Table 2 Mortality data at 1 year reported for THA (n/N) vs HA (n/N) in 3 reviews

Dorr 1986 [22]

3/39 vs 4/50

NR

NA (event data for each arm NR)

Error:

Dorr reports 7 deaths across both arms; BMJ review categorises this as 3 and 4 in each arm [10]

Skinner 1989 [17]

21/91 vs 24/89

18/80 vs 27/100

20/89 vs 25/91

Error and Selection difference:

(1-year data)

As Table 1 for these data

Ravikumar 2000 [18]

(13-year data)

Baker 2006 [12]

NR

NR

NR

NA

Keating 2006 [11]

4/69 vs 11/111

4/69 vs 6/69

4*/69 vs 6/69

Selection difference:

As Table 1

Blomfeldt 2007 [9]

4/60 vs 3/60

4/60 vs 3/60

4/60 vs 3/60

Identical

Macaulay 2008 [23]

1/17 vs 5/23

NR

NR

Selection difference:

BMJ analyses reported 6-month data as 1 year data; Cochrane, HTA and Liang [7] all omit it from the 1-year analysis; Cochrane analysis uses these data for 6-month follow-up only. It is unclear when the additional deaths reported for up to 2 years (4/17 vs 7/23) occurred. They may have occurred in the 6–12 month period.

Mouzopoulos 2008 [8]

10/43 vs 13/43

6/43 vs 6/43

6/39 vs 6/38

Selection difference:

Denominators: HTA applied an intention-to-treat analysis, excluding 4 from THA arm (2 data lost; 2 not satisfy inclusion criteria) and 5 from HA arm (none satisfied trial’s inclusion criteria), as they did not satisfy the study’s inclusion criteria and no follow-up data were collected. [13] If these are included, then explicit best and worst case analyses perhaps should have been performed, with imputations explained [14]

4/24 errors = 17%

2/16 errors = 13%

4-6 analysed studies = a/B vs c/D = 16 or 24 variables:

10/24 selection differences = 42%

4/16 selection differences = 25%