Low-intensity blue-enriched white light (750 lux) and standard bright light (10 000 lux) are equally effective in treating SAD. A randomized controlled study

In the winter of 2008-2009 patients of the SAD outpatient clinic of the University Medical Center Groningen, the Netherlands were asked to participate in the study. Eight patients were recruited by means of an advertisement in a local newspaper. Potential participants were sent written information and were invited for an intake interview at the clinic in order to obtain a diagnosis by an experienced clinical psychologist. The participants were informed about the goal of the study: to investigate the effects of low-intensity blue-enriched light treatment compared to standard light treatment. If patients were suffering from winter depression, they were given information about the research project. After they had signed the informed consent form, a screening visit was scheduled. Twenty-three patients were included in the study. After a few days of light treatment, one patient dropped out for reasons unrelated to her depression (concussion) and was excluded from the study, leaving 22 patients. In the SLT condition 3 men and 8 women participated (mean age 39.9 yrs ± 12.7), in the BLT condition this amounted to 2 men and 9 women (mean age 41.7 yrs ± 13.1).

The research protocol was approved by the Medical Ethical Committee of the University Medical Center Groningen.

Light treatment consisted of 2 weeks of SLT (correlated colour temperature 5000°K, vertical illuminance at eye position: 10 000 lux) with the EnergyLight (Philips Consumer Lifestyle B.V., Drachten, The Netherlands) or 2 weeks of BLT (correlated colour temperature 17 000°K) with a vertical illuminance of 750 lux at eye position, administered with an EnergyLight equipped with a special prototype lamp emitting white light with a high proportion of short wavelengths with a correlated colour temperature of 17 000°K, as in the Philips ActiViva Active lamps. An international standard for retinal blue-light hazard risk has been defined to protect participants against retinal photochemical injury from chronic blue-light exposure [17]. All light conditions used in this study remain far below the exposure limits as defined by this standard.

During light therapy, patients were sitting at equal distances (20 cm) from the EnergyLight in both conditions. In Figure 1 the spectral-power distributions of the standard light and of the blue-enriched white light lamps are shown.

Subjects came to the university hospital on 10 workdays (days 4-8 and days 11-15 in the protocol) and were given either 30 minutes SLT or BLT between 7:45 and 8:45 a.m. Treatment was given to one patient at a time, without staff or other people being present. The SLT condition had an illuminance of 10 000 lux and a correlated colour temperature of 5000 °K. In the 380-740 nm wavelength band the SLT irradiance was 3207 microW/cm2, with a total photon flux of 8.84 × 10¹⁵ photons/cm²/s. The SLT irradiance within the 424-532 nm band equals 1135 microW/cm2, with a photon flux of 2.7 × 10¹⁵ photons/cm²/s. The BLT illuminance equals 750 Lux, with a correlated colour temperature of 17 000 °K. The BLT irradiance within the 380-740 nm range equals 307 microW/cm2, with a total photon flux of 7.81 × 10¹⁴ photons/cm²/s. The BLT irradiance within the 424-532 nm band equals 168 microW/cm2, with a photon flux of 3.9 × 10¹⁴ photons/cm²/s.

During the screening visit, patients were assessed by means of a standardized structured interview, the Mini-International Neuropsychiatric Interview (M.I.N.I) [18]. Subjects meeting the criteria of a major depressive disorder, seasonal pattern, winter type, according to the DSM-IV-TR [19] were subsequently assessed by means of the Structured Interview Guide for the Hamilton Depression Rating Scale-Seasonal Affective Disorder, 24-item version (SIGH-SAD) [20]. After that they were asked to complete the Beck Depression Inventory-II-NL (BDI-II-NL) [21], and a questionnaire aiming to evaluate subjects' expectations of the effects of light therapy. On a 5-point scale this questionnaire rated both for SLT and BLT whether subjects expected to benefit from the therapy, whether they thought it was a logical treatment, and whether they would recommend this therapy to a friend. Subjects who met all inclusion criteria were randomly assigned to one of the two conditions, with gender and age distributed evenly over the groups. They were not told which of the two conditions they were going to participate in.

Each of the two conditions started at day 1 (Friday) with a baseline measurement consisting of a SIGH-SAD interview, the BDI-II-NL and a fatigue self-rating questionnaire (Short Fatigue Questionnaire, SFQ) [22]. The SIGH-SAD interviewers were unaware of the experimental conditions. The SIGH-SAD, the BDI-II-NL and SFQ were repeated at day 8 (directly after the 5^th light session), at day 15 (directly after the 10^th light session) and at day 22 (1 week after light treatment had ended).

On day 22, an evaluation questionnaire was added to check the outcome of subjects' expectations on day 1. Furthermore, subjects were asked which condition they thought they had been treated in.

Starting at day 1, before 8.00 a.m. and at least 30 minutes after waking up subjects rated their mood and sleep quality of the previous night on a daily basis using the Adjective Mood Scale (AMS)[23, 24] and the Groninger Sleep Quality Scale (GSQS)[25, 26]. Also, the following four components of activation were measured, using the Activation-Deactivation Check List (AD-ACL)[27]: General Activation (GA; i.e. energetic, vigorous, full of pep, active, and lively), Deactivation-Sleep (DS; i.e. sleepy, tired, drowsy, wide awake, and wakeful), High Activation (HA; i.e. jittery, intense, fearful, clutched-up, and tense), and General Deactivation (GD; i.e. placid, calm, at rest, still and quiet). To describe their current feelings, subjects were asked to rate these 20 adjectives on a 4-point scale. The scores of the first 3 days on the daily questionnaires (before light treatment) were considered baseline.

Baseline differences between the two conditions were tested by means of t-tests (continuous outcomes) and chi-square tests (dichotomous outcomes).

Effect sizes [28] were calculated for each condition. These effect sizes reflect the differences between baseline (day 1) and day 22. Results were based on the weekly assessments of the two conditions and were compared by means of repeated measures ANOVA. This was done for the patients who had complete data for these measures (n= 11 vs.11).

Linear Mixed Models were used to compare the two conditions on the basis of the daily self-rating questionnaires.

An advantage of these models is that all available data can be used, including those of subjects with one or more missing values. Consequently, in these analyses data of all 22 subjects were used. Another advantage of linear mixed models is that they allow for including random effects; i.e. parameters are allowed to vary across individuals. This may reveal heterogeneity in individual growth curves. We used models with time, condition, and the interaction between time and condition, with the baseline score as a covariate (baseline score = mean of the 3 pre-intervention scores). We fitted models with the 22 days as the repeated measures and allowed the slope to vary across individuals. Maximum likelihood estimation was used. We compared models with different variance-covariance matrices. Selection of the final model was based on the Bayesian Information Criterion (BIC; with lower values indicating better models). If the random effect for slope was found to be non-significant, this term was removed from the model (unless this resulted in a higher value of the BIC criterion). Regression assumptions were checked by performing residual diagnostics on the final models.

In a secondary analysis, we examined the potential impact of the initial severity of the complaints on outcome. To this end, we added the interaction baseline*time to the models. We also examined whether this effect of baseline severity differed for the different conditions by adding the interaction condition*baseline*time to the models, including all lower-order terms.

A responder was defined as a subject who improved at least by 50%. Analyses were carried out using SPSS 17. A two-tailed alpha level of 0.05 was used to determine statistical significance.

The results of the weekly assessment procedures are summarized in Table 1. Although subjects in both conditions improved after exposure to light treatment, there were no statistical differences between these improvements.

In both conditions, depressive complaints decreased during the 3-week period (Table 1, SIGH-SAD 24 items, main effect "time" F(3,18) = 30.2, p < 0.001), with no significant differences between conditions (main effect "condition" F(1,20) = 0.012, ns) nor over time between conditions (interaction effect "time*condition" F(3,60) = 0.95, ns). The same pattern emerged when the SIGH-SAD was subdivided into "typical symptoms" (17-item Hamilton rating, Table 1, main effect "time" F(3,18) = 28.2, p < 0.001; main effect "condition" F(1,20) = 0.00, ns; interaction effect "time*condition" F(3,60) = 0.62, ns) and "atypical items" (7 atypical items, Table 1, main effect "time" F(3,18) = 18.84, p < 0.001; main effect "condition" F(3,20) = 0.039, ns; interaction effect "time*condition" F(3,60) = 0.99, ns). Calculations based on the BDI-II-NL scores showed similar results (main effect "time" F(3,18) = 31.4, p < 0.001; main effect "condition" F(1,20) = 0.78, ns; interaction effect "time*condition" F(3,60) = 1.57, ns). Calculations based on the SFQ scores point in the same direction (main effect "time" F(3,18) = 39.6, p < 0.001; main effect "condition" F(1,20) = 0.21, ns; interaction effect "time*condition" F(3,60) = 1.42, ns). Although the number of responders differs in the two conditions (measurements based on the weekly ratings), this difference was not statistically significant.

The evaluation questionnaire taken at day 22 shows that participants of the blue-enriched white-light condition experienced the treatment as less comfortable than participants of the standard bright-light treatment (F(1,20) = 9.61, p = 0.006). After treatment, 2 subjects in the SLT condition thought they had been treated in the other condition, another 2 were unsure. In the BLT condition 1 subject thought he had been treated in the other condition and another one was unsure.

As can be seen from Table 2, the results of the daily self-rating questionnaires are in line with the results of the weekly assessment procedures. There was no statistically significant difference in the way subjects improved. The interaction time*condition was not significant in any of the models. The time effect, on the other hand, was significant in all models. Thus, mood, sleep quality and energy levels improved in both conditions (Figure 2 and Table 2).

We also examined the influence of gender and age on the outcomes. No interaction effects were found between gender or age on the one hand and time or time*condition on the other. Adjustments for gender and age did not cause any substantial changes in the results either. Therefore, gender and age have not been included in the final models.

As can be seen from Table 2, in all models baseline severity was related to outcome, with higher baseline scores predicting higher overall follow-up scores. The interaction between baseline scores and time was significant in the models for Mood (p < 0.001), Sleep (p < 0.05), General Activation (p < 0.005), and General Deactivation (p < 0.05). Consequently, baseline severity was also related to change in symptoms over time. The regression coefficients for the interaction effects had a negative sign, which implies that higher baseline severity predicted steeper slopes (more improvement).

The interaction condition*baseline*time was non-significant in all models. This indicates that the effects of baseline severity on outcome did not differ for the different conditions.