|
Dorr 1986 [22]
|
2/39 vs 4/50
|
2/39 vs 4/50
|
2/39 vs 4/50
|
Identical
|
|
Skinner 1989 [17]
|
NR
|
3/80 vs 13/100
|
4/89 vs 12/91
|
As Table 1 above, plus an additional error: BMJ calculates 25% rather than the reported 24% for numerator
|
|
(1-year data)
| | | |
|
Ravikumar 2000 [18]
|
6/91 vs 22/89
|
NR
|
6/89 vs 22/91
|
|
(13-year data)
| | | |
|
Baker 2006 [12]
|
1/40 vs 6/41
|
1/40 vs 3/41
|
1/40 vs 6/41
|
Selection difference:
|
|
BMJ, HTA and Liang [7] all report 1/40 vs 6/41; Cochrane omits 3/41 which were classified as planned or awaiting revision; the event had not taken place but was “planned” only, so was not counted
|
|
Keating 2006 [11]
|
NR
|
NR
|
NR
|
Identical
|
|
Blomfeldt 2007 [9]
|
4/60 vs 3/60
|
1/60 vs 0/60
|
0/60 vs 0/60
|
Selection difference:
|
|
Cochrane 1/60 is a “revision” described by Blomfeldt as a “wound revision”; only revision of implant counts as a revision in the HTA; the BMJ review figures include re-operations both on the contra-lateral side, not related to the implant, and for trauma of the lower limb. [10]
|
|
Macaulay 2008 [23]
|
1/17 vs 0/23
|
1/17 vs 0/23
|
1/17 vs 0/23
|
Identical
|
|
Mouzopoulos 2008 [8]
|
1/43 vs 3/43
|
1/43 vs 5/43
|
1/39 vs 5/38
|
Selection difference:
|
|
Numerators: BMJ “excluded” 2 HA revisions from the analysis (so 3 rather than 5) but this was arguably not justified as the individuals had the outcome of interest and so should have been included;
|
|
Denominators: As Table 2.
|
| |
3/24 errors = 13%
|
2/24 errors = 8%
| |
6 analysed studies = a/B vs c/D = 24 variables
|
|
6/24 selection differences = 25%
|
5/24 selection differences = 21%
|
