The case against aggressive treatment of type 2 diabetes: critique of the UK prospective diabetes study

The United Kingdom prospective diabetes study (UKPDS) is seriously flawed, and its results do not justify a policy of aggressive treatment of type 2 diabetes.¹,² The study's flaws arise from changes that were made in the trial as it progressed.

Summary points

• During the course of the UK prospective diabetes study the length of follow up was changed, and changes seem to have been made to the end points and the groups under analysis

• These changes are not in keeping with accepted scientific principles and make the results of the trial suspect

• An independent review of the trial's design and analysis is needed

• The results of the study do not justify aggressive treatment of type 2 diabetes

Methods

I searched Medline (Ovid's CD version) from 1976 to March 2000, using various combinations of the keywords “UKPDS” and “prospective diabetes study” and the names of the principal authors of the study's reports. I reviewed the resulting articles for any description of the design of the UKPDS. I reviewed reference lists for other references.

Changes in the end points

The authors have presented various end points over the life of the study (box). The UKPDS grew originally out of the authors' interest in the use of basal rather than postprandial glucose in monitoring diabetes. They concluded, “A prospective controlled trial of different ways of obtaining basal normoglycaemia is needed to determine whether the improved control of mild diabetes is beneficial.”³ In the first report of the study, published in 1983, the authors set out their rationale: “If, after dietary therapy, the fasting plasma glucose continues to be raised, there is little information available to determine whether one should continue with diet alone, or add a sulphonylurea, biguanide or insulin.”⁴ Specific end points were not laid out in this original description but were set out the following year in a letter discussing the paper.⁴,⁵ The authors later argued (1999) that this “brief letter” had only “minor differences in wording.”⁶ It seems, however, that the letter was written specifically to clarify the vagueness and ambiguities of the 1983 paper: the authors wrote, “In the initial [1983] report, it was not feasible to mention all details of the study.”⁵

Changes in the end points and stopping points in the UK prospective diabetes study

19834	199510
•No specific end points reported: “morbidity and excess mortality of the disease,” “complications including macrovascular events or retinopathy, nephropathy or peripheral neuropathy”	Main comparison and interdrug comparison
19845	• “Diabetes-related mortality—death from: heart attacks, sudden death, stroke, complications of peripheral vascular disease or amputations, renal failure, or hyperglycemic or hypoglycemic coma”
Main comparison and interdrug comparison	• “Diabetes-related mortality and major clinical endpoints including non-fatal myocardial infarct, clinical angina, . . . heart failure, . . . major stroke, . . . amputation, retinal photocoagulation, vitreous hemorrhage, blindness, . . . and renal failure”
•“Deaths from vascular events, sudden death or renal failure”	• “Total mortality”
•“Complication-free interval, including avoidance of death from any cause, heart attack, angina, renal failure, blindness, major stroke or amputation”	19969
19917 (interpretation using the stopping criteria set out in tables 4 and 5 of the 1991 paper)	Main comparison and interdrug comparison
Main comparison and interdrug comparison	• “Diabetes-related mortality—death from: heart attacks, sudden death, stroke, complications of peripheral vascular disease or amputations, renal failure, hyperglycemic or hypoglycemic coma”
•“Diabetes-related deaths, ie vascular, renal, hyper- or hypoglycemia or sudden death”	• “Diabetes-related mortality and major clinical endpoints”:
•“Diabetes-related death and major morbitity”: myocardial infarction, angina and ischaemic heart disease, major stroke, major limb complications requiring amputation, blindness in one eye, renal failure (“death from any cause” removed)	• Macrovascular: fatal and non-fatal myocardial infarction, fatal and non-fatal stroke, ischaemic heart disease, heart failure
19917 (interpretation using the aggregates set out in table 5 of the 1991 paper)	• Microvascular: fatal and non-fatal renal disease, ophthalmic (blindness, retinal photocoagulation, vitreous haemorrhage), peripheral neuropathy (amputation)
Main comparison and interdrug comparison	• Cataract
•Non-fatal end points: myocardial infarction, angina and ischaemic heart disease, major stroke, major limb complications requiring amputation of a digit or limb, blindness in one eye, renal failure (“death from any cause” removed)	• “Total mortality”
•“Clinical events not included in stopping criteria” (cataract extraction, vitreous haemorrhage, heart failure, photocoagulation)	19981
•Total mortality	Main comparison
19938 (translated from the French)	• “Diabetes-related death (death from myocardial infarction, stroke, peripheral vascular disease, renal disease, hyperglycaemia or hypoglycaemia,sudden death)”
Main comparison and interdrug comparison	• “Any diabetes-related endpoint (sudden death, death from hyperglycaemia or hypoglycaemia, fatal or non-fatal myocardial infarction, angina, heart failure, stroke, renal failure, amputation, . . . vitreous hemorrhage, photocoagulation, blindness in one eye, or cataract extraction)”
•Diabetes related deaths: cardiovascular events, sudden death, hypoglycaemia, hyperglycaemia, renal failure	• “All-cause mortality”
•Diabetes related major morbidity: non-fatal cardiovascular accidents (heart attacks, strokes, and lower limb amputations), renal insufficiency, blindness	19981
•(Total mortality removed)	Interdrug comparison
	• Main comparison end points (as above) and:
	• “Myocardial infarction (fatal and non-fatal) and sudden death”
	• “Stroke (fatal and non-fatal)”
	• “Amputation or death due to peripheral vascular disease”
	• “Microvascular complications (retinopathy requiring photocoagulation, vitreous hemorrhage, and fatal and non-fatal renal failure)”

In 1991, after the study had been going 14 years and after numerous interim analyses,¹,⁵ the authors restated their end points.⁷ They presented two possible interpretations of the end points, but neither matches the end points given in the final publications.¹,² Nevertheless, the authors repeatedly refer to the 1991 paper as setting out the final end points.¹,⁸,⁹ The end points set out in 1993 are almost identical to those published in 1984.⁵,⁸ It is not until 1995 that the end points take on a form similar to those given in the final publication in 1998.¹,¹⁰ The 1995 paper is also the first time that total mortality is set out as an end point and the first time that a clear distinction is made between microvascular and macrovascular end points (although this distinction is mentioned briefly in earlier reports). By 1996 the end points are, apart from differences in wording, identical to those seen in the final report—the only change is that “diabetes-related mortality and major clinical endpoints” is renamed “any diabetes-related endpoint.”¹,⁹

The final report defines, for the first time, four additional end points to be used when comparing intensive treatments (see box).¹ These end points do not, as far as I could discover, appear in any other report. Every previous publication implies or states that the end points of the main comparison were to be used in the secondary comparison among the different agents—this was stated explicitly in 1984 and 1993.⁵,⁸ In the main body of the final report the authors state that these secondary end points are to be used only when comparing different intensive regimens, not when comparing intensive and conventional regimens (although such a comparison is made in figure 4 of the report).¹ Thus the authors' claim that the “intensive treatment group had a substantial, 25% reduction in the risk of microvascular endpoints” (the only one of the four additional end points that was significantly different in the retrospective comparison of conventional and intensive treatment) is not supported by their study design, and the result must be viewed simply as hypothesis generating.¹

Summary of the changes

The authors made substantial changes to the end points as the study progressed. In particular, cataract extraction, vitreous haemorrhage, heart failure, and retinal photocoagulation were not included in 1984, were mentioned ambiguously (as “events,” not end points) in 1991, were not included in 1993, and then were included from 1995 on.⁵,⁷,⁸ In fact, none of the publications before 1995 specifically set out the end points that were used in the final analysis. It also seems that the decisions about which of the aggregations of end points to include were based on the results of the interim analyses that were available to the authors.¹,⁵

Soft end points

A second point is that the study was not blinded. This is less an issue with the hard end points (death) than with the soft end points (cataract extraction and retinal photocoagulation), where the decision to perform a procedure might have been influenced by the degree of glucose control in a patient. This is important, as the decrease in these soft end points accounts for all of the significant beneficial results claimed in the study.¹

Analysis of subgroups

The main purpose of the study was always to compare intensive and conventional (diet) treatment. This aim is consistent throughout the early reports of the study and was stated as the overall objective in the authors' 1998 review.⁴,⁷,⁸,¹¹,¹²

Over the years, however, a subtle change was made. A question based primarily on outcome (does lowering blood sugar decrease morbidity and mortality?), with a secondary question based on mechanism (does the way in which blood sugar is lowered matter?), was changed to a question based primarily on mechanism, and the question based on outcome was simply ignored. This change is reflected in the treatments that were analysed (initially conventional versus intensive treatment, then later conventional treatment versus sulphonylurea and insulin and conventional treatment versus metformin). The authors' argument, first put forth in 1996, that separate comparisons based on mechanism should be made is interesting and worthy of study, particularly with the development of the thiazolidinedione drugs, but it does not justify the decision not to publish the study as originally designed.⁹

Length of follow up

The study was originally planned to end in 1992, with a median follow up of seven years.⁵ In 1987 an interim analysis showed negative results; the study was therefore expanded in size and also in length of follow up.⁷ In 1990 the study was “due to report in 1995, by which time a total of 5000 patients will have been followed for a median of 8 yr.”¹³ In 1991 the study was “planned to finish in 1994 with a median follow-up of 9 years.”⁷ In 1993 it was also planned to finish the study in 1994.⁸ It was not until 1995 that the authors stated, “The clinical study will end in 1997 when the 4,209 patients will have had a median time since randomization of 11 years.”¹⁰

Interim analyses were planned to occur every six months to 1985 and yearly thereafter.¹,⁵ It seems that the authors continued the study until they obtained a result that was significant, without adjusting for repeatedly looking at the data. Although it is acceptable to extend a study for a set period and to have predetermined stopping rules, it is not acceptable to repeat interim analyses and to delay publication until a significant result is found.¹⁴

Recommendations for screening

Can the findings of this study be generalised to screening for asymptomatic diabetes? The patients in the study were aged 25-65 years, had type 2 diabetes that was newly diagnosed, and were referred by local general practitioners. No attempt was made to screen patients for diabetes, and at diagnosis 50% of the patients had evidence of diabetic tissue damage.¹⁰ The study was not designed to show a benefit from screening and would not have been powerful enough to do so even if it had been. Screening detects cases of diabetes much earlier in the course of the disease, and it is not logical to imply that beneficial treatment (if there is any) given later in the course of the disease would give the same benefit (either absolute or proportional) when given earlier. Furthermore, the ethics of screening are different in several respects from those of routine medical investigation and treatment and require that more attention be paid to issues of harm, consent, and cost.¹⁵

Are the results clinically significant?

The authors present details of their power calculations and state that the study had a good chance of detecting a 20% or 15% benefit.¹,⁷,⁹,¹⁰,¹² They further state, “This reduction has been accepted as being a clinically significant gain,”⁷ and, “A protective effect of 15% has been judged to be clinically relevant,”¹² implying that lower reductions should not be considered clinically significant.

Not one of the main results of the study even approaches these numbers. The best result is in the “any diabetes-related end point,” a reduction of 12% (95% confidence interval 1% to 21%).¹ The risk of diabetes related death—clinically the most important result—was not significantly reduced.¹

Conclusions

In 1999 the authors restated their position that after 1981 they made no substantive changes in the design of the study (apart from those discussed in the 1998 paper).¹,⁶,¹⁶ The ambiguities and contradictions in the various reports of the study cannot, however, be denied. Whether the authors' conclusions are supported by the data cannot be resolved by debate but only by an independent review of the study's design and analysis. Meanwhile it is not unreasonable to ask for the results to be published as outlined in 1984—that is, as a comparison between the intensive treatment and conventional treatment groups, using the two end points “deaths from vascular events, sudden death or renal failure” and “complication free interval, including avoidance of death from any cause, heart attack, angina, renal failure, blindness, major stroke or amputation.”⁵ It would not be unreasonable to add total mortality to this list, with the caveat that it was not included in the initial design.

If it is true that substantial changes, derived from ongoing reanalysis of the data, were made, and if reanalysis of the data according to the original design shows no significant benefit, then we must call into question the recommendations for screening and more aggressive treatment that have flowed from the publication of this study.¹⁷,¹⁸ In the interim we should return to the position that, although management of the symptoms of type 2 diabetes is reasonable (that is, with the intention of keeping blood sugar concentrations below about 11-14 mmol/l (200-250 mg/dl), a recommendation in favour of screening and more aggressive care is not supported by the evidence presented in this study.