Strengths and limitations
One of the main strengths of this research is that the study population is representative of the current working demographic in Carmarthenshire, South Wales, who volunteered to participate in a workplace-based diabetes risk assessment. The details of these employees would not be routinely available if not for the Prosiect Sir Gâr initiative and therefore this information provides an insight into the current diabetes ‘risk’ of the workforce in Wales.
This study is also the first to compare directly the proportion of individuals that was predicted to be at high risk by four validated and routinely used risk assessments. The risk assessments were chosen in this study primarily because they feature in the NICE guidance,2 and also have five common risk variables (age, sex, BMI, family history of diabetes, and currently prescribed antihypertensive medication; Table 1) that make direct comparisons feasible.
One of the limitations to this study and also to the current literature is that, unlike in validated CVD risk prediction algorithms,8 no prospective studies have compared diabetes risk prediction models to measure the accuracy and false–positive rates of these current models. Thus, this admission from the current literature offers a suggestion for important future research.
Comparison with existing literature
It is acknowledged that in the development and validation stages of some of the risk assessment models included in this study some comparisons have been previously made between the risk-assessment tools. During the validation of the QDiabetes model, comparisons were made with the Cambridge Risk Score. This validation demonstrated that the QDiabetes model improved discrimination; however, because the Cambridge Risk Score does not give a prediction of absolute risk, calibration measures between the two risk scores could not be determined.3 The inclusion of ethnicity in the QDiabetes model could explain these differences, with previous research concluding that ethnic-specific cut points need to be established when using the Cambridge Risk Score in a multi-ethnic population.14 Ethnic group is an important consideration in the FINDRISC model, which was developed in a white population. There is a tendency for diabetes risk questionnaires developed in white populations to underperform in multi-ethnic populations,15 and this observation provided the rationale for the Leicester Risk Assessment, which was based on the FINDRISC example2 and validated for use in a multi-ethnic population in the UK.6 Interestingly, in comparison with the FINDRISC model, a score of ≥16 on the Leicester Risk Assessment increased the number of individuals identified with impaired glucose regulation rather than a score of ≥9, which was indicative of drug-treated diabetes using the FINDRISC questionnaire.2 This offers a suggestion of why differences were observed in the numbers of individuals predicted to be high risk between these two risk assessments.
Other previous research has established that the Cambridge Risk Score had no advantage in identifying diabetes risk compared with BMI alone.16 This finding demonstrates the importance of waist circumference in predicting diabetes risk, especially given that previously evidence has demonstrated a clear association between ‘central obesity’ (waist circumference of ≥102 cm in males and ≥88 cm in females) and diabetes risk, regardless of BMI values.17
Implications for research and practice
The findings from this study raise a question about the validated risk assessments currently advocated by NICE and the correct approaches to reduce the ever-increasing prevalence of type 2 diabetes in the UK.2 There are three apparent options in terms of risk-assessment tools that can be taken from the present observations. First, an aggressive approach could be taken by favouring the Cambridge Risk Score. Although the limitation of this model has been discussed, this approach potentially could benefit males who develop type 2 diabetes at a lower BMI value than their female counterparts.18,19 The second method could be taking a conservative approach by prioritising adoption of the Leicester Risk Assessment, which also allows layperson completion. This risk tool did predict the lowest proportion of males, females, and, subsequently, all participants at high risk of developing type 2 diabetes, however, which could overlook some at-risk individuals. The third, more practical, and cost-effective approach would be to use one of the two tools that predicted ∼6% of all participants at high risk, either FINDRISC or QDiabetes.
Interestingly, lifestyle intervention through exercise and diet has been proven to prevent type 2 diabetes in high-risk individuals,20 and has also been shown to be more effective than metformin at reducing the incidence of diabetes in a group of high-risk individuals with impaired glucose regulation.21 This intervention focused on weight loss of at least 7% and individuals partaking 150 minutes of physical activity per week.21 Impaired fasting glucose (IFG) where concentrations are ≥6.1 mmol/L1 have also shown strong associations with individuals developing type 2 diabetes compared with those individuals with fasting blood glucose below this threshold value.22,23 Only the FINDRISC tool accounts for physical activity and history of high blood glucose, both of which are positive aspects to the questionnaire. However as discussed previously, the FINDRISC tool does not account for ethnic group, which is fine if the population is Europid, but with a large South Asian population in the UK reservations should be given to prioritising this assessment. Moreover, since the introduction in 2011 of HbA1c as a diagnostic criterion for type 2 diabetes,24 the performance of FINDRISC has reduced.25
From a clinical standpoint, given the discrepancies in the numbers of predicted high risk individuals, a more practical approach would be to focus more on isolated risk factors (for example, adverse family histories, physical inactivity, elevated waist circumference), irrespective of diabetes risk prediction values. There is merit for everyone to benefit from lifestyle advice that reinforces the benefits of regular physical activity and a balanced diet, rather than wait for the individual to be deemed high risk. This strategy also has the potential to convert those individuals currently predicted at ‘intermediate’ or ‘increased’ risk to ‘low’ risk. Nevertheless, while diabetes risk assessments remain the recommended primary tool to identify individuals at high risk,2 the risk assessments that include waist circumference as a risk factor should be prioritised in primary care and prevention because of the overwhelming evidence of ‘central obesity’ being a better indicator of diabetes risk than BMI.17,26
A further finding was no significant concomitant relationship between age and changes in risk scores, with no apparent rise in diabetes risk prediction from 50 years old in males and females. This is somewhat surprising given the inclusion of age as a risk factor and the varying weightiness by coefficient in each of the diabetes risk assessments.3–6 This finding is important, however, as it provides additional evidence to the current targeted age groups as documented in current government guidelines,2 and potentially offers a suggestion to slightly amend this documentation to target those individuals aged ≥50 years instead, while still providing lifestyle advice to adults aged <50 years old. In addition, epidemiological research has previously reported that diabetes risk increases with age, with one explanation being that glycaemic control as reflected in HbA1c scores revealed a significant increase from 50 years onwards.27
Unfortunately, it seems that in the 10 years since the FINDRISC model was introduced, and despite new models for predicting risk of type 2 diabetes being introduced in great numbers globally and annually with an increase of focus on layperson completion (for a review, see Noble and colleagues28), discrepancies remain in numbers of individuals at high risk. The changes in diagnostic criteria for diabetes, which now incorporate HbA1c values, also have been shown to reduce performance in some of the more established diabetes prediction tools. Encouragingly, an emerging model has been developed recently that has accounted for these diagnostic changes involving HbA1c in its predictive capability.29 Differences were observed in predicted high risk individuals using the currently advocated risk assessments, and therefore caution should be addressed when categorising such individuals at high risk. In primary care, until a consensus is made on the diabetes risk prediction ‘model of choice’, more focus should be on isolated risk factors, especially regarding lifestyle choices and evidence of ‘central obesity’ irrespective of diabetes risk prediction score. Furthermore, with the change in diagnostic criteria, it is also important that any new risk prediction assessments allow for these in their respective algorithms.
