Table 1 Summary of impact of excluding trials not found with brief RCT search.

Lymphoedema of the limbs [20]

4

1. Kasseroller (1996) [21]

Kasseroller study involved in 2 of 17 meta-analyses; exclusion makes no significant difference to outcome of either meta-analysis.

Ventricular pacemakers [22]

34

1. Davis (1985) [23]

Davis study included in 1 of 12 meta-analyses; exclusion makes no significant difference.

Early intervention for psychosis [13]

7

1. Linszen (1998) trial comprised 5 papers; 2 not found with brief search.

a) Linszen (1998a) [24]

b) Linszen (1998b) [25]

Linszen study included in 1 of 12 meta-analyses. Excluding Linszen would mean that there would be no data for this outcome.

Interventions for impetigo [15]

60

1. Arata (1989a) [26]

2. Arata (1989b) [27]

3. Bass (1997) [28]

4. Koranyi (1976) [29]

5. Moraes Barbosa (1986) [30]

6. Park (1993) [31]

7. Pruksachat. (1993) [32]

8. Sutton (1992) [33]

9. Tamayo (1991) [34]

The 9 trials were included in 8 of the 19 meta-analyses in this review.

Outcome 1: excluding Sutton, Bass and Tamayo studies individually made no significant difference. The removal of all 3 together causes meta-analysis to lose statistical significance.

Outcome 2: excluding Bass, Park, Koranyi; no significant change

Outcome 3: excluding Park; no significant change

Outcome 4: excluding Arata 1989a; no significant change

Outcome 5: excluding Pruksachat.; no significant change

Outcome 6: excluding Arata 1989b; only trial providing data to this outcome, so no data now available.

Outcomes 7 & 8: Moraes-Barbosa 1986 is the only trial providing data for both outcomes, so no data now available.

Adherence to treatment in patients with high blood pressure [12]

38

1. Gabriel (1977) [35]

2. Hamilton (1993) [36]

3. Kerr (1985) [37]

4. Morisky (1985) [38]

5. Rehder (1980) [39]

58 different interventions were tested on 15519 patients. Review did not do meta-analysis due to heterogeneity between studies in terms of interventions and the methods used to measure adherence. Missing trials were all small studies of poor methodological quality. Their exclusion would have little impact on the final conclusions.

Preventing infection in nephrotic syndrome [14]

5

1. Dou (2000) [40]

2. Li (2000) [41]

3. Zhang (2000) [42]

4. Tong (1998) [43]

The 4 missing trials were included in 3 of 4 of the meta-analyses.

Outcome 1: Excluding Dou and Tong; no significant change.

Outcome 2: Tong was the only trial providing data, so no data now available.

Outcome 3: Zhang was the only trial providing data, so no data now available.

Outcome 4: Li was the only trial providing data, so no data now available.

Probiotics for treating infectious diarrhoea [17]

25

Sugita (1994) [44]

Sugita included in 7 of 27 meta-analyses in review. Exclusion of Sugita makes no significant difference to any of the clinical outcomes. The only outcome to lose statistical significance is not a clinical outcome, but a sensitivity analysis based on a methodological characteristic (blinding).

Psychological interventions for coronary heart disease [16]

56

1.Gallacher (1997) [45]

2. Mitsibounas (1992) [46]

1. Gallacher was included in 8 of 28 meta-analyses. In six, omission makes no difference.

In 2, confidence intervals are much wider; in the one of these statistical significance is lost. 2. Mitsibounas was included in 5 of 28 meta-analyses. Excluding Mitsibounas makes no significant difference in four outcomes. In the other, it cause causes the meta-analysis to lose statistical significance

Insulin analogues versus human insulin [47]

43

Iwamoto (2001) [48]

Iwamoto in 4 of the 8 meta-analyses. Exclusion of Iwamoto makes no significant difference to any of the outcomes.

Tramadol for neuropathic pain [49]

5

Leppert (2001) [50]

Leppert not used in either of 2 meta-analyses in the review. The review says: 'Given the small (even if unknown) number of subjects and the non-blinded nature of the trial it is probably not possible to draw and conclusions about their relative efficacy from this study.