to P = 0.8 can be used while still maintaining balance. For one prognostic variable there is no significant benefit in terms of predictability in reducing the value of P. However, for more than one prognostic variable, significant reduction in levels of predictability can be achieved with the appropriate choice of P for the given trial design.

AB - BackgroundThe use of restricted randomisation methods such as minimisation is increasing. This paper investigates under what conditions it is preferable to use restricted randomisation in order to achieve balance between treatment groups at baseline with regard to important prognostic factors and whether trialists should be concerned that minimisation may be considered deterministic.MethodsUsing minimisation as the randomisation algorithm, treatment allocation was simulated for hypothetical patients entering a theoretical study having values for prognostic factors randomly assigned with a stipulated probability. The number of times the allocation could have been determined with certainty and the imbalances which might occur following randomisation using minimisation were examined.ResultsOverall treatment balance is relatively unaffected by reducing the probability of allocation to optimal treatment group (P) but within-variable balance can be affected by any P <1. This effect is magnified by increased numbers of prognostic variables, the number of categories within them and the prevalence of these categories within the study population.ConclusionsIn general, for smaller trials, probability of treatment allocation to the treatment group with fewer numbers requires a larger value P to keep treatment and variable groups balanced. For larger trials probability of allocation values from P = 0.5 to P = 0.8 can be used while still maintaining balance. For one prognostic variable there is no significant benefit in terms of predictability in reducing the value of P. However, for more than one prognostic variable, significant reduction in levels of predictability can be achieved with the appropriate choice of P for the given trial design.

KW - randomisation

KW - minimisation

KW - allocation imbalance

KW - allocation predictability

U2 - 10.1186/1745-6215-14-86

DO - 10.1186/1745-6215-14-86

M3 - Article

C2 - 23537389

VL - 14

JO - Trials

JF - Trials

SN - 1745-6215

M1 - 86

ER -