2.1 Study Population

The study population consisted of individuals with PTB of all age groups and gender diagnosed between 2003 and 2007 who had resided in Mexicali for at least one year. The diagnosis of PTB (ICD-10: A15) required the presence of the following parameters: Ziehl-Neelsen stain positive sputum smear, clinical and chest x-ray manifestations characteristic for TB, and a suspected epidemiological history.

2.2 Clinical Data

Time-series data from 2003-2007 on PTB were made available by the Epidemiology Department (ED) of the Mexican Institute of Social Security, which records all new PTB cases. Individuals with PTB and HIV coinfection or Diabetes mellitus were excluded from the study. As per WHO guidelines (^{WHO, 1998}), the ED obtains and evaluates three sputum smears and defines a new PTB case as a person with at least two out of three Ziehl-Neelsen stain positive sputum smears. In some instances, sputum smear-negative cases require a sputum culture to confirm PTB. However, those cases were not included in the current study, leaving the possibility of missing false-negative PTB cases; and in very rare cases including false positive cases with Nocardia sp. infection (^{Muricy et al., 2014}).

Data in the current study is organized into weekly overall counts and counts by gender. We describe the total distribution by age groups (i.e. 0-14, 15-24, 25-34, 35-44, 45-54, 55-64, > 65 years) and age groups according to gender.

2.3 Meteorological Data

We worked with meteorological data [i.e. temperature (Temp) and relative humidity (Rh)] obtained from the Department of Meteorology at the Engineering Institute of the Autonomous University of Baja California, as published previously (^{Reyna et al., 2012}). The data are organized as time-series (2003-2007) of daily averages and weekly averages.

2.4 PM ₁₀ and PM _2.5 data

PM₁₀ and PM_2.5 data were provided by the Environmental Protection Secretary (SPA) of the State Government of Baja California, the National Institute of Ecology and Climate Change (INECC), and the United States California Air Resources Board (CARB). Time-series of daily averages and weekly averages were calculated as reported previously (^{Reyna et al., 2012}).

2.5 Biostatistics

The associations between PM₁₀ or PM_2.5 and PTB were calculated using Poisson regression models and are expressed as relative risk (RR) and estimated 95% confidence intervals (CI), adjusted for over-dispersion. The effects of PM on PTB were analyzed by gender and for the entire study population.

The baseline models were first estimated and then applied to assess PM₁₀ and PM_2.5 exposures. Baseline models were constructed by separately testing each of the dependent variables (e.g. PTB time-series according to gender) and adjusting for different independent control variables (see below). The models were specified with the Poisson regression algorithm in the NCSS9® software (^{NCSS, 2013}).

We used the following control variables for the estimation of our baseline models: unlagged (i.e. Lag0) and lagged versions up to 12 weeks (i.e. Lag1 - Lag12) of Temp, Rh, the square of Temp and Rh and sinusoidal structural terms for seasonality (^{Katsouyanni et al., 1996}), as well as the long-term linear and squared trends; time indicator variables were created for each year, month and week of the PTB diagnosis, taking as reference the first week, first month and first year; for age, the proportion of 65< year olds was taken as reference; for gender, the proportion of females was taken as reference. The control variables were included one-by-one when adjusting the baseline models. We started with the meteorological variables, then the structural and time variables, and, finally, age and gender. Only those variables contributing to a statistically significant (p<0.05) increase in the deviance were accepted for the baseline model specification.

We assumed having adequate models when the residual autocorrelation coefficients did not exceed the confidence limits (α = 0.05).

Once we developed the models, the baseline models were tested separately for PM₁₀ and PM_2.5 at Lag0 and up to 12 weeks (Lag1 - Lag12).

The PM levels were assessed as continuous or as categorical exposure (i.e. low and high pollution). To evaluate categorical exposure, we created two indicator variables. The “Low Pollution” category took the value of 1 when PM levels were equal to or less than the federal 24-hr average limit of 75 μg/m³ for PM₁₀ or 45 μg/m³ for PM_2.5 as regulated in the Official Mexican Standards (^{SSA, 2014}). Otherwise, it was assigned the value 0. The “High Pollution” variable took the value of 1 when PM levels were above the federal limits for PM₁₀ or PM_2.5 and otherwise, took the value of 0.

To obtain the effect of the PM pollutant on PTB in the weeks of high pollution, the indicator variable “Low Pollution” and its interaction with the respective pollutant (i.e., PM₁₀ or PM_2.5) were included in the model. The coefficient associated with the variable of the PM pollutant without interaction is taken as the effect on PTB in the weeks of high pollution. Thus, the PM variable’s coefficient would show the main effect of the category for a high level of PM pollution.

As with the baseline model estimation, it was assumed that the final models constructed were adequate when the residual autocorrelation did not show statically significant coefficients (α = 0.05). Any significant residual autocorrelation coefficient was controlled using the corresponding autoregressive PTB terms.

3.1 Population and PTB

During the five year study period (2003-2007), 1,748 new PTB cases were recorded in Mexicali, which had 896,394 inhabitants in 2007 (^{INEGI, 2020}). Of the 1,748 PTB cases studied, 1,253 were male and 495 female, with 61 PTB cases under the age of 15 years; 32 male and 29 female. There were noticeable differences in the PTB incidence rate over the study period, with the highest PTB case rate recorded in 2006 and the lowest in 2004. The incidence of PTB was higher in the period from winter to spring in 2004 and 2006, and the period from spring to summer in 2003, 2005, and 2007. Table I shows the PTB incidence rates estimated for 2003 - 2007, and by season.

In general, the highest number of PTB cases occurred in men (2.5 males per female). Most of the cases occurred in the 25-34 year age group. The 35-44 year age group had the highest rate of men (4.0 males per female). Table II shows the PTB incidence according to age group and gender.

3.2 Meteorology and Pollution

In general, the dry-hottest seasons occurred during spring (March 21-Jun 20) and summer (June 21-September 22), and the wet-coolest seasons occurred between fall (September 23-December 21) and winter (December 22-March 20). Table III shows the Temp, Rh and PM in Mexicali according to year and season for the entire period studied.

The highest PM₁₀ and PM_2.5 levels occurred during the wet-coolest seasons, with concentrations exceeding the federal annual average limits of 40 μg/m³ and 12 μg/m³, respectively (^{SSA, 2014}), for every year studied. The federal 24-hr average limit of 75 μg/m³ for PM₁₀ was exceeded on 219 days in winters, 181 days in springs, 201 days in summers, and 268 days in falls, resulting in a total of 869 exceedances during the study period. The federal 24-hr average limit of 45 μg/m³ for PM_2.5 was exceeded on 302 days in winters, 248 days in springs, 262 days in summers, and 351 days in falls, resulting in a total of 1,163 exceedances during the study period.

Figures 1 and 2 present the distribution of PTB cases and PM levels between 2003 and 2007. Figure 1 presents the number of PTB cases and mean PM concentrations according to the months for the duration of the entire study. Figure 2 presents the distribution of PTB cases for each month, according to PM₁₀ (Figure 2A) and PM_2.5 (Figure 2B) concentrations.

Fig. 1 Time series of the monthly mean PM₁₀ (blue line) and PM_2.5 (orange line) concentrations and the monthly number of PTB cases (gray bars) in Mexicali between 2003 and 2007. Interestingly, peak particle concentration preceded a peak in the number of cases suggesting seasonality. The models used in the study identified a time related statistically significant association in which existing elevated PM concentrations anteceded a ~3% increased risk for the diagnosis of PTB. 

Fig. 2 Distribution of all PTB cases (colored dots) diagnosed by month (1 = January to 12 = December) according to the PM concentration at the diagnosis time. This graphical representation shows the increased PM concentration (A = PM₁₀, B = PM_2.5) values during the cold months and the increase in the number of cases during the warmer months. The short horizontal red line indicates the position of the median for each dispersion cloud and the numbers in parentheses indicate the total monthly count of PTB cases. 

3.3 Pollution Exposure

The models showed statistically significant results (p<0.05) in some of the Poisson regression coefficients associated with the PM studied for the full population and for both genders. Table IV shows every most significant lag time regression coefficient associated with the PM and their statistics by continuous exposure and by high pollution exposure category (i.e. pollution levels above and below the 24-hr average limit for PM₁₀ (75 μg/m³) and PM_2.5 (45 μg/m³)). The Poisson regression coefficients were used to calculate changes in the RRs of PTB if the concentration of the PM₁₀ or PM_2.5 were increased by 10 µg/m³ of air, also shown in Table IV. Figure 3 graphically depicts the RR percentage increases of PTB by groups, for the 10 µg/m³ increment in PM₁₀ or PM_2.5.

Fig. 3 RR increases of PTB by group due to an increment of 10 µg/m³ in PM. 95% confidence interval. * p<0.10. ** p<0.05. 

The RR for the development of PTB in males and the totality of cases remained similar and was statistically significant (p<0.05), and borderline significant (p<0.10) for the continuous PM₁₀ exposure. The RRs associated with the PM₁₀ or PM_2.5 were slightly higher for the male PTB cases. Female PTB cases had an increased RR with broad confidence intervals compared with the RR for all the cases. Observed increased RRs were statistically significant only when considering continuous PM₁₀.

After model adjustments, the analysis showed a 10-week delayed effect for PM₁₀ and PM_2.5 in all PTB and the male PTB cases (p<0.05), and borderline significance (p<0.10) for continuous PM₁₀ exposure. An 11-week delayed effect occurred in the female PTB case population when considering continuous PM₁₀ exposure (p < 0.05). In the full PTB case population, the RR rose by 2.4% (95% CI: 2.1, 2.6) for each 10 µg/m³ increase of PM₁₀ in the continuous exposure and by 3.6% (CI: 3.3, 4.0) in the high pollution exposure category, and by 3.2% (CI: 2.9, 3.4) for each 10 µg/m³ increase of PM_2.5 in the continuous exposure and by 3.9% (CI: 3.6, 4.3) in the high pollution exposure category. In male PTB cases, the RR rose by 2.8% (CI: 2.5, 3.1) for each 10 µg/m³ increase of PM₁₀ in the continuous exposure and by 4.6% (CI: 4.2, 5.0) in the high pollution exposure category, and by 3.4% (CI: 3.1, 3.7) for each 10 µg/m³ increase of PM_2.5 in the continuous exposure and by 4.2% (CI: 3.8, 4.6) in the high pollution exposure category. In female PTB cases, the RR for development of PTB rose by 5.1% (CI: 4.7, 5.5) for each 10 µg/m³ increase of PM₁₀ in the continuous exposure and by 5.3% (CI: 4.7, 5.8) in the high pollution exposure category, and by 4.3% (CI: 3.8, 4.8) for each 10 µg/m³ increase of PM_2.5 in the continuous exposure category and by 5.3% (CI: 4.8, 5.9) in the high pollution exposure category.

As an example, Table V shows the Poisson regression coefficients with their respective confidence intervals and p-values obtained by including the continuous exposure of PM_2.5 delays in the baseline model for the full population. The most significant effect was in Lag 10.