主题：【原创】在医学上什么是证据？送给无知的人们 -- 虽远必诛

We assessed the risk of bias of each trial in terms of generation

of allocation sequence, allocation concealment, blinding, uncompleted

data and selective reporting; and classified themas ’lowrisk’,

’moderate risk’, or ’high risk’ according to the guidelines of the

Cochrane Handbook for Systematic Reviews of Interventions 5.0.0 (

Higgins 2008) and as described in Wu 2007a. There was no disagreement

in this process.

Sequence generation

An adequate approach for generating allocation sequence with a

low risk of selection bias should be by using a random numbers table

or computer software, or other simple randomisationmethods,

for example, coin tossing or card shuffling. We considered a trial

which only mentions ’random’ but does not include a description

of the approach used as a moderate risk of selection bias.

Low risk of selection bias: adequate measures to conceal allocation

sequence is defined as the person who generates an allocation

sequence not recruiting the participants, for example, by central

randomisation. Examples of concealing allocation sequences are

using sealed opaque envelopes or storing allocation sequences in

a locked computer.

Moderate risk of selection bias: where concealment of the allocation

sequence is mentioned but the approach used is not reported.

High risk of selection bias: allocation concealment is reported

inadequately, or is not performed at all.

Masking of both the participants and results assessor was considered

as a low risk of performance or/and detection bias.

Single blinding of the results assessor was considered as amoderate

risk of performance or/and detection bias. If single blinding was

performed on the participants but not the results assessor, it was

considered as a high risk of detection bias.

Non-blinding for detection of outcomes includes quality of life

(QoL); adverse events were considered as a high risk of detection

bias. Blinding was not considered necessary for reporting mortality.

We analysed the data using Review Manager 5 (RevMan 2008).

We compared outcome measures for binary data using risk ratios.

For continuous data, we used the mean difference.We conducted

a pooled analysis for two trials only (Chang 2002; Yu 2005).

All 17 included trials used a parallel group design. Four trials (

Chang 2007; Li 1999a; Wang 2008a; Zhang 2008) were multicentre

trials.

A total of 3212 participants were included in the 17 trials, with

numbers of participants in each trial varying from58 to 463. Only

three trials (Chang 2005; Chang 2007; Wang 2008a) mentioned

that the sample size was calculated according to the SFDA’s regulation

about sample size of non-inferiority test studies (that a total

number of 200 participants be included); extra participants were

also included in these trials to avoid possible attrition bias.

Nine trials included children aged from six months to 14 years (

Chen 2004; Li 1998; Liu 2002;Ma 2000; Pan 2000; Song 2004a;

Wang 1998; Yang 2000; Yu 1997). Eight trials included adultsaged from 18 to 65 or 72 years old (Chang 2002; Chang 2005;

Chang 2007; Li 1999a;Wang 2004;Wang 2008a; Yu 2005;Zhang

2008).

Eleven trials included participants according to TCM signs. Nine

trials (Chang 2005; Chang 2007; Chen 2004; Li 1998; Li 1999a;

Ma 2000; Wang 2008a; Yang 2000; Zhang 2008) included patients

with “fever cold”. One trial (Yu 1997) included both “fever

cold” and “chills cold” patients. Three trials (Chang 2002; Wang

2004; Yu 2005) included “chills cold”.Two trials (Pan 2000;Wang

1998) did not sort the patients by TCM signs.

Two trials (Chang 2002; Yu 2005) compared the Chinese herbs

Sanhan Jiere Koufuye and (Fenghan) Biaoshi Ganmao Chongji.

Other trials used different interventions and comparators. These

are listed below:

Yu 1997 comparedCaichenQinreWeixinGuanchangji

with virazole and acetaminophen.

Li 1998 compared Qinwen Keli granule with Kangbingdu

Koufuye.

Wang 1998 compared JianerQinjie YewithQinre Jiedu

Koufuye.

Li 1999a compared Qinkailing injection with Lincomycine.

Ma 2000 compared Shuanghua Penhuji with

Shuanghuanglian Qiwuji.

Pan 2000 compared Xiaoer Reganning with Vitamin C

Yinqiao Chongji.

Yang 2000 compared Huanghu Jiere Daipaoji with

Shiqi Ganmao Daipaoji.

Liu 2002 compared Kangbingdupian with Banlangen

Chongji.

Chen 2004 comparedGegenCenlianweiWanwith Yinqiao

Jiedupian.

Song 2004a compared self-prepared TCM cream with

penicillin or lincomycin.

Wang 2004 compared Sufeng Ganmao Koufuye with

Ganmao Qinre Koufuye.

Chang 2005 compared Jinlian Qinre capsules with Jinlian

Qinre granules; that is the same ingredients in two

different forms. The principle of selecting the control

drug was that its “effect was commonly recognised”. A

double-dummy placebo was used in both arms.

Chang 2007 compared ChaigeQingre granule with Fufang

Shuanghua granule.

Wang 2008a used a double-dummy placebo in both

arms, and compared Yiqing Shuangjie granule and

Chaihuang tablet placebo with Chaihuang tablet, and

compared Yiqing Shuangjie tablet and Caihuang tablet

placebo with Chaihuang tablet.

Zhang 2008 comparedQingyin injectionwithQinkailing

injection

Recovery (expressed as a dichotomous event)was used as a primary

outcome in all trials, except for Pan 2000 and Zhang 2008. Inefficacy

was reported in 15 trials. Fever clearance time was reported in

six trials (Chang 2002; Chang 2005; Li 1998;Wang 1998;Wang

2008a; Yu 1997). The time point at which the temperature started

to abate was reported in four trials (Chang 2005; Li 1998; Wang

1998; Wang 2008a). Six trials (Chang 2002; Chang 2005; Chen

2004; Li 1999a;Wang 2008a; Yu 1997) compared the prevalence

of viral respiratory tract infection by a throat swab culture. Five

trials reported that liver and kidney function tests were carried

out to look for side effects or adverse events (Chang 2002; Chang

2005; Chang 2007; Wang 2008a; Zhang 2008).

Eight trials (Chang 2002; Chang 2005; Chang 2007; Chen

2004; Song 2004a;Wang 2004;Wang 2008a; Yu 2005) evaluated

changes of TCM signs, which included “e feng han” (fear of wind

and cold); “fa re” (fever); “bi sai liu ti” (snuffles and runny nose);

examination of the colour and surface of the tongue; and “mai

xiang” (examining the pulse).

Risk of bias in included studies

Randomisation

Three trials (Chang 2002; Song 2004a; Yang 2000) mentioned

that a random numbers table was used to generate the allocation

sequence. Three trials (Chang 2005; Chang 2007; Wang 2008a)

mentioned that SAS software was used to generate the sequences.

The other 11 trials mentioned “random allocation”, but no description

about which methods were used. After telephoning the

trial authors, we learned that a random numbers table was used

in seven trials (Chen 2004; Li 1998; Liu 2002; Ma 2000; Song

2004a; Yu 1997; Yu 2005). Five trials used computer software (Li

1999a; Pan 2000;Wang 2004; Zhang 2001; Zhang 2008) and one

trial (Wang 1998) used the method of “mo jiu” (drawing straws).

Implications for practice

We could not find a well-designed trial to provide strong evidence

to support

the use of any Chinese traditional herbal medicines for

the treatment of the common cold.