Association of glycaemia with macrovascular and microvascular complications of type 2 diabetes (UKPDS 35): prospective observational studyBMJ 2000; 321 doi: https://doi.org/10.1136/bmj.321.7258.405 (Published 12 August 2000) Cite this as: BMJ 2000;321:405
All rapid responses
Percent vs. Percentage Point, Re: Association of glycaemia with macrovascular and microvascular complications of type 2 diabetes (UKPDS 35): prospective observational study
Consider this simple math. If 20% of a company’s employees are smokers at baseline, and only 10% are smokers after a company-wide smoking cessation intervention, how much of a decline does this represent?
If your answer is “10%,” you’re wrong, but in good company. Many physicians and a decent swath of the medical literature get this wrong, too. I got it wrong up until a few years ago when a friend of mine in finance set me straight.
The correct answer is either “10 percentage points” or “50%.” Consider the implications. An intervention that reduces smoking by 50% is far more compelling than one that reduces the habit by only 10%.
Now that I am attuned to this important distinction—particularly critical in the comparison of small percentages—it bugs me to notice the error on a regular basis.
I recently came across this conclusion in the referenced BMJ article: “Each 1% reduction in haemoglobin A1c was associated with a 37% decrease in risk for microvascular complications....” At first I was surprised that such a tiny change (only 1%!) could have such a strong impact.
However, given that hemoglobin A1c (HbA1c) is itself expressed as a percentage, this 1% statistic refers to a comparison of percentages. What the authors meant to express was a reduction in HbA1c from, for example, 10% to 9%, which is actually a 1 percentage point difference. (I corresponded with one of the authors to confirm that this is what they meant.) Expressed accurately as a percentage, this decrease in HbA1c from 10% to 9% is actually a reduction of 10%, not 1%.
This is not just annoying semantics. It affects how we think about risk, outcomes, and the efficacy of interventions. Ideally, we ought to make the distinction clear, not only in finance but also in medicine.
Competing interests: No competing interests
Removing the influence of LADA patients will add more value to UKPDS35 findings on Type 2 diabetes related complications.
UKPDS35  a publication of United Kingdom Prospective Diabetic
Study (UKPDS) concludes that there is a direct relation between the risk
of complications of diabetes and glycaemia over time. How ever unadjusted
complication rates provided in the table2 of the paper shows a significant
variation from the above mentioned pattern at very high HbA1c values. In
fact mortality rates and myocardial infarction [MI] rates are lower for
HbA1c >10% group compared to few other groups with better glycimic
control. UKPDS35 introduces an adjustment to this abnormal variation by
applying a statistical technique which focuses on white men aged 50 to 54.
How ever a more meaningful correction is necessary and seems to be
available to explain the true nature of variation in complications with
glycimic control applicable for the Type 2 diabetic patients.
Most of the UKPDS reports including its main report UKPDS33  and
UKPDS35 do not differentiate the LADA patients from the Type 2 diabetic
patients. How ever 9•4% of the patients included in the randomized sample
had LADA . As LADA is different from Type 2 diabetes, almost all of
them will have high HbA1c ratings unless they were provided with
sufficient quantity of insulin. Protocol adopted in the UKPDS has
directed LADA patients in the conventional groups to keep Fasting Plasma
Glucose(FPG) values closer to 15 even after converting them into
sulfonylurea or insulin .
Fast response to UKPDS41  provides a simple method of separating
total diabetes related complications in to two groups named as latent
autoimmune diabetes in adults (LADA) group and Type 2 diabetes group. The
method assumes in spite of LADA patients having relatively higher micro
vascular complications but lower macro vascular complications, the
combined affect will produce more or less similar number of total
complications compared to Type 2 diabetic patients. As the abnormal
variation patterns appearing in UKPDS35 are related to macro vascular
deceases where complication rates are significantly different additional
factors should be considered to obtain a more meaningful result.
UKPDS66  provides the simple risk factors and their sensitivities
derived from the trial sample applicable to fatal myocardial infarction.
MI risk increases with age, HbA1c %, systolic blood pressure (SBP) and the
period elapsed after diagnosis. Although the sensitivities to these
parameters are not linear, cumulative effect of above parameters is
governed by the arithmetic sum of four figures each one representing one
of the identified risks. Figure representing the affect of a risk is
computed by multiplying the actual parameter by corresponding sensitivity
coefficient (coefficients for fatal MI as given in the UKPDS66 are, Age
: 0.048, HbA1c: 0.178, SBP: .0141. Although the non fatal MI
sensitivity parameters can be some what different, their relative values
should be similar. Therefore above figures will be used as general MI risk
sensitivity coefficients). As the post detection period does not vary with
the type of disease it will not be considered in the current discussion.
UKPDS70 provides the comparative details of LADA and Type 2
diabetic patients included in the trial. LADA patients were 4.5 years
younger and their SBP values were 7 units lower compared to Type 2
diabetic patients. (LADA age 48.2: SBP 129, Type 2 diabetic patients age
52.7 : SBP 136).
As shown in the fast response to UKPDS41 LADA patients in the
conventional treatment group had HbA1c value closer to 10% and LADA
patients in the intensive control group had low HbA1c values? As a
consequence of above findings it can be concluded that there cannot be
many LADA patients with intermediate level of HbA1c values. In other words
patients having medium level of HbA1c values should be Type 2 diabetic
Therefore change of complication rate applicable for HbA1c change
from 7-8 range to 8-9 range represents a property of Type 2 diabetic
patients included in the sample. However change of complication rate
applicable for HbA1c change from 9-10 to 10+ has got influenced by the
presence of LADA patients. In fact this is the range that shows reverse
trend or reduction in some of the very important complication rates such
as MI and diabetes related death with increased HbA1c %.
Although the sample selected in the UKPDS35 is a sub set of total
UKPDS trial, patients were not selected by any special qualifying
criteria. Therefore they should have more or less same properties as the
UKPDS total group.
UKPDS had 1138 patients in the conventional treatment group. As 9.4 %
of them had LADA there should be 107 LADA patients in the conventional
treatment group. As there were 5102 patients in the total group 2.097% of
them were LADA patients assigned to the conventional treatment.
MI patient years in UKPDS35 sample were 42880 (can be obtain by
adding patient years of different sub groups). If LADA patients and Type
2 diabetic patients had similar survival rates then LADA patient years in
conventional treatment group computed according to above % is 899. How
ever according to UKPDS70 availability of LADA patients at the end of 10
years were 10% higher compared to type 2 diabetic patients( 51% LADA
patients: 46% Type 2 diabetic patients). Therefore it is more correct to
apply a 5% upward revision to above figure to accommodate the increase in
recorded LADA patient years due to higher availability for monitoring. As
a result of above correction number of LADA patient years becomes 944.
Total patient years applicable for the HbA1c >10 range for MI is
1490. Therefore reasonable part of the patient years counted under HbA1c
>10 range should be LADA patient years. (If 50% of the LADA patients
had HbA1c >10 then 472 LADA patient years are included in the
HbA1c>10 range patient years. This represents 31.7% of the total
patient years of the highest HbA1c category. Actual % of LADA patient
years in the HbA1c>10 range is likely to be much higher.)
As LADA patients were younger and had lower SBP values their MI
complication rate is much less. From the sensitivity parameters given in
the UKPDS66 it is quite clear MI complication rate of the average LADA
patient with HbA1c % closer to 10 is much less compared to a Type 2
diabetic patient with HbA1c% closer to 9 , This happens because LADA
patients are younger and they had lower SBP. As an example when HbA1c
increase of 1, age decrease of 4.5 and SBP decrease of 7 will result in a
reduction in the cumulative risk indicating figure [arithmetic addition (-
.1367)= HbA1c factor(+.178)+age factor(-.216)+ SBP factor(-.0987)]
Above calculation clearly shows that presence of LADA patience has
distorted the picture of relationship between complication rates and
HbA1c % variation. How ever in addition to SBP, HbA1c%, Age and period
after diagnosis, type of illness should also be considered as a separate
risk factor. Actually sensitivity factors applicable for LADA can be
significantly different from the sensitivity factors applicable for Type 2
diabetes. It is to be noted that the sensitivity factors used here were
derived from the mixed sample and they should be different from the actual
factors corresponding to LADA and Type 2 diabetes both.
Actual pattern variation should have got influanced by the well known
property of LADA patients having less macro vascular complications. As
these patients does not have problems related to excess insulin net
reduction in macro vascular complications should be much higher than the
computed result representing the effect of SBP, Age and HbA1c only.
As mentioned earlier UKPDS70 clearly indicates that percentage of
LADA patients available for follow up at 10 years were higher compared to
Type 2 diabetic patients. Considering the probability of dropout by other
factors to be equal it can be concluded that death rate of LADA patients
is much lower compared to the Type 2 diabetic patients. In fact this
conclusion is strongly related to the findings of UKPDS35 and the above
Presence of LADA patients should have affected all the complication
types. As discussed earlier it is a different decease. Categorizing
their properties separately will help to design better treatment options
to both patient groups. As the LADA patient percentage is much higher than
Type 1 patient percentage in many communities separation of UKPDS35 data
in to two groups should be considered as a priority. To collect such a
volume of data from a new study will take a long time.
1. Stration M IM, Adlet AI, Neil HA, Matthews DR, Manley SE, Cull CA,
et al Association of glycaemia with macrovascular and microvascular
complications of type 2 diabetes (UKPDS 35) BMJ 2000 321:405-12.
2. UKPDS Group (1998) Intensive blood-glucose control with
sulphonylureas or insulin compared with conventional treatment and risk of
complications in patients with type 2 diabetes. (UKPDS 33) Lancet
3. Davis TM Wright AD, Mehta ZM et al (2005) Islet autoantibodies in
clinically diagnosed type 2 diabetes: prevalence and relationship with
metabolic control (UKPDS 70). Diabetologia 48:695-702
4 UKPDS Group (1991)UK Prospective Diabetes study (UKPDS) VIII
Study design, progress and performance. Diabetologia 34:877-890
5. LADA patients in the UKPDS sample has influenced the ukpds41
conclusion: Arya K Kumarasena : Rapid Response to Cost effectiveness of
an intensive blood glucose policy in patients with type 2 diabetes:
economic analysis alongside randomised controlled trial (ukpds 41). Gray
a,Raikou M, McGuire A et al(2000). BMJ 320:1373-13
6. Stevens RJ Coleman RL Adler AI Stratton IM Matthews DR Holman RR
(2004) Risk Factors for Myocardial Infarction Case Fatality Type 2
Diabetes. Diabetes Care 27:201-207
Competing interests: No competing interests
Editor - Articles recently published in BMJ (UKPDS 35 & 36)1,2
by the UKPDS group appear methodologically problematic, and are
It appears the authors followed analytical procedures used previously for
type 1 diabetes data from the DCCT3, involving proportional hazards and
Poisson regression models of the relative risks of developing diabetic
complications due to lifetime exposure to hyperglycaemia and hypertension.
However, there are important differences between the two studies with
consequences for their analysis. In particular, in the DCCT cohort there
were insufficient events to allow any analysis of either macrovascular
complications or mortality: results were only reported for three
Sadly, insufficient information is provided in the UKPDS papers to allow
full assessment of the statistical models obtained. Nonetheless, the
limited detail supplied suggests that diabetes-related mortality risk
increases considerably faster than all- causes mortality with both HbA1c
and SBP, so that all non-diabetic mortality ceases for patients with HbA1c
> 13% or SBP >186mm Hg. Clearly this is unrealistic and casts doubt
on the appropriateness of this model formulation for mortality. A
competing risks model, with non-diabetes related deaths assumed to be
independent of both glycaemia and blood pressure, would be preferable. We
have estimated such models using both linear and log-linear equations for
diabetes-related mortality risk, and obtain plausible relationships
consistent with the published data and credible for all values. If the
DCCT approach fails for mortality, it also unlikely to work for macro-
vascular disease generally, the dominant cause of diabetes-related
Use of lifetime exposure to glycaemia and blood pressure as principal
independent variables in the analyses implies a very strong general
assumption on the nature of incidence and progression of diabetic
complications. Though reasonable to consider that microvascular damage
accumulates continuously over time, in the case of heart disease and
stroke there are probably different mechanisms operating in the
development of chronic disease, incidence of acute events, and case
fatality rates so that current values of clinical variables may be at
least as influential as accumulated exposure. There is merit in exploring
a range of alternative exponentially-smoothed independent variables.
Another consequence of using updated mean HbA1c in regression models is
that incremental risk ratios reported in Table 3 are not comparable with
those shown for the equivalent baseline variables. The linear upward
trend in glycaemia observed in the trial means that models using a
cumulating-average mean variable will generate risk gradients
approximately double those obtained with single observations. Thus from a
clinical perspective, the models which use updated mean HbA1c
significantly overstate the risk gradient.
Most disturbingly, these publications give no details of the statistical
models employed, by which other researchers may judge their
appropriateness and reliability, in stark contrast to the DCCT study which
presented full diagnostics of their models.3 Though such information may
not be relevant to all BMJ readers, it could have been provided as an eBMJ
appendix, or a supplement obtainable from the authors. At present the
Diabetes Trial Unit at Oxford is refusing to release such details to
individual researchers, and it is a matter of regret that custodians of a
major trial database, largely funded from public sources, seem unwilling
to make the fruits of their work more freely available for scrutiny by the
wider academic and clinical community.
York Health Economics Consortium,
University of York, UK.
1 Stratton IM, Adler AI, Neil HAW, Matthews DR, Manley SE, Cull CA,
et al. Association of glycaemia with macrovascular and microvascular
complications of type 2 diabetes (UKPDS 35): prospective observational
study. BMJ 2000; 321:405-12.
2 Adler AI, Stratton IM, Neil HAW, Yudkin JS, Matthews DR, Cull CA,
et al. Association of systolic blood pressure with macrovascular and
microvascular complications of type 2 diabetes (UKPDS 36): prospective
observational study. BMJ 2000; 321:412-9.
3 DCCT Research Group. The relationship of glycaemic exposure
(HbA1c) to the risk of development and progression of retinopathy in the
Diabetes Control and Complications Trial. Diabetes 1995; 44:968-83.
Competing interests: No competing interests
Stratton, et. al. have documented that as glycaemic exposure
increases, diabetic complications increase.(1) They conclude that
treatment of hyperglycemia will have substantial benefit, a conclusion
reiterated by Dr. Tuomilehto.(2) Yet reduction of glycaemic exposure
failed to have such benefit in the UKPDS randomized trial (3, 4) For
example, Stratton's data would suggest that reducing mean HgA1C by 1%
would reduce diabetes related deaths by 21%. Intensive treatment of
hyperglycaemia for 10 year in UKPDS reduced HgA1C by nearly 1% (from 7.9
to 7.0%), yet failed to reduce diabetes-related deaths significantly. The
conventionally treated group, with greater glycaemic exposure, suffered
diabetes-related deaths at a rate of 11.5 deaths/1,000 person-years.
Based on Stratton's data, the intensively treated group should have
experienced diabetes-related death at a rate of 9.0 deaths/1,000 person-
years. However, intensive treatment was associated with only a non-
significant decrease in diabetes-related mortality.(4) Similarly,
Stratton's data suggests that intensive treatment would result in
significant reductions in adverse outcomes that include all-cause
mortality, stroke, myocardial infarction, and amputation. Reducing HgA1C
by nearly 1% in the UKPDS, however, was not associated with significant
reductions in any of these adverse outcomes.
Thus, treatment that significantly improves glycaemic control does
not achieve the predicted benefit. Does this mean that greater glycaemic
exposure is a marker for adverse outcomes--but not a cause? This would
imply that the higher the HgA1C, the more one needs to pay attention to
non-glycemic treatment of diabetic patients--such as controlling blood
pressure. Or does it mean that the treatments currently available to
lower glucose harm diabetic patients as much as the lowering of blood
glucose helps them?
I wonder if McCormack and Greenhalgh are correct when they suggest
that metformin treatment improves outcomes in diabetic patients, not
necessarily resulting from its glucose lowering effect, but that lowering
glucose per se is of little to no value in type 2 diabetes.(3)
(1)Stratton IM, Adler AI, Neil AW, et. al. Association of glycaemia with
macrovascular and microvascular complications
of type 2 diabetes (UKPDS 35): prospective observational study. BMJ
(2)Tuomilehto, J. Controlling glucose and blood pressure in type 2
diabetes. BMJ 2000;321: 394-395.
(3)McCormack J, Greenhalgh T. Seeing what you want to see in randomised
controlled trials: versions and perversions of UKPDS data. BMJ
(4)UKPDS Group. Intensive blood glucose control with sulphonylureas or
insulin compared with conventional treatment and risk of complications in
patients with type 2 diabetes Lancet 1998;352:837-853.
Competing interests: No competing interests