In this natural history cohort, we found significantly increased levels of disease activity, as defined by new T2 lesion occurrence, during the spring and summer seasons. Our results agree with seasonality of clinical variables from studies in Japan,3
and the United States (Ohio20
). Relapsing and progressive MS exhibited different seasonal patterns, with peak prevalence shifting toward spring for progressive MS. We also found associations of activity prevalence with climate data, particularly temperature and solar radiation levels, although this does not suggest a causal relationship. In contrast, as with other studies,10,11
the CEL count was unable to show a seasonal effect. The number of attacks recorded likewise did not have sufficient power to show a seasonal bias. All variables were controlled for individual levels of disease activity as well as variable number and frequency of observations.
We tried to distinguish between the likelihood of new activity and the intensity of activity. Both were clearly elevated during the warmer months in all analyses. Lesion fraction measured the new lesions observed in a season relative to each subject's total amount, emphasizing disease intensity, whereas the active scan metric disregarded the number of new lesions per examination and thus emphasized more the likelihood of the presence of an inflammatory process. When further excluding subjects with no activity in either season altogether, the lesion count remained significantly elevated in the warmer months. Together these findings suggest that both likelihood and intensity of active MS are modulated by seasonal effects.
There are limitations to this retrospective study arising from cohort size and unrecorded variables: the 44 subjects studied here are insufficient to extrapolate prevalence and range of susceptibility for a seasonal effect onto the entire MS population, as well as a putative link to vitamin D. However, the exquisite longitudinal follow-up of this cohort makes these results particularly valuable, because they demonstrate that a robust effect can be captured with sufficient longitudinal coverage and a sensitive metric such as new T2 lesions. Ethnicity was not recorded in all subjects and hence could not be controlled in analysis. Given that the patient population at our center is over 90% Caucasian,21
any confounding effect of ethnicity on the results is likely small. This is also validated in part by the results of the random split shown in , indicating that the observed seasonality did not originate from a particular subgroup of the cohort. Age and steroid treatment, which can influence MRI metrics such as brain volume,22
were not controlled directly in analysis due to the limited statistical power, but separate sensitivity analyses were performed to evaluate the effect of each: the seasonal modulation prevailed when splitting the cohort based on median age, making age an unlikely confounder. Also no change in seasonal prevalence was observed when excluding all steroid-treated subjects from the above analysis.
Because follow-up frequency was highest at study entrance, an irregular distribution of subjects entering the study could be a concern. The distribution of high-frequency follow-up examinations across the year is important in assessing potential bias from too many examinations concentrated in a particular period. lists the number of subjects entering the study over the year and the percentages of high-frequency examinations in each month. On average 66% (range 53%–78%) of all scans performed in a month were high-frequency examinations. The subject and examination frequencies are also shown in .
The observed seasonality with 2-fold to 3-fold shifts in activity may raise concerns for design and analysis of clinical trials with MRI outcome measures. If left unaccounted, this effect could bias longitudinal assessment both at individual as well as group level. Factors like genetic affinity, disease phenotype, and geographic location are likely to contribute to the magnitude of seasonality effects. The latter will have particular implications for multicenter trials that pool data from geographically distant locations. Studies with prescreening MRI will experience selection bias. Crossover trials also could be biased, depending on the timing of the trial arms, if both disease likelihood and intensity are nonuniform across the calendar year. Similarly, inaccurate power calculations may result from the assumption of a uniform yield of activity.
New T2 lesions served as the main MRI surrogate for active disease in this study. New T2 lesions commonly leave a long-lasting residual area of hyperintensity,14,15
whereas CEL enhance only briefly for about 1–3 weeks.23
New T2 lesions therefore provide an integrated view of disease activity that is much less sensitive toward the interval between MRI assessments and consequently a more robust outcome measure for longitudinal trials. Only one study thus far reports testing for potential seasonal bias of trials with prescreening MRI and CEL, finding seasonal variation but of insignificant impact on the main trial variables.11
The effect on the final power to detect a treatment effect, however, remains inaccessible. We also consider it critical to distinguish in this context between effects that influence the likelihood for new activity and effects that modulate the intensity of activity when it occurs. Seasonal effects, as far as the present results suggest, modulate both, hence trials with active disease enrollment criteria and pooled data are likely to be affected.12
Thus far only clinical and immunologic variables showed seasonal effects in MS. Neither, however, unlike MRI, strictly represents the CNS, making observations on MRI an important complementary finding. For example, the seasonal variations in immunoregulatory cytokine production, reported for interferon-γ, tumor necrosis factor–α, and interleukin-10,6-8
represent the response of peripheral blood mononuclear cells stimulated in vitro and do not reflect immune regulation in the CNS.