BE study designs [Design Issues]
Hello Essar,
at least the basics of bioequivalence are not that complicated
Ok, BE gives us information about the performance (generally extent [AUC] and rate [Cmax/tmax] of bioavailability of a test treatment (e.g., a new formulation of an innovator, a generic drug...) compared to a reference treatment (e.g., an approved formulation of an innovator...). If we accept some assumptions (i.e., constant clearances in cross-over designs, clinical meaningful validity of a chosen acceptance range...] and a formulation was shown to be bioequivalent, we conclude that clinical data used in the registration process for the 'old' formulation is also valid for the 'new' one. In other words bioequivalence is an accepted surrogate for clinical equivalence.
Since generally the intra-subject variability CVintra (the variability of a given PK parameter in a given group of subjects if measured on at least two occasions) is lower than inter-subject variability CVinter (the variability between different subjects), cross-over designs (where both test and reference treatments are given to the same subject) are preferable to parallel designs (where one group of subjects is treated with the test, and another group with the reference) in terms of statistical power. In other words you will need (much) less subjects in a cross-over study compared to a paralled study.
The most simple cross-over design is the 2×2×2 (2 treatment, 2 period, 2 sequence) design:
Half of the subjects are treated in the first period with test, the other half with reference; in the second period the first half is treated with reference, and the second half with test.
Therefore we have two sequences (or groups): T-R for the first half of subjects, and R-T for the second half. Subjects are randomized between these two sequences.
In a parallel design half of the subjects will be treated with test, the other half with reference (only one study period). In the BE setting parallel designs are only rarely used (e.g., for drugs with very long half-lives, or studies in patients).
Before starting the study you have to perform a proper sample-size estimation. You will need the following information beforehand: The acceptance range and confidence level (mainly given by a regulatory authority, e.g., 80% – 125% and 5%), the producer's risk (e.g., 20%), the variability of any PK metrics, and the anticipated deviation of test from reference (e.g., ±5%).
In more statistical terms:
The error type I (alpha-risk, patient's risk of being treated with a product which was erroneously claimed bioequivalent) is set to 5%. Since there is a 5% chance of the bioavailability in a particular patient being either lower or higher than expected, the risk for the entire population of patients is 2×alpha=10% (therefore the confidence interval is set to 90%).
The error type II (beta-risk, producer's risk of getting a formulation rejected, although it is 'bioequivalent') generally is set to 10% – 20%. Since statistical power is 1-beta, 80% or 90% are used. Lower or higher values may raise ethical issues.
Information on variability may come from previous studies, a (not too small!) pilot study, or literature. The reliability decreases in the given order, therefore you should add some 'safety margin'.
The same applies to the expected deviation from the reference.
After analyzing our biosamples we calculate PK metrics from individual plasma profiles, which subsequently may be subjected to an appropriate statistical method, in order to get point estimates (e.g., 103% of the reference) and a 90% confidence interval (e.g., 93% – 113% of the reference). For any particular PK metric an acceptance range must be defined in the protocol (e.g., 80%-125% for AUC). If your confidence interval is entirely included in the acceptance range, you may claim your formulation to be bioequivalent.
The chosen statistical method depends on the assumed distribution of the PK metrics, i.e., ANOVA on log-transformed AUC/Cmax-data and a nonparametric method on untransformed tmax-data.
The acceptance ranges should be based on clinical grounds, but since generally no 'hard data' supporting such a range exist for the majority of drugs, 80%-125% is used in many legislations for AUC and Cmax. You must check your local guidelines, since exceptions exist (e.g., in some countries these 'goalpost' may be narrower or wider, or even 95% confidence intervals instead of 90% CIs should be used for some drugs).
at least the basics of bioequivalence are not that complicated
❝ I want to know abt the basic BE designs...can you suggest me some reference (preferably something on the net). i'm pretty much confused...2-way crossover, 2 period, 2 sequence, replicate, parallel...could not understand a word...
Ok, BE gives us information about the performance (generally extent [AUC] and rate [Cmax/tmax] of bioavailability of a test treatment (e.g., a new formulation of an innovator, a generic drug...) compared to a reference treatment (e.g., an approved formulation of an innovator...). If we accept some assumptions (i.e., constant clearances in cross-over designs, clinical meaningful validity of a chosen acceptance range...] and a formulation was shown to be bioequivalent, we conclude that clinical data used in the registration process for the 'old' formulation is also valid for the 'new' one. In other words bioequivalence is an accepted surrogate for clinical equivalence.
Since generally the intra-subject variability CVintra (the variability of a given PK parameter in a given group of subjects if measured on at least two occasions) is lower than inter-subject variability CVinter (the variability between different subjects), cross-over designs (where both test and reference treatments are given to the same subject) are preferable to parallel designs (where one group of subjects is treated with the test, and another group with the reference) in terms of statistical power. In other words you will need (much) less subjects in a cross-over study compared to a paralled study.
The most simple cross-over design is the 2×2×2 (2 treatment, 2 period, 2 sequence) design:
Half of the subjects are treated in the first period with test, the other half with reference; in the second period the first half is treated with reference, and the second half with test.
Therefore we have two sequences (or groups): T-R for the first half of subjects, and R-T for the second half. Subjects are randomized between these two sequences.
In a parallel design half of the subjects will be treated with test, the other half with reference (only one study period). In the BE setting parallel designs are only rarely used (e.g., for drugs with very long half-lives, or studies in patients).
Before starting the study you have to perform a proper sample-size estimation. You will need the following information beforehand: The acceptance range and confidence level (mainly given by a regulatory authority, e.g., 80% – 125% and 5%), the producer's risk (e.g., 20%), the variability of any PK metrics, and the anticipated deviation of test from reference (e.g., ±5%).
In more statistical terms:
The error type I (alpha-risk, patient's risk of being treated with a product which was erroneously claimed bioequivalent) is set to 5%. Since there is a 5% chance of the bioavailability in a particular patient being either lower or higher than expected, the risk for the entire population of patients is 2×alpha=10% (therefore the confidence interval is set to 90%).
The error type II (beta-risk, producer's risk of getting a formulation rejected, although it is 'bioequivalent') generally is set to 10% – 20%. Since statistical power is 1-beta, 80% or 90% are used. Lower or higher values may raise ethical issues.
Information on variability may come from previous studies, a (not too small!) pilot study, or literature. The reliability decreases in the given order, therefore you should add some 'safety margin'.
The same applies to the expected deviation from the reference.
After analyzing our biosamples we calculate PK metrics from individual plasma profiles, which subsequently may be subjected to an appropriate statistical method, in order to get point estimates (e.g., 103% of the reference) and a 90% confidence interval (e.g., 93% – 113% of the reference). For any particular PK metric an acceptance range must be defined in the protocol (e.g., 80%-125% for AUC). If your confidence interval is entirely included in the acceptance range, you may claim your formulation to be bioequivalent.
The chosen statistical method depends on the assumed distribution of the PK metrics, i.e., ANOVA on log-transformed AUC/Cmax-data and a nonparametric method on untransformed tmax-data.
The acceptance ranges should be based on clinical grounds, but since generally no 'hard data' supporting such a range exist for the majority of drugs, 80%-125% is used in many legislations for AUC and Cmax. You must check your local guidelines, since exceptions exist (e.g., in some countries these 'goalpost' may be narrower or wider, or even 95% confidence intervals instead of 90% CIs should be used for some drugs).
—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz
The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz
The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
Complete thread:
- BE study designs Essar 2006-01-12 04:47
- BE study designsHelmut 2006-01-12 18:38
- BE study designs Essar 2006-01-13 04:40
- BE study designs Essar 2006-01-16 13:20
- BE study designs Helmut 2006-01-16 15:39
- BE study designs Essar 2006-01-17 04:53
- BE study designs Helmut 2006-01-17 09:27
- BE study designs nguyenvo 2006-04-30 21:40
- sample size reference Helmut 2006-05-01 11:59
- BE study designs olacy 2006-06-23 09:28
- Reference (direct link) Helmut 2006-06-24 12:45
- BE study designs Essar 2006-01-17 04:53
- BE study designs Helmut 2006-01-16 15:39
- BE study designs shiv 2006-01-25 06:46
- ANVISA?! Helmut 2006-01-25 13:37
- BE study designsHelmut 2006-01-12 18:38