Thursday, March 12, 2015

The Figures Used in the Meta-Analyses

In this post I’m going to report the figures that the meta-analyses have used for the trials.  The format is that I’ll mention some figures, followed by which ones each of the meta-analyses have used.

I also included three other meta-analyses in this post:

1.      Truswell (1994)

Some things to remember:

  • Mozaffarian and Ramsden only reported the RR and total number of events/deaths, which I used to calculate the number of events/deaths in each group
  • While Mozaffarian pooled the results for CHD mortality and total mortality, they didn’t report the RRs or the number of deaths for each trial in their paper.  The figures I used came from personal communication
  • Chowdhury didn’t do any analysis for CHD mortality or total mortality; Truswell didn’t do any analysis on CHD mortality; and Harcombe didn’t do any analysis on CHD events
  • Hooper and Schwingshackl both used CVD events and CVD mortality as their outcome measures rather than CHD events and CHD mortality, so their figures are often inconsistent with other meta-analyses
  • Reporting data from FMHS and MCS in person years is more appropriate given the nature of those trials.  Mozaffarian was the only one to do this, but only for mortality in FMHS and didn’t always use deaths per age-adjusted person years, which is important as FMHS is an inadequately randomised trial so differences in age are more likely to occur and did happen
  • Truswell often made errors in calculating the RR due to ignoring differences in group sizes
  • Skeaff also often made errors in calculating the RR, though I’m sure not why this is the case
  • E = number of CHD/CVD events (multiple events in one participant is counted each time).  P = number of participants who have had CHD/CVD events (multiple events in one participant is counted once)



Rose Corn Oil Trial


Experimental
Control
Relative Risk
Major CHD Events (E&P) [1]
12
6
1.86
Total CHD Events [1]
15
11
1.27
CHD/CVD/Total Mortality [1]
5
1
4.64
* E=P for major events because once participants experienced a major CHD event they were removed from the trial

Exp N = 28
Con N = 26
CHD Events
CHD Mortality
Total Mortality
Exp
Con
RR
Exp
Con
RR
Exp
Con
RR
Skeaff & Miller
15
11
1.17






Mozaffarian, et al









Ramsden, et al
16
8
1.86
5
1
4.64
5
1
4.64
Hooper, et al
26
11
1.14
8
1
3.85
8
1
3.85
Chowdhury, et al









Truswell
12
6
2.00



5
1
4.71
Schwingshackl
26
11
1.14
8
1
3.85
8
1
3.85
Harcombe, et al



5
1
4.64
5
1
4.64

  • Skeaff didn’t include data on mortality for some reason
  • Mozaffarian excluded RCOT because of “multiple interventions”, but included STARS even though STARS had three groups too
  • Hooper and Schwingshackl combined the corn oil and olive oil groups (so the Exp N = 54) because the purpose of both meta-analyses was to ‘fat modification’ rather than simply replacing SFA with PUFA
  • Truswell listed the total number of participants as 52 rather than 54
  • Chowdhury excluded RCOT because it had less than 50 CHD events, but included STARS even though by their interpretation STARS only had 7 events (RCOT had 18-26 depending on interpretation)



Los Angeles Veterans Administration Trial


Experimental
Control
Relative Risk
Major CHD Events (E) [1] [2] [3] [4]
60
78
0.766
Major CHD Events (P) [1] [2] [3] [4]*
52/53
65/71
0.796/0.743
Total CHD Events (E) [1] [4]
89
105
0.843
Major CVD Events (E) [1] [2] [3] [4]
85
119
0.711
Major CVD Events (P) [1] [2] [3] [4]
66
96
0.684
CHD Mortality [1] [2] [3] [4]**
41/42
50/51
0.816/0.820
CVD Mortality [1] [2] [3] [4]
48
70
0.682
Total Mortality****
174
177
0.978
* The lower number is the correct figure for the number of participants with CHD events, while the higher number counts participants who had two different major CHD events twice (for example see table 1 in [2]).
** The figure for CHD mortality varies between papers.  It’s reported as 41 vs. 50 [1] [2] and 42 vs. 51 [3] [4], due to a later recategorisation of one death in each group from ‘miscellaneous’ to ‘acute myocardial infarct’
*** The figure for total mortality is a bit ambiguous.  The original papers reported it as being 174 vs. 177 [1], 177 vs. 174 [3] and 174 vs. 178 [5].  Also, the graph in [2] suggests total mortality is higher in the experimental group, while the graph in [4] suggests it’s lower.  I’m using the figure from the monograph ([1]), which is also the figure used most often in the meta-analyses

Exp N = 424
Con N = 422
CHD Events
CHD Mortality
Total Mortality
Exp
Con
RR
Exp
Con
RR
Exp
Con
RR
Skeaff & Miller
52
65
0.82



174
177
0.98
Mozaffarian, et al
53
71
0.74
42
51
0.82
170
174
0.97
Ramsden, et al
61
79
0.77
42
51
0.82
174
178
0.97
Hooper, et al
97
122
0.79
57
81
0.70
174
177
0.98
Chowdhury, et al
53
71
0.74






Truswell
60
78
0.77



174
177
0.98
Schwingshackl









Harcombe, et al



41
50
0.82
174
177
0.98

·         CHD events in Ramsden seems to be the number of major CHD events, but has a minor mistake as they report the number of events as 140 rather than 138
·         CVD events in Hooper seems to be major CVD events (P) plus a combination of soft CVD events (which may include (based on the numbers): possible angina, definite angina, possible MI by ECG, possible overt MI, possible stroke and new aneurysm) [1] [4]
·         Skeaff didn’t include data on CHD mortality for some reason
·         CVD mortality in Hooper comes from deaths from acute CVD events (48 vs. 70) and atherosclerotic complications (9 vs. 11), where atherosclerotic complications is often not the sole cause of death [3]
·         Unsure where total mortality in Mozaffarian comes from.  It may be an error or due to excluding certain causes of death given that their total mortality in FMHS excludes ‘Accidents, poisonings and violence’.  Also, the LAVAT studies that were referenced in Mozaffarian do not provide numbers for total mortality, only graphs (which contradict each other)
·         Schwingshackl excluded LAVAT because it was not a pure secondary prevention trial

Medical Research Council Trial


Experimental
Control
Relative Risk
Major CHD Events (P) [1]
45
51
0.860
Total CHD Events (P) [1]
62
74
0.817
CHD Mortality [1]
25
25
0.975
CVD Mortality [1]
27
25
1.053
Total Mortality [1]
28
31
0.881

Exp N = 199
Con N = 194
CHD Events
CHD Mortality
Total Mortality
Exp
Con
RR
Exp
Con
RR
Exp
Con
RR
Skeaff & Miller
62
74
0.86
15
14
1.04
27
25
1.05
Mozaffarian, et al
45
51
0.86
25
25
0.97
28
31
0.88
Ramsden, et al
45
51
0.86
25
25
0.97
28
31
0.88
Hooper, et al
62
74
0.82
27
25
1.05
28
31
0.88
Chowdhury, et al
45
51
0.86






Truswell
45
51
0.88



28
31
0.88
Schwingshackl
62
74
0.82
27
25
1.05
28
31
0.88
Harcombe, et al



27
25
1.05
28
31
0.88

·         CHD mortality in Skeaff only includes fatal first relapses (table 5 [2]) and not the total number of fatal CHD events
·         Total mortality in Skeaff only includes CVD mortality and non-CVD mortality (1 cancer death in the experimental group and 6 cancer deaths in control group; table 5 [2])
For CHD events Skeaff lists the Con N as 794, which is clearly a typo

Oslo Diet Heart Study


Experimental
Control
Relative Risk
Major CHD Events (E) [1]
70
91
0.769
Major CHD Events (P) [1] [2]*
56/61
70/81
0.800/0.753
Total CHD Events (E) [1]
80
120
0.667
Total CHD Events (P) [1]
64
90
0.711
CHD Mortality [1] [2]
37
50
0.740
CVD Mortality [2]
38
52
0.731
Total Mortality [2]
41
55
0.745
* The lower number is the correct figure for the number of participants with CHD events, while the higher number counts participants who had two different major CHD events twice

Exp N = 206
Con N = 206
CHD Events
CHD Mortality
Total Mortality
Exp
Con
RR
Exp
Con
RR
Exp
Con
RR
Skeaff & Miller
34
54
0.68
10
23
0.46
41
55
0.79
Mozaffarian, et al
61
81
0.75
37
50
0.74
41
55
0.75
Ramsden, et al
61
81
0.75
37
50
0.74
41
55
0.75
Hooper, et al
64
90
0.71
38
520.73
48
65
0.74
Chowdhury, et al
34
54
0.56






Truswell
79
94
0.84



101
108
0.94
Schwingshackl
61
81
0.75
38
52
0.73
41
55
0.75
Harcombe, et al



79
94
0.84
101
108
0.94


·         CHD events in Skeaff and Chowdhury only includes the number of participants with myocardial infarction (table 1 [1])
·         CHD events in Truswell and CHD mortality in Harcombe came from CHD mortality in the 11 year report (table 5 [2])
·         CHD mortality in Skeaff only includes fatal myocardial infarction (table 1 [1] and table 2 [2]), but not sudden death
Unsure about total mortality in Hooper
·         Total mortality in Truswell and Harcombe came from the 11 year report (table 5 [2])

Sydney Diet Heart Study


Experimental
Control
Relative Risk
CHD Events (E&P) [1]*
36
24
1.609
CHD Mortality [1]**
36
24
1.609
CVD Mortality [1]**
38
26
1.567
Total Mortality [2]
39
28
1.494
* CHD events are made up of: (1) CHD mortality (sudden death and fatal myocardial infarction); (2) non-fatal myocardial infarction; and (3) non-major CHD events such as angina.  Therefore in SDHS, where CHD events wasn’t reported, CHD mortality could be used as the CHD events, not sure whether this ‘correct’
** Calculated from the percentages reported in [1]

Exp N = 221
Con N = 237
CHD Events
CHD Mortality
Total Mortality
Exp
Con
RR
Exp
Con
RR
Exp
Con
RR
Skeaff & Miller









Mozaffarian, et al









Ramsden, et al






39
28
1.49
Hooper, et al






39
28
1.49
Chowdhury, et al
35
23
1.74






Truswell
35
25




39
28
1.49
Schwingshackl
35
22
1.71
35
22
1.71
39
28
1.49
Harcombe, et al



35
25
1.50
39
28
1.49

·         Skeaff excluded SDHS because “they reported only cardiovascular disease and not CHD endpoints”.  This isn’t correct as the original paper reported the total CHD, CVD and total mortality, but did not provide a breakdown by groups for CHD and CVD mortality nor did they report CHD events (that weren’t fatal).  They still could have included the trial in their analysis of total mortality as 60/67 deaths were due to CHD
·         Mozaffarian excluded SDHS because of “Non-CHD endpoint”, which was correct at the time, but they still could have included the trial in their analysis of total mortality as 60/67 deaths were due to CHD
·         Unsure about CHD events and mortality in Chowdhury and Schwingshack.  Both of these meta-analyses were published after the recovered data on CHD and CVD mortality.  I calculated CHD and CVD mortality from the percentages.  The original paper also reports that there were a total of 60 deaths from CHD and another 3 from stroke which is greater than what either of them reported (Chowdhury = 58, Schwingshackl = 57)
·         CHD events in Truswell and Harcombe came from Truswell’s assumption of “60/67 deaths from CHD) say 35/25”, which isn’t an unreasonable assumption as the RR remained very similar, but Harcombe should have used CHD mortality from [2]
·         At the time of the Hooper and Ramsden meta-analyses the data for CHD and CVD mortality hadn’t been recovered

Finnish Mental Hospital Study (Men)


Experimental
Control
Relative Risk*
Major CHD Events (E&P) [1]**
8
26
0.330
Total CHD Events [1]***
25
47
0.556
CHD Mortality [2]
34
76
0.469
CVD Mortality [2]
62
104
0.608
Total Mortality [2]
188
217
0.882
* Relative risk is calculated using person years
** Includes coronary death and major ECG change
*** Includes coronary death, major ECG change and intermediate ECG change

Exp N = 444/2276
Con N = 478/1902
CHD Events
CHD Mortality
Total Mortality
Exp
Con
RR
Exp
Con
RR
Exp
Con
RR
Skeaff & Miller
8
26
0.34
6
12
0.54
187
217
0.74
Mozaffarian, et al
25
47
0.55
34
76
0.49
175
199
0.95
Ramsden, et al









Hooper, et al









Chowdhury, et al
?
?
?






Truswell









Schwingshackl









Harcombe, et al










·         The RR for CHD events in Mozaffarian came from pooling the numbers together rather than averaging the hospitals
·         Chowdhury only had one entry for FMHS, which seems to refer to the analysis in women (27 vs. 46)
·         CHD mortality in Skeaff only refers to the deaths from CHD among men aged 34-64 “whose initial electrocardiogram was free from coronary patterns”.  In addition the RR came from pooling the numbers together rather than averaging the hospitals
·         The RR for CHD mortality in Mozaffarian comes from using non-age adjusted person years
·         Total mortality in Skeaff has a minor mistake and the RR comes came from pooling the numbers together rather than averaging the hospitals and using age adjusted person years
·         Total mortality in Mozaffarian comes from ‘all diseases’ and excludes deaths from ‘accidents, poisonings and violence’.  The age-adjusted RR using person years is = 0.890 and non-age adjusted RR using person years is 0.944, so this may be a minor error [2]
·         Ramsden, Truswell and Harcombe all excluded FMHS because it was not randomised.  Hooper excluded FMHS because it was cluster randomised with < 6 clusters and used a crossover design.  Schwingshackl excluded FMHS because it was not a pure secondary prevention trial

Finnish Mental Hospital Study (Women)


Experimental
Control
Relative Risk*
Major CHD Events (E&P) [1]
3
8
0.393
Total CHD Events (E&P) [1]
27
46
0.635
CHD Mortality [2]
73
129
0.659
CVD Mortality [2]
156
204
0.859
Total Mortality [2]
415
465
1.064
* Relative risk is calculated using person years
** Includes coronary death and major ECG change
*** Includes coronary death, major ECG change and intermediate ECG change

Exp N = 372/3598
Con N = 341/2836
CHD Events
CHD Mortality
Total Mortality
Exp
Con
RR
Exp
Con
RR
Exp
Con
RR
Skeaff & Miller
3
8
0.35
3
3
0.92
415
465
0.73
Mozaffarian, et al
27
46
0.64
73
123
0.68
404
450
1.06
Ramsden, et al









Hooper, et al









Chowdhury, et al
27
46
0.59






Truswell









Schwingshackl









Harcombe, et al










·         The RR for CHD events in Skeaff and Mozaffarian came from pooling the numbers together rather than averaging the hospitals
·         For CHD events Chowdhury reported a total of 612 participants in the experimental group and 610 participants in the control group, which made the RR different ((27/612)/(46/610)=0.585).  I don’t know where those figures came from
·         CHD mortality in Skeaff only refers to the deaths from CHD among men aged 34-64 “whose initial electrocardiogram was free from coronary patterns”  In addition the RR came from pooling the numbers together rather than averaging the hospitals
·         The RR for CHD mortality in Mozaffarian comes from non-age adjusted person years
·         Total mortality in Skeaff comes came from pooling the numbers together rather than averaging the hospitals
·         Total mortality in Mozaffarian comes from ‘all diseases’ and excludes deaths from ‘accidents, poisonings and violence’ [2].  They reported the correct RR for total mortality using age adjusted person years, which is isn’t very different to all diseases  (1.064 vs. 1.068)
·         Ramsden, Truswell and Harcombe all excluded FMHS because it was not randomised.  Hooper excluded FMHS because it was cluster randomised with < 6 clusters and used a crossover design.  Schwingshackl excluded FMHS because it was not a pure secondary prevention trial

Minnesota Coronary Survey


Experimental
Control
Relative Risk*
Major CHD Events (E) [1]
131
121
1.058
Major CVD Events (P)**
134
129
1.016
CHD Mortality*** [1]
61
54
1.105
CHD Mortality + Stroke Deaths [1]
83
75
1.082
CVD Mortality [1]
157
157
0.978
Total Mortality [1]
269
248
1.061
* Relative risk is calculated using person years
** Comes from the third paper listed here
*** CHD and CVD mortality isn’t reported but can be calculated using table 7 [1].  CHD mortality = ‘arteriosclerotic heart disease’ + ‘cardiac arrest, heart block’.  (table 7 [1])
**** Not reported but can be calculated from deaths due to ‘arteriosclerotic heart disease’ down to ‘pulmonary embolism’ (depending on your criteria) (table 7 [1])
***** CVD mortality exceeds CVD events because CVD only includes stroke

Exp N = 4541
Con N = 4516
CHD Events
CHD Mortality
Total Mortality
Exp
Con
RR
Exp
Con
RR
Exp
Con
RR
Skeaff & Miller
131
121
1.07



269
248
1.07
Mozaffarian, et al
131
121
1.08
61
54
1.12
269
248
1.08
Ramsden, et al
131
121
1.08
61
54
1.12
269
248
1.08
Hooper, et al
134
129
1.03
157
147
1.06
269
248
1.08
Chowdhury, et al
131
121
1.08






Truswell
131
121
1.08



269
248
1.08
Schwingshackl









Harcombe, et al










·         Skeaff didn’t report CHD mortality (you do have to calculate it, but it’s not hard)
·         Hooper used counted 10 fewer CVD deaths among women in the control group
·         Harcombe, et al excluded MCS because the results were not published prior to 1983
·         Schwingshackl excluded MCS because it was not a pure secondary prevention trial

Diet and Reinfarction Trial


Experimental
Control
Relative Risk
Major CHD Events [1]
132
144
0.914
CHD Mortality [1]
97
97
0.997
CVD Mortality [2]
101
100
1.007
Total Mortality [1]*
111
113
0.979*
* Adjusting for confounding variables increased the RR for total mortality to 1.00 [1]

Exp N = 1018
Con N = 1015
CHD Events
CHD Mortality
Total Mortality
Exp
Con
RR
Exp
Con
RR
Exp
Con
RR
Skeaff & Miller
132
144
0.92
97
97
1.00
111
113
0.98
Mozaffarian, et al
132
144
0.91
97
97
1.00
111
113
0.98
Ramsden, et al









Hooper, et al
136
147
0.92
101
100
1.01
111
113
0.98
Chowdhury, et al
132
144
0.91






Truswell
132
144
0.92



111
113
0.98
Schwingshackl
132
144
0.91
97
97
1.00
111
113
0.98
Harcombe, et al










·         Ramsden excluded DART because, there isn’t sufficient dietary information to calculate omega 3 and 6 intakes in DART, which is understandable since the aim of their meta-analysis was to compare SFA vs. omega 6 trials with SFA vs. omega 3+6 trials
·         CVD events and CVD mortality in Hooper also includes stroke deaths which was reported in [2]
·         Harcombe, et al excluded DART because the results were not published prior to 1983

St Thomas Atherosclerosis Trial


Experimental
Control
Relative Risk
Major CHD Events (E) [1]
2
5
0.415
Total CHD Events (E)* [1]
3
9
0.346
Total CVD Events (E)* [1]
3
10
0.311
CHD/CVD/Total Mortality [1]
1
3
0.346
* Includes ‘coronary surgery’ and ‘angioplasty’, but not angina

Exp N = 27
Con N = 28
CHD Events
CHD Mortality
Total Mortality
Exp
Con
RR
Exp
Con
RR
Exp
Con
RR
Skeaff & Miller
3
9
0.41
1
3
0.37
1
3
0.37
Mozaffarian, et al
2
5
0.41
1
3
0.35
1
3
0.35
Ramsden, et al
2
5
0.42
1
3
0.35
1
3
0.35
Hooper, et al
8
20
0.41
1
3
0.35
1
3
0.35
Chowdhury, et al
2
5
0.41






Truswell
3
10
0.31






Schwingshackl
3
10
0.31
1
3
0.35
1
3
0.35
Harcombe, et al










·         CHD events in Hooper comes from total CVD events and ‘requiring increased anti-anginal treatment’ (5 vs. 10, page 567 [1])
·         Harcombe, et al excluded STARS because the results were not published prior to 1983
·         Truswell included STARS among the ‘Secondary Dietary Prevention Trials: Plasma Cholesterol Lowering with Angiographic End Points’ and only reported CHD events and not CHD/total mortality

No comments:

Post a Comment