Full text of DOI:10.1348/014466500163356

349
British Journal of Clinical Psychology (2000), 39, 349±362 Printed in Great Britain
#
2000 The British Psychological Society
Diabetes mellitus and the rate of cognitive ageing
This work was carried out at the Age and Cognitive Performance Research Centre, University of Manchester, UK
Nuala Bent*
Research School of Medicine, Rheumatology and Rehabilitation Research Unit, University of Leeds, UK
Patrick Rabbitt and David Metcalfe
Age and Cognitive Performance Research Centre, University of Manchester, UK
Objectives. There is some evidence that besides aåecting peripheral neural
function diabetes may also cause more widespread changes in the central nervous
system which reduce cognitive e¬ciency and so, also, independence and quality of
life. The present study explores whether diabetes mellitus is a compounding factor
in average declines in cognitive performance observed in old age.
Design. A sample of diabetics and controls were compared on a battery of
cognitive tasks previously used in cognitive ageing research.
Methods. Thirty-three insulin dependent (IDDM), 135 non-insulin dependent
(NIDDM) diabetics and 2191 non-diabetics aged between 50 and 91 years were
compared on two tests of general intellectual ability, and on three tests of verbal
memory.
Results. Overall, the combined IDDM and NIDDM groups had signi®cantly lower
average scores than the controls group on all cognitive tasks. Detailed analyses
revealed most cognitive impairment for the NIDDM sub-group whose condition
was managed by hypoglycaemic drugs, slightly less for those managed by diet, and
no impairment for the IDDM group. These eåects were independent of age,
depression, socio-economic status, and presence of other illnesses.
Conclusions. Together with other recent studies these data emphasize the need for
early detection and eåective management of diabetes in older patients.
Changes in cognitive ability in old age are brought about both by processes of
`normal’ or `usual’ ageing (see Ricklefs & Finch, 1995) and by a steadily
accumulating burden of pathologies. A main task of cognitive gerontology is to
compare the relative contributions of these two factors and their interactions.
Illnesses certainly contribute to changes in the average levels of cognitive ability with
age because a number of pathologies that become increasingly common in old age
are also known to aåect mental e¬ciency (see review by Holland & Rabbitt, 1991).
This leads us to ask further questions: (a) Do the eåects of pathologies and of
`normal ageing’ interact to change cognitive function and, if they do, are the impacts
of pathologies greater when they occur in old age? (b) Because average durations of
chronic pathologies are likely to be longer in older adults than in younger adults, we
* Requests for reprints should be addressed to Dr Nuala Bent, Research School of Medicine, Rheumatology and
Rehabilitation Research Unit, University of Leeds, 36 Clarendon Road, Leeds LS2 9NZ, UK

350
Nuala Bent et al.
need to know how far any age pathology interactions that may be observed re¯ect
¬
either increases in average durations or their more marked eåects if they ®rst occur
late in life. (c) Does the eåect of any given pathology on cognition vary with its
particular mode of expression, with its degree of severity and, an especially important
practical question, with the eåectiveness of its clinical management.
In practice, these questions are best asked by examining the eåects of one
pathology at a time by recruiting a large group of older people who suåer from it
and comparing them with matched controls. Screens of even very large
`demographically normal’ elderly samples are unsatisfactory because incidence of any
particular pathology is usually quite low, individual diåerences in durations and age
of ®rst onset of illnesses and in severity and style of case management are usually
considerable and variations in these factors across age groups are usually too great
to reveal interactions of interest.
Diabetes provides a useful test case. Because it is often associated with peripheral
neuropathy (Worrall, Moulton, & Briåett, 1993; Zaslavsky, Gross, Chaves, &
Machado, 1995), we may assume that it is also likely to compromise central nervous
system function and, indeed, the majority of studies have found that diabetics score
signi®cantly lower than non-diabetics on tests of mental ability (e.g. Holland &
Rabbitt, 1991; Perlmuter et al., 1984; Tun, Perlmuter, Russo, & Nathan, 1987). It
is important to distinguish between the cognitive eåects of two kinds of diabetes,
`insulin dependent diabetes mellitus’ (IDDM) and `non-insulin dependent diabetes
mellitus’ (NIDDM). NIDDM is of particular interest to cognitive gerontologists
because its onset typically occurs late in life, its incidence increases sharply with
population age (Harris, 1990; Krieger & Rudberg, 1994; WHO, 1984), and most
studies have found that it does aåect cognitive ability (e.g. Mooradian, Perryman,
Fitten, Kavonian, & Morley, 1988; Perlmuter et al., 1984; Perlmuter, Tun, Sizer,
McGlinchey, & Nathan, 1987; Tun et al., 1987). This implies that the in¯uence of
NIDDM as a factor determining variance in cognitive function is likely to increase
with the average age of the population sampled. This also means that when assessing
the contributions of diabetes to cognitive loss in later life it is important to
distinguish between the eåects of NIDDM, which typically has late onset and so will
not have lasted long, compared to those of IDDM which, because of its early onset,
will have had a long duration and so, possibly, cumulative eåects. Recent studies
suggest that there may be both independent and interactive eåects of duration of
illness and age of onset of diabetes on mental ability. Prescott, Richardson, and
Gillespie (1990) found that cognitive performance declined with IDDM duration in
patients aged between 24 and 67 years. Ryan, Vega, Longstreet, and Drash (1984)
and Ryan, Vega, and Drash (1985) found that adolescents who have become diabetic
before the age of 5 years learn new information less e¬ciently and forget more during
a 30-minute delay than those whose illnesses had later onsets. Many other studies
have also found evidence for cognitive impairment consequent on IDDM in younger
adults (Ewing, Deary, McCrimmon, Strachan, & Frier, 1998; Franceschi et al., 1984;
Holmes, 1986) and in children (Holmes & O’Brien, 1995). A review by Richardson
(1990) suggests that in young people these eåects, although statistically signi®cant,
may be only slight. However, because IDDM is usually acquired early in life its
eåects may become increasingly evident in old age when its average duration is much

Diabetes mellitus and cognitive ageing
351
greater. All of these factors make it necessary to identify and compare the eåects of
age at time of cognitive testing, of age at onset of illness, and of duration of illness
in both IDDM and NIDDM.
The absolute duration of diabetes may be less important than how well metabolic
control has been maintained. There is evidence that adults with poorly controlled
IDDM show subtle decrements in tasks requiring sophisticated information
processing strategies (Ryan, 1988). Animal models and studies of elderly patients
with NIDDM suggest that ine¬cient glucose regulation impairs learning and
memory (Messier & Gagnon, 1996; Messier, Desrochers, & Gagnon, 1999). The
question is worth pursuing because consensus is still incomplete. A well conducted
study by Robertson-Tchabo, Arenberg, Tobin, and Plotz (1986) found that an
NIDDM sample did not diåer from controls in performance on the vocabulary sub-
test of the Wechsler Adult Intelligence Scale (WAIS) or on the Benton Visual
Retention Test and, incidentally, also found no evidence that NIDDM accelerates
cognitive ageing. Prescott et al. (1990) found no signi®cant diåerences in mental
ability between poorly controlled and well-controlled diabetics and concluded that
degree of glycaemic control is probably not correlated with cognitive performance.
Lowe, Tranel, Wallace, and Welty (1994) also found no relationship between indices
of mental ability and of metabolic control and others reported no consistent
association between mild diabetes and cognitive function in community dwelling
older adults (Scott, Kritz-Silverstein, Barrett-Conner, & Wiederholt, 1998). A more
recent study (Deary, 1999) concluded that repeated episodes of hypoglycaemia were
not related to cognitive decrement. Further work is also needed to explore even more
radical inconsistencies in the literature. Not all studies have found diåerences
between diabetics and non-diabetics. For example, Lawson et al. (1984), found no
marked diåerences between diabetic and non-diabetic adults on standard neuro-
psychological tests.
Discrepancies between studies may arise from a number of methodological
diåerences. As we have seen, results are likely to diåer between studies in which
diabetic samples have included only IDDM patients, only NIDDM patients or a
mixture of both. Results are also likely to vary with the average ages of patient
groups, which will possibly aåect average age of onset and certainly aåect average
duration of illness. The nature, and success, of hyperglycaemic management may also
aåect estimates of cognitive change. Evidence that NIDDM may often be associated
with neuropathy and loss of visual function (Zaslavsky et al., 1995) raises the caveat
that diabetic groups must be screened to detect sensory impairments which may
interfere with their performance on cognitive tests (Holmes, Koepke, & Thompson,
1986). It is also possible that the cognitive changes observed in some diabetics may
re¯ect not only cumulative damage to their central nervous systems, but also the
emotional eåects of suåering from a life-threatening illness, or the direct eåects of the
illness on lifestyle and so life-satisfaction (Prescott et al., 1990, in a study of IDDM).
Even mild depression and dysphoria have been shown to be associated with poorer
cognitive performance in elderly populations (Rabbitt, Donlan, Watson, Bent, &
McInnes, 1995). Thus, it is plausible that cognitive performance may be aåected by
emotional response to diabetes, and by the lifestyle changes it may entail and it is
therefore important to investigate the contribution of depression (Silberman et al.,

352
Nuala Bent et al.
1995) to cognitive change in diabetes. As Prescott et al. (1990) suggest, this can best
be done by studying variations in performance within samples of diabetic patients
rather than by gross comparisons between groups of diabetics and non-diabetics.
Finally, as in all explorations of individual diåerences in mental ability, the relative
sensitivity of the tasks used to assess cognitive status is a constant methodological
problem. Absence of a detectable eåect on one task does not imply that there are no
diåerences on others. For example, Lichty and Klacko (1985) found no detectable
eåects of IDDM on simple tasks that had been practised to the point at which they
could be carried out `automatically’. There is also some evidence that diabetes may
selectively aåect some mental abilities but spare others. For example, Richardson
(1990) suggests that cognitive impairment associated with NIDDM seems to be
especially evident on tests of memory e¬ciency. In short, a fundamental
methodological di¬culty in cognitive epidemiology is that the psychometric tests
that are used to detect the eåects of illnesses may give discrepant results both because
they are unequally di¬cult and also because they assess diåerent mental skills, some
of which are aåected by the conditions under examination, while others are not.
The present study tried to take all of these methodological problems into
consideration while exploring the relative eåects of IDDM and NIDDM on variance
in mental ability in old age, whether and how the eåects of these illnesses interact
with age, whether these illnesses aåect some mental abilities but not others and
whether e¬ciency of glycaemic control in diabetes determines the kind and severity
of cognitive changes that occur. To do this relatively large samples of both IDDM
and NIDDM diabetics were recruited and screened by a psychometric battery of a
wide range of tests of diåerent mental ability which had already been shown to detect
changes in cognitive function associated with normal, healthy ageing. The possible
eåects of e¬ciency of glycaemic control were considered by taking into account the
types of treatment that NIDDM patients had undergone to control their condition.
Patients and controls were screened and compared on a depression questionnaire, a
gross index of lifestyle, and level of socio-economic advantage, all of which were
included in the ®nal statistical model used to examine the data. Interactions between
age and the eåects of all of these variables were explored.
Method
Participants
Diabetics were recruited from GP Practices in Greater Manchester in collaboration with the University
of Manchester’s Department of General Practice. The criteria for inclusion in the study included a
diagnosis of diabetes, an age of 50 years or older and freedom from auditory or visual problems, which
would restrict performance on pen-and-paper tests. Patients were initially approached by their GP who
explained the nature of the research project to them. Those who expressed interest in taking part were
then asked to come to a university laboratory or take batteries of cognitive tests. After screening for
perceptual problems a group of 168 diabetics consisting of 94 men and 74 women was recruited. It
follows that these individuals were atypically mobile, and were su¬ciently competent and motivated to
make eåorts to come to an unfamiliar institution on two successive occasions for intensive cognitive
testing. The diabetic sample consisted of IDDM and NIDDM participants whose diabetes was
controlled either by drugs or by diet. A demographic picture of the IDDM and NIDDM samples is
shown in Table 1.

Diabetes mellitus and cognitive ageing
353
Table 1. Demographic details of diabetic sample (mean (SD))
Diabetes
IDDM
Insulin
NIDDM
NIDDM
Diet
F-value
Hypoglycaemic
drugs
Treatments
Age (years)
Age at diagnosis
(years)
64±33 (7±65)
46±91 (15±46)
65±55 (8±31)
58±21 (10±04)
66±20 (7±71)
61±37 (8±37)
F(2, 165) ±43 n.s.
¯
F(2, 165) 15±60***
¯
Duration of illness
(years)
17±59 (14±12)
7±17 (5±38)
4±92 (5±06)
F(2, 165) 24±05***
¯
N
33
100
35
!
*** p ±001.
The control group was 2191 healthy elderly volunteers, resident in Greater Manchester, all of whom
had been tested for the ®rst time on the same battery of psychometric tests and who were aged 50 years
and over.
Procedure
The psychometric test battery used had been standardized during the course of a longitudinal study of
cognitive ageing on more than 5000 healthy elderly volunteers. All tasks had been shown to sensitively
detect age-related cognitive change. The battery required volunteers to attend the centre for two
separate test sessions each lasting 75 minutes. On their ®rst visit they were given both parts 1 and 2 of
the Heim (1970) AH 4 intelligence test, the Mill Hill Vocabulary Test (Raven, 1965) and the Beck
Depression Inventory (BDI; Beck, Ward, Mendelson, Mock, & Erbaugh, 1961).
Tests administered
The AH4, parts 1 (AH41) and 2 (AH42) are two, 10-minute tests of 65 questions each, the ®rst assessing
the ability to solve verbal, numerical, and logical problems and the second assessing spatial reasoning,
mental rotation and addition and subtraction of shapes and other problems involving visuo-spatial
intelligence.
The Mill Hill A (MHA) and B (MHB) scales are vocabulary tests in which suitably selected words
are presented arranged in descending order of relative frequency of use in everyday life. In scale A,
volunteers are asked to select the most exact synonym for each of 33 words among 6 plausible
alternatives; in scale B, they are asked to present concise and exact de®nitions for each of 33 words. In
contrast with performance on tests of general ¯uid mental ability and on memory tests, performance on
these and other vocabulary tests is known not to decline throughout adult life until age 80 or beyond
(Rabbitt, 1993). However, although previous research has shown that patients with severe closed head
injuries were not impaired when compared to control patients on Mill Hill Vocabulary tests scores
(Sunderland, Harris, & Baddeley, 1983), in the main, very little is known about the robustness of
vocabulary scores in the face of the presence of pathologies.
On their second visit both diabetics and their controls were given three memory tasks: a cumulative
learning task, a picture recognition task and a free recall task. However, not all diabetics returned for
the second test battery and consequently, the reduced sample is re¯ected in the reporting of the memory
tasks. In all tasks words and pictures were clearly projected on to a large, 3 m¬ 3 m screen so that they
were very easily legible by all participants. Presentation durations were timed accurately by an automatic
electronic control device ®tted to a Kodak `Carousel’ projector. In the Cumulative Learning task (CL),
15 words were presented visually, in sequence, one at a time for 3 seconds each. Participants were then
immediately asked to write down as many as they could recall, in any order. The same 15 words were

354
Nuala Bent et al.
again presented, in new random orders, on three successive occasions on which the same procedure was
followed. Scores analysed were total numbers of words correctly recalled over all four trials.
For the Picture Recognition task (PR) 40 line drawings of persons and actions were presented
successively for 3 seconds each. Immediately following this 10 of these drawings were immediately
shown, randomly embedded among 10 other, very similar, distracter pictures that had not been
presented. Volunteers saw drawings one at a time and were allowed 3 seconds to decide whether or not
each had appeared in the original set of 40. This procedure was repeated three times with diåerent target
and distracter pictures at intervals of 20 minutes during which the other memory tasks were performed.
PR results are scored in terms of the derived signal detection measure, d, a measure of recognition
accuracy which takes into consideration both failures to identify previously presented pictures, and false
identi®cations of new pictures.
For the verbal free recall task (VFR) 30 words were projected one at a time on a large screen in
random order for 2 seconds each. Volunteers were then immediately asked to recall as many as possible
in any order. Apart from the cognitive tasks, questionnaire data was also collected from both volunteers
and diabetics on general health, problems including vision and hearing, socio-economic history, family
mortality, mobility, hobbies and occupations and other demographic and lifestyle factors.
Results
When comparing the cognitive performance of diabetics and healthy controls it is
necessary to be sure that diåerences between them are not caused by the visual and
auditory problems from which diabetics are likely to suåer, or by a higher incidence
of depression among diabetics than among non-diabetics, or by the fact that diabetics
are also more likely to suåer from other, additional illnesses. Diabetics were recruited
by their GPs on our behalf, and were also screened by them for any visual and
auditory problems, which might give rise to di¬culty during group testing
conditions. Only those who satis®ed these criteria were included in the study. If
spectacles had been prescribed, these were worn for all tasks. Diåerences between
groups do not re¯ect perceptual changes with diabetes.
Scores on the cognitive tests were analysed in two diåerent ways. To avoid
problems of post hoc matching an initial analysis compared all diabetics against 2191
non-diabetic controls. Note that levels of experience, and test sophistication of
patients and controls are precisely comparable because scores for diabetics were
compared with scores obtained from controls on their ®rst screenings at the
Manchester laboratory. These are given in Table 2. Diabetics had signi®cantly lower
scores than non diabetics on both AH4 parts (1) and (2) and on the CL, VFR and PR
tasks and on the Mill Hill A and Mill Hill B (see Table 2).
The diabetics as a group performed less well than all non-diabetics on all of the
tests of memory and general mental ability. They also scored less well on both parts
of the Mill Hill vocabulary test, which in a well sample would suggest educational
diåerences. However, as noted earlier, very little is known about the robustness of
vocabulary scores in the presence of pathologies.
The question of whether these results could be explained by the combined eåects
of age and diabetes were explored next. A series of 2 2 ANOVAs was carried out
¬
to check for interactions between health status (diabetic non-diabetic) and age on
}
performance on each of the cognitive tasks. In order to do this, age was re-coded as
a categorical variable, with volunteers divided into two groups, older and younger
}
than 65 years. Main eåects of age and health status (diabetic non-diabetic) were

Diabetes mellitus and cognitive ageing
355
Table 2. Means for diabetics and controls on the cognitive tasks
Diabetics
(N ¯ 168)
Controls
(N ¯ 2191)
Cognitive measures
t value
AH4 part 1
AH4 part 2
Mill Hill A
Mill Hill B
Cumulative learning
Verbal free recall
d
23±23 (12±20)
22±18 (11±27)
18±95 (6±04)
12±59 (6±95)
37±02 (10±01)
6±53 (2±65)
±82 (±52)
31±72 (1±51)
29±15 (10±89)
22±90 (4±53)
18±33 (6±04)
42±96 (8±41)
8±54 (3±52)
1±20 (±61)
t(2357)¯ 9±18***
t(2357)¯ 7±97***
t(2357)¯ 8±28***
t(2357)¯ 10±37***
t(2120)¯ 6±38***
t(2120)¯ 7±91***
t(2120)¯ 7±81***
t(1982)¯ 1±39 n.s.
t(2357)¯ 0±68 n.s.
BDI
Age (years)
8±34 (8±05)
65±17 (8±02)
7±26 (5±82)
64±74 (7±79)
!
*** p ±001.
Table 3. Means for health status by age
Cognitive
measures
Diabetic Non-diabetic
Age groups Means (SD) Means (SD)
Age¬ diabetes
interaction term
!
!
!
!
!
AH4 Part 1
65 years 25±36 (14±18) 35±01 (11±12) F(3, 2355) ¯ 1±77 p ¯ ±18
-
65 years 21±19 (9±60) 28±46 (10±95)
AH4 Part 2
65 years 25±45 (12±50) 33±03 (10±41) F(3, 2355) ¯ 0±66 p ¯ ±42
-
65 years 19±07 (8±97) 25±31 (9±97)
Cumulative learning
Verbal free recall
d scores
65 years 39±36 (9±25) 45±68 (7±32) F(3, 2122) ¯ 0±63 p ¯ ±43
-
65 years 35±16 (10±27) 40±27 (8±57)
65 years
-
65 years
7±06 (2±62)
9±52 (3±61) F(3, 2122) ¯ 2±52 p ¯ ±11
7±58 (3±13)
65 years 6±12 (2±61)
!
±88 (±62)
±76 (±41)
1±39 (±61)
1±01 (±54)
F(3, 2122) ¯ 5±51 p .05
-
65 years
signi®cant on all of the cognitive tasks with participants younger than 65 years
performing better than those older than 65 years, and non-diabetics performing
better than diabetics. However, a signi®cant interaction between age and health
}
status (diabetic non-diabetic) was found only for the signal detection theory measure
d (PR) task (see Table 3).
It was noted earlier that it is important to investigate the contribution of
depression to observed cognitive changes in diabetes. Scores on the BDI were
compared between the diabetics and controls and there was no signi®cant diåerence
in depression found between the two groups (t(1982) 1 39; p 17). Missing BDI
¯ ±
¯
±
scores were due to the fact that some respondents did not complete the
questionnaire because of the sensitive nature of the questions. Multiple regression of
age, diabetes and BDI on the cognitive variables showed ®nal models with age, BDI
and diabetes as signi®cant independent predictors of the cognitive scores. Although

356
Nuala Bent et al.
the impact of diabetes was reduced when BDI was included in the models, the
contribution of diabetes remained statistically signi®cant for all variables. Age was
the strongest predictor accounting for 14% of the variance in AH41 scores (F1, 1981
!
!
001); 14 9% in CL
±
325 68; p
001); 21% in AH42 (F1, 1981 541 69; p
¯
±
±
¯
±
±
!
!
001); and
±
(F1, 1979 345 43; p
12 3% in d scores (F1, 1979 278 02; p
diabetes accounting for a further 3 5% variance in AH41 (F1, 1980 80 55; p
001) and AH42 (F1, 1980 84 54; p
001); 10 60% in VFR (F1, 1979 233 44; p
¯
±
±
±
¯
±
!
001); BDI was a better predictor than
!
2% in
±
¯
±
±
±
¯
001); but it accounted for only
±
!
!
±
¯
±
±
memory scores. Diabetes accounted for 1% of variance in AH41 (F1, 1979 33 05;
¯
±
!
22 89; p
!
001); 1% of CL scores (F1, 1977
±
p
001) and AH42 (F1, 1979 28 44; p
001); 1% of VFR (F1, 1977 16 63; p
1977 22 60; p
±
¯
±
¯
!
!
001) and 1% of d scores (F1,
!
001) showing that the eåects of diabetes on cognitive
±
±
±
¯
±
±
¯
±
performance is independent of the eåects of depression.
The second way the data was analysed considered whether we could identify any
speci®c factors associated with diabetes such as types of treatment that might
separately or jointly contribute to these results. We compared the eåects of three
types of treatment used in the management of the conditionÐinsulin, drugs and
dietÐand again, also included age and socio-economic circumstances. The statistical
package, GLIM was used and test scores were regressed on management type, age
and social class to study main eåects and second- and third-order interactions.
Variables were deleted in hierarchical sequence, with the lowest non-signi®cant F
ratio at the 5% signi®cance level being deleted at each step, until all variables
remaining in the model had signi®cant F ratios. Social class, as assessed by HM
Registrar-General’s occupational scale (HMSO, 1980) and diabetes type were
declared as factor variables and age was treated as a continuous variable. The ®ve
sub-divisions of the HMSO scale were re-coded as (1 & 2) 1; (3N) 2 (3m) 3;
¯
and (4 & 5) 4 and declared as a 4-factor variable. Diabetes treatment management
¯
¯
¯
was also declared as a 4-factor variable, i.e.: (1) non-diabetics, (2) insulin dependent
Table 4. Regression steps in Glim analysis on cognitive measures
Measure
Age (years)
SC
AH41
AH42
F(1, 2136) ¯ 266±9***
F(3, 2136) ¯ 127±1***
F(3, 2136) ¯ 9±72**
F(1, 2136) ¯ 399±2***
F(3, 2136) ¯ 104±9***
F(3, 2136) ¯ 10±82**
d type
Measure
Cumulative learning
F(1, 2050) ¯ 249±8***
F(1, 2049) ¯ 67±3***
F(1, 2048) ¯ 9±5**
Visual free recall
F(1, 2050) ¯ 172±6***
F(1, 2049) ¯ 55±83***
F(1, 2048) ¯ 5±16*
Age (years)
SC
d type
Measure
Picture recognition
F(1, 2050) ¯ 210±3***
F(1, 2049) ¯ 40±48***
F(1, 2048) ¯ 8±62**
Age (years)
SC
d type
Note. d type diabetes type.
¯
!
!
!
*p ±05; ** p ±01; ***p ±001.

Diabetes mellitus and cognitive ageing
357
Table 5. Final regression models of age, social class and diabetes type
AH41ÐFinal model
Diabetes type 2 ¯ Beta (2) ¯ 2±25 (1±94) n.s. (insulin treatment group)
®
Diabetes type 3 ¯ Beta (3) ¯ 5±34 (1±10) (drug treatment group)**
®
Diabetes type 4 ¯ Beta (4) ¯ 5±05 (0±48) n.s. (diet treatment group)
®
Social Class 2 ¯ Beta (2)¯ 10±55 (0±64)**
®
Social Class 3 ¯ Beta (3)¯ 13±35 (0±81)**
®
Social Class 4 ¯ Beta (4)¯ 5±39 (0±48)**
®
Beta (age) ¯ 0±46 (0±03)**
®
AH42ÐFinal model
Diabetes type 2 ¯ Beta (2) ¯ 0±38 (1±80) n.s. (insulin treatment group)
Diabetes type 3 ¯ Beta (3) ¯ 5±36 (1±02) (drug treatment group)**
®
Diabetes type 4 ¯ Beta (4) ¯ 5±01 (2±06) (diet treatment group)**
®
Social Class 2 ¯ Beta (2)¯ 8±27 (0±59)**
®
Social Class 3 ¯ Beta (3)¯ 11±03 (0±75)**
®
Social Class 4 ¯ Beta (4)¯ 5±24 (0±44)**
®
Beta (age) ¯ 0±56 (0±02)**
®
Cumulative learningÐFinal model
Diabetes type 2 ¯ Beta (2) ¯ 1±55 (1±78) n.s. (insulin treatment group)
®
Diabetes type 3 ¯ Beta (3) ¯ 4±86 (0±94) (drug treatment group)**
®
Diabetes type 4 ¯ Beta (4) ¯ 2±40 (1±78) n.s. (diet treatment group)
®
Social Class 2 ¯ Beta (2)¯ 5±28 (0±44)**
®
Social Class 3 ¯ Beta (3)¯ 1±19 (0±38)**
®
Beta (age) ¯ 0±37 (0±02)**
®
Visual free recallÐFinal model
Diabetes type 2 ¯ Beta (2) ¯ 1±55 (1±78) n.s. (insulin treatment group)
®
Diabetes type 3 ¯ Beta (3) ¯ 4±86 (0±94) (drug treatment group)**
®
Diabetes type 4 ¯ Beta (4) ¯ 2±40 (1±78) n.s. (diet treatment group)
®
Social Class 2 ¯ Beta (2)¯ 5±28 (0±44)**
®
Social Class 3 ¯ Beta (3)¯ 1±19 (0±38)**
®
Beta (age) ¯ 0±37 (0±02)**
®
Picture recognitionÐFinal model
Diabetes type 2 ¯ Beta (2) ¯ 0±2226 (0±1309) n.s. (insulin treatment group)
®
Diabetes type 3 ¯ Beta (3) ¯ 0±2456 (0±07) (drug treatment group)**
®
Diabetes type 4 ¯ Beta (4) ¯ 0±4493 (±1310) (diet treatment group)**
®
Social Class 2 ¯ Beta (2)¯ 0±2982 (0±03)**
®
Social Class 3 ¯ Beta (3)¯ 0±1282 (0±03)**
®
Beta (age) ¯ 0±37 (0±02)**
®
!
!
* p ±05; **p ±01.
diabetics, (3) diabetics on hypoglycaemic drugs, and (4) diabetics treated by diet. In
some cases information on occupational status was not supplied which resulted in
slightly reduced numbers in this analysis, because social class could not be assessed.
There were signi®cant main eåects of age, social class and diabetes type and these
results are reported in Table 4.

358
Nuala Bent et al.
Final interaction term models were ®tted and tested for statistical adequacy by
checking the normality of the residuals using a Filliben correlation. All correlations
were found to be at 99 and above, indicating a close ®t to the normal distribution.
±
Having established that there were signi®cant main eåects of age, social class and
diabetes type, the ®nal regression model results demonstrate which groups show
signi®cant decrements on cognitive scores when compared with their own reference
groups (see Table 5). The reference groups used in the analysis were as follows: for
diabetes type, the reference group was the non-diabetics or group (1); the reference
group for social class was a re-coded combination of occupational categories one and
two.
The eåects of age and socio-economic advantage were as expected; older
volunteers had lower scores than younger volunteers on all cognitive tests, and
individuals in the lower social class categories had lower scores than those with
greater socio-economic advantage. The regression of treatment type showed that the
signi®cant diåerences between average scores for the combined group of all diabetics
and for all controls were due mainly to the relatively poor performance of group 3
(Beta 3), the group on hypoglycaemic drug treatment. Group 4 (Beta 4), the diet
treatment group, also performed less well than non-diabetics, but only on the AH42
intelligence test and the PR task; Group 2 (Beta 2), insulin dependent diabetics, did
not diåer from the non-diabetic controls on any cognitive tasks. There were no
signi®cant interactions found between age, social class and diabetes on any of the
"
tasks (in all cases p
1).
Duration of illness and age of diagnosis were then examined. There was no
±
"
signi®cant eåect of duration of illness on performance on any cognitive task (p
1
±
in all cases). As might be expected the three treatment groups diåered signi®cantly
in terms of age at initial diagnosis and, in particular, members of the IDDM group
had been diagnosed much earlier than the other two NIDDM groups (see Table 1).
However, age at diagnosis showed no eåect on any of the cognitive scores in our
sample.
Diabetes is often complicated by the co-occurrence of other pathologies. However,
only 45 8% of diabetics in this sample reported having other illnesses as well as
±
diabetes. There was no association between types of treatment and presence of other
#
v
illnesses ( (2)
between diabetics with and without other illnesses showed no impact of additional
87; p
99). A Mann±Whitney U test used to compare scores
¯ ±
¯ ±
"
illnesses on any of the cognitive tasks (in all cases, p
cognitive diåerences re¯ect diabetes, rather than diåerences in general health.
1). We may thus assume that
±
Discussion
Though of modest size, the sample of diabetics was larger than those reported in
most other studies. In keeping with previous results (Franceschi et al., 1984; Holmes,
1986; Holmes & O’Brien, 1995; Lichty & Klacko, 1985; Perlmuter et al., 1984;
Perlmuter, Tun, Sizer, McGlinchey, & Nathan, 1987; Richardson, 1990; Tun et al.,
1987), we found that, in comparison with 2191 non-diabetic volunteers, diabetics in
our sample were cognitively impaired on the two parts of the AH4 intelligence test
and on three memory tasks. Most previous studies have examined diåerences

Diabetes mellitus and cognitive ageing
359
}
between patients with IDDM and or NIDDM but none have yet explored the
possibility that severity of diabetes, as evidenced by the diåerent treatment regimens
given to members of the NIDDM patients, may contribute to the extent and nature
of cognitive impairment associated with NIDDM. This study compared three groups
of diabetics distinguished by their treatment regimens: treatment by insulin (IDDM
patients), by hypoglycaemic drugs (NIDDM patients) and by diet (NIDDM
patients).
The group of patients who had been treated with insulin (IDDM group)
performed as well as non-diabetics on most cognitive tasks. Eåects of illness on
cognition were apparent only in the two NIDDM groups. They were more marked
in patients on drug treatment than in patients whose illness was managed by diet.
This is consistent with earlier ®ndings that elderly non-insulin dependent diabetics
are impaired in retrieval of recently learned material, and in immediate recall (Messier
& Gagnon, 1996; Mooradian et al., 1988; Perlmuter et al., 1987; Ryan, 1988), to
which it adds evidence of a possible decrement in general mental ability.
Earlier studies have suggested that the eåects of NIDDM may interact with age in
such a way as to accelerate cognitive ageing. The present study speci®cally tested for,
but did not ®nd, interactions between the eåects of NIDDM and age on cognitive
performance. Eåects of date of diagnosis, and hence duration of illness, were also
examined, but were not detected. These last ®ndings bear on the general question as
to what are the relative contributions of `normal’ or `usual’ ageing and of pathology
to cognitive changes in later life. In the diabetic group age, as a factor on its own,
accounted for between 10 56% and 21% of variance in scores on the cognitive tasks.
±
The maximum eåect of diabetes, observed in the NIDDM group whose condition
was controlled by medication, was much less, accounting for from 1% to 2% of
variance. Judged in terms of total contribution to variance in performance the eåects
of diabetes were greatest on the cumulative learning task and equally distributed
amongst the other tasks. Diåerences in scores on the Mill Hill A and B tests may
re¯ect an eåect of diabetes, possibly through impaired retrieval from semantic
memory. However, we also have to consider that they may also be re¯ecting the
presence of other unidenti®ed factors associated with a low Mill Hill score and a
predisposition to diabetes, but as yet, there is little known to support a theory either
way. However, the eåects of diabetes on cognitive performance persist even when
socio-economic circumstances are controlled for in the model.
There was no apparent interaction between the cognitive eåects of diabetes and of
age. Thus these data suggest that diabetes makes a signi®cant, but comparatively
minor contribution to the average variance in cognitive ability between members of
an elderly population but, unlike previous results, they provide no evidence that
diabetes `accelerates the rate of cognitive ageing’.
The ®nding that NIDDM patients who were treated by diet control showed less
impairment than patients who were treated with hyperglycaemic drugs may, at ®rst
sight, seem to be evidence for diåerential success of case management. This is not a
necessary conclusion because a complicating factor is that members of the
hypoglycaemic drug group were late onset diabetics whose illness had been diagnosed
relatively later in their lives. They may have suåered from the condition for some
time before it was ®rst detected, and so have suåered cognitive impairment before

360
Nuala Bent et al.
treatment began. The authors acknowledge that stronger evidence is required to
suggest that any decrements in general mental ability found to be associated with
NIDDM are directly related to types of management, but the possibility needs to be
explored further.
More than 8% of people aged 65 years and older are aåected by NIDDM (Tun
et al., 1987). Because of its relatively high prevalence NIDDM is an important factor
in the management of healthy and competent old age. The high rate of association
of NIDDM with other age-related pathological processes is also cause for concern
(Bodansky, 1989; Goldberg & Coon, 1987; Perlmuter et al., 1984). Our replication
of ®ndings that NIDDM is associated with some degree of cognitive impairment in
later life, and the suggestion that it may be linked with the style of management,
provides a good argument for early screening for diabetes in elderly populations.
Conclusion
In contrast with earlier studies on children and young adults (Franceschi et al., 1984;
Holmes, 1986; Holmes & O’Brien, 1995) this study found no detectable cognitive
eåects of IDDM in a relatively mobile and independent elderly population. NIDDM
was associated with cognitive decrements, but these varied with treatment regimen
and were most pronounced in patients whose condition was controlled by
hypoglycaemic drugs. As diabetes had been diagnosed later in the drug treatment
group it is possible that some, or all, of their impairment had occurred before
treatment was initiated. Although the proportion of variance in cognitive
performance associated with NIDDM was signi®cant, it was much lower than the
proportion of variance associated with age alone; these results suggest that NIDDM
does contribute signi®cantly, but slightly, to variance in cognitive ability observed
in elderly populations but the absence of any interactions between the eåects of
NIDDM and age fails to support previous results that NIDDM `accelerates
cognitive ageing’. Because of the suggestion that early detection and treatment of
NIDDM may mitigate its cognitive sequelae, and so prolong independence and
quality of life, these results speak for the importance of early and more stringent
screening for diabetes in elderly populations.
References
Beck, A. T., Ward, C. H., Mendelson, M., Mock, J., & Erbaugh, J. (1961). An inventory for measuring
depression. Archives of General Psychiatry, 4, 561±567.
Bodansky, H. J. (1989). Diabetes. Pocket picture guide to diabetes. London: Gower Medical Publishing.
Deary, E. J. (1999). Eåects of repeated hypoglycemia on cognitive functionÐA psychometrically
validated reanalysis of the Diabetes Control and Complications Trial Data. Diabetes Care, 22,
1273±1277.
Ewing, F. M. E., Deary, I. J., McCrimmon, R. J., Strachan, M. W. J., & Frier, B. M. (1998). Eåect of
acute hypoglycemia on visual information processing in adults with type 1 diabetes mellitus.
Physiology & Behaviour, 64, 653±660.
Franceschi, M., Ceccetto, R., Minicucci, F., Smizne, S., Baio, G., & Canal, N. (1984). Cognitive
processes in insulin-dependent diabetes. Diabetes Care, 7, 228±231.
Goldberg, A. P., & Coon, P. J. (1987). Non-insulin-dependent diabetes mellitus in the elderly.
In¯uence of obesity and physical inactivity. Endocrinology & Metabolism Clinics of North America, 16,
843±865.

Diabetes mellitus and cognitive ageing
361
Harris, M. I. (1990). Epidemiology of diabetes mellitus among the elderly in the United States. Clinics
in Geriatric Medicine, 6, 959±970.
Heim, A. W. (1970). AH4 Test. Windsor: NFER-Nelson.
HMSO (1980). Classi®cation of occupations. London: O¬ce of Population Census and Surveys.
Holland, C. A., & Rabbitt, P. (1991). The course and causes of cognitive change with advancing age.
Social and Psychological Gerontology, reviews in Clinical Gerontology, 1, 81±96.
Holmes, C. S. (1986). Neuropsychological pro®les in men with insulin-dependent diabetes. Journal of
Consulting and Clinical Psychology, 54, 386±389.
Holmes, C. S., & O’Brien, B. (1995). Cognitive-functioning and academic achievementin Children with
insulin-dependent Diabetes-Mellitus (IDDM). School Psychology Quarterly, 10, 329±344.
Holmes, C. S., Koepke, K. M., & Thompson, R. G. (1986). Simple versus complex performance
impairments at three blood glucose levels. Psychoneuroendocrinology (England), 11, 353±357.
Krieger, L. M., & Rudberg, M. A. (1994). Health care costs of older persons with non-insulin
dependentDiabetes Mellitus : A perspective from the United States. Facts and Research in Gerontology, 8,
287±293.
Lawson, J. S., Williams-Erdahl, D. L., Monga, T. N., Bird, C. E., Donald, M. W., Surridge, D. H. C.,
& Letemendia, F. J. J. (1984). Neuropsychological function in diabetic patients with neuropathy.
British Journal of Psychiatry, 145, 263±268.
Lichty, W., & Klacko, D. (1985). Memory in type 1 Diabetics (abstract). Diabetes 35, (Suppl. 1) 19A.
Lowe, L. P., Tranel, D., Wallace, R. B., & Welty, T. K. (1994). Type 11 diabetes and cognitive
function: A population based study of Native Americans. Diabetes Care, 17, 891±896.
®
Messier, C., Desrochers, A., & Gagnon, M. (1999). Eåect of glucose regulation, and word imagery
value on human memory. Behavioral Neuroscience, 113, 431±438.
Messier, C., & Gagnon, M. (1996). Glucose regulation and cognitive functionsÐrelation to
Alzheimers-disease and Diabetes. Behavioural Brain Research, 75, 1±11.
Mooradian, A. D., Perryman, K., Fitten, J., Kavonian, G. D., Morley, J. E. (1988). Cortical function
in elderly non-insulin dependent diabetic patients. Behavioural and electrophysiologic studies.
Archives of Internal Medicine, 148, 1369±1372.
Perlmuter, L. C., Hakami, M. K., Hodgson Harrington, C., Ginsberg, J., Katz, J., Singer, D. E., &
Nathan, D. M. (1984). Decreased cognitive function in ageing non insulin-dependent diabetic
patients. American Journal of Medicine, 77, 1043±1048.
Perlmuter, L. C., Tun, P. A., Sizer, N., McGlinchey, R. E., & Nathan, D. M. (1987). Age and diabetes
related changes in verbal ¯uency. Experimental Aging Research, 13, 9±14.
Prescott, J. H., Richardson, J. T. E., & Gillespie, C. R. (1990). Cognitive function in diabetes mellitus:
The eåects of duration of illness and glycaemic control. British Journal of Psychology, 29, 167±175.
Rabbitt, P. (1993). Does it all go together when it goes? The nineteenth Bartlett Memorial Lecture. The
Quarterly Journal of Experimental Psychology, 46A, 385±434.
Rabbitt, P. M. A., Donlan, C., Watson, P., Bent, N., & McInnes, L. (1995). Unique and interactive
eåects of depression, age, socio-economic advantage and gender on cognitive performance of normal
healthy older people. Psychology and Aging, 10, 307±313.
Raven, J. C. (1965). The Mill Hill Vocabulary Scale. London: H. K. Lewis.
Richardson, J. T. E. (1990) Cognitive function in diabetes mellitus. Neuroscience and Behavioral Reviews,
14, 385±388.
Ricklefs, R. E., & Finch, C. E. (1995). Aging: A natural history. New York: Scienti®c American Library.
Robertson-Tchabo, E. A., Arenberg, D., Tobin, J. D., & Plotz, J. B. (1986). A longitudinal study of
cognitive performance in non-insulin dependent (type 11) diabetic men. Experimental Gerontology, 21,
459±467.
Ryan, C., Vega, A., & Drash, A. (1985). Cognitive de®cits in adolescents who developed diabetes early
in life. Pediatrics, 75, 921±927.
Ryan, C., Vega, A., Longstreet, C., & Drash, A. (1984). Neuropsychological changes in adolescents
with insulin-dependent diabetes. Journal of Consulting and Clinical Psychology, 52, 335±342.
Ryan, C. M. (1988). Neurobehavioural complications of type 1 diabetes: Examination of possible risk
factors. Diabetes Care, 11, 86±93.

362
Nuala Bent et al.
Scott, R. D., Kritz-Silverstein, D., Barrett-Connor, E., & Wiederholt, W. C. (1998). The association
of non-insulin-dependent diabetes mellitus and cognitive function in an older cohort. Journal of the
American Geriatrics Society, 46, 1217±1222.
Silberman, C., Souza, C., Wilhems, F., Kipper, L., Wu, V., Diogo, C., Schmitz, M., Stein, A., &
Chaves, M. (1995). Cognitive de®cit and depressive symptoms in a community school of elderly
peopleÐA preliminary study. Revista De Saude Publica, 29, 444±450.
Sunderland, A., Harris, J. E., & Baddeley, A. D. (1983). `Do laboratory tests predict everyday
memory? A neuropsychological study’. Journal of Verbal Learning and Verbal Behavior, 22, 341±357.
Tun, P. A., Perlmuter, L. C., Russo, P., & Nathan, D. M. (1987). Memory self-assessment and
performance in aged diabetics and non-diabetics. Experimental Aging Research, 13, 151±157.
Worrall, G., Moulton, N., & Briåett, E. (1993). Eåect of type 11 diabetes mellitus on cognitive
function. Journal of Family Practice, 36, 639±643.
WHO (1984). The uses of epidemiology in the study of the elderly. Report of a WHO Scienti®c Group
on the Epidemiology of Aging. World Health Organization Technical Report Series 706, Geneva: World
Health Organization.
Zaslavsky, L. M. A., Gross, J. L., Chaves, M. L., & Machado, R. (1995). Memory disfunction and
autonomic neuropathy in non-insulin dependent (type 2) diabetic-patients. Diabetes Research and
Clinical Practice, 30, 101±110.
Received 19 June 1998; revised version received 1 December 1999