Implementation of a road traffic crash injury surveillance tool in a trauma referral hospital in Kisumu City, Western Kenya: Lessons learned from the pilot study

Abstract Objective Road traffic injury (RTI) is a major public health concern in Kenya with more than 13,000 deaths estimated annually. The primary objective of this study was to design and implement an injury surveillance tool for the collection of injury data, and assess the tool’s feasibility for the development of a hospital-based trauma registry in a leading Kenyan referral hospital. Secondarily, an epidemiologic profile was created to characterize RTI in the region. Methods An injury surveillance tool was developed and implemented, on a pilot basis, in a level 5 trauma hospital, Jaramogi Oginga Odinga Teaching and Referral Hospital (JOOTRH), in Kisumu City, Western Kenya, for a 4-week period from 2019-07-15 to 2019-08-11. A descriptive statistical analysis summarized injury frequency counts and percentages. Results Over the pilot phase, 371 patients had forms completed, but 488 official injury-related hospital registrations, indicating that 117 injury patients (24%) were missed. A process evaluation of the tool implementation revealed issues in the collection protocol that required revisions, resulting in improved data form completeness rates. For the 368 cases with cause documented, the most common mechanisms of injury were RTI (46.5%; n = 171), assaults (23.9%; n = 88), and falls (14.9%; n = 55). For RTI patients, the median age was 28 years (IQR = 16) and 77% (n = 132) were males, with motorbike collision injuries (n = 91; 53.2%) the leading RTI mechanism. There were 348 injuries for 171 patients. The most common anatomical regions for RTI were the lower limb 32.8% (n = 114), upper limb (15.2%; n = 53), followed by head lacerations 8.6% (n = 30) and concussions 7.2% (n = 25). Two-thirds of patients (n = 113; 66.1%) were discharged from ED, just over a quarter (n = 46; 26.9%) were admitted to hospital and 9 patients succumbed to RTI (5.3%). Conclusions This injury surveillance pilot study produced the first injury dataset in Kisumu City, demonstrating the significant magnitude of RTI in Western Kenya, the leading cause of injury for the region. This dataset can be replicated in other hospitals to create an injury surveillance system for the collection of trauma data, needed for the development of countermeasures for the reduction of trauma, as well as for quality initiatives to improve patient outcomes.


Introduction
Road traffic injuries (RTIs) and fatalities are a preventable and predictable public health issue, yet, RTIs are the leading killer of children and adults worldwide (World Health Organization 2018). Worldwide, 14,000 lives are lost each day due to injury (approx. 5 million), with 1.35 million lives lost solely due to road traffic deaths (RTD). Over 93% of these deaths occur in low and middle-income countries (LMIC), with Africa having the highest RTD rate at 26.6 per 100,000 population, 1.5 times the global average (World Health Organization 2018).
Kenya, located in the WHO African Region, is a LMIC with a population of 52 million (World Health Organization 2018). In 2016, Kenyan RTD rates were 27.8 deaths per 100,000 population, more than 3 times higher than highincome countries. WHO estimated that 13,463 lives were lost in RTD in Kenya in 2016(World Health Organization 2018. The high burden of RTIs underlines the need for evidence-based road safety strategies. Implementation and support of data systems for on-going monitoring and evaluation of road safety status is a key activity highlighted in WHO's Decade for Road Safety Action. In Kenya, the true magnitude and health burden of RTI is unknown. Like many other LMICs, RTD are grossly underreported and often inaccurate, resulting in a huge variation between traffic police documented road deaths and WHO estimates (World Health Organization 2018). The number of RTD derived from the country's official police statistics was 4.5 times lower than WHO estimates in 2016 (2,965 versus 13,463, respectively) (World Health Organization 2018).
The ability to develop context-specific and cost-effective RTI prevention strategies requires an understanding of the magnitude of risk factors leading to injury and death (World Health Organization 2018). LMIC need improved local reliable trauma data to accurately define the injury problem before countermeasures can be developed and implemented (Zargaran et al. 2014(Zargaran et al. , 2018. A trauma registry, data collection tool for unintentional and intentional injury composed of standardized data elements, can provide critical epidemiological and clinical information surrounding an injury. They improve the quality of care and clinical outcomes for patients; inform public officials about trauma as a public health problem; are vital for advocacy, legislative policy and regulatory efforts; and provide detailed information required for the creation of injury prevention strategies (American College of Surgeons 2014). An injury surveillance strategy to collect accurate, reliable and valid local epidemiological injury data is critical in defining the local burden of injury for the implementation of RTI prevention strategies (Botchey et al. 2017;Stevens et al. 2013).
The primary objective of this study was to design and implement, on a pilot basis, an injury surveillance tool for the collection of epidemiologic injury data and assess the tool's feasibility for the development of a hospital-based trauma registry in Kisumu County, Western Kenya. Secondarily, a RTI epidemiologic profile was undertaken.

Study setting
This study was conducted in Jaramogi Oginga Odinga Teaching and Referral Hospital (JOOTRH), a 200-bed facility in Kisumu City. JOOTRH is the primary referral hospital and has sufficient injured patients with a case-mix to be able to define RTIs in the region. Kisumu City is a major transportation hub in East Africa and is the third largest city in Kenya (Appendix C). Kisumu has three major highways entering and leaving the city with only 40% paved road coverage for the region. The outskirts of the city are surrounded by agricultural land and dirt roads leading to farmland. An injury surveillance form was implemented to allow for prospective data collection on all consecutive injury cases that sought care in the ED of JOOTRH during a 4-week pilot period (2019-07-15 to 2019-08-11).

Tool development
Data were collected on a one-page questionnaire designed to be an adaptable, inexpensive and easy-to-use injury surveillance tool, suitable for use in a resource-constrained low-income setting. Variables were based on best practice literature, and the minimal dataset (MDS) created by the WHO (Holder 2020) including: injury mechanism; patient demographics; occupation categories, [based on a pilot Uganda orthopedic registry, as a proxy to stratify socio-economic groups (Kisitu et al. 2016)]; and the Kampala Trauma Scale (KTS), an injury severity scale based on: 1) the total number of serious injuries the patient sustained; 2) age; 3) systolic blood pressure (SBP); 4) respiratory rate; and 5) neurologic status (Macleod et al. 2007). Other variables included time since injury (in hours); hospital arrival day of the week; mode of transportation; first aid provided; and location of injury. For RTIs, road user type, collision counterpart and reason for collision were collected. KTS has been specifically adapted for use in resource-constrained health care facilities in sub-Saharan Africa (Macleod et al. 2007). Scores for KTS range from 5 to 16. Lower values correspond to more severe injury. KTS < 11 ¼ severe injury, KTS ¼ 11-13 moderate injury, KTS ¼ 14-16 mild injury). Clinical injury type, treatment and outcome were also collected. There were two tool versions of this tool, one for injury survivors used as an adjunct to the hospital ED registration process, and a second adapted version for fatalities (both are identified in Appendix B).

Process evaluation and data collection
A total population sampling strategy was utilized. All injured patients who accessed healthcare at the JOOTRH ED during the pilot period were included in the study, along with patients declared dead at the injury scene or en route. Consent was obtained from all participants. Excluded from study were noninjury-related ED visits or nonconsenting patients/substitute decision makers.
Injury information was collected, utilizing the standardized data collection tool, by research team members, following a two-day training session. Sporadic data form audits were done to monitor data quality and timeliness. Adjustments were made to the process to adjust for data collection issues and to increase the completion rates. Part i of Appendix Figure 1A presents the data collection process, as designed. All data were collected from trauma patients' health records, progress reports, diagnostic reports, and the patients themselves or witnesses for injury details. For deceased patients, autopsy and police reports were also utilized for injury data. To monitor trauma patient inclusion rates, completed pilot forms were compared to injured patient daily hospital logbooks, then a follow up data collection strategy was used to include any missed patients. All data were entered into the REDCap (Research Electronic Data Capture) platform, a web-based data capturing tool that allows the secure storage and encryption of data (Harris et al. 2009). REDCap was chosen because data can be saved offline without internet connection and used globally. Instructional videos are available on the internet and can be accessed easily by partners in a multi-site project. Other injury surveillance platforms can be used such as the electronic Trauma Health Record (eTHR), that is a low-cost, downloadable, multiplatform, web-based application, for use on an iPad (Apple Computers) or other tablets.

Statistical analysis
Data quality assessments for completeness of data elements and data validity were undertaken prior to commencing with the statistical analyses, in accordance with WHO's data quality review recommendations (Holder 2020). Data with a high proportion of missing entries were not reported. While multiple imputation can be utilized for data with over 5% missing data or up to even 90% of incomplete data elements, (World Health Organization 2018) this method was not an option for incomplete data in this study due to nonrandom nature of the missing data and the lack of previous data on which to base reliable imputation.
An epidemiologic profile of descriptive data summarizing frequency counts and percentages was created. All data were screened for normality and skewed data points were presented as medians with Interquartile Range (IQR). Pearson's Chi Square test was used to analyze categorical variables in pairwise comparisons, and the non-parametric Mann Whitney U test was used to analyze continuous variables (Dawson-Saunders and Trapp 1994). A sub-analysis of RTI cases was undertaken for any patient involved in a collision. For all analyses, a p value of < 0.05 was considered statistically significant. All analyses were performed using IBM V R SPSS V R Statistics Version 26.0 (IBM Corporation, Armonk, NY).

Results
The sample size with completed forms over the pilot was 371 patients. The number of official injury-related hospital registrations was 488, indicating 117 patients were missed (24.0%). Data were missing for these individuals because they either 1) presented to hospital during evenings and weekends when hospital labor shortages meant that ED staff had insufficient time to get written consent, or 2) were critically ill and went to the ICU bypassing the ED before written consent could be obtained. Moving forward, mandating injury surveillance data be collected in the hospital record to improve patient care and outcomes would eliminate the need for consent and result in more representative injury data. All deaths required postmortem examination and were included in the study. Ninety-one percent of patient inclusion occurred from 7 am to 7 pm, when the PI was in-hospital overseeing data collection. The overall temporal pattern representative of patient volumes in the ED could not be ascertained due to a lack of accurate data collection and poor documentation at the hospital. Trauma data from other LMICs suggest the present study's data were not temporally representative because of the lack of cases occurring during weekends when trauma incidence peaks (Matheka et al. 2015).

Process evaluation of data collection tool implementation
Utilizing the data collection tool revealed process issues that required changes in the protocol, as depicted in (Appendix Figure 1A). Results of the protocol adaption included a provisional diagnosis made at the first patient encounter (Appendix Figure 1A ii: Block A) and collecting all data forms prior to patient investigations (Appendix Figure 1A ii: Block C). To ensure that the provisional diagnosis was confirmed (Appendix Figure 1A ii: Block D), these adaptations were necessary due to difficulties in data form retention with patients' departures after excessive wait-times or unexpected cost of investigations. The KTS score was calculated by the Emergency Medical Service (EMS)/research team instead of the nursing staff (Appendix Figure 1A ii: Block B) with initial triage vital signs, demographic information, and provisional diagnosis collected by the research team or EMS nurses at initial hospital arrival. This protocol increased KTS completion rates from 50.0% to 84.2% (144/171) at the end of the pilot and improved total data form completeness from 89.5% to 96.0%. Figure 1 provides a summary of the mechanism of injury. Of the 368 cases with mechanism documented, 46.5% (n ¼ 171) were RTIs; 23.9% (n ¼ 88) assaults, 14.9% (n ¼ 55) were falls. The mortality rate for injuries presenting to the ED was 3.8% (n ¼ 14).

Epidemiologic profile of RTI
Appendix Table A2 describes the 171 RTI cases collected during the study period. Males predominated (n ¼ 132; 77%). The majority of RTI cases was in the 24-64 age group (n ¼ 100; 59%). For weekday injury distribution, 79.5% of all injury-related cases occurred from Monday-Friday with only 20.5% on weekends. Pre-hospital transport information was available for 163 (95%) of RTI cases. Public transport was the most common mode of transportation (n ¼ 67; 39.2%) to reach the hospital. Public transport was mainly by motorbike taxis (boda bodas) and matatus, a common public transport minibus in Kenya. For all RTI patients arriving to the ED alive [of which 27 (15.8%) were missing], most RTI patients were in a mild injury category with a KTS 14-15, (n ¼ 119; 69.6%); 20 (11.7%) classified as moderate in the KTS range of 12-13 and 5 (2.9%) cases were severe injuries with a KTS 11. Of the 9 RTI deaths, 4 patients (2.4%) were pronounced dead on arrival and 5 (2.9%) died while receiving ED care. There were an additional 5 assault deaths for a total of 14 injury-related ED deaths. No patient died following admission.
Slightly more than half (n ¼ 91; 53.2%) of RTI that sought care were for motorbike collision injuries (Figure 2). Age-sex differences were greater for motorbike injuries than any other road user with 94.7% of all motorcyclists identifying as male. Helmet use for motorbike and non-motorized vehicles, primarily two-wheeled bicycles (n ¼ 103), was known for 84 (81.6%) of riders. Less than one-quarter (n ¼ 20; 23.8%) of these road users wore a helmet.
There was a total of 348 injuries documented for 171 patients indicating an injury to patient ratio of approximately 2:1 (Appendix Figure 3A). The most common injured anatomical region was the lower limb (n ¼ 114; 32.8%), followed by upper limb (n ¼ 53;15.2%), and head lacerations and concussions (n ¼ 30; 8.6% and n ¼ 25; 7.2%, respectively). The most common fatal injury was subdural hemorrhage (SDH), which accounted for n ¼ 7 (77.8%) of all deceased RTI patients (n ¼ 9). The remaining two deceased patients had multi-organ system injury, specifically intestinal rupture, and internal hemorrhage.

Discussion
The primary study objective was to create a descriptive hospital-intake process for injury surveillance at JOOTRH. Creating a preliminary epidemiological profile of RTI in Kisumu City County, Kenya was a secondary objective.
For any low-income injury surveillance, six key themes were identified as indicators for successful trauma registry development: 1) strong data quality; 2) simplicity; 3) integration into local resources; 4) adequate funding and trained staff; 5) the need for a local champion; and 6) the active functional use of data (O'Reilly et al. 2016). For this pilot, and the descriptive analysis of injury tool implementation, the first three themes will be addressed, while the latter three will be addressed in next steps.
The main challenge of our pilot was obtaining high completion rates and data quality of the injury surveillance form since there was difficulty integrating the new injury tool within the existing hospital processes and staff roles. Study protocol adaptations improved data quality and completion rates, but increased labor for the research team, which was unsustainable and led to lower form completion rates on nights and weekends when labor shortages peaked. This study adaptation improved the quality of data collected, but also led to an underrepresentation of injury visits since the tool was seen as an extraneous adjunct of hospital registration, and not as a vital part of an initial trauma triage process. For discharged patients, no formal documentation existed in the hospital records to allow a retrospective review.
Hospital clerks ranked simplicity and adaptability as the two strongest features of our injury surveillance form, however, there was variability in the subjective interpretation of some data variables among clerks. Examples included time waiting in triage, access to first aid, and alcohol use/toxicology, which were subsequently omitted from the form. For future injury surveillance tools, a data dictionary comprising the exact definitions and the use of objective criteria where possible (i.e., blood alcohol level) of all variables will be a prerequisite for improved simplicity and data reliability. A data dictionary with an accurate and multipronged approach to define each RTI-related ED variable is critical to ensure clear communication and consistency.
Our RTI sub-analysis yielded mixed results compared to the published literature. Because the sample size was limited, the results cannot be extrapolated to other LMIC settings; however, the observed RTI patterns mirror the trends seen in larger sample size LMIC studies (Hamadani et al. 2019;Matheka et al. 2015;Ogendi et al. 2013). The most collected variables in these LMIC injury tool surveillance studies were age, sex, injury mechanism and location (Gathecha et al. 2018). Our results showed similar findings with age, sex, occupation, and injury mechanism as the most completed variables (over 90%); while pre-hospital care and circumstances surrounding the RTI were less than 70% completed. This study mirrored others with RTIs and assaults as the leading mechanisms of injury-related trauma visits, accounting for over 50% of all ED injuries (Bachani et al. 2012;Botchey et al. 2017;Matheka et al. 2015).
The main difference in our pilot compared to published LMIC data was that our study had higher rates of motorbike injuries compared to pedestrians (Ogendi et al. 2013). Many Kenyan injury studies have shown that pedestrians are the most frequent vulnerable road user group, accounting for more than half of all RTI presenting to hospital (Matheka et al. 2015;Ogendi et al. 2013). In Kisumu City, motorbikes dominated ED visits and represented the highest frequency of collision counterparts for RTI road users at 24% (n ¼ 41). Our study's increased proportion of motorbikes, both as the impacting and impacted vehicle, could be due to: 1) Kisumu City is not on a major tourist destination likely accounting for a lower frequency of motor vehicle users; and 2) JOOTRH is in urban area of Kisumu City where motorbikes are the main mode of transport due to their availability and affordability (Bachani et al. 2012;Gathecha et al. 2018). Our cause of collision variable for motorbikes mirrored similar Kenya studies which have also demonstrated the majority of motorbike collisions were due to speeding, negligence on the road and a lack of injury prevention education (Bonnet et al. 2017). Even with this small sample size, there was a trend to low helmet use (24%) highlighting a need for public health intervention on helmet availability and use as an injury prevention target.
Our initial poor KTS completion rate of 50% out-performed other injury surveillance studies (Hamadani et al. 2019;O'Reilly et al. 2016). The KTS was chosen over alternative injury scoring systems (e.g., Injury Severity Score [ISS]) because KTS does not require automated calculation to determine a triage score simplifying the scoring process but still retaining strong utility in predicting mortality (Macleod et al. 2007). To accurately score individual injuries, ISS, requires extensive resources which do not exist at JOOTRH (Macleod et al. 2007;Weeks et al. 2016). A recent 20-year review of KTS's use has shown it to be an effective predictor of mortality and a better indicator of all other outcomes compared to most other severity scores. KTS's continued use in emerging trauma registries in LMICs is recommended (Zargaran et al. 2018).
This study had limitations. Data collection occurred only in a four-week summer period; therefore, our data may not have reflected the annual pattern of RTIs. The RTI cases in this study were people seeking hospital care and did not include individuals who may not have sought care due to financial constraints, a study drawback reported in other resource-constrained countries (Kisitu et al. 2016). Consequently, the data in this study may represent only a small proportion of RTI cases from an epidemiologic perspective. The hospital data were derived from persons coming from various areas inside Kisumu and may not be a complete picture of the urban RTI pattern. Additionally, the requirement for written consent led to an underrepresentation of injury-related visits, and trauma-related deaths. Realtime written consent required significant time and labor in an already resource-limited setting, which skewed the data toward milder injury inclusion since patients with lifethreatening injuries bypassed hospital registration. Lastly, research assistants were trained to identify and include all different types of traumas; however, due to the absence of a specific injury-related guideline for inclusion, a subjective inclusion bias toward RTIs and assaults could have been present as they were the most identified injury in the ED.
The next steps for this project and others in LMIC are based on the last three themes identified which would prioritize buy-in from frontline ED health care and clerical staff, led by a local frontline champion, in recognizing the importance of data and facilitating its collection as a necessary step for public health prevention strategies (O'Reilly et al. 2016). A successful implementation of this registry will be the active and functional use of the injury data integrated with data from police and medicolegal death investigations (Mehmood and Razzak 2009;O'Reilly et al. 2016). Attention to data utilization and its importance in targeting public health interventions ensures that a trauma registry/local injury surveillance system is sustainable (Matheka et al. 2015). However, hospital commitment is required for the long-term prioritization of integrated data collection for better patient outcomes and use of hospital resources. The sustainability of trauma registries are tied to the successful implementation of integrated Electronic Medical Record (EMR) systems such as the model used in South Africa, where the overall estimated cost of EMR implementation was 10 to 15 thousand dollars, a reasonable cost for a hospital to absorb to improve data collection (Zargaran et al. 2014(Zargaran et al. , 2018.
Following the implementation of an injury surveillance system, this study produced the first injury dataset in Kisumu. It demonstrated a realistic data snapshot of the impact of RTIs in Western Kenya, and the difficulties obtaining accurate data needed for road safety. Countermeasures for RTI cannot be designed and implemented without local data to define the problem. This dataset can be replicated in other hospitals to implement an injury surveillance system for the collection of trauma data, development of injury prevention strategies and improvement of patient health outcome. Schulich School of Medicine and Dentistry, Western University funded by Transport Canada [Contract T8056-160026/004/SS]. The Western University SROP grant, and the African Research Bursary was used for the travel expenses, accommodations and ethics approval for the lead researcher, BR, during study protocol creation in Kenya in 2018. The Dr. Barry Davidson Global Opportunities Scholarship was used for travel expenses and accommodation by the lead researcher, BR, during the study pilot implementation process in Kisumu City, 2019. Finally, the Motor Vehicle Safety Research Team funding was used to cover the administration fees for Kenyan ethics approval, project expenses such as copying and printing, and the stipend for a research assistant during study pilot implementation in Kisumu City, Kenya, 2019. Additionally, Transport Canada funding was used for conference fees and accommodation to cover the expenses for lead researcher, BR, to attend and present this research at the Canadian Global Health Conference, Ottawa, Ontario in 2019.