Spatial and spatio-temporal patterns of road traffic accident mortality in Bangladesh,2022–2024: A district-level geospatial analysis

Abstract

Objective

Road traffic accidents (RTAs) are a leading cause of injury deaths in Bangladesh, but local geographic patterns remain poorly understood. This study aimed to identify high- and low-risk districts for RTA mortality and track changes from 2022 to 2024 to inform targeted interventions.

Methods

In this district-level ecological study, we analyzed DGHS official RTA mortality data across all 64 districts. Population-adjusted rates were computed; Moran’s I and Getis-Ord Gi* assessed spatial autocorrelation for hotspots/coldspots; SaTScan detected space-time clusters.

Results

Raw counts were randomly distributed, but adjusted rates showed significant dispersion (Moran’s I = -0.143, p=0.006 in 2023), with high-risk districts bordering low-risk ones. Persistent northwestern spatial hotspots included Naogaon, Natore, and Bogra. Spatio-temporal analysis identified high-rate clusters in 2024 for Dhaka (RR 6.14, p=0.001) and Rajshahi (RR 10.07, p=0.001), alongside stable low-mortality clusters in northern (RR 0.07) and southern (RR 0.19) regions. Methodological limitations include the use of uniform population projection rates and the ecological design, which precludes causal inference.

Conclusion

RTA mortality reflects distinct local epidemics, not a uniform crisis. Geographic disparities support a precision public health approach, with prioritization of high-risk corridors and urban zones for targeted interventions, alongside investigation of protective factors in low-risk clusters.

Keywords

road traffic accident mortality spatial analysis spatio-temporal clustering SaTScan Bangladesh geospatial epidemiology hotspot detection public health intervention systems thinking emergency medicine

Significance for public health

While RTAs are a known major cause of mortality in Bangladesh, existing evidence lacks granular geographic detail. This study addresses this important gap by providing the first district-level spatio-temporal analysis across all 64 districts. By revealing a “checkerboard” pattern of risk dispersion, pinpointing alarming emerging urban clusters (Dhaka: RR 6.14; Rajshahi: RR 10.07), and identifying stable low-mortality regions, it moves beyond national averages to expose notable local inequalities. These findings offer an evidence-based roadmap for precision public health, enabling targeted allocation of limited resources to high-risk zones for enforcement and trauma care, while investigating protective factors in safer areas for replication, thereby maximizing the impact of road safety interventions.

Introduction

Road traffic collisions represent a critical and ongoing global health crisis, resulting in approximately 1·19 million fatalities annually worldwide.¹ Injuries related to road traffic accidents remain the leading cause of death among individuals aged 5–29 years globally, despite a modest 5% reduction in total deaths since 2010, even in the context of rapid population growth and increased motorization.^1,2 A marked disparity exists between country income groups: the risk of death is approximately three times higher in low-income countries compared to high-income countries, with low- and middle-income countries (LMICs) accounting for around 90% of global traffic fatalities.^1,2 In contrast, high-income countries (HICs) have demonstrated the most substantial long-term improvements, with regions such as Western Europe achieving a 76% decline in death rates since 1990.² Regionally, the South-East Asia Region bears the largest share of global road traffic deaths (28%) and is the only region where pedestrian fatalities have increased, rising by 42% since 2010; this trend is accompanied by a 273% increase in powered two- and three-wheeler usage since 2011.¹ The African Region continues to have the highest fatality rates globally, with total deaths increasing by 17% since 2010.¹ Conversely, the European Region represents the safest road environment, reporting the largest decline in deaths, with a 36% reduction since 2010.¹ In Bangladesh, the scenario is particularly alarming, with official estimates suggesting that the country experiences one of the highest road traffic injury (RTI) burdens globally.³ Prior reporting indicates road accidents are a serious and growing issue, with an average of 20 people dying daily.^4,5 The global affliction of road traffic accidents are pictured in Table 1.

Table 1.

Global burden of road traffic accidents.

Category/region	Annual deaths (approx.)	Age-standardized death rate per 100,000 people
World (global)	1.19 million¹, 1.27 million²	15¹, 15.54²
Global LMICs (low- & middle-income countries)	92% of global deaths occur in low- and middle-income countries¹	16–21¹; 17.27–23.96²
Global HICs (high-income countries)	8% of global deaths occur in high-income countries¹	8¹; 8.36²
South-East Asia (WHO)/South Asia (GBD)	330,222¹; 299,354,² 28% of global burden	16¹; 16.92².
Africa (WHO region)	∼225,482 deaths; 19% share of global burden¹	19¹, 26.55²
Europe (WHO region)	∼62,670 deaths; 5% share of global burden¹	7¹; 3.96 in Western Europe²

Effective road safety requires a comprehensive, systems-thinking approach that targets higher-level factors such as institutional controls, policy development, and enforcement, moving beyond a sole focus on road users.⁶ Local environmental factors and road infrastructure strongly influence where accidents cluster, indicating that sustainable road safety requires not just education and enforcement but also robust public investment in safer infrastructure.⁷

In LMICs, however, these infrastructure improvements and the implementation of effective safety strategies are often hindered by complex sociotechnical system deficiencies, including organizational fragmentation, competing financial priorities, and significant political interference.⁸ To maximize the efficacy of limited resources and overcome these systemic challenges, road safety interventions, such as, targeted enforcement, infrastructure modifications, and trauma care upgrades ─ all must be strategically and pragmatically deployed to the precise geographical areas of highest risk.⁵

Systematic identification of high risk road locations, commonly termed black spots, is a foundational step in evidence based road safety management.^9,10 Various methods exist for this purpose, including sliding window analysis, spatial autocorrelation, and empirical Bayesian approaches, with method performance varying by road type and speed characteristics.⁹ For instance, spatial clustering techniques such as Getis Ord Gi* are particularly effective on low speed urban roads where accidents tend to cluster around conflict points, whereas sliding window methods may perform better on high speed roads where accident patterns are more dispersed.⁹ A systems thinking perspective, which considers interactions between road users, vehicles, infrastructure, and the external environment, is essential for comprehensive black spot auditing.¹¹ Comparative analyses across European countries reveal substantial variation in black spot identification practices, with no single universal approach; however, coupling statistical methods with accident severity indices yields more holistic assessments of road infrastructure safety.¹⁰

Traditional analyses often rely on aggregated national data, which can obscure underlying geographical variations and complex local patterns in collision risk.⁵ To address this deficit, this study performs a spatial and spatio-temporal analysis of RTA deaths in Bangladesh. Utilizing district-level mortality data from the Directorate General of Health Services (DGHS) Medical Certification of Cause of Death (MCCD) Dashboard spanning 2022 to 2024, we applied rigorous geospatial methodologies, including Moran’s I, Getis-Ord Gi*,¹² and SaTScan.^13,14 The primary objective is to delineate the geographical structure of RTA mortality, identifying emerging spatial hotspots and high-risk spatio-temporal clusters as well as the cold ones to provide an evidence-based foundation for prioritising and tailoring public health and road safety interventions.

Methodology

Study design and data sources

This ecological study analyzed district-level road traffic accident (RTA) deaths across 64 districts of Bangladesh from 2022 to 2024.

Mortality data

Data on RTA fatalities were obtained from the Directorate General of Health Services (DGHS) Medical Certification of Cause of Death (MCCD) Dashboard,¹⁵ a national registry of medically certified deaths. Age and sex breakdowns were not available in the aggregate DGHS data releases used for this study.

Population data

Population denominators were derived from the Population and Housing Census 2022 Preliminary Report, compiled and published by the Bangladesh Bureau of Statistics.¹⁶ Populations for 2023 and 2024 were projected by applying a uniform 1.12% annual national growth rate to the 2022 census figures. The projection was calculated as:

{Population}_{t} = {Population}_{t - 1} \times (1 + 0 \cdot 0112)

(1)

For reference, populations for 2024 were also projected using the same constant annual growth rate applied over three years:

{POP}_{2024} = {POP}_{2022} \times {(1 + 0 \cdot 0112)}^{2}

(2)

The average national population for the period 2021–2025 was 167, 003, 180. The population data used in the study are incorporated in Supplementary Table 5.

Geographic data

The geographic boundaries of Bangladesh’s 64 districts, including longitude and latitude in decimal degrees, were sourced from the Bangladesh Subnational Administrative Boundaries (ADM0–ADM4) dataset. This dataset, provided in the form of shapefiles (.shp) and associated geospatial data, is available from the United Nations Office for the Coordination of Humanitarian Affairs (OCHA) through the Humanitarian Data Exchange (HDX) portal.¹⁷ The dataset is released under the Creative Commons Attribution 3·0 Intergovernmental Organization (CC BY 3·0 IGO) license.¹⁷ Centroids for each district were calculated and saved using GeoDa software.¹⁸ For consistent visualization across all maps and plots, the EPSG:3857 projection was used. Contextual basemaps were obtained from OpenStreetMap¹⁹ tiles via the contextily library.²⁰ The geographic data used in the study are incorporated in Supplementary Table 2.

Data cleaning and validation

All district-level case and population data were cross-checked for consistency across DGHS weekly bulletins and press releases. The absence of any missing values was confirmed via DGHS archival reports (Directorate General of Health Services (DGHS), 2019). Data entry errors were carefully avoided throughout the compilation process.

Operational definitions

Hotspots & coldspots (local analysis)

• Hotspots: Districts with significantly high RTA mortality compared to their immediate neighbors, identified by Getis-Ord Gi* (positive z-score, p < 0·05).¹²

• Coldspots: Districts with significantly low RTA mortality compared to their immediate neighbors, identified by Getis-Ord Gi* (negative z-score, p < 0·05).¹²

Hot & cold clusters (space-time analysis)

• Hot Clusters: Geographic areas showing significantly higher-than-expected RTA deaths during specific time periods, identified by SaTScan (RR > 1, p < 0·05).^13,14

• Cold Clusters: Geographic areas showing significantly lower-than-expected RTA deaths during specific time periods, identified by SaTScan (RR < 1, p < 0·05).^13,14

Hotspots and coldspots signify localized, static risk relative to immediate neighbors, whereas space-time clusters represent dynamic, regional risk relative to the national average.

Rate calculation

RTA death rates were calculated as deaths per 100,000 population using the formula:

Rate per 100, 000 = (\frac{Number of RTA Deaths}{Population}) \times 100, 000

(3)

Spatial autocorrelation analysis

Global clustering was assessed using Moran’s I to test whether RTA deaths were randomly distributed or showed clustering/dispersion across districts. Values range from -1 (dispersion) to +1 (clustering); 0 indicates random distribution. Computations were performed in Python using PySAL with 999 Monte Carlo permutations for significance testing (p < 0·05).

Local hotspots and coldspots were identified using Getis-Ord Gi*, where a positive Gi* z-score with p < 0·05 indicates a hotspot and a negative z-score indicates a coldspot. Four spatial weighting schemes were employed: Queen contiguity (districts sharing a border or corner, primary method), Rook contiguity (sharing a border only), and K-Nearest Neighbors (KNN-4, KNN-6). All Gi* analyses were conducted in Python (PySAL, GeoPandas, NumPy). Random seed was set at 12345 to ensure reproducibility.

It is important to note that Getis-Ord Gi* detects local clustering relative to immediate neighbors and can identify statistically significant hotspots even in low-count areas if they form a local peak. In contrast, SaTScan relative risk assesses deviation from the national expected rate.

Space-time cluster detection

SaTScan v10·3·3 was used to detect emerging clusters of high or low RTA death rates over space and time (prospective analysis).^13,14 To ensure the robustness of our findings, we tested four different parameter combinations for the maximum spatial and temporal window sizes. Maximum spatial cluster size was capped at 50% of the population at risk in all models; maximum temporal window varied as specified. 999 Monte Carlo simulations assessed significance (p < 0·05). To ensure reproducibility, the random seed was manually set to 12345 for each simulation by editing the parameter file (.prm). Relative Risk (RR) > 1 indicates higher-than-expected deaths; RR < 1 indicates lower. Full sensitivity results are summarized in Table 5 and visualized in Figure 3. The Discrete Poisson model was selected over the Space-Time Permutation model because it incorporates population-at-risk data, allowing for the calculation of relative risk and the identification of areas with elevated mortality rates, rather than simply an excess of case counts. Confidence intervals for relative risk estimates are not reported, as SaTScan does not routinely produce them for cluster-level output; instead, statistical significance was assessed using p-values from Monte Carlo simulations (999 replications), which are reported for each identified cluster.

Statistical software

Analyses utilised Python 3·9 (PySAL, GeoPandas, Matplotlib, Seaborn) for Moran’s I, Getis-Ord Gi* and visualizing the plots; SaTScan v10·3·3 for space-time clusters.^13,14

Ethics statement

This study analyzed exclusively aggregated, anonymized, and publicly available data from official government portals. No individual-level data was accessed. Following international guidelines (e.g., CIOMS) for epidemiological research using public surveillance data, this study was considered exempt from ethical board approval. The research was conducted with rigorous adherence to scientific integrity and the principles of the Declaration of Helsinki.

Workflow

The data flow (workflow) has been presented in Supplementary Figure 4.

Results

Our analysis of district-level road traffic accident (RTA) mortality in Bangladesh from 2022 to 2024 revealed distinct and evolving geographical patterns. The key findings highlight the critical difference between raw death counts and population-adjusted risk, identify persistent and emerging high-risk areas, and reveal regions of sustained safety.

Global spatial patterns: A checkerboard of risk

The analysis of global spatial autocorrelation yielded a crucial insight. When examining raw death counts, we found no significant clustering across any of the three years (Moran’s I ≈ -0.055, p > 0.05), indicating that the absolute number of fatalities was randomly distributed across the country. However, after adjusting for population size, the mortality rates exhibited a significant negative spatial autocorrelation (Moran’s I ranging from -0.100 to -0.143, p < 0.05). This signifies a dispersed pattern, akin to a checkerboard, where districts with high mortality rates are typically surrounded by districts with low rates. This underscores that local, district-level factors are primary drivers of mortality risk, rather than broad regional trends. The full results of the global spatial analysis are provided in Table 2.

Table 2.

Moran’s I results for raw deaths and rates per 100,000 for 2022, 2023, 2024.

Year	Measures	Moran’s I	P value	Z value
2022	Raw Deaths	-0.0543	0.125	-0.87
2023	Raw Deaths	-0.0570	0.097	-0.91
2024	Raw Deaths	-0.0574	0.118	-0.82
2022	Rate per 100,000	-0.1207	0.052	-1.34
2023	Rate per 100,000	-0.1434	0.006	-1.79
2024	Rate per 100,000	-0.1004	0.031	-1.21

It is important to distinguish between two complementary analytical approaches. Hotspots, identified by Getis-Ord Gi, represent districts with significantly high mortality rates compared to their immediate neighbors in a given year (purely spatial, static). Clusters, identified by SaTScan, represent geographic areas with significantly higher or lower mortality rates compared to the national average over a specific time period (spatio-temporal, dynamic). A district may appear as a hotspot, a cluster, both, or neither, depending on the scale and nature of its risk pattern (Figure 1).

Figure 1.

Significant hotspots of RTA deaths (Getis-Ord G* analysis over 2022-2024). Legend: Hot spot: orange, cold spot: light blue, non significant: grey.

Local hotspots and coldspots: A persistent northwestern corridor

Local hotspot analysis (Getis-Ord Gi*) revealed sharply different patterns depending on whether raw counts or population-adjusted rates were used, with striking convergence over time. In 2022 the country displayed a classic urban–rural divide: rate-adjusted analysis uncovered hidden elevated risk in low-population northwestern and southern districts plus the Chittagong Hill Tracts, whereas raw-count hotspots were almost exclusively concentrated in the Greater Dhaka ring (Gazipur, Manikganj, Munshiganj, Narayanganj, Tangail). From 2023 onward, a distinct western border-belt hotspot emerged that rapidly dominated both metrics simultaneously. By 2024, Chapainawabganj, Naogaon, Meherpur, and Satkhira constituted the clearest and most intense hotspot in the entire dataset, achieving the highest Gi* values under both raw counts and per-capita rates (Chapainawabganj registering the single highest raw Gi* peak of 0.122 across all years and weights). Chapainawabganj’s hotspot status, however, should be interpreted with caution given its extremely low absolute mortality (one death over three years); this finding illustrates the method’s sensitivity to localized relative elevations rather than indicating high absolute burden. Naogaon stood out as the only district that qualified as a significant hotspot by both metrics in 2023 and 2024, while Chapainawabganj also met statistical significance despite recording only one death over three years (see caveat above). The earlier northwestern hotspot visible only in rate-adjusted analysis (Bogra, Natore, Pabna, Rajshahi, Dinajpur, Rangpur) proved far less persistent than the western border-belt hotspot when absolute numbers are considered. Greater Dhaka remained a massive raw-count hotspot throughout the period but lost exceptional status once population size was accounted for. No districts emerged as statistically robust coldspots in any year, indicating that low-risk areas tend to form broader regional patterns rather than isolated districts. A summary of robust hotspots by metric and year is presented in Table 3, and their geographical evolution from 2022 to 2024 is mapped in the supplementary figures (S1–S3).

Table 3.

List of Robust Hotspots (High-Risk Districts, from spatial analyses only) for each year.

Year	Robust districts (with type)	Example Gi* range (across weights)	Interpretation
2022	Rate-only: Bogra, Kurigram, Pabna, Rajshahi, Barisal	Rate 0.019–0.081 Raw 0.050–0.111	Classic split: quiet rural northwest + south show hidden high risk (rate), while all absolute events pile into mega-urban Dhaka ring + CHT. Only Satkhira and the Hill Tracts are genuinely hot by both metrics.
2022	Raw-only: Gazipur, Manikganj, Munshiganj, Narayanganj, Tangail Both: Satkhira, Cox’s Bazar, Bandarban, Khagrachhari, Rangamati, Faridpur	Rate 0.019–0.081 Raw 0.050–0.111
2023	Rate-only: Bogra, Natore, Joypurhat, Pabna, Rajshahi, Sylhet	Rate 0.032–0.089 Raw 0.048–0.110	Dhaka still dominates raw volume, but the same districts start appearing in both columns → early warning of convergence.
2023	Raw-only: Gazipur, Manikganj, Munshiganj, Narayanganj, Tangail Both: Naogaon, Cox’s Bazar, Khagrachhari	Rate 0.032–0.089 Raw 0.048–0.110
2024	Rate-only: Dinajpur, Natore, Rangpur, Sylhet	Rate 0.041–0.122 Raw 0.061–0.122	Full convergence on the western border belt with India. Naogaon, Meherpur, and Satkhira are now the clearest and most intense hotspots by BOTH per-capita risk and absolute volume (★ = single highest peaks in the entire dataset in 2024). Greater Dhaka is still huge in raw numbers but no longer exceptional per person.
2024	Raw-only: Gazipur, Manikganj, Munshiganj, Narayanganj, Tangail Both (very strong): Naogaon ★, Meherpur ★, Satkhira ★	Rate 0.041–0.122 Raw 0.061–0.122

*There is no district common in all 3 years.

Space-time clusters: Emerging crises and sustained safe havens

The space-time analysis identified specific periods and locations where mortality risk surged or plummeted relative to the national average. Dhaka and Rajshahi districts were identified as the most stable and critical high-risk clusters. In these clusters, observed deaths were 6 to 10 times higher than expected (Relative Risk [RR] 6.14 for Dhaka; RR 10.07 for Rajshahi).

In stark contrast, large, stable clusters of significantly low mortality were consistently detected. A prominent cluster of 10 northern border districts, including Dinajpur and Nilphamari, demonstrated a dramatically lower risk, with observed deaths only 7% of what was expected (RR 0.07). Another large cluster of 16 southern districts, including Lakshmipur and Barisal, also showed significantly low mortality (RR 0.19). These findings confirm a substantial spatial inequality in RTA mortality risk across Bangladesh. The details of the significant space-time clusters from our base model are shown in Table 4. A sensitivity analysis confirmed that the core findings for the Dhaka and Rajshahi hotspots, as well as the northern cold cluster, were robust across different analytical parameters, as summarized in Table 5. The geographical location of these clusters is visualized in Figure 2, and a comparison across sensitivity parameters is shown in Figure 3.

Table 4.

Significant space-time clusters from spatiotemporal analysis (Base Model: 50% Spatial/50% Temporal).

Rank	Type	Location	Time	Observed	Expected	RR	p-value
1	Hot	Dhaka	2024	1255	235·93	6.14	0·001
2	Hot	Rajshahi	2024	446	46·68	10.07	0·001
3	Cold	16 southern districts*	2024	119	578·90	0.19	0·001
4	Cold	10 northern districts**	2024	26	341·90	0.073	0·001

*Southern: Lakshmipur, Noakhali, Chandpur, Shariatpur, Barisal, Feni, Bhola, Comilla, Madaripur, Munshiganj, Jhalokathi, Narayanganj, Patuakhali, Pirojpur, Gopalganj, Narsingdi.

**Northern: Dinajpur, Nilphamari, Thakurgaon, Rangpur, Lalmonirhat, Joypurhat, Panchagarh, Naogaon, Gaibandha, Kurigram.

Table 5.

Sensitivity summary of SaTScan clusters across four models.

Cluster type/Location	30/70	50/50 (base)	30/50	50/70	Consistency	RR range	LLR range
Hot: Dhaka (single district)	Yes	Yes	Yes	Yes	4/4	6·14 to 6·75	1150 to 2181
Hot: Rajshahi (single district)	Yes	Yes	Yes	Yes	4/4	8·46 to 10·07	618 to 876
Hot: Bogra (single district)	Yes	No	Yes	No	2/4	4·63 to 4·98	230 to 380
Cold: Northern border (10 to 11 districts, Rangpur Div.)	Yes	Yes	Yes	Yes	4/4	0·067 to 0·10	255 to 538
Cold: Southern/central (14-16 districts)	No	Yes	Yes	No	2/4	0·19 to 0·22	286 to 420
Cold: Eastern central (12-13 districts)	Yes	Yes	Yes	Yes	2/4	0·22	510 to 590
Cold: Central western (Pabna + 9 others)	Yes	No	No	No	1/4	0.22	291

Figure 2.

Significant clusters of RTA deaths (prospective, space-time SaTScan analysis over 2022-2024), clusters localized to only one district may not show as circles, only star will be indicative of presence of that cluster. Legend: high rate cluster (RR>1): orange; low rate cluster (RR<1): light blue; not in cluster: grey. Single district/small sized clusters are not shown and indicated with a star. For these districts, use color bar to determine the RR level. Note: Clusters (orange/blue) represent spatio-temporal areas with significantly different mortality compared to national average (SaTScan).

Figure 3.

Space-time clusters under different parameters (prospective, space-time SaTScan analysis over 2022-2024), cluster localized to only one district may not show up, see text for further clarification. Legend: high rate cluster (RR>1): orange; low rate cluster (RR<1): light blue; not in cluster: grey. Single district/small sized clusters are not shown.

The local Gi* hotspot in Chapainawabganj, despite its extremely low national RR, illustrates how localized relative elevations can occur within broader safe regions. This does not indicate high absolute burden but highlights the value of multi-scale analysis.

The concentration of national burden

Aggregating data over the entire three-year period starkly illustrates the geographical concentration of the RTA mortality burden. A handful of districts were responsible for a disproportionate share of national deaths. Dhaka, Rajshahi, and Bogra alone accounted for nearly two-thirds of the national RTA death burden, with each district exhibiting a sustained relative risk exceeding 4.0 over the study period. Conversely, several districts demonstrated remarkably low mortality. For example, Jhalokathi recorded zero deaths, while Chapainawabganj and Gaibandha had observed deaths over 97% lower than expected. This distribution highlights the profound geographic inequality in road traffic mortality across the country. The top high-risk and safest districts based on aggregate relative risk are listed in Table 6.

Table 6.

Top 5 high-risk and 3 safest districts (2022–2024 Aggregate).

Rank	District	RR	Interpretation
High risk 1^st	Dhaka	7·25	Highest risk nationally; residents experienced over 7 times the expected mortality burden.
High risk 2^nd	Rajshahi	6·63	Extremely elevated risk; more than sixfold excess compared to expectation.
High risk 3^rd	Bogra	4·36	Persistently high-risk northern district with over fourfold elevation.
High risk 4^th	Khulna	3·04	High metropolitan/industrial mortality pattern, roughly three times expectation.
High risk 5^th	Sylhet	2·76	Significantly elevated pattern indicating nearly three times the expected level.
Safest 1	Chapainawabganj	0·011	Among the lowest in the country; deaths were 99% lower than expected.
Safest 2	Jhalokathi	0	Reported no observed deaths, yielding the lowest possible risk.
Safest 3	Gaibandha	0·024	Mortality was 97–98% lower than expected.

Note. 1. Chapainawabganj’s appearance as both a ‘safest district’ (by aggregate RR) and a spatial hotspot (by Gi*) reflects the distinct analytical perspectives. The aggregate RR captures absolute risk relative to the national average over three years, while Gi* captures localized risk relative to immediate neighbors in a given year. These findings are complementary, not contradictory. 2. Relative risks shown are derived from the aggregate SaTScan analysis (2022-2024) and reflect observed-to-expected ratios. District-level p-values are not produced by SaTScan; cluster-level significance is reported in Tables 4 and 5.

Sensitivity analysis

Full sensitivity analysis across four space-time windows confirmed core findings. Dhaka hotspot appeared in 4/4 models (RR 6·14–6·75). Rajshahi hotspot appeared in 4/4 models (RR 8·46–10·07). Bogra hotspot appeared in 2/4 models (RR 4·63–4·98; only in 30% spatial settings). Northern cold cluster (Rangpur Division) appeared in 4/4 models (RR 0·067–0·092). Southern and eastern coldspots were regionally robust (4/4) but parameter-sensitive in composition (southern in 50% temporal windows, eastern central in 70% temporal). The Pabna-western cold cluster was highly sensitive (1/4), appearing only in the 30% spatial/70% temporal model.

Discussion

The spatial and spatiotemporal analysis of road traffic accident (RTA) mortality in Bangladesh from 2022 to 2024 reveals a landscape of risk that is profoundly localised and geographically unequal. A pivotal finding of our study is the critical distinction between raw death counts and population-adjusted mortality rates. The random distribution of raw counts suggests fatalities are, to a degree, a function of population density. However, the significant negative spatial autocorrelation in the adjusted rates reveals a dispersed, “checkerboard” pattern where high-risk districts are consistently bordered by low-risk neighbours.^21,22 This pattern is a powerful indicator that localised determinants, such as specific road infrastructure, the intensity of traffic law enforcement, vehicle mix, and access to timely emergency care, are likely important contributors to mortality risk, overshadowing broader regional influences. While the Moran’s I values (-0.100 to -0.143) indicate modest but statistically significant negative spatial autocorrelation, this dispersed ‘checkerboard’ pattern suggests that local determinants play a measurable role in shaping mortality risk, though regional factors are not entirely absent.

Consistent with the operational definitions in the Methods section, the term ‘hotspot’ hereafter refers to local spatial associations identified by Getis-Ord Gi, while ‘cluster’ refers to spatio-temporal patterns identified by SaTScan.

Our spatial analysis (2022–2024) identifies a single, increasingly dominant hotspot: a narrow western belt comprising Naogaon, Meherpur, and Satkhira. By 2024 this belt recorded the highest Gi* values under both raw counts and population-adjusted rates. The same area already showed elevated per-capita risk in earlier years, but the dramatic rise in absolute numbers confirms a genuine and intensifying concentration of crashes and deaths. Greater Dhaka divisions (Gazipur, Manikganj, Munshiganj, Narayanganj, Tangail) consistently registered very high raw counts due to population size, yet were unexceptional after rate adjustment. No robust high-rate clusters appeared in Dhaka or Rajshahi cities, and no statistically significant coldspots emerged in any year (low-risk areas form broader regional patterns, visible only in complementary SaTScan results). The western belt, now dominant by both metrics, is Bangladesh’s clearest road safety priority. The vehicle types prevalent in this region, particularly unregulated informal vehicles and motorcycles which are responsible for a majority of RTA deaths in Bangladesh, may contribute to elevated risk.²³ Furthermore, the year 2024 saw the alarming emergence of intense, high-rate clusters in the major urban centres of Dhaka and Rajshahi, where observed deaths were six to ten times higher than expected. This suggests an emerging urban safety challenge associated with extreme population density, traffic congestion, and potentially overburdened emergency medical systems.²⁴ It also challenges the assumption that congestion inherently reduces severe casualties, suggesting that dense urban environments in Bangladesh generate their own unique and lethal risk profiles.²⁵ The discrepancy between hotspots according to the raw count and the rate-adjusted counts is a key finding, revealing that high-fatality urban areas (like Greater Dhaka) require volume management, while high-risk regional corridors (like the Northwest) need systemic safety fixes.

Equally important to the public health narrative are the large, stable clusters of significantly low mortality we identified, particularly in the northern districts. These “colds clusters” represent a natural experiment in effective road safety. The remarkably low and consistent risk in these areas, which form the most significant protective cluster in the country, demands rigorous investigation. Understanding the underlying protective factors, whether they relate to superior road geometry, effective regional governance, stringent local enforcement, or a safer road-user culture, is essential. The lessons from these safe havens can provide a blueprint for interventions nationwide.²⁶

Another notable pattern is that while some northwestern districts appeared as hotspots in the spatial analysis (e.g. Rajshahi, Bogra, Naogaon, Meherpur, Satkhira) and were supported by SaTScan hot clusters (Rajshahi), the adjoining northern districts formed the country’s most prominent SaTScan cold cluster (RR < 0.10), a pattern not detected in the hotspot analysis. This contrast suggests a sharp regional divide between high-risk and low-risk areas and may be attributed to location of busy highways through relatively underdeveloped rural area with limited number of accessible health facilities adept at trauma management.

We could read one more meaning into this. The hotspots that were present only for a year and temporarily not sustainable and therefore do not appear in clusters, probably happen due to factors that get mitigated within that year or so. This could be a manifestation of temporary appearance of a risk factor, such as, maintenance or appearance of some road condition that was being neglected throughout that year e.g. patches of road not repaired. However, inclusion in a cluster which is temporally longer might signal the presence of some risk factors that are greater than some ‘rough patches of road’ or the necessity of systems upgradation as a whole. We can further deduce that hotspots according to the raw death counts identify where the most people die, primarily driven by population density and traffic volume in urban centers. In contrast, hotspots according to population-adjusted rates reveal where an individual’s risk is highest, uncovering systemic safety failures in regional corridors with infrastructure related issues, such as faulty roads and poor trauma care. Again, the high-rate clusters identify even broader and durable systemic faults that needed greater attention and adequate action.

The stark geographical disparities we document are symptomatic of wider sociotechnical system deficiencies common in low- and middle-income countries. Challenges such as institutional fragmentation, competing financial priorities, and political interference often undermine coordinated road safety efforts.^8,27 This can lead to improvised and suboptimal interventions that fail to address the root causes of high risk. The mortality in identified hotspots is further amplified by critical limitations in the emergency response chain. Road traffic injuries are the leading cause of traumatic brain injury in Bangladesh, and such injuries are associated with a threefold increase in mortality odds, a risk dramatically worsened by delays in receiving definitive care.³ Therefore, a high-risk cluster is not merely a marker of where crashes occur frequently, but also a stark indicator of where the entire system, from crash scene to hospital, is failing to save lives.

The methodological choices in black spot identification have important implications for study findings. As demonstrated in comparative studies, the performance of different identification methods varies by road context; spatial clustering techniques are better suited for low-speed urban environments where accidents aggregate, while sliding window approaches may be more appropriate for high-speed rural corridors where crash patterns are more dispersed.⁹ The present study’s use of multiple complementary methods, including Moran’s I for global patterns, Getis-Ord Gi* for local hotspots, and SaTScan for spatio temporal clusters, aligns with recommendations for combining statistical and spatial approaches to achieve more robust and holistic risk assessments.¹⁰ Furthermore, the systems thinking framework that underpins this analysis, which considers interactions between road users, vehicles, infrastructure, and environment, is increasingly recognized as essential for understanding the multifactorial nature of black spot formation.¹¹

Population data for 2023 and 2024 were projected from the 2022 census using a uniform 1.12% annual national growth rate. While this assumes homogeneous demographic change across districts—a limitation given that urban districts like Dhaka and Gazipur likely experience higher growth—this approach was necessary due to the absence of district-level intercensal population estimates from the Bangladesh Bureau of Statistics. Importantly, this uniform projection is conservative; if urban districts experienced higher growth than projected, our rate-based risk estimates for those areas would be underestimated, meaning the true risk in high-urban clusters (e.g., Dhaka) may be even greater than reported. Conversely, rural hotspot districts with slower growth would have rates slightly overestimated. This potential bias does not undermine the core findings of geographic disparity but should be considered when interpreting absolute rate values.

Several additional limitations should be acknowledged. First, the ecological study design precludes causal inference; observed associations between geographic patterns and potential risk factors are hypothesis-generating and require confirmation through individual-level or mechanistic studies. Second, confidence intervals for relative risk estimates from SaTScan are not reported, as the software does not routinely produce them for cluster-level output; instead, statistical significance was assessed using p-values from Monte Carlo simulations (999 replications). Third, age- and sex-stratified mortality data were not available in the DGHS dashboard releases used for this study, precluding analysis of demographic subgroups. Fourth, the analysis is limited to three years (2022-2024), which captures recent patterns but may not reflect longer-term trends.

Conclusion and recommendations

In conclusion, the RTA mortality crisis in Bangladesh is not a homogenous national issue but a collection of distinct local epidemics. Our findings demonstrate a dramatic geographical inequality, with critical emerging clusters in major urban centers and a persistent high-risk corridor in the northwest, existing in stark contrast to large regions of sustained safety. This evidence necessitates a decisive departure from blanket national policies towards a precision public health approach, where resources and interventions are meticulously tailored to the specific vulnerabilities of each high-risk district.

Based on our geospatial evidence, we propose a three-pronged strategic response:

First, public health and road safety resources must be urgently directed to the identified high-risk zones. For the emerging urban clusters of Dhaka and Rajshahi, this requires immediate enhancements to pre-hospital trauma care systems, targeted traffic law enforcement focused on high-risk vehicle types, and comprehensive safety audits of high-incidence road corridors. For the persistent northwestern spots, including districts like Naogaon, Natore, and Bogra, interventions should prioritize infrastructure upgrades on major highways and a significant strengthening of trauma care capabilities at the district hospital level.

Second, the consistent safety observed in the large “cold” clusters, particularly in the northern districts, must be treated as a vital learning opportunity. We recommend launching a systematic, qualitative investigation into these regions to identify the protective factors, whether related to road design, enforcement culture, or community-based initiatives, that are keeping mortality low. The successful practices identified in these natural experiments are a national public health asset and should be studied and actively replicated in high-risk areas to save lives on a national scale.

Third, a development of nationwide distributed emergency response teams and designated chains of referral systems are now long overdue and warrant immediate action.

By adopting this geographically informed, targeted strategy, Bangladesh can transform its response to road traffic injuries, ensuring that policy and action are as precise and effective as possible in mitigating this enduring public health crisis. The methodological framework employed in this study—combining global spatial autocorrelation (Moran’s I), local hotspot detection (Getis-Ord Gi), and prospective spatio-temporal cluster analysis (SaTScan)—is replicable in other low- and middle-income countries where district-level mortality data and geographic boundaries are available. The approach requires only aggregated case counts, population denominators, and shapefiles, making it feasible even in settings with limited data infrastructure. Other countries facing similar road safety challenges can adopt this framework to identify their own high-priority geographic areas for targeted interventions, moving beyond national averages to precision public health approaches. Despite the limitations inherent to ecological study design and data availability, these findings provide actionable evidence for prioritizing road safety investments.

Supplemental material

Supplemental material - Spatial and spatio-temporal patterns of road traffic accident mortality in Bangladesh, 2022–2024: A district-level geospatial analysis

Supplemental material for Spatial and spatio-temporal patterns of road traffic accident mortality in Bangladesh, 2022–2024: A district-level geospatial analysis by Pratyay Hasan, Tazdin Delwar Khan, Ishteaque Alam, Mohammad Emdadul Haque, and Minhajul Abedin in Journal of Public Health Research.

Footnotes

Acknowledgements

This study was conducted using exclusively publicly available, aggregated data from official government portals. The authors gratefully acknowledge the open data policies of the Directorate General of Health Services (DGHS) and the Bangladesh Bureau of Statistics (BBS), which made this research possible. We also acknowledge the United Nations Office for the Coordination of Humanitarian Affairs (OCHA) for providing the Bangladesh Subnational Administrative Boundaries dataset through the Humanitarian Data Exchange (HDX) platform, released under the Creative Commons Attribution 3.0 Intergovernmental Organization (CC BY 3.0 IGO) license. We acknowledge the OpenStreetMap contributors () for providing the base map data used in the visualizations for this study, which is available under the Open Database License (ODbL). Additionally, we acknowledge the developers of the open-source software (Python, GeoPandas, and SaTScan) that enabled this analysis. The authors are solely responsible for the analyses, interpretations and conclusions presented in this manuscript.

ORCID iDs

Pratyay Hasan

Tazdin Delwar Khan

Mohammad Emdadul Haque

Author contributions

Pratyay Hasan: Conceptualization, Methodology, Software, Validation, Formal Analysis, Data Curation, Writing – Original Draft, Writing – Review & Editing, Visualization, Supervision, Project Administration. Tazdin Delwar Khan: Conceptualization, Methodology, Validation, Formal Analysis, Data Curation, Writing – Original Draft, Writing – Review & Editing, Visualization, Supervision. Ishteaque Alam: Methodology, Software, Validation, Formal Analysis, Data Curation, Writing – Review & Editing, Visualization. Mohammad Emdadul Haque: Conceptualization, Writing – Review & Editing, Supervision, Project Administration. Minhajul Abedin: Methodology, Software, Validation, Formal Analysis, Data Curation, Writing – Review & Editing, Visualization. Pratyay Hasan, Tazdin Delwar Khan, Ishteaque Alam and Mohammad Emdadul Haque directly accessed and verified the underlying data reported in the manuscript. The decision to submit the manuscript was made by all authors.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. Dr. Pratyay Hasan (MBBS, MPH) serves as an Editorial Review Board Member for the Journal of Public Health Research (SAGE). This role was declared to the journal prior to submission.

Data Availability Statement

Will individual participant data be available? No. This study analyzed aggregated, public surveillance data at the district level. No individual participant data was collected or used.*

What data will be available? The analysis-ready, compiled dataset used for this study will be shared. This file contains the merged district-level data on case counts (fatalities), population denominators, and geographic identifiers.

What other documents will be available? The methodology is fully described within the manuscript. No separate study protocol exists.

When will data be available? Immediately upon publication.

Where will data be available? The compiled dataset will be provided as a Supplementary Appendix to the published article.

By what access criteria will data be shared? Freely available under the same Creative Commons license as the article.

The original public data sources are:

• Mortality data: Directorate General of Health Services (DGHS) Medical Certification of Cause of Death (MCCD) Dashboard¹¹

(https://dashboard.dghs.gov.bd/pages/dashboard_mccod_test.php).

• Population data: Bangladesh Bureau of Statistics (BBS) Population and Housing Census 2022 Preliminary Report.¹²

• Geospatial boundary data: United Nations OCHA Humanitarian Data Exchange (HDX)¹³ (https://data.humdata.org/dataset/cod-ab-bgd).

By what access criteria will data be shared?

Freely available. The compiled dataset will be provided under a Creative Commons CC-BY license.

Transparency declaration

The lead author (Pratyay Hasan) affirms that this manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and, if relevant, registered) have been explained. We (the authors) confirm that we have adhered to relevant EQUATOR guidelines, and the reporting method is referenced in the abstract and methods section of our paper. All authors have read and approved the final version of the manuscript. Pratyay Hasan had full access to all of the data in this study and takes complete responsibility for the integrity of the data and the accuracy of the data analysis.

Patient and public involvement statement

Patients and/or the public were not involved in the design, conduct, analysis, or dissemination of this research due to its ecological nature, relying solely on aggregated, publicly available surveillance data from the Directorate General of Health Services (DGHS) and Bangladesh Bureau of Statistics (BBS). No individual patient data or direct human subjects were accessed, making direct PPI inappropriate. Future studies on dengue interventions will prioritize PPI to ensure community relevance. The corresponding author takes responsibility for ethical considerations.

Supplemental material

Supplemental material for this article is available online.

Appendix

References

World Health Organization . Global Status Report on Road Safety 2023. World Health Organization, 2023. https://www.who.int/teams/social-determinants-of-health/safety-and-mobility/global-status-report-on-road-safety-2023

Wan

. Global and regional assessments of the burden of transport injuries and associated risk factors, 1990–2021: Insights from the 2021 Global Burden of Disease study. Medicine (Baltimore) 2025; 104(16): e42157. https://doi.org/10.1097/MD.0000000000042157

Rahman

Das

Kader

, et al. Epidemiology, outcomes, and risk factors of traumatic brain injury in Bangladesh: a prospective cohort study with a focus on road traffic injury-related vulnerability. Front Public Health 2025; 13: 1514011. https://doi.org/10.3389/fpubh.2025.1514011

Accident Research Institute (ARI) . ARI Database: Accident Statistics and Characteristics Document. Bangladesh University of Engineering and Technology (BUET), 2014, pp. 1–6.

Islam

Dinar

. Evaluation and Spatial Analysis of Road Accidents in Bangladesh: an Emerging and Alarming Issue. Transportation in Developing Economies 2021; 7(10): 10. https://doi.org/10.1007/s40890-021-00118-3

Salmon

Read

GJM

Stevens

. Who is in control of road safety? A STAMP control structure analysis of the road transport system in Queensland, Australia. Accid Anal Prev 2016; 96: 140–151. https://doi.org/10.1016/j.aap.2016.05.025

Flahaut

. Impact of infrastructure and local environment on road unsafety: Logistic modeling with spatial autocorrelation. Accident Analysis & Prevention 2004; 36(6): 1055–1066. https://doi.org/10.1016/j.aap.2003.12.003

Hamim

Debnath

Hasanat-E-Rabbi

, et al. Resolving the differences between system development and system operation using STAMP: A road safety case study in a low-income setting. Ergonomics 2021; 64(7): 839–855. https://doi.org/10.1080/00140139.2021.1876928

Ghadi

Török

. Comparison Different Black Spot Identification Methods. Transportation Research Procedia 2017; 27: 1105–1112. https://doi.org/10.1016/j.trpro.2017.12.104

10.

Karamanlis

Nikiforiadis

Botzoris

, et al. Towards Sustainable Transportation: The Role of Black Spot Analysis in Improving Road Safety. Sustainability 2023; 15(19): 14478. https://doi.org/10.3390/su151914478

11.

Kurakina

Kravchenko

Brylev

, et al. Systemic approach to auditing road traffic accident black spots. Transportation Research Procedia 2020; 50: 330–336. https://doi.org/10.1016/j.trpro.2020.10.039

12.

Getis

Ord

. The Analysis of Spatial Association by Use of Distance Statistics. Geographical Analysis 1992; 24(3): 189–206. https://doi.org/10.1111/j.1538-4632.1992.tb00261.x

13.

Kulldorff

. SaTScanTM User Guide for Version 10.1. 2022. https://www.satscan.org/techdoc.html, Accessed 10 November 2025.

14.

Kulldorff

. Prospective Time Periodic Geographical Disease Surveillance Using a Scan Statistic. J R Stat Soc Ser A Stat Soc 2001; 164(1): 61–72. https://doi.org/10.1111/1467-985X.00186

15.

Directorate General of Health Services (DGHS) . Medical Certificate Cause of Deaths Module (Real Time Health Information Dashboard). 2023. https://dashboard.dghs.gov.bd/pages/dashboard_mccod_test.php accessed 7 May 2025.

16.

Bangladesh Bureau of Statistics (BBS) . Population & Housing Census 2022 Preliminary Report. Bangladesh Bureau of Statistics, 2022.

17.

United Nations Office for the Coordination of Humanitarian Affairs (OCHA) . Bangladesh - Subnational Administrative Boundaries (ADM0-ADM4) 2022. https://data.humdata.org/dataset/cod-ab-bgd

18.

Anselin

Rey

. GeoDa: An Introduction to Spatial Data Science. 2024. Center for Spatial Data Science, University of Chicago. https://geodacenter.github.io/

19.

OpenStreetMap contributors . OpenStreetMap. 2025. https://www.openstreetmap.org

20.

Conrtibutors of Contextility Arribas-Bel

. contextily: context geo tiles in Python https://github.com/geopandas/contextily (2025). 2023.

21.

Anselin

. Spatial Econometrics. Springer, 1988.

22.

Nowosad

Meyer

. Investigating Moran’s I Properties for Spatial Machine Learning: A Preliminary Analysis. AGILE: GIScience Series 2025; 6: 1–6. https://doi.org/10.5194/agile-giss-6-40-2025

23.

Saha

Khandaker

Saha

, et al. Public Health Consequences of Road Traffic Accident Patterns in Bangladesh: A Comprehensive Analysis. Fortune Journal of Health Sciences 2025; 8(2): 462–469. https://doi.org/10.26502/fjhs.296

24.

Wang

Zou

, et al. Applying OHSA to Detect Road Accident Blackspots. Int J Environ Res Public Health 2022; 19(24): 16970. https://doi.org/10.3390/ijerph192416970

25.

Noland

Quddus

. Congestion and safety: A spatial analysis of London. Transportation Research Part A: Policy and Practice 2005; 39(7): 737–754. https://doi.org/10.1016/j.tra.2005.02.022

26.

Hossain

Faruque

. Road traffic accident scenario, pattern and forecasting in Bangladesh. Journal of Data Analysis and Information Processing 2019; 7(1): 29–45. https://doi.org/10.4236/jdaip.2019.71003

27.

Shafie

Das

Hamim

, et al. Exploring improvisations in road safety in a low-income setting. Ergonomics 2024; 67(2): 168–181. https://doi.org/10.1080/00140139.2023.2216407

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.60 MB