Using the outlined model-building strategy, the best-fit candidate model included natural cubic splines with degrees of freedom (df) = 3 for month and df = 4 for number of days since 1 June 2002. After controlling for time trends, ACF between consecutive days was reasonably low (no-lag, 0.17; 1-day lag, 0.17; 2-day lag, 0.17). Linear functions of elemental carbon, carbon monoxide, nitrogen dioxide, and iron were included in the model because they improved model fit, quantified by AIC. Summary statistics for zinc and asthma ED visits and hospitalizations (health care visits) are found in . Overall, there were 3,786 pediatric asthma ED visits and hospitalizations in Maryland during the 183-day study period by children residing in the Greater Baltimore area. During the 183-day period, the median number of daily urgent health care visits was 16 [interquartile range (IQR), 11–30].
Summary statistics for zinc and ED visits/hospital admissions for children, Baltimore, Maryland, 1June 2002–30 November 2002.
Zinc was measured on 93 of the 183 days from 1 June 2002 to 30 November 2002, with a median (IQR) of 14.71 (7.53–25.30) ng/m3. The characteristics of children included in the models who visited EDs or were hospitalized are shown in . The number of visits included in the models were 1,813 (no-lag model), 1,819 (1-day lag model) and 1,784 (2-day lag model). The sex, race, and age distribution of the patients included in analyses did not vary by lag of zinc and did not vary over time (data not shown). Approximately 60% of visits included in analyses were made by males, 79% of visits were made by African Americans, and the most common age group presenting were school-age children (36% of visits). The random intercept accounted for 30% of the variation of health care utilizations after controlling for the variables included in the best-fit model. The random slopes for the effects of zinc accounted for < 1% of the variation of health care utilization.
Characteristics of children hospitalized or who visited the emergency department for asthma, Baltimore, Maryland, 1 June 2002–30 November 2002 [no. (%)].
shows that, adjusted for time trends, same-day medium concentrations of ambient air PM2.5 zinc are associated with an increased risk of urgent health care utilization of 1.12 [95% confidence interval (CI), 0.98–1.28] times that on days with low levels of zinc (p = 0.09). Risk for asthma ED visits and hospitalizations for previous-day medium levels of zinc was 23% higher [relative risk (RR) = 1.23; 95% CI, 1.07–1.41] than for previous days with low levels of ambient zinc (p = 0.005). Same-day high levels of zinc have risks for asthma ED visits and hospitalizations that are 9% higher (RR = 1.09; 95% CI, 0.91–1.30) than same-day low levels of zinc. Previous-day high levels of zinc are associated with a risk for urgent health care that is 1.16 (95% CI, 0.97–1.39) times higher than that with previous-day low levels of zinc (p = 0.10). Risks of asthma health care utilization are 1.15 (95% CI, 0.96–1.38) times higher when 2-day lag zinc is high compared with when 2-day lag zinc is low. Last, the risk of asthma health care utilization is also elevated when 2-day lag zinc is at medium levels, compared with low levels (RR = 1.11; 95% CI, 0.94–1.30).
RR (95% CI) and p-values for zinc, Baltimore, Maryland, 1 June 2002–30 November 2002.
The number of ED visits and admissions over time on days included in the analyses and their estimated trend using all three lag models are displayed in , showing lower numbers of visits, on average, in the summer than in the fall; the models fit the data well. Each line in represents the estimates from the different models. The levels of zinc and their estimated time trend fit with a LOESS smoother are shown in , suggesting peaks in August and November and troughs in June and October; however, this is partially attributable to large outliers in August and November.
Figure 2 Asthma health care visits/admissions by lag and level of zinc, Baltimore, Maryland, 1 June 2002–30 November 2002. The number of ED visits and admissions are shown using all three lag models, with each line representing the estimates from the different (more ...)
Zinc levels over time with LOESS smoother, Baltimore, Maryland, 1 June 2002–30 November 2002. The levels of zinc suggest peaks in August and November and troughs in June and October.
Air pollution is composed of many constituents. During the period of study, we found relatively strong correlations between zinc and other pollutants including nickel, chromium, iron, carbon monoxide, elemental carbon, and nitrogen dioxide, but relatively weak correlation between zinc and sulfate and ozone (), even after accounting for time trends. Also, a moderate correlation between zinc and temperature was found. After accounting for elemental carbon, carbon monoxide, iron, and nitrogen dioxide (the pollutants included in the best-fit Poisson regression model), the correlations between zinc and the remaining pollutants are attenuated. A sensitivity analysis was conducted that involved refitting each lag model excluding copollutants (including time trends only) and including nickel and chromium in addition to time trends, nitrogen dioxide, elemental carbon, carbon monoxide, and iron (). Except for same-day and 1-day lag models comparing high to low levels of zinc, the results show little sensitivity to choice of copollutants included in the models.
Pearson’s correlation coefficient between zinc and other candidate pollutants and weather variables, Baltimore, Maryland, 1 June 2002–30 November 2002.
RR (95% CI) and p-values for sensitivity analysis models of zinc and potential confounders, Baltimore, Maryland, 1 June 2002–30 November 2002.
Results assessing interactions of copollutants elemental carbon, carbon monoxide, nitrogen dioxide, and iron with zinc are shown in . The data showed little evidence that the RRs of asthma ED visits and hospitalizations comparing high and medium levels of same-day zinc to low levels of same-day zinc depend on values of copollutants (p for interaction > 0.20). However, the 1-day and 2-day lag models showed evidence of interaction of elemental carbon and nitrogen dioxide with zinc (p for interaction < 0.01). In both cases, the RR of asthma health care utilization comparing medium zinc to low zinc is higher on days with low levels of elemental carbon (≤ 0.99 μg/m3) and nitrogen dioxide (≤ 22 ppb). Last, the 2-day lag model showed evidence of an interaction of carbon monoxide with zinc (p for interaction = 0.048). The RR of asthma health care utilization comparing high zinc to low zinc levels is higher on days with high levels of carbon monoxide (> 0.4 ppm).
Figure 4 RRs (95% CIs) for high and medium levels of zinc (vs. low levels), separately for days with high (> median) and low (≤ median) levels of co-pollutants, Baltimore, Maryland, 1 June 2002–30 November 2002. Asthma ED visits and hospitalizations (more ...)