TB trends in regional and national scales
This study was to assess the characteristics of TB trends for the period 2004-2008 by month, year, gender, age, temperature, seasonality, aborigines/nonaborigines and to provide a best-fit generalized regression model to examine the potential predictors for TB incidences in Taiwan with regional and national scales. Since TB is caused by infection of the pathogen M. tuberculosis, the activity of pathogen can be influenced by weather and environmental conditions. The host behavior and healthy status can also cause the susceptibility of TB infection. Thus we examined different factors which may affect TB events.
We found that Hwalien (124 per 100,000 population) and Taitung (104 per 100,000 population) Counties situated at eastern Taiwan had the highest TB incidences among Taiwan regions in the period 2004-2008. Generally, more than 90% of TB cases found in adults aged 25-64 and older, with the largest fraction among > 64-year olds. The eastern Taiwan had the largest proportion of TB relapse cases (8.17%) in which Hwalien County gave 8.68% relapse proportion in the period 2004-2008. Overall, the average TB incidence, mortality, and relapse proportion were estimated to be 68 per 100,000 population, 0.036 person-1 yr-1, and 4.38%, respectively, in Taiwan for the period 2004-2008. Our result indicates that there is a trend of an increasing proportion of TB among the elderly. The plausible reason may be that with the falling of birth rate and increasing of longevity, the aging of populations is rising in Taiwan.
Fares et al., [
19] indicated that lower temperature during winter seasons may induce the susceptibility to respiratory epithelium infection. Furthermore, temperature is also an important climatic factor that influences TB seasonal trends. Specifically, the mean temperature with 1-month lag had highly significant effect on TB trends in Taiwan, whereas mean temperature with a 3-month lag showed highly significant in Hwalien County at eastern Taiwan. Therefore, in accordance with the idea of the impact of weather variations on TB infections, temperature emerged as the primary environmental correlate of TB incidence patterns at regional and national scales tested here. Based on this finding, changes in temperature may have strong consequences for the patterns of TB incidence.
There were various reasons that cause temperature lagged effects on TB trends. Naranbat et al., [
20] hypothesized that temperature may change the indoor/outdoor (I/O) ratio for TB susceptible and infected population, and further influence the transmission probability for
M. tuberculosis. Therefore, the longer incubation period may subsequently delay the TB detection and notification. The interval between observable immunological response and infection may delay over 7 weeks [
20]. The I/O ratio can also affect Vitamin D intake by sunshine levels which can decrease the risk for TB illness [
14,
20]. Our results found that there were different lagged effects with temperature on TB incidence such as 1 month-lag in national scale in Taiwan and 3 month-lag in regional scale in Hwalien County. We speculated that climatic properties were the important factor influenced by geographic location, for example, the relative lower temperature and higher rainfall in eastern Taiwan than other regions. The climate factor may further influence the sunshine proportion to residents on Vitamin D intake.
Our findings revealed that there had weak relationship between (lagged) temperature and TB incidence. We thus further considered the seasonality impact on TB incidence separately with temperature effect. The seasonality may exist in many infectious diseases due to variety of human behaviors, environmental conditions, and other uncertain factors. Greenman et al., [
21] indicated that external forcing can cause some cycles, oscillations, and even chaotic phenomenon in disease dynamics of populations. Therefore, we considered the seasonality as a major factor which can be seemed as predictor to better understand Taiwan TB trend. The five cycles of seasonality has been applied to investigate periodic trend of monthly new TB cases and possible causes of seasonal trend in previous China TB study [
18]. In this study, we also found that the Taiwan TB seasonality had similar pattern with TB epidemic in China. Thus we used five-cycle pattern as seasonality factor to improve the predictability of TB incidence rate. Although the seasonality showed insignificant in regional scales, the properties were detected significantly in national scale.
In regional scale of Hwalien County, age was the only predictor of TB trends identified as statistically significant among considered parameters in the Poisson regression model. Our analysis of TB trends at the regional scale reveals that the appearance of irregular annual TB incidences invariably coincides with the seasonality. None of the considered parameters generally explained variation in TB incidence in Taitung County. Yet, seasonality, aboriginal subpopulation, male, and age groups all showed significant relationships with TB trends in Taiwan. It is interesting to note that gender, time trend, and 2-3-month lag maximum temperature showed stronger association with TB trends in aboriginal subpopulations.
Our study showed that seasonal peaks of TB incidences generally occurred at late spring to early summer seasons in Taiwan. Thus dynamic consequences of seasonal variation in TB incidence appeared in Taiwan. This result was consistent with the study in China [
18]. The relatively high significant association of seasonality, gender, age group with national scale TB trends leads to more regular TB time-series dynamics in Taiwan than those in regional Hawlien and Taitung Counties. This emphasizes the potential disadvantages of extrapolating TB trends for these sorts of highly non-linear systems without a detailed understanding of regional parameters. Therefore, better data related to regional scale are required to account for these outcomes. Furthermore, the quality of the regional data allows us a rare opportunity to generate data-driven statistical models to assess TB trends in Taiwan.
TB infection appears to persist throughout several years even in relatively small populations, possibly through recurring infections in adults. If individuals experiencing their first infection are the primary drivers of endemics, then demographic changes will have a strong influence on endemic time-series. Therefore, differences in population demographics and epidemiology of TB diseases, and, potentially, vaccine effectiveness, would need to be carefully considered when estimating the TB trends in local regions. Hsueh et al., [
3] found that TB incidence in male patients was 2.2-fold higher than that in female patients. In this study, we found the similar results for male as higher prevalence group that can influence the TB trends in Taiwan (
p < 0.001). Thus we thought that gender differences could be a factor accounting for the TB incidence.
We used AIC and RMSE to assess the model fitness in this study. The RMSE is a measure of the differences between the original datasets and predictive values by optimal model. The RMSE can also measure the average magnitude of the error by quadratic scoring function. The differences were being aggregated into a single measure to test the model predictive power. The AIC is a measure of the relative goodness of fit for statistical model. It is often used to describe the trade-off between bias and variance in model construction, or loosely speaking between accuracy and complexity of the model. AIC is an assessment for model selection not only reflects goodness of fit, but also includes a criterion that is an increasing function of the number of estimated parameters. Based on AIC method, our result found that Hwalien County had more optimal fitted models than Taitung County.
Our Poisson regression analysis indicated that the performance of RMSE did not obviously improve the predictability of TB trends in Taitung County by adding other predictors as age and gender. This revealed that these factors had only weak effects on the TB incidence in regional scale. Therefore, the age and gender show significant influence on TB trends in Hwalien County than in Taitung County.
Limitations and implications
There are some limitations in our analyses. First, our database is limited to regional scale for which sufficient records were accessible to determine more significant predictors. A large gap is the age- and gender-specific TB cases in local scale, where data remain scarce. Likewise, we did not consider subpopulation effects, and had limited TB burden and trends data before 2004.
Second, our results are based on relatively coarse TB data provided by limited sources. The assessment of regional trends in incidence requires judgments about the reliability of case notifications reported by individual regional counties [
22]. With these limitations, the results and forecasts present herein should be interpreted with caution.
Last, we analyzed only a subset of mechanisms that may shape TB trends regionally and nationally. We did not consider the factors such as numbers of alcoholics, diabetics, HIV, nature resistance-associated macrophage protein 1 (NRAMP 1), smokers, and military and inmate subpopulations, that have proven critical to TB trends variability [
3,
15,
23‐
28]. Thus, those factors may be incorporated into the TB trends models to improve the predictability but they could have important dynamic consequences, which are worth exploring in future research. Generalized regression models rooted in regional data are important for providing clear recommendations for control strategies.
Although it was parameterized on the basis of observations from 2004-2007, the proposed Poisson regression model predicts the qualitative patterns of TB incidence rate at the regional and national scale in 2008. Assessing the impact of potential predictors on TB trends in Taiwan cannot neglect the self-intrinsic errors of missing variables for the selected Poisson regression model. This study suggests that more detailed surveillance data could well explain the peak values rather than missing information. Therefore, seeking novel mechanisms and providing both biological plausibility and epidemiological evidence for existing theories are still needed in future studies [
22,
29]. Assessment of TB trends in eastern Taiwan presents an important opportunity to understand the time-series dynamics and control of TB infections, given that this is the typical host demography in regions where these infections remain major public health problems.