1
00:00:00,180 --> 00:00:07,440
I'm Richard. I'm one of the team running the course this week where money one of the modules on this program,

2
00:00:07,530 --> 00:00:12,749
which is a central medical statistics, about half the audience of students from that module.

3
00:00:12,750 --> 00:00:25,520
Whenever we run a module, we try to have a seminar speaker who is relevant but diverting and off syllabus, something that's fun, Balaban examined.

4
00:00:26,610 --> 00:00:32,700
And to be honest, when you want to talk about statistics, you have to try really hard not to invite Martin Landau.

5
00:00:34,110 --> 00:00:38,370
Last time we spoke, it was about the life expectancy of left handed cricketers.

6
00:00:40,620 --> 00:00:42,389
And we're pleased to have Martin back.

7
00:00:42,390 --> 00:00:49,440
And just this morning, our students were studying plant opening plots or planned plots, depending where your loyalties lie.

8
00:00:49,440 --> 00:00:56,310
But either way, it's one of the most highly cited papers as the most highly cited paper in The Lancet ever.

9
00:00:56,460 --> 00:01:00,120
Yeah. Yeah. You're probably sick of hearing people mentioned that it's one of the 30 most

10
00:01:00,120 --> 00:01:04,199
highly cited papers ever of any kind on anything anywhere in the universe.

11
00:01:04,200 --> 00:01:07,290
Now, you got that right.

12
00:01:08,130 --> 00:01:12,030
Not not as highly cited as David Cox. Yeah.

13
00:01:12,450 --> 00:01:16,349
Okay. Well, thank you for coming. But if you make it stop. Okay.

14
00:01:16,350 --> 00:01:22,380
Thank you very much. All right. Well, I'm going to talk today about diabetes, blood sugar and red wine.

15
00:01:22,710 --> 00:01:26,280
This is a personal study and you'll see why it's a personal study.

16
00:01:26,490 --> 00:01:30,810
All the data we're not only collected by me, but provided by me as well.

17
00:01:31,540 --> 00:01:40,470
It was published in significance the the magazine of the Royal Statistical Society, which I recommend highly.

18
00:01:40,800 --> 00:01:44,670
It's very interesting and very publication.

19
00:01:44,940 --> 00:01:53,250
And I found it to be that a place where I can publish things that no respectable journal is going to take but are often,

20
00:01:53,640 --> 00:01:57,630
you know, much more illuminating than what I get into the Lancet.

21
00:01:58,050 --> 00:02:02,910
So I'm an emeritus professor, so that means they're not paying me any more.

22
00:02:03,210 --> 00:02:07,050
I just do it for fun. So we'll see if we can have some.

23
00:02:08,160 --> 00:02:11,160
Okay. Now, this is all about me.

24
00:02:12,390 --> 00:02:20,640
I have diabetes, I have type two diabetes, and I was diagnosed around the beginning of 2001.

25
00:02:21,330 --> 00:02:33,030
It was a shock because I expect to be told that that was what it was, but not a great surprise because I am the descendant of many type two diabetics.

26
00:02:33,360 --> 00:02:39,570
Both my mother and father had Type two diabetes. Both my grandmothers had Type two diabetes.

27
00:02:39,870 --> 00:02:48,419
My grandfathers, we don't know one died as a young man in an accident and the other died in the trenches in the First World War.

28
00:02:48,420 --> 00:02:57,959
So we don't know about them. Several other members of my family have had type two diabetes or currently have it over.

29
00:02:57,960 --> 00:03:02,190
My two sisters, one has Type two diabetes, the other doesn't.

30
00:03:02,520 --> 00:03:09,360
She has hypothyroidism, which an endocrinologist assured me is just a different expression of the same genes.

31
00:03:09,750 --> 00:03:22,110
So, you know, we're full of it. And one of the things I had to I was asked to do a course of my new status as a diabetic was to make my watch,

32
00:03:22,140 --> 00:03:29,490
to measure my blood glucose every morning, fasting blood glucose, that is before I've had breakfast.

33
00:03:30,060 --> 00:03:34,710
And you do this with a little finger prick to raise a drop of blood.

34
00:03:35,040 --> 00:03:44,969
And that is a picture. I do all my own stunts, that's my blood and that's my metre and that was my glucose, but not fasting.

35
00:03:44,970 --> 00:03:58,140
Of course I did this specially for the presentation and the and so a number appears that and it's supposed to be below.

36
00:03:58,260 --> 00:04:01,890
Well, currently, I'll tell you, I'll come to it now. I love data.

37
00:04:02,010 --> 00:04:14,309
So I began to I began to enter this into a computer and I began to analyse them and watch the graph falling as the blood glucose fell.

38
00:04:14,310 --> 00:04:24,750
Because I was taking the medication and I was watching my diet and not taking things which I was told would deadly, which really surprised me.

39
00:04:24,750 --> 00:04:27,930
The most deadly thing I had in my diet was orange juice.

40
00:04:28,860 --> 00:04:38,460
I thought it was supposed to be good for me, but apparently it's terrible because it's full of sugar and my blood glucose came under control.

41
00:04:38,670 --> 00:04:46,920
The initial measurement by my GP was 24 millimetres per litre and my first fasting blood glucose that I made

42
00:04:47,130 --> 00:04:53,490
measured before breakfast was 19 milli miles per litre and the target was less than 7 million gallons per litre.

43
00:04:53,490 --> 00:04:59,940
So I was quite a long way to go, but I made it and I eventually reached that mark.

44
00:05:00,290 --> 00:05:08,360
Seven. And then for several years my fasting blood glucose remained comfortably below this limit and I stopped worrying about it.

45
00:05:08,660 --> 00:05:13,910
I was so use by then too, this diet without sugar and so on.

46
00:05:14,570 --> 00:05:18,979
And yeah, you know, everything was going very smoothly.

47
00:05:18,980 --> 00:05:26,030
Nothing interesting was happening and I stopped recording it. Oh, what do I do then?

48
00:05:27,050 --> 00:05:32,340
Hit the wrong key. Right. There's a second measurement used to monitor diabetes.

49
00:05:32,360 --> 00:05:36,370
HPI 1c0 glycosylated haemoglobin.

50
00:05:36,740 --> 00:05:40,800
And this requires a much larger sample of blood.

51
00:05:40,820 --> 00:05:49,550
They stick a needle in your vein, suck it out and send it to a lab and you get the result a few days later.

52
00:05:49,730 --> 00:05:51,290
And so that's always done as a clinic.

53
00:05:51,290 --> 00:05:59,599
You can't do it at home and it measures the glucose exposure over the past few weeks rather than over the past few hours,

54
00:05:59,600 --> 00:06:03,500
which is what the blue blood glucose measurement does.

55
00:06:03,950 --> 00:06:10,220
And it's the measurement that doctors prefer to use to monitor diabetes.

56
00:06:13,770 --> 00:06:17,700
Now the method of measuring blood glucose changed.

57
00:06:17,970 --> 00:06:29,490
It changed from the concentration of glucose in the blood to the concentration in serum that is in the liquid component of the blood.

58
00:06:29,610 --> 00:06:37,560
And glucose is highly soluble, as I'm sure you all know, and it is nearly all in the liquid.

59
00:06:37,770 --> 00:06:41,550
So when they changed the ace, it went up.

60
00:06:42,040 --> 00:06:49,140
I didn't notice that it had been changed. I noticed that the colour of the plastic strips I put into my little metre had changed,

61
00:06:49,650 --> 00:06:57,299
but my glucose went up and I thought, that's funny, I haven't changed my diet or anything.

62
00:06:57,300 --> 00:07:00,660
I was still taking all the drugs. Why has it gone up?

63
00:07:01,380 --> 00:07:10,680
And then my wife said, You haven't looked at the leaflet in your in this little box because if you do, you'll see that it's changed.

64
00:07:11,100 --> 00:07:15,630
And so I realised, yes, of course it's gone up because the way of measuring it has gone up,

65
00:07:15,780 --> 00:07:23,610
not because the actual thing has gone up and the upper limit was now around eight months.

66
00:07:23,790 --> 00:07:34,110
Sorry. Yeah. So eight miles per litre and the units of measurement to HPI well see also changed from a percentage to barely moles per litre.

67
00:07:34,290 --> 00:07:44,890
I went to see my GP and she said to me, your HPI one C is 75 and I said, I guess 75.

68
00:07:44,910 --> 00:07:48,930
She said, What did you think it was going to be? I said, 6.5%.

69
00:07:50,010 --> 00:07:54,960
She said, Oh yes, we've changed it to millimetres per litre. So many miles per litre it became.

70
00:07:55,350 --> 00:07:59,219
And so that was that was all right.

71
00:07:59,220 --> 00:08:05,520
But it was still a shock because 75 millimetres per litre was a very high value, she said.

72
00:08:05,520 --> 00:08:12,540
It should be 59 or less. So I was out of control again and I knew it had gone higher.

73
00:08:12,540 --> 00:08:17,730
But you know what, with the change of units and so I haven't really appreciated it very much.

74
00:08:18,150 --> 00:08:22,110
And so I thought, right, you've got to do something about this.

75
00:08:23,160 --> 00:08:26,700
If you don't, they'll put you on insulin, you'll have to inject yourself.

76
00:08:26,970 --> 00:08:34,890
And I didn't fancy that. I thought, I'll leave that to I'm older. So I decided to go for a more rigorous diet instead.

77
00:08:35,160 --> 00:08:39,480
Not only would avoid sugar, I would cut out most of the carbohydrate.

78
00:08:39,490 --> 00:08:44,010
So usually now one piece of bread in the morning is okay.

79
00:08:44,340 --> 00:08:50,340
That's it. That's all I have. Just to show you what a health freak I am.

80
00:08:50,550 --> 00:08:58,290
I bake my own bread without any salt because I also have very high blood pressure, although the drugs keep that down to something really good.

81
00:09:00,090 --> 00:09:03,419
So I thought I'll also. Yeah, no more potatoes.

82
00:09:03,420 --> 00:09:08,159
I'll try harder to avoid beer and wine. I won't cut them out altogether.

83
00:09:08,160 --> 00:09:12,780
That is asking too much, but they are vehicles for carbohydrate and sugar.

84
00:09:12,990 --> 00:09:15,990
And so when I drink alcohol, I'll drink red wine.

85
00:09:16,740 --> 00:09:22,170
And I also resume recording of the daily fasting glucose to monitor the effects.

86
00:09:22,950 --> 00:09:31,080
Oh, I've done it again there. And this is what it looks like over several years, day since the 1st of January 2012.

87
00:09:31,350 --> 00:09:39,120
And you can see that at first it goes down in a very good and encouraging way and you can see that's going down over that first year,

88
00:09:39,810 --> 00:09:49,500
down below that new limits of eight. And then the next year it went up again, hovering around it and then below and it chops and changes.

89
00:09:49,500 --> 00:09:57,060
It goes up and down and occasionally that shoots the go shoots up like that.

90
00:09:57,690 --> 00:10:02,579
I know what that was. That was at the International Biometrics Conference in Florence.

91
00:10:02,580 --> 00:10:16,170
Believe it or not, we went to a 2 to 1 of the restaurants that a tasting restaurant, and it was tasting enormous amounts of pasta.

92
00:10:16,490 --> 00:10:21,030
It was a nightmare. Oh, okay.

93
00:10:21,300 --> 00:10:25,560
So it rises and falls, occasional spikes.

94
00:10:26,670 --> 00:10:30,330
The medication was increased. Was Del changed at several points.

95
00:10:30,720 --> 00:10:37,110
And as you can see, the HCA one see rises and falls with the fasting glucose, as you might expect.

96
00:10:37,530 --> 00:10:46,710
Yeah. So I thought, oh it might be interesting to see whether I can predict the HPI one C from the fasting glucose.

97
00:10:47,610 --> 00:10:52,139
You know, I don't know what my HPI one C is, but maybe I can predict it.

98
00:10:52,140 --> 00:11:01,620
I am a statistician after all. So I thought, well the pre, you know, they said the preceding few weeks, how about six weeks.

99
00:11:01,770 --> 00:11:05,339
I said right 42 days and it worked.

100
00:11:05,340 --> 00:11:09,299
Then I tried a longer period, tried 60 days and it worked better.

101
00:11:09,300 --> 00:11:12,540
So that's what I use now. And there is a predictive.

102
00:11:12,620 --> 00:11:20,879
Shin of my blood glucose to predict my HPI one C and you can see that quite a lot of measurements in

103
00:11:20,880 --> 00:11:27,000
there and that's the regression line and the confidence interval for the prediction and the one in read

104
00:11:27,000 --> 00:11:36,809
the 59 that is the value of HPI one C predicted for the 16th of August 2017 when I had blood taken and I

105
00:11:36,810 --> 00:11:41,130
had to arrange it because I was going off on some conference visit and I thought it's going to go up,

106
00:11:41,430 --> 00:11:49,230
so I'll get it measured before that happens. And, and the measured value is 59.

107
00:11:49,380 --> 00:11:54,750
So I thought that was fantastic, you know, prediction and measurement coinciding.

108
00:11:54,750 --> 00:12:02,320
Exactly. My wife, who features again in this story, Shira,

109
00:12:02,340 --> 00:12:09,400
one day that she noticed that my blood glucose seemed to be a bit lower on days after I'd been drinking.

110
00:12:10,740 --> 00:12:15,480
And so I thought, yes, you might be right there.

111
00:12:15,870 --> 00:12:23,909
So I thought, well, I will add a record of my wine drinking to my daily glucose record and a prospective data

112
00:12:23,910 --> 00:12:30,210
collection to test the hypothesis that red wine was associated with reduced blood glucose.

113
00:12:30,480 --> 00:12:33,629
And I thought if this works out well, that be fantastic.

114
00:12:33,630 --> 00:12:36,720
You know, I can have a drink and think I'm doing myself good.

115
00:12:38,340 --> 00:12:42,749
I also recorded an other alcohol variable, which might be white wine, beer,

116
00:12:42,750 --> 00:12:48,000
whisky, whatever, because I thought that these might increase blood glucose.

117
00:12:48,470 --> 00:12:55,620
Uh, yeah. Sherry gets counted as other alcohol because it's white wine.

118
00:12:56,000 --> 00:13:01,650
The pork gets counted as red wine because that's what it is.

119
00:13:02,910 --> 00:13:12,120
And so I thought all that alcohol might increase blood glucose because white wine and beer are vehicles for sugar and carbohydrate.

120
00:13:14,580 --> 00:13:20,879
So I planned statistical analysis with I would do a two sample tea method for red wine days versus

121
00:13:20,880 --> 00:13:26,910
no red wine days and a multiple regression of glucose on red wine and other alcohol jointly.

122
00:13:27,510 --> 00:13:34,230
And the hypotheses for that were that the effect of red wine and other alcohol would be opposite,

123
00:13:34,500 --> 00:13:42,480
and also the alcohol consumption would not be independent of red wine because on social occasions I might well drink both.

124
00:13:43,560 --> 00:13:53,000
So I decided to write an article about this in 2017, March the 27th.

125
00:13:53,010 --> 00:14:02,940
It was I know this because it was at the meeting to celebrate the 20th anniversary of the Centre for Statistics in Medicine.

126
00:14:03,450 --> 00:14:10,140
And at the dinner, I was sitting next to the editor of the BMJ and we were chatting.

127
00:14:10,280 --> 00:14:17,160
I told her about this and we agreed that it might make an interesting article for the Christmas BMJ.

128
00:14:18,210 --> 00:14:25,500
And I was very happy about that.

129
00:14:26,310 --> 00:14:36,750
And on the morning after this dinner, I had a very high fasting blood glucose of 6 to 9.0 millimetres per litre.

130
00:14:36,960 --> 00:14:42,030
And that was because I'd been drinking red wine and sparkling white wine and lots of food.

131
00:14:43,470 --> 00:14:52,920
And now I'd been monitoring the comparison between my red wine days and other days, you know, just doing the occasional t test.

132
00:14:53,310 --> 00:14:57,299
And this day increased the running value for my significance test.

133
00:14:57,300 --> 00:15:05,850
And I thought, Right, well, we know that repeated significance testing is wrong, but I'd only be doing it for my own interests, so that didn't matter.

134
00:15:06,300 --> 00:15:13,770
But if I was going to publish this, I would need to set a date for a formal closing of the data set and not look at them again.

135
00:15:14,220 --> 00:15:25,360
So that's what I did. And I didn't want to invalidate the whole thing by choosing the final date just because the results fit the theory.

136
00:15:25,380 --> 00:15:29,910
I know that there are people who do this, but I hope none of them are statisticians.

137
00:15:30,960 --> 00:15:39,300
So I chose to use June the 30th because that would allow time to prepare an article for the September deadline for the Christmas BMJ.

138
00:15:39,450 --> 00:15:49,620
This is how real science works. And so that now you go.

139
00:15:50,880 --> 00:15:58,060
That's June the 30th, 2017, going from the 12th of December when I first started doing this.

140
00:15:58,080 --> 00:16:01,710
So, you know, it's nearly 200 days.

141
00:16:01,980 --> 00:16:09,060
And as you can see, the difference is small but apparent.

142
00:16:09,990 --> 00:16:19,170
It is statistically significant. The confidence interval is -5.51, 2.06 many miles per litre.

143
00:16:19,650 --> 00:16:25,500
And I thought that that was good evidence for a relationship between red wine and glucose.

144
00:16:28,350 --> 00:16:36,360
Normal distribution with uniform variance followed quite closely, apart from a couple of outliers of one rather dramatic outlier.

145
00:16:36,570 --> 00:16:42,030
That was the. Obviously another wild night wasn't the one in Florence, but a different one.

146
00:16:44,580 --> 00:16:51,570
And and curiously, on a day when there was no wine, I don't know how that came about.

147
00:16:52,340 --> 00:16:56,640
And the sample is quite large, you know, nearly 200 people.

148
00:16:57,270 --> 00:17:00,900
Equal sized groups. And I know you want to know.

149
00:17:01,140 --> 00:17:05,450
And I drank red wine on slightly more than 50% of the day.

150
00:17:06,900 --> 00:17:15,660
And that so that seemed perfectly valid, except that the assumption of independence then, of course,

151
00:17:16,470 --> 00:17:20,910
the analysis ignores the fact that the glucose measurements from day to day are related.

152
00:17:21,210 --> 00:17:25,380
A high glucose day tends to be followed by another high glucose day.

153
00:17:25,590 --> 00:17:28,730
You saw the way that thing goes up and down. Stuff.

154
00:17:29,430 --> 00:17:42,410
But so the analysis is only approximate ignores the auto correlation and there are the sequential P values.

155
00:17:42,420 --> 00:17:45,660
I thought that would be interesting. As you can see, when I started to do it,

156
00:17:47,310 --> 00:17:54,719
the P value was not significant because there were only a few observations and it fell and fell very satisfactorily.

157
00:17:54,720 --> 00:18:04,620
That's why I decided to write the article and then it went up again and for a few days it was above the red line and then it went down again.

158
00:18:05,490 --> 00:18:15,110
And if there were a genuine difference, we would expect the p value to tend to get progressively smaller as data were added, which it does.

159
00:18:15,110 --> 00:18:24,180
So I think that that's fine. If there were no true difference, we might expect the p value to go up and down in a random way,

160
00:18:24,240 --> 00:18:33,330
except bearing in mind that each p value is linked to the one before, so they would sort of zigzag rather than be just scattered.

161
00:18:37,030 --> 00:18:46,030
The test was statistically significant when I decided to write the article on day 87, but around day 140 it ceased to be so.

162
00:18:46,180 --> 00:18:50,260
But it became significant again before my predetermined analysis.

163
00:18:50,500 --> 00:18:55,540
That's very important. My pre-determined analysis pie and the trend was down with.

164
00:18:55,570 --> 00:19:00,260
I illustrates quite well how dangerous repeated significance testing is.

165
00:19:00,280 --> 00:19:03,290
Don't do it a thing, you know.

166
00:19:03,850 --> 00:19:07,900
If you keep doing it and then pick one of them out, it can be highly misleading.

167
00:19:08,170 --> 00:19:11,650
We should decide what we're going to do and when.

168
00:19:12,130 --> 00:19:19,540
Before we start now, I didn't in this case, but I decided well before I decided when I was going to write the paper.

169
00:19:21,700 --> 00:19:27,010
We must not keep adding data and testing it until we get the results we want.

170
00:19:28,030 --> 00:19:36,519
But the overall evidence to support the red wine hypothesis, the other alcohol that was associated with red wine,

171
00:19:36,520 --> 00:19:46,239
is that it will be being consumed now 14.1% of red wine days and 5.4% of days without red wine and the multiple

172
00:19:46,240 --> 00:19:55,070
regression estimate so that the estimate for red wine is -0.31 miles per litre on days following wine.

173
00:19:55,360 --> 00:19:58,719
And that was highly significant piece point nought, nought.

174
00:19:58,720 --> 00:20:05,710
And I'm very close to the point, not one that I had for the T test and the confidence interval is also very similar.

175
00:20:06,370 --> 00:20:12,130
So very similar results, the just slightly more significant, slightly narrower interval.

176
00:20:12,550 --> 00:20:16,690
And the estimate for the other alcohol was plus .26.

177
00:20:16,690 --> 00:20:24,130
So it went up, but it wasn't significant. The confidence interval goes from minus point 1 to 2 plus 0.64.

178
00:20:24,460 --> 00:20:29,590
Hence the other alcohol was associated with an increase in blood glucose, though not significant.

179
00:20:31,360 --> 00:20:35,200
And there are the serial P's for that. As you can see, it's quite weird.

180
00:20:35,530 --> 00:20:44,230
For the red wine it goes, it bubbles about falls and stays down below the limit all the time.

181
00:20:45,220 --> 00:20:50,860
For the other alcohol it goes down is much more dramatic, but then it goes up again.

182
00:20:51,430 --> 00:20:54,310
So why that should be? I have absolutely no idea.

183
00:20:54,760 --> 00:21:03,640
And it had an effect very early on, but that significance disappears towards the end of the period, saying it may be spurious.

184
00:21:04,090 --> 00:21:09,459
However, again, it illustrates how misleading these significance tests are.

185
00:21:09,460 --> 00:21:20,380
Repeated significance tests are well supports in my theory that other alcohol would increase blood glucose, but it just didn't prove it.

186
00:21:20,830 --> 00:21:26,110
The evidence crumbled as I continued, and it's not refuted by the non-significant fact.

187
00:21:26,890 --> 00:21:36,610
The data are consistent with an increase of 2.64 millimetres per litre with no difference and consistent with a decrease of .12 millimetres per litre.

188
00:21:39,160 --> 00:21:44,799
More data in the future may cast further light on this. Now there are a few caveats here.

189
00:21:44,800 --> 00:21:48,160
This is an observational study, not an experiment.

190
00:21:48,190 --> 00:21:52,480
We must be very cautious about inferring cause and effect.

191
00:21:52,810 --> 00:21:56,790
Drinking red wine is not random, okay.

192
00:21:56,860 --> 00:22:01,630
Tends to happen. Weekends tends to happen when eating out, for example.

193
00:22:01,990 --> 00:22:07,390
So my wife and I had a night out on Tuesday and we shared a bottle of red wine.

194
00:22:07,690 --> 00:22:10,810
And how can I put this? We didn't waste any of it.

195
00:22:13,040 --> 00:22:16,800
I'd have guessed that eating out would, if anything, increase glucose.

196
00:22:16,810 --> 00:22:20,860
But this is pure, pure speculation. It did. That particular one certainly did.

197
00:22:21,760 --> 00:22:25,419
We've also noticed that my glucose appears to be higher after I've eaten.

198
00:22:25,420 --> 00:22:32,440
Correct. Why that should be? I have no idea, even though I usually do this without rice or other carbohydrates.

199
00:22:32,440 --> 00:22:40,750
But because curry contains a number of pharmacologically active substances such as turmeric and who knows what they do?

200
00:22:41,590 --> 00:22:45,580
Well, I'm sure many people do know what they do. I don't know what they do.

201
00:22:49,060 --> 00:22:53,080
And Curry implies no alcohol consumption. If I'm out. How often?

202
00:22:53,080 --> 00:23:00,040
A beer if I'm out. Now the details of food are not recorded.

203
00:23:00,060 --> 00:23:08,160
All I can be said is this rather weak evidence of an association between blood glucose and red wine consumption on the previous day,

204
00:23:08,280 --> 00:23:14,040
even one person with diabetes. So I'll have a look at the literature.

205
00:23:14,640 --> 00:23:21,780
If you're going to send this paper off. You know, having done all this, you better say you find out has anybody else done it?

206
00:23:22,650 --> 00:23:29,220
So I did a search and I looked at red wine of fasting blood glucose.

207
00:23:29,670 --> 00:23:37,770
And here is a systematic review by the Oxford team, I should say.

208
00:23:37,770 --> 00:23:41,880
Geoff Aronson commented on this article for me very kindly.

209
00:23:43,020 --> 00:23:51,270
And they reported a systematic review of randomised trials of the effect of alcohol intake upon glucose.

210
00:23:51,660 --> 00:23:56,400
They reported no overall evidence of an effect of alcohol on blue blood glucose,

211
00:23:56,640 --> 00:24:00,900
but did not distinguish between red wine and other sources of alcohol.

212
00:24:01,440 --> 00:24:11,640
Gatineau as well managed to randomise 224 people with diabetes who usually abstain from alcohol to a daily drink of red wine,

213
00:24:12,000 --> 00:24:17,430
white wine or mineral water. You know, I mean, how do you get people to agree to this?

214
00:24:18,170 --> 00:24:23,760
Yeah, it's amazing. Managed to retain 87% of them for two years.

215
00:24:23,970 --> 00:24:31,020
Whether they stuck to this for two years, I don't know. But they allowed the measurements to be used.

216
00:24:31,260 --> 00:24:41,370
They reported significantly lower fasting plasma glucose with white wine compared with water and a small and non-significant reduction with red wine.

217
00:24:41,700 --> 00:24:44,970
But there do appear to be problems with the design of this study.

218
00:24:45,150 --> 00:24:48,900
As Dirk pointed out to me, he published on this.

219
00:24:48,930 --> 00:24:53,130
Why don't I read all these articles he was implying? But it was a letter.

220
00:24:53,370 --> 00:24:59,070
And he said the actual allocation was pretty ropey. And when I looked at it in detail, I agreed with him.

221
00:24:59,400 --> 00:25:10,370
So that's a dodgy study. Shares how randomised 109 people with type two diabetes to red wine or alcohol free diet beer for three months.

222
00:25:10,380 --> 00:25:17,160
Why? I don't know. But they got a significant reduction in fasting blood glucose for the red wine

223
00:25:17,160 --> 00:25:23,040
group compared to the beer group codeine as how they had a crossover trial.

224
00:25:23,040 --> 00:25:27,569
And I randomised 20 women to drink 190 miles of red wine.

225
00:25:27,570 --> 00:25:36,560
That's a large loss on five days per week for ten weeks and abstain from alcohol for ten weeks, you know.

226
00:25:37,700 --> 00:25:44,969
Yeah. Oh I just give people drugs. They reported that mean a nice day.

227
00:25:44,970 --> 00:25:54,330
Fasting glucose was 91.1 and 9.2 milligrams per decilitre at baseline 91.6

228
00:25:54,330 --> 00:26:00,389
following red wine and 88.5 following abstention with no significant differences.

229
00:26:00,390 --> 00:26:05,130
So they certainly didn't agree with me, but they disagree.

230
00:26:06,090 --> 00:26:10,100
And an uncontrolled study of 42 subjects setting out.

231
00:26:10,100 --> 00:26:17,520
After all, this is the hardest paper to get hold of reported that after daily consumption of 25 miles of red wine for two weeks,

232
00:26:17,850 --> 00:26:23,129
blood glucose increased slightly. And this one I like Ginette.

233
00:26:23,130 --> 00:26:25,980
How nominative determinism in action.

234
00:26:26,970 --> 00:26:34,770
Jen has reported the amount which glucose rose after a meal taken with red wine was less than when compared to water.

235
00:26:35,220 --> 00:26:40,200
A similar result was found for tannin taken after the meal, but no difference for alcohol.

236
00:26:40,200 --> 00:26:48,470
Which was very interesting because it suggests they might be the tannin rather than the alcohol in the red wine, which may have an effect.

237
00:26:48,480 --> 00:26:54,330
So it's not the wine this, but the redness. So that's an interesting idea.

238
00:26:54,570 --> 00:26:58,050
But they didn't report fasting blue glucose, so we don't know about that.

239
00:27:00,600 --> 00:27:05,850
I could find no report of the effect of a single dose of red wine on fasting glucose.

240
00:27:06,240 --> 00:27:14,160
I came across this paper by holds that are reporting that regular alcohol consumption decrease the risk of developing type

241
00:27:14,160 --> 00:27:22,230
two diabetes and the lowest risk it was estimated to occur at 14 drinks per week for men and nine drinks per week for women.

242
00:27:22,920 --> 00:27:24,989
They didn't distinguish red wine consumption.

243
00:27:24,990 --> 00:27:33,720
While it's very nice that their optimum figures coincide so well with the Department for Health recommendations for what you should be drinking.

244
00:27:33,960 --> 00:27:39,330
So, yeah, so when people tell you one drink is enough to kill you, save would be might stop me dying of diabetes.

245
00:27:39,570 --> 00:27:44,340
So that would be a good thing. You know, every I know that we I will die.

246
00:27:44,550 --> 00:27:48,840
You know that will happen. Yeah. So what next?

247
00:27:48,840 --> 00:27:53,729
I thought. Would it be feasible to do it? Oh, yeah, but before I thought about what next.

248
00:27:53,730 --> 00:28:01,590
Oh, right. Okay, so you've reviewed the. You've done the analysis, you've collected the data, it's SEP, send it to the BMJ.

249
00:28:01,600 --> 00:28:06,490
So I sent it to the BMJ and they turned it down.

250
00:28:06,850 --> 00:28:15,489
It's not scientific enough. They said I was not scientific about it, but they didn't tell me that.

251
00:28:15,490 --> 00:28:20,470
It just said it wasn't scientific enough. So I thought, Well, I'm not going to worry.

252
00:28:21,160 --> 00:28:32,400
I met the editor of the BMJ later at a very sad occasion and she she said, Well, how about that red wine paper?

253
00:28:32,410 --> 00:28:38,170
I said, Oh, I sent it to significance and they published it. She said, Oh, yeah, I thought you'd have something like that up your sleeve.

254
00:28:38,440 --> 00:28:41,620
So it's all right. So everybody's happy. We're all friends.

255
00:28:42,550 --> 00:28:46,330
So what could we do? How could we do a study? Could we do a trial?

256
00:28:46,660 --> 00:28:51,310
I thought, well, in principle, I could randomise myself to drink on some days, but not others.

257
00:28:51,610 --> 00:28:55,200
But I think this will be unlikely to work. I don't think my wife would like it.

258
00:28:55,210 --> 00:29:01,360
For example, I have a social life and I'm honestly not that dedicated to this particular search for knowledge.

259
00:29:01,690 --> 00:29:07,530
And if we said, Oh yeah, we're going out to dinner, but this is one of the days I'm randomised not to drink, you know,

260
00:29:07,540 --> 00:29:14,469
now it's not going to work and I thought I'd be even less likely to be able to randomise others to drink or not drink and give

261
00:29:14,470 --> 00:29:22,570
a day's probably much easier to randomise them to drink every day for a given month as the various researchers have done.

262
00:29:23,080 --> 00:29:30,520
Perhaps there's an animal model, I don't know. But like most nutritional risk factors, red wine and glucose is very hard to study.

263
00:29:30,790 --> 00:29:35,470
It's really difficult to study nutritional risk factors apart from anything else.

264
00:29:35,680 --> 00:29:40,600
If you drink red wine, you're not drinking something else. And so this is always the problem.

265
00:29:40,600 --> 00:29:45,850
It's like saying nuts are good for you. Yeah, if you eat lots of nuts, you're not eating that much REM steak.

266
00:29:46,120 --> 00:29:51,190
So, you know, it's very difficult to pin down one factor.

267
00:29:51,490 --> 00:29:57,309
Oh, I've done it again. Now, the study was prospective.

268
00:29:57,310 --> 00:30:01,180
It appears to demonstrate an association, but the sample is rather small.

269
00:30:01,180 --> 00:30:04,210
Just one person can't really get smaller.

270
00:30:04,570 --> 00:30:14,530
And even if this relationship were causal and not associated with other dietary things, you know, might be just an idiosyncratic reaction.

271
00:30:14,530 --> 00:30:19,720
It might be just me or a particular type of type two diabetes I have,

272
00:30:19,960 --> 00:30:27,370
because type two diabetes is a very variable condition and there is some evidence that it's in fact several different diseases.

273
00:30:27,910 --> 00:30:33,129
Perhaps the next step, I thought, would be see how other statistically minded people with type two diabetes.

274
00:30:33,130 --> 00:30:39,940
So get the diabetic statisticians of the world to unite, but only the ones who like red wine.

275
00:30:40,150 --> 00:30:41,920
And to replicate this study.

276
00:30:42,950 --> 00:30:51,730
I thought it might be nice to explore this within my own family, but that can't be done because it's a condition of later life.

277
00:30:52,390 --> 00:31:03,070
Most of the people I know I knew with type two diabetes in my family have died and there's only my sister left that I know of.

278
00:31:03,070 --> 00:31:10,270
And she she prefers white. Perhaps a few statisticians with diabetes might be moved to add to knowledge.

279
00:31:10,480 --> 00:31:13,990
How have the response since my significance article has been known?

280
00:31:15,550 --> 00:31:20,560
Be interesting to see how this works in type one diabetes two, which is a completely different disease.

281
00:31:22,840 --> 00:31:28,450
Ask for myself. I continue to record my blood glucose.

282
00:31:29,080 --> 00:31:33,700
I did it this morning. It was 10.7, quite high.

283
00:31:34,900 --> 00:31:45,790
I've added other data, the type of other alcohol, beer, whisky, white wine, the number of drinks, and some information about food consumption.

284
00:31:45,790 --> 00:31:51,100
How much carbohydrate not measured in a scientific way, but equivalence of a slice of bread.

285
00:31:52,330 --> 00:32:00,610
And approximately whether I exercised on the day before as well, because I have an exercise bike,

286
00:32:00,610 --> 00:32:06,970
I try to go on it most days and of course I'll drink red wine as usual.

287
00:32:07,540 --> 00:32:10,690
It seems to be rather good for me. Oh, okay.

288
00:32:10,840 --> 00:32:24,810
Thank you very much. At the time development in Jeff Aronson and the Significance editorial board for their helpful comments on earlier drafts.

289
00:32:25,230 --> 00:32:30,150
Has anybody ever thanked anybody for their extremely annoying comments on an earlier draft?

290
00:32:30,360 --> 00:32:38,219
They were extremely helpful, all of them. I'd also like to thank my two GP's, Sam Patel, who diagnosed me and Alison Hunter,

291
00:32:38,220 --> 00:32:44,820
who now looks after me for their excellent care, stuffing me full of drugs and keeping it within the bounds of reason.

292
00:32:45,990 --> 00:32:46,440
Thanks.