The effect of the introduction of ICD-10 on trends in mortality from injury and poisoning in England and Wales

Transcription

1 The effect of the introduction of ICD-10 on trends in mortality from injury and poisoning in Clare Griffiths and Cleo Rooney, Office for National Statistics This article examines the effect of the introduction of ICD-10 on mortality from injury and poisoning in. The article first examines the effect of the changes between ICD-9 and ICD-10 on deaths from injury, looking both at external causes, including accidental falls and land transport accidents, and secondary causes, focusing on fractures of the skull and femur. The effect of the classification change on poisoning is then examined. The main changes are highlighted and the article explains how data can be adjusted to take account of these changes so that trends over time can be analysed. The article then looks at the effect of the changes on baselines for government targets to reduce mortality from injury and poisoning. INTRODUCTION From January 2001 deaths in have been coded to the Tenth Revision of the International Classification of Diseases (ICD-10). This is the first revision to the ICD since 1979, and the most important since Information on the main changes to the classification and to the coding rules including Rule 3, and analysis of their effect on mortality data have been published by ONS. 1,2 These are summarised in Boxes 1 and 2. This article looks specifically at the effect on deaths from injury and poisoning of moving to ICD-10. We have examined both the effects on underlying cause of death (external cause) and also on secondary cause (also called nature of injury or main injury). Coroners certify virtually all deaths from external causes of mortality, apart from some due to falls and fractures of unspecified cause. The coroner s verdict is used by coders to determine the underlying cause of death. In addition, the level of detail provided by coroners on the forms submitted after inquest is crucial to the coding of cause of death for these individuals. Overall external cause mortality rates are not affected by the change to ICD-10. However, there are changes for particular causes of death. This article examines these changes in more detail and explains how data can be adjusted to allow trends over time to be examined. The effect of the change in classification on external causes of injury is examined, focusing on the effects on accidental falls and land transport accidents, including motor vehicle traffic accidents. The effect on secondary causes is explored, focusing on fractures of the femur and skull. Poisoning is also examined, looking at external and secondary causes together. These causes were chosen because they account for large numbers of injury and poisoning deaths, and in some cases because they have major changes in ICD-10 compared with ICD-9. National Statistics 10

2 Box one Box two BRIDGE CODING STUDY: METHODS To understand trends in cause-specific mortality spanning the change from ICD-9 to ICD-10, we need to measure the effect of this change on the proportion of deaths attributed to different causes of death. This is done using bridge coding, that is coding a sample of death certificates independently to both ICD-9 and ICD-10, and comparing the resulting underlying causes of death. The first step in this process is to identify equivalent codes or code groups in the two revisions which represent the same causes. In most cases this is not contentious, and the same groupings have been used by various authors and national statistics offices. The results are then presented as comparability ratios of the numbers of deaths assigned to a given disease or group of diseases in the two revisions. These are simply the ratio of the number of deaths coded to a cause in ICD-10 to the number coded to the equivalent cause in ICD-9. They measure the net effect of all changes to a particular cause of death. Confidence intervals have also been produced for these ratios; the method used to do this has been described elsewhere. 1 These comparability ratios can then be used to adjust comparisons of cause-specific mortality rates at or near the time of the changeover. They cannot be used before 1993, as ONS used a different interpretation of the rules to select the underlying cause of death from the international interpretation between 1984 and EXTERNAL CAUSES OF MORTALITY The external cause of injury codes form Chapter 20 of the ICD, and are prefixed with the letters V, W, X and Y. They are no longer a MAIN CHANGES IN ICD-10 ICD-10 represents the greatest change in the ICD in over 50 years. The main changes are: The first character of each code is now alphabetic rather than numeric this has enabled the expansion of the number of codes to provide for recently recognised conditions and more detail about common diseases. Some diseases and groups of conditions have been moved between broad groups (ICD Chapters) to another to reflect current ideas of aetiology and pathology. There have been several changes to the rules governing selection of the underlying cause of death. There are now only 5 rules instead of 9. The changes in the application of Rule 3 have the biggest impact. This rule allows a condition which is reported in either Part I or II of the death certificate to take precedence over the condition selected using the other coding rules if it is obviously a direct consequence of that condition. In ICD-10 the list of conditions affected by Rule 3 is more clearly defined than in ICD-9 and is also broader in scope. This internationally agreed interpretation is used in the automated coding software produced by the National Centre for Health Statistics (NCHS) in the US, used in, Scotland and an increasing number of other countries. The impact of this is to reduce the number of deaths assigned to conditions such as pneumonia and to increase the number of deaths assigned to chronic debilitating diseases. In, about 20 per cent of deaths mention pneumonia so the effect of this rule change is large. supplementary chapter of the ICD, as they were in ICD-9, when they were prefixed with an E. Table 1 gives comparability ratios for the external causes examined in this article. Table 1 Comparability ratios for selected external causes of mortality, 1999 ICD-10 code ICD-9 code Comparability Ratio (with 95 per cent confidence interval) Males V01 Y89 E800 E999 External causes of mortality (0.994, 1.003) V01 X59 E800 E928, excluding E870 E879 Accidents (0.989, 1.003) V01 V99 E800 E848 Transport accidents (0.991, 1.005) V01 V89 E800 E829 Land transport accidents (0.994, 1.007) W00 W19 E880 E888 Falls (and fractures, cause unspecified in ICD-9) (0.694, 0.744) W65 W74 E910 Accidental drowning and submersion (0.942, 1.049) X00 X09 E890 E899 Exposure to smoke, fire and flames (0.934, 1.003) X40 X49 E850 E869 Accidental poisoning by and exposure to noxious substances (0.987, 1.031) X40 X44 E850 E858 Accidental poisoning by drugs and medicaments (0.983, 1.020) X60 X84 E950 E959 Intentional self-harm (0.998, 1.001) X85 Y09 E960 E969 Assault (0.981, 1.041) Y10 Y34 E980 E989 Injury and poisoning of undetermined intent (1.000, 1.020) Females V01 Y89 E800 E999 External causes of mortality (0.999, 1.017) V01 X59 E800 E928, excluding E870 E879 Accidents (1.000, 1.023) V01 V99 E800 E848 Transport accidents (0.997, 1.017) V01 V89 E800 E829 Land transport accidents (0.995, 1.014) W00 W19 E880 E888 Falls (and fractures, cause unspecified in ICD-9) (0.437, 0.476) W65 W74 E910 Accidental drowning and submersion (0.904, 1.057) X00 X09 E890 E899 Exposure to smoke, fire and flames (0.929, 1.016) X40 X49 E850 E869 Accidental poisoning by and exposure to noxious substances (0.966, 1.042) X40 X44 E850 E858 Accidental poisoning by drugs and medicaments (0.964, 1.037) X60 X84 E950 E959 Intentional self-harm (0.996, 1.001) X85 Y09 E960 E969 Assault (0.972, 1.009) Y10 Y34 E980 E989 Injury and poisoning of undetermined intent (0.999, 1.029) 11 National Statistics

3 Suicide, homicide and injury/poisoning of undetermined intent (ICD-10 X60 X84, X85 Y09, Y10 Y34; ICD-9 E950 E959, E960 E969, E980 E989) Suicide, homicide and injury/poisoning of undetermined intent are based on the verdict of the coroner, so overall numbers allocated to each of these causes have not changed in ICD-10 compared with ICD- 9. The codes do vary at the third and fourth digit level. Table 2 gives comparable codes for the most common methods for suicide, homicide and open verdicts (undetermined intent). One major difference between ICD-9 and ICD-10 is that suicides and poisoning of undetermined intent involving carbon monoxide poisoning from car exhausts are no longer identifiable in ICD-10. In ICD-9, external cause codes E952.0 and E982.0 specified motor vehicle exhausts. This could be combined with main injury code 986, toxic effect of carbon monoxide. In ICD-10 codes X67 and Y17 are used for the external cause for these deaths. These are simply poisoning with other gases. The main injury code T58 specifies carbon monoxide, but not the source of it. The same is true for accidental mortality, where E868.2 specified a motor vehicle exhaust in ICD-9, but in ICD-10 the broader X47, accidental poisoning with other gases, is used instead. Codes for late effects of suicide, homicide and injury of undetermined intent are now combined in one code, Y87. They are not placed with the relevant sections of suicide, homicide or undetermined intent as they were in ICD-9. There are small numbers of deaths allocated to these codes in any one year, so their exclusion has a negligible effect on analysis of trends. This also applies to late effects of accidents. However, there are subdivisions for suicide, homicide and undetermined intent within the late effects section, so these deaths can be added to the main groups if required. This has not been done in routine tabulations published by ONS. Accidental drowning and submersion (ICD-10 W65 W74, ICD-9 E910) The number of deaths coded to accidental drowning does not change as a result of introducing ICD-10. At a more detailed level of coding (3 or 4 digit level) the focus of the classification has shifted. In ICD-10 the detail is on the type of water involved (for example bathtub, swimming pool, river, lake) and, in the case of bathtub and swimming pool incidents, the classification distinguishes between normal activity (bathing, swimming) from unintended activity (falls). In ICD-9 the emphasis was on obtaining detail of the nature of the activity being undertaken at the time of the event, for example swimming, water-skiing, diving, other sport or recreational activities, and in bathtub. Therefore, information on the activity prior to the incident is lost in ICD-10. Accidental exposure to smoke, fire and flames (ICD-10 X00 X09, ICD-9 E890 E899) Accidental exposure to smoke, fire and flames also has comparability ratios that are not significantly different from 1. There are some changes within the broad category, however. Secondary fires caused by explosions are included in the group in ICD-9 but not in ICD-10, where they are coded to W35 W40. Falls (ICD-10 W00 W19, ICD-9 E880 E888) The standard grouping for falls in ICD-10 identifies 30 per cent fewer deaths for males and over 50 per cent fewer for females when compared to the ICD-9 standard group. Table 3 shows that this is because in ICD-9 deaths coded to fracture of unspecified cause (E887) were included in the total for falls deaths (E880 E888), whereas in ICD-10 they are coded to exposure to unspecified factor (X59), along with other deaths where not enough information was provided to assign a more specific code. Some deaths due to injury do have very little information provided about them on the death certificate from the coroner or doctor. In these cases in ICD-10 the code X59 is assigned to the death. In ICD-9, the limited detail provided may have been sufficient to assign an apparently more specific code. Our analysis shows that over 1,700 female deaths were assigned to the X59 code in the bridge coding study, compared to just over 700 male deaths (Table 4). This is likely to be as a result of the fact that more of these deaths occur at older ages, and more deaths at older ages are to females. These deaths do not always have inquests, as many coroners take the view that no public interest is served by investigating them through a judicial process. For example, in, in 2000, 43 per cent of deaths from falls did not have an inquest. Most of the deaths coded to X59 in ICD-10 (1,200 for females and nearly 400 for males) were previously coded as fracture of unspecified cause (E887) in Table 2 Classification of methods: Suicide, Open verdicts, Homicide Suicide Open Homicide ICD-10 ICD-9 ICD-10 ICD-9 ICD-10 ICD-9 Drug poisoning X60 X64 E950.0 E950.5 Y10 Y14 E980.0 E980.5 X85 E962.0 Other poisoning, including alcohol, gases and vapours X65 X69 E950.6 E952.9 Y15 Y19 E980.6 E982.9 X86 X90 E962.1 E962.9 Hanging, strangulation and suffocation X70 E953 Y20 E983 X91 E963 Drowning and submersion X71 E954 Y21 E984 X92 E964 Firearms and explosives X72 X75 E955 Y22 Y25 E985 X93 X96 E965 Smoke, fire, flames and exposure to steam and hot objects X76 X77 E958.1 E958.2 Y26 Y27 E988.1 E988.2 X97 X98 E968.0, E968.3 Cutting/piercing/sharp and blunt objects X78 X79 E956 Y28 Y29 E986 X99 Y00 E966, E968.2 Falling from a high place X80 E957 Y30 E987 Y01 E968.1 Falling/lying/being pushed before moving object X81 E958.0 Y31 E988.0 Y02 No equivalent Crashing of motor vehicle X82 E958.5 Coded to relevant section Coded to Y03 if murder verdict given. If verdict is in Land Transport Accidents causing death by dangerous driving, coded to relevant section in Land Transport Accidents Other and unspecified X83 X84 E958.3 E958.4, Y33 Y34 E988.3 E988.4, Y04 Y09 E967, E968.4 means E958.6 E958.9 E988.6 E988.9 E968.9 National Statistics 12

4 Table 3 Numbers of deaths allocated to falls in ICD-9 and ICD-10, 1999 ICD-10 code ICD-9 code Falls W00 W19 Osteoporosis /pathological Unspecified accident X59 Other Total fracture M80 M81 Falls E880 E886, E888 2, ,237 Fracture of unspecified cause E , ,819 Osteoporosis /pathological fracture , ,343 Unspecified accident E Other Total 2,253 1,710 2,473 Table 4 Numbers of deaths coded to X59 in ICD-10 by their ICD-9 code, 1999 Figure 1 Age-standardised* accidental falls mortality rates, ICD-9 code Males Females Total Disease ( ) Fracture of unspecified cause (E887) 392 1,214 1,606 Unspecified accident (E928.9) Other accident Total 764 1,709 2,473 ICD-9, so this particularly affects trends in falls deaths. A further 400 were assigned to unspecified accident (E928.9) in ICD-9. Overall, there were 4.5 times as many deaths assigned to the code for unspecified injury in ICD-10 than in ICD Rate per 100, Males W00 W19, E880 E888 Females W00 W19, E880 E888 Males W00 W19, E880 E886,E888 When fractures of unspecified cause are included in the ICD-9 grouping for falls, the comparability ratios are 0.72 for males and 0.46 for females, compared to 1.02 and 1.00 when they are excluded. Figure 1 shows the trends in falls deaths from 1993 to 2001 when these deaths are both included and excluded. This clearly illustrates that excluding E887 deaths from the falls group in ICD-9 accounts for the differences between ICD-9 and ICD-10. Table 5 shows age-specific comparability ratios for falls. This shows that the ratio declines with age, indicating that older age groups are much more affected by this change than younger age groups. This is because at younger ages, fracture of unspecified cause is much less common as a cause of death, under 1,000 deaths in under 75s compared with over 11,000 in those aged 75 and over. In order to examine trends in mortality from falls, it is therefore important to exclude those deaths coded to E887 in ICD-9. However, it is often thought to be likely that many of the deaths coded to E887 were the result of a fall, but that the fall had simply not been stated on the death certificate. Around 70 per cent of deaths coded to E887 in ICD-9 had fractured neck of femur as their secondary cause. Most were at advanced ages, when these fractures can follow relatively minor force, or even occur without any appreciable force, because of osteoporosis. It is very unlikely that the external cause of injury would not be mentioned Table 5 Age-specific comparability ratios for falls (ICD-10 W00 W19, ICD-9 E880 E888), 1999 Comparability Ratio (with 95 per cent confidence interval) Under 75s and over Males (0.924, 0.985) (0.621, 0.709) (0.372, 0.469) Females (0.721, 0.815) (0.507, 0.583) (0.311, 0.360) Females W00 W19, E880 E886, E * Directly age-standardised to the European Standard Population. on the certificate if they were due to motor vehicle crashes, assaults or incidents requiring investigation by police or inquest. These deaths were grouped with falls in ICD-9, because preventing osteoporosis, falls and their consequences was considered to be the most useful public health approach to preventing these fracture deaths. The secondary cause codes for fractures, or for fractured neck of femur, could therefore be used in ICD-10 to identify these possible falls from all of the deaths coded to X59 as the underlying cause of death. Eighty per cent of deaths where the underlying cause was X59 and the secondary cause was a fracture had been coded to E887 in ICD-9, with the others mainly being coded to diseases in ICD-9. If deaths coded to X59 with a fracture as secondary cause are counted with falls in ICD-10, this leads to an overall increase of about 4 per cent in the number of deaths attributed to falls. This is due to the application of Rule 3 to some deaths previously coded as diseases in ICD-9. These are conditions, such as pneumonia or cardiac arrest, which would have appeared alone in Part I of the death certificate with an accidental injury in Part II. In ICD-9, in England and Wales, the condition in Part I is selected as the underlying cause of death if the death was certified without an inquest being held (i.e. the death did not have a verdict of accidental death). In ICD-10 pneumonia is considered to be a direct consequence of the accidental injury, so the injury is selected as the underlying cause of death. Additionally, cardiac arrest is regarded as an ill-defined condition in ICD-10, and therefore 13 National Statistics

5 preference is given to the injury in Part II. Box 3 gives some examples of this. This illustrates that changes to the classification and changes to the rules can interact, making interpretation of trends more difficult. Comparability ratios measure the net effect of all these changes together. In other countries, falls deaths have been affected by another change in coding rules in ICD-10, which allows fractures to be due to osteoporosis even if a fall was also mentioned on the death certificate. This approach was already being used in prior to the introduction of ICD-10 and so only a very small number of deaths classified as falls in ICD-9 were reclassified as due to osteoporosis in ICD-10. Box three EXAMPLES OF DEATHS WITH TERMINAL CONDITIONS AS THE UNDERLYING CAUSE IN ICD-9 AND X59 IN ICD-10 1a 1b II Bronchopneumonia Supracondylar fracture right femur, dementia Underlying cause ICD Bronchopneumonia, organism unspecified ICD-10 X59.9 Exposure to unspecified factor 1a 1b II Cardiac arrest Atrial fibrillation Fractured neck of femur (repaired) Underlying cause ICD Atrial fibrillation and flutter ICD-10 X59.9 Exposure to unspecified factor 1a 1b II Pulmonary embolism Deep vein thrombosis Fracture neck of left femur Underlying cause ICD Pulmonary embolism ICD-10 X59.9 Exposure to unspecified factor Land transport accidents (ICD-10 V01 V89, ICD-9 E800 E829) The changes to mortality from land transport accidents are caused by two separate factors. These are a change in the axis of the classification compared with ICD-9 and an increase in the level of detail required to assign a specific code to the death. These changes are described in this section. In ICD-10 the axis of classification focuses on the injured party, rather than the type of vehicle involved. The first subdivisions (the second digit) are for the victim s mode of transport, for example pedestrian, cyclist or occupant. The third digit refers to the circumstances of the accident, for example collision or non-collision. The fourth digit identifies the activity of the victim, for example driver or passenger and whether the incident occurred in a traffic (on the road) or non-traffic situation. This is different from ICD-9, where the first piece of information was the vehicle and whether or not the incident was traffic-related, so the classification referred to rail and motor accidents. Only at the fourth digit was the situation of the victim (pedestrian, car occupant, motor cyclist, or pedal cyclist) described. It is therefore possible in ICD-10 to quickly identify the number of pedestrians or cyclists injured. Consequently, it is not however possible to obtain an exact equivalent in ICD-10 to the motor vehicle traffic accidents group in ICD-9. Table 6 clearly shows the change in the axis of classification, with motor vehicle traffic accidents being spread throughout the V01 V89 codes. The majority of these deaths are to car occupants, pedestrians or motorcycle riders. Annex 1 identifies codes in ICD-9 and ICD-10 for some of the main different types of land transport accident. Using ICD-10, as described above, the situation of the victim is the main axis of classification. Figure 2 shows trends in land transport accidents involving pedestrians, pedal cyclists, motor cyclists, drivers and passengers. Comparability between ICD-9 and ICD-10 is good for pedestrians, pedal cyclists and motor cyclists, with ratios very close to or exactly one. The identification of passengers and drivers is more complex. In ICD-9 they were grouped together at the fourth digit level. In ICD-10 at the third digit level, the type of vehicle is specified, eg car, van, bus, and again at the fourth digit level passengers and drivers are identified. Table 6 Number of deaths assigned to land transport accidents in ICD-9 and ICD-10, 1999 ICD-9 code and name E800 E807 E810 E819 E820 E825 E826 E829 ICD-10 code and name Railway Motor Vehicle Motor Vehicle Other Road Other Total Accidents Traffic Accidents Non-traffic Accidents Vehicle Accidents V01 V09 Pedestrian injured in transport accident V10 V19 Pedal cyclist injured in transport accident V20 V29 Motorcycle rider injured in transport accident V30 V39 Occupant of three-wheeled motor vehicle injured in transport accident 2 2 V40 V49 Car occupant injured in transport accident 1, ,376 V50 V59 Occupant of pick-up truck or van injured in transport accident V60 V69 Occupant of heavy transport vehicle injured in transport accident V70 V79 Bus occupant injured in transport accident V80 V89 Other land transport accidents Other Total 56 2, National Statistics 14

6 In ICD-10 the number of drivers identified is reduced by about 15 per cent, and the number of passengers by 35 per cent (Table 7). This is because some of the deaths coded as passenger or driver in ICD-9 have been coded to car occupant [any] injured in unspecified traffic accident (V49.9). This is because they are vehicle accidents which do not state whether or not they involved a collision. The codes allocated in ICD-10 only allow the passenger or driver to be identified if it is stated on the death certificate whether or not the accident involved a collision. Figure 3 therefore shows adjusted rates for drivers and passengers. This shows an apparent decline in mortality in 2001 for male and female drivers is translated into an increase when the data are adjusted using comparability ratios, and a substantial apparent decline in mortality for female passengers becomes less pronounced. Another issue surrounding the level of detail required is that for a death to be coded as a motor vehicle accident, ICD-10 coding rules require the word motor or the type of vehicle, such as car or van, to be listed on the death certificate. For example, if the death certificate lists road traffic accident or head-on collision, but the type of vehicle is not stated, the Table 7 Comparability ratios for land transport accidents by victim type, 1999 Comparability Ratio (with 95 per cent confidence interval) Males Females Pedestrians (1.020, 1.069) (0.995, 1.037) Cyclists (0.958, 1.028) (1.000, 1.000) Motorcyclists (0.990, 1.023) (0.869, 1.150) Drivers (0.806, 0.860) (0.779, 0.902) Passengers (0.592, 0.722) (0.606, 0.730) cause of death is coded as other land transport accident. A reduction of 3 5 per cent is seen in road traffic accidents and 6 7 per cent in motor vehicle traffic accidents due to the allocation of these deaths to nonspecific codes (Table 8). Table 8 Comparability ratios for land transport accidents, road traffic accidents and motor vehicle traffic accidents, 1999 Comparability Ratio (with 95 per cent confidence interval) Males Females Land transport accidents (0.994, 1.007) (0.995, 1.014) Road traffic accidents (0.976, 0.992) (0.950, 0.981) Motor vehicle traffic accidents (0.947, 0.967) (0.935, 0.970) In routine publications, ONS will therefore use the land transport accidents group as a whole to present trend information. When looking at different types of these accidents, it is important to be aware of potential inconsistencies between ICD-9 and ICD-10, as described above. In, more than 95 per cent of land transport accidents are motor vehicle traffic accidents. Other data can also be examined to look at trends, for example from the Department for Transport. Christophersen and colleagues have looked at the comparability of these data with routine mortality data. 3 SECONDARY CAUSES OF MORTALITY The secondary causes of mortality, which describe the main injury incurred, are found in Chapter 19, prefixed with an S or a T. S codes classify injuries relating to a single body region. T codes classify Figure 2 Age-standardised* land transport accident mortality rates, Males 3.5 Females Rate per 100, Rate per 100, Year Year Pedestrian Pedal cyclist Driver Passenger Motorcyclist * Directly age-standardised to the European Standard Population. 15 National Statistics

7 Figure 3 Age-standardised* land transport accident mortality rates drivers and passengers adjusted and unadjusted, Males 3.5 Females Rate per 100, Rate per 100, Driver unadjusted Passenger unadjusted Driver adjusted Passenger adjusted * Directly age-standardised to the European Standard Population. injuries to multiple or unspecified body regions; poisoning; and other consequences of external causes, for example burns. These codes have been restructured so that the body region is first, and then the type of injury. This is the reverse of what was found in ICD-9 where the nature of injury was first. This means it is now therefore much more straightforward to identify all injuries to a particular body region, for example the head, but less straightforward to group similar types of injury. Table 9 shows comparability ratios for major secondary causes of mortality. These causes, together with poisoning, account for about 60 per cent of all mortality from injury and poisoning. For asphyxiation and burns, using ICD-10 makes no difference to the number of deaths coded to these categories. However, for skull fracture and intracranial injury and fracture of femur, changes are seen. These are discussed in the following two sections. Skull fracture and intracranial injury (ICD-10 S02, S06; ICD , ) For females the comparability ratio for skull fracture and intracranial injury (referred to as skull fracture) was 0.634, so 37 per cent fewer deaths are allocated to this injury type in ICD-10 compared with ICD-9. For males the ratio was This is because 75 per cent of the deaths coded to skull fracture in ICD-9, but not ICD-10, are coded to unspecified injury of the head in ICD-10, and 11 per cent to other specified injuries involving multiple body regions. This is because ICD- 10 coding rules for injuries are different from those used in ICD-9. ONS has interpreted the rules in ICD-10 to mean that if more than one injury of the same body part is mentioned on the death certificate the death is coded to unspecified injury of that body part and if more than one injury of different body regions is mentioned the death is coded to other specified injuries involving multiple body regions. Some examples of this are shown in Box 4. In ICD-9 a precedence list was used to obtain the main injury. Table 9 Comparability ratios for selected injuries, 1999 ICD-10 code ICD-9 code Comparability Ratio (with 95 per cent confidence interval) Males S02, S , Skull fracture and intracranial injury (0.516, 0.563) S Fracture of femur (1.026, 1.115) T Asphyxiation (0.994, 1.009) T20 T Burns (0.939, 1.019) Females S02, S , Skull fracture and intracranial injury (0.604, 0.666) S Fracture of femur (1.072, 1.124) T Asphyxiation (0.994, 1.032) T20 T Burns (0.942, 1.062) National Statistics 16

8 Examining multiple cause data shows that often the skull fracture is not coded at all, but the unspecified code is used instead. Table 10 shows that the number of deaths with skull fracture coded in the multiple cause codes in ICD-10 is about two thirds of the number in ICD-9. Table 11 shows the numbers of deaths which had skull fracture coded in ICD-9, by the injury codes used instead in ICD-10. This shows very clearly the replacement of the skull fracture code with an unspecified code in a large number of cases. The changes to skull fractures were significant for all age groups, with the largest effect seen in under 75s (Table 12). This is likely to be Table 10 Skull fracture and intracranial injury multiple cause data, ICD-9 and ICD-10, 1999 ICD-9 ICD-10 Coded anywhere on the death certificate 3,348 2,017 Selected as secondary cause 2,881 1,647 Ratio of multiple cause to secondary cause Percentage of multiple cause selected as secondary cause Box four Table 11 Number of deaths with skull fracture or intracranial injury coded in ICD-9 (multiple cause data) by the ICD-10 code used instead, 1999 EXAMPLES OF DEATHS WITH MULTIPLE INJURIES AS THE MAIN INJURY 1a 1b Fracture of skull Multiple injuries Motorcyclist fell from his bike when he lost control negotiating a bend Secondary cause T06.8 Other specified injuries involving multiple body regions 1a 1b Hemothorax and pneumothorax Fracture of ribs Pedestrian hit by car whilst crossing the road Secondary cause S29.7 Multiple injuries of thorax 1a Fracture skull, spine and ribs Pedestrian hit by car Secondary cause T02.8 Fractures involving other combinations of body regions Number of deaths Skull fracture or intracranial injury (S02, S06) 1,936 Unspecified head injury (S09.9) 1,239 Multiple head injuries (S09.7) 24 Unspecified multiple injuries (T07) 6 Other injury 143 Total 3,348 ICD-10 compared with ICD-9 (Table 9). However, only changes in the over 75s were statistically significant (Table 12). These changes are seen because around 300 deaths coded to the disease chapters in ICD-9 have been coded to exposure to unspecified factor (X59) in ICD-10. As this code is within the external causes chapter of the ICD, these deaths have been given a secondary cause code. In ICD-9 they would not have had a secondary cause because they were coded within the disease chapters of the ICD. About 60 per cent of these 300 deaths had fractured neck of femur as their main injury in ICD-10, implying that the death may have followed a fall. The main diseases these deaths were coded to in ICD-9 are pneumonia and other terminal conditions. This is due to the application of Rule 3 to these deaths, as described in the section on falls. because when young adults are killed in accidents it is often in one which results in them sustaining more than one injury, for example in a motor vehicle crash. Using ICD-10 means that instead of one of these injuries being selected as the main injury, the main injury is coded to one of a number of multiple injury codes. Fracture of femur (ICD-10 S72, ICD ) The comparability ratios for fracture of femur were and for males and females respectively, signalling a 9.7 per cent and 6.9 per cent increase in the number of deaths allocated to this injury type when using POISONING The ICD-10 codes for poisoning, both as external and secondary causes, are given in Table 13. There is no change to the total number of deaths allocated to these groups. However, the way specific drugs/poisons are grouped together has changed, with the most frequently abused drugs grouped together, rather than being grouped by therapeutic class as in ICD-9. So for example X42 accidental poisoning by narcotics and psychodysleptics includes heroin, methadone, other opiates, cocaine, LSD and cannabis. Amphetamines are not included in this group but are coded to X41 accidental poisoning by antiepileptic, sedative-hypnotic, antiparkinsonism and psychotropic drugs, not elsewhere classified. Table 12 Age-specific comparability ratios for skull fracture and intracranial injury and fracture of femur, 1999 Comparability Ratio (with 95 per cent confidence interval) Under 75s and over Males Skull fracture and intracranial injury (0.468, 0.521) (0.623, 0.743) (0.700, 0.888) Fracture of femur (0.973, 1.235) (1.004, 1.154) (1.001, 1.123) Females Skull fracture and intracranial injury (0.485, 0.572) (0.714, 0.828) (0.702, 0.825) Fracture of femur (0.971, 1.168) (1.072, 1.182) (1.059, 1.121) 17 National Statistics

9 Because ICD-10 has less detail than ICD-9 in the underlying cause codes for drug poisoning, it is impossible to identify particular drugs without using the secondary cause. Codes T36 to T50 have much more detail than was available in ICD-9 and this allows identification of important specific drugs, for example the main drugs of abuse, antidepressants and paracetamol. In 75 per cent of drug poisoning deaths in England and Wales, only one drug is mentioned. However, analysis using the ICD codes alone is problematic if more than one drug is mentioned on the death certificate, where in 75 per cent of cases, using data for 2001, the secondary cause is coded as T50.9 other and unspecified drugs, medicaments and biological substances. For this reason, and because the level of detail given in the ICD on drug-related poisonings is not sufficient to meet the requirement in for data on specific substances, ONS uses a specially developed database of drug-related poisoning mortality, which uses the text of death certificates to identify the specific substances involved, rather than using the underlying or secondary codes alone. These data are unaffected by the change to ICD-10. Data for 2001 have been published in Health Statistics Quarterly GOVERNMENT TARGETS ON MORTALITY FROM INJURY AND POISONING The Government has a number of targets that relate to reducing mortality from injury and poisoning. These are shown in Box 5. The public health strategy, Saving Lives: Our Healthier Nation, 5 has targets on accidental mortality and suicide, and uses routine mortality data to monitor them. Neither of these targets are affected by changes in ICD-10, and so the overall targets can be monitored without the need to adjust the baseline figures, which were produced in ICD-9, although for accidental mortality it is important to ensure that the correct equivalent codes are used, as the ICD-9 grouping was a broader one than the one used in ICD-10. However, changes in more detailed groupings do Table 13 ICD-10 code ICD-10 codes for poisoning with ICD-9 equivalents ICD-9 code External cause X40 X49 E850 E869 Accidental poisoning by and exposure to noxious substances X40 X44 E850 E858 Accidental poisoning by drugs, medicaments and biological substances X60 X69 E950 E952 Intentional self-poisoning X60 X64 E950.0 E950.5 Intentional self-poisoning by drugs, medicaments and biological substances X85 X90 E962 Assault by poisoning X85 E962.0 Assault by poisoning with drugs, medicaments and biological substances Y10 Y19 E980 E982 Poisoning of undetermined intent Y10 Y14 E980.0 E980.5 Poisoning by drugs, medicaments and biological substances, undetermined intent Secondary cause T36 T Poisoning by drugs, medicaments and biological substances T51 T Toxic effects of substances chiefly nonmedicinal as to source cause problems in monitoring and planning specific strategies to reduce accidental mortality. Obviously there is a need to know what kind of accidental deaths are happening in order to plan strategies to prevent them. Therefore, coding deaths to unspecified codes, such as exposure to unspecified factor, because of the level of detail required by the classification, is unhelpful, both to policy makers and those planning public health interventions. There is a similar issue for suicides, with the loss of specific codes identifying carbon monoxide poisoning from motor Box five GOVERNMENT TARGETS RELATING TO EXTERNAL CAUSES OF MORTALITY Routine mortality data used to monitor targets: Saving Lives: Our Healthier Nation (Department of Health) To reduce the death rates from accidents (ICD-10 V01 X59; ICD-9 E800 E928, excluding E870 E879) by at least one fifth by (Baseline: ) To reduce the death rate from suicide and undetermined injury (ICD-10 X60 X84, Y10 Y34 excluding Y33.9; ICD-9 E950 E959, E980 E989 excluding E988.8) by at least a fifth by (Baseline: ) Other sources of data used to monitor targets: Tomorrow s Roads: Safer for Everyone (Department for Transport) To reduce the number of people killed or seriously injured in road accidents by 40 per cent by (Baseline: ) To reduce the number of children killed or seriously injured by 50 per cent by (Baseline: ) Safe as Houses The Report of the Community Fire Safety Task Force (Office of the Deputy Prime Minister) Reduce accidental fire-related deaths in the home by 20 per cent by March (Baseline: Five-year average to March 1998) vehicle exhaust gases. This has been a common method of suicide among men, 6 and data are needed to monitor the effect of any interventions. However, it may be possible to identify these deaths using a combination of ICD codes and the text of death certificates. The road safety strategy, Tomorrow s Roads: Safer for Everyone, 7 has targets on reducing deaths and serious injuries on the road, both overall and in children. This target is monitored using data from the Department for Transport, based on personal injury accidents on public roads (including footways) which became known to the police. Routine mortality data can be used to provide background to this, but care must be taken, as there is a 3 7 per cent reduction in deaths allocated to road accidents, caused by the introduction of ICD-10. In response to the recommendations set out in the fire safety strategy, Safe as Houses The Report of the Community Fire Safety Task Force, 8 a target on reducing accidental deaths from fires in the home was established in Data compiled from reports submitted to the Office of the Deputy Prime Minister (ODPM) on fires attended by local authority fire brigades are used to monitor this target. These deaths are validated against death certificates supplied by ONS to ensure that they were indeed fire-related. Overall data from ONS on deaths from fires are not affected by the changes to ICD-10, so they can be used in addition to the ODPM data where they may provide different information to planners about how to prevent these deaths and whether their interventions are working. National Statistics 18

10 DISCUSSION This article has demonstrated some of the changes to injury and poisoning mortality seen following the move to coding cause of death using ICD-10 in from In, overall levels of mortality from accidents and mortality from accidental poisoning, fire and drowning are unaffected by the move to ICD-10. Suicide, homicide and undetermined intent are coded based on the verdict of the coroner and so do not change as a result of the move to ICD-10. All of these groups have changes at the detailed level of the ICD, and some of the main changes have been described above. Large changes are seen in falls and in motor vehicle traffic accidents. Nearly all of the problems in making comparisons are due to classification to ICD-10 requiring more detail than ICD-9 and more than is often provided on death certificates. This detailed information cannot always be obtained from routine data based on the administrative system of death registration which has to serve a wide variety of purposes in most countries. The coding of deaths to unspecified injury codes has therefore become more problematic with the use of ICD-10. Examples of this, described in this paper, include the coding of motor vehicle accidents, fractures of unspecified cause and skull fracture. Allocating deaths to unspecified codes gives no information on which to plan preventive measures or health services, or to measure their effectiveness. Using the secondary cause in addition to underlying cause may provide some additional useful information in the case of deaths coded to X59. For example, many of these deaths have a fracture as their main injury and therefore the most probable cause of death was a fall. Grouping these with the standard falls group in ICD-10 identifies 4 per cent more deaths from falls than are identified by using the standard falls group in ICD-9. Trends in mortality from falls can also be analysed across the change in revision by excluding deaths coded to E887 from the ICD-9 analyses, as recommended by the International Collaborative Effort on Injury Statistics approved framework for presenting injury mortality data, or applying comparability ratios. Deaths coded to X59 with fracture as the secondary cause could be added to the standard falls codes in ICD-10 and compared with the full E880-E888 range in ICD-9, however this will identify additional deaths in ICD-10 compared with ICD-9, as described above. The Mortality Reference Group of the World Health Organisation (MRG) (Box 6) proposes to introduce new codes in a future update of ICD-10, which would separate X59 into two parts X59.0 would be the equivalent of the E887 code in ICD-9, and X59.9 would be used for other deaths currently coded to X59. The analysis described here shows that combining deaths coded to X59.0 with W00 W19 to look at total falls would increase falls deaths by around 4 per cent in England and Wales. However, further bridge coding would be needed to quantify any changes. The precise detail required to classify land transport accident deaths is sometimes problematic, and the change in axis of the classification makes it more difficult to identify traffic deaths. It is therefore important to examine trends using all land transport accidents, of which over 95 per cent are motor vehicle traffic accidents, and also to be aware of potential inconsistencies at the more detailed level of classification. In addition it is possible to look at other sources of data, for example data from the Department for Transport. 9 It is also important to recognise when looking at land transport accidents that these data include deaths that occurred in rail accidents. While these happen less frequently than road accidents, they can affect particular years when large-scale accidents have occurred. This should be borne in mind when examining trends. Box six ROLE OF THE UPDATE REFERENCE COMMITTEE AND MORTALITY REFERENCE GROUP The World Health Organisation and the international network of Collaborating Centres for the Family of International Classifications have established a framework for updating, adding to and correcting ICD-10. The Update Reference Committee (URC) has been set up to oversee this process. The Mortality Reference Group (MRG) is a subcommittee of the URC, which recommends changes to ICD-10 as it is used for mortality coding and statistics. Details of the structures, a list of cumulative updates and annual reports of the URC can be found on the website of the Australian National Centre for Classification in Health at WHO%20URC/who_urc.html. Some countries have experienced more severe problems than those in. Sweden, which generally has very good mortality data, found that 60 per cent fewer deaths were classified as falls in ICD-10 than in ICD-9. Most of this was because of the shift from E887, fracture of unspecified cause, to X59, and some because of a shift to pathological fracture and osteoporosis. 10 Mexico initially found that the change to ICD-10 caused a drop of 40 per cent in deaths classified as motor vehicle traffic accidents. Again, most of these deaths were classified in ICD-10 to categories too non-specific to be of any use for public health purposes. 11 Looking at secondary causes of mortality, there are increases in the number of deaths allocated to fracture of femur, due to the reclassification of deaths previously coded as disease deaths to X59, exposure to unspecified factor, which then have a secondary code attached. Substantial declines are seen in skull fracture and intracranial injury with reductions of nearly 50 per cent. This is because of differences in coding rules between ICD-9 and ICD-10. The MRG is looking at proposals to change the way the main injury is selected in ICD-10 so as to retain useful detail and specificity. Briefly, they propose that combination codes not be used where a more specific code for a major injury is present. Clear consistent rules for selecting a single main injury where more than one potentially lethal injury is mentioned need to be developed. The impact of the change to ICD-10 on monitoring injury mortality for specific Government targets is limited. Overall accidental mortality and suicide rates are not affected by the move to ICD-10, and different sources of data are used to monitor the targets on road traffic and fire deaths. Routine mortality data can be used to provide background, but for road traffic accidents it must be borne in mind that data are not completely comparable between ICD-9 and ICD-10. Monitoring specific interventions to reduce unintentional injury is made more difficult by the level of detail required by the ICD in some situations. FURTHER INFORMATION More information on the bridge coding study and its results can be found at or in Health Statistics Quarterly volumes 8, 13 and ,1,2 Any further queries on the change to ICD-10 should be directed to icd10.mortality@ons.gov.uk. Queries on any aspect of the ICD should be sent to who@ons.gov.uk. 19 National Statistics

11 Key findings The overall number of deaths classified to external causes is unaffected by the introduction of ICD-10. Numbers of deaths classified to several groups of external causes are also unaffected (all accidents combined, accidental poisoning, accidental deaths in fires, accidental drowning, suicide, injury and poisoning of undetermined intent, and homicide). Changes in the way some types of accidental death are coded in ICD-10 have resulted in: the number of deaths coded as falls being reduced by 30 per cent for males and 50 per cent for females compared with ICD-9; around 7 per cent fewer deaths being coded to motor vehicle traffic accidents compared with ICD-9; the number of deaths in land transport accidents identified as being to drivers falling by about 15 per cent, and the number of deaths of passengers falling by 35 per cent compared with ICD-9; and about 40 per cent fewer deaths having skull fracture and intracranial injury as their main injury compared with ICD-9. Changes in the way rules for selecting cause of death are applied in ICD-10 have resulted in about 10 per cent more deaths having fractured femur as their main injury compared with ICD-9. The changes to ICD-10 have no impact on baselines for government targets to reduce deaths from external causes of mortality. However, some of the detail needed to monitor specific interventions is lost. 9. Department for Transport (2002) Transport Statistics Bulletin. Road Casualties in Great Britain Main Results 2001 (Provisional), Department for Transport: London. 10. Johansson L A (2001) Swedish Bridge Coding Study, ICD-9 ICD- 10: Design and Preliminary Results, Proceedings of the International Collaborative Effort on Automating Mortality Statistics Volume II; , National Center for Health Statistics: Hyattsville, Maryland. 11. Lozano R (2001) Motor vehicle traffic accidents in Mexico: Changes resulting from implementation of ICD-10, Mexican Centre for the Classification of Diseases. Paper no. WHO/GPE/CAS/C/01.86, presented at Annual Meeting of Heads of WHO Collaborating Centres for the Family of International Classifications, Bethesda, Maryland, Oct Rooney C and Smith S (2000) Implementation of ICD-10 for mortality data in from January Health Statistics Quarterly 08, ACKNOWLEDGEMENTS The bridge coding study was managed by Sue Smith, Lin Shane and Rosemary Coward. Without their leadership, and the skill and hard work of the cause coding and ICD-10 project teams in ONS Titchfield Office, the analyses presented in this article would not have been possible. REFERENCES 1. Rooney C, Griffiths C and Cook L (2002) The implementation of ICD-10 for cause of death coding some preliminary results from the bridge coding study. Health Statistics Quarterly 13, Office for National Statistics (2002) Report: Results of the ICD-10 bridge coding study, Health Statistics Quarterly 14, Christophersen O, Dix D and Rooney C (1999) Road traffic deaths: trends and comparison with DETR figures. Health Statistics Quarterly 03, Office for National Statistics (2003) Report: Deaths related to drug poisoning: results for, Health Statistics Quarterly 17, Department of Health (1999) White Paper, Saving Lives: Our Healthier Nation, TSO: London. 6. Kelly S and Bunting J (1998) Trends in suicide in England and Wales, Population Trends 92, Department of the Environment, Transport and the Regions (2000) Tomorrow s roads: safer for everyone. The Government s road safety strategy and casualty reduction targets for 2010, DETR: London. 8. Community Fire Safety Task Force (1997) Safe as Houses, Home Office: London. National Statistics 20

12 Annex 1 ICD CODES USED TO IDENTIFY DIFFERENT TYPES OF LAND TRANSPORT ACCIDENT ICD-9 ICD-10 Pedestrian E800 E807 (.2), E810 E819 (.7), E820 E825 (.7) V01 V09 Pedal cyclist E800 E807 (.3), E810 E819 (.6), E820 E825 (.6), E826 V10 V19 Driver E810 E819 (.0), E820 E825 (.0) V30 V38 (.0) (.5), V39 (.0) (.4), V40 V48 (.0) (.5), V49 (.0) (.4), V50 V58 (.0) (.5), V59 (.0) (.4), V60 V68 (.0) (.5), V69 (.0) (.4), V70 V78 (.0) (.5), V79 (.0) (.4), V86 (.0) (.5) Passenger E810 E819 (.1), E820 E825 (.1) V30 V38 (.1) (.6), V39 (.1) (.5), V40 V48 (.1) (.6), V49 (.1) (.5), V50 V58 (.1) (.6), V59 (.1) (.5), V60 V68 (.1) (.6), V69 (.1) (.5), V70 V78 (.1) (.6), V79 (.1) (.5), V86 (.1) (.6) Motorcyclist E810 E819 (.2) (.3), E820 E825 (.2) (.3) V20 V29 Road traffic accident E810 E819, E826 E829 V01 V06 (.1), V09 (.2) (.3), V10 V29 (.4) (.5) (.9), V19.6, V29.6, V30 V38 (.5) (.6) (.7) (.9), V39 (.4) (.5) (.6) (.9), V40 V48 (.5) (.6) (.7) (.9), V49 (.4) (.5) (.6) (.9), V50 V58 (.5) (.6) (.7) (.9), V59 (.4) (.5) (.6) (.9), V60 V68 (.5) (.6) (.7) (.9), V69 (.4) (.5) (.6) (.9), V70 V78 (.5) (.6) (.7) (.9), V79 (.4) (.5) (.6) (.9), V80 (.3) (.4) (.5), V81 V82 (.1), V83 V86 (.0) (.1) (.2) (.3), V87, V89 (.2) (.3) Motor vehicle traffic accident E810 E819 V02 V04 (.1), V09.2, V12 V14 (.4) (.5) (.9), V19 (.4) (.5) (.6), V20 V28 (.4) (.5) (.9), V29 (.4) (.5) (.6) (.9), V30 V38 (.5) (.6) (.7) (.9), V39 (.4) (.5) (.6) (.9), V40 V48 (.5) (.6) (.7) (.9), V49 (.4) (.5) (.6) (.9), V50 V58 (.5) (.6) (.7) (.9), V59 (.4) (.5) (.6) (.9), V60 V68 (.5) (.6) (.7) (.9), V69 (.4) (.5) (.6) (.9), V70 V78 (.5) (.6) (.7) (.9), V79 (.4) (.5) (.6) (.9), V80 (.3) (.4) (.5), V81 V82 (.1), V86 (.0) (.1) (.2) (.3), V87 (.0) (.8), V National Statistics