Eric_Chen_final

Post on 15-Aug-2015

20 views 0 download

Tags:

Transcript of Eric_Chen_final

Introduction Results ConclusionsMethods

A Spatial Analysis of the Iowa Child Passenger Safety Survey

Based on Generalized Linear Mixed Models

Presented by Zunqiu Chen

Advisor: Dr. CavanaughDepartment of Biostatistics

University of Iowa

Introduction Results Conclusions Background Restraints Study DesignMethods

Child Passenger Safety

• According to statistics from the National SAFE KIDS Coalition, motor vehicle crashes are the leading cause of death in the United States among children from 1 to 14 years of age. (CDC, 2006)

• During 2006, 32 children under age 11 were killed in Iowa traffic crashes. (Iowa Department of Public Safety, 2007)

• 40-60 young children are seriously injured in Iowa vehicle crashes each year. (Iowa Department of Public Safety, 2007)

Introduction Results Conclusions Background Restraints Study DesignMethods

• Child safety seats reduce the risk of death by 71% for infants, and by 54% for toddlers age 1 to 4 years. (National Highway Traffic Safety Administration, 2005)

• For children age 4 to 7 years, booster seats reduce injury risk by 59% compared to seat belts alone. (Durbin et al., 2003)

• The proper use of child safety seats, booster seats, and seat belts is the best protection we have to keep children safe.

Importance of Safety Protection

Introduction Results Conclusions Background Restraints Study DesignMethods

• In 2001, the National SAFE KIDS Coalition graded each state’s child restraint laws. Iowa received an "F", ranking 46 out of 51 (50 states and the District of Columbia).

• In July of 2004, the Iowa child passenger restraint law was strengthened.

Iowa History

Introduction Results Conclusions Background Restraints Study DesignMethods

Proper Child Safety Seat Use Chart(Buckle everyone. Children age 10 and under in back!)

  INFANTS TODDLERSYOUNG

CHILDREN

AGE birth to 1 year age 2-5 years Age 6-10 years

TYPE of SEAT

rear-facing safety seatsafety seat/booster seat

booster seat/ seat belt

SEAT POSITION

rear-facing only forward-facing forward-facing

Using a restraint is not the same as using a restraint properly.

Proper Restraint

Introduction Results Conclusions Background Restraints Study DesignMethods

Rear-Facing Safety Seat

From 0-1 year old, child should be put in a rear-facing safety seat.

Introduction Results Conclusions Background Restraints Study DesignMethods

Front-Facing Safety Seat

From 2- 5 years old, child should be put in a safety seat or a booster seat.

Introduction Results Conclusions Background Restraints Study DesignMethods

Booster Seat / Seat Belt

From 6 – 10 years old, child should be put in a booster seat and/or restrained with a seat belt.

Introduction Results Conclusions Background Restraints Study DesignMethods

Proper Restraint vs. Restraint

Proper Restraint Use by Age Level Restraint Use by Age Level

0-1 2-5 6-100%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

96.57%

70.73%79.47%

3.43%

29.27%20.53%

Not Properly Restrained Properly Restrained

0-1 2-5 6-100%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

98.89%90.01%

79.47%

1.11%

9.99%20.53%

Not Restrained Restrained

Introduction Results Conclusions Background Restraints Study DesignMethods

Purpose of Investigation

• Investigate if spatial correlation exists among the pattern of proper restraint use.

• Discover the factors that influence proper restraint use.

• Evaluate compliance with the new regulation by comparing the results of years 2005, 2006, and 2007.– Is there any improvement?

Introduction Results Conclusions Background Restraints Study DesignMethods

Study Design

• Conducted by University of Iowa Injury Prevention Research Center. (UI-IPRC)

• Based on collecting samples from 36 Iowan communities (sites).– Site population sizes from 1,000 to 200,000;

classified by four levels.

Population level Category

1,000-2,499 Rural

2,500-9,999 Town

10,000-49,999 Suburban

50,000+ Urban

Introduction Results Conclusions Background Restraints Study DesignMethods

Study Area

All observations within each site share same latitude and longitude coordinates.

Introduction Results Conclusions Background Restraints Study DesignMethods

Data Structure

Driver Belted No Yes

Urban/Rural Rural Town Suburban Urban

Year 2005 2006 2007

Vehicle Type Small Medium Large

9093 observations in 2005-2007 survey

Variables collected and used in survey analysis

Age Level 1: ≤1 2: 2-5 1: ≤1 2: 2-5 3: 6-10

Restraint TypeNot rear facing

SSBelted

Rear facing

SS

SSBooster

Belted

Proper Restraint Use No Yes

Introduction Results Conclusions GLMM PROC GLIMMIXSpatial CovarianceMethods

Spatial CorrelationResidual mean based on fitted GLM without consideration of

urban/rural covariate

Introduction Results Conclusions GLMM PROC GLIMMIXSpatial CovarianceMethods

Generalized Linear Mixed Model

– Distribution: Bernoulli (Binary)• Proper restraint use

– Link: Logit– Fixed effects:

• Control for Driver belted, Age level, Year, Vehicle type

– Random effect:• Site location

GLMM model ( | ) by

Link function: = ( )

E y

g X Z

var( ) 'Z Z

• Modeling for this analysis

Introduction Results Conclusions GLMM Spatial CovarianceMethods PROC GLIMMIX

Random Effect Covariance

• Spatial random effect will be considered in our model– account for between site correlation

• Covariance assumption– Isotropic exponential spatial covariance structure is

assumed.– First law of cognitive geography: closer sites tend to be more

similar. The covariance between two observations is:

The covariance decreases with the increase of the Euclidean distance between two observations:

2

1

k

ij i j mi mjm

d c c c c

ijd

2ij where exp ij

ij

d

Introduction Results Conclusions GLMM Spatial CovarianceMethods PROC GLIMMIX

Spatial Variance-Covariance Structure

2 2 2 2 2 2 2 2 212 12 12 13 13 13

2 2 2 2 2 2 2 2 212 12 12 13 13 13

2 2 2 2 2 2 2 2 212 12 12 13 13 13

2 2 2 2 221 21 21

2 2 2 221 21 21

2 2 2 221 21 21

2 2 2 223 23 23

2 2 2 2 223 23 23

2 2 2 2 223 23 23

2 2 2

2 2 2

2 2 2

Site 1

Site 2

Site 3

Site 1 Site 2 Site 3

var( ) 'Z Z

Introduction Results Conclusions GLMM Spatial CovarianceMethods PROC GLIMMIX

GLMM Structure and GLIMMIX Code

| ~ ,

GLMM: |

y dist R

g y X Z

proc glimmix; class variables; model <resp>=<fixed effects> / dist= link= ; random <random effects> / <options>;run;

Type=sp(exp) (lat long) V=1;

Introduction Results ConclusionsMethods

Random Effect

• Euclidean distance is calculated using latitude and longitude.

• Covariance parameter estimates:

• The effective range is the distance beyond which the correlations fall below 0.05. It is estimated by:

2ˆ 0.09574 ˆ 0.2621

ˆ3 0.7863

Introduction Results ConclusionsMethods

Random EffectThe output suggests spatial correlation exists between sites within a certain distance. For example:

2 20.36; =0.249; exp 0.024ijij ij ij

dd

Introduction Results ConclusionsMethods

Fixed EffectsSignificant odds ratios:

Variables Category Odds Ratios

Driver belted No to yes 0.107

Age level Less than 1 to 6-10 7.78

2-5 to 6-10 0.497

Vehicle size Large to small 3.119

Middle to small 1.503

Year 2005 to 2007 0.464

2006 to 2007 0.675

Introduction Results ConclusionsMethods

Comparison Among 2005, 2006, and 2007 Surveys

The data shows progress in increasing the proper restraint use for child passengers.

2005 2006 200760%

65%

70%

75%

80%

85%

0.710724827812398

0.782853359814631

0.832558139534884

Properly restrained percentage

Properly restrained

Introduction Results ConclusionsMethods

Conclusions• Spatial correlation/clustering pattern does exist in the

data.

• Cautious drivers pay more attention to the safety of their child passengers.

• Drivers are more careful with the youngest children.• Drivers ignore or don’t fully understand what is proper

restraint use for children age 2 to 5 years.• Vehicle type is a potential risk factor.

• There is a statistically significant improvement in proper restraint use from 2005 to 2007.

Introduction Methods Results Conclusions

Acknowledgement

• Dr. Joseph Cavanaugh• John Lundell• Lucas Bohnett• Jing Xu• All others who have helped this project

Introduction Methods Results Conclusions

Thank you!!

Introduction Methods Results Conclusions

Proper Use of Restraint Devices (2005-2007)

Y2005-Y2007Restraint

Device

Properly Restraint (No/Yes/Total)

Age levelsTotal

Age 0-1 Age 2-5 Age 6-10

Propriety Propriety Propriety

No Yes Total No Yes Total No Yes Total No Yes Total

Belted 0 0 0 766 0 766 0 2936 2936 766 2936 3702

Booster 6 0 6 0 1220 1220 0 311 311 6 1531 1537

CSS 17 957 974 0 1591 1591 0 23 23 17 2571 2588

No 11 0 11 397 0 397 845 0 845 1253 0 1253

Total /Missing 34 957

991/2

227 28113974/

2845 3270

4115/ 9

2042 70389080/

13

Two major sources of inappropriate restraints• Use of safety belt on children between age 2 and 5 (766/2042 =

37.5%)• Not using safety belt (1253/2042 = 61.36%)

Introduction Methods Results Conclusions

Driver belted: yes Driver belted: no0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

83.97%

36.67%

16.03%

63.33%

Not properly restained

Properly restrained

Proper Restraint vs. Driver Belted Status

AssociationX2 = 1363.26d. f. = 2p-value < .0001

Year 2005-2007

Introduction Methods Results Conclusions

0-1 2-5 6-100%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

96.57%

70.73%79.47%

3.43%

29.27%20.53%

Not Properly Restrained

Properly Restrained

Proper Restraint Use by Age Level

Year 2005-2007

AssociationX2 = 320.189d. f. = 2p-value < .0001

Introduction Methods Results Conclusions

Restraint Use by Age Level

0-1 2-5 6-100%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

98.89%90.01%

79.47%

1.11%9.99%

20.53% Not re-straint

Age_level

AssociationX2 = 339.56d. f. = 2p-value < .0001

Year 2005-2007

Introduction Methods Results Conclusions

Small Midum Large0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

57.78%71.19%

85.38%

42.22%28.81%

14.62% Not Properly Restrained

Properly Restrained

Proper Restraint Use by Vehicle Type

AssociationX2 = 334.64.07d. f. = 2p-value < .0001

Year 2005-2007

Introduction Methods Results Conclusions

Rural Town SubUrban Urban0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

72.98% 78.09% 75.73% 80.71%

27.02% 21.91% 24.27% 19.29% Not Properly RestrainedProperly Restrained

Proper Restraint Use by Urban/Rural

AssociationX2 = 45.15d. f. = 3p-value < .0001

Year 2005-2007

Introduction Methods Results Conclusions

Rural areas Small town core Larger town core Metro_area high commuting

Metro_area core0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

79.62%70.72%

76.32% 78.07% 80.98%

20.38%29.28%

23.68% 21.93% 19.02%

Not Properly Restrained

Properly Restrained

Proper Restraint Use by RUCA

Higher usage in heavier traffic areas except rural areas

AssociationX2 = 73.73d. f. = 4p-value < .0001

Year 2005-2007

Introduction Methods Results Conclusions

Population Classification by RUCAYear 2005-2007

  RUCA

Col Pct Rural areas Small town core Larger town coreMetro_area

high commuting Metro_area core Total

Rural 749 301 600 0 150 1800

  29.27 15.98 36.43 0 5.71  

Town 452 683 0 0 674 1809

  17.66 36.25 0 0 25.65  

SubUrban 593 900 300 0 300 2093

  23.17 47.77 18.21 0 11.42  

Urban 765 0 747 375 1504 3391

  29.89 0 45.36 100 57.23  

Total 2559 1884 1647 375 2628 9093

Introduction Methods Results Conclusions

Proper Use of Restraint Devices (2005)

Two major sources of inappropriate restraintsUse of safety belt on children between age 2 and 5 (331/882 = 37.5%)Not using safety belt (549/882 = 62.2%)

Y2005Restraint

Device

Properly Restraint (No/Yes/Total)

Age levelsTotal

Age 0-1 Age 2-5 Age 6-10

Propriety Propriety Propriety

NoYes

Total

No Yes Total No Yes Total No Yes Total

Belted 0 0 0 331 0 331 0 903 903 331 903 1234

Booster 1 0 1 0 274 274 0 52 52 1 326 327

CSS 1329

330 0 602 602 0 7 7 1 938 939

No 3 0 3 225 0 225 321 0 321 549 0 549

Total /Missing

5329

334/1

556 8761432/

1321 962

1283/ 4

882 21673049/

6

Introduction Methods Results Conclusions

Proper Use of Restraint Devices (2006)Two major sources of inappropriate restraintsUse of safety belt on children between age 1 and 5 (265/656 = 40.4%)Not using safety belt (381/656 = 58.1%)

Y2006Restraint

Device

Properly Restraint (No/Yes/Total)

Age levelsTotal

Age 0-1 Age 2-5 Age 6-10

Propriety Propriety Propriety

No Yes Total No Yes Total No Yes Total No Yes Total

Belted 0 0 0 265 0 265 0 1042 1042 265 1042 1307

Booster 4 0 4 0 564 564 0 106 106 4 670 674

CSS 6 249 255 0 401 401 0 3 3 6 653 659

No 5 0 5 115 0 115 261 0 261 381 0 381

Total /Missing 15 249 264 380 965

1345/1

261 11511412/

4656 2365

3021/5

Introduction Methods Results Conclusions

Proper Use of Restraint Devices (2007)

Y2007Restraint

Device

Properly Restraint (No/Yes/Total)

Age levelsTotal

Age 0-1 Age 2-5 Age 6-10

Propriety Propriety Propriety

No Yes Total No Yes Total No Yes Total No Yes Total

Belted 0 0 0 170 0 170 0 991 991 170 991 1161

Booster 1 0 1 0 382 382 0 153 153 1 535 536

CSS 10 379 389 0 588 588 0 13 13 10 980 990

No 3 0 3 57 0 57 263 0 263 323 0 323

Total /Missing 14 379

393/1

227 970 1197 263 11571420/

1504 2506

3010/2

Two major sources of inappropriate restraintsUse of safety belt on children between age 1 and 5 (170/504 = 33.7%)Not using safety belt (323/504 = 64.1%)

Introduction Methods Results Conclusions

Generalized Linear Mixed Model

• General Model considerations– A Model is a description of the components of an

observation

• observation = systematic + random• Alternative: random = design components +

remaining variation• “All models are wrong but some are useful” –

G.E.P Box

Introduction Results Conclusions GLMM Spatial CovarianceMethods PROC GLIMMIX

Variance-Covariance Structure

2 2 2 2 2 21 12 12 12

2 2 2 2 2 22 12 12 12

2 2 2 2 2 23 12 12 12

2 2 2 2 2 221 21 21 4

2 2 2 2 2 221 21 21 5

2 2 2 2 2 221 21 21 6

( )

( )

( )

( )

( )

( )

v

v

v

v

v

v

Site 1

Site 2

Site 1 Site 2

( ) var( ) ' ; iVar y Z Z R R diag v

Introduction Methods Results Conclusions

GLMM Estimation

1 12 2

12

GLMM is model of ( | )

Link form: ( | )

Inverse link form: ( | ) = ( )

More general expression of distribution of |

|

( ) is "working correlation matrix"

E

i

E y u

g E y u X Zu

E y u h X Zu

y u

Var y u R R AR

R diag V A

stimation: as with LMM, may choose to focus on

1. only GLS equations in LMM;

Generalized Estimating Equations with GLMM

2. and several approachesu

Introduction Methods Results Conclusions

Pickup Pickup club cab

Car SUV Van0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

57.78%73.06% 70.90%

83.23% 86.58%

42.22%26.94% 29.10%

16.77% 13.42% Not Properly RestrainedProperly Restrained

Proper Restraint Use by Vehicle Type

Year 2005-2007

AssociationX2 = 342.46d. f. = 4p-value < .0001

Introduction Methods Results Conclusions

Limitation of the survey data and future considerations

• Multiple children in one vehicle may introduce correlation

• There appears spatial correlation and clustering in directional pattern. More sophisticated spatial analysis method could be involved to investigated the direction clustering pattern -- Kriging.

• we clearly need to increase our efforts to educate parents about what is the proper restraint device, especially for young children age from 2 to 5.

Introduction Methods Results Conclusions

References

• CDC. Web-based Injury Statistics Query and Reporting System [online]. (). National Center for Injury Prevention and Control, Centers for Disease Control and Prevention (producer). Available from URL: www.cdc.gov/ncipc/wisqars. [2006 Dec 8].

• Department of Transportation (US), National Highway Traffic Safety Administration (NHTSA), Traffic Safety Facts 2005: Children. Washington (DC): NHTSA; 2006. Available from URL: http://www-nrd.nhtsa.dot.gov/pdf/nrd-30/NCSA/TSF2005/810618.pdf.

• Durbin DR, Elliott MR, Winston FK. Belt-positioning booster seats and reduction in risk of injury among children in vehicle crashes. JAMA 2003;289(14):2835–40.

• Iowa Department of Public Safety , Child Restraint Fact Sheet http://www.dps.state.ia.us/commis/gtsb/fact-2childrestraint.pdf