€¦ · web viewwe collected information on whether rcts explained the method of ... in each...
TRANSCRIPT
TITLE PAGE
Title
Characteristics of randomized trials focusing on stroke due to intracerebral hemorrhage:
systematic review
Authors
Arina Tamborska BSc,1 Michael T.C. Poon MSc,2 Rustam Al-Shahi Salman PhD.2
1 College of Medicine and Veterinary Medicine, University of Edinburgh, Chancellor's
Building, 49 Little France Crescent, Edinburgh, EH16 4SB, UK
2 Centre for Clinical Brain Sciences, University of Edinburgh, Chancellor's Building, 49 Little
France Crescent, Edinburgh, EH16 4SB, UK
Author for correspondence
Prof. Rustam Al-Shahi Salman, Centre for Clinical Brain Sciences, University of Edinburgh,
Chancellor's Building, 49 Little France Crescent. Edinburgh. EH16 4SB. UK. Phone: +44
(0)1314659602. Fax: +44 (0)1315372944. Email: [email protected]
Cover title
Characteristics of RCTs focusing on ICH
Tables and figures
Tables 4; Figures 2.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
1
Indexing terms
Randomized Controlled Trial; Meta-analysis; Intracerebral Hemorrhage; Bias; Publication
Bias; Sample Size.
Subject terms
Cerebrovascular Disease/Stroke; Intracranial Hemorrhage; Meta Analysis.
Word Count
4664
26
27
28
29
30
31
32
33
34
2
ABSTRACT
Background and purpose: The absence of treatments for intracerebral hemorrhage (ICH)
with significant consistent benefit in randomized controlled trials (RCTs) could be due to
lack of treatment efficacy or the design of RCTs.
Methods: We searched the Cochrane Stroke Group Trials Register in December 2015 for
completed and published RCTs reporting clinical outcome in adults with ICH. We collected
data on publication year and language, study characteristics, and effect size. We regarded
RCTs to be at lower risk of bias if they performed 2 of describing randomization, using
blinding, or specifying the primary outcome. We registered this systematic review
(PROSPERO CRD42016051103).
Results: We found 136 eligible RCTs: 57% were phase II, 76% were single-center, 98%
studied acute treatments, 49% involved drug interventions, 24% were placebo-controlled, the
primary outcome was death or disability in 30%, and median sample size was 77 (inter-
quartile range 47-160). 46% explained randomization, 24% blinded treatment allocation, and
24% specified the primary outcome such that 38 (28%) were at lower risk of bias. RCTs at
lower risk of bias were more likely to use multicenter recruitment (adjusted odds ratio [OR]
6.95, 95% CI 2.2-21.5) and be published in English (adjusted OR 12.9, 95% CI 2.7-62.5).
RCTs with larger sample sizes were independently more likely to be phase III/IV (p<0.01)
and use multicenter recruitment (p<0.01). RCTs at lower risk of bias had smaller pooled
treatment effects on death/disability (p=0.02).
Conclusions: ICH RCTs have often been at high risk of bias, and these RCTs have been
characterized by small sample sizes and larger effect sizes.
1
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
3
INTRODUCTION
Half of the adults affected by stroke due to intracerebral hemorrhage (ICH) are dead within
one year1 and two-thirds of survivors are dependent on others.2 Sadly no interventions for
acute ICH have shown significant consistent benefit in randomized controlled trials (RCTs),3,
4 which could be due to the lack of treatment efficacy or due to RCT design.
Large sample sizes are needed to detect small, but meaningful treatment effects and to
conduct relevant subgroup analyses. However, the rarity and severity of ICH limit
recruitment. Increasing numbers of eligibility criteria reduce recruitment to ICH RCTs,5, 6 but
other determinants of ICH RCT sample size are unknown.
In general, RCTs lacking adequate sequence generation,7, 8 double-blinding,7-9 or specification
of the primary outcome10 can systematically overestimate (bias) effect sizes.7 These
characteristics have not been uncommon among RCTs in acute stroke.11-13 However, the risk
of bias and association with effect size in ICH RCTs is unknown.
Therefore, we systematically searched for ICH RCTs to summarize their characteristics and
investigate independent associations with their sample sizes and effect sizes.
METHODS
This systematic review was prospectively registered in PROSPERO (CRD42016051103) and
is reported according to the PRISMA Statement. No ethical approval was required. We
2
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
4
declare that all supporting data are available within the article and its online supplementary
files.
Eligibility criteria
We included all completed RCTs of interventions in adults who suffered from spontaneous
ICH reporting clinical outcomes (e.g. mortality, disability, or impairment). We included
RCTs of all stroke types, from which data on ICH alone were extractable. We did not apply
time or language restrictions. We excluded studies that were 'quasi-randomized' or not
randomized, and RCTs that did not report clinical outcomes.
Information sources
On 15th December 2015, the Cochrane Stroke Group Managing Editor searched the Cochrane
Stroke Group’s Trial Register for all RCTs coded with the term 'hemorrhagic stroke’ (this
search can be replicated via an online search of the register at http://www.askdoris.org). We
also reviewed authors' personal collections to identify additional RCTs.
Study selection
One reviewer (AT) screened all titles and abstracts retrieved from the search for eligibility.
Another reviewer (MTCP) assessed eligibility of titles and abstracts only available in
Chinese. Potentially eligible RCTs were read in full to determine eligibility, and uncertainties
were resolved by discussion with another reviewer (RA-SS). When a single RCT resulted in
multiple publications, we selected the primary publication for data collection. On 5th July
3
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
5
2016 we checked Medline to see if RCTs, which were ongoing at the time of the initial
search, had been completed and published.
Data collection & items
We developed and piloted a data collection spreadsheet using a subset of eligible RCTs. Two
reviewers used the spreadsheet to extract data from English and Chinese RCTs (AT and
MTCP respectively). A third reviewer (RA-SS) resolved uncertainties or disagreements by
discussion. We collected data on RCT characteristics including: year of publication,
language of publication, phase of trial (II and III/IV, as there were no randomized phase I
trials), sample size, recruitment duration, recruitment region (single center, one-country
multicenter, and multi-country multicenter), type of study (acute treatment, rehabilitation, or
prevention), type of intervention (drug(s), surgery, new system of care, rehabilitation,
traditional Chinese medicine, traditional Chinese procedure, and others), type of control
comparator (placebo, standard of care, active comparator), and time to the last follow-up.
Risk of bias assessment
We collected information on whether RCTs explained the method of randomization, used
blinding (none, single, or double), or specified their primary outcome measure and time-
point. We classified RCTs as being at higher risk of bias if they fulfilled 2 of these criteria:
method of randomization not explained, blinding not used, or primary outcome not specified.
Summary measures
4
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
6
In order to investigate associations with effect sizes, we calculated odds ratios (ORs) for
unfavorable outcome in all included two-arm RCTs. We excluded RCTs if they had more
than two arms or if they did not report sufficient data for the OR calculation. Because some
RCTs used multiple outcome measures without a specified primary outcome, we pre-
specified a hierarchical system for selecting an outcome to analyze, in the following
descending order of preference: death or disability (e.g. modified Rankin Score [mRS],
Glasgow Outcome Scale [GOS-E]); death; disability (e.g. modified Barthel Index); or
impairment (e.g. National Institutes of Health Stroke Scale, Glasgow Coma Scale, or another
neurological deficit scale). If multiple outcomes at one level in the hierarchy were available,
preference was given to the outcome collected after the longest follow-up. In RCTs that
primarily aimed to prevent adverse events (e.g. seizures or deep vein thrombosis), we used
the number of participants experiencing adverse events to calculate the OR. If the outcome of
interest was ordinal, we dichotomized the data for OR calculation. If the outcome of interest
was continuous, we converted the standardized mean difference and its corresponding
standard deviation to OR using an online calculator
(http://www.campbellcollaboration.org/escalc/html/EffectSizeCalculator-Home.php).
Statistical analysis
We compared the characteristics of RCTs using chi-squared tests (or Fisher’s exact test if
covariates had cell counts <5) for nominal independent variables and non-parametric Mann-
Whitney tests for continuous independent variables. We investigated associations of RCTs’
characteristics with risk of bias using multiple logistic regression, and with the sample size
using multiple linear regression. We log-transformed sample size in the linear regression
because of a skewed distribution. We included all RCTs with complete data for each analysis,
5
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
7
so the total number of RCTs varies in each analysis. We pre-specified the inclusion of
recruitment duration in the multivariable analysis of associations with sample size. We used
DerSimonian and Laird random effects models to pool ORs in higher vs. lower risk of bias
RCTs and Wald tests to compare the pooled estimates. We assessed publication bias using a
funnel plot, and plot asymmetry using the Egger test. We performed statistical analyses using
StatsDirect statistical software V2.7.2 and STATA 13.0 (StataCorp). We set a statistical
significance threshold of p<0.05.
RESULTS
Search results
We screened 1,920 records, identified 159 potentially eligible RCTs, and included 136 RCTs
in analyses of associations with risk of bias, sample size, and language of publication (Online
Supplementary Figure http://stroke.ahajournals.org). Of the 136 included RCTs, eight had
multiple treatment arms and three had insufficient data for OR calculation, so 125 RCTs were
available for analyses of associations with effect size.
General characteristics
Of the 136 RCTs published 1961-2016 (Table 1), 80% were published in 2000 or later and
54% were published in English. Most RCTs were phase II (57%) and recruited at a single
center (76%) over a median of two years for a median sample size of 77 participants. 98% of
RCTs evaluated acute treatments for ICH (49% of which were drugs) and most (60%)
compared them to standard care, with a primary outcome measured at a median 90 days that
6
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
8
was death or disability in only 30% of RCTs.
Risk of bias
46% of RCTs described the method of randomization, 24% used some form of blinding, and
24% specified their primary outcome. 38 (28%) of RCTs had at least two of these
characteristics and were classified as being at lower risk of bias. RCTs at lower risk of bias
were significantly more likely to be published in English, published recently, phase III/IV,
multicenter, larger in sample size, placebo-controlled, and use death or disability as the
primary outcome (Table 1). RCTs published in English shared similar associations (Online
Supplementary Table I) because few RCTs at lower risk of bias were published in languages
other than English. All of the 14 multicenter, multi-country RCTs14-27 were at lower risk of
bias.
Characteristics associated with lower risk of bias
We included 136 RCTs in multiple logistic regression analyses, in which English language of
publication and multicenter recruitment were the only characteristics independently
associated with RCTs at lower risk of bias (Table 2).
Characteristics associated with larger sample sizes
ICH RCT sample sizes have increased over time (Figure 1). Only three RCTs have recruited
≥1000 participants16, 22, 25 and seven RCTs have recruited 500-999 participants.14, 18-20, 23, 24, 28 All of
these large RCTs were multicenter and all of them – with one exception28 – were international
7
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
9
collaborations published in English. We included 92 RCTs in multiple linear regression
analyses (after excluding 44 that did not report recruitment duration), in which phase III/IV
and multicenter design were the only characteristics independently associated with a larger
sample size (Table 3). These RCT attributes have become more common over time,
accounting for the apparent secular trend in sample size (Figure 1). Multiple linear regression
models stratified by the phase of the trial confirmed the association of multicenter
recruitment with a larger sample size in phase III/IV but not phase II trials (Online
Supplementary Tables II and III).
Association between RCT quality and effect size
125 RCTs reported clinical outcomes that could be converted to ORs permitting investigation
of association between risk of bias and pooled effect size, stratified by type of outcome or
intervention (Table 4). The 36 (29%) RCTs at lower risk of bias had significantly smaller
pooled effect sizes than RCTs at the high risk of bias, in the two most commonly used
outcome categories (death or disability, and neurological impairment) and regardless of the
type of intervention (Table 4). The results were similar when stratified by the phase of the
trial, with RCTs at a lower risk of bias reporting smaller pooled effect sizes across a range of
outcomes and intervention types (Online Supplementary Tables IV and V).
Small study effects
We assessed small study effects in the 63 RCTs of drugs in which we could estimate the OR,
as this was the largest group of RCTs with the same type of intervention and a similar
number of RCTs in English or another language (Online Supplementary Table I
8
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
10
http://stroke.ahajournals.org). We found evidence of funnel plot asymmetry (Egger test
p<0.01), seemingly driven by RCTs published in languages other than English (Figure 2).
DISCUSSION
In this systematic review of 136 ICH RCTs published over a 55-year period, we found that
only just over one quarter of RCTs were at lower risk of bias. RCTs at lower risk of bias were
more likely to use multicenter recruitment and to be published in English. The median sample
size of ICH RCTs has been only 77; sample size seems to have increased over time, but the
only independent associations with larger sample sizes were phase III/IV RCTs with
multicenter recruitment, and these RCTs have increased in frequency over time. RCTs at
lower risk of bias reported smaller treatment effects. Small phase II RCTs at low risk of bias
with promising treatment effects can be useful; readers may find them in online
supplementary table VI, and may wish to consider confirming the effects of these
interventions in multicenter phase III RCTs.
The strengths of this review are that it included all relevant RCTs in a comprehensive trial
register, without time or language restrictions, and we were able to extract information on a
broad range of RCT characteristics. However, our study has some limitations. Restriction of
our search strategy to ‘hemorrhagic stroke’ RCTs in the Cochrane Stroke Group Trials
Register focused our search on RCTs that were dedicated to ICH, thereby omitting RCTs of
all types of stroke (which may have also included ICH). We did not undertake a full risk of
bias assessment, but focused on three domains that correspond to trial quality29 and have been
used by other similar reviews.7, 11, 12, 30
9
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
11
In previous studies of RCTs of ischemic stroke or acute stroke in general, 74% used
blinding12 and 57% defined the primary outcome.13 Indirect comparison with our findings
suggests that the risk of bias of ICH RCTs has been higher than for stroke in general. The
median ICH RCT sample size is similar to median sample sizes reported for ischemic stroke
RCTs (n=73)12 and acute stroke trials in general (n=80).11 While almost half of the ischemic
stroke trials in the 20th century used multicenter recruitment,12 paradoxically this was used by
only a quarter of RCTs focused on the less frequent ICH subtype of stroke. Others found that
multicenter acute stroke RCTs were of higher quality11 and we have confirmed this for ICH.
Our finding that multicenter RCTs are more likely to achieve larger sample sizes underscores
the many advantages of multicenter multinational RCTs recognized by others.31
We found that 4/62 (6%) RCTs published in languages other than English were at lower risk
of bias, which is slightly better than a systematic review of 9,061 acute stroke RCTs
published in Mainland China, where only 6% used adequate randomization, 2.2% used
blinding and 0.2% defined the primary outcome.30 Limited endorsement of CONSORT
Statement in Chinese neurological journals30 may account for higher risk of bias in Chinese-
language ICH RCTs.
In conclusion, we have found that many ICH RCTs have been small or at risk of bias. Future
research could explore methods to encourage investigators to reduce bias and encourage
collaborators to increase recruitment in ICH RCTs. Small RCTs may miss modest treatment
effects, whilst RCTs at higher risk of bias may overestimate treatment effects. Reducing the
risk of bias and increasing sample size using multicenter recruitment will help future ICH
RCTs; support to do this from funders and regulators would maximize the likelihood of RCTs
finally detecting unequivocal true effect of treatments for ICH.32
10
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
12
ACKNOWLEDGEMENTS
We are very grateful to Hazel Fraser, the Managing Editor of the Cochrane Stroke Group,
who carried out the database search and helped us obtain full manuscripts of selected articles.
SOURCES OF FUNDING
None.
DISCLOSURES
RA-SS received a special project grant from the British Heart Foundation paid to the
University of Edinburgh for the REstart or STop Antithrombotics Randomised Trial.
11
285
286
287
288
289
290
291
292
293
294
295
296
13
REFERENCES
1. van Asch CJ, Luitse MJ, Rinkel GJ, van der Tweel I, Algra A, Klijn CJ. Incidence, case fatality, and functional outcome of intracerebral haemorrhage over time, according to age, sex, and ethnic origin: a systematic review and meta-analysis. The Lancet Neurology. 2010;9(2):167-76.2. Poon MTC, Fonville AF, Salman RA-S. Long-term prognosis after intracerebral haemorrhage: systematic review and meta-analysis. J Neurol Neurosurg Psychiatry. 2013:jnnp-2013-306476.3. Steiner T, Al‐Shahi Salman R, Beer R, Christensen H, Cordonnier C, Csiba L, et al. European Stroke Organisation (ESO) guidelines for the management of spontaneous intracerebral hemorrhage. International journal of stroke. 2014;9(7):840-55.4. Hemphill JC, Greenberg SM, Anderson CS, Becker K, Bendok BR, Cushman M, et al. Guidelines for the management of spontaneous intracerebral hemorrhage. Stroke. 2015;46(7):2032-60.5. Fonville AF, Samarasekera N, Hutchison A, Perry D, Roos YB, Salman RA-S. Eligibility for Randomized Trials of Treatments Specifically for Intracerebral Hemorrhage. Stroke. 2013:STROKEAHA. 113.001493.6. Hansen BM, Ullman N, Norrving B, Hanley DF, Lindgren A. Applicability of clinical trials in an unselected cohort of patients with intracerebral hemorrhage. Stroke. 2016;47(10):2634-7.7. Kjaergard LL, Villumsen J, Gluud C. Reported methodologic quality and discrepancies between large and small randomized trials in meta-analyses. Annals of internal medicine. 2001;135(11):982-9.8. Savović J, Jones HE, Altman DG, Harris RJ, Jüni P, Pildal J, et al. Influence of reported study design characteristics on intervention effect estimates from randomized, controlled trials. Annals of internal medicine. 2012;157(6):429-38.9. Schulz KF, Chalmers I, Hayes RJ, Altman DG. Empirical evidence of bias: dimensions of methodological quality associated with estimates of treatment effects in controlled trials. Jama. 1995;273(5):408-12.10. Chan A-W, Krleža-Jerić K, Schmid I, Altman DG. Outcome reporting bias in randomized trials funded by the Canadian Institutes of Health Research. Canadian Medical Association Journal. 2004;171(7):735-40.11. Bath FJ, Owen VE, Bath PM. Quality of full and final publications reporting acute stroke trials. Stroke. 1998;29(10):2203-10.12. Kidwell CS, Liebeskind DS, Starkman S, Saver JL. Trends in acute ischemic stroke trials through the 20th century. Stroke. 2001;32(6):1349-59.13. Duncan PW, Jorgensen HS, Wade DT. Outcome measures in acute stroke trials. Stroke. 2000;31(6):1429-38.14. Chapman N, Huxley R, Anderson C, Bousser M, Chalmers J, Colman S, et al. Effects of a Perindopril-Based Blood Pressure–Lowering Regimen on the Risk of Recurrent Stroke According to Stroke Subtype and Medical History The PROGRESS Trial. Stroke. 2004;35(1):116-21.15. Mayer SA, Brun NC, Broderick J, Davis S, Diringer MN, Skolnick BE, et al. Safety and feasibility of recombinant Factor VIIa for acute intracerebral hemorrhage. Stroke. 2005;36(1):74-9.16. Mendelow AD, Gregson BA, Fernandes HM, Murray GD, Teasdale GM, Hope DT, et al. Early surgery versus initial conservative treatment in patients with spontaneous supratentorial intracerebral haematomas in the International Surgical Trial in Intracerebral Haemorrhage (STICH): a randomised trial. Lancet. 2005;365(9457):387-97.
12
297
298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345
14
17. Mayer SA, Brun NC, Begtrup K, Broderick J, Davis S, Diringer MN, et al. Recombinant activated Factor VII for acute intracerebral hemorrhage. New England Journal of Medicine. 2005;352(8):777-85.18. Haley EC, Thompson JLP, Levin B, Davis S, Lees KR, Pittman JG, et al. Gavestinel does not improve outcome after acute intracerebral hemorrhage. An analysis from the GAIN International and GAIN Americas studies. Stroke. 2005;36:1006-10.19. Lyden PD, Shuaib A, Lees KR, Davalos A, Davis SM, Diener HC, et al. Safety and tolerability of NXY-059 for acute intracerebral hemorrhage. The CHANT trial. Stroke. 2007;38(8):2262-9.20. Mayer SA, Brun NC, Begtrup K, Broderick J, Davis S, Diringer MN, et al. Efficacy and safety of recombinant activated Factor VII for acute intracerebral hemorrhage. New England Journal of Medicine. 2008;358(20):2127-37.21. Anderson CS, Huang Y, Wang JG, Arima H, Neal B, Peng B, et al. Intensive blood pressure reduction in acute cerebral haemorrhage trial (INTERACT): a randomised pilot trial. Lancet Neurology. 2008;7(5):391-9.22. Anderson CS, Heeley E, Huang Y, Wang J, Stapf C, Delcourt C, et al. Rapid blood-pressure lowering in patients with acute intracerebral hemorrhage. New England Journal of Medicine. 2013;368(25):2355-65.23. Mendelow AD, Gregson BA, Rowan EN, Murray GD, Gholkar A, Mitchell PM, et al. Early surgery versus initial conservative treatment in patients with spontaneous supratentorial lobar intracerebral haematomas (STICH II): a randomised trial. Lancet. 2013;382(9890):397-408.24. Krishnan K, Woodhouse L, Scutt P, Berge E, Sprigg N, Dineen RA, et al. Glyceryl trinitrate for acute intracerebral hemorrhage: results from the efficacy of nitric oxide in stroke (ENOS) trial. International Journal of Stroke. 2015;10(Suppl 2):84.25. Qureshi AI, Palesch YY, Barsan WG, Hanley DF, Hsu CY, Martin RL, et al. Intensive Blood-Pressure Lowering in Patients with Acute Cerebral Hemorrhage. New England Journal of Medicine. 2016.26. Baharoglu MI, Cordonnier C, Salman RA-S, de Gans K, Koopman MM, Brand A, et al. Platelet transfusion versus standard care after acute stroke due to spontaneous cerebral haemorrhage associated with antiplatelet therapy (PATCH): a randomised, open-label, phase 3 trial. The Lancet. 2016.27. Steiner T, Poli S, Griebe M, Hüsing J, Hajda J, Freiberger A, et al. Fresh frozen plasma versus prothrombin complex concentrate in patients with intracranial haemorrhage related to vitamin K antagonists (INCH): a randomised trial. The Lancet Neurology. 2016;15(6):566-73.28. Cheng XC, Wu JS, Zhou XP, et al. The randomized multicentric prospective controlled trial in the standardized treatment of hypertensive intracerebral hematomas: the comparison of surgical therapeutic outcomes with conservative therapy. Chinese Journal of Clinical Neurosciences. 2001;9(4):365-8.29. Moher D, Jones A, Cook DJ, Jadad AR, Moher M, Tugwell P, et al. Does quality of reports of randomised trials affect estimates of intervention efficacy reported in meta-analyses? The Lancet. 1998;352(9128):609-13.30. He S, Wu S, Zeng Q, Zhang S, Lin S, Zhang C, et al. Assessment of methodological quality and outcome measures of acute stroke randomized controlled trials in China in recent 15 years. Journal of evidence-based medicine. 2012;5(3):174-82.31. Djurisic S, Rath A, Gaber S, Garattini S, Bertele V, Ngwabyt S-N, et al. Barriers to the conduct of randomised clinical trials within all disease areas. Trials. 2017;18(1):360.32. Berge E, Salman RA-S, van der Worp HB, Stapf C, Sandercock P, Sprigg N, et al. Increasing value and reducing waste in stroke research. The Lancet Neurology.
13
346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395
15
2017;16(5):399-408.
14
396397
16
FIGURE LEGENDS
Figure 1. Scatter plot of sample size (on a logarithmic scale) versus year of publication for
136 RCTs, stratified by language of publication and distribution of recruiting sites
Figure 2. Funnel plot of 63 RCTs of drug interventions, stratified by language of publication
15
398
399
400
401
17
TABLE 1. Characteristics of 136 included RCTs and comparison of trials at higher
versus lower risk of biasAll 136 RCTs (%)
Higher risk of biasRCTs (n=98)
Lower risk of biasRCTs (n=38)
Statistical significance of higher vs. lower risk of bias RCTs
Randomization explained <0.01Yes 62 (45.6) 27 (27.6) 35 (92.1)No 74 (54.4) 71 (72.5) 3 (7.9)
Blinding <0.01Blinded (single or double) 33 (24.3) 89 (90.8) 14 (36.8)Un-blinded (open) 103 (75.7) 9 (9.2) 24 (63.2)
Primary outcome specified <0.01Yes 33 (24.3) 4 (4.1) 29 (76.3)No 103 (75.7) 94 (95.9) 9 (23.8)
Language of publication <0.01English 74 (54.4) 40 (40.8) 34 (89.5)Other* 62 (45.6) 58 (59.2) 4 (10.5)
Phase of trial 0.03Phase II 77 (56.6) 61 (62.2) 16 (42.1)Phase III or IV 59 (43.4) 37 (37.8) 22 (57.9)
Year of publication <0.01Before 1994 17 (12.5) 16 (94.1) 1 (5.9)
1995-2004 48 (32.3) 45 (93.8) 3 (6.3)2005-2016 71 (52.2) 41 (57.8) 30 (42.3)
Distribution of recruiting sites <0.01Single center 103 (75.7) 90 (87.4) 13 (12.6)Multicenter 33 (24.3) 12 (36.4) 21 (63.6)
Recruitment duration (months)Median (IQR) 24.5 (17-37) 24 (15-36) 25 (20-39) 0.21Trials specified recruitment 92 (67.6) 63 (64.3) 29 (76.3)Unspecified 44 (32.4) 35 (35.7) 9 (23.7)
Follow-up duration (days)Median (IQR) 90 (21-90) 28 (20-90) 90 (90-90) <0.01Trials specified follow-up 127 (93.4) 89 (90.8) 100 (100)Unspecified 9 (6.6) 9 (8.2) 0 (0)
Sample sizeMedian (IQR) 77 (47-160) 72 (47-125) 134.5 (42-404) 0.04
Type of intervention 0.19Acute treatment 133 (97.8) 97 (99.0) 36 (94.7)Rehabilitation 2 (1.5) 1 (1.0) 1 (2.6)Secondary prevention 1 (0.7) 0 (0) 1 (2.6)
Intervention tested 0.22Drugs 67 (49.3) 44 (44.9) 23 (60.5)Surgery 24 (17.6) 20 (20.4) 4 (10.5)New system of care 11 (8.1) 4 (4.1) 7 (18.4)Rehabilitation 3 (2.2) 2 (2.0) 1 (2.6)TCM drug 16 (11.8) 14 (14.3) 2 (5.3)TCM procedure 7 (5.1) 7 (7.1) 0 (0)Others 8 (5.9) 7 (7.14) 1 (2.6)
Comparator <0.01Placebo 33 (24.3) 12 (12.2) 21 (55.3)Standard care 81 (59.6) 66 (6.4) 15 (39.5)Active comparator 22 (16.2) 20 (20.4) 2 (5.3)
Outcome category† <0.01Death or disability 37 (29.6) 18 (20.2) 19 (52.8)Death 29 (23.2) 21 (23.6) 8 (22.2)Disability 11 (8.8) 10 (11.2) 1 (2.78)Neurological impairment 34 (27.2) 30 (33.7) 4 (11.1)Others 14 (11.2) 10 (11.2) 4 (11.1)
*61 trials were published in Chinese and 1 in Russian. † Based on 125 trials that reported an outcome of interest.
16
402
403
18
TABLE 2. Multiple logistic regression analysis of 136 included RCTs, investigating
multivariable, adjusted associations with trials at lower risk of bias
RCTs Number of RCTs at lower risk of
bias (%)
Multiple logistic regression
Odds ratio 95% CI
Language of publication
English 74 32 (43.2) 12.9 2.7-62.5
Other 62 2 (3.2) 1.00 -
Year of publication
Before 1994 17 1 (5.9%) 1.00 -
1995-2004 48 3 (6.3%) 3.87 0.3-47.7
2005-2016 71 30 (42.3) 17.6 1.9-160
Phase of trial
Phase II 77 16 (20.8) 1.00 -
Phase III/IV 59 18 (30.5) 0.65 0.2-1.9
Distribution of recruiting sites
Single center 103 13 (12.6) 1.00 -
Multicenter 33 21 (63.6) 6.95 2.2-21.5
Multiple logistic regression model adjusted for language of publication, year of publication, phase of trial, and the distribution of recruiting sites.
1
404
405
406407
19
TABLE 3. Simple and multiple linear regression models of the associations between the characteristics of 136 RCTs and log (sample size)
RCTs Simple linear regression Multiple linear regression*Coefficient 95% CI Coefficient 95% CI
Phase of trialPhase II 77 ref - ref -Phase III/IV 59 0.91 0.61-1.20 0.85 0.50-1.19
Distribution of recruiting sitesSingle 103 ref - ref -Multicenter 33 1.12 0.79-1.46 0.71 0.26-1.15
RandomizationNot explained 74 ref - ref -Explained 62 0.52 0.20-0.84 0.16 -0.21-0.52
BlindingNot double-blinded 112 ref - ref -Double-blinded 24 -0.09 -0.53-0.35 -0.10 -0.53-0.33
Primary outcome definedNo 103 ref - ref -Yes 33 0.64 0.26-1.01 -0.07 -0.57-0.42
Log (duration of recruitment)Each unit increase 92 0.34 0.02-0.65 0.12 -0.16-0.40
Year of publicationEach year 136 0.03 0.01-0.04 0.004 -0.02-0.03
* Multiple linear regression based on 92 trials that reported the duration of recruitment.
2
408
409
20
TABLE 4. Univariate associations between the risk of bias of 125 RCTs and pooled
estimates of intervention effects
RCTs Pooled OR 95% CI Inconsistency (I2)
Type of outcome Death or disability
All RCTs 37 0.75 0.52-0.98 95.8%Higher risk of bias 18 0.45 0.23-0.67 75.0%Lower risk of bias 19 0.88 0.78-0.98 34.4%
DeathAll RCTs 29 0.41 0.30-0.52 0%Higher risk of bias 21 0.48 0.29-0.68 10.7%Lower risk of bias 8 0.39 0.23-0.54 0%
Disability All RCTs 11 0.09 0.03-0.15 72.4%Higher risk of bias 10 0.08 0.02-0.15 74.0%Lower risk of bias 1 - - -
Neurological impairmentAll RCTs 34 0.12 0.09-0.16 72.3%Higher risk of bias 30 0.09 0.05-0.12 64.3%Lower risk of bias 4 0.36 0.10-0.62 75.0%
Other outcomesAll RCTs 14 0.26 0.11-0.40 22.6%Higher risk of bias 10 0.15 0.02-0.29 0%Lower risk of bias 4 0.42 0.22-0.62 0%
Type of interventionDrug
All RCTs 63 0.32 0.26-0.38 89.6%Higher risk of bias 42 0.10 0.06-0.13 63.6%Lower risk of bias 21 0.65 0.43-0.88 86.6%
SurgicalAll RCTs 22 0.31 0.22-0.39 87.0%Higher risk of bias 18 0.14 0.07-0.21 78.1%Lower risk of bias 4 0.83 0.64-1.02 4.0%
OtherAll RCTs 40 0.38 0.28-0.47 87.6%Higher risk of bias 29 0.16 0.10-0.22 51.2%Lower risk of bias 11 0.69 0.50-0.87 60.9%
3
410
411
412
21
FIGURE 1
Sample size is plotted on a logarithmic scale because of its skewed distribution. Legend refers to language of
publication and distribution of recruiting sites. The trend line represents the regression line of sample size on
year of publication.
4
413
414
415
416
417
22
FIGURE 2
5
418
419
23