[這篇是我被那兩個人聯手趕出來以後，心裡很不服氣，所以就把老師口中「不可能做得更好」的研究找出來仔細研讀，發現其實裡面破綻不少，而且資料的運用分析大有改進的空間。於是我就寫了這樣的一封信給英國牛津大學的研究人員，跟他們說明我的想法，並表達如果能讓我拿到原始資料，我就有辦法解出一個醫界爭論已久、教科書積非成是的迷團…]

Hi,

I am sorry to bother you. I know that you must be very busy. I am a Ph.D. student in the Dept. of Epidemiology, University of Washington, Seattle, WA, USA. I read your published study, UKPDS 39 (BMJ 1998;317:713-20), yesterday. I found it was very intriguing, especially in the aspect of hypoglycaemia. I am interested in the effect of antihypertensives in increasing/decreasing the risk of hypoglycemia in DM patients.

The main purpose of the trial was to determine the effect of tight control of blood pressure by a beta-blocker (atenolol) or an ACE inhibitor (captopril) in preventing the macrovascular and microvascular complications of type II diabetes. The side effect of hypoglycaemia was not the major concern. I found it was not so well to address hypoglycaemia as an outcome, although much more subjects than other clinical trials were enrolled (eg. Kerr D et al. Br J Clin Pharmacol.
1990;29:685-93, Viberti GC et al. Metabolism. 1980;29:866-72, or Clausen-Sjobom N et al. Acta Med Scand. 1987;222:57-63.). The results about hypoglycaemia contained possible biases. And, you didn’t calculate relative risk of hypoglycaemia, either, but concluded that beta-blockers and ACE inhibitors had similar effects in causing hypoglycemia (similar fault made by EUCLID study: Randomised placebo-controlled trial of lisinopril in normotensive patients with insulin-dependent diabetes and normoalbuminuria or microalbuminuria. The EUCLID Study Group. Lancet. 1997;349(9068):1787-92.).

The issues in concern would be:

1. Could randomization really control ALL possible confounders? The answer might be NO. It still could happen by chance. In intention-to-treat comparisons (Table 1), 38% patients in captopril group were treated by sulphonylurea, but only 29% was treated by sulphonylurea in atenolol group. And, 20% were treated by insulin in captopril group, but 29% were treated by insulin in atenolol group. So the treatment of DM was related to the exposure by the process of randomization (by chance, in other words). And, it’s surely related to the outcome, hypoglycaemia. It could be a potential confounder and should be controlled in presenting the results of hypoglycaemia.

2. The difference of compliance between captopril group and atenolol group was huge (P-value less than 0.0001). 78% patients in captopril group took it till the end of trial. But, only 65% patients in atenolol group did, mostly because of intolerance by impaired peripheral circulation or bronchospasm. I believe it’s non-selective (to the outcome of hypoglycaemia). So, in the analysis of intention-to-treatment, this was a non-selective misclassification of exposure which resulted in relative risk shifting to null. Since only 65% compliance was performed in atenolol group, the effect estimation of beta-blocker was very dubious. Since the lost to follow-up was not selective to the side effect of hypoglycaemia, I think the analysis of really treatment group would be fine in estimating the risk of hypoglycaemia and, thus, estimate the RR more reasonably.

3. I tried to calculate the relative risk of major episode for ACE inhibitor and beta-blocker group relative to less tight control group. They were 1.49 (95% CI: 0.79, 2.93) and 1.29 (95% CI: 0.64, 2.61) separately. The estimations were suffered from non-selective misclassification of exposure (making the estimations shift to null, especially for the beta-blocker group). Although they were not significant, the differences were still unforgettable. And then, I tried to calculate the detectable RR by PS program (Power and Samples Size Calculation, a free software made by William D. Dupont and W. Dale Plummer, Jr., Division of Biostatistics, Department of Preventive Medicine, Vanderbilt University School of Medicine, 1996). Under the condition of 80% statistical power, significance level at 0.05, sample size 400 in ACE inhibitor group, and 4.4% probability of events in controls (less tight control group), the minimum detectable RR was 2.21. In beta-blocker group (n = 358), it’s 2.30. Insufficient sample size was still a crucial issue in this study.

Based on the points above, I think you are on an excellent stand point to address the effect of ACE inhibitor and beta-blocker to type II DM patients in causing hypoglycaemia. If the study is still going on, the total duration of follow-up would be even more (around 11 years, right?). If we could analyze the data with a more powerful statistical method (ex. survival analysis or non-parametric method), it would be very helpful to clarify the possible effects of antihypertensives to hypoglycaemia in your cohort! Meanwhile, possible confounders by the treatments of DM could be controlled and the analysis for real treatment group could be performed, too.

So, could I get the raw data of the study from you? We could cooperate to perform another paper in addressing the issue of hypoglycaemia in hypertension intensively treated DM patients. What you think??

Chin-Kuo Chang

[很有說服力吧？他們那邊很快地也有了反應，而且說說我的想法是「…impressed with how well referenced it was…」喔！不過他們的研究目的並不只是糖尿病人的低血糖血症而已。以他們UKPDS研究團隊的作法，是不會把原始資料給外人分析的…]

Dear Chin-Kuo Chang:

Thank you very much for your letter. I was impressed with how well referenced it was.

I would agree with you that the study is not, in fact, powered to detect a difference in hypoglycaemia between captopril and atenolol. The study was designed initially to test the glucose control hypothesis, and so powered. The study initially overestimated the risk difference, modifying (lessening) en route and thereby extending the study. Any power in the hypertension are was in the tight versus less tight comparison, and then only for the primary endpoints, not hypoglycaemia itself. Moreover, hypoglycaemia was symptomatic - and not necessarily documented by blood glucose testing. As such, if patients on beta-blockers did not detect their hypoglycemia, then differential misclassification could have occurred. For these reasons, this, at present, is not high on the list of research priorities.

While I welcome you offer of analysing the data collected by the UKPDS, it is not currently our policy to provide data to outside groups.

In answer to your question, we are doing post study monitoring, but, of course, given the results associated with good control of blood glucose and blood pressure, there is no effort to maintain a difference between intensive vs. conventional, and tight vs. less tight.

Again, I appreciate your thoughtful communication.

Please give my regards to Noel Weiss.

Sincerely,

Dr. Amanda Adler

[不過他們研究的總管說他們可以開會一下來討論看是不是要把原始資料給我，開會時間是兩千年的七月十二號，似乎又燃起了希望…]

Dear Dr Chang

Thank you for you e-mail regarding our paper (UKPDS 39), and your request for access to our data.

All such requests are considered by our management team, which meets monthly. The next meeting is to be held on July 12th: we will consider your request at that time, and contact you again shortly afterwards.

Sincerely
Philip Bassett
UKPDS Administrator
Diabetes Trials Unit
University of Oxford
Radcliffe Infirmary
Woodstock Road
Oxford OX2 6HE

[不過結果還是失望的，他們已經自己決定要針對我的想法作一個分析「…already working on analyses similar to those outlined in your letter…」，唉，又一次為人作嫁了…]

I’m afraid that the answer is ”no”, on two grounds. Firstly we have a policy of not releasing raw data, and secondly our group is already working on analyses similar to those outlined in your letter.

I am sorry that you have not received a reply. We discussed your proposal on 12th July, and Professor Holman was due to write to you after that meeting to convey our decision. I will ask him to ensure that you do receive a formal reply. However, he has been out of the office for considerable periods over the past four weeks.

I am sorry to be the bearer of bad news.

Sincerely
Philip Bassett

[兩個禮拜後，那個Professor Holman也正式地回應我說他們要自己拿去做了，唉…。兩年多過去了，他們也沒發表出這篇東西，真是的，搶著要又不努力做，這是在幹什麼呢？？]

Re: Hypoglycaemia in the UKPDS
Thank you for your recent email which was considered by the UKPDS Executive Committee on Wednesday 12th July 2000. We were sympathetic to your request and impressed by the very sensible questions you asked, but unfortunately we are not in position to release raw data to external researchers at this time. I can tell you however, that this particular topic is the subject to one of our on-going analyses which we hope will be published within the next twelve months.

With many thanks for your interest.

Yours sincerely,
Rury R Holman FRCP
Professor of Diabetic Medicine
Head of Diabetes Trials Unit