Background Image
Table of Contents Table of Contents
Previous Page  24 / 48 Next Page
Information
Show Menu
Previous Page 24 / 48 Next Page
Page Background

REVIEW

SA JOURNAL OF DIABETES & VASCULAR DISEASE

66

VOLUME 12 NUMBER 2 • NOVEMBER 2015

Figure 6. B:

Forest plots of ethnicity subgroups.

C:

RCT or non-RCT subgroups.

B

SES

BMS

Odds ratio

Odds ratio

Study or subgroup

Events

Total

Events

Total

Weight (%)

M–H, random, (95% CI)

M–H, random, (95% CI)

1.1.1 European Subgroup

Aoki J,

et al.

27

112

37

118

18.5

0.70 (0.39–1.24)

Baumgart D,

et al.

15

94

38

96

17.2

0.29 (0.15–0.58)

Maresta A,

et al.

15

68

28

70

16.4

0.42 (0.20–0.90)

Subtotal (95% CI) 274 284

52.1

0.45 (0.27–0.77)

Total events

57

103

Heterogeneity: Tau

2

= 0.10, Chi

2

= 3.71, df = 2 (

p

= 0.16);

I

2

= 46%

Test for overall effect: Z = 2.95 (

p

= 0.003)

1.1.2 American and Asian

Chan C,

et al.

8

54

12

29

12.8

0.25 (0.09–0.71)

Daemen J,

et al.

44

206

54

252

20.1

1.00 (0.64–1.56)

Jimenez-Quevedo P,

et al.

8

80

31

80

15.0

0.18 (0.07–0.41)

Subtotal (95% CI) 340 361

47.9

0.37 (0.11–1.27)

Total events

60

97

Heterogeneity: Tau

2

= 1.02, Chi

2

= 15.55, df = 2 (

p

= 0.0004);

I

2

= 87%

Test for overall effect: Z = 1.58 (

p

= 0.11)

Total (95% CI) 614 645

100.0

0.42 (0.24–0.74)

Total events

117

200

Heterogeneity: Tau

2

= 0.36, Chi

2

= 20.14, df = 5 (

p

= 0.001); I

2

= 75%

Test for overall effect: Z = 3.00 (

p

= 0.003)

Test for subgroup differences: Chi

2

= 0.09, df = 1 (

p

= 0.77);

I

2

= 0%

C

SES

BMS

Odds ratio

Odds ratio

Study or subgroup

Events

Total

Events

Total

Weight (%)

M–H, random, (95% CI)

M–H, random, (95% CI)

5.1.1 RCT

Baumgart D,

et al.

15

94

38

96

17.2

0.29 (0.15–0.58)

Chan C,

et al.

8

54

12

29

12.8

0.25 (0.09–0.71)

Jimenez-Quevedo P,

et al.

8

80

31

80

15.0

0.18 (0.07–0.41)

Maresta A,

et al.

15

68

28

70

16.4

0.42 (0.20–0.90)

Subtotal (95% CI) 296 275

61.4

0.28 (0.19–0.42)

Total events

46

109

Heterogeneity: Tau

2

= 0.00, Chi

2

= 2.40, df = 3 (

p

= 0.49);

I

2

= 0%

Test for overall effect: Z = 6.14 (

p

= 0.00001)

5.1.2 Non-RCT

Aoki J,

et al.

27

112

37

118

18.5

0.70 (0.39–1.24)

Daemen J,

et al.

44

206

54

252

20.1

1.00 (0.64–1.56)

Subtotal (95% CI) 318 370

38.6

0.87 (0.61–1.24)

Total events

71

91

Heterogeneity: Tau

2

= 0.00, Chi

2

= 0.92, df = 1 (

p

= 0.34);

I

2

= 0%

Test for overall effect: Z = 0.76 (

p

= 0.45)

Total (95% CI) 614 645

100.0

0.42 (0.24–0.74)

Total events

117

200

Heterogeneity: Tau

2

= 0.36, Chi

2

= 20.14, df = 5 (

p

= 0.001);

I

2

= 75%

Test for overall effect: Z = 3.00 (

p

= 0.003)

Test for subgroup differences: Chi

2

= 16.81, df = 1 (

p

< 0.0010);

I

2

= 94.1%

Heterogeneity is one major concern with regard to the

validity of meta-analyses.

26,34

Non-homogeneous data can easily

give misleading results. In our study, the

Q

and

I

2

statistics were

performed to test heterogeneity. For all samples, there was

significant heterogeneity for major adverse cardiac events in the

SES and BMS groups.

We further conducted subgroup analysis according to sample

size, ethnicity and studymethod. It demonstrated that in the studies

where sample size was ≤ 90, method was a RCT and population

was European, the overall major cardiac events were significantly

different between the SES and BMS groups. Heterogeneity

between the studies was decreased after stratifying the samples.

No significant heterogeneity was observed with RCTs, suggesting

an RCT is important for good results. More high-quality RCTs are

therefore warranted.

Another concern for meta-analyses is publication bias, due

to selection of the studies included. In this study, using funnel

plots and Egger’s test,

28,35,36

we found publication bias for overall

major cardiac events, target-lesion revascularisations and

myocardial infarction, but not for overall mortality. Furthermore,

the sensitivity analysis confirmed there was no change if one

study was removed at a time. Although more studies would

have produced better results, overall, our results were statistically

reliable.