Does Seasonal Affective Disorder Exist? A Commentary on Traffanstedt,Mehta,and LoBello (2016)

Issues About the Diagnosis and Measurement of Depression

One major focus of the article is a critique of various diagnostic constructs and their measurement. However, the article presents a picture that in some ways is confused and inaccurate. The constructs of concern are (a) major depressive episode (MDE) as defined by criteria of the Diagnostic and Statistical Manual of Mental Disorders (DSM), (b) DSM diagnosis of major depressive disorder with seasonal pattern (MDD-SP), and (c) seasonal affective disorder (SAD). In addition, the authors raise issues about use of the Seasonal Pattern Assessment Questionnaire (SPAQ; Rosenthal, Bradt, & Wehr, 1984a).

The authors state, “The SAD construct and the typical method of measurement it (i.e., the SPAQ) have little in common with the DSM construct of major depression” (p. 5). The meaning of this statement depends on whether “major depression” refers to MDE or to MDD-SP. There has always been an inconsistency between MDD-SP and what researchers in the field mean by SAD. For MDD-SP, when annually recurrent episodes begin and end need to be consistent for a given individual but can occur at any time of the year. In contrast, SAD research is almost always concerned with only winter depression, that is, regular onsets in the fall/winter and offsets in the spring. Thus, inconsistent results in research that uses SAD and MDD-SP are to be expected rather than a point of criticism—a critique of MDD-SP does not necessarily constitute a critique of SAD. Although Traffanstedt et al. state their questions primarily as about SAD, their design focuses heavily on MDD-SP.

The authors correctly note that the SPAQ does not include all MDE criteria symptoms (also see previous similar criticisms by Steinhausen, Gundelfinger, Winkler, & Metzke, 2009, and Young, Hutman, Enggasser, & Meesters, 2015) and asks the respondent to make a retrospective judgment about the seasonality of the symptoms it does assess. They do not note, however, that many SAD studies, whether they use the SPAQ or not, also use standard methods for assessing MDE, such as the Structured Clinical Interview for DSM-IV (First, Spitzer, Gibbon, & Williams, 1996), that include all nine DSM symptoms. In addition, however, it is important to note that, in contrast to recurrent MDD in general, sleep and appetite/eating/weight in SAD are usually increased and that hypersomnia, hyperphasia, and fatigue often are considered the cardinal symptoms of SAD (Rosenthal et al., 1984b; Young, Watel, Lahmeyer, & Eastman, 1991). Although MDE criteria do include hypersomnia and hyperphasia, because they are combined into a single criterion with hyposomnia and hypophasia, cases identified using MDE criteria may include a large number of cases that would not usually be considered SAD—another contrast between MDD-SP and SAD.

The SAD Construct and What Varies Seasonally

Following up on their diagnosis critique, Traffanstedt et al. state that “between the publication of DSM-IV and that of DSM-5, evidence accumulated contradicting the claim that some recurrent episodes of major depression are linked to seasonal changes” (p. 2). Given that only one of these studies (Kerr et al., 2013) assessed participants in more than 1 year and only one (different) study used MDE diagnostic criteria (Posternak & Zimmerman, 2002), drawing conclusions about “recurrent episodes” is problematic (also see the discussion later in this article about detection of seasonality). In addition, careful attention to what symptoms were assessed in these five studies is very informative. In one sample, Kerr et al. (2013) used the Center for Epidemiological Studies Depression scale (Radloff, 1977), which has only sleep and appetite decreases, and in another sample, they used 13 items from the SCL-90R (Derogatis, 1994), which do not include sleep or appetite at all. Magnusson, Axelsson, Karlsson, and Oskarsson (2000) employed the Hospital Anxiety and Depression Scale (HADS; Zigmond & Snaith, 1983), which also does not include sleep or appetite changes. Nayyar and Cochrane (1996) also failed to observe seasonal changes with the HADS; however, they did observe seasonality with the SPAQ, which does include these symptoms. Finally Pasternak and Zimmerman (2002) did not find a winter worsening in overall symptom severity, but did find it for hypersomnia and hyperphagia. Thus, in these studies cited by Traffanstedt et al., seasonality was not observed when the “reverse vegetative” symptoms were excluded but was observed when they were specifically examined. Although as noted by Traffanstedt et al. (2016, p. 2), the scales used in the studies more “closely approximate DSM major depression diagnostic criteria,” by including MDE symptoms not included in the SPAQ, they also less closely approximate DSM criteria by excluding the reverse vegetative symptoms that are in the criteria. Of note, the Hospital Anxiety and Depression Scale (PHQ-8) used in the Traffanstedt et al. study did include increased sleep and/appetite. However, as in DSM criteria, these were combined with decreases so that seasonal changes specific to those symptoms would be obscured.

Not included in the review by Traffanstedt et al. is a study by Murray, Allen, and Trinder (2001) in which members of a community sample in Australia were assessed in summer and winter repeatedly over several years. The authors conclude that “in the population sampled, mood seasonality seems to be a small effect, the demonstration of which depends on subtleties of measurement” (p. 887). Specifically, seasonality was observed for only “pure positive affect” items of the behavioral engagement scale (sociability, sensation seeking, sensory awareness, and physical movements; Spoont, Depue, & Krauss, 1991) and not on items that loaded on both positive and negative affect. Although low energy was interpreted as a mixed-loading item, it had the second highest loading on positive affect and a loading of only .16 on negative affect and so easily could have been considered a pure positive affect item instead. Thus, seasonality was observed in the energy/motivation aspect of symptomatology. Seasonality was not observed in negative affect itself, which represents depressive phenomena such as sadness, distress, and guilt.

The issues of what symptoms constitute SAD and what varies seasonally is not new to the field. The specific focus of Traffanstedt et al. on the seasonality of MDD-SP, although a meaningful question in itself, is not the only SAD-relevant question and actually may obscure our understanding of SAD. As early as 1986 Mueller and Davies suggested that SAD is a seasonal energy syndrome. Several studies of specific SAD episodes, both retrospectively during the episode (Young et al., 1991) and prospectively (McCarthy, Tarrier, & Gregg, 2002; Whitcomb-Smith et al., 2013), have observed that episodes begin with vegetative changes (energy, sleep, appetite) and that cognitive and affective depressive symptoms develop later.

The dual vulnerability model of SAD (Young, 1999; Young et al., 1991; Young, Reardon, & Azam, 2008) proposes that the etiology of SAD consists of two mechanisms, a physiological vegetative response to seasonal environmental changes and a psychological response to those vegetative changes based on psychological depressive vulnerabilities. Among the support for this model is that there are individuals in the general population who report seasonal changes in sleep, appetite, weight, and energy only (White & Terman, 2004). Thus, one might expect the primary seasonal variation in the general population to be in these vegetative functions, as has been the case. The observed variability in the course of SAD across years (Sakamoto, Nakadiara, Kamo, Kamo, & Takahashi, 1995) might also be accounted for by variability in the engagement of psychological depressive vulnerabilities following vegetative changes, although this has not been examined empirically. It also is worth noting that if the etiology of SAD occurs in two stages as proposed in the dual vulnerability model, and if the probability of having vegetative symptoms in the winter following a SAD episode is .90 (very high) and the probability of having psychological symptoms if vegetative symptoms occur is .80 (still high), then the probability of having an episode that second winter (having both vegetative and psychological symptoms) is only .72.

Issues About Assessing Seasonality and About Predictors of Seasonal Depression

Season in the Traffanstedt et al. study was represented using a polynomial approximation of a continuous sinusoidal time variable across the entire year. What this method might capture depends on what is assumed about the seasonal distribution of symptomatology. This continuous variability definition of seasonality does not correspond precisely to the conceptualization of seasonality of the clinical syndrome of SAD, that is, fall/winter onset and spring offset. This approach is even more problematic if one applies it to MDD-SP, in which peak depressive symptoms are not synchronized at all across individuals, but can occur at any time of the year. The more that the timing of seasonality differs across individuals, the less seasonal pattern would be observed using this method in the population as a whole. In fact, if every participant had a seasonal pattern, but the patterns were distributed uniformly across the year, the result would be absolutely no seasonal pattern. Similarly, to the extent that the trajectories of different symptoms differ, the seasonal effect will be blunted if all the symptoms are combined into a total symptom severity score. For example, Whitcomb-Smith et al. (2013) found that the trajectory of mood symptoms lagged several weeks behind that of vegetative symptoms. The point here is not that all these situations are necessarily the case, but that the method used is not well suited to address the question posed. Arguing for the null hypothesis (there is no seasonality) based on nonsignificant findings is always a challenge, but particularly so in this case.

In contrast to seasonality, latitude and day length (“sunlight”) as predictors of depression were inexplicable reduced from truly continuous variables to three and four categories, respectively. The authors provide no rationale for categorizing these variables or for the category boundaries used. It is well-known that categorizing continuous variables decreases statistical power and confuses the meaning of the variables (e.g., MacCallum, Zhang, Preacher, & Rucker, 2002).

The authors note that latitude and season are hypothesized to act as proxy variables for photoperiod (“sunlight exposure duration”). However, they fail to mention research that specifically examined this issue using photoperiod itself. Across five latitudes (39°N to 60°N) and 20 fall/winter weeks, Young, Meaden, Fogg, Cherin, and Eastman (1997) found that the risk of SAD onset in any given location-week was highly associated with photoperiod. This pattern was observed equally for North American and European locations and as far north as Oslo. The only exception was the one location-week with the very shortest days (~7 hours). This finding suggests that photoperiod is not irrelevant in northern (or European) locations or that there is a general cultural or genetic selection effect reducing a photoperiod effect, but that the biological responses to short photoperiod may have already hit their maximum in the most extreme conditions wherever or whenever they occur (Young et al., 1997).

Also related to photoperiod, the Traffanstedt et al. abstract begins, “Seasonal affective disorder (SAD) is based on the theory that some depressions occur seasonally in response to reduced sunlight” [italics added]. Despite the fact that light was discussed as an etiology from the beginning of modern considerations of SAD (including several different hypotheses about light; Rosenthal et al., 1984b) and that light has been central to the scientific and lay discourse about SAD, a hypothesis about photoperiod being the etiology of SAD is not a hypothesis about the existence of SAD. The histories of medicine and psychology are rife with examples of valid disorders that had many mistaken proposed theories of etiology.

Issues About Study Designs to Detect SAD

The authors admit that it might be difficult to detect a seasonal pattern in the general population and so repeat their analyses (with similar results) in a subsample of people with higher depression scores. However, this approach only weakly addresses this issue. SAD is not a phenomenon that varies in intensity across the entire population or even in the population of people who suffer from depression. Therefore, it is neither conceptually nor methodologically an effective strategy to try to detect the presence of a subpopulation as a trend in the population as a whole.

The issue of the existence of SAD is a question about the pattern of an individual’s symptoms of depression across time. In contrast, the study by Traffanstedt and colleagues (2016) was cross-sectional, with each person being assessed once at some point during the year. Thus, the data contain no information about any of the symptomatology assessed being recurrent, seasonally or otherwise. Use of this design is understandable given the large sample size, a definite strength of the study. However, the implications of results of a cross-sectional between-subjects study for within-subject seasonal variability are limited (Murray, 2001). An ideal study would include a sample assessed frequently across the year (e.g., weekly). The goal of such a design is to separate variability due to (a) individual differences in mean depression levels, (b) patterns across time, and (c) individual differences in patterns across time. (Statistically, this would involve multilevel models [Hox, 2002].) The study partially accounted for individual differences with the control variables, which was appropriate. However, it is unlikely that all of the simple between-subject variability in depression was isolated, and individual differences in time patterns were not addressed.

Conclusions

As described earlier, the study by Traffanstedt and colleagues (2016) failed to address the question of whether SAD exists for both conceptual and methodological reasons. A thorough review of the evidence supporting and not supporting the existence of SAD is beyond the scope of this commentary. However, it is important to note that the literature cited by the authors was quite limited and selective, being focused almost exclusively on studies that do not support SAD (although see the more detailed consideration of those results presented earlier) and, to a lesser extent, studies with methodological limitations they wished to highlight. In particular, by focusing narrowly on the operational definition of MDD-SP, the authors failed to consider the existing theoretical and empirical literature that has attempted to understand the nature of SAD (if it exists) and explain the time course of SAD symptoms. Furthermore, the fact that something has received popular attention does not in itself make it merely a “folk” phenomenon (e.g., Beethoven’s symphonies). Overall, rese-arch that addresses the issues Traffanstedt et al. (2016) raise about the existence of SAD needs to attend to specifics about in what people, in what symptoms, and at what times and have a methodology that matches the theoretical formulations in the field. To a large extent, the Traffanstedt et al. study attempted to find a needle in a haystack with a faulty definition of the needle, a faulty definition of the haystack, and a faulty definition of “find.”