To the best of our knowledge this paper is the first extensive description of non-participants in a population-based MRI study. The present study demonstrate that participants volunteering and successfully completing MRI scanning were not widely different from those who did not participate, and self-reported health did not differ between them. Notably, however, participants were less often obese, had a higher level of education, and were somewhat younger than MRI-np and MRI-ni. Risk factors for cardiovascular disease (high BP and cholesterol) were less prevalent among participants. The participants were also more likely to be employed. Additionally, HADS-score was found to be lower among participants, indication less psychological symptoms. It appears that most of the differences were present at all survey points. However, cholesterol and HADS-score were only available from the last two surveys, therefore, it is difficult to ascertain differences between the groups with regard to these factors were present from the first study or had developed over time.
The main objective of this study of non-participants was to enable a careful evaluation of the generalizability of results from future MRI-HUNT analyzes, which has rarely been possible in previous population-based MR-studies [2
]. Strengths of the study were the large number of participants, the population-based design, and the long follow-up (>20
years) with three data points for each participant.
A limitation of the study was that some questions were not filled out by every participant, but the problem with missing data was not extensive on each question, and was unlikely to influence results. It should be noted that 66 out of 1560 eligible and selected candidates were not invited due to stratification, but we had to count them as invited, because it was impossible to trace them in the de-identified data file. We cannot see that this could markedly influence the results. One may also note that all three groups consisted of individuals who had participated in all three HUNT-studies and therefore might be more compliant than the rest of the population. Also, multiple comparisons increase the risk of type I error. To avoid false positive results, a Bonferroni-adjusted p-value of 0.001 was chosen for the univariate analyses.
Participation rates have declined in HUNT 1, 2 and 3(88%, 71% and 54%). Such decline in epidemiological studies seems to be a general tendency in later years [1
]. Therefore, it has become increasingly important to analyze characteristics of non-participants. However, evaluation of non-participants in MRI studies is in general lacking, and if done, is mostly restricted to demographic variables [14
]. A Finnish study examining non-participation rates among patients with psychiatric illnesses suggested that subjects with psychosis were less likely to participate in an MRI-study [15
]. Similarly, in the present study participating women had lower HADS-D than non-participating, possibly indicating a lower burden of psychiatric illness. One may speculate whether subjects with higher level of anxiety or depression tend to avoid MRI for fear of the investigation itself, or of the result.
There is a considerable decline in participation rates from the first to the last HUNT study. It seems that the reasons for not participating were quite similar in HUNT 1 and 2, but with some differences. In both, being busy and having moved were the main reasons, but having health problems were specific for HUNT 1, and forgetting the invitation and not having the desire to participate, were only reported in HUNT 2. Self-reported reasons for not participating are not available in HUNT 3. Thus it is not possible to ascertain whether the reasons for not participating differed from those in the first two studies.
There were slight, but significant, differences with regard to clinical characteristics and presence of risk factors between the three groups. This finding shows the need to take into account differences in risk factor profiles at baseline in participants versus non-participants in future population based MRI studies. Importantly, power might be weakened due to lower prevalence of people with risk factors in the study population. This will, however, not have any effect on associations or risk analyses. One of the exclusion criteria was weight >150
kg, but this probably does not explain the lower BMI among participants, since only one individual was above this weight in HUNT 3.
However, place of living within Nord-Trøndelag is a factor that probably accounts for part of the difference. In the MRI-HUNT study, the participants had to live <45 minutes of travel from the town where the scans were performed (Levanger), due to budget restraints, and to increase participation. In all three HUNT-studies, a higher proportion with obesity and lower education levels has been found in rural communities [8
], and the higher BMI may also explain the higher BP and cholesterol among MRI-ni.
This cannot explain differences between MRI-np and MRI-p, because both groups lived in the same area. The lower level of education and increased BMI and BP among MRI-np may be explained by generally lower participation rates among individuals with lower education and poorer health [16
]. Conceivably, higher BMI among MRI-np compared to MRI-p may also be a result of overweight people, even those well below 150
kg, tend to refrain from participation in fear of being too big for the scanner. Different proportions of obese individuals might further have contributed to differences in other health related measures (cholesterol and BP).
Cardiovascular risk factors (like obesity and hypertension) are related to a risk of stroke and TIA, and also to alterations in brain morphometry [17
].Lower participation rates among those with high cardiovascular risk could therefore lead to an underestimation of vascular brain changes in the general population. The prevalence of these changes in the MRI-p will therefore most probably represent the minima, and to some extent one can correct for the bias. In other population-based MRI-studies, various types of bias may be present, probably related to the mode of recruitment and a host of other factors, but their direction and magnitude are largely unknown.