An investigation of the population impact of variation in HbA_1clevels in older people in England and Wales: From a population based multi-centre longitudinal study

The Medical Research Council Cognitive Function and Ageing study (CFAS) is a multi-centre, longitudinal study of people aged 65 years and over (over 18,000 individuals at baseline). It is predominantly a two-phase population based interview study, with annual or biennial follow-ups. There are six CFAS centres, two in rural areas (Cambridgeshire and Gwynedd, Wales) and four in urban areas (Nottingham, Newcastle, Oxford and Liverpool). Details of the CFAS study design have been published previously [14, 15] (see Figure 1 for a simplified flow diagram).

Blood was collected at the third wave of interview in 1996 – 1998, between four to six years after the start of the study in all centres except Oxford. All those who had previously received an assessment interview including the study diagnosis were invited to participate in this wave of the study. Sampling at prevalence assessment was weighted towards cognitive impairment. Blood samples were requested from all respondents, and samples were collected from 69% (1,150 out of 1654). An EDTA-anticoagulated blood sample was sent by post to Cambridge, and assayed at the Department of Clinical Biochemistry, University of Cambridge. The HbA1c assays were carried out using high-performance liquid chromatography (HPLC) on a Bio-Rad Diamat Automated Glycosylated Haemoglobin Analyser (Hemel Hempstead, United Kingdom) that was DCCT-aligned. The HbA_1c readings were fed back to the GPs. 11 samples were undated, and were excluded. The HbA_1c measurements from 1,139 respondents were used in the analysis. A cut-off point of 7% has been used during this investigation for diagnosis of diabetes, as it has been reported to provide produced the maximum sum of sensitivity and specificity [16] and has been used in similar studies such as EPIC-Norfolk study [8]. The respondents were divided into five categories: those with self reported diabetes, those with a HbA_1c level ≥7% but who did not report they had been previously diagnosed with diabetes, as we believe that some of the people who were diagnosed with diabetes have been treated and monitored, therefore there will be some different impacts on cognitive function and physical function between these two groups. Three further groups defined by classifying the remaining respondents as low, medium and high tertiles of HbA_1c, the ranges for the three groups are 3.7%–5.2%, 5.3%–5.7% and 5.8%–6.9%,

Two years after the blood collection interview, a subgroup of respondents was interviewed again, and after another two years, all surviving respondents (excluding Liverpool) were interviewed (year 10 interview). 524 out of the 1139 respondents with a HbA1c measurement had at least one follow up interview. The median follow-up time from blood collection was 5 years.

Information on residence, marital status, main occupation, social and service contacts, physical health and well-being was collected at baseline interview. Change in any of these variables was obtained from follow-up interviews. Data on health conditions including detailed cognitive function and physical function are available both before and after the HbA_1c measurement in those individual with follow up interviews.

Risk factors for cognitive impairment measured by self report or informant at baseline interview included history of diabetes, heart attack, stroke and hypertension. The participants with self reported diabetes were defined as those answered "Yes" to the question: "Have you ever had sugar diabetes?" in any interviews. Smoking status was classified into three categories: non-smoker (never smoked), ex-smoker (stopped at least 5 years ago) and current smoker.

All participants were flagged for death notification at the Office of National Statistics (ONS). There had been 619 deaths in the HbA_1crespondents by the end of 2004 (version 8.0 of data). Cause of death was established from death certificates. Cardiovascular death was defined as International Classification of Disease Revision 9 (ICD9) codes 400–438, ischaemic heart disease death as ICD 9 codes 410–414 and cerebro-vascular disease death as ICD 9 codes 430–438 anywhere on the death certificate.

Cognitive function was measured using the Mini Mental State Examination (MMSE) score [17], which tests a range of cognitive abilities and provides a total score from 0 to 30, and dementia of diagnosis using the computerized algorithm AGECAT organicity level of the Geriatric Mental State Examination (GMS Automated Geriatric Examination for Computer Assisted Taxonomy) [18].

Respondents have been classified as demented, not demented and missing diagnosis using the criteria:

Individuals unable to complete the full interview were diagnosed using vignette, informant information and interviewer observations.

An incident dementia case was defined as a respondent becoming demented after the interview at which blood was collected.

The activities of daily living (ADL) impairment has been defined using the hierarchy of IADL (instrumental activities of daily living) and/or ADL [19] and based upon the concept of interval of need [20]. Participants' mobility was also rated by the interviewer, as usually ambulant non-housebound, usually ambulant housebound, chairfast permanently, and bedfast permanently. An incident ADL impaired case was defined as a respondent becoming impaired after the interview when blood was collected.

Due to differences in study design and data collected, individuals in Liverpool (n = 168) only contributed to the analysis of mortality.

Descriptive analysis was carried out to explore the characteristics of the population, the distribution of variables and the relationship between HbA_1c and vascular diseases risk factors. Analysis of variance (ANOVA) was used to investigate differences on HbA_1c levels between categories of demographic variables and risk factors. Linear regression was used to investigate trends in continuous and ordinal variables.

Survival analysis was used to investigate the association between HbA_1c level and mortality from all causes, cardiovascular disease, cerebral disease, ischaemic heart disease and other causes. Cox proportional hazards models were used to determine the contribution of risk factors to mortality. Logistic regression models have been used to examine of relationships between HbA_1c level and incidence of dementia, and incidence of physical disability. Proportional hazards assumptions were tested using Kaplan-Meier curves. The analysis uses inverse probability weighting to adjust for the study design using methods described previously [21]. Analysis was performed with STATA version 8.

All CFAS interviewing and blood collection have been covered under local and multi-centre ethical approval.

All research has been undertaken independently of the funding bodies.

HbA_1c levels ranged from 3.7% to 13.9% with a skewed distribution with median 5.6%, and mean 5.8%. 91% of values were below 7%, and no differences were found between centres (p = 0.83).

Table 1 shows the characteristics of 1,139 respondents categorised by HbA_1c level and self reported diabetes. Men without previously diagnosed diabetes were more likely to have a HbA_1c level ≥7%. Respondents in HbA_1c level ≥7% category were older and had lower MMSE scores. Those respondents who were in either self reported diabetes or HbA_1c level ≥7% categories were more likely to have IADL and/or ADL impairment. Respondents with diagnosed diabetes were more likely to report having had a heart attack or stroke.

The mean value of HbA_1c for smokers and those who reported high blood pressure was systematically higher in the 1,004 respondents whose HbA1c was under 7%. Otherwise there were no systematic or significant differences between the groups with 'normal' HbA_1c levels.

By the end of 2004, 54% of the 1139 respondents had died. The number of deaths was similar in each the first five follow up years, and increased sharply thereafter. The median follow up time was 3 years from blood collection and interquartile range was 1.4 to 5.0 years for those who had died.

Figure 2 displays the cumulative mortality curves from all cause deaths by five categories of respondents: those with self reported diagnosed diabetes, those without diagnosed diabetes but had HbA_1c ≥7% and the three tertiles of the population with HbA1c<7%. Individuals without diagnosed diabetes but who had HbA_1c ≥7% had higher mortality than all others. Table 2 shows age and sex adjusted mortality hazard ratios for these groups. Respondents whose HbA_1c was greater or equal to 7% but had not previously been diagnosed with diabetes had twice the rate of death from all causes and dying from cardiovascular disease compared with respondents whose HbA_1c was in the lowest tertile. Respondents who reported diabetes also had nearly double the death rate from cardiovascular disease compared with respondents whose HbA_1c was in the lowest tertile. The relative risks of ischaemic heart disease and cerebro vascular disease mortality were higher in the respondents who reported diabetes or whose HbA_1c was greater or equal to 7% but had no self-reported diabetes than the lowest HbA_1c tertile, although they are not significant at 5% level. There was no significant difference between the three groups in which respondents' HbA_1c was less than 7%, although there may be a slight trend with increasing tertiles. Adjusting for smoking status did not change the patterns in the mortality hazard ratios.

Table 3 shows the relations between HbA_1c level and/or diabetes status and mortality from all cause, cardiovascular, ischaemic, cerebro vascular and non-cardiovascular disease modeled using Cox multivariable regression models adjusted for age and sex only and additionally for history of heart attack, stroke, hypertension and smoking status respectively. There is a continuous relationship between HbA_1c and mortality on all cause, cardiovascular and ischaemic heart disease after taking of account of age, sex and some known risk factors even including diabetes status. An increase of 1% in HbA_1c was associated with about 10% (95% CI: 1% to 24%) increase in all cause mortality and 20% increase in cardiovascular (95% CI: 7% to 37%) or ischaemic disease (95% CI: 4% to 44%) mortality. However there was no significant association between self reported diabetes status and mortality. The risk associated with HbA1c is independent of self-reported diabetes. After excluding the respondents with self-reported diabetes or with a HbA_1c level ≥ 7%, the relative risks of all cause, cardiovascular and ischaemic heart disease were similar, but were no longer significant.

There were 99 people with self reported diagnosed diabetes and of these one third had HbA_1c concentration levels below 7%. To examine whether this group of respondents differed with respect to mortality from respondents who reported diabetes but had HbA_1c concentration levels above 7%, we separated the respondents into four groups:

Table 4 shows that respondents who had HbA_1c ≥ 7% in self reported diabetes group had significant higher hazard rate of dying from cardiovascular disease and ischaemic heart disease compared with the group of respondents who had HbA_1c < 7% and without self reported diagnosis of diabetes even after adjusting for age and sex. Individuals who reported diabetes but were well controlled according to HbA_1c level had similar rates of death compared with individuals who do not have diabetes, although the confidence intervals are wide.

In this sample, dementia increased with age and was more common in women (9.5%) than men (2.6%). There was no cross sectional association between HbA_1c level and prevalent dementia, adjusted for sex and age at baseline.

There were 67 incident cases of dementia identified in 453 individuals during an average 5 years follow up since HbA1c measurement. Dementia in this sample increased with age and more women developed dementia (14.6%) than men (4.9%). Table 5 shows the percentage of respondents developing dementia over the 5 years, in HbA_1c ≥ 7% category this percentage was considerably higher than other groups. The percentages in the three other HbA_1c groups were similar. The percentage developing dementia in the self reported diabetes group was low. The crude death rate by 5 years was 70% among people with self reported diabetes compared to 52% in participants with HbA_1c level < 7%.

A logistic regression model found that respondents in the group HbA_1c ≥ 7% who had not been diagnosed with diabetes had significant higher risk (odds ratio = 4.8, 95% CI:1.1 to 21.6) of developing dementia than the respondents with HbA_1c < 7% even after adjusting for age and sex.

Table 1 shows that respondents who were in both self reported diabetes category and no self reported diabetes but with HbA_1c level ≥7% category were more likely to have IADL and/or ADL impairment at the time blood was collected, even after adjusting for age and sex. But there was no relationship between IADL and/or ADL impairment and levels of HbA_1cunder 7%.

114 respondents with no IADL or ADL impairment at the time of blood collection had developed one in the five follow up years. Overall there was no relationship between the incidence of IADL and/or ADL impairment and HbA_1c level or diabetes status. Women with HbA_1c level ≥7% had 60 % higher risk than women with HbA_1c level in the lower tertiles after adjusting for age, but confidence intervals are wide (95% CI: 0.3, 9.5).