Article Text
PDF
Abstract
Background Limited evidence exists about associations between road crash injury and economic status in sub-Saharan Africa from large, population-based data sets. Existing studies generally do not incorporate fatal crashes. This study aims to understand the relationship between relative wealth and road crash injury and severity using population-representative cross-sectional data from Uganda’s 2016 Demographic and Health Survey .
Methods One-year road crash risk was flexibly modelled as a function of wealth using fractional polynomial models, stratified by sex and rural/urban residence. Wealth was operationalised as 1/20th quantiles of the first principal component from a polychoric principal component analysis. Injury severity was coded as a three-level ordinal variable; associations with wealth were modelled with ordinal logistic regression on quintiles of relative wealth, stratified by residence.
Results Overall, injury risk peaked in the upper middle of the wealth distribution. Rural resident injury risk increased monotonically with wealth. Urban resident risk had an upside-down U shape. Risk peaked in the distribution’s middle at about double the lowest levels. Only urban men had higher risk among the least wealthy than most wealthy (3.2% vs 1.7%; difference=1.5 percentage points, 95% CI 0.2 to 2.7). Among those with road crash injuries, greater relative wealth was associated with decreased likelihood of more severe injury (33.2 percentage points lower in the highest category than lowest, 95% CI 18.4 to 48.1) or death (5.9 percentage points, 95% CI −0.1 to 11.8) for urban residents but not rural residents.
Conclusion Relationships between relative wealth and injury risk and severity are complex and different for urban and rural Ugandans.
- public health
- descriptive epidemiology
- low-middle income country
Data availability statement
Data are available in a public, open access repository. Data are freely available from the Demographic and Health Survey programme at https://dhsprogram.com/data/. Statistical code to prepare data sets for analysis and replicate all analyses reported in this paper is provided as an online supplement.
Statistics from Altmetric.com
Introduction
Each year in East Africa, road crashes kill over 50 000 people and cause the loss of approximately 3 million disability-adjusted life years (DALYs). Unlike in higher income settings, the share of deaths and lost DALYs attributable to road crash in East Africa—and in sub-Saharan Africa in general—have increased consistently for the last 20 years. In Uganda, deaths and lost DALYs attributable to road crashes have increased by at least 20% since 2010.1 These trends are likely to continue as economic growth increases access to motorised transport but safer vehicles, roadways and driving environments take longer to develop.2 Indeed, a recent systematic review by Balikuddembe and colleagues found that Uganda’s efforts to reduce road crash morbidity and mortality, while laudable, are unlikely to achieve United Nations Decade of Action for Road Safety targets.3 While road crash injury throughout sub-Saharan Africa is a growing public health challenge, epidemiological research has not kept pace with the problem’s urgency.4
One area of ongoing controversy is the relationship between individuals’ wealth or poverty and the likelihood of road crash injury. Qualitative research from sub-Saharan Africa often identifies mechanisms by which relative poverty increases risks of being in crashes and exacerbates barriers to emergency care.5–8 However, quantitative research has led to mixed conclusions, with wealth and various factors associated with wealth—such as education, job type and urban versus rural location—being associated varyingly with higher and lower crash risks.9–17
A recent analysis of a large, population-representative Kenyan data set found that non-fatal road crash injury risk was highest in the middle of the wealth distribution, but that resulted from injury increasing with wealth in rural areas but generally decreasing with wealth in urban areas.16 However, that study did not include data on fatal crashes so it could not exclude the possibility that non-fatal crash systematically underestimated crash incidence for those who were poorest and thus faced the greatest barriers to postcrash survival.
This uses population-representative data that include fatal crash data from Uganda’s most recent national Demographic and Health Survey (DHS), with the aim of answering two questions. First, what is the association between relative wealth and road crash incidence in Uganda, overall and disaggregated by rural or urban place of residence and sex? In this, it seeks to extend the Kenyan study referenced above with a more complete data set. Second, what is the association between relative wealth and injury severity among those in road crashes, disaggregated by place of residence?
Methods
Participants, sampling and survey administration
This study uses secondary data from the 2016 Uganda DHS, which were made publicly available in December 2017 and downloaded from a public repository maintained by ICF Macro at https://www.dhsprogram.com/Data/. Data elements, sampling and the conduct of the survey have been fully described previously,18 but briefly summarised here.
The Ugandan DHS was a stratified, two-stage cluster sample, with 696 clusters chosen by selection probability proportionate to size from 34 rural/urban-by-region strata at the first stage and 30 households per cluster at the second stage. Enumerators who received 1 month of training collected data through interviewer-administered surveys between June and December 2016. Surveys consisted of the internationally standardised DHS version 7 modules and are publicly available.18 19 They were administered in English and eight local languages. In addition to the standardised modules, the Uganda DHS included an additional injury section within the household-level modules in which heads of households were asked, ‘Was anyone in your household killed in a road traffic accident in the past 12 months or injured in a road traffic accident with injuries severe enough that for at least one day they could not carry out their normal daily activities?’ If the household head reported an injury, enumerators recorded which member had been injured to allow linkage to individual-level data elements captured in other survey modules, as well as additional information about vital status, demographics and what sort of road user the injured member was (eg, pedestrian, motorcyclist, and so on).
Measures
There are two main outcomes of interest: road crash injury incidence over the 12-month period and injury severity. Injury incidence is calculated as the proportion of household members for whom a road crash injury or fatality was reported. Injury severity was categorised as a three-level ordinal variable based on the most severe injury reported. Levels were death, higher severity (report of being paralysed, brain damage, disfigurement, loss of limb, loss of limb function, loss of eyesight, burns, or broken bones) and lower severity (report of bruising, emotional trauma, cuts, chronic pain, or other). These categories were adapted from the National Safety Council’s KABCO scale.20 Because injury categories match KABCO imperfectly, a categorisation was created as a sensitivity analysis. In it, lacerations were recategorised as more severe because the DHS did not record extent of blood loss as KABCO requires.
Relative household wealth was the principal predictor variable. It was calculated in the same way as previously by Kraemer.16 Household wealth was measured as the score for first principal component produced by Kolenikov and Angeles’ approach for principal component analysis (PCA) of the standard DHS set of household assets.21 This approach substantially parallels the standard DHS approach for calculating relative household wealth,22 23 except that it treats multicategory assets (such as housing materials) as ordinal rather than a series of dichotomous variables and then uses polychoric correlations in the PCA to correctly handle ordinal variables. Kolenikov and Angeles have shown that this approach outperforms the standard DHS approach.21 Household wealth was then categorised in two ways: into 1/20th ventiles (overall and stratified by urban and rural status) for descriptive analyses and quintiled for further analysis of relationships with injury severity. More details on wealth index construction and the specific items from which it was constructed are provided in online supplementary material 1.
Supplemental material
Additional covariates included urban versus rural location of residence (dichotomised using the Ugandan DHS definition of locales with more than 2000 people being urban), sex and age (categorised as <18, which is the age at which one can obtain a driving licence in Uganda, and quintiles above age 18).
Statistical methods
The relationship between 1 year road crash injury incidence and ventiles of relative wealth was estimated by fitting fractional polynomial models, which allow for flexible investigation without a strong a priori expectation about the shape of relationships. The approach recommended by Royston et al was followed.24 25 Logistic regression models with all possible first and second-degree fractional polynomials with power transformations from −2 to 3 were fitted. An adaptation of the Bayesian information criterion for survey data was used to choose the best model.26 Modelled road crash injury probability was then calculated to convert model output from a log-odds scale and graphed for ease of interpretation. Additional information on these models is provided in online supplementary material 1.
Among those with a road crash injury, the association between degree of injury severity and wealth quintile, stratified by rural and urban location, was investigated. Standard cross-tabulation approaches were used for unadjusted analyses. Ordinal logistic regression was used for adjusted analyses, with the injury severity variable regressed on age, sex, wealth, location and the interaction of wealth and location. Because ordinal logistic regression model output is difficult to interpret (and especially so with interaction terms), results were converted to adjusted probabilities using predictive margins and graphed.
All analyses incorporated sampling weights and used Taylor linearised SEs to account for clustering. Analyses used Stata V.16.1. Code to replicate analyses is provided as online supplementary material 2.
Supplemental material
Patient and public involvement
Because of the secondary analysis nature of this study, it was not feasible to include members of the public in the design, conduct, reporting, or dissemination of this research.
Results
Characteristics of the 91 409 people whose data were collected from 19 588 households are shown in table 1. Urban respondents were wealthier, slightly older and slightly more likely to be female than rural respondents. Urban respondents also reported being slightly more likely to have been injured in a road crash badly enough that they could not carry out normal daily activities for at least a day over the last year (2.8% vs 2.1%). Of those injured in a crash, 2.3% were killed. Injury severity was similar between rural and urban areas. Of the 1999 people reported to be injured, information about severity was missing for four.
Participants’ demographic characteristics and road crash occurrences
There was no sex by residence location subpopulation for which those with the least wealth were estimated to be at the highest risk of road crash injury (figure 1). In the overall sample, injury risk peaked in the upper middle of overall household wealth distribution. In the rural subpopulation, injury risk monotonically increased with relative wealth, and injury risk was consistently higher for males than females. In the urban subpopulation, injury risk had an upside-down U shape, with risk peaking in the middle of the distribution at about double the lowest levels, which were approximately the same at the highest and lowest wealth extremes. For urban females, risk increased monotonically with wealth, but urban males’ risk increased through the middle of the distribution and then fell among the wealthiest. Urban men comprise the sole subpopulation where those at the poorest extreme of the wealth distribution had higher road crash injury risk than those at the wealthiest extreme (3.2% vs 1.7%; difference=1.5 percentage points, 95% CI 0.2 to 2.7).
One-year cumulative incidence of road crash injury as a function of relative household wealth in Uganda and stratified by urban and rural residence.
Among those with road crash injuries, greater relative wealth was associated with decreased injury severity for the urban subpopulation, but the risk of death or more severe injury remained constant in the rural subpopulation (figure 2 and online supplementary material 1). Adjusting for age and sex, urban residents in the lowest wealth quintile were 5.5 percentage points (95% CI −0.4 to 11.5) more likely to die if injured in a crash than their rural counterparts, but urban residents in the wealthiest quintile were 0.8 percentage points (95% CI −0.1 to 1.7) less likely to die (difference-in-differences 6.3 percentage points, 95% CI 0.2 to 12.3). Similarly (and again adjusting for age and sex), those from urban areas in the highest wealth quintile who were injured in a crash were 5.9 percentage points less likely to die (95% CI −0.1 to 11.8) and 33.2 percentage points less likely to endure a more serious injury (95% CI 18.4 to 48.1) than those in the lowest quintile. The poorest urban residents injured in road crashes were an adjusted 28.2 percentage points (95% CI 13.7 to 42.8) more likely to have an injury categorised as more severe than the poorest rural residents; however, by the highest wealth quintile, urban residents were 9.0 percentage points (95% CI 0.3 to 17.8) less likely to have a more severe injury (difference-in-differences 37.2 percentage points, 95% CI 19.8 to 54.7). These associations were approximately the same in unadjusted models (figure 2). Adjusting for wealth and location of residence, injury severity increased with age and was lower for females than males (figure 3 and online supplementary material 1).
Probability of injury severity among those with a road crash injury in Uganda by urban or rural residence and relative household wealth.
Probability of injury severity among those with a road crash injury in Uganda by age and by sex, adjusting for urban or rural residence and relative household wealth.
In the sensitivity analysis where severe injury was classified differently, relationships with wealth did not change meaningfully (online supplementary material 1).
Discussion
Using a large, population-representative data set, this study finds a complex relationship between relative wealth and road crash injury risk and severity in Uganda. Overall, the risk of road crash injury over the prior year increases with wealth except at the highest part of the wealth distribution, where it begins to decrease again. For no subpopulation was the poorest segment the most likely to have been injured in a road crash. For rural residents, risk consistently increases with wealth for both men and women. For urban residents risk consistently increases with wealth for women but, for men, risk increases from through the middle of the urban wealth distribution and then falls such that the wealthiest urban men have lower risk than the poorest.
These findings are broadly consistent with those of a similar study from Kenya that explored only non-fatal injuries using slightly older data from 2014.16 In that study, as in this one, urban men comprised the only subpopulation for which the wealthiest residents were the least likely to report a road crash injury. Similarly, that study found consistently increasing risk of road crash injury for rural residents. On the other hand, risk of the road crash injury—both overall and for each subpopulation—was about twice as high in Kenya as in Uganda. Further, peak road crash risk for both men and women is in higher parts of the wealth distribution in Uganda than in Kenya.
To the extent that road crash risk is partially a result of both occupation (eg, professional driving) and access to and mode of motorised transport,5 7 16 the differences between Ugandan and Kenyan risk may reflect underlying differences in wealth between the two countries. Per capita income at purchasing power parity is about twice as high in Kenya as in Uganda,27 so any given point in the household wealth distribution in Uganda would be lower than in the wealth distribution in Kenya. This points to a challenge with measuring economic position using household asset indices. While asset indices are well validated and amenable to data collection via household surveys, they produce a score that has little absolute meaning so they are mainly only interpretable in relative terms, but that makes them difficult to translate across contexts.21 22 Augmenting with measures of absolute poverty, like Afrobarometer’s Lived Poverty Index, may be fruitful.28
The second major finding of this study is that wealth is associated with injury severity among those who are injured in road crash among urban but not rural residents. This likely reflects two factors. First, the share of transport shifts from foot or bicycle to motorbike to enclosed vehicle with wealth, but a significantly smaller fraction of the rural population than urban population can afford regular transport in enclosed vehicles.29 Using the same underlying data as this study, the Ugandan government has previously reported that 20% of urban resident crashes resulting in injury were in cars, trucks, or buses, compared with 11% for rural residents.18 At the same time, the wealth disparity in access to prehospital emergency services and hospital-based trauma care may be greater in urban areas because both services are highly limited by geography and travel time for everyone in most rural areas.30 Unsurprisingly, one study from Uganda has found distance to be a principal determinant of elapsed time between crashes and trauma care,31 though recent evaluations of lay prehospital care and task shifting of trauma care to lower level health professionals have shown favourable outcomes in rural Uganda.32 33 However, even when trauma care can be reached in time, which is more likely in urban areas, previous research has documented significant financial barriers to care in Uganda.8 34 This is consistent with extensive research across sub-Saharan Africa documenting poverty as a barrier to emergency care.35–37 Large population-based surveys focused specifically on injury in this context (or enhanced injury modules in existing surveys) would help to identify more nuance about the circumstances contributing to crashes and care provided after.
This study has some limitations. First, one should be cautious about time order in cross-sectional studies. The conclusions of this study assume that households’ wealth ranking at the time of the survey approximates their ranking at the time of the recalled road crash, but injuries often cause a loss of income.38 However, because household wealth measured by asset indices (as opposed to recent income) is a relatively long-run measure of economic position, this risk should be mitigated.22 Further, if households tended to be wealthier at the time of crashes than when surveyed, then the extent to which crash risk increases with wealth would be underestimated but it would not affect the overall finding that crash risk is not highest among the poorest part of any subpopulation. The risk that households lost relative economic position after a crash is a larger threat to the finding that the likelihood of severe injury or death falls with relative wealth among urban respondents. However, there is strong theoretical reason to believe this association is real, as discussed above, and the amount of reverse causation would have to be enormous to explain the trends observed in this study.
A second limitation is that two studies from sub-Saharan Africa have found that 12-month recall periods result in underestimation of minor injuries (though not more severe injuries). In one study, recall errors were more common with greater recall periods for rural residents than urban residents,39 while the other study did not find recall to be differentially associated with any participant characteristics.40 Because the analyses reported in this study are stratified by location of residence, it is likely that the longer recall period did not bias its main findings. However, it would be useful for future population-based surveys either to shorten the period of recall (though that would require larger sample sizes to acquire information about the same number of injuries) or ask respondents about the date of injury so that sensitivity analyses could be restricted to respondents reporting injury within a shorter time period.
Overall, this study finds evidence of a complex association between relative wealth and road crash injury risk and severity in Uganda. Insofar as it is consistent with prior research suggesting that road crash risk increases with wealth in rural areas, it suggests an urgent need for rural development efforts to focus on mitigating unanticipated potential increased risk of injury while maintaining focus on reduced transport-related risks and increased access to emergency services overall.
What is already known on the subject
Qualitative and quantitative studies of the relationship between wealth and road crash injury in East Africa come to contradictory conclusions.
Limited research using large, population-based data sets has found different relationships between wealth and road crash injury in rural and urban areas. However, existing population-based surveys have not incorporated fatal crash which may confound observed relationships.
What this study adds
The relationship between relative wealth and road crash risk and severity is complex in Uganda.
For rural residents, injury risk increases with wealth, but risk of more severe injury or death among those in a crash is not associated with wealth.
For urban residents, injury risk peaks in the upper middle part of the wealth distribution, but the risk of more severe injury or death drops consistently with wealth.
Data availability statement
Data are available in a public, open access repository. Data are freely available from the Demographic and Health Survey programme at https://dhsprogram.com/data/. Statistical code to prepare data sets for analysis and replicate all analyses reported in this paper is provided as an online supplement.
Ethics statements
Ethics approval
The original collection and use of the data used in this study was approved by ICF’s Institutional Review Board. Because this study only analysed publicly available and anonymised data further review was not required by Georgetown University’s Institutional Review Board.
Acknowledgments
JK is grateful to Georgetown University for giving him sabbatical time during which he worked on this study. He is also grateful to the many, many Washington, DC coffee shops that let him use their space and kept him caffeinated while working on it. JK hopes they all remain after COVID-19.
References
Footnotes
Contributors JK designed and conducted the analyses and drafted the manuscript.
Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
Competing interests None declared.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review Not commissioned; externally peer reviewed.