In this very large population-based sample of middle-aged Danish adults with 898,285 person-years of follow-up, a single round of diabetes screening and cardiovascular risk assessment was associated with a 21% reduction in all-cause mortality rate and a 16% reduction in CVD events between 2001 and 2012 in individuals diagnosed with diabetes between 2001 and 2009. Individuals with clinically diagnosed diabetes were identified on average 2.2 years later than individuals with diabetes detected by screening.
One argument for considering screening for type 2 diabetes is the historical observation [22] that there is an extended latent or pre-clinical phase (lead time) in which people could be diagnosed and during which earlier treatment might have a beneficial long-term effect. More recently, changes in clinical practice with greater testing and public awareness have probably led to a shortening of this latent period. Data from the parallel-group population-based Ely study suggested that the lead time is relatively short at 3.3 years [23]. This is comparable with our more contemporary estimate of 2.2 years. However, the lead time may be longer in less developed health systems and/or in more deprived populations. In addition, the historical estimate of 9 to 12 years by Harris et al. [22] and the more recent estimate of 6 years from Porta et al. [24] relates to the true point of onset of diabetes. This is not the same as the point at which diabetes is detectable by screening, especially if screening is infrequent and not 100% sensitive. The period between true onset and clinical diagnosis of diabetes may be long precisely because there are few clinical manifestations during this period [23].
Even with a relatively modest lead time, our screening programme was associated with a significant reduction in mortality and incident CVD in individuals with diabetes over 6 years of follow-up. As only 10.1% of individuals in the screening group were actually diagnosed by screening, it is likely that the programme had wider effects in this cohort e.g. by delaying diagnosis and providing lifestyle advice (and perhaps treatment) among those screened and found to be at risk who were later diagnosed clinically. The difference in mortality and CVD might also have been driven by screening practices that were vigilant for diabetes even after the programme had finished, contributing to continued earlier detection and the higher diabetes incidence observed in screening compared with no-screening practices.
Our findings support results from modelling studies showing that screening programmes could contribute to a reduction in risk of mortality and cardiovascular morbidity in screen-detected individuals [1,2,3,4,5]. Herman et al. used a validated computer simulation model in ADDITION-Europe to show that screening and routine care, compared with a 3 year delay in diagnosis and routine care, was associated with a 17% relative risk reduction in all-cause mortality after 5 years [5]. They argue that the benefits of screening and treatment primarily accrue from early diagnosis and by hastening the treatment of CVD risk factors in the lead time [5]. We observed a rapid increase in the proportion of screen-detected individuals who redeemed cardioprotective medication during follow-up. However, larger proportions of clinically diagnosed individuals in the no-screening group redeemed medication compared with clinically diagnosed individuals in the screening group. As individuals in the no-screening group were diagnosed at a later stage in the disease trajectory, they may have had higher cholesterol, blood glucose and blood pressure values at diagnosis compared with the screening group, necessitating higher levels of cardioprotective medication. It is likely that promotion of healthy behaviour change also impacted on CVD and mortality rates in the screening group. Those who attended screening reported their smoking status at baseline (28%), which was similar to national self-reported smoking prevalence data in 2004 (Danish National Health Service survey) [25]. One-third of screen-detected individuals in ADDITION-Denmark reported that they had stopped smoking at 5 year follow-up. Furthermore, this cohort lost an average of 2 kg in weight [12]. If similar behavioural responses were observed among other individuals diagnosed with diabetes in the screening group, this suggests potential mechanisms for the risk reduction observed other than prescribed treatment. We also observed lower rates of cancer incidence in the screening group, which might be linked to changes in health behaviour and prescribing [26].
In a separate paper [11], we examine the impact of the ADDITION-Denmark screening programme at the population level, e.g. comparing all individuals aged 40 to 69 years in the screening and no-screening groups, and showed no long-term reduction in mortality or CVD. As such, our results mirror those from trials of screening for other conditions, which have shown reductions in disease-specific mortality but not in overall mortality [27]. There appeared to be beneficial effects for all those diagnosed with diabetes in the screening practices, regardless of the mode of diagnosis e.g. by screen detection or by clinical diagnosis. However, this benefit is too small to impact on overall population risk of CVD events and mortality [10, 11].
Strengths and limitations
This very large controlled trial with long-term follow-up included all individuals aged 40 to 69 years diagnosed with diabetes in Denmark between 2001 and 2009. Outcome ascertainment was robust. The National Death Registry estimates 100% coverage of mortality based on death certificates. All-cause mortality is an all-inclusive measure that addresses both direct and indirect effects of screening, and puts disease-specific mortality reduction in the context of other competing risks [27]. We were able to ascertain which individuals were living in Denmark in 2001 and censor those who emigrated during follow-up. Deaths and CVD events were coded blind to study group.
Our definition of clinically diagnosed diabetes was a proxy measure based on date of inclusion in the Danish National Diabetes Register, where individuals are classified as having diabetes according to a number of criteria [19, 28]. Using registry-defined diabetes ensures that the entire Danish population is covered by uniform inclusion criteria and the dropout rate is nil. However, we did not have formal clinical diagnosis of diabetes or the date of diagnosis. A recent validation of the algorithm for including individuals in the Register suggests that it has a sensitivity ≥95% and a positive predictive value of around 80% [19]. The same report also suggests that around 20% of diabetes diagnoses in the Register may represent false-positive inclusions of people with frequent measurements of blood glucose who do not have diabetes. This may help account for the higher incidence of diabetes in the screening group and help explain the lower levels of cardioprotective medication redeemed by clinically diagnosed individuals in the screening group. Many high-risk individuals would have undergone frequent measurements of blood glucose during the screening programme, delineating them with a diabetes diagnosis on the Register, when they did not in fact have diabetes and were therefore unlikely to receive cardioprotective treatment.
In total, 1406/1533 (92%) of individuals diagnosed with screen-detected diabetes in ADDITION-Denmark were added to the Diabetes Register with a median delay of 56 days (R. K. Simmons, unpublished data). While this proxy date means our estimate of diabetes duration is probably shorter than the actual length, this is unlikely to be differential by group. By only including individuals aged 40 to 69 years in our study, we assume the number of clinically diagnosed type 1 diabetes cases is likely to be low and similar in both groups.
A limitation of our study is the non-randomised design; we cannot eliminate the possibility of selection bias and residual confounding. Groups were well balanced for most characteristics at baseline. However, our findings might have been influenced by the higher levels of education and the slightly lower levels of pre-existing chronic disease in the screening group. We did adjust for age, sex, education and prevalent chronic disease, which had a small impact on the effect size. Including adjustment by county had a large impact, reducing the effect size considerably, though the hazard remained significant. It is likely that adjusting for county took account of some of the potential socioeconomic differences across different regions in Denmark.
We tried to minimise lead and length time biases by comparing outcomes for all individuals diagnosed with diabetes in the screening and no-screening groups. Further, we extended the inclusion period to 3 years beyond the end of the formal screening programme so that people who could have been detected by screening (had they been in the screening group) were included. However, the small difference in the overall incidence between the groups suggests that some of the observed effect may be due to residual lead and length time bias. Participation in the programme may also have impacted on subsequent diabetes detection rates in screening practices.
While we were able to compare trends in redeemed cardioprotective medication to explore a potential mechanism for the observed difference in outcomes, we did not have population-level data on dietary, physical activity or smoking behaviour. The majority of participants were white, the main ethnic group in Denmark, which also limits generalisability to other settings.
In conclusion, a single round of diabetes screening and cardiovascular risk assessment in middle-aged Danish adults performed in general practice was associated with a 21% reduction in all-cause mortality rates and a 16% reduction in CVD events between 2001 and 2012 among people diagnosed with diabetes between 2001 and 2009. Screening resulted in cases being identified, on average, 2.2 years earlier.