How to Derive Collision Type from Reported Crash Events

We want to compare types of collisions in order to increase the probability of true links with Name or Birth Date disagreements and also decrease the probability of false links for the same person in two different crashes. Even if collision type is not available on a crash record we should be able to derive collision type from Crash Events for comparison with Standard Collision derived from trauma E Codes.

Table 1 Std. Collide vs. Crash Events

(Most Harmful Event, First Harmful Event, or Sequential Event 1)

Std.

Collide

Pairs

Pedestrian

Pedal

Cycle

Non

Collision

Ran

Off

Road

Left

Ran

Off

Road

Right

Overturn

Rollover

Any

Non

Events

Any

Objects

Only

Non

Events

Only

Objects

Most

Vehicles

First

Vehicles

Seq.

Vehicle

31

1

5

0

2

3

6

9

3

9

3

14

21

2

N/A

23

1

1

0

0

3

0

3

6

2

5

14

19

4

NO

289

1

1

27

53

103

103

194

79

135

20

65

118

11

OBJ

96

1

0

3

20

33

10

60

57

22

19

13

36

7

OTH

69

0

24

1

1

7

2

10

5

7

2

28

47

7

PED

69

46

8

0

20

3

0

23

1

23

1

10

40

1

VEH

814

5

5

9

25

52

59

139

68

101

30

544

704

135

I analyzed high probability linked pairs from a preliminary Crash to Trauma linkage as shown in Table 1. Std. Collide lists the collision types derived from E Codes and Pairs shows the number of pairs with that value. The other columns show the number of pairs for particular events or groups of events. Events were counted if found in Most Harmful Event, First Harmful Event, or Sequential Event 1. A single pair might be counted in multiple columns depending on reported events. Ideally, one group of crash events should show high counts for one and only one Std. Collide value. Clearly, this doesn’t happen very often. For example, 84% of Pedestrian go to PED (46 / 55) with 16% errors, 79% of Most Harmful Vehicle go to VEH (544 / 688) with 21% errors, and 45% of Only No Collision Events go to NO (135 / 299) with 55% errors.

I found a number of reasons for disagreements and tried to correct them. First, based on details about  disagreements observed, I collapsed collision types produced by Std. Collide to PED, VEH (old OTH and VEH), and ONE (old NO and OBJ) and reassigned some E Codes to PED based on their fourth digit. Second, I used number of vehicles to predict collision type except for Pedestrian and Pedal Cycle events.

Table 2 Std. Collide vs. Crash Events for Num. Vehicles = 1

(Most Harmful Event, First Harmful Event, or Sequential Event 1)

New

Collide

Pairs

Pedestrian

Pedal

Cycle

Non

Collision

Ran

Off

Road

Left

Ran

Off

Road

Right

Overturn

Rollover

Any

Non

Events

Any

Objects

Only

Non

Events

Only

Objects

Most

Vehicles

First

Vehicles

Seq.

Vehicle

11

0

0

0

1

3

5

7

1

7

1

1

4

0

N/A

8

1

1

0

0

2

0

2

4

2

4

0

4

0

ONE

282

2

1

24

68

117

106

221

117

134

30

24

76

2

PED

96

49

39

0

22

3

0

25

0

25

0

7

48

0

VEH

60

2

1

6

13

18

10

41

26

22

7

10

27

1

Table 2 shows that comparisons are better for one vehicle cases. 92% of Pedestrian and Pedal Cycle go to PED (88 / 96) with 8% errors and 78% of other one vehicle cases go to ONE (282 / 361) with 22% errors.

Table 3 Std. Collide vs. Crash Events for Num. Vehicles > 1

(Most Harmful Event, First Harmful Event, or Sequential Event 1)

New

Collide

Pairs

Pedestrian

Pedal

Cycle

Non

Collision

Ran

Off

Road

Left

Ran

Off

Road

Right

Overturn

Rollover

Any

Non

Events

Any

Objects

Only

Non

Events

Only

Objects

Most

Vehicles

First

Vehicles

Seq.

Vehicle

12

0

0

0

0

0

1

1

2

1

2

10

11

2

N/A

15

0

0

0

0

1

0

1

2

0

1

14

15

4

ONE

103

0

0

6

5

19

7

33

19

23

9

54

78

16

PED

15

1

2

0

0

0

0

0

1

0

1

9

12

1

VEH

788

0

0

4

12

41

51

107

47

85

25

559

709

140

Table 3 shows that comparisons are also better for multiple vehicle cases. 84% of multiple vehicle cases go to VEH (788 / 933) with 16% errors.

Comparison results with these reclassifications were the basis for a better program to derive type of collision from number of vehicles and reported crash events. The largest number of disagreements is 103 multiple vehicle cases classified ONE in trauma. It might be that these cases involve a vehicle that is not a motor vehicle. If so, these disagreements could be corrected based on vehicle type. I added two custom functions to the User Library database.

Public Function NewCollide(varStdCollide As Variant, varECode As Variant) As Variant

    NewCollide = Null

    Select Case Nz(varECode, “”)

        Case
“812.6”, “812.7”, “813.6”, “826.0”, 826.1″

            NewCollide = “PED”

        Case Else

            Select
Case Nz(varStdCollide, “”)

                Case
“NO”, “OBJ”

                    NewCollide = “ONE”

                Case
“OTH”, “VEH”

                    NewCollide = “VEH”

                Case
Else

                    NewCollide = varStdCollide

            End Select

    End Select

End Function

Public Function CrashCollide(varNumVehicles As Variant, varEvents As Variant) As Variant

    CrashCollide = Null

        If IsNumeric(varNumVehicles) Then

        Select Case Val(varNumVehicles)

            Case 1

                CrashCollide = “ONE”

            Case Else

                CrashCollide = “VEH”

        End Select

    End If       

    If InStr(1, Nz(varEvents, “”),
“PEDAL-CYCLE”) Or InStr(1, Nz(varEvents, “”),
“PEDESTRIAN”) Then

        CrashCollide = “PED”

    End If

End Function

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

w

Connecting to %s