Population/regional differences in efficacy of 3 drug categories (antidiabetic, respiratory and psychotropic agents) among East Asians: A retrospective study based on multiregional clinical trials

Abstract

Aims

This study aimed to identify population/regional differences in drug efficacy and the influencing factors among East Asians to be considered when planning multiregional clinical trials (MRCTs) to facilitate rapid drug approval in Asians.

Methods

A retrospective analysis of efficacy (intergroup difference in endpoint between control and study drug treatment) among East Asian populations for 3 drug categories, antidiabetic, respiratory and psychotropic agents, was conducted in collaboration with pharmaceutical companies using their MRCT data. Common endpoints by drug category were selected; background factors that commonly affected the endpoints among regions were analysed first; then the population/regional differences were evaluated by the interaction term region‐by‐treatment using an analysis of covariance model after adjusting for background factors.

Results

Among 17 endpoints for eight pharmaceutical products from 3 drug categories, no substantial population/regional differences were detected in the 3 drug categories examined (P > .05), except for haemoglobin A1c change between Japan and Korea for an antidiabetic drug, insulin glulisine (P = .0068). However, no such regional differences were evident in patients with clinically important higher haemoglobin A1c baseline values (majority subgroup). Variability in disease severity at baseline and concomitant drugs were determined to be potential influencing factors for regional differences.

Conclusions

This study suggests that the regional variability in efficacy of these 3 drug categories is not large among East Asians, and reveals the importance of considering background factors when planning MRCTs. Further studies are needed to evaluate regional variability in the efficacy of other drug categories and clarify the factors leading to regional differences in East Asians.

What is already known about this subject

• Multiregional clinical trials (MRCTs) are encouraged worldwide as an efficient strategy for resolving drug lag; however, ethnic factors are a key issue when planning MRCTs.

• Involvement of East Asian countries in MRCTs has been rapidly increasing; however, knowledge of regional variability in drug efficacy is limited.

What this study adds

• No substantial population/regional differences in efficacy (intergroup difference between control and study drug treatment) were found for 3 drug categories among East Asian populations.

• This study identified some background factors related to regional differences that should be considered when planning MRCTs in East Asian countries.

1. INTRODUCTION

Drug lag, a delay in availability of a new drug already approved in other countries, has been a critical issue in many countries including Japan. Multiregional clinical trials (MRCTs) are an efficient drug development strategy to resolve drug lag issues; accordingly, regulatory authorities encourage pharmaceutical companies to conduct such trials.1, 2, 3, 4 However, ethnic factors are a key issue when planning MRCTs. The International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH) E5 guidelines define ethnic factors as those relating to the intrinsic (e.g. genetic and physiological) and extrinsic (e.g. medical practice, cultural, and environmental) characteristics of a population.5 The ICH E17 guideline was developed to provide general principles for planning and designing MRCTs to encourage their use in global regulatory submissions. This guideline additionally points out the importance of pre‐considering regional variabilities and their potential impact on efficacy and safety when designing MRCTs.6

Currently, most MRCTs have been conducted in the USA and Europe; however, the involvement of Asian countries, especially East Asian countries, is expected to promote rapid and efficient drug development in Asian countries, assuming similar genetic backgrounds and pharmacokinetic (PK) profiles. Currently, several studies comparing East Asian populations have indicated no large differences in genetic polymorphisms for major drug‐responsive genes.7, 8 Further, no clinically relevant differences in PK and safety profiles of 4 compounds with different characteristics, based on analysis of global clinical trial data, have been found.9 Moreover, the percentage rate of MRCTs for drug development from 2008 to 2015 increased to a higher extent in East Asia than in the USA and Europe.10 Along with this trend, international discussion on the importance of collaborations in drug development and research regarding ethnic differences among East Asians has been accelerated following the Asia‐Pacific Economic Cooperation and the East Asian Tripartite (China, Korea and Japan) Health Minister meeting.11 According to the joint statement of the first tripartite Health Ministers meeting in April 2007, a collaboration project on PK was conducted; in contrast to previous results, this joint project revealed no significant differences in PK parameters among 3 East Asian populations under the strict control of extrinsic factors, such as food and drinks, indicating the importance of extrinsic factors as well as intrinsic factors.12, 13

Although knowledge regarding factors influencing PK has been accumulating, regional variability in pharmacodynamics (PD; drug efficacy and safety) and influential intrinsic and extrinsic factors that are common or specific among drug types still need to be studied.

In this study, we conducted a secondary retrospective efficacy analysis of 3 drug categories among East Asian populations to determine the population/regional differences in efficacy and influencing factors among East Asians. This work was performed in collaboration with pharmaceutical companies, using their MRCT data.

2. METHODS

2.1. Cooperation system for the study

This study was conducted in collaboration with the following pharmaceutical companies in Japan: Sanofi K.K. (Tokyo, Japan); AstraZeneca K.K. (Osaka, Japan); Novo Nordisk Pharma Ltd. (Tokyo, Japan); Novartis Pharma K.K. (Tokyo, Japan); Eli Lilly Japan K.K. (Kobe, Japan); and Otsuka Pharmaceutical Co. Ltd. (Tokyo, Japan). All analyses using individual data were conducted in‐house by the company that performed the MRCT (Table S1) according to a protocol specifically prepared by our project team for this study and the aggregated data were provided by each company to the project team. The collaborating companies and project team discussed the results and interpretation and agreed on the publication. This study was based on a retrospective analysis, with secondary use of data of MRCTs that were not specifically designed to evaluate population/regional differences. Thus, the current analysis was conducted using a small number of subjects from each region. This study was approved by all collaborating companies and the ethics committee of the National Institute of Health Sciences (NIHS approval No. 295).

2.2. Study drugs and countries included in MRCTs

Eight drugs, namely insulin glulisine, exenatide, insulin degludec, indacaterol, glycopyrronium, olanzapine, aripiprazole and atomoxetine, which have been approved for marketing in Japan were selected from 3 therapeutic categories (antidiabetic, respiratory and psychotropic agents), as their development has been increasing worldwide and assessed in MRCTs in Asian countries, including Japan and East‐Asian countries/areas (China, Hong Kong, Taiwan and South Korea). In regard to exenatide, both Byetta (control) and Bydureon (study drug) contain the same active ingredient. The MRCTs included in this study conformed to recognized standards and appropriate informed consent was obtained. Detailed information regarding each pharmaceutical product and participating countries/areas in the MRCTs is shown in Table 1, with alphabetic drug identifiers (ID) which are used in Figures 2, 3, 4, 5, 6, 7, 8, 9 and Table 3. For some of the MRCTs, the number of subjects from Hong Kong or Taiwan was small; therefore, data on these populations were combined with those from China. Further, southeast Asian or non‐Asian countries were classified into 1 subgroup, assuming a high degree of racial similarity. Although the dropout rates varied among regions, especially for antidiabetic agents, we assumed that this would not substantially affect the current evaluation of population/regional difference of the endpoints, which was based on the last observation carried forward as conducted in MRCT.

Table 1.

Study drugs and participating countries in multi‐regional clinical trials

Drug category	Drug		Indication	Treatment group a	Country/area b
	ID	Nonproprietary name (brand name)			Japan	China/Hong Kong	Taiwan	South Korea	Southeast or South Asia	Non‐Asians
Antidiabetic agent	A	Insulin glulisine (Apidra)	Diabetes: need for insulin therapy	Drug‐A + OHA vs OHA	132 (51%)	(−)	(−)	125(49%)	(−)	(−)
					3.1%			11.2%
	B	Exenatide (Bydureon)	Type 2 diabetes	Drug‐B (prolonged action) vs Byetta (BID)	155 (23%)	207 (31%)	74 (11%)	62 (9%)	180 (27%)c	(−)
					12.3%	14.0%	12.2%	35.5%	14.4%
	C	Insulin degludec (Tresiba)	Diabetes: need for insulin therapy	Drug‐C vs insulin glargine	133 (31%)	31 (7%)d	31 (7%)d	142(33%)	98 (23%)e	(−)
					3.8%	6.5%	12.9%	12.7%	12.2%
Respiratory agent	D	Indacaterol (Onbrez)	Chronic obstructive pulmonary disease	Drug‐D vs placebo	152 (46%)	7 (2%)d	45 (14%)d	100 (31%)	41 (12%)c	(−)
					10.5%	28.6%	13.3%	10.0%	14.3%
	E	Glycopyrronium (Seebri)	Chronic obstructive pulmonary disease	Drug‐E vs placebo	96 (12%)	(−)	(−)	31 (4%)	123 (16%)c	531 (68%)f
					11.5%			9.7%	5.7%	13.9%
Psychotropic agent	F	Olanzapine (Zyprexa)	Bipolar disorder	Drug‐F vs placebo	156 (30%)	210 (41%)d	28 (5%)d	30 (6%)	(−)	90 (18%)g
					16.7%	24.3%	28.6%	16.7%		28.9%
	G	Aripiprazole (Abilify)	Bipolar disorder	Drug‐G vs placebo	79 (32%)	56 (23%)d	36 (14%)d	(−)	77 (31%)h	(−)
					49.4%	50.0%	54.3%		41.0%
	H	Atomoxetine (Strattera)	Attention‐deficit hyperactivity disorder	Drug‐H vs placebo	247 (64%)	(−)	68 (18%)	73 (19%)	(−)	(−)
					10.1%		20.6%	26.0%

^aIntention‐to‐treat or full analysis set.

^bUpper: number of patients at start (percent by country/area). Lower: dropout rate (percent of the number of patients at start compared to the number of patients at end).

^cIndia.

^dData were combined for analysis by country/area.

^eMalaysia and Thailand.

^fCanada, Argentina, Australia, Netherlands, Romania, Russia, Spain, Turkey, USA.

^gUSA.

^hIndonesia, Malaysia, Thailand.

OHA: oral hypoglycaemic agent; BID: twice daily; (−): did not participate.

Figure 2.

Comparison of changes in haemoglobin A1c (HbA1c) after treatment with an antidiabetic agent (Drug‐A) between Japan and South Korea. An analysis of covariance model by treatment group was adopted for visual evaluation of regional differences in HbA1c changes after treatment with Drug‐A (insulin glulisine) by stratification with baseline HbA1c values, a potential influencing factor. P‐value of interaction term Region‐by‐Treatment is shown in the upper panel. Number of subjects in each region is indicated in parentheses. OHA: oral hypoglycaemic agent, LS: least square, CI: confidence interval

Figure 3.

Comparison of changes in haemoglobin A1c (HbA1c) after treatment with antidiabetic agents (Drugs B and C) among regions. No significant regional differences in changes to HbA1c values after treatment with Drug‐B (exenatide) or Drug‐C (insulin degludec) were observed using an analysis of covariance (ANCOVA). P‐value of interaction term Region‐by‐Treatment is shown in upper panel for each drug. Mean and 95% CIs for the overall treatment differences are indicated as thin and dotted lines, respectively, in lower panels. Number of subjects in each region is indicated in parentheses. LS: least square, SE: standard error, CI: confidence interval, HK: Hong Kong

Figure 4.

Comparison of changes in fasting blood or plasma glucose (FG) after treatment with antidiabetic agents (Drugs A–C) among regions.No significant regional differences in changes in FG after treatment with Drug‐A (insulin glulisine) Drug‐B (exanatide) or Drug‐C (insulin degludec) were observed using an analysis of covariance (ANCOVA). P‐value of interaction term Region‐by‐Treatment is shown in upper panel for each drug. Mean and 95% CIs of overall treatment differences are indicated as thin and dotted lines, respectively, in lower panels. Number of subjects in each region is indicated in parentheses. LS: least square, SE: standard error, CI: confidence interval, HK: Hong Kong

Figure 5.

Comparison of trough forced expiratory volume in 1 second (FEV₁) after treatment with respiratory agents (Drugs D and E) among regions. An analysis of covariance (ANCOVA) model by treatment group was adopted to visually evaluate regional differences attributable to Drug‐D (indacaterol) via stratification with baseline FEV₁ reversibility, an influential factor. Baseline value for stratification is indicated as a dotted line in the upper left panel for Drug‐D. No significant regional differences in trough FEV₁ attributable to Drug‐E (glycopyrronuium) were observed using ANCOVA. P‐value of interaction term Region‐by‐Treatment for Drug‐E is shown in upper panel. Mean and 95% CIs of overall treatment differences in Drug‐E are indicated as thin and dotted lines, respectively, in lower panels. Number of subjects in each region is indicated in parentheses. LS: least square, SE: standard error, CI: confidence interval, HK: Hong Kong

Figure 6.

Comparison of number of puffs of rescue medication after treatment with respiratory agents (Drugs D and E) among regions. Analysis of covariance (ANCOVA) model by treatment group was adopted to visually evaluate regional differences attributable to Drug‐D (indacaterol) using stratification with baseline number of puffs of rescue medication, an influencing factor. Baseline value of stratification is indicated as a dotted line in the upper left panel for Drug‐D. No significant regional differences in number of puffs of rescue medication after treatment with Drug‐E (glycopyrronuium) were observed using ANCOVA. P‐value of interaction term Region‐by‐Treatment for Drug‐E is shown in upper panel. Mean and 95% CIs of overall treatment differences in Drug‐E are indicated as thin and dotted lines, respectively, in lower panels. Number of subjects in each region is indicated in parentheses. LS: least square, SE: standard error, CI: confidence interval, HK: Hong Kong

Figure 7.

Comparison of changes in Montgomery–Asberg depression rating scale (MADRS) scores attributable to bipolar disorder agents (Drugs F and G) among regions. No significant regional differences in changes to MADRS scores attributable to Drug‐F (olanzapine) or Drug‐G (aripiprazole) were observed using analysis of covariance (ANCOVA). P‐value of interaction term Region‐by‐Treatment is shown in upper panel for each drug. Mean and 95% CIs of overall treatment differences are indicated as thin and dotted lines, respectively, in lower panels. Number of subjects in each region is indicated in parentheses. LS: least square, SE: standard error, CI: confidence interval

Figure 8.

Comparison of changes in Young mania rating scale (YMRS) scores attributable to bipolar disorder agents (Drugs F and G) among regions.No significant regional differences in changes to YMRS scores attributable to Drug‐F (olanzapine) or Drug‐G (aripiprazole) were observed using analysis of covariance (ANCOVA). P‐value of interaction term Region‐by‐Treatment is shown in upper panel for each drug. Mean and 95% CIs of overall treatment differences are indicated as thin and dotted lines, respectively, in lower panels. Number of subjects in each region is indicated in parentheses. LS: least square, SE: standard error, CI: confidence interval

Figure 9.

Comparison of changes in Conners' adult attention deficit hyperactivity disorder (ADHD) rating scale‐investigator rated: screening version (CAARS‐Inv:SV) and adult ADHD quality of life questionnaire (AAQOL) scores attributable to an ADHD agent (Drug H) among regions. No significant regional differences in changes to CAARS‐Inv:SV and AAQOL scores after treatment with Drug‐H (atomoxetine) were observed using analysis of covariance (ANCOVA). P‐value of interaction term Region‐by‐Treatment is shown in upper panel for each drug. Mean and 95% CIs of overall treatment differences are indicated as thin and dotted lines, respectively, in lower panels. Number of subjects in each region is indicated in parentheses. LS: least square, SE: standard error, CI: confidence interval

Table 3.

Summary of influential factors and regional differences for drug efficacy based on ANCOVA

Drug category Endpoint	1) Influence of background factors		2) Population/regional differences		3) Factors potentially related to population/region differences
	Influential background factor	Interaction with treatment group	Population/region	Interaction with treatment group
Antidiabetic agent (Drug ID in parentheses; A: insulin glulisine B: insulin glulisine, C: insulin degludec
Changes in HbA1c	HbA1c↓(B, C)	ns (all)	ns (all)	S. Koreaa (A)ns (B, C)	OHA (type/dose)
	FG↓(C)
	OHA co‐therapy↑ (C)
Changes in FG	HbA1c ↑(B)	ns (all)	ns (all)	ns (all)
	FG↓(B, C)
	Sulfonyl urea co‐therapy↑(B)
Respiratory agent (Drug ID in parentheses; D:indacaterol, E: glycopyrronuim)
Trough FEV1	Trough FEV1↑(D, E)	FEV1 reversibility(D)b	ns (E)	ns (E)	FEV1 reversibility (D)
	Severity↓(D, E)
	FEV1 reversibility↑↓(D)
	Age ↑↓(D)
	Test sitec(E)
Number of puffs of rescue medication	Number of puffs of rescue medication↑(D, E)	Number of puffs of rescue medication (D)b	S. Korea (E)	ns (E)	Number of puffs of rescue medication (D) Severity (E)
	Test sitec(D, E)
	Severity ↑(E)
Psychotropic agent (Drug ID in parentheses; F: olanzapine, G: aripipurazole, H; atomoxetine)
Changes in MADRS score (F, G)	MADRS↓(F, G)	ns (F, G)	ns (F, G)	ns (F,G)	Age (G)
	Female↑ (F)
Changes in YMRS score (F, G)	YMRS↓(F)	ns (F, G)	USA (F) ns (G)	ns (F,G)	YMRS (F) Age (G)
	MADRS↓ (G)
	Age↓ (G)
Changes in CAARS‐Inv:SV score (H)	AAQOL↓ (H)	ns (H)	Taiwan, S. Korea (H)	ns (H)	ADHD‐inattentive type (H)
Changes in AAQOL score (H)	CAARS‐Inv:SV↑ (H)	ns (H)	Taiwan, S. Korea (H)	ns (H)	ADHD‐inattentive type (H)
	AAQOL↓(H)

Arrows indicate influential direction on endpoint: ↑: upward, ↓downward, ↑↓:different by dosage. Significance was evaluated as described in Methods and Figure 1.

^aVisual evaluation based on the ANCOVA model by treatment group including baseline HbA1c, a potentially influential factor in regional differences.

^bRegional differences in the endpoints for Drug‐D (indacaterol) were visually evaluated using the ANCOVA model by treatment group, attributable to a significant interaction with this background factor.

^cOmitted, attributable to too many sites.

HbA1c: haemoglobin A1c; FG: fasting glucose of blood or plasma; OHA: oral hypoglycaemic agent; FEV₁: forced expiratory volume in 1 second; MADRS: Montgomery‐Asberg depression rating scale; YMRS: Young mania rating scale; ADHD: attention‐deficit hyperactivity disorder; CAARS‐Inv:SV: Conners' adult ADHD rating scale‐investigator rated: screening version; AAQOL: adult attention deficit hyperactivity disorder quality of life questionnaire.

2.3. Efficacy evaluation and potential influencing factors

For evaluating population/regional difference, efficacy was defined as intergroup differences in endpoint between control and study‐drug treatment. In this study, common endpoints of the same drug category, except for atomoxetine, were chosen from the primary and secondary endpoints evaluated in the MRCTs (Table 2). The influence of each patient background factor on each endpoint was examined before establishing the final ANCOVA model for assessing population/regional differences. For each drug, we preliminarily analysed regional variations of all patient demographics and baselines related to disease phenotypes among each region, where the regional difference was regarded as large enough if the mean baseline value or proportion of subjects in 1 region was lower than 80% or higher than 125% compared with that of Japan. We also selected the potential confounders/effect modifiers in general: age, sex, body weight, endpoint baseline and renal function; and drug‐specific items: genetic polymorphisms of drug metabolizing enzyme, concomitant drugs, smoking history, disease phenotypes and duration. All selected background factors for pre‐ANCOVA are listed in Table 2.

Table 2.

Endpoints and background factorsa for evaluation of efficacy

Drug category	Antidiabetic agent	Respiratory agents	Psychotropic agent
			Bipolar disorder agent	ADHD agent
Endpointb	Changes in HbA1c and FG	Trough FEV1 and Number of uses of rescue medication	Changes in MADRS and YMRS scores	Changes in CAARS‐Inv:SV and AAQOL
Background factor or test condition for evaluation of efficacy	Age Sex Body weight HbA1c FG BMI Renal function (% of normal patients)c Duration of diabetes Pre‐treatment of diabetes co‐treatment with sulfonylurea agent or OHA Test site (arbitrary)	Age Sex Body weight Trough FEV 1 Number of puffs of rescue medication Duration of respiratory disease Smoking history (habitual) Amount of smoking (pack–year) Severity (severe or very severe) Test site	Age Sex Body weight MADRS YMRS Dose reduction CYP2D6 EM	Age Sex Body weight CAARS‐Inv:SV AAQOL ADHD‐inattentive type

Bold: common item within a drug category

^aincluding common item (age, sex, body weight, baseline of evaluation item) and item with large difference in the mean value from Japan (lower than 80% or 125% and higher).

^bLOCF: last observation carried forward.

^cDefined as creatinine clearance >80 mL/min.

OHA: oral hypoglycaemic agent; HbA1c: haemoglobin A1c based on National Glycohemoglobin Standardization Program; FG: fasting glucose of blood or plasma; BMI: body‐mass index; MRCT: multiregional clinical trial; FEV₁: forced expiratory volume in 1 second; MADRS: Montgomery‐Asberg depression rating scale; YMRS: Young mania rating scale; ADHD: attention‐deficit hyperactivity disorder; CAARS‐Inv:SV: Conners’ adult ADHD rating scale‐investigator rated: screening version; AAQOL: adult attention deficit hyperactivity disorder quality‐of‐life questionnaire.

2.4. Analysis of population/regional differences

The analyses of population/regional differences in efficacy (intergroup difference), as shown in Figure 1, were performed by comparing results for Japan, the reference country, to those for other regions. Our evaluation was based on the formal interaction test.14 To identify common influencing background factors in all regions, effects of each background factor listed in Table 2 and resulting interactions with treatment group (Treat) were first evaluated for each endpoint using the overall data using an analysis of covariance (ANCOVA), with the following model (equation 1):

$$\mathrm{E}={\mathrm{\beta }}_0+{\mathrm{\beta }}_1\times \left[\text{Treat}\right]+{\mathrm{\beta }}_2\times \left[\mathrm{BG}\right]+{\mathrm{\beta }}_3\times \left[\text{Treat}\right]\times \left[\mathrm{BG}\right]$$ (1)

where E, Treat and BG represent endpoint, treatment group and background factor, respectively.

Figure 1.

An analysis flowchart and evaluation of the influence of background factors and population/regional differences in drug efficacy. ANCOVA: analysis of covariance

The influence of each background factor, and its interaction with Treat were first assessed and selected based on P‐values with a significance level of 0.15 using type III sum of squares. All screened covariates and corresponding interaction terms were incorporated into the new ANCOVA model, and significant variables were selected via a stepwise, forward and backward selection procedure at a significance level of 0.05 (the second selection). If results among the selection methods were different, data from the stepwise method were adopted. If no interaction term was significant in the second selection model, the terms of region (Region) and its interaction with Treat were added to the second selection model. Finally, the ANCOVA model was used to evaluate population/regional differences (equation 2). Using the final model (equation 2), the population/regional differences affecting efficacy were assessed using the P value of the interaction term (between Treat and Region) with type III sum of squares at a significance level of 0.05. The statistical significance of Region was also assessed to determine possible regional differences in overall endpoint levels from Japan.

$$\mathrm{E}={\mathrm{\beta }}_0+{\mathrm{\beta }}_1\times \left[\text{Treat}\right]+{\mathrm{\beta }}_2\times \left[\text{BG1}\right]+···+{\mathrm{\beta }}_{\mathrm{x}}\times \left[\text{Region}\right]+{\mathrm{\beta }}_{\mathrm{y}}\times \left[\text{Treat}\right]\times \left[\text{Region}\right]$$ (2)

where E, Treat, BG and Region represent endpoint, treatment group, background factor and various regions, respectively.

If the interaction term in the second selection model or the final ANCOVA model was significant, the ANCOVA model by treatment group was prepared, excluding the term Treat and its interaction with background factor (equation 3). Using these models, intergroup differences between regions were visually evaluated; this was carried out by comparing the means and 95% confidence intervals (CIs) between treatment groups in all populations and subpopulations stratified by the influencing background factor to determine how the background factor influenced the intergroup differences (comparison between treatment groups) within and between regions. When the background factor was a discrete influential variable, the population was separated into 2 groups based on the background factor. When the background factor was a continuous influential variable, the stratification point was set based on the mean value of background factor for all regions, or 125% of the value of the upper normal limit of the endpoint.

$$\begin{array}{c}\mathrm{E}\left(\text{study drug}\right)={\mathrm{\beta s}}_0+{\mathrm{\beta s}}_1\times \left[\mathrm{BG}1\right]+···+{\mathrm{\beta s}}_{\mathrm{x}}\times \left[\mathrm{BGx}\right]\\ \mathrm{E}\left(\text{control}\right)={\mathrm{\beta c}}_0+{\mathrm{\beta c}}_1\times \left[\mathrm{BG}1\right]+···+{\mathrm{\beta c}}_{\mathrm{x}}\times \left[\mathrm{BGx}\right]\end{array}$$ (3)

where E, study drug, control and BG represent endpoint, treatment group with study drug, control group and background factor, respectively.

Considering the limitation of small sample size in each region, comprehensive evaluation was conducted in this study based on both P‐value and visual assessment using 95% CIs, as well as other global clinical trial data, such as PK values.

3. RESULTS

3.1. Antidiabetic agents

3.1.1. Changes in haemoglobin A1c

Changes in haemoglobin A1c (HbA1c) after treatment with insulin glulisine revealed a significant interaction between Treat (Drug‐A + hypoglycaemic agents [OHA] vs OHA) and Region (S. Korea vs Japan) according to the final ANCOVA model (P = .0068). These results suggested a potential regional difference in this endpoint. As the baseline HbA1c values were assumed to be a potential factor related to this observed regional difference (left upper panel in Figure 2, and Table 3), the ANCOVA model by treatment group including baseline HbA1c was used to visually evaluate regional differences. The visually evaluated intergroup differences in HbA1c were larger for Japan (right upper panel in Figure 2), with estimated HbA1c values (%) of −2.252 (95% CIs: −2.438 to −2.066) in the Drug‐A + OHA group and −0.483 (−0.713 to −0.254) in the OHA group, than for Korea at −1.930 (−2.174 to −1.685) in the Drug‐A + OHA group and − 0.779 (−1.044 to −0.513) in the OHA group. Stratification analyses using baseline HbA1c values of 8.6% revealed more evident regional differences (the larger trend of intergroup differences in Japan) in patients with a lower baseline Hb1Ac (< 8.6%) value. This observation was related to the much smaller changes in HbA1c values in the control OHA group in Japan, although fewer patients had lower HbA1c baseline (<8.6%) (lower left panel in Figure 2). In contrast, no such regional differences between Japan and Korea were evident in the higher baseline (majority) subgroup (≥8.6%) who exhibited a more clinically important disease phenotype (lower right panel in Figure 2). In regard to changes in HbA1c by exenatide, insulin degludec or both, associations were detected in baseline values of Hb1Ac, fasting plasma glucose and co‐administration of OHA in the MRCT (Table 3). The final ANCOVA models for changes in HbA1c revealed no significant regional differences, i.e. no interactions between Treat and Region were observed (Table 3 and Figure 3). This finding suggests that intergroup differences in this endpoint, attributable to treatment with exenatide and insulin degludec, were equivalent among the regions (lower panels in Figure 3).

3.1.2. Changes in fasting glucose

Baseline Hb1Ac, fasting glucose (blood or plasma) values, and co‐administration of sulfonylurea agents in the MRCTs were associated with changes in fasting glucose after treatment with exenatide, insulin degludec or both. Although smaller changes in fasting glucose were observed in the Hong Kong plus Taiwan population than in other regions (Figure 4), the final ANCOVA models revealed no significant interactions between Treat and Region for all antidiabetic agents (Table 3 and Figure 4), indicating that changes in intergroup differences in fasting glucose were equivalent among the regions (lower panels in Figure 4).

3.2. Respiratory agents

3.2.1. Trough forced expiratory volume in 1 second

For the trough forced expiratory volume in 1 second (FEV₁) after treatment with indacaterol, glycopyrronium, or both, the trough FEV₁ baseline values, severity (severe or very severe), FEV₁ reversibility, age and test sites were identified as influencing background factors. Regarding indacaterol, a significant interaction between FEV₁ reversibility and Treat was detected (Table 3). Accordingly, regional differences in trough FEV₁ in response to indacaterol were visually assessed using ANCOVA models by treatment group. Equal or very slightly higher trough FEV₁ values were commonly observed in the low‐dose group (Drug‐D150) than the high‐dose group (Drug‐D300; Figure 5 upper left). Stratified analysis revealed that this trend was also generally observed among the 4 Asian countries/regions at both lower (<15%) and higher (≥15%) baselines. Although the trends in the Hong Kong plus Taiwan population for the higher baseline (≥15%) and both Indian populations were different from the above trend, the number of subjects in these stratified populations was too small to evaluate (lower panels of Drug‐D in Figure 5). For glycopyrronium, the final ANCOVA model revealed no significant regional differences in trough FEV₁ (Table 3), indicating the same degree of intergroup differences in this endpoint among regions (Drug‐E in Figure 5).

3.2.2. Frequency of rescue medication use

The baseline number of puffs of rescue medication, a short acting β‐2 agonist, severity of disease, and test sites influenced the number of puffs after treatment with indacaterol, glycopyrronium or both. For indacaterol, a significant interaction between baseline number of puffs and Treat was detected (Table 3). Accordingly, regional differences in the number of puffs of rescue medication for indacaterol were visually assessed using the ANCOVA models by treatment group. Effectiveness—a reduction in the number of puffs of rescue medication—was noted for the low‐dose group (Drug‐D150); however this was less obvious for the high‐dose group (Drug‐D300; Figure 6 upper left) in the East Asian populations. Stratified analysis divided by the mean baseline number of puffs (2.15) revealed that this trend was also common among East Asian countries at both lower (<2.15) and higher (≥2.15) baselines. Although the trends in the Hong Kong plus Taiwan population for the higher baseline (≥2.15) and both subgroup of Indian populations were different from the above trend, the number of those stratified populations was too small to evaluate (lower panels of Drug‐D in Figure 6). For glycopyrronium the final ANCOVA model showed a significant difference for Region (S. Korea vs Japan); however, no interactions between Region and Treat were detected (Table 3). This result indicates that although the number of puffs of rescue medication in S. Korea was higher in both treatment groups (placebo and glycopyrronium) than in the Japanese population (upper panel for Drug‐E in Figure 6), no substantial intergroup differences were observed for this endpoint between regions (lower panel for Drug‐E in Figure 6).

3.3. Psychotropic agents

3.3.1. Changes in Montgomery–Asberg depression rating scale scores for bipolar disorder agents

Changes in Montgomery–Asberg depression rating scale (MADRS) scores15 after treatment with olanzapine or aripiprazole were associated with baseline MADRS scores and female sex (Table 3). Although a lower response to olanzapine was observed in S. Korea, the final ANCOVA model revealed no interactions between Region and Treat, indicating that intergroup differences were equivalent among regions (Drug‐F in Figure 7). For aripiprazole, the final ANCOVA model for changes in MADRS scores showed no interaction between Region and Treat. Although the term Region was not significant in the final model (Table 3), greater changes in MADRS scores occurred in the placebo groups; thus, smaller intergroup differences were observed in the China plus Taiwan population and the Southeast Asian population than in the Japanese population (Drug‐G in Figure 7). The trend of differential responses among placebo groups did not solely correlate with the proportion of subjects with a dose reduction or who dropped out in the placebo groups (left panel in Figure S2); however, mean baseline age of the placebo groups was lower in these populations than in the Japanese population (right panel in Figure S2).

3.3.2. Changes in Young mania rating scale scores for bipolar disorder agents

Changes in Young mania rating scale (YMRS) scores16 after olanzapine or aripiprazole treatment were associated with baseline values of MADRS scores, YMRS scores, and age (Table 3). The final ANCOVA models demonstrated no interactions between Region and Treat for either drug, indicating the same degree of intergroup differences in changes to YMRS scores among regions after treatment with olanzapine or aripiprazole (Table 3). In addition, after treatment with olanzapine, but not aripiprazole, a significant difference in change of YMRS scores in both treatment groups was detected between the USA and Japan (Table 3 and upper panel of Drug‐F in Figure 8). This trend may be correlated with baseline YMRS scores for both treatment groups (i.e. the highest in the USA and the lowest in Japan; Figure S3).

3.3.3. Changes in Conners' adult attention deficit hyperactivity disorder (ADHD) rating scale‐investigator rated: screening version and adult ADHD quality of life questionnaire scores for ADHD agents

Changes in Conners' adult ADHD rating scale‐investigator rated: screening version (CAARS‐Inv:SV)17 and adult ADHD quality of life questionnaire (AAQOL) scores18 were assessed as endpoints for atomoxetine, an ADHD agent. Baseline AAQOL and CAARS‐Inv:SV scores were identified as influencing factors for changes in CAARS‐Inv:SV and AAQOL scores, respectively (Table 3). The final ANCOVA models showed significant differences between Japan and Taiwan or S. Korea, but no interactions between Region and Treat (Table 3). These results indicate that, although greater CAARS‐Inv:SV scores and lower AAQOL scores, which may have been caused by severe disease phenotype, were observed for both treatment groups in Japan (upper panels in Figure 9, Table 3), the same degree of intergroup differences for both endpoints among regions was demonstrated (lower panels in Figure 9).

4. DISCUSSION

4.1. Study subject and approach

The extent of regional variability of drug responses among East Asians is an important issue. Research on this subject is also expected to help guide l when considering the possible application of a pooled population concept to East Asians, as introduced in the ICH E17 guidelines.6 As many endogenous and exogenous factors affect regional differences in drug responses, it is most desirable to compare the same endpoint using a single MRCT protocol among the target countries. However, because the MRCTs selected in this study were not initially designed to specifically investigate population/regional differences, the number of subjects from each country was small; this may lead to lower statistical power and is a major limitation of this study. Therefore, although we used the most common ANCOVA model approach, our evaluation was based on not only the P‐value but also visual assessment, and other supporting information, such as PK data (from the same MRCT or other global clinical trials), was also considered.

4.2. Antidiabetic agents

The importance of appropriate control of both HbA1c and fasting glucose levels in the prevention of various diabetic complications has been highlighted.19 Our analyses suggested that no overall substantial regional differences in efficacy (intergroup difference) were observed for the 3 antidiabetic drugs, except for differences between S. Korea and Japan with respect to changes in HbA1c level after treatment with insulin glulisine (Figure 2). This observed regional difference may be attributable to a difference in the OHA products or doses administered as a comparator between the 2 countries (data not shown). Notably, no such regional differences were evident in the majority of patients with clinically important higher HbA1c baseline value (≥8.6 mg/dLl; lower right panel of Figure 2). In addition, no significant regional differences in changes in fasting glucose after treatment with insulin glulisine were observed (Figure 4). The exact reason for the differences in results between HbA1c and fasting glucose is not clear; however, this finding may be related to the recently demonstrated nonlinear relationship between HbA1c and fasting glucose after treatment with antidiabetic agents in an agent‐class related manner.19 Considering the similarity of HbA1c changes in higher risk patients (majority subgroup) between S. Korea and Japan as well as fasting glucose change, the observed regional difference in efficacy attributable to insulin glulisine may not be clinically meaningful. In addition, this study suggests that comparator selection should be carefully considered during MRCT planning.

4.3. Respiratory agents

Analysis of respiratory agents showed significant interactions between treatment with indacaterol and specific background factors related to severity of disease (Table 3), suggesting that evaluation of efficacy by stratification using severity of respiratory disease is important. Stratified analysis revealed no substantial regional differences among East Asian countries/regions including subpopulations with severer disease phenotype (lower left in Figure 5 for Drug‐D), although there were few subjects from some subpopulations. These results reveal no clinically meaningful differences in the efficacy of indacaterol between Japan and other East Asian regions/populations. Further, no significant population/regional differences were detected after treatment with glycopyrronium; however, some variability in responses, especially for the number of puffs of rescue medication, was observed in the placebo groups (Drug‐E in Figures 5 and 6). Although the reason for the different responses among the placebo groups is not clear, the trend toward a higher response in the placebo group in Japan may be related to a lower proportion of severe cases at baseline (Figure S1). Thus, consideration of the disease status of subjects may be important for group assignment. The mean number of puffs of rescue medication in all groups at baseline or after drug treatment, or both was higher in S. Korea for both drugs (Figure 6); this was probably attributable to cultural differences, differences in physician instructions, or both among countries.

4.4. Psychotropic agents

No significant population/regional differences in efficacy were observed in psychotropic agents (Figures 7, 8, 9). However, there was some variability in changes in MADRS scores in the placebo group associated with aripiprazole (Figure 7). The reason for this observed regional difference was unclear; however, the relatively lower response in the placebo group in Japan may be related to a higher mean age among the placebo groups (Figure S2). The level of changes in YMRS scores after treatment with olanzapine in both groups was higher in the USA than in Japan; these data may be correlated with differences in the mean baseline YMRS scores (Figure S3). Target indications in terms of bipolar disease phenotypes (depression or mania) were different between olanzapine and aripiprazole; therefore, the regional variability in the baseline values for disease phenotype was larger in the index of nontarget disease phenotype (data not shown). This may have led to a regional variability in the endpoint for nontarget indications. This finding supports the importance of considering baseline alignments for both disease phenotypes among regions when planning for a MRCT of bipolar disease agents. Regarding atomoxetine, a trend of slightly different background values (greater CAARS‐Inv:SV scores and lower AAQOL scores) was observed in both treatment groups in Japan. This observation may be associated with a higher proportion of patients with ADHD‐inattentive type at baseline in Japan than in other regions (Figure S4).

4.5. Other relevant information

The common technical documents or review reports of drug application indicated no meaningful population/regional differences in PK parameters among East Asians in the analysed MRCT or other global clinical trials for all study drugs. Moreover, no clinically meaningful variations in efficacy, at least, between the Japanese subpopulations and total/non‐Japanese population had been estimated in their drug approval. Our analysis of all drugs showed that the mean values of treatment difference in the primary endpoint for each region fell within the variations (95% CIs) of that of the total population, which was in an acceptable range for drug approval.

5. CONCLUSION

Overall, no substantial population/regional differences were observed in 17 efficacy endpoints for eight pharmaceutical products from 3 drug categories among East Asians, except for 1 endpoint (change in HbA1c) for an antidiabetic drug (insulin glulisine). However, no such regional differences were evident in the majority of patients with clinically important higher HbA1c baseline value; thus, the clinical importance of this observed difference is considered to be low. Our study also revealed the potential influencing factors that should be considered in MRCT design: (i) severity of disease at baseline; (ii) class and dosage of the control/concomitant drugs; (iii) possible instruction difference of drug usage in general practice (e.g. the number of puffs of rescue medication); (iv) disease types in complex disorders (e.g. ADHD). Further studies of regional variability in efficacy, including investigation of other drug categories, and the potential factors influencing such regional differences are needed.

COMPETING INTERESTS

This study was a collaboration between the authors and the following pharmaceutical companies: Sanofi K.K. (Tokyo, Japan); AstraZeneca K.K. (Osaka, Japan); Novo Nordisk Pharma Ltd. (Tokyo, Japan); Novartis Pharma K.K. (Tokyo, Japan); Eli Lilly Japan K.K. (Kobe, Japan); and Otsuka Pharmaceutical Co. Ltd. (Tokyo, Japan). All analyses using individual data were conducted in‐house by the company that performed the MRCT according to the protocol prepared for this study. The results were interpreted based on discussions with the companies.

CONTRIBUTORS

Y.S., T.H. and K.S. contributed to the study design. K.S., A.Y. and M.T. primarily contributed to preparation of the protocol and detailed analysis plan. K.S. mainly wrote the manuscript. All authors discussed the design, protocol and interpretation of results, and commented on the manuscript.

Supporting information

Table S1. Registration information of multiregional clinical trials used for analysis on population/regional difference in this study

Figure S1. Comparison of the proportion of severe cases (%) among regions at the starting point of clinical trials for Drug‐E (glycopyrronium). The proportion of severe patients in severity of chronic obstructive pulmonary disease at baseline in the placebo group was lower in Japan than that in other regions, which correlated with a higher response in the placebo group and lower intergroup differences in endpoints attributable to Drug‐E (glycopyrronium) in Japan. Number of subjects in each region is indicated in parentheses.

Figure S2. Comparison of the proportion of patients with dose reduction or dropout (A) and age at the starting point of clinical trials (B) for Drug‐G (aripiprazole) among regions. Age at baseline in the placebo group was higher in Japan than that in other regions, which correlated with different placebo responses for Drug‐G (aripiprazole); however, few regional differences were observed in the proportion of dose reductions and dropouts during testing among placebo groups. Number of subjects in each region is indicated in parentheses. SD: standard deviation

Figure S3. Comparison of baseline Young mania rating scale (YMRS) scores from patients participating in clinical trials for a bipolar disorder agent (Drug‐F: olanzapine) among regions. Mean baseline YMRS scores for both groups were the lowest in Japan and the highest in the USA and correlated with a significant difference in changes to YMRS scores between the USA and Japan. Number of subjects in each region is indicated in parentheses.

Figure S4. Comparison of the proportion of patients with attention‐deficit hyperactivity disorder (ADHD)‐inattentive type at the starting point of clinical trials for an ADHD agent (Drug‐H: atomoxetine) among regions. The proportion of patients with ADHD‐inattentive type at baseline was higher in Japan than that in other regions and correlated with a relatively lower response for both groups in Japan. Number of subjects in each region is indicated in parentheses.

Sai K, Yoshida A, Hanatani T, et al. Population/regional differences in efficacy of 3 drug categories (antidiabetic, respiratory and psychotropic agents) among East Asians: A retrospective study based on multiregional clinical trials. Br J Clin Pharmacol. 2019;85:1270–1282. 10.1111/bcp.13893