Matching of police and hospital road crash casualty records – a data-linkage study in Malaysia

ABSTRACT

Underreporting of road crashes hampers the development of appropriate road safety countermeasures in many countries. In this study, police and hospital records from road crash casualties in the Melaka Tengah district in Malaysia from 2014 were collected to determine their matching and reporting rates. Based on authentic personal identifiers from both types of records, Microsoft SQL was used to reveal how the matching rate varies due to multiple factors. The results showed that 311 cases (of 7625 hospital records) could be linked to both databases, yielding a 4.1% matching rate and a 4.7% police reporting rate. Both the reporting and matching rates increased with the level of injury severity. The significant underreporting in the police database showed that complementary data are necessary for enhancing the current official crash data records.

KEYWORDS:

• Road crash victims
• police register
• hospital register
• matching rate
• reporting rate

1. Introduction

Malaysia is one among the countries with the highest road traffic crash death rate in the Association of the South East Asian Nations (ASEAN) (World Health Organization [WHO], 2015). In 2015, 489,606 road crashes were registered in Malaysia, with 6706 fatalities, 4120 severe injuries and 7432 slight injuries (Royal Malaysia Police [RMP], 2017). These figures are based on police records (see Figure 1), but the real numbers are certainly much higher than the recorded ones, especially in cases involving low injury severity or minor damage (see e.g. Alsop & Langley, 2001; Janstrup, Kaplan, Hels, Lauritsen, & Prato, 2016).

Figure 1. Road crash records for year 2010–2015 in Malaysia (RMP, 2016, 2017).

National road traffic crashes are normally documented based on police reports. Information on road traffic crashes can also be retrieved from hospital databases via emergency room or hospital discharge records. Generally, hospital data provide detailed and accurate demographic and injury information, while police data provide details on crash characteristics and the vehicles involved (Rosman & Knuiman, 1994). Thus, the records complement each other, and both are necessary for obtaining a good picture of the road safety situation. However, road safety analyses in transport engineering are typically conducted based on cases contained in the police database. The dependence on police data for road safety analyses can be problematic, as not all road crash injuries are reported to the police, resulting in an underestimation of the overall burden of road crash injuries (Watson, Watson, & Vallmuur, 2015). This underreporting could potentially introduce biases relating to specific groups of road users.

A study of road traffic crashes based solely on police records may be quite misleading, resulting in the wrong setting of prevention measures (Amoros, Martin, & Laumon, 2006). It is known that not all motorists report crashes in which they are involved (Salifu & Ackaah, 2012). For instance, Abdul Manan and Várhelyi (2012) found that, in Malaysia, severe injuries were underreported by up to 6 times and slight injuries up to 14 times compared with highly developed motorized countries like Sweden, where police and hospital records are combined.

The gap between the hospital and police records can be identified by performing a data-linkage procedure, as suggested by several researchers in previous studies (see e.g. Abay, 2015; Alsop & Langley, 2001; Amoros et al., 2006; Aptel et al., 1999; Rosman & Knuiman, 1994). Data linkage is the process of linking information from two independent records that are expected to belong to the same individual. Data linkage of several databases would increase the reliability of the data. This would give an opportunity to widen the scope of data available for research and safety improvement measures.

Computer-assisted deterministic data-linkage methods are useful for matching several data sources employing probabilistic methods (Rosman, 1996), especially for larger samples (Clark, 2004). The easiest computer-assisted method is linking cases that have the same identification number, or alternatively, some other element or group of elements uniquely identifying a given person or episode (Clark, 2004). However, when this unique identifier is absent, probabilistic methods that simulate human pattern recognition may be useful when deciding whether a record from one source refers to the same person or event as a record from another source.

The challenge in data-linkage research is the lack of common unique personal attributes. Various studies matched police and hospital records based on matching characteristics (mostly date, gender and age) in the absence of individual identification numbers (see e.g. Amoros et al., 2006; Aptel et al., 1999; Rosman, 2001; Watson et al., 2015). In some countries, such as Sweden, a unique personal identifier is included in data sources, making the matching process easier. Similarly, in Denmark, a study conducted by Janstrup et al. (2016) linked police and hospital records using civil registration numbers. Rosman (1996) reported that there was a trade-off between the proportion of records linked and level of confidence in the linked records without using names or other unique identifiers. There is a challenge in matching/rematching records from various databases without knowing an exact individual identifier, as this involves a high risk of matching error. One of the challenges was highlighted by Kudryavtsev et al. (2013), who encountered frequent inconsistencies between dates of crashes in the police database and dates of injuries in the hospital database. Therefore, a minimum level of identification is required to match the hospital and police records of road crash casualties to obtain a reliable matching rate, for example anonymous civil registration number (Abay, 2015; Janstrup et al., 2016), surname (Alsop & Langley, 2001) and name (Rosman & Knuiman, 1994), especially when the matched records are highly similar, but they do not derive from the same crash.

Comparison of police and hospital databases in various countries found matching rates varying between 24% and 65% (Abay, 2015; Alsop & Langley, 2001; Amoros et al., 2006; Aptel et al., 1999; Rosman, 2001; Rosman & Knuiman, 1994; Watson et al., 2015; Wilson, Begg, & Samaranayaka, 2012). The results of the matching rates derived from those studies show an apparent underreporting of road crashes in the current database systems, in both the hospital and police records. The findings from many previous studies were subject to a certain rate of false positives, as strong personal identifiers, such as an identification number and/or full name, were absent. Such information about the discrepancies between police and hospital databases does not seem to exist in Malaysia. If underreporting of road casualties is also significant in Malaysia, then decisions on road safety interventions may be biased, resulting in misguided road safety countermeasures and the wasting of the limited resources available for road traffic crash and injury prevention. Thus, there is a need to perform a data-linkage study by using authentic personal identifiers to accurately match the two data sources without risk of false positive identification of similar crashes to determine the matching rate of recorded crashes in police and hospital databases. To our knowledge, this is the first such study conducted in an ASEAN country, or at least in Malaysia, and the results for the matching rate, reporting rate and underreporting rate using authentic personal identifiers will be a novel contribution of the study for developing countries.

2. Aim

The aim of the research was studying the road crash casualty data linkage between the hospital registers and police records in a representative region of Malaysia.

3. Method

To control the variables and possible disturbances, a limited area should be satisfactory to determine the matching rate between the two databases of hospital and police records. To ensure that the study area represents the country as a whole, the region of Melaka was selected based on the following criteria: (1) the number of registered injuries per 100,000 population is close to the average for Malaysia (see Table 1); (2) the region represents a well-delimited study area and (3) the volume of data is manageable.

Table 1. Number of injury crashes (registered by the police), per 100,000 population by state in Malaysia in 2015 (Department of Statistics Malaysia, 2017b; RMP, 2017).

Melaka is situated in the southern region of the Malay Peninsula, next to the Straits of Melaka, and covers approximately 1650 km². It is divided into three districts under separate jurisdictions, namely Alor Gajah, Jasin and Melaka Tengah. Approximately 60% of the population in Melaka is concentrated in Melaka Tengah (Department of Statistics Malaysia, 2017a); therefore, this area is an appropriate site for a data-linkage study.

There are three government hospitals located in Melaka – Alor Gajah Hospital, Jasin Hospital and Melaka Hospital (i.e. Melaka Hospital is located in the Melaka Tengah district). Melaka Tengah is surrounded by the Alor Gajah and Jasin districts, so there may be crashes occurring close to the border or in neighbouring districts where the victims are taken to a Melaka Tengah hospital for medical treatment or vice versa. Hence, the registered crash cases in the Melaka Tengah hospital database were screened to identify crashes possibly occurring in neighbouring districts. There is a ‘reference column’ in the hospital registration books indicating whether the patients were referred from other districts. The cases marked as coming from outside the Melaka Tengah district were removed. Crash cases that did not have any reference notation were assumed to have occurred in the Melaka Tengah district. The hospital registers were then compared with the police records to screen out crashes that did not belong to Melaka Tengah district, as the police records contain the location of the crashes from all districts in Melaka. Crash cases coming from neighbouring districts, as determined by the police records, were removed from the hospital registers.

Prior to hospital data acquisition, an ethical approval request was submitted. The analysed data were classified as private and confidential; therefore, they were handled with extra security measures to ensure that individual sensitivity and personal information were protected. The data were encrypted and stored only in designated storage. For this reason, only aggregated data are presented in this study. Only the main author of this paper had access to personal data.

Data were exclusively extracted from RMP database and Melaka Tengah district's hospital registers. All traffic crashes occurring in the country are registered in the RMP database, which is the only traffic crash database maintained by police in the country. Traffic crash records for 2014 were used, as this was the latest available data in both databases during the study. Hospital data were retrieved from three different departments as follows: (1) the emergency and trauma; (2) medical records and (3) forensic departments. All data were combined to gather all crash victim cases in the hospital record.

As a distinct measure in this study, unique personal identifiers were used to link crash data in the police and hospital databases, allowing elimination of false positives. These identifiers were personal identification numbers, given names and surnames. Secondary attributes (i.e. date of crash, age and gender) were also used for verification.

Data matching was performed using the Microsoft SQL (MSSQL) relational database management system software to ease the estimation of the matching rate based on authentic and/or common attributes from both records. Both deterministic and probabilistic approaches were used to obtain maximum linkages. The deterministic approach was based on exact matches of the personal identification number and/or given name and/or surname. The probabilistic approach used a combination of asterisks and/or wildcards to substitute characters in the personal identification number, given name and/or surname to ensure that alternative spellings and numbers were included. The probabilistic approach is important for identifying spelling variations or inconsistency so that false negatives can be identified.

MSSQL was used with several query algorithms with an authentic identifier and/or similar variables belonging to the same individual and case. The matching process was initiated by linking the authentic personal identification number prior to creating a masking identity, which was used to protect personal privacy information in further analysis. The process was repeated for the same dataset for given names and surnames for obtaining variations of matched sets and improve the matching rate. Several algorithm code trials were run to increase the probability of the matching percentage with different variable combinations.

A mixed matching procedure technique was used, comprising two parts – structured query join and traditional matching. Prior to the analyses, the data underwent a cleaning process to eliminate incomplete information and redundancy. Structured query join combines rows from two or more data tables based on a 100% matching of individuals’ attributes. Traditional matching represents manual data linkage for pairs with similar shared attributes other than the dataset determined by the structured query join. In this step, it is crucial to ensure that possible links are not missing or overlooked. All matched data are then manually checked to eliminate false positives.

A comparable injury classification by hospital and police force helps in easing the data-linkage process. Police classify traffic injuries into four groups based on their severity, as follows: (1) no injury; (2) slight injury; (3) severe injury and (4) fatal. The hospital classifies traffic injuries according to ICD10 codes (for documentation) and groups cases based on the seriousness and urgency of treatment as follows: (1) slight injury (green: non-critical), with the subgroups of G1–G5 (very urgent–not urgent); (2) severe injury needing priority of treatment (yellow: semi-critical) and (3) critical condition, requiring immediate treatment and intensive monitoring (red: critical). The red group has two subcategories, the first comprising patients who survive the injury and are hospitalized and the second including those who die on the way to hospital or in the ‘red’ group room. Here, the injury classification was simplified and regrouped to follow the police's injury scale: The ‘green’ group was categorized as slightly injured, ‘yellow’ and ‘red’ as severely injured and victims who were brought in dead or died in the hospital as fatal.

The formulae used to calculate the reporting rate (based on Elvik, Høye, Vaa, & Sørensen, 2009) and matching rate of hospital records and police records are shown in Equations (1(1) $Reportin\mathrm{g }Rate \left(\%\right)\mathrm{ }=\frac{\mathrm{Victims recorded by the police}}{\mathrm{Total cases in }(\mathrm{hospital records only}+\mathrm{police only}+\mathrm{Both police }\&\mathrm{ hospital})}$(1) )–(3(3) $Matchin\mathrm{g }Rat\mathrm{e }\mathrm{of} Polic\mathrm{e }Records \left(\%\right) =\frac{\mathrm{Overlaps of police and hospital records}}{\mathrm{Total cases in police records}}$(3) ).(1) $Reportin\mathrm{g }Rate \left(\%\right)\mathrm{ }=\frac{\mathrm{Victims recorded by the police}}{\mathrm{Total cases in }(\mathrm{hospital records only}+\mathrm{police only}+\mathrm{Both police }\&\mathrm{ hospital})}$(1) (2) $Matching Rate of Hospital Records \left(\%\right) =\frac{\mathrm{Overlaps of police and hospital records}}{\mathrm{Total cases in hospital records}}$(2) (3) $Matchin\mathrm{g }Rat\mathrm{e }\mathrm{of} Polic\mathrm{e }Records \left(\%\right) =\frac{\mathrm{Overlaps of police and hospital records}}{\mathrm{Total cases in police records}}$(3)

Matched and unmatched road traffic crash data in the police and hospital databases were validated by Chi-square tests regarding victim attributes, injury severity and time of crash.

4. Results and discussion

In 2014, 7625 crash victims were registered in the hospital database, while only 362 victims were recorded by the police in the Melaka Tengah district. There were 311 cases that could be linked in both databases (120 fatalities, 132 severe injuries, 58 slight injuries and 1 unknown injury; see Figure 2, Table 2). Thus, 4.1% of the hospital records were found in the police database and 86% of the police records were found in the hospital database (Table 2), which indicates an extremely low reporting rate of injury-causing crashes to the police (4.7%). The number of cases decreased dramatically with increased injury severity in the hospital database (Table 2), but it was almost the opposite in the police database. The finding that the reporting rate in police registers was proportional to injury severity confirms the results from several other studies (e.g. Abay, 2015; Rosman & Knuiman, 1994; Wilson et al., 2012).

Figure 2. Matching hospital (H)-registered and police (P)-registered victims.

Table 2. Crash victim characteristics in Melaka Tengah district for 2014.

		Police records			Hospital records
	Matched records	Total	Matching rate (%)	P-value*	Total	Matching rate (%)	P-value*
Crash victim cases	311	362	85.9	–	7625	4.1	–
Gender				<0.05			<0.05
Male	251	299	83.9	–	5541	4.5	–
Female	60	63	95.2	–	2070	2.9	–
Unknown^a	0	0	0	–	14	0	–
Age (years)							<0.05
<12	9	–	–	–	279	3.2	–
13–17	36	–	–	–	1135	3.2	–
18–20	47	–	–	–	1906	2.5	–
21–30	74	–	–	–	1766	4.2	–
31–40	34	–	–	–	811	4.2	–
41–50	24	–	–	–	551	4.4	–
51–60	20	–	–	–	508	3.9	–
61–70	33	–	–	–	279	11.8	–
>70	19	–	–	–	162	11.7	–
Unknown^a	15	–	–	–	228	6.6	–
Injury severity				<0.05			<0.05
Slight injury	58	74	78.4	–	5815	1.0	–
Severe injury	132	156	84.6	–	1418	9.3	–
Fatal	120	131	91.6	–	124	96.8	–
Unknown^a	1	1	100	–	268	0.4	–
Day of crash				0.98			0.08
Weekdays	212	253	83.8	–	5191	4.1	–
Weekend	99	109	90.8	–	2417	4.1	–
N/A^a	0	0	0	–	17	0	–
Time of day							<0.05
00:00–06:00	44	–	–	–	1300	3.4	–
06:01–12:00	73	–	–	–	1448	5.0	–
12:01–18:00	66	–	–	–	1998	3.3	–
18:01–23:59	69	–	–	–	2766	2.5	–
N/A^a	59	–	–	–	113	52.2	–

* P-value from Chi-square test for independency (i.e. assessed between matched and unmatched variables).

^a Unknown and N/A values were not included in P-value calculations.

The characteristics of crash victims recorded in the police and hospital databases in the Melaka Tengah district are shown in Table 2. There was a statistically significant difference for gender in both databases, p < 0.05. The numbers of male and female victims in the hospital database were approximately 19 and 33 times those in the police database, respectively. Males constituted the great majority of crash victims in both databases, representing the vast majority of all injury types. These results are in agreement with studies conducted by Amoros et al. (2006) and Salifu and Ackaah (2012), and they are in line with other findings showing that females are less involved in traffic crashes (Abay, 2015) and serious traffic crashes (Rosman & Knuiman, 1994; Salifu & Ackaah, 2012) than males are. There was a lower reporting rate for female victims (3%) than male victims (5.3%), possibly due to the lower injury severity level. It is known that the probability of reporting a slight injury crash is low compared with crashes of a higher severity level (see Abay, 2015; Alsop & Langley, 2001; Aptel et al., 1999; Wilson et al., 2012). However, this contradicts other studies that found a higher reporting rate for females (e.g. Janstrup et al., 2016; Watson et al., 2015).

The matching rate of the police database was highest for the highest severity level, and the difference was statistically significant, p < 0.05 (Table 2), which is in line with earlier findings (Aptel et al., 1999; Elvik & Mysen, 1999; Elvik et al., 2009; Wilson et al., 2012). The matching rate of fatalities for both genders was approximately 90% (Figure 3). Unmatched male fatalities in the hospital database were about 2%, and in the police database they were about 10%. The higher percentage of unmatched fatalities in the police database could have arisen because the deceased were sent somewhere other than Melaka Hospital. All female fatality records in the police database were present in the hospital database, while 8.3% of the fatality cases in the hospital database were not present in the police database. These discrepancies may be due to the 30-day internationally recognized road traffic crash policy used by the hospital, established for deceased who suffered from polytrauma, where crash victims who die 30 days or more after the crash are not classified as victims of a motor vehicle collision. For fatal crashes, a police investigation officer is always present at the crash scene; therefore, fatal crashes were always recorded in the police database. It can be concluded that the fatality records in both databases are reliable as a basis for designing road safety strategies and counter-measures.

Figure 3. Matched and unmatched records of the police database and the hospital database.

The matching rates for severe injuries were 10% in males and 6.5% in females, respectively (Figure 3). Moreover, 88% of severely injured males recorded in the hospital database were not registered in the police database, and 2% of male cases in the police database did not appear in the hospital database. In addition, 93% of severely injured females did not appear in the police database, while only 0.5% of severely injured females in the police database were not registered in the hospital database. These figures show alarming underreporting of traffic crashes for severe injury in the police database, which could give policymakers and authorities a wrong picture if they base their road-safety decisions solely on police data.

For slight injury, miniscule matching proportions (approximately 1%) were seen for both genders (Figure 3); 98% and 99% of slightly injured males and females, respectively, in the hospital database were not registered in the police database. This apparently shows that slight injury cases recorded by police can be questioned even more.

An important observation from this study is that, although a unique identifier was used, the matching rate between the police database and hospital database of 4.1% is extremely low compared with findings from developed countries; for example, researchers have reported matching rates of 24.3% in Denmark (Abay, 2015), 37.7% in France (Amoros et al., 2006), 46% in New Zealand (Wilson et al., 2012) and 62% in Australia (Rosman, 2001). This could be due to underreporting to or by the police in Malaysia. Action needs to be taken, as the police must ensure the citizens’ safety, including road-traffic safety, and organizations at all levels should cooperate to improve safety in the community. However, it should also be noted that different studies have unique characteristics, complicating their comparability. For example, Elvik and Mysen (1999) highlighted the challenge of comparing countries due to different reporting levels and different procedures for injury classification. In Malaysia, the corresponding law states that if more than one motor vehicle is involved in a road traffic crash, the crash should be reported to the nearest police station as soon as reasonably practicable within 24 h (Law of Malaysia, 2006). This probably contributes to underreporting of single-vehicle crashes and may partly explain the overall low crash reporting rate. The non-availability of information on the type of crashes (i.e. single- or multiple-vehicle crashes) limits our ability to draw conclusions on specific crash types. Regulatory requirements for reporting all road traffic crashes, including single-vehicle crashes and crashes involving vulnerable road users (e.g. pedestrian, cyclist), are necessary.

Regarding the distribution of the injury cases by age, Table 2 shows that the highest number of casualties was in the age group of 18–20 years. Interestingly, the matching rate for this age group was the lowest, and the difference was statistically significant, p < 0.05). Thus, young road users are less likely to report their crash involvement to the police. In fact, novice drivers in Malaysia must hold a probationary driving license for two years before receiving a full driver's license. Drivers can apply for the probationary driving license and motorcycle driving license at 18 and 16 years, respectively. Novice drivers are subject to the 10-point KEJARA demerit point system, where points are deducted according to the types of traffic offences (Road Transport Department Malaysia, 2018). The low reporting rate for the age group of 18–20 years could be due to a fear of losing demerit points on their probationary driving license, leading to their license being suspended or revoked. The number of injured victims gradually increased from the age group of younger than 12 years, peaking at 18–20 and 21–30 years, and then gradually decreasing until the group of older than 70 years. This pattern was statistically significant for both matched and unmatched data, p < 0.05. This is in line with the WHO's (2015) finding that road traffic crashes are the leading cause of death in those aged 15–29 years. Our results also show that the number of fatalities was highest in the age group of 61–70 years, which is in line with Abay (2015). The high fatality for the elderly indicates this group's vulnerability.

No statistically significant difference in the frequency of registered victims between the days of the week could be seen in either database, p > 0.05 (Table 2). These results contradict Alsop and Langley's (2001) finding that Sunday had the highest crash frequency. In our study, the hospital registered about 16% more victims/day during the weekend (1209 cases/day) than on weekdays (1039 cases/day), while the police recorded averages of 55 cases/day during the weekend and 51 cases/day during weekdays.

In terms of the time of day, the early morning (00:00–06:00 h) had the lowest frequency of casualties in the hospital database, while late evening/night (18:01–23:59 h) had the highest, and the difference was statistically significant, p < 0.05 (Table 2). The highest matching rate concerning the time of day was for 06:01–12:00 h (5.0%), while the lowest was for 18:01–23:59 h (2.5%). Nevertheless, 06:01–12:00 h only accounted for one-fifth of the total injury cases, while one-third of injury cases were registered at 18:01–23:59 h. A similar trend was found in the registered severity of injuries (slight, severe, fatal) for each quarter of the day in the hospital database. Two-thirds of injury cases were registered after midday. A possible explanation for this fact is that most private/other health centres are closed during these hours (especially at night).

Various factors (e.g. type of road user, type of crash, injury severity, time of day, crash site, cost of damages) may have contributed to the underreporting, as has been discussed extensively in previous research (e.g. Alsop & Langley, 2001; Amoros et al., 2006; Aptel et al., 1999; Dandona, Kumar, Ameer, Reddy, & Dandona, 2008; Rosman & Knuiman, 1994; Salifu & Ackaah, 2012; Watson et al., 2015). The socioeconomic characteristics of the region (Aptel et al., 1999) and Malaysian citizens probably contribute to the non-reporting, as a driver found to be at fault will be fined 300 Malaysian Ringgit (≈USD 70), and those involved in a crash often prefer to reach a mutual agreement on paying for damages, as highlighted by Salifu and Ackaah (2012). Expected insurance compensation may also influence the rate of reporting to the police (Salifu & Ackaah, 2012; Watson et al., 2015). Serious traffic crashes are normally prone to police investigation, and they may therefore result in a higher matching rate.

There were discrepancies in injury classification between the police and hospital records (Table 3). Comparing the slight and severe injury cases in the two databases showed that the police tend to underestimate the injury level. This discrepancy may be due to internal injuries in the crash victims that are difficult to assess without a thorough medical inspection. Underestimation of injury levels in police records was also found in previous studies (see Abay, 2015), although our findings contradict the findings of Aptel et al. (1999), who showed that police tends to overestimate the level of injury in France.

Table 3. Injury discrepancies of matched cases.

Injury severity	Hospital record	Police record
Slight injury	40	59
Severe injury	151	132
Fatal	120	120
Total	311	311

Our lack of access to data from private/other health centres may be seen as a limitation of our study. Nevertheless, most of the injury cases registered by private/other health centres ought to be slight injuries. If these additional injuries were considered, the reporting rate would be even lower.

Some further challenges to comparing police reports and hospital records may be that hospital records will record off-road crashes that the police would not be expected to report. This is also a major issue for cyclists and pedestrians, since these casualties are normally recorded in hospital records, but not police reports.

Our study covers a limited geographical area, namely the Melaka Tengah district in Malaysia, which was selected due to its similar number of registered injuries per 100,000 inhabitants to the average for the whole country. However, there may be some other features (e.g. socioeconomic characteristics, population composition, road infrastructures, traffic composition, road user behaviour) that could restrict the generalizability of our findings. Still, the results are significant and alarming, and they should be taken with great seriousness.

Our findings confirm the anecdotal ‘rule of thumb’ that for every fatality case recorded in the police database, there are 10 cases with severe injury and 100 with slight injury – including both recorded or unrecorded injuries. Because there are relatively few fatal crashes, decisions on road safety interventions based on such crashes may be biased, resulting in neglect of certain road-user groups’ safety problems. Such a situation occurred in Sweden, where car users comprised most road fatalities, and building middle barriers on rural roads became the central road safety intervention; this resulted in a decrease in car-user fatalities, but vulnerable road users’ fatalities and injuries remained constant (Trafikverket, 2015).

If decision makers have the wrong picture concerning road crash victims, they cannot make the right decision about where to direct road safety countermeasures, resulting in wasting of the resources available for crash injury prevention. Understanding the problem of road crash underreporting in official national databases can help policymakers and practitioners to improve the planning of preventive road safety measures and prioritize budget allocations in higher risk areas.

To overcome the underreporting issue, complementary data are necessary (e.g. self-reporting of crashes) and an integrated road traffic crash database that links police reports, hospital registers and self-reports would be beneficial. The use of an authentic identification number for the linkage process would likely enhance the reliability and accuracy of the integrated road traffic crash database.

5. Conclusions

Based on the study findings, it can be concluded that the police database, with its reporting rate of 4.7%, suffers from severe underreporting of crash victims as compared to the hospital database. The hospital database contains 86% of police cases, while the police database only contains 4.1% of hospital cases. The matching rate between police and hospital records is proportional to the level of injury severity, while the level of underreporting in the police records increases with decreasing injury severity. It can be observed that the police database tends to underestimate the injury severity of the casualties; therefore, police-recorded crash data for traffic safety analyses should be used with caution. Integration of police and hospital databases should be introduced in Malaysia as currently done in Sweden (Swedish Traffic Accident Data Acquisition) to enhance the reliability of crash data. Furthermore, the quality of data entry and means of archiving in the databases need to be improved to ease the process of data extraction and analysis. In addition, enforcement of laws and citizens’ awareness of their responsibility to report crash involvement for all types of crashes should be improved. To further understand the traffic safety situation, complementary data (e.g. self-reported accidents) are necessary to enhance the current crash databases, which could help the policymakers and practitioners in improving the planning of preventive road safety measures.

Acknowledgments

The authors would like to thank the Director of Health, Malaysia, for permission to publish this paper. None of the sponsors had any involvement in study, the design or in the collection, analysis, and interpretation of the data; neither in the writing of the report nor in the decision to submit the article for publication. Thanks to the Royal Malaysia Police; the Melaka Police Contingent Headquarters; the Ministry of Health Malaysia; the National Institute of Health, Malaysia; the Department of Health, Melaka; Hospital Melaka; Hospital Alor Gajah; and Hospital Jasin for the access to their data. Thanks to the anonymous reviewer(s) who gave useful comments. The authors would also like to thank everyone who was directly or indirectly involved in the success of this research.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was financially supported by the Ministry of Higher Education Malaysia and Universiti Teknologi MARA.