Among the 537 participants at the first (baseline) study visit, the average age was 71 years, BMI was 25 kg/m2, and creatinine-adjusted urine cotinine as a proxy of exposure to either passive or active smoking was 255 mg/g (). The percentage of female participants was 74%. The proportion of individuals who reported regular alcohol consumption at least once a month for ≥ 10 years was 23%, and 63% of participants reported moderate exercise at least once a week. Hypertension was the most common preexisting disease (51%) followed by diabetes (18%) ().
Description of the study population at the first study visit (n = 537).
The average SGDS-K score out of a possible 15 was 3.6 at the first visit, 2.4 at the second visit, and 1.1 at the third visit (). The proportion of participants who responded negatively to each SGDS-K item was highest at the first visit. In the Supplemental Material, Table S1 (http://dx.doi.org/10.1289/ehp.1104100
), we show correlations among SGDS-K items, factor loading values from a factor analysis, and reliability coefficients for the test-retest analysis. For the factor loading values, the first factor consisted of six items (1, 5, 6, 7, 8, and 11), which may be interpreted as a dimension of emotional symptoms. The second factor was related to somatic symptoms and covered three items (9, 10, and 13). The third factor included six items (2, 3, 4, 12, 14, and 15) related to affective symptoms. The results of test-retest analysis at a 1-week interval showed a high correlation (Pearson r
= 0.92) of SGDS-K scores (see Supplemental Material, Table S1).
Depressive symptom scoresa [mean ± SD (median, range)].
shows characteristics of environmental factors in Seongbuk-Gu during the study period (August through December 2008, April through December 2009, and March through August 2010). The mean ± SD temperature was 17.3°C ± 8.1, PM10
concentration was 43.7 µg/m3
[interquartile range (IQR) = 24.2], SO2
concentration was 4.0 ppb (IQR = 2.3), NO2
concentration was 36.2 ppb (IQR = 5.0), maximum CO was 0.79 ppm (IQR = 4.0), and O3
was 48.1 ppb (IQR = 37.0) (). Pollutants, except O3
, showed a high correlation with other pollutants in repeatedly measured data and follow-up data [see Supplemental Material, Table S2 (http://dx.doi.org/10.1289/ehp.1104100
)]. Although PM10
concentrations did not show distinct patterns depending on hours of exposure, NO2
and CO concentrations were slightly higher during the morning rush hour (0700–0900 hours) and midnight through early morning (2300–0100 hours) than in other hours of a 24-hr day [see Supplemental Material, Figure S1a–e]. During the daytime (1400–1700 hours), high O3
concentrations were measured and were positively correlated with other pollutants [see Supplemental Material, Figure S1(f)].
Overall characteristics of environmental factors in Seongbuk-Gu, Korea (August–December 2008, April–December 2009, and March–August 2010).
displays the estimated percent change in SGDS-K scores per IQR increase in air pollutants at different lag days. The best fitting lag structures for PM10
, and O3
for the overall depression score were lag 0–2 days, lag 0–7 days, and lag 0–2 days, respectively. IQR increases in PM10
, and O3
for the best selected lag days were significantly associated with SGDS-K scores, with estimated increases of 17.0% [95% confidence interval (CI): 4.9%, 30.5%], 32.8% (95% CI: 12.6%, 56.6%), and 43.7% (95% CI: 11.5%, 85.2%), respectively. SO2
and CO were not significantly associated with SGDS-K scores at any moving average lag days within 4 weeks. IQR increases in SO2
apparently had a marginally protective effect at lag 0–21 days [–20.0% (95% CI: –36.6%, 0.9%)] and null effects at other lag days. Increases in CO showed positive associations at lag days 0–5, 0–7, and 0–14, and null associations at other lag days [see Supplemental Material, Figure S2 (http://dx.doi.org/10.1289/ehp.1104100
)]. When we performed cross-sectional analysis using data from the first visit only, estimated effects of PM10
on depressive symptoms were slightly attenuated but similar to associations based on the repeated measures analysis, although the magnitude of estimated effects of O3
were not statistically significant for any lag days and were considerably lower (see Supplemental Material, Figure S3).
Figure 1 Percent change in SGDS-K scores per interquartile range (IQR) of air pollutants by increasing lag day. Solid circles indicate the best-selected lag days based on minimal Akaike Information Criterion. IQRs for PM10, NO2, and O3 were 24 µg/m3, 15 (more ...)
We examined whether the association between depression and air pollution was modified by a history of CVD. suggests that participants without a history of hypertension may have been more susceptible to air pollution exposures than were those with a history of CVD. However, associations with O3 were stronger among participants with a history of hyperlipidemia than among those without this history. We observed little difference between participants with and without CVD () or MI ().
Figure 2 Percent change in SGDS-K scores by increasing lag days per IQR of air pollutants with data stratified by responses to a history of CVD [yes (solid circle)] vs. no (triangle)] (A), hypertension (B), MI (C), and hyperlipidemia (D). 0 d, concurrent exposure (more ...)
depicts the association of each individual item in the SGDS-K with IQR increases in PM10, NO2, and O3. All items in the first factor (emotional symptoms) showed significant associations with some or all of the three pollutants (PM10, NO2, and O3). ORs associated with responding negatively to items related to emotional symptoms (1, 5, 6, 7, 8, and 11) increased significantly with an IQR increase of NO2. Some of these items also showed a significant relationship with O3 and PM10. However, in the second and third factors that were related to somatic and affective symptoms, only “life is empty” was associated with three air pollutants, whereas “problems with memory,” “full of energy,” and “others are better off” were associated with one pollutant, and other items were not associated with any pollutants. Similarly, among the factor-specific composite scores (emotional, somatic, and affective symptoms), which were calculated as the loading value–weighted sum of the item responses within each factor structure, we observed that emotional symptom composite scores increased with increasing concentrations of PM10 [38.2% (95% CI: –3.6%, 98.1%)], NO2 [118.2% (95% CI: 37.9%, 245.3%)], and O3 [132.5% (95% CI: 32.0%, 309.3%)]; however, somatic and affective symptoms were associated with an increase of PM10, and nonsignificantly but positively associated with increasing NO2 and O3.
Estimated effectsa of responding negatively to each SGDS-K itemb and composite scoresc per IQR of air pollutants (95% CI).
In our sensitivity analysis, we compared single- and two-pollutant models using moving average lag 2 days for pollutants. Associations of PM10
with depression symptoms were attenuated when the model was adjusted for NO2
[17% (95% CI: 5%, 30%) vs. 9% (95% CI: –8%, 30%)] and for PM10
[20% (95% CI: 5%, 38%) vs. 11% (95% CI: –10%, 37%)], respectively; however, controlling for other pollutants did not substantially change the risk estimates [see Supplemental Material, Figure S4 (http://dx.doi.org/10.1289/ehp.1104100
)]. In an additional sensitivity analysis, we compared estimates in a full model with unweighted or unadjusted estimates. We found that weighted or unweighted estimates were similar, whereas crude effects of PM10
were substantially different from adjusted estimates, although no such difference was evident for O3
(see Supplemental Material, Figure S5). The significant confounders in the association with PM10
were age, sex, number of years of school, day of week, and follow-up time.