A
A
Prediction and Spatiotemporal Heterogeneity Pulmonary Tuberculosis in Iran using Geographically Weighted Machine LearningSaber Ghaffari fam 1, Leili Tapak 2,3, Erfan Ayubi 4, Mahshid Nasehi 5, Salman Khazaei 1,6*
1- Department of Epidemiology, School of Public Health, Hamadan University of Medical Sciences, Hamadan, Iran
Running title: Spatial Machine Learning of PTB in Iran
2- Modeling of Noncommunicable Diseases Research Center, Institute of Health Sciences and Technologies, Hamadan University of Medical Sciences, Hamadan, Iran
3- Department of Biostatistics, School of Public Health, Hamadan University of Medical Sciences, Hamadan, Iran
4- Social Determinants of Health Research Center, Hamadan University of Medical Sciences, Hamadan, Iran
5- Centre for Communicable Diseases Control, Ministry of Health and Medical Education, Tehran, Iran
6- Research Center for Health Sciences, Hamadan University of Medical Sciences, Hamadan, Iran
Corresponding Author:
Salman Khazaei,
School of Public Health,
Hamadan University of Medical Sciences
Address
Hamadan University of Medical Sciences Headquarters, Shahid Fahmideh St.
Phone
081-31310000
Zip code
6517838736
Email
salman.khazaei61@gmail.com
Abstract
Background Spatial analyses of pulmonary tuberculosis (PTB) have garnered significant attention due to the inherent spatial dependence and heterogeneity of this infectious disease. In the present study, we employed the Geographically Weighted Random Forest (GWRF) model to rigorously evaluate the effects of meteorological variables and the Human Development Index (HDI) on PTB incidence throughout the study period, in addition, we predicted the incidence of PTB over the next five years. Methods This study utilizes publicly available PTB incidence data from 31 provinces of Iran spanning 2009 to 2023. We employed the GWRF model to investigate the local associations between the standardized incidence ratio (SIR) of PTB and various influencing factors, including the HDI, temperature, relative humidity, dew temperature, and wet temperature, all obtained from multiple data sources.
Results
Temperature showed a stronger influence in the southern and northern regions, while HDI exhibited high importance in several southern and central provinces. In addition, humidity demonstrated localized effects, particularly in southern and eastern areas. prediction analyses indicated an increasing trend in PTB incidence in provinces such as Qom, Kerman, and Ilam over the next five years, whereas a declining trend is anticipated in provinces including Sistan and Baluchestan, Kermanshah, and North Khorasan.
Conclusions
These findings highlight the critical role of spatially varying metrological and socioeconomic factors in shaping PTB incidence and underscore the need for region-specific prevention and control strategies.
Keywords:
Pulmonary Tuberculosis
Spatial machine learning
Meteorological
Human Developmental Index
A
A
Background
It is estimated that approximately one-quarter of the global population has been infected with Mycobacterium tuberculosis (1). In 2023, tuberculosis (TB) re-emerged as the leading cause of death worldwide attributable to a single infectious agent, having previously ranked second to COVID-19 during the pandemic years of 2020–2022 (2). TB can infect various organs throughout the human body; however, it primarily affects the lungs, a manifestation referred to as PTB (3). TB continues to pose a significant public health concern in developing countries, including Iran. In Iran, the reported overall incidence rate of all forms of TB has declined from 9.6 per 100,000 population in 2019 to 7.6 per 100,000 population in 2024 (4).
Spatial epidemiology describes and analyzes the geographic distribution of diseases and is currently widely used in epidemiology (5). The spatiotemporal analysis of infectious diseases is generally conducted using Geographic information systems software, providing powerful spatial analysis capabilities and ideas and methods for the spatial analysis of infectious diseases (6). TB incidence is typical of spatiotemporal data, and many studies have confirmed that TB incidence is characterized by spatial aggregation (7, 8). These methods help identify disease clusters, dynamically visualize changes in disease occurrence and cluster trends over time and space, and elucidate geographic distribution patterns and risk levels in different regions (8).
Recent evidence indicates that climate change may contribute to increased susceptibility to TB by adversely affecting key social and environmental determinants, including food security, nutritional status, water availability, and access to healthcare services. These disruptions may, in turn, intensify both TB transmission dynamics and disease severity. Consistent with this perspective, epidemiological studies conducted in Zhejiang Province, China, have reported significant associations between meteorological conditions and TB incidence. Specifically, lower ambient temperatures and reduced relative humidity were found to be associated with elevated TB risk, with these relationships further influenced by contextual factors such as regional economic development, population density, and geographic latitude (9). Another investigation conducted in Changde City, Hunan Province, identified a positive association between TB incidence and selected meteorological parameters, particularly mean ambient temperature and duration of sunshine (10).
The incidence of PTB in middle-income countries such as Iran is shaped not only by meteorological conditions but also by levels of HDI. This relationship operates through interconnected pathways, as socially and economically disadvantaged populations are disproportionately exposed to a range of environmental, biological, and structural risk factors that increase vulnerability to TB (11).
Recently, Bayesian inference approaches have been increasingly applied to the modeling of spatially aggregated colorectal and gastric cancer data. These models are particularly effective in capturing spatial dependence and heterogeneity by incorporating spatially structured random effects and covariance structures, thereby enabling a more accurate representation of geographic variation in disease risk (12). However, Bayesian spatial models require prior knowledge to select appropriate prior distributions for random parameters; otherwise, it may lead to inaccurate modeling results.
Another commonly applied approach is geographically weighted regression (GWR), which extends conventional regression frameworks by explicitly incorporating spatial heterogeneity into the modeling process. This method operates by fitting a series of localized regression models, each calibrated using spatially weighted observations, thereby allowing relationships between variables to vary across geographic space (13). GWR models offer strong theoretical interpretability, providing a direct and transparent understanding of how pulmonary tuberculosis incidence relates to various influencing factors across different regions. More recently, advances in machine learning have facilitated the integration of geographically weighted frameworks with machine learning algorithms, enhancing predictive performance while accounting for spatial heterogeneity (14). Meanwhile, the GWRF approach aims to enhance predictive accuracy compared with a conventional RF model by explicitly incorporating spatial heterogeneity in the effects of influencing factors.
This study aims to evaluate the impact of climatic variables and the HDI on PTB incidence over the study period using a RF and GWRF model. Furthermore, it seeks to forecast PTB incidence at the provincial level across Iran for the next five years, up to 2028. Additionally, this research endeavors to project future PTB incidence and assess Iran's progress toward achieving the World Health Organization's End TB Strategy targets for 2030. To accomplish these aims, a GWRF model is employed to project incidence trends and provide scientific evidence to facilitate timely adjustments in public health planning and resource allocation.
Method
This study utilizes an ecological design, with provinces serving as the unit of analysis rather than individual-level data.
Study area
Iran is administratively divided into 31 provinces, which are further subdivided into counties, districts, and rural districts. Geographically, the country is located between 25°N and 40°N latitude and 44°E and 63°E longitude.
Prediction of Meteorologic Indicators
Decadal climate predictions from 22 international climate models were utilized in this study. Following the World Meteorological Organization (WMO) standard protocols, outputs from three global decadal models—MPI-ESM1.2-LR, MIROC6, and CNRM-ESM2-1 were extracted, including summer and winter dry temperatures (with winter temperature representing the annual mean), mean annual relative humidity (%), mean dew point temperature, and mean wet temperature. These datasets were obtained from the Global Precipitation Climatology Centre and the European Centre for Medium-Range Weather Forecasts (15).
Subnational HDI
The HDI, a composite measure encompassing health, education, and standard of living, was used to assess provincial-level development. HDI data and its subcomponents were obtained from the Global Data Lab (https://globaldatalab.org/shdi/table/).
PTB Data
The study included both sputum smear-positive and sputum smear-negative PTB cases. Smear-positive cases were defined as those in which Mycobacterium tuberculosis was detected via acid-fast bacilli (AFB) in sputum samples using microscopy, whereas smear-negative cases were confirmed PTB cases with negative sputum smear results. The dataset, provided in Excel format following an official request to the Vice Presidency for Research at Hamadan University of Medical Sciences, contained variables including TB type, patient nationality, province, affiliated medical university, and year of diagnosis.
Population Data
Population data were obtained from the Statistical Center of Iran (www.amar.org.ir) based on the national censuses conducted in 2011, 2016, and 2017 (16). For inter-census years, population estimates provided by the Statistical Center were used (17).
Population Prediction Between Census Years
To estimate provincial populations for intercensal years, the average annual growth rate was calculated. For instance, the population of West Azerbaijan Province increased from 3,080,576 in 2011 to 3,265,219 in 2016, corresponding to an average annual growth rate of 1.17%. The average annual growth was computed using the following formula:
In this regard,
Pn: Population at the end of the period
P0: Population at the beginning of the period
r: Average annual population growth
n: Time interval between the beginning and end of the period in years
Meteorological indicators
Meteorological data were accessed from the Iran Meteorological Organization website (www.irimo.ir) via the "Custom Data Request" section available at https://data.irimo.ir. For each province, monthly data were extracted from counties with meteorological stations, and provincial averages were computed.
Number of Synoptic Weather Stations
For example, in West Azerbaijan Province, data from 19 of the 28 available synoptic weather stations were included. Stations were selected based on the criterion that they were located in county centers reporting PTB cases.
Meteorological variables were selected based on a review of previous studies and their impact on model performance. For instance, among three humidity-related measures-maximum humidity, monthly average humidity, and mean relative humidity-the most appropriate variable was chosen to optimize model fit.
SIR
The number of observed PTB cases in each province (Yi) was assumed to follow a Poisson distribution,
Yi~Poisson (Ei θi)
represents the SIR for province i. For each province i, (i = 1,…, n, the SIR was calculated using the following formula:
SIRi = Yi/Ei
Where Ei represents the expected cases, which can be calculated as follows:
Ei = ni
, i = 1, 2, …I
In this equation, Yi denotes the observed number of PTB cases in province i, and ni represents the population of that province. An SIR value greater than 1 indicates that the observed number of cases exceeds the expected number.
Random Forest
RF is an ensemble learning method that constructs multiple decision trees using random feature selection and bootstrap sampling, enhancing model robustness and generalizability. By building each tree on random subsets of samples and features, RF effectively handles large datasets and reduces the impact of feature correlations and noise in high-dimensional data. Furthermore, RF does not require assumptions about the underlying statistical distribution of the data or predefined relationships between the dependent and independent variables, making it particularly suitable for capturing nonlinear effects of predictors.
Each decision tree in the RF is independently generated and trained using a bootstrap sample drawn with replacement from the original training dataset, typically comprising approximately two-thirds of the data. The remaining one-third of the data, referred to as the out-of-bag (OOB) set, is excluded from training and used to evaluate model performance. For each tree, a random subset of predictor variables (m) is selected from the total set of k variables for splitting at each node. Each decision tree is grown to its maximum depth without pruning, continuing until no further splits are possible. Prediction errors are subsequently calculated using the OOB samples.
GWRF
The variance inflation factors (VIF) are calculated for all factors to execute a multi-collinearity check. As depicted in Fig. 1(a), the VIF values for all variables are less than 2.34, indicating no multi-collinearity with other variables. The results of the correlation test, as shown in Fig. 1(b), reveal a strong positive correlation between the dew temperature and wet temperature (r = 0.50), temperature and wet temperature (r = 0.47).
Fig. 1
Correlation and collinearity test results of variables
Model Accuracy Metrics
In the training process of the GW-RF model, the optimal hyperparameters are fine-tuned using the K-fold cross-validation method. The hyper-parameters of the GWRF (‘‘ntree’’: the number of trees, and ‘‘mtry’’: the number of variables randomly sampled) are determined using Random Grid Search (RGS). Then, these hyper-parameters are kept fixed to train the local GWRF model. The bandwidth is also determined using the ten-fold cross-validation method. During the parameter tuning process, both the global RF and local GWRF models are trained with the aforementioned PTB data.
To assess the predictive performance of the GWRF model and other models, several evaluation metrics are used, including Mean Square Error (MSE), Akaike Information Criterion (AIC) and goodness of fit (𝑅2):
AIC = 2K-2 ln (L)
where 𝑦𝑖 is the true value for observation 𝑖,
denotes the predicted value of observation 𝑖,
means the average value of the dependent variable, 𝑛 is the total sample size, 𝑘 is the number of factors, 𝐿 means the maximum likelihood estimate of the model.
The GWRF model facilitates the assessment of feature importance for explanatory variables at each location, which aids in exploring the spatial heterogeneity in the effects of a factor across different zones. Local permutation feature importance is calculated for each zone based on the local RF, providing feature importance values for a factor in different zones.
Descriptive analyses were performed to summarize the study variables using measures of central and variability, including the mean, median, standard deviation, and range (minimum and maximum values).
Software
The entire study was conducted in the R programming environment (18) and RF (19) was used for RF calculations, for mapping, ggplot2 for visualizing of data, and caret R packages for data preparation and separation, using the “GWmodel” and “spatialRF” packages (20).
Ethical statement
This study received ethical approval from the Hamadan University of Medical Sciences Ethics Committee (IR.UMSHA.REC.1403.577). Patient consent was waived, as all data were obtained from publicly accessible databases in Iran.
Results
Epidemiological characteristics description
We analyze the local effects of the influencing factors using the GWRF model. The results of GWRF are summarized in Table 1, presenting descriptive statistics of the estimated coefficients for influencing factors across various provinces. These statistics provide general views on the variances in the effects of influencing factors. Across all variables, local coefficients exhibit both positive and negative values. For example, the local coefficients for HDI range from 0.64 to 0.82, with a median value of 0.77.
A
A
Fig. 2
shows the spatial distribution of PTB among province-level during 15 years mean (2009–2023), hot spot analysis shows high clustering of PTB in the Sistan and Baluchestan, Golestan, Khuzestan, and Khorasan Razavi. While there is cold spot of PTB were observed in Chaharmahal and Bakhtiari, Shiraz, Isfahan, Kohgiluyeh and Boyer-Ahmad, and Qazvin.
A
Fig. 2
Identification of hot-spot and cold-spot provinces based on the Moran's Index.
Comparative Predictive Performance of RF and GWRF Models
The R2 of the global RF model is 0.97, which serves as a baseline for comparison. The GWRF model exhibits a higher average adjusted R2 values (0.97) and a lower AIC value [see Table 2], indicating superior predictive performance over the global RF model. The GWRF model shows a similar MSE value than that of the global RF model.
A
A
Fig. 3
Average changes in the SIR of PTB in Iran (2009 to 2023)
GWRF-Based Prediction of PTB by Province
Table 3 showed the prediction of SIR of PTB over the next 5 years based on the GWRF model, indicating that 16 provinces had an increasing trend, with the highest increases belonging to Qom, Kerman, and Ilam provinces. Fifteen provinces showed a decreasing trend, with the smallest decreases belonging to Sistan and Baluchestan, Kermanshah, and North Khorasan provinces.
A
Fig. 4
Spatial heterogeneity distribution of local effects of first five influencing factors on SIR of PTB.
Partial Effects of Key Variables on PTB SIR
The interesting results of five important variables are demonstrated in Fig. 5. When controlling for the influence of other factors, the HDI (Fig. 5(b)), are all negatively related to area-adjusted SIR of PTB.
A
A
Fig. 5
Partial dependency profiles of the Five important variables of the global random forest model.
Discussion
We analyzed the spatiotemporal distribution characteristics of TB cases from 2009 to 2023 and conducted the first-ever nationwide predictive research on the of PTB in Iran during 2024 to 2028. Additionally, the inclusion of predictive analyses enhances the ability to evaluate the effectiveness of existing TB health policies and prediction.
Following the GWRF analysis, Sistan and Baluchestan and Golestan provinces were identified as hotspot areas for PTB during the study period, leading us to select these regions for model development. In line with this, a study by Salek et al. reported that the incidence of smear-positive PTB in Golestan Province in 2003 was 22.1 per 100,000 population, compared to the national incidence rate of 7.8 per 100,000 (21). The elevated TB incidence in Golestan Province may be partly attributed to the influx of immigrants from Sistan and Baluchestan Province. The subsequent decline in cases is likely associated with improvements in healthcare services and the implementation of public health interventions (22).
The local importance of the remaining variables did not show substantial spatial heterogeneity. Among these, HDI and temperature were identified as having a strong influence on PTB SIR across the study area. We observed a positive association between regions with low HDI and a higher PTB burden, a finding that aligns with previous research. For instance, Feliciano et al. reported that increases in chronic diseases are often linked to greater social disparities, a pattern observable in southern and southeastern regions where higher TB incidence corresponds with socioeconomic indicators such as local HDI (23). The results of Muniyandi et al. (24), and Rodríguez-Moraleset al. (25) showed a higher concentration of TB in countries with low HDI. According to some studies, poverty, income inequality and lack of social capital were important predictors of an increase in TB incidence (26). According to the GWRF model, HDI was one of the most influential factors driving the spatial variability of TB in Iran. In the subsequent five-year incidence predictions, HDI ranked second in importance among all input variables. Our findings indicate a strong inverse relationship between HDI increases over time and TB incidence rates. Globally, countries with higher HDI have experienced more rapid declines in TB incidence, and investments in healthcare, rather than social protection alone, are key determinants of reduced TB burden (27).
The GWRF model outperformed conventional RF models in predicting the spatial distribution of PTB incidence. By incorporating spatial heterogeneity, GWRF effectively captures local variations in the effects of socioeconomic and meteorological factors, which may be obscured in global models. Among meteorological variables, temperature was found to have the strongest influence on TB incidence, exhibiting a positive correlation with case rates (28). Without considering lag effects, globally, each unit increase in national annual average temperature was associated with a 0.89% reduction in TB age-SIR (95% CI: 0.60–1.18%) and a 1.61% reduction in age-standardized mortality rate (95% CI: 1.27–1.95%) (29). However, the results of the studies in various regions were not the same, such as North China, where the incidence of TB was negatively correlated with temperature (28), and Jiangsu Province, where temperature had no significant effect on the incidence of TB (30). These areas already bear a high TB burden, making the population more vulnerable to environmental changes. Warmer temperatures may worsen health conditions such as malnutrition, HIV co-infection, and other comorbidities, increasing vulnerability to TB, while socio-economic factors like overcrowding, poor sanitation, and limited healthcare access exacerbate its spread (31, 32).
Conclusions
This study demonstrates that the GWRF model provides a robust framework for capturing spatial heterogeneity in PTB incidence across Iran, surpassing the predictive performance of the global random forest model. Analysis revealed persistent provincial hotspots, temporal stability in PTB of SIR, and notable local variations in key determinants. Meteorological factors, including temperature and humidity, exerted stronger effects in southern and northern provinces, whereas higher HDI consistently corresponded to lower PTB risk. Five-year prediction highlighted provinces with increasing PTB trends, emphasizing the need for targeted, region-specific interventions.
Declarations
A
Author Contribution
SGh: Conceptualization, writing – original draft, Methodology, writing – review & editing, Formal analysis. EA: Conceptualization, writing – original draft, Methodology, Writing – review & editing, Formal analysis. LT: Formal analysis, Writing – review & editing . MN: Data curation . [SKh](https:/www.benthamdirect.com/search?value1=Salman+Khazaei&option1=author&noRedirect=true&sortField=prism_publicationDate&sortDescending=true) : Formal analysis, Writing – review & editing.
A
Funding
The author(s) declare financial support for the research, authorship, and/or publication of this article. This work was supported by grants from Hamadan University of Medical Sciences Ethics Committee (IR.UMSHA.REC.1403.577).
|
Variables
|
Min
|
Max
|
Mean
|
Median
|
Std
|
|---|---|---|---|---|---|
|
Temperature
|
9.40
|
43.90
|
26.97
|
26.10
|
7.00
|
|
Dew temperature
|
1.43
|
31.74
|
12.59
|
11.61
|
7.33
|
|
Wet temperature
|
3.32
|
33.52
|
20.01
|
19.09
|
4.92
|
|
Relative humidity
|
10.46
|
98.83
|
68.65
|
69.23
|
16.73
|
|
HDI
|
0.64
|
0.82
|
0.76
|
0.77
|
0.03
|
| Analysis of hot and cold spot provinces | |||||
|
Models
|
Total SIR of PTB
|
Area-adjusted for independent variables
|
||||||
|---|---|---|---|---|---|---|---|---|
|
AIC
|
MSE
|
R2
|
OOB R2
|
AIC
|
MSE
|
R2
|
OOB R2
|
|
|
RF
|
-2031.2
|
0.01
|
0.97
|
0.87
|
-2032.9
|
0.01
|
0.97
|
0.87
|
|
GWRF
|
-2051.7
|
0.01
|
0.97
|
0.90
|
-2152.4
|
0.009
|
0.98
|
0.89
|
| Temporal Variation in PTB SIR | ||||||||
| Figure [3], for the period from 2009 to 2023, the highest mean SIR was in the years 2009 and 2013, and the lowest SIR was in 2020. Province-level SIR of PTB remains relatively stable through these years, with notable decrease in the year 2020 concentrated. A total decrease of 0.04 units was observed in both models. | ||||||||
|
Name of provinces
|
2024
|
2025
|
2026
|
2027
|
2028
|
|---|---|---|---|---|---|
|
Sistan va Balochestan
|
3.34
|
3.27
|
3.20
|
3.14
|
3.08
|
|
Golestan
|
2.80
|
2.74
|
2.69
|
2.64
|
2.58
|
|
Khuzestan
|
1.72
|
1.69
|
1.65
|
1.62
|
1.59
|
|
Guilan
|
1.66
|
1.63
|
1.60
|
1.56
|
1.53
|
|
Khorasan Razavi
|
1.61
|
1.58
|
1.55
|
1.52
|
1.49
|
|
Mazandaran
|
1.26
|
1.23
|
1.21
|
1.18
|
1.16
|
|
Kermanshah
|
1.13
|
1.11
|
1.08
|
1.06
|
1.04
|
|
Khorasan North
|
1.11
|
1.09
|
1.07
|
1.05
|
1.03
|
|
Hormozgan
|
1.08
|
1.06
|
1.04
|
1.02
|
1.00
|
|
Kurdistan
|
1.08
|
1.06
|
1.04
|
1.02
|
1.00
|
|
Khorasan South
|
0.87
|
0.85
|
0.84
|
0.82
|
0.80
|
|
Lorestan
|
0.84
|
0.82
|
0.81
|
0.79
|
0.77
|
|
Qom
|
0.83
|
0.82
|
0.80
|
0.79
|
0.77
|
|
Yazd
|
0.83
|
0.81
|
0.79
|
0.78
|
0.76
|
|
Markazi
|
0.79
|
0.77
|
0.76
|
0.74
|
0.73
|
|
Kerman
|
0.78
|
0.76
|
0.74
|
0.73
|
0.71
|
|
Ardebil
|
0.76
|
0.75
|
0.74
|
0.72
|
0.71
|
|
Alborz
|
0.75
|
0.74
|
0.72
|
0.71
|
0.70
|
|
Tehran
|
0.75
|
0.73
|
0.72
|
0.70
|
0.69
|
|
Semnan
|
0.72
|
0.70
|
0.69
|
0.68
|
0.66
|
|
Ilam
|
0.67
|
0.66
|
0.60
|
0.63
|
0.62
|
|
Bushehr
|
0.56
|
0.55
|
0.54
|
0.53
|
0.51
|
|
Hamadan
|
0.52
|
0.51
|
0.50
|
0.49
|
0.48
|
|
Zanjan
|
0.52
|
0.51
|
0.50
|
0.49
|
0.48
|
|
West Azerbaijan
|
0.51
|
0.49
|
0.48
|
0.48
|
0.47
|
|
East Azerbaijan
|
0.46
|
0.45
|
0.44
|
0.43
|
0.42
|
|
Qazvin
|
0.43
|
0.42
|
0.41
|
0.41
|
0.40
|
|
Isfahan
|
0.38
|
0.37
|
0.36
|
0.35
|
0.35
|
|
Kohgiluyeh and Boyerahmad
|
0.35
|
0.35
|
0.34
|
0.33
|
0.32
|
|
Shiraz
|
0.28
|
0.27
|
0.26
|
0.26
|
0.25
|
|
Chaharmahal Bakhtiari
|
0.19
|
0.19
|
0.19
|
0.18
|
0.18
|
| Spatial Heterogeneity of PTB Determinants | |||||
| For the PTB incidence factors, Fig. 4(a) shows that importance of temperature on PTB incidence varies significantly in different provinces The high SIR values of ‘‘temperature’’ are mainly concentrated in the southern and northern regions, indicating the temperature has higher impacts on PTB incidence in these areas than other areas. Figure 4(b) demonstrate that the importance of HDI is high in some provinces in the southern regions such as Sistan va Balochestan, Hormozgan, Shiraz, Yazd, Kurdistan, Mazandaran, Tehran, Alborz, Markazi, and Kohgiluyeh and Boyerahmad. Figure 4(c–e) demonstrate that the importance of relative humidity, wet temperature and dew temperature is high in some cities in the southern and eastern regions. | |||||
Ethics approval and consent to participate
A
This study utilized publicly available data from the Global Data Lab, the Statistical Center of Iran, and the Iran Meteorological Organization. The study protocol was approved by the Ethics Committee of Hamadan University of Medical Sciences (IR.UMSHA.REC.1403.577). Patient consent was not required, as all data were aggregated at the provincial level and publicly accessible.A
Data Availability
The datasets used in this study are publicly available from the Global Data Lab, the Statistical Center of Iran, and the Iran Meteorological Organization.
Abbreviations
Pulmonary Tuberculosis
PTB
Geographically Weighted Random Forest
GWRF
Human Development Index
HDI
Standardized Incidence Ratio
SIR
Consent to publish
All the authors consent to publish the article in its present form.
Competing of Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
The authors would like to express their sincere appreciation to all staff involved in the diagnosis and reporting of PTB cases at healthcare centers.
References:
1.
World Health Organization (WHO). Global tuberculosis report 2024 [Internet]. WHO. 2024 [cited 2024 Dec 30].
2.
United Nations (UN). Sustainable development goals [Internet]. UN. 2024 [cited 2024 Dec 30]. Available from: https://sdgs.un.org/goals
3.
Churchyard G, Kim P, Shah NS, Rustomjee R, Gandhi N, Mathema B, et al. What we know about tuberculosis transmission: an overview. J Infect Dis. 2017;216(suppl6):S629–35.
4.
Management CfID. https://icdc.behdasht.gov.ir/TB_status/%D8%A2%D8%AE%D8%B1%DB%8C%D9%86-%D9%88%D8%B6%D8%B9%DB%8C%D8%AA-%D8%A8%DB%8C%D9%85%D8%A7%D8%B1%DB%8C-%D8%B3%D9%84-%D8%AF%D8%B1%D8%A7%DB%8C%D8%B1%D8%A7%D9%86-%D8%B3%D8%A7%D9%84-1403 2023 [.
5.
Xue M, Zhong J, Gao M, Pan R, Mo Y, Hu Y, et al. Analysis of spatial–temporal dynamic distribution and related factors of tuberculosis in China from 2008 to 2018. Sci Rep. 2023;13(1):4974.
6.
Li X, Chen D, Zhang Y, Xue X, Zhang S, Chen M, et al. Analysis of spatial-temporal distribution of notifiable respiratory infectious diseases in Shandong Province, China during 2005–2014. BMC Public Health. 2021;21(1):1597.
7.
Romanyukha AA, Karkach AS, Borisov SE, Belilovsky EM, Sannikova TE, Krivorotko OI. Small-scale stable clusters of elevated tuberculosis incidence in Moscow, 2000–2015: Discovery and spatiotemporal analysis. Int J Infect Dis. 2020;91:156–61.
8.
Amsalu E, Liu M, Li Q, Wang X, Tao L, Liu X, et al. Spatial-temporal analysis of tuberculosis in the geriatric population of China: An analysis based on the Bayesian conditional autoregressive model. Arch Gerontol Geriatr. 2019;83:328–37.
9.
Khorshid MR, Behzadi S, Sharifi A, Vafaeinejad A, Abbasian Z, Naderi H. Evaluation of spatial and non-spatial factors on tuberculosis using geospatial information system and fuzzy logic. Spat Spatio-temporal Epidemiol. 2025;54:100729.
10.
Sun H, Ren J, Wang J, Yu J, Zuo L, Wu X, et al. Spatial distributions, potential sources, and ecological risks of heavy metals in stream sediments in Zambia. J Geochem Explor. 2025;270:107659.
11.
Okhovat-Isfahani B, Bitaraf S, Mansournia MA, Doosti-Irani A. Inequality in the global incidence and prevalence of tuberculosis (TB) and TB/HIV according to the human development index. Med J Islamic Repub Iran. 2019;33:45.
12.
Ayubi E, Niksiar S, Amlashi ZK, Talebi-Ghane E. Spatiotemporal Mapping of Colorectal and Gastric Cancer Incidence in Hamadan Province, Western Iran (2010–2019). J Res Health Sci. 2025;25(2):e00650.
13.
Liu J, Khattak AJ, Wali B. Do safety performance functions used for predicting crash frequency vary across space? Applying geographically weighted regressions to account for spatial heterogeneity. Accid Anal Prev. 2017;109:132–42.
14.
Quiñones S, Goyal A, Ahmed ZU. Geographically weighted machine learning model for untangling spatial heterogeneity of type 2 diabetes mellitus (T2D) prevalence in the USA. Sci Rep. 2021;11(1):6955.
15.
WCRP. (2022). WMO-WCRP portal: https://www.wcrp-climate.org/dcp-overview; available in: 24/1/2020.
16.
Statistical center of Iran (SCI). General population census, 2016. Iran, Tehran. Available from: https://www.amar.org.ir/
17.
https://www.amar.org.ir/ ScoISEtpotiscftIScoISAf.
18.
Team R. R: A language and environment for statistical computing. R Found Stat Comput. 2016;1:409.
19.
Liaw A, Wiener M. Classification and regression by randomForest. R news. 2002;2(3):18–22.
20.
Georganos S, Kalogirou S. A forest of forests: a spatially weighted and computationally efficient formulation of geographical random forests. ISPRS Int J Geo-Information. 2022;11(9):471.
21.
Salek S, Salek S, Emami H, Masjedi MR, Velayati AA. Epidemiologic status of tuberculosis in Golestan province. 2008.
22.
Yousefi R. Tuberculosis Incidence in Iran and Neighboring Countries from 2010 to 2023. Int J Adv Stu Hum Soc Sci. 2025;14(4):270–89.
23.
Feliciano SCC, Villela PB, Oliveira GMM. Associação entre a mortalidade por doenças crônicas não transmissíveis e o Índice de Desenvolvimento Humano no Brasil entre 1980 e 2019. Arquivos brasileiros de cardiologia. 2023;120:e20211009.
24.
Muniyandi M, Ramachandran R. Socioeconomic inequalities of tuberculosis in India. Expert Opin Pharmacother. 2008;9(10):1623–8.
25.
Rodríguez-Morales AJ, Castañeda-Hernández DM. Relationships between morbidity and mortality from tuberculosis and the human development index (HDI) in Venezuela, 1998–2008. Int J Infect Dis. 2012;16(9):e704–5.
26.
Arenas NE, Quintero-Álvarez L, Rodríguez-Marín K, Gómez-Marín JE. Análisis sociodemográfico y espacial de la transmisión de la tuberculosis en la ciudad de Armenia (Colombia). Infectio. 2012;16(3):154–60.
27.
Költringer FA, Annerstedt KS, Boccia D, Carter DJ, Rudgard WE. The social determinants of national tuberculosis incidence rates in 116 countries: a longitudinal ecological study between 2005–2015. BMC Public Health. 2023;23(1):337.
28.
Chang M, Emam M, Chen X, Lu D, Zhang L, Zheng Y. An investigation of the effects of meteorological factors on the incidence of tuberculosis. Sci Rep. 2024;14(1):2088.
29.
Liu Q, Wang Y, Liu M, Zhao Y, Liu J. The influence and lag-effect of temperature and precipitation on the incidence and mortality of tuberculosis, 2000–2021: an observational study. Front Public Health. 2025;13:1572422.
30.
Liu W, Ji H, Shan J, Bao J, Sun Y, Li J, et al. Spatiotemporal dynamics of hand-foot-mouth disease and its relationship with meteorological factors in Jiangsu Province, China. PLoS ONE. 2015;10(6):e0131311.
31.
Mori T, Leung CC. Tuberculosis in the global aging population. Infect Disease Clin. 2010;24(3):751–68.
32.
Byng-Maddick R, Noursadeghi M. Does tuberculosis threaten our ageing populations? BMC Infect Dis. 2016;16(1):119.