Using Analytical Models to Site Medical Distribution Centers€¦ · an Emergency event, what is...

Using Analytical Models to Site Medical Distribution Centers

Nicholas A. Jackson and Julie KanzlerArlington County and City of Alexandria, VA

Medical Distribution Centers Model

Presentation Outline

• City of Alexandria– GIS and stakeholders– Project and model

• Arlington County– Project and model– Application and joint effort


The Big Picture

When we fail to make Spatial Analysis part of the pre-event planning strategy, we can overlook critical selection criteria, such as the population each center might serve and its relative accessibility.

Under the City's default strategy, which uses schools as Emergency Medical Distribution Centers, portions of the City are effectively "cut off" from pedestrian access. To develop a better alternative, we generated a city-wide boolean raster model of pedestrian barriers ("Walking Accessibility Surface") and optimized allocations to centers based on walking accessibility, evenness of allocation, and landmark quality.


City Readiness Initiative

• Originated from CDC City Readiness Initiative– Aids state and local officials in developing plans that

support mass dispensing drugs to 100% of the identified population within 48 hours of a decision to do so.

– Provides funding to states, whose CRI jurisdictions cover 500 counties. This means that 56% of the US population lives within a CRI jurisdiction.

– Awarded Metropolitan Statistical Areas (MSA) close to $60 million in aid for 2006-2007

– For more detailed information: http://www.bt.cdc.gov/planning/guidance05/pdf/appendix3.pdf

• Transportation Modes for Citizens– Walking only

• Between 1 and 1½ Mile Walking Distance


Evolution

• Both jurisdictions received this project as a basic “buffer analysis”– Stakeholders in each City’s Health Department

identified the need for a GIS role in the decision-making, but didn’t know the best way to answer the question at hand: how do we best choose and evaluate which sites we choose for emergency medical distribution centers?

• Alexandria developed the model, and Arlington became interested in participating through a presentation among regional health departments


The Starting point - Schools

Mean Walking Distance (ft)

1000

2000

3000

4000

5000

6000

7000

8000

051015202530354045

No. of Facilities

% Households outside 1 mi Walking Distance from Selected Facility

0%

5%

10%

15%

20%

25%

30%

35%

40%

45%

50%

051015202530354045No. of Facilities


The Question

• Assuming that the population is on foot during an Emergency event, what is the optimal location and number of Medical Distribution Centers?– Whether the emergency event is an Anthrax Plume

or the Aftermath of a Hurricane, Medical supplies and services (e.g., Vaccines, Medications, Supplies, and Basic Triage) need to be distributed efficiently and equitably during and after it. This is why we need to make Spatial Analysis part of the pre-event planning strategy.


Walking v. “as the crow flies”

• E.g., people tend to avoid crossing into private residential property– We still need

references for these assumptions


Tools & Inputs

• Tools– ArcGIS ModelBuilder– ArcGIS Spatial Analyst

• Inputs– Boolean raster surface based on 100-scale

planimetric data (our “Walking Accessibility Surface”)

• Paved areas and roads, sidewalks and crosswalks, building footprints, rail lines, parcels, even fences

• Rasterize and classify barriers based objective rules

– Housing Unit data• A proxy for family count

– Assumption: one person from each family to user the center

• Rasterize into housing unit density

COSTALLOCAT ION

GRIDAllo catio n

GRIDDis ta nce

GRID Co stBack link

ZST AT houunits fo r e a ch

a llo ca tio n

DBF co s t suma llo c zo nes

JOIN

LAYER fac il i tie sjo in 1

MFL 1

LAYERse lec tio n

ZST ATdis tma sk

ve ctor a lloc

DBF Allo cZone zs ta t

JOIN vec to rzs tat

LAYERfa c il i tie s

COPYFEAT URES 2

FEAT UR ES SelFa c il i ties A lloc

SOMAExc lude

Build ing s

GRIDd is tmask

Ra ster toPo lyg on

FEAT URESAllo catio n

COPY

GR IDd istOut

SELECTMod e ledFac il itie s

FEAT UR ESSe le cte d

SQL Se le c tion o fFac il ie s by ID

MFL 2


ZST ATc itywid e

DBF Citywid ezs ta t

FEAT URESUnive rse o f

Fac il i tie s

SOMAremo ve b ld gs

GRID CostSurfa ce

FEAT URESUniverse o fBuild ings

T EST dis t > 1mi

GRID outs id e1 mi

ZST AT ho uunits o utsid e1 mi p er a llo c

DBF o uts id e 1mi a llo c zo ne

MFL 3

LAYER fa c il i tiesjo in 4

ZST AT ho uunits outs id e1 mi c itywid e

DBF Allo cZone zs ta t

SOMABo o lea n

Bld gs

GRID B ldgFo otp rints

Ma p Alg eb raExp re ss ion

GRID ho us ing unitd ens ity

GRIDBarriers

GRIDco nstant

COPYSe lec tio n

FEAT URESSele c te d

Summa rySta tis tics (2)

DBFsumma ry o f

JOIN HOA


COPYFEAT U RES 3

FEAT URESfa c il i tie s jo ine d

SummarySta tis tics

D BFsumma ry

MFL 4

LAYER fa c il i tie sjo ined to a llo c


The optimization

• How does the optimization work? – It starts with a 40-facility universe of potential sites for

use as Emergency Medical Distribution Centers. To select the optimal number and configuration of these sites, we developed a model to iterate through siting alternatives and successively drop the lowest performer.

• What makes one configuration of Emergency Medical Distribution Centers better than another?– We assess the differences and optimize based on

four metrics


Metrics

• (m4) Percentage of households outsidemaximum walking distance (1 mile) to the nearest site (ensures centers are located where the people are)

• (m3) Mean Walking Distance from any given location in the city to the nearest facility

• (m2) Evenness of the Population Allocated to each site (assuming individuals will tend to walk to the closest facility)

• (m1) Aggregate Landmark Quality of the selected sites (e.g., a neighborhood school is more readily located than a church)


Key Assumptions

• Barriers to movement on foot include physical impediments such as fences, buildings, and water, and social boundaries such as private residential property lines

• Individuals will tend to walk freely over all other unfenced property, private or public

• Underpasses and overpasses provide the only bypasses for pedestrian access through major highways


Results

• Results


1000

2000

3000

4000

5000

6000

7000

8000

051015202530354045

No. of Facilities


0%

5%

10%

15%

20%

25%

30%

35%

40%

45%

50%



Results

• Metrics – Before and After optimization– Baseline = “schools only” scenario

Landmark Quality(ave. landmark rank*)

1

2.35

0

0.5

1

1.5

2

2.5

Schools Optimized

Allocation Evenness (hhld std. dev)

2404

1562

0

500

1000

1500

2000

2500

3000

Schools Optimized

% Households outside 1 mi

6.51%

0.30%

0%

1%

2%

3%

4%

5%

6%

7%

Schools Optimized


3231

2678

0

500

1000

1500

2000

2500

3000

3500

Schools Optimized

(m4)(m3)

(m2)

(m1)


Results by Iteration


1000

2000

3000

4000

5000

6000

7000

8000

051015202530354045

No. of Facilities


0%

5%

10%

15%

20%

25%

30%

35%

40%

45%

50%


(m4) (m3)


Arlington’s Medication Model-Overview

•Based on the Alexandria Model•Model Scenario from the County’s Health

Planner and the CDC City Readiness Initiative •Initially Two Models: 1 mile and 1½ mile walking

distance•Model relied heavily on Human Input


Arlington Project Scenario

• Emergency Centers– Medication Centers (MC)

• Deliver medicine in first 48 hours• Medication Center’s should be less than a 1½ mile walk for Head

of House Hold one way

• Medication Center Staffing– ~20 staff/MC– Estimated 1 Head of House Hold Processed, from start to finish,

in 5 Minutes• Phase Approach for Opening MC’s

– Two Phases (Phase 1: 1½ Miles Model; Phase 2: 1 Mile Model)– Phase 1 only covered in this presentation

• Arlington Demographics– 2000 Census: 86,352 House Holds (HH)– 2008 Estimates: 96,069 House Holds (HH)


Medication Center Selection Criteria

• Medication Centers– Elementary School only model would not provide the

coverage that is needed.– Medication Centers were selected primarily by the County’s

Health Planner-• Community Centers were identified as prime locations due to the

fact that they are County Property and there is no need for a legal agreement between the County and the Owner of the facility.

• Elementary Schools were also identified as prime locations sincethey are also County Property, but they hold certain restrictions to the use of the building, and thus, they are secondary to Community Centers.

• Polling Places were considered as possible Medication Centers, but due to the fact that Arlington County does not own all the Polling Places in the county could make it difficult to open thefacility in an medical emergency event.


Arlington Tools & Inputs Arlington

• Tools– ArcGIS Model Builder– ArcGIS Spatial Analyst– ArcGIS for Cartographic Products

• Inputs– Vector Point Data

• Point Locations of Potential Medication Centers

– Community Centers, Elementary Schools, and Voting Locations

– Raster Surface• Boolean Raster representing Open Space

and Barriers – Barriers were Represented as Buildings, Divided

Highways, Single Family Parcels, Federally Owned Land, State Owned Land, Water Features, and Railroad/Metro Tracks

– Household Data• An Estimate of how many households would

be served in each Medication Center• Based on Census 2000 data and Estimated

Demography data for 2008

Co n

barop t03a lta

Input trueraste r o r

Input fa lseras te r o r

ba rrie rs(2)

Co stAlloc a tion

AlloOp t3a lta

Dis tOp t03Al ta

Back Op t03a l ta

QDC_Location_Op t3_

Sing le OutputMap A lgeb ra

DsMs kOpt3Aa

T e st

D is tOut1_5MAa

Zona lSta tis tics as

T ab le

Allo c_Zone_Sta t_1_

Raste r toPo lyg on

Alloca tion_Area sOp t

Add Fie ld


Ca lcula teFie ld


Po lygo n toRaste r

HHDe nsityAl t

Census HH_Data_200

Zona lSta tis tics a s

T a b le (2)

Op t3_ZStatAlta

back g round

Zona lSta tis tics as

T ab le (3)

Cost_Sum_Alloc_Zo

Make Fea tureLa ye r

QDC_Location_Op t3_

Add Jo in

QDC_Location_Op t3_Lay e r (2)

Add Fie ld (2)

QDC_Location_Op t3

Ca lcula teFie ld (2)

QDC_Location_Op t3

Zo na lSta tis tics as

T ab le (4)

T b l_A lloc_Zone _ZSt

Zo na lSta tis tics as

T ab le (5)

Zona lSt_Alloca t1_Op t

Add J o in (2)

QDC_Location_Op t3_


Arlington Model Differences

• Arlington County Medication Center Model– Arlington County’s Model is based on the Model

created by Alexandria– Major differences are the method of which we

selected our Medication Centers.• Arlington relied heavily on Human Input for the selection of

the Medication Center that would be used (Alexandria relied on a large collection of sites and used iteration to select Medication Centers)

• Arlington’s physical landscape is different from Alexandria’s, and thus, created a barrier to use the original Model from Alexandria

– Large number of parcels in the County are Zoned for Single Family Residential in the Northern Arlington, and Multiple Family Residential (i.e. Condos and Apartment Buildings)


Arlington Model ResultsAverage & Maximum Walking Distance per QDC Phase 1A

Max

imum

Dis

tanc

e, 7

,457

Max

imum

Dis

tanc

e, 7

,482

Max

imum

Dis

tanc

e, 8

,635

Max

imum

Dis

tanc

e, 8

,174

Max

imum

Dis

tanc

e, 4

,724

Max

imum

Dis

tanc

e, 8

,222

Max

imum

Dis

tanc

e, 7

,272

Max

imum

Dis

tanc

e, 5

,924

Max

imum

Dis

tanc

e, 9

,465

Max

imum

Dis

tanc

e, 7

,218

Max

imum

Dis

tanc

e, 7

,294

Max

imum

Dis

tanc

e, 6

,449

Max

imum

Dis

tanc

e, 7

,953

Ave

rage

, 3,3

52

Ave

rage

, 3,1

89

Ave

rage

, 3,1

78

Ave

rage

, 4,0

19

Ave

rage

, 1,9

14

Ave

rage

, 3,6

98

Ave

rage

, 3,3

33

Ave

rage

, 2,8

59

Ave

rage

, 4,1

54

Ave

rage

, 2,8

70

Ave

rage

, 4,0

66

Ave

rage

, 2,9

73

Ave

rage

, 3,7

36

0

1,000

2,000

3,000

4,000

5,000

6,000

7,000

8,000

9,000

10,000

Mad

ison

Com

mun

ityC

ente

r

Auro

ra H

illsC

omm

unity

Cen

ter

Wal

ter R

eed

Com

mun

ity C

ente

r

Swan

son

Mid

dle

Scho

ol

Fire

Sta

tion

10

Barr

ett E

lem

enta

rySc

hool

Key

Elem

enta

rySc

hool

Long

Bra

nch

Elem

enta

ry S

choo

l

Not

tingh

amEl

emen

tary

Sch

ool

Oak

ridge

Elem

enta

ry S

choo

l

Tayl

or E

lem

enta

rySc

hool

Abin

gdon

Elem

enta

ry S

choo

l

Barc

roft

Elem

enta

rySc

hool

QDC Location

Wal

king

Dis

tanc

e (f

eet)


Arlington Model Results

Total House Holds Served (2008) per QDC Phase 1A2,

026

10,8

31

8,09

7

6,97

9

7,52

9

14,4

66

11,5

79

4,69

4

4,72

9

3,24

6 5,21

9 7,81

4

8,85

8

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

Mad

ison

Com

mun

ityC

ente

r

Auro

ra H

illsC

omm

unity

Cen

ter

Wal

ter R

eed

Com

mun

ity C

ente

r

Swan

son

Mid

dle

Scho

ol

Fire

Sta

tion

10

Barr

ett E

lem

enta

rySc

hool

Key

Elem

enta

rySc

hool

Long

Bra

nch

Elem

enta

ry S

choo

l

Not

tingh

amEl

emen

tary

Sch

ool

Oak

ridge

Elem

enta

ry S

choo

l

Tayl

or E

lem

enta

rySc

hool

Abin

gdon

Elem

enta

ry S

choo

l

Barc

roft

Elem

enta

rySc

hool

QDC Location

Hou

se H

olds

Ser

ved


Arlington Model Results

Distance from QDCdistance in feet

0 - 984

985 - 1,771

1,772 - 2,509

2,510 - 3,198

3,199 - 3,887

3,888 - 4,576

4,577 - 5,265

5,266 - 5,904

5,905 - 6,495

6,496 - 7,134

7,135 - 7,920

7,921 - 12,547


Key Similarities and Differences

• Similarities– Both adopted essentially the same raster cost

allocation model, built in ModelBuilder– Both used an extensive and detailed

cadastral/planimetric database to derive areas accessible by walking

• Differences– Alexandria identified a large universe of sites and

used model iteration to select an “optimal”distribution of sites

– Arlington used local knowledge and human inputs to identify sites and evaluate site distribution


Model into Reality

• With many geospatial analysis projects, we rarely get the opportunity to witness their model being implemented in the real world.

• This coming Fall, Arlington County and the City of Alexandria will be conducting a joint cross jurisdictional public health exercise to test the validity of the Medication Center Models.

• The participants of this joint exercise will be asked to fill out a survey to identify certain key questions, such as:– The Beginning Location of their trip– Did they walk or utilize public transportation– Route taken to reach medication center– Was this an easy trip to make– Was the center easily found

• With the Survey we will be able to better model the real world in our GIS Models for the Future.


Lessons Learned so Far

• For best optimization, identify a large universe of potential sites (but don’t forget that each site adds hours of run time)

• Start with good data (e.g., spatially explicit population data)

• Make performance a key concern – changing scenarios ideally should not induce days of computer crunching

• Gather literature to backup key assumptions


Future Developments

• Update population data (2000 Census does not capture some of our new and growing population centers)

• Test the sensitivity of model assumptions to metric weighting and review the process– Performance is a big concern

• Refine assumptions on walking behavior through field testing and literature review– The “perceptual” component is difficult to capture

• Incorporate terrain and/or barrier gradient into the cost allocation analysis


How is this being applied?

• Sometimes GIS people don’t get to see how their work is applied

• This fall, there is going to be an exercise to apply and test the selected sites

Questions???

Using Analytical Models to Site Medical Distribution Centers€¦ · an Emergency event, what is...

Documents

Transcript of Using Analytical Models to Site Medical Distribution Centers€¦ · an Emergency event, what is...