jleo, v30 n2 electronic discovery and the adoption of ...€¦ · technologies that store...

Electronic Discovery and the Adoption of Information

Technology

Amalia R. Miller*

University of Virginia and RAND Corporation

Catherine E. Tucker

MIT Sloan School of Management and NBER

After firms adopt electronic information and communication technologies, their

decision-making leaves a trail of electronic information that may be more exten-

sive and accessible than a paper trail. We ask how the expected costs of litiga-

tion affect decisions to adopt technologies, such as electronic medical records

(EMRs), which leave more of an electronic trail. EMRs allow hospitals to docu-

ment electronically both patient symptoms and health providers’ reactions to

those symptoms and may improve the quality of care that makes the net impact

of their adoption on expected litigation costs ambiguous. This article studies the

impact of state rules that facilitate the use of electronic records in court. We find

evidence that hospitals are one-third less likely to adopt EMRs after these rules

are enacted. (JEL K41, O33, I12).

1. Introduction

When firms adopt new information and communication technologies(ICTs), they hope to increase profits by reducing communication andarchiving costs. However, they may incur hidden costs among which isan increased likelihood of detrimental evidence being uncovered from anelectronic “paper trail.” To take a classic example, in United States v.Microsoft, 87 F. Supp. 2d 30 (D.D.C 2000), prosecutors used e-mailssent between Microsoft executives as evidence of anticompetitive intenttoward Netscape. In this article, we explore how policies that change in therisks and expected costs of litigation stemming from the existence of anelectronic document trail influence firms in their decisions to adopt newtechnologies.

*University of Virginia and RAND Corporation. Email: [email protected]

We thank Healthcare Information and Management Systems Society Analytics for pro-

viding the data used in this study and the RAND Institute for Civil Justice for support. We

also thank Bill Encinosa, Avi Goldfarb, Patrick O’Doherty, Doug Staiger, seminar partici-

pants at Carnegie Mellon, the Federal Trade Commission, Georgia State, Texas A&M, and

Virginia, as well as the editors and anonymous reviewers ofThe Journal of Law, Economics, &

Organization for useful feedback on earlier versions of this article. All errors are our own.

The Journal of Law, Economics, and Organization, Vol. 30, No. 2doi:10.1093/jleo/ews038Advance Access published November 21, 2012� The Author 2012. Published by Oxford University Press on behalf of Yale University.All rights reserved. For Permissions, please email: [email protected]

JLEO, V30 N2 217

at MIT

Libraries on N

ovember 23, 2015

http://jleo.oxfordjournals.org/D

ownloaded from

http://jleo.oxfordjournals.org/

We ask how the prospect of electronic data being used in the courtaffects the decisions of US hospitals to adopt electronic medical record(EMRs) systems. EMRs allow health providers to store and exchangeinformation about their patients’ medical and treatment histories electron-ically rather than using paper. EMRs can improve patient’s care andreduce administrative costs (Miller and Tucker 2011a), but their effecton expected malpractice costs is ambiguous.

On the one hand, by automating documentation of a patient’s care,EMR systems can help to protect health providers in a malpractice case,by documenting comprehensively and reliably that hospital protocolswere followed. By helping to prevent medical mistakes, such as dosageerrors, EMRs may actually reduce the risk of a malpractice lawsuit beinglaunched by reducing the risk of malpractice and adverse health outcomes.

On the other hand, EMRs include more detailed information aboutpatient care than traditional paper records. Therefore, plaintiff attorneysmay make extensive discovery requests for “relevant” electronic informa-tion in medical malpractice litigation. Plaintiff access to this electronicinformation can increase the probability of malpractice litigation afteran adverse medical event, and also increase the cost of trial and thechance of a negative outcome for the defense. In a case described inVigoda and Lubarsky (2006) and Dimick (2007), surgery left a patientquadriplegic. The patient’s lawsuit initially focused on the surgeon’s com-petence, but it switched to focusing on the anesthesiologist’s competenceafter pretrial discovery, which released EMRs to the patient’s attorneys.These records contained an electronic timestamp that cast doubt onwhether the anesthesiologist was present for the entire procedure. Thisanecdote illustrates the risk to health providers of EMRs being releasedduring discovery. An increase in information can improve patient care andaid rebuttal in court, but it can also increase the chance that the plaintiffs’lawyers will find evidence of wrongdoing.

To analyze whether the threat of increased medical malpractice costsdeter adoption of EMRs, we use panel data on EMR adoption by hos-pitals from 1994 to 2007. To measure the likelihood that electronic datawill be used in medical malpractice trial proceedings, we exploit differ-ences over time in state court procedural rules governing the scope anddepth of general electronic document discovery, or “e-discovery,” in pre-trial proceedings. E-discovery refers to the use of electronic materials inthe discovery stage of court proceedings. These rules increase the likeli-hood of extensive electronic metadata being preserved, such as damagingevidence of timestamps when records were accessed and modified. We findthat the enactment of such state rules decreases the propensity of hospitalsto adopt EMRs by one-third. We check the robustness of this result byadding control variables to the main model and by employing a set offalsification tests.

We then examine which hospitals were most deterred by these proced-ural rules. We find that e-discovery rules hinder adoption more in states

218 The Journal of Law, Economics, & Organization, V30 N2

at MIT

Libraries on N

ovember 23, 2015


ownloaded from


with more malpractice litigation stemming from allegations that could be

bolstered with data from EMRs. This finding that malpractice concernsaffects technology adoption builds on the literature in health economics

that attempts to assess how the risk of malpractice litigation affects health-care provider choices. The bulk of this research concerns physician

responses to the malpractice environment, and considers location

(Matsa 2007) and treatment decisions (Kessler and McClellan 1996;Dubay et al. 1999; Currie and MacLeod 2008; Avraham et al. 2012).

We also find evidence that smaller hospitals that were most deterred by

these e-discovery rules. This may be because small hospitals find it harder

to cover the fixed cost, such as specialized software and personnel proced-ures, associated with maintaining electronic records if they face a malprac-

tice suit.By showing that the use of electronic data in court cases can inhibit the

diffusion of healthcare IT, we also contribute to a growing literature that

examines the determinants of the diffusion of health care IT in the United

States (Kim and Michelman 1990; Devaraj and Kohli 2000, 2003; Menonet al. 2000; Angst et al. 2008; Miller and Tucker 2009, 2011b). The finding

that the use of electronic data in court appears to deter hospitals from

adopting EMRs has particular policy relevance now, because the 2009HITECH Act offers incentives of roughly $44,000 per physician to pro-

mote EMR adoption.Our findings imply that these are costs, in terms of delayed or reduced

EMRs adoption, from allowing e-discovery in the court. This does not

mean that it would be best to prevent e-discovery. There may be important

benefits from e-discovery in improving the medical malpractice system.For example, e-discovery may identify and penalize truly negligent care.

However, from the perspective of providing incentives that promote the

adoption of new technologies, it is important to clarify the proceduresunder which EMRs can be used in court cases. It may be advantageous

for medical professionals to be well-informed about how much additionalrisk they will expose themselves to, rather than having them perceive those

risks as unquantifiable.

2. Background

2.1 State E-Discovery Rules

The discovery phase of a medical malpractice case starts after the plaintiff

files the lawsuit. During discovery, both the defense and the plaintiff havethe opportunity to obtain relevant information and documents from the

other parties in the lawsuit. The standard for discovery of paper records isgenerally very broad. Documents are “discoverable” if they are likely to

lead to the uncovering of admissible evidence. They do not have to be

necessarily admissible at trial. Requests for discovery are generally statu-torily predetermined requests that must be produced without objection.

Electronic Discovery and the Adoption of Information Technology 219

at MIT

Libraries on N

ovember 23, 2015


ownloaded from


Therefore, all parties in a lawsuit must respond to discovery requests or

face being found in contempt.In the past decade, many states have adopted rules that govern

“e-discovery.” These rules generally add electronic documents as an add-

itional class of documents that are governed by existing rules on discovery

in pretrial proceedings. This means that such materials fall, without any

room for dispute, into the class of materials that must be automatically

produced without objection in pretrial proceedings. As shown by Figure 1,

these rules are geographically diverse. The rules originate both from stat-

utes and courts. Table 1 summarizes the rules that have been enacted.In the absence of these rules, the plaintiff’s and the defendant’s teams of

lawyers must hash out the use of electronic materials between themselves.

Interviews with medical malpractice attorneys to provide evidence on this

informal and undocumented process suggest that they often reach an

agreement with the other party to exclude electronic evidence from the

discovery process. Lawyers rationalize such behavior by pointing out that

without clarification from the courts about how e-discovery should be

conducted, e-discovery becomes costly and may not produce worthwhile

evidence that offsets these costs. A recent “2008 Litigation Survey of

Fellows of the American College of Trial Lawyers” (ACTL and IAALS

(2009)) suggested that nearly 77% of the courts did not understand the

difficulties associated with e-discovery and that 87% of the trial lawyers

said that e-discovery increases the costs of litigation.In our regressions, we use an indicator variable to signal the existence of

a state rule. In all cases, we focus on the dates the rules became effective.1

We do not exploit the variation in the wording of the rule. We did check

the robustness of our results to the exclusion of Texas, which appears to

have the least “plaintiff-friendly” rules for e-discovery of the states in our

sample, and obtained similar, if marginally higher, estimates of the effects

of e-discovery rules than before.In 2007, there were sweeping changes to the Federal Rules of Civil

Procedure (effective December 2006) that encompassed electronic meta-

data, such as creation dates and modification dates. Though not directly

applying to state courts, there is the possibility that some state courts may

have used these new rules as a reference. As a robustness check, we

1. Since these rules are generally changes to the court code of civil procedure, there is not a

large gap between the enactment date and the effective date as sometimes it occurs with

changes in state law. Furthermore, while these state rules often formalize and codify the

procedures that have been followed in a prior case, they do represent a discontinuity in

terms of legal practice. This is for two reasons. First, before such rules are enacted, it is

not clear whether in that state practices that have been followed in complex corporate civil

cases apply to all civil procedure. Second, while lawyers who specialize in such areas of

corporate law may be well aware of the issues of e-discovery, the enactment of such rules is

ordinarily attended by both publicity, and offers of workshops and training in matters of

e-discovery by the state bar association, which improve the ability of the broad spectrum of

the legal profession to take advantage of electronic data mining techniques.


at MIT

Libraries on N

ovember 23, 2015


ownloaded from


estimate an alternative version of our main specification excluding 2007data with similar results.

We also investigated the origin of the state rules. It would be problem-atic for our interpretation of causal effects if these rules stemmed fromlobbying by large hospitals or medical malpractice litigation experts inthat state. A search of early case law that mentions e-discovery suggeststhat rather than reflecting activity in the health sector, cases instead reflectlitigation practices in the financial sector. Banks and other financial firmswere the first industrial sector to embrace a large number of ICT-typetechnologies that store electronic data. For example, Merrill Lynch’sJonathan Eisenberg noted that in his experience, 98% of the records indiscovery cases involving Merrill Lynch are electronic (Losey 2008). Anexample of this relationship between the presence of financial firms andthe enactment of state-level e-discovery rules isZubulake v. UBSWarburg,217 F.R.D. 309 (S.D.N.Y. 2003), where UBS had to pay $29.2 million,partly because they failed to store data properly. Conversations withe-discovery experts suggest that it was the holes in the existing discoverytrial procedure guidelines exposed by Zubulake, which prompted the NewYork courts to issue clarifying procedural rules in 2006.

2.2 Effects of E-Discovery on Malpractice Litigation

The impact of greater use of electronic data in malpractice cases on the netexpected malpractice costs of hospitals is ambiguous. The majority of theEMR policy literature emphasizes that EMRs can document that hospitalprotocols were followed, and consequently provide a more complete and

Figure 1. Distribution of State E-discovery Rules in 2007.


at MIT

Libraries on N

ovember 23, 2015


ownloaded from


Ta

ble

1.

Sta

teR

ule

sG

ove

rnin

gE

-dis

co

ve

ry

Sta

teR

ule

Da

teS

ou

rce

De

sc

rip

tion

CT

Co

nn

ec

ticu

tP

rac

tice

Bo

ok,

Su

pe

rio

rC

ou

rt—

pro

ce

du

res

inc

ivil

ma

tte

rsS

ec

tion

13

-9.

Re

qu

ests

for

pro

du

ctio

n,

insp

ec

tion

an

de

xa

min

atio

n;

ing

en

era

l

(se

esu

bse

ctio

n(d

),a

tp

.1

92

of

25

9-p

ag

e.p

df

do

cu

me

nt)

Eff

ec

tive

Ja

nu

ary

1,

20

06

Co

urt

Am

en

dm

en

tsc

rea

ting

ap

roc

ed

ure

toa

dd

ress

ho

we

-dis

co

ve

ry

will

take

pla

ce

IDId

ah

oR

.C

iv.

P.

34

Eff

ec

tive

Ju

ly1

,2

00

6C

ou

rtA

me

nd

me

nts

ad

de

lec

tro

nic

ally

sto

red

info

rma

tion

(ES

I)to

exis

ting

rule

sre

latin

gto

dis

co

ve

ryo

fd

oc

um

en

ts

ILIll

ino

isS

up

rem

eC

ou

rtR

ule

s2

01

(b)(

1)

an

d2

14

Eff

ec

tive

Ja

nu

ary

1,

19

96

Co

urt

Ad

ds“re

trie

va

ble

”E

SI

toe

xis

ting

rule

so

nd

isc

ove

ry;

req

uir

es

pro

vis

ion

inp

rin

ted

form

of

su

ch

ES

I,a

nd

tha

tth

ey

pro

du

ce

tho

se

do

cu

me

nts

org

an

ize

din

the

ord

er

inw

hic

hth

ey

are

ke

pt

inth

eu

su

al

co

urs

eo

fb

usin

ess

or

org

an

ize

da

nd

lab

ele

dto

co

rre

sp

on

dw

ithth

ec

ate

go

rie

sin

the

req

ue

st.

LA

CC

P1

42

4—

sc

op

eo

fd

isc

ove

ry;

tria

lp

rep

ara

tion

;m

ate

r-

ials

:C

CP

14

60

—o

ptio

nto

pro

du

ce

bu

sin

ess

rec

ord

s:

CC

P1

46

1—

pro

du

ctio

no

fd

oc

um

en

tsa

nd

thin

gs;

en

try

up

on

lan

d;

sc

op

e:

CC

P1

46

2—

pro

du

ctio

no

f

do

cu

me

nts

an

dth

ing

s;

en

try

up

on

lan

d;

pro

ce

du

re

Ap

pro

ve

dJu

ne

25

,2

00

7,

eff

ec

tive

Oc

tob

er

20

07

Sta

tute

Ad

ds

ES

Ito

de

fin

itio

no

f“w

ritin

g”

ina

cla

use

exc

lud

ing

dis

-

co

ve

ryfo

rm

ate

ria

lsp

rep

are

din

ad

va

nc

eo

flit

iga

tion

or

in

pre

pa

ratio

nfo

rtr

ial.

Allo

ws

inte

rro

ga

tori

es

tob

ea

nsw

ere

db

y

sp

ec

ifyin

gth

eE

SI

fro

mw

hic

hth

ea

nsw

er

ma

yb

eo

bta

ine

d.

Ad

ds

ES

Ito

exis

ting

rule

sc

ove

rin

gp

rod

uc

tion

of

do

cu

-

me

nts

.A

dd

sE

SI

toe

xis

ting

rule

sc

ove

rin

gp

rod

uc

tion

of

do

cu

me

nts

;re

qu

ire

sE

SI

tob

ep

rod

uc

ed

inth

efo

rmin

wh

ich

itis

“o

rdin

ari

lym

ain

tain

ed

”o

ris

rea

so

na

bly

in

usa

ble

form

.

MN

Am

en

dm

en

tsto

Ru

les

of

Civ

ilP

roc

ed

ure

16

,2

6,

33

,3

4,

37

,4

5

Eff

ec

tive

Ju

ly1

,2

00

7C

ou

rtA

dd

sa

sa

va

lidm

att

er

for

pre

tria

lc

on

fere

nc

es

the

rea

ch

ing

of

ag

ree

me

nts

on

e-d

isc

ove

ry

MS

Mis

s.

R.

Civ

.P

.2

6(b

)(5

)E

ffe

ctiv

eM

ay

29

,2

00

3C

ou

rtR

eq

uir

es

tha

t“T

oo

bta

ind

isc

ove

ryo

fd

ata

or

info

rma

tion

tha

t

exis

tsin

ele

ctr

on

ico

rm

ag

ne

ticfo

rm,

the

req

ue

stin

gp

art

y

mu

st

sp

ec

ific

ally

req

ue

st

pro

du

ctio

no

fe

lec

tro

nic

or

ma

g-

ne

ticd

ata

an

dsp

ec

ifyth

efo

rmin

wh

ich

the

req

ue

stin

g

pa

rty

wa

nts

itp

rod

uc

ed

.T

he

resp

on

din

gp

art

ym

ust

pro

du

ce

the

ele

ctr

on

ico

rm

ag

ne

ticd

ata

tha

tis

resp

on

siv

eto

the

re-

qu

est

an

dis

rea

so

na

bly

ava

ilab

leto

the

resp

on

din

gp

art

yin

itso

rdin

ary

co

urs

eo

fb

usin

ess.”

(co

ntin

ue

d)


at MIT

Libraries on N

ovember 23, 2015


ownloaded from


Ta

ble

1.

Co

ntin

ue

d

Sta

teR

ule

Da

teS

ou

rce

De

sc

rip

tion

MT

Mo

nt.

R.

Civ

.P

.1

6(b

).S

ch

ed

ulin

ga

nd

pla

nn

ing

,2

6(b

).

Dis

co

ve

rysc

op

ea

nd

limits

,2

6(f

).D

isc

ove

ryc

on

fer-

en

ce

,3

3(c

).O

ptio

nto

pro

du

ce

bu

sin

ess

rec

ord

s,

34

(a).

Sc

op

e,

34

(b).

Pro

ce

du

re,

37

(e).

Ele

ctr

on

ica

lly

sto

red

info

rma

tion

,4

5(a

).F

orm

issu

an

ce

,4

5(c

).

Pro

tec

tion

of

pe

rso

ns

su

bje

ct

too

ra

ffe

cte

db

ysu

b-

po

en

as,

45

(d).

Du

ties

inre

sp

on

din

gto

su

bp

oe

na

Eff

ec

tive

Fe

bru

ary

28

,

20

07

Co

urt

Ad

ds

as

ava

lidm

att

er

for

pre

tria

lc

on

fere

nc

es

the

rea

ch

ing

of

ag

ree

me

nts

on

e-d

isc

ove

ry,

limits

dis

co

ve

ryo

fE

SI

tha

t

wo

uld

be“u

nre

aso

na

bly

bu

rde

nso

me

or

exp

en

siv

e,”

or“u

n-

rea

so

na

bly

cu

mu

lativ

eo

rd

up

lica

tive

,”a

dd

sa

sa

va

lid

ma

tte

rfo

rp

retr

ial

co

nfe

ren

ce

sth

ere

ac

hin

go

fa

gre

em

en

ts

on

e-d

isc

ove

ry.

Allo

ws

inte

rro

ga

torie

sto

be

an

sw

ere

db

y

sp

ec

ifyin

gth

eE

SI

fro

mw

hic

hth

ea

nsw

er

ma

yb

eo

bta

ine

d.

Ad

ds

ES

Ito

exis

ting

rule

sc

ove

rin

gp

rod

uc

tion

of

do

cu

-

me

nts

.A

dd

sa

ne

xe

mp

tion

fro

mn

orm

al

sa

nc

tion

sfo

rfa

ilure

toc

oo

pe

rate

ind

isc

ove

ry,

for

ES

I“lo

st

as

are

su

lto

fth

ero

u-

tine

,g

oo

dfa

itho

pe

ratio

no

fa

ne

lec

tro

nic

info

rma

tion

sys-

tem

”A

dd

sE

SI

tola

ng

ua

ge

go

ve

rnin

gsu

bp

oe

na

s.

Allo

ws

su

bp

oe

na

sto

co

ve

rsa

mp

ling

of

ES

I.E

SI

mu

st

be

pro

vid

ed

as“ke

pt

inth

eu

su

al

co

urs

eo

fb

usin

ess

”o

ro

rga

niz

ed

an

d

lab

ele

d“to

co

rre

sp

on

dw

ithth

ec

ate

go

rie

sin

the

de

ma

nd

.”

Ifsu

bp

oe

na

do

es

no

tsp

ec

ifyfo

rm,

ES

Im

ust

be

pro

vid

ed

“in

afo

rmo

rfo

rms

inw

hic

hth

ep

ers

on

ord

ina

rily

ma

inta

ins

it,o

rin

afo

rmo

rfo

rms

tha

ta

rere

aso

na

bly

usa

ble

.”

Exe

mp

tsE

SI

fro

mso

urc

es

tha

ta

ren

ot

ac

ce

ssib

leb

ec

au

se

of“u

nd

ue

bu

rde

no

rc

ost”

NC

Ru

les

for

Su

pe

rio

rC

ou

rtJu

dic

ial

Dis

tric

t1

5B

:R

ule

6—

Dis

co

ve

ry

Eff

ec

tive

Ju

ly1

,2

00

6C

ou

rtR

eq

uir

es

ES

Ito

be

pro

vid

ed

ina“re

aso

na

bly

usa

ble

”fo

rm

NH

Su

pe

rio

rC

ou

rtR

ule

62

.(I

)In

itia

lS

tru

ctu

rin

gC

on

fere

nc

e

(se

esu

bse

ctio

n(C

)(4

))

Eff

ec

tive

Ma

rch

1,

20

07

Co

urt

Ad

ds

as

ava

lidm

att

er

for

pre

tria

lc

on

fere

nc

es

the

rea

ch

ing

of

ag

ree

me

nts

on

e-d

isc

ove

ry

NJ

Pa

rtIV

Ru

les

Go

ve

rnin

gC

ivil

Pra

ctic

ein

the

Su

pe

rio

r

Co

urt

,T

ax

Co

urt

,a

nd

Su

rro

ga

tes

Co

urt

s,

Ru

le1

:9.

Su

bp

oe

na

s,

Ru

le4

:5B

.C

ase

Ma

na

ge

me

nt;

Co

nfe

ren

ce

s,

Ru

le4

:10

.P

retr

ial

Dis

co

ve

ry,

Ru

le4

:17

.

Inte

rro

ga

tori

es

toP

art

ies,

Ru

le4

:18

.D

isc

ove

rya

nd

Insp

ec

tion

of

Do

cu

me

nts

an

dP

rop

ert

y;

Co

pie

so

f,

Eff

ec

tive

Se

pte

mb

er

1,

20

06

Co

urt

Allo

ws

su

bp

oe

na

sfo

rE

SI

tha

tis

no

t“u

nre

aso

na

ble

or

op

pre

s-

siv

e,”

ad

ds

as

ava

lidm

att

er

for

pre

tria

lc

on

fere

nc

es

the

rea

ch

ing

of

ag

ree

me

nts

on

e-d

isc

ove

ry,

allo

ws

cla

ims

tha

t

ES

Iis

no

tre

aso

na

bly

ac

ce

ssib

leif

pa

rty

pre

-sp

ec

ifie

sth

e

so

urc

es

tha

ta

ren

ot

ac

ce

ssib

le,

allo

ws

inte

rro

ga

torie

sto

be

an

sw

ere

db

ysp

ec

ifyin

gth

eE

SI

fro

mw

hic

hth

ea

nsw

er

ma

y

be

ob

tain

ed

,a

dd

sE

SI

toe

xis

ting

rule

sc

ove

rin

gp

rod

uc

tion

(co

ntin

ue

d)


at MIT

Libraries on N

ovember 23, 2015


ownloaded from


Ta

ble

1.

Co

ntin

ue

d

Sta

teR

ule

Da

teS

ou

rce

De

sc

rip

tion

Ru

le4

:23

.F

ailu

reto

Ma

ke

Dis

co

ve

ry;

Sa

nc

tion

s

Do

cu

me

nts

of

do

cu

me

nts

.S

pe

cifi

es

tha

tth

eE

SI

mu

st

be

ina

form

or

form

sin

wh

ich

itis

ord

ina

rily

ma

inta

ine

do

rin

afo

rmo

r

form

sth

at

are

rea

so

na

bly

usa

ble

.A

dd

sa

ne

xe

mp

tion

fro

m

no

rma

lsa

nc

tion

sfo

rfa

ilure

toc

oo

pe

rate

ind

isc

ove

ry,

for

ES

I“lo

st

as

are

su

lto

fth

ero

utin

e,

go

od

faith

op

era

tion

of

an

ele

ctr

on

icin

form

atio

nsyste

m”

NY

Un

iform

Civ

ilR

ule

so

fth

eS

up

rem

ea

nd

Co

un

tyC

ou

rts,

20

2.7

0C

om

me

rcia

lD

ivis

ion

of

the

Su

pre

me

Co

urt

,

Ru

le8

.C

on

su

ltatio

np

rio

rto

Pre

limin

ary

an

d

Co

mp

lian

ce

Co

nfe

ren

ce

s

Eff

ec

tive

Ja

nu

ary

17

,2

00

6C

ou

rtR

eq

uir

es

me

etin

go

fc

ou

nse

lb

efo

rep

relim

ina

ryc

on

fere

nc

eto

ad

dre

ss

e-d

isc

ove

ry,

inc

lud

ing

“(i)

imp

lem

en

tatio

no

fa

da

ta

pre

se

rva

tion

pla

n;

(ii)

ide

ntifi

ca

tion

of

rele

va

nt

da

ta;

(iii)

the

sc

op

e,

exte

nt

an

dfo

rmo

fp

rod

uc

tion

;(i

v)

an

ticip

ate

dc

ost

of

da

tare

co

ve

rya

nd

pro

po

se

din

itia

la

lloc

atio

no

fsu

ch

co

st;

(v)

dis

clo

su

reo

fth

ep

rog

ram

sa

nd

ma

nn

er

inw

hic

hth

ed

ata

ism

ain

tain

ed

;(v

i)id

en

tific

atio

no

fc

om

pu

ter

syste

m(s

)u

ti-

lize

d;

(vii)

ide

ntifi

ca

tion

of

the

ind

ivid

ua

l(s)

resp

on

sib

lefo

r

da

tap

rese

rva

tion

;(v

iii)

co

nfid

en

tialit

ya

nd

privile

ge

issu

es;

an

d(i

x)

de

sig

na

tion

of

exp

ert

s.”

TX

Te

x.

R.

Civ

.P

.1

96

.4E

lec

tro

nic

or

Ma

gn

etic

Da

taE

ffe

ctiv

eJa

nu

ary

1,

19

99

Co

urt

ES

Im

ust

be

sp

ec

ific

ally

req

ue

ste

da

nd

itsfo

rmp

resp

ec

ifie

d;

limite

dto

ES

I“re

aso

na

bly

ava

ilab

leto

the

resp

on

din

gp

art

y

inits

ord

ina

ryc

ou

rse

of

bu

sin

ess

”

UT

Uta

hR

.C

iv.

P.

26

.G

en

era

lp

rovis

ion

sg

ove

rnin

gd

is-

co

ve

ry:

P.

33

.In

terr

og

ato

rie

sto

pa

rtie

s:.

P.

34

.

Pro

du

ctio

no

fd

oc

um

en

tsa

nd

thin

gs

an

de

ntr

yu

po

n

lan

dfo

rin

sp

ec

tion

an

do

the

rp

urp

ose

s:

P.

37

.F

ailu

re

tom

ake

or

co

op

era

tein

dis

co

ve

ry;

sa

nc

tion

s

Eff

ec

tive

No

ve

mb

er

1,

20

07

Co

urt

Lim

itsd

isc

ove

ryo

fE

SI

tha

tp

ose

sa

n“u

nd

ue

bu

rde

no

rc

ost,

”

ad

ds

pro

vis

ion

sc

ove

rin

gin

ad

ve

rte

nt

dis

clo

su

reo

fp

rivile

ge

d

info

rma

tion

.A

llow

sin

terr

og

ato

rie

sto

be

an

sw

ere

db

ysp

e-

cify

ing

the

ES

Ifr

om

wh

ich

the

an

sw

er

ma

yb

eo

bta

ine

d.

Ad

ds

ES

Ito

exis

ting

rule

sc

ove

rin

gp

rod

uc

tion

of

do

cu

-

me

nts

.A

dd

sa

ne

xe

mp

tion

fro

mn

orm

al

sa

nc

tion

sfo

rfa

ilure

toc

oo

pe

rate

ind

isc

ove

ry,

for

ES

I“lo

st

as

are

su

lto

fth

ero

u-

tine

,g

oo

dfa

itho

pe

ratio

no

fa

ne

lec

tro

nic

info

rma

tion

syste

m”

So

urc

efo

rd

ate

so

fe

na

ctm

en

to

fla

ws/r

ule

s:

K&

LG

ate

sL

LP

,E

-Dis

co

ve

ryA

na

lysis

&T

ec

hn

olo

gy

Gro

up“C

urr

en

tL

istin

go

fS

tate

sT

ha

tH

ave

En

ac

ted

E-D

isc

ove

ryR

ule

s”

Oc

tob

er

20

08

.D

esc

rip

tion

of

ea

ch

rule

ba

se

do

na

ctu

al

rule

or

sta

tute

text.


at MIT

Libraries on N

ovember 23, 2015


ownloaded from


easily “provable” paper trail for the defense. EMRs may prevent mistakesby standardizing care and patient histories. However, there are numerousreasons to think that the admission of EMRs into discovery may increase

the expected malpractice costs of the hospital.The presence of electronic data makes it more likely that a mistake, if it

were made, would be recorded. For example, metadata captures when andwhether physicians refer to a patient’s history. This is a change from thecurrent paper-based systems, where failure to obtain past clinical history

has had to be judged based on a comprehensive review of the individualcase files. Blumenthal et al. (2006) discuss the concern that widespread useof electronically stored information (ESI) in medical malpractice casesintroduces ambiguity in the definition of a legal “medical record” and

extends the scope of discovery.2

ESI also opens the possibility that even if the initially alleged error or

negligence did not occur, the plaintiff could data mine the electronic in-formation until it found another piece of data that might indicate a mis-take (Terry 2001). Hospitals may also fear that in the course of discoveryfor a single malpractice case, the electronic information will reveal a

system-wide error in the EMR system’s clinical guidelines and alertsthat would affect a large class of patients, thereby amplifying the riskrelative to paper documents.

Requests for ESI may cause problems for defendants because the meta-data contained in the electronic record can give the impression that med-

ical records have been tampered with. This is a particular point ofvulnerability for defendants in court. One example, given to us by anindustry insider, was the case of a radiologist in Arizona who prior to adeposition accessed the patient’s record to review its contents. By doing

so, they inadvertently created a timestamp which suggested that the recordhad been “modified” just prior to the deposition. In Arizona, the absenceof clear rules meant that this inadvertent electronic data were not admittedas evidence, but legal experts say that under new state rules, this electronic

evidence of possible “tampering” could be used effectively by the plaintiffto have medical evidence dismissed. Electronic timestamps may also revealinappropriate corrections to a medical record, inaccurate data entry, andunauthorized access.

Further, if there are informal practices in a hospital, which go againstofficial protocol, they will also be recorded. For example, in some hos-

pitals when nurses are called away in an emergency from making regularchecks on patients recovering from surgery, they can “correct” paper re-cords after the fact to indicate that they had made the checks. That is notpossible with EMRs. Similarly, Williams (2009) reports a case where the

2. This extension goes beyond the medical record information that hospitals are required

to provide to patients upon request under the Privacy Rule of the 1996 Health Insurance

Portability and Accountability Act (AHIMA e-HIM Work Group on Defining the Legal

Health Record 2005; AHIMA e-HIM Work Group on e-Discovery 2006).


at MIT

Libraries on N

ovember 23, 2015


ownloaded from


focus of a medical malpractice suit was the fact that a nurse made entriesto an electronic record after the patient had died. If the patient records hadbeen in paper, there would have been no way of knowing for sure whenthese entries were made.

More generally, there is anecdotal evidence that e-discovery is viewed asa burden for medical providers. For example, at the IQPC’s 3rd ObstetricMalpractice Conference, held on November 10–11, 2009 at the SwissotelChicago, one session described e-discovery as “an aggressive enemy lurk-ing at the door of every hospital in the United States” and “a friend of theplaintiff’s bar,” explaining that “a wealth of digital information can ac-cumulate about a patient that is housed outside an organization’s legalmedical record. Attorneys in search of additional information pertinent toa lawsuit may view this data as containing digital Easter eggs waiting to bediscovered.”3

The existing academic literature has come to mixed conclusions.Feldman (2004) quotes survey evidence that shows that in 41 malpracticecases, there were no reported cases where an “automated record” hinderedthe defense process. He also discusses quotes from survey participants thatsuggest positive legal outcomes, such as “I know of three cases where theanesthesia record directly contributed to the anesthesiologist being dis-missed (from the suit).” Lane (2005) points out that it would be unwise toconclude from anecdotal evidence that electronic systems do not increasemalpractice exposure, because Feldman (2004) ignores the additionalrisk created by additional data stored in the electronic record. To try tounderstand how “malpractice risk” and adoption of EMRs may correlate,Virapongse et al. (2008) sent surveys to 1140 physicians in Massachusettsbut, after controlling for demographics, did not find a significantrelationship.

3. Data

We use technology data from the 2008 release of the HealthcareInformation and Management Systems Society AnalyticsTM Database(HADB). To control for time-varying hospital characteristics, we matchedthe HADB data with the American Hospital Association (AHA) surveyfrom 1994 to 2007. We study the initial EMR adoption decisions using theHADB–AHA-matched samples of 3599 hospitals that had not adoptedEMRs before 1994. The hospitals in our data were generally larger thanthe hospitals we could not match in the AHA data. On average, they had7530 when compared to the 2717 average annual admissions of thehospitals for which the HADB data did not contain information on IT

3. The full session description is available at http://www.iqpc.com/Event.

aspx?id¼222598.


at MIT

Libraries on N

ovember 23, 2015


ownloaded from

http://www.iqpc.com/Event.aspx?id=222598




adoption.4 Therefore, our estimates should be taken as representative onlyof larger hospitals. Looking ahead, in Section 5, we show that smallerhospitals respond more negatively to the presence of e-discovery laws.Thus, it seems likely that by omitting smaller hospitals, we are understat-ing the average impact of the potential for the use of electronic data in thecourt on EMR adoption.

Table 2 describes the main variables in our regressions, including themultiple control variables that we use to control for hospital-level hetero-geneity. Appendix Table A1 reports summary statistics for the initial yearof the sample, separately for hospitals in states that did and did not adopte-discovery rules by 2007. On average, the two sets of hospitals appearquite similar.

We measure adoption of EMRs by whether a hospital has installed orhas entered into a contract installing an “enterprise EMR” system.Installing EMRs in a hospital setting may take many years, so we defineadoption by whether or not the hospital has contracted with a EMRsvendor to set up an EMR system in that year. This is because it is thedecision to choose to install an EMRs, which will be influenced by thepotential of electronic data in litigation rather than the EMRs’ completiondate. Enterprise EMRs software provides the software skeleton thatunderlies other potential add-ins, such as clinical decision support, a clin-ical data repository, and order entry. EMRs software can therefore pro-vide the electronic metadata, such as timestamps, file modification dates,and user access details, which increase the amount of information avail-able to lawyers in a medical malpractice case when compared to paperrecords. Of the 3599 hospitals in our sample, 1400 hospitals adoptedEMRs during the sample period between 1994 and 2007 and 2199 hadnot adopted by the end of 2007.5 The average annual adoption rate ofEMRs among hospitals that had not previously adopted the technologywas 3.3%.

We focus on how the expected malpractice costs associated with EMRsaffect hospital adoption decisions, because the EMR system ishospital-wide. However, malpractice cases may not be hospital-wide.Litigation resulting from hospital care may be directed at the hospital;at the physician and other members of the medical team as well as the hos-pital; or at individual members of the medical team. Therefore, thehospital adoption decision may reflect not only its direct perception ofthe negative consequences of electronic data being used in the court foritself, but also the negative consequences faced by its physicians, if the

4. Also, we do not have information on hospitals that were in operation during the sample

period but that were closed before 2007, but this represents <1% of the AHA sample. Two

hundred and four hospitals in the matched sample first appear in the AHA data after 1994.

5. Nonadopting hospitals are included in the sample for all years with available AHAdata

and contribute 29,729 of the hospital-year observations. Hospitals that adopt during the

sample period are observed for an average of 8.8 years and contribute the remaining

12,377 observations.


at MIT

Libraries on N

ovember 23, 2015


ownloaded from


hospital worries that this could interfere with employing high-caliberphysicians.

4. Results

We first examine the aggregate impact of rules that clarify the use ofe-discovery in the pretrial stages of medical malpractice suits on adoptionof EMRs. As described in Section 2.1, these rules increase how muchelectronic information a plaintiff’s lawyer may receive automatically aspart of pretrial disclosure. As such, the rules are expected to magnify theimpact of electronic information on the litigation process.

Table 2. Summary Statistics for Full Sample

Variable Mean (SD) Min Max

Adopt EMR 0.033 (0.18) 0 1

Adopt business intelligence software 0.0083 (0.091) 0 1

Adopt financial data warehousing software 0.0067 (0.082) 0 1

E-discovery rule 0.11 (0.31) 0 1

Years opened 34.3 (34.4) 0 190

Staffed beds 182.9 (174.9) 3 1875

Admissions (000) 7.53 (8.12) 0.0030 98.2

Inpatient days (000) 42.7 (48.2) 0.0040 582.0

Medicare inpatient days (000) 19.2 (21.1) 0 476.9

Medicaid inpatient days (000) 8.43 (14.3) 0 302.8

Births 896.9 (1257.0) 0 16463

Total inpatient operations (000) 2.26 (2.87) 0 83.1

Total operations (000) 5.88 (6.64) 0 213.4

Emergency outpatient visits (000) 23.5 (21.8) 0 290.1

Total outpatient visits (000) 112.5 (147.2) 0 2935.9

Total payroll expenses (USDm) 35.7 (52.1) 0.037 1116.5

Employee benefits (USDm) 8.19 (12.7) 0 294.5

Total expenses (USDm) 85.0 (125.2) 0.095 2393.8

Length of stay 1.01 (0.078) 1 2

No. of doctors 15.5 (65.0) 0 2067

No. of nurses 219.4 (279.2) 0 3325

No. of trainees 19.0 (82.2) 0 1347

Nonmedical staff 608.7 (769.2) 0 12,054

PPO 0.64 (0.48) 0 1

HMO 0.56 (0.50) 0 1

Member hospital system 0.48 (0.50) 0 1

Speciality hospital 0.036 (0.19) 0 1

Nonprofit 0.60 (0.49) 0 1

Gross state product (USDtr) 0.38 (0.37) 0.014 1.81

Gross state product per capita (USD000) 32.5 (4.99) 20.9 58.8

EMR-related malpractice payouts (USDm) 0.26 (0.11) 0.022 0.92

EMR-unrelated malpractice payouts (USDm) 0.18 (0.072) 0.041 0.81

Cap punitive 0.55 (0.50) 0 1

Cap noneconomic 0.36 (0.48) 0 1

Joint + several liability 0.79 (0.41) 0 1

Contingency fee 0.39 (0.49) 0 1

No. of observations 42,106


at MIT

Libraries on N

ovember 23, 2015


ownloaded from


We model hospitals as maximizing an objective function that includes

net revenues and patient outcomes, including potential costs associatedwith malpractice lawsuits. Hospitals choose to adopt EMRs, if the net

benefits are positive. We model adoption of EMRs as an irreversible,

absorbing state and exclude hospitals that have previously adopted

from the sample. Examining the past years of HADB data confirms thatthere are no observations where a hospital divests itself of an EMR system

without seeking a replacement. We use a discrete-time hazard model since

our survival time data are discrete (the year of adoption). Discrete survival

time models can be estimated using standard binary choice methods, if thepanel is limited to time periods for each firm when it is still at risk of the

event (Allison 1982). We use a probit specification to model adoption

decisions. Our results are robust to using a linear probability model,

a logit specification, or a Cox proportional hazards model (AppendixTable A2).

We model the probability that a hospital i in state j adopts EMR in year

t as

ProbðAdoptEMRijt ¼ 1jE-discoveryRulejt,XijtÞ

¼ �ðE-discoveryRuleijt,Xijt, �Þð1Þ

E-discoveryRulejt, our key variable of interest is an indicator for the

presence of an e-discovery rule in that state in year t. Xijt is a vector ofhospital characteristics in year t as described in Table 2, which affect the

propensity to exchange information, g is a vector of unknown parameters,

and � is the cumulative distribution function of the standard normal

distribution.Robust standard errors (SEs) are clustered at the state level. This allows

for the potential correlation within states over time and between hospitals.

This robustness is especially important for our study because the policy

variation occurs at the state level (Bertrand et al. 2004). Clustering errors

at the hospital level produces similar estimates but they are somewhatsmaller for the main variable of interest.6 This suggests that clustering

at the state level provides a more conservative and appropriate test for

the impact of the state-level policies we study.Table 3 reports our initial results. Column (1) reflects the results of our

initial panel specification that includes the full set of state and year fixed

effects. Here, the presence of a rule that facilitates e-discovery is associated

with a statistically significant 0.24 reduction in the latent variable captur-

ing the net value of adoption of EMRs. At the sample mean, this translatesto a marginal effect of e-discovery rules lowering the likelihood of adopt-

ing EMRs by 0.011 each year. This represents a large (one-third) decrease

relative to the average propensity to adopt 0.033 each year. These

6. Double clustering at the state and hospital level produces comparable estimates that are

significant at the p< 0.01 level.


at MIT

Libraries on N

ovember 23, 2015


ownloaded from


Ta

ble

3.

Ho

wE

-dis

co

ve

ryR

ule

sA

ffe

ct

Ho

sp

ital

Ad

op

tion

of

EM

Rs

Va

ria

ble

(1)

(2)

(3)

(4)

(5)

(6)

E-d

isc

ove

ryru

le�

0.2

39

**�

0.2

36

**�

0.2

33

**�

0.2

35

**�

0.2

34

***

�0

.19

1**

(0.1

14

)(0

.11

7)

(0.1

10

)(0

.10

9)

(0.0

71

1)

(0.0

81

8)

Ye

ars

op

en

ed

(00

)0

.19

6**

*0

.19

6**

*0

.19

6**

*0

.19

6**

*0

.19

6**

*

(0.0

33

3)

(0.0

33

0)

(0.0

33

0)

(0.0

38

3)

(0.0

38

3)

Sta

ffe

db

ed

s(0

00

0)

0.8

54

0.9

13

0.9

17

0.9

17

0.9

39

(3.0

61

)(3

.07

0)

(3.0

68

)(2

.98

1)

(2.9

80

)

Ad

mis

sio

ns

(00

00

0)

0.7

31

0.7

36

0.7

18

0.7

13

0.7

26

(0.8

47

)(0

.84

5)

(0.8

31

)(0

.68

3)

(0.6

82

)

Inp

atie

nt

da

ys

(00

00

0)

0.1

11

0.1

08

0.1

07

0.1

08

0.1

04

(0.1

16

)(0

.11

7)

(0.1

18

)(0

.13

7)

(0.1

37

)

Me

dic

are

inp

atie

nt

da

ys

(00

00

0)

�0

.24

5�

0.2

45

�0

.23

7�

0.2

36

�0

.23

7

(0.2

16

)(0

.21

6)

(0.2

15

)(0

.17

5)

(0.1

75

)

Me

dic

aid

inp

atie

nt

da

ys

(00

00

00

)0

.99

91

.02

11

.02

61

.03

31

.04

6

(1.3

99

)(1

.40

0)

(1.4

01

)(1

.70

3)

(1.6

99

)

Birth

s(0

00

00

0)

�7

.84

1�

7.8

61

�7

.63

4�

7.6

71

�7

.66

9

(20

.69

)(2

0.6

7)

(20

.61

)(1

5.7

1)

(15

.70

)

To

tal

inp

atie

nt

op

era

tion

s(0

00

0)

�0

.12

9�

0.1

32

�0

.13

3�

0.1

32

�0

.13

3

(0.1

26

)(0

.12

6)

(0.1

25

)(0

.11

6)

(0.1

16

)

To

tal

op

era

tion

s(0

00

00

)0

.35

30

.35

00

.34

80

.34

50

.34

7

(0.4

30

)(0

.43

2)

(0.4

30

)(0

.35

6)

(0.3

55

)

Em

erg

en

cy

ou

tpa

tien

tvis

its(0

00

00

)0

.20

1*

0.2

00

*0

.20

1*

0.2

01

**0

.20

0**

(0.1

19

)(0

.11

8)

(0.1

17

)(0

.09

42

)(0

.09

41

)

To

tal

ou

tpa

tien

tvis

its(0

00

00

0)

0.1

01

0.1

02

0.1

00

0.1

01

0.1

01

(co

ntin

ue

d)


at MIT

Libraries on N

ovember 23, 2015


ownloaded from


Ta

ble

3.

Co

ntin

ue

d

Va

ria

ble

(1)

(2)

(3)

(4)

(5)

(6)

(0.1

21

)(0

.12

1)

(0.1

22

)(0

.11

1)

(0.1

11

)

To

tal

pa

yro

lle

xp

en

se

s(U

SD

b)

0.4

53

0.4

54

0.4

54

0.4

56

0.4

57

(0.8

90

)(0

.88

4)

(0.8

87

)(1

.25

9)

(1.2

58

)

Em

plo

ye

eb

en

efits

(US

Db

)5

.44

7**

5.3

32

**5

.28

2**

5.2

39

*5

.23

7*

(2.5

39

)(2

.45

9)

(2.4

60

)(2

.82

6)

(2.8

22

)

To

tal

exp

en

se

s(U

SD

b)

�0

.18

6�

0.1

76

�0

.16

6�

0.1

66

�0

.16

5

(0.5

07

)(0

.50

7)

(0.5

08

)(0

.51

6)

(0.5

15

)

Le

ng

tho

fsta

y0

.46

4**

0.4

62

**0

.46

1**

0.4

61

***

0.4

62

***

(0.1

91

)(0

.19

2)

(0.1

91

)(0

.17

0)

(0.1

70

)

No

.o

fd

oc

tors

(00

00

)0

.21

50

.23

50

.14

40

.16

40

.15

8

(2.1

23

)(2

.12

6)

(2.1

12

)(2

.59

4)

(2.5

91

)

No

.o

fn

urs

es

(00

0)

�0

.24

3**

�0

.24

2*

�0

.24

3*

�0

.24

2*

�0

.24

2*

(0.1

24

)(0

.12

5)

(0.1

25

)(0

.14

5)

(0.1

45

)

No

.o

ftr

ain

ee

s(0

00

)�

0.7

71

***

�0

.77

0**

*�

0.7

65

***

�0

.76

3**

*�

0.7

63

***

(0.2

64

)(0

.26

4)

(0.2

63

)(0

.24

6)

(0.2

45

)

No

nm

ed

ica

lsta

ff(0

00

0)

0.3

07

0.3

11

0.3

17

0.3

14

0.3

18

(0.4

72

)(0

.47

1)

(0.4

71

)(0

.50

5)

(0.5

05

)

PP

O0

.02

62

0.0

25

60

.02

67

0.0

26

30

.02

61

(0.0

41

2)

(0.0

40

9)

(0.0

40

6)

(0.0

42

8)

(0.0

42

8)

HM

O0

.00

52

70

.00

59

00

.00

57

80

.00

60

40

.00

58

7

(0.0

48

1)

(0.0

48

0)

(0.0

48

0)

(0.0

41

7)

(0.0

41

7)

Me

mb

er

ho

sp

ital

syste

m0

.12

3**

*0

.12

5**

*0

.12

4**

*0

.12

4**

*0

.12

4**

*

(0.0

37

4)

(0.0

37

3)

(0.0

37

3)

(0.0

28

7)

(0.0

28

7)

(co

ntin

ue

d)


at MIT

Libraries on N

ovember 23, 2015


ownloaded from


Ta

ble

3.

Co

ntin

ue

d

Va

ria

ble

(1)

(2)

(3)

(4)

(5)

(6)

Sp

ec

ialit

yh

osp

ital

0.1

55

**0

.15

5**

0.1

56

**0

.15

6**

0.1

55

**

(0.0

78

3)

(0.0

78

4)

(0.0

78

7)

(0.0

67

7)

(0.0

67

7)

No

np

rofit

0.3

12

***

0.3

12

***

0.3

12

***

0.3

12

***

0.3

12

***

(0.0

40

8)

(0.0

40

9)

(0.0

40

9)

(0.0

33

0)

(0.0

33

0)

Gro

ss

sta

tep

rod

uc

t(U

SD

m)

�0

.24

9�

0.2

57

�0

.33

0*

�0

.23

9�

0.2

25

(0.2

03

)(0

.19

2)

(0.1

81

)(0

.24

0)

(0.2

41

)

Gro

ss

sta

tep

rod

uc

tp

er

ca

pita

(US

D0

00

0)

0.2

09

0.1

79

0.1

98

0.1

85

0.1

88

(0.2

44

)(0

.23

5)

(0.2

29

)(0

.15

2)

(0.1

52

)

Ca

pp

un

itive

�0

.08

94

�0

.08

42

�0

.08

63

�0

.08

62

(0.0

96

3)

(0.0

93

6)

(0.0

67

9)

(0.0

68

0)

Ca

pn

on

ec

on

om

ic�

0.0

24

5�

0.0

27

8�

0.0

20

6�

0.0

18

3

(0.0

80

4)

(0.0

75

1)

(0.0

52

5)

(0.0

52

4)

Jo

int+

se

ve

ral

liab

ility

�0

.05

64

�0

.05

68

�0

.04

76

�0

.05

77

(0.0

59

2)

(0.0

58

4)

(0.0

80

5)

(0.0

81

2)

Co

ntin

ge

nc

yfe

e0

.01

03

0.0

11

50

.00

93

40

.00

49

6

(0.1

40

)(0

.14

5)

(0.1

64

)(0

.16

5)

La

gg

ed

insta

lled

ba

se

0.0

07

69

*0

.00

78

9**

0.0

07

68

**

(0.0

04

04

)(0

.00

36

3)

(0.0

03

62

)

Sta

tee

ffe

cts

Ye

sY

es

Ye

sY

es

Ye

sY

es

Ye

ar

effe

cts

Ye

sY

es

Ye

sY

es

Ye

sY

es

Priva

cy

law

sN

oN

oN

oN

oY

es

Ye

s

Lin

ea

rtr

en

dN

oN

oN

oN

oN

oY

es

No

.o

fo

bse

rva

tion

s4

2,1

06

42

,10

64

2,1

06

42

,10

64

2,1

06

42

,10

6

Lo

glik

elih

oo

d�

57

66

.2�

55

75

.3�

55

73

.9�

55

71

.9�

55

71

.6�

55

71

.1

SE

sc

luste

red

at

the

sta

tele

ve

l.*p<

0.1

0,

**p<

0.0

5,

***p<

0.0

1.


at MIT

Libraries on N

ovember 23, 2015


ownloaded from


estimates are identified from within-state variation in adoption ratesaround the time that state e-discovery rules are put in place.

In Column (2), we add a full set of hospital- and state-level controlvariables to capture differences in hospital characteristics over time.This means that we can control explicitly for changes in hospital charac-teristics that may be correlated with the rules. For example, if states thatwere more likely to enact an e-discovery rule also had hospitals that werelikely to adopt earlier, because they were growing, then the potential foronly small hospitals to have not adopted after the enactment of the rulecould give rise to a spurious negative correlation between the enactment ofthe rule and adoption in the state. However, adding all of these controlvariables leaves the key effect of the e-discovery rule largely unchanged.

Many of the coefficients on the hospital-level control variables are in-significant. Being a speciality hospital, a nonprofit hospital, a member of ahospital system, having higher employee benefits, having longer averagepatient length of stays, more emergency room visits, and relatively fewertrainees and nurses are all linked with hospitals that are more likely toadopt EMR.

Column (3) of Table 3 reports the results of a specification where wecontrol directly for changes in other laws that may affect medical mal-practice costs. These changes may be problematic for the interpretation ofthe results in earlier columns, if they are correlated with the enactment ofe-discovery. Our data on state-level tort reforms are based on an updatedversion of Avraham (2006), which documents changes in state-level regu-lations that affected caps on medical malpractice payouts, the use of con-tingency fees, and the allocation of liability. We include separate controlvariables for limits on punitive and economic damages.

The addition of tort reform has an insignificant impact on the estimatedcoefficient for e-discovery rules. The e-discovery coefficient is �0.233. Themarginal effect at the sample mean is �0.011, or a one-third decrease inthe propensity to adopt, as in Columns 1 and 2. The reforms themselvesare generally unrelated to adoption of EMRs.

Column (4) of Table 3 reports the results of a specification in which wecontrol for differences in the level of the installed base of EMR in theprevious year for that state. This directly addresses the concern that ourresults are an artifact of states that pass these rules having many earlyadopters, meaning that after the rule is passed, there are few hospitals leftwho actually would benefit from adopting EMRs. Although the laggedinstalled-base variable itself has a positive and significant relationship withEMR adoption, its inclusion has a negligible impact on the main estimateof interest.

Two additional robustness checks are reported in the last two Columnsof Table 3. Miller and Tucker (2009) document the importance of privacyregulation as a driver of EMRs adoption. Column (5) shows that control-ling for the presence of state-level privacy laws limiting the disclosure ofpersonal medical information by hospitals does not alter the estimated


at MIT

Libraries on N

ovember 23, 2015


ownloaded from


impact of e-discovery. Column (6) shows that the main effect also remainsnegative and significant after allowing for separate linear time trends inEMR adoption for states that pass e-discovery rules during the sampleperiod.

4.1 Timing of Rules

To address the possibility that the enactment of e-discovery rules iscorrelated with unobserved trends in technology adoption in that state,in Table 4, we report results from a falsification exercise in which we use a“false” adoption date for state e-discovery rules. For each state with ane-discovery law, we create three “false” e-discovery rules passed 1, 2, and 3years before the actual enactment date. The idea of doing such a falsifi-cation test is that if our estimates are reflecting some change in the timetrend of adoption behavior in the states that enact rules but that is notrelated to the actual rule, then the placebo will pick up some of this timetrend. For easy comparison, we report the estimate from our main model(Column (4) of Table 3) in Column (1). In Column (2), we report estimatesfrom a model that includes the true e-discovery adoption date, as well asthe false dates in each of the three preceding years. The lack of a relation-ship with the false law suggests that the changes in adoption followede-discovery rules rather than preceding them.7

Column (3) of Table 4 builds on these estimates to investigate the dy-namic impact of e-discovery rules on EMR adoption. The main finding ofColumn (2) is repeated in that there are no significant pre-trends leadingup to the law. There are also no significant differential trends followinge-discovery rules either. The lack of any significant coefficient for rulesstarting 1, 2, or 3 years after the actual enactment date suggests that thelonger term effects of the rule on adoption rates resemble the immediateeffects. This implies that the total impact on adoption accumulates overtime, but the annual reduction shows no systematic pattern of increasingor decreasing magnitude. Column (4) provides additional evidence for theconstancy of the incremental impact by reporting the estimated impact ofthe e-discovery rules on a sample that excludes data from Illinois andTexas, the two states with e-discovery rules in place by 2000.8

4.2 Different Technologies

Table 5 investigates how e-discovery rules affects different types of tech-nology. Column (1) repeats the previously reported main estimate for

7. In other estimation, when we measure separate impacts of each of the false rules in

isolation, we obtain negative point estimates that are both smaller than the main estimate

(ranging from �0.0395 to �0.0562) and statistically insignificant.

8. The immediate negative impact of the e-discovery rule on adoption is consistent with

anecdotal evidence from changes in medical malpractice insurance premiums for physicians

in Illinois around the time its rule was enacted, as reported in the Medical Liability Monitor.

Premiums increased for one insurer by 28–40% (depending on specialty and region) over the 2

years that ended after enactment, but by only 8–12% over the following 2 years.


at MIT

Libraries on N

ovember 23, 2015


ownloaded from


adoption of EMRs. Column (2) shows a similar negative impact ofe-discovery rules on the adoption of a related technology, ComputerizedPhysician Order Entry (CPOE) systems. These systems can record anddisseminate instructions for patient treatments and often include safetyfeatures, such as error checking and clinical decision support tools (such asautomated warnings for potential drug interactions). As with EMRs, theincreased record-keeping associated with CPOEs has been identified as apotential risk in malpractice cases where e-discovery is present (Greenbergand Ridgely 2011). The reduction in CPOE adoption provides further

Table 4. Dynamic Effects and Timing of Rules

Variable Main Placebo

rules

Placebo

rules

Exclude

early rules

(1) (2) (3) (4)

False e-discovery rule: 3 years before �0.0242 �0.0234

(0.102) (0.101)

False e-discovery rule: 2 years before �0.0206 �0.0206

(0.144) (0.145)

False e-discovery rule: 1 year before 0.124 0.124

(0.112) (0.112)

E-discovery rule �0.235** �0.301** �0.313*** �0.273*

(0.109) (0.131) (0.112) (0.162)

False e-discovery rule: 1 year after 0.0617

(0.190)

False e-discovery rule: 2 years after �0.151

(0.173)

False e-discovery rule: 3 years after 0.111

(0.0967)

Observations 42,106 42,106 42,106 36,814

Log likelihood �5571.9 �5570.9 �5570.5 �4935.9

All estimates in this table are from models that include the full set of state and year fixed effects as well as the full

set of hospital control variables (years opened, staffed beds, admissions, inpatient days, medicare inpatient days,

Medicaid inpatient days, births, total inpatient operations, total operations, emergency outpatient visits, total outpa-

tient visits, total payroll expenses, employee benefits, total expenses, length of stay, number of doctors, number of

nurses, number of trainees, nonmedical staff, PPO, HMO, hospital system member, specialty hospital, and non-

profit), state control variables (gross state product and gross state product per capita), lagged installed base of

EMRs, and tort law control variables (punitive cap, noneconomic damages cap, joint and several liability, and

contingency fee limits). SEs clustered at the state level. *p< 0.10, **p< 0.05, ***p< 0.01.

Table 5. Effects on Different Hospital Information Technologies

Variable EMR CPOE Financial data

warehouse

Business

intelligence

(1) (2) (3) (4)

E-discovery rule �0.235** �0.258** 0.150 0.191

(0.109) (0.119) (0.159) (0.196)

Observations 42106 46252 46180 47252

Log likelihood �5571.9 �4267.7 �2024.1 �2231.9


set of hospital, state, and tort law control variables (listed in Table 4), and lagged installed base. SEs clustered at

the state level. **p< 0.05.


at MIT

Libraries on N

ovember 23, 2015


ownloaded from


evidence of a decline in clinical IT adoption following the enactment ofe-discovery rules.

In contrast, in the spirit of our previous falsification tests, in the nexttwo columns of the table, we report estimates from different placebo testusing alternative technologies that are unlikely to affect pretrial discovery:Financial Data Management Software and Business IntelligenceSoftware. These technologies, while used in hospital record-keeping, donot produce data that is part of a legal medical record. We found nosignificant effects from state e-discovery rules on these technologiesand the point estimates are positive. This suggests that the main resultin Table 3 is not being driven by a spurious negative correlation betweene-discovery rules and healthcare IT adoption more generally.

5. Heterogeneous Effects of E-Discovery Rules

5.1 E-Discovery and Different Types of Malpractice Cases

This section explores how the effect of the e-discovery rule is mediated bythe size and nature of the medical malpractice payments associated withpractitioners working in that state. Our hypothesis is that higher malprac-tice payments by practitioners translate into greater financial risk to hos-pitals from malpractice, either because hospitals themselves face increasedfinancial exposure in medical malpractice suits or because hospitals com-pete for physicians and would need to compensate them for increasingtheir exposure to electronic data in malpractice cases. We use data fromthe national practitioner databank of all medical malpractice payments.9

The files are the universe of all claims paid in the United States, but do notinclude information on complaints and litigation that did not result in apayment.10 We use payments from the previous year to predict adoption,avoiding the potential reverse causality that could otherwise run fromadoption of EMRs to malpractice payments.

Using the three-digit allegation claim category code, we determinewhether each payment is potentially related or unrelated to EMRs.Claims related to EMRs are from mistakes that might be prevented ordocumented by EMRs. An example of a claim that could be theoreticallysupported by electronic metadata in an EMR is a “failure to monitor” apatient sufficiently. An example of a claim that could be theoreticallyprevented by EMRs is a claim of a “wrong dosage” being administeredby a nurse, since EMRs theoretically remove the uncertainties introducedby a physician’s handwriting and idiosyncratic use of unit abbreviations.Other claims in the first category include a “failure to diagnose,” whereeasy access to a patient’s previous medical history may make diagnosis

9. Downloaded from http://www.npdb-hipdb.hrsa.gov/.

10. The practitioner databank is the most comprehensive source of malpractice payments,

with full coverage of practitioners and inclusion of both settlements and verdicts. The Jury

Verdict Research data exclude settlements and the Physician Insurer Association of America

Data Sharing Project contains only �12% of the claims.


at MIT

Libraries on N

ovember 23, 2015


ownloaded from

http://www.npdb-hipdb.hrsa.gov/


easier, but a failure to use the history would also be documented and could

be used in the court. The second category of claims are those that are not

likely to be affected by EMRs. An example is “Surgical or Other Foreign

Body Retained.” It is unlikely that the presence of an EMRs would affect

the likelihood of a surgeon leaving behind a piece of operating equipment

within a patient, and if the surgeon is unaware he or she has done so, it

would not be in the EMR.The results in Table 6 show how the presence of an e-discovery rule is

mediated by the average payment in that state for a medical case for each

of these claims classes. To ensure comparability, we use a state-specific

mean standardized and centered measure of the average payment data.11

The average size of claims that are associated with practices that might be

related to EMRs has a statistically significant negative interaction with the

presence of a law. The sizes of the estimates imply that increasing the value

of EMR-related claims from its mean to a value 1 SD above the mean

increases the e-discovery coefficient by�50% (from�0.214 to�0.325). In

contrast, unrelated claims have statistically insignificant effects on the

estimated impact of an e-discovery law. This suggests that when a hospital

is in a state where there are large medical malpractice payouts for lawsuits

that would be documented by EMRs, a rule facilitating e-discovery would

be incrementally negatively correlated with adoption. These separate ef-

fects are confirmed in Column (3) where a single model is estimated with

both types of payouts and interactions. As pointed out by Ai and Norton

(2003), care is needed when evaluating the significance of interaction terms

in nonlinear models. Results from a linear probability model, however, are

reassuringly similar.Column (1) of Table 6 reports both a positive correlation between mal-

practice payments and adoption of EMRs in that Column and a negative

estimate of the effect of the interaction between e-discovery rules and

malpractice payments. Though this is just a correlation that may be sub-

ject to the usual confounds, one interpretation is that having an EMR

system in place can be an advantage for hospitals in documenting their

compliance with standard practices. However, this benefit is eliminated

when e-discovery rules put all electronic information by default in the

hands of the plaintiffs.In Columns (2) and (3) of Table 6, the interaction terms between

e-discovery rules and other types of malpractice payments are statistically

insignificant. However, consistent with the relationship in Column (1),

there is a positive relationship between EMR adoption and malpractice

cases stemming from allegations that might have been prevented by EMRs

in Column (3). For unrelated allegations, there is no such association.

11. We created the re-scaled variables by subtracting out the average payment amount in

the state over the entire sample period and then dividing by the state-specific standard devi-

ation across all payments in the sample period.


at MIT

Libraries on N

ovember 23, 2015


ownloaded from


5.2 Which Hospitals Are Affected by E-Discovery Rules?

We now consider how hospital characteristics may affect the correlations

between hospital EMRs adoption and the presence of e-discovery laws.

We find evidence that the most statistically significant moderator along

various dimensions is hospital size. Table 7 provides a summary of our

results. The three columns summarize specifications that include inter-

actions with an indicator variable that measures whether or not the hos-

pital has a below median number of total admissions, expenses, and

nonmedical staff. In all cases, we find a negative interaction with whether

a hospital is small by that measure and whether a state has an e-discovery

rule in place. The largest interaction by size is the interaction with the size

of the support staff. To control for the Ai and Norton (2003) critique, we

also estimated a linear probability model with similar results.We speculate that this result may reflect the costs and difficulties asso-

ciated with preparing electronic data for discovery for civil trial. There is

anecdotal evidence that some hospitals with EMRs attempt to gain con-

trol over this risk by engaging in costly activities, such as retaining dupli-

cate paper records in order to provide supplementary evidence in the case

that timestamps are unreliable, or purchasing expensive customized soft-

ware modules to limit the content included in the legal medical record.

Hospitals that do not employ large enough legal and IT teams may be

therefore placed at a relative disadvantage, if EMRs are brought into the

discovery process. Losey (2008) urges institutions that face substantial

e-discovery risks to establish an internal e-discovery preparedness and

response team, consisting of one or more outside attorneys who specialize

in e-discovery as well as representatives from the legal team, IT depart-

ment, key business departments as well as records and compliance units.

Such costly and complex organizational requirements may deter smaller

hospitals more. In addition, hospitals that face the prospect of e-discovery

Table 6. Different Types of Medical Malpractice Claims Mediate the Effect of

E-discovery Rules on Hospital Adoption of EMRs

Variable (1) (2) (3)

E-discovery rule �0.214** �0.200** �0.208**

E-discovery rule*EMR-related

malpractice payouts

�0.111*** �0.102*

EMR-related malpractice payouts 0.0458** 0.0489**

E-discovery rule *EMR-unrelated

malpractice payouts

�0.0777 �0.0120

EMR-unrelated malpractice payouts �0.00987 �0.0219

Observations 42106 42106 42106

Log likelihood �5567.9 �5570.7 �5567.3



the state level. *p< 0.10, **p< 0.05, ***p< 0.01.


at MIT

Libraries on N

ovember 23, 2015


ownloaded from


would need to invest in additional storage and IT management systems toprevent unintentional data loss.

6. Conclusion

This article documents how the presence of state e-discovery laws affectsthe adoption of EMRs. On the one hand, it seemed possible that the use ofelectronic records might facilitate a hospital’s defense, by providing abroader and more robust standard of documentation. On the otherhand, it seemed possible that the increase in the breadth of evidence andthe possibility of “data-mining” by the plaintiff’s lawyers might increasehospitals’ perception of the potential for negative consequences of elec-tronic medical data being used in the court.

Our empirical analysis suggests that state rules that clarify the use ofelectronic evidence in discovery are associated with a one-third decrease inadoption of EMRs by hospitals. The implication of this finding is thatthere may be previously ignored welfare effects from the risk and expectedcost of litigation on the spread of certain new technologies that storeelectronic data. Although we have focused on the adoption of healthIT, this deterrence effect may be present in other sectors of the economywhere companies make choices about converting records from paper toelectronic methods of storage.

There has been a substantial federal push to ensure widespread adop-tion of EMRs, providing financial incentives of approximately $44,000 perphysician under the 2009 HITECH Act. However, such policies have as ofyet not addressed this issue of the potential use of electronic data in mal-practice cases when designing incentives. Our research suggests that hos-pitals’ concerns about the use of electronic data in malpractice cases maylimit the effectiveness of such financial subsidies. If the efforts to promoteadoption of EMRs are to be effective, they should be coupled with efforts

Table 7. Which Hospitals’ Adoption is Affected by E-discovery Laws?

Variable (1) (2) (3)

E-discovery rule �0.153 �0.159 �0.149

(0.103) (0.107) (0.121)

E-discovery rule*low admissions �0.221***

(0.0853)

E-discovery rule*low total budget �0.196**

(0.0779)

E-discovery rule*low nonmedical staff �0.228**

(0.0925)

Observations 42,106 42,106 42,106

Log likelihood �5568.4 �5569.1 �5568.2



the state level. **p< 0.05, ***p< 0.01.


at MIT

Libraries on N

ovember 23, 2015


ownloaded from


to streamline and guide the use of electronic data in court proceedings, toreduce hospitals’ perceived costs from malpractice claims enabled byEMRs.

Several limitations are worth noting. First, we only study the effect ofe-discovery rules on the adoption of EMRs, not how they are actuallyused. It is possible that hospitals that fear litigation end up not fully usingtheir EMRs, for fear of future data mining. If this is the case, then ourestimates may understate the extent of the problem. Second, when a stateenacts an e-discovery rule, the local medical and legal press commonlypublish articles that address issues of e-discovery and the potential coststhey entail. We recognize that, therefore, our regressions measure theeffect of the enactment of an e-discovery law (and the attention that sur-rounds the enactment) as opposed to the pure causal effect of an unpub-licized change in the wording of the law.

ReferencesACTL and IAALS. 2009. Final report. Technical report, Joint Project of the American

College of Trial Lawyers Task Force on Discovery and the Institute for the

Advancement of the American Legal System http://www.actl.com/AM/Template.

cfm?Section=Home&template=/CM/ContentDisplay.cfm&ContentID=4008.

AHIMA e-HIM Work Group on Defining the Legal Health Record. 2005. “The Legal

Process and Electronic Health Records,” 76 Journal of AHIMA 96A–D.

AHIMA e-HIMWork Group on e-Discovery. 2006. “New Electronic Discovery Civil Rule,”

77 Journal of AHIMA 68A–H.

Ai, C., and E. C. Norton. 2003. “Interaction Terms in Logit and Probit Models,” 80

Economics Letters 123–9.

Allison, P. D. 1982. “Discrete-time Methods for the Analysis of Event Histories,” 13

Sociological Methodology 61–98.

Angst, C., R. Agarwal, and V. Sambamurthy. 2008. Propensity, Susceptibility, Infectiousness,

and Proximity: Predicting Organizational Diffusion of Electronic Medical Records.

Mimeo, Maryland University.

Avraham, R. 2006. Database of State Tort Law Reforms (DSTLR 2nd). University of Texas

Law, Law and Econ Research Paper No. 184.

Avraham, R., L. S. Dafny, and M. M. Schanzenbach. 2012. “The Impact of Tort Reform on

Employer-sponsored Health Insurance Premiums,” 28 Journal of Law, Economics, and

Organization 10.1093/jleo/ewq017.

Bertrand, M., E. Duflo, and S. Mullainathan. 2004. “How Much Should we Trust

Differences-in-differences Estimates?,” 119 Quarterly Journal of Economics 249–75.

Blumenthal, D., C. DesRoches, and K. Donelan. 2006.Health Information Technology in the

United States: the Information Base for Progress. Princeton, NJ: Robert Wood Johnson

Foundation.

Currie, J., and W. B. MacLeod. 2008. “First Do No Harm? Tort Reform and Birth

Outcomes,” 123 Quarterly Journal of Economics 795–830.

Devaraj, S., and R. Kohli. 2000. “Information Technology Payoff in the Healthcare

Industry: A Longitudinal Study,” 16 Journal of Management Information Systems

241–68.

———. 2003. “Performance Impacts of Information Technology: Is Actual Usage the

Missing Link?,” 49 Management Science 273–89.

Dimick, C. 2007. “E-discovery: Preparing for the Coming Rise in Electronic Discovery

Requests,” 78 Journal of AHIMA 24–9.


at MIT

Libraries on N

ovember 23, 2015


ownloaded from

http://www.actl.com/AM/Template.cfm?Section=Home&template=/CM/ContentDisplay.cfm&ContentID=4008

http://www.actl.com/AM/Template.cfm?Section=Home&template=/CM/ContentDisplay.cfm&ContentID=4008


Dubay, L., R. Kaestner, and T. Waidmann. 1999. “The Impact of Malpractice Fears on

Cesarean Section Rates,” 18 Journal of Health Economics 491–522.

Feldman, J. M. 2004. “Do Anesthesia Information Systems Increase Malpractice Exposure?

Results of a Survey,” 99 Anesthesia & Analgesia 840–3.

Greenberg, M., andM. S. Ridgely. 2011. “Clinical Decision Support andMalpractice Risk,”

306 JAMA: The Journal of the American Medical Association 90–1.

Kessler, D., and M. McClellan. 1996. “Do Doctors Practice Defensive Medicine?,” 111 The

Quarterly Journal of Economics 353–90.

Kim, K. K., and J. E. Michelman. 1990. “An Examination of Factors for the Strategic Use of

Information Systems in the Healthcare Industry,” 14 MIS Quarterly 201–15.

Lane, P. E. 2005. “AIS Current Role in Anesthesiology Risk Management Remains

Uncertain,” 100 Anesthesia & Analgesia 1537–1538.

Losey, R. 2008. E-Discovery: Trends and Cases. Chicago, IL: ABA.

Matsa, D. A. 2007. “DoesMalpractice Liability Keep the Doctor Away? Evidence from Tort

Reform Damage Caps,” 36 The Journal of Legal Studies S143–82.

Menon, N. M., B. Lee, and L. Eldenburg. 2000. “Productivity of Information Systems in the

Healthcare Industry,” 11 Information Systems Research 83–92.

Miller, A. R., and C. Tucker. 2009. “Privacy Protection and Technology Diffusion: The Case

of Electronic Medical Records,” 55 Management Science 1077–93.

———. 2011a. “Can Health Care Information Technology Save Babies?,” 119 Journal of

Political Economy 289–324.

———. 2011b. “Encryption and the Loss of Patient Data,” 30 Journal of Policy Analysis and

Management 534–56.

Terry, N. P. 2001. “An eHealth Diptych: The Impact of Privacy Regulation on Medical

Error and Malpractice Litigation,” 27 American Journal of Law and Medicine .

Vigoda, M. M., and D. A. Lubarsky. 2006. “Failure to Recognize Loss of Incoming Data in

an Anesthesia Record-Keeping System May Have Increased Medical Liability,” 102

Anesthesia & Analgesia 1798–802.

Virapongse, A., D. W. Bates, P. Shi, C. A. Jenter, L. A. Volk, K. Kleinman, L. Sato, and S.

R. Simon. 2008. “Electronic Health Records and Malpractice Claims in Office Practice,”

168 Archives of Internal Medicine 2362–7.

Williams, L. 2009. “Routine Complication from Surgery Turns Fatal,” San Francisco

Chronicle, San Francisco.


at MIT

Libraries on N

ovember 23, 2015


ownloaded from


Appendix A

Table A1. Summary Statistics in 1994 by E-Discovery Rule in 2007

No rule Rule

Mean (SD) Mean (SD)

Adopt EMR 0.011 (0.10) 0.017(0.13)

E-discovery rule 0 (0) 0 (0)

Years opened 35.2 (34.6) 37.2 (36.7)

Staffed beds 189.1 (168.3) 206.4 (198.2)

Admissions (000) 6.89 (6.89) 7.07 (7.49)

Inpatient days (000) 42.5 (44.5) 50.1 (58.0)

Medicare inpatient days (000) 19.3 (20.1) 22.3 (24.8)

Medicaid inpatient days (000) 7.99 (13.3) 10.7 (20.2)

Births 875.9 (1205.0) 920.3(1226.6)

Total inpatient operations (000) 2.24 (2.67) 2.28 (2.82)

Total operations (000) 5.29 (5.86) 5.09 (5.58)

Emergency outpatient visits (000) 20.5 (18.8) 20.5 (18.9)

Total outpatient visits (000) 83.7 (100.5) 86.8 (110.5)

Total payroll expenses (USDm) 25.8 (32.6) 29.4 (42.5)

Employee benefits (USDm) 5.83 (8.00) 6.55 (10.1)

Total expenses (USDm) 60.0 (78.6) 64.6 (91.4)

Length of stay 1.01 (0.086) 1.01 (0.100)

No. of doctors 10.2 (42.3) 13.7 (51.3)

No. of nurses 198.1 (234.3) 205.4 (256.4)

No. of trainees 16.1 (70.3) 22.7 (87.1)

Nonmedical staff 559.7 (642.9) 612.7 (783.0)

PPO 0.57 (0.50) 0.59 (0.49)

HMO 0.45 (0.50) 0.49 (0.50)

Member hospital system 0.40 (0.49) 0.36 (0.48)

Speciality hospital 0.030 (0.17) 0.046 (0.21)

Nonprofit 0.63 (0.48) 0.61 (0.49)

Gross state product (USDtr) 0.27 (0.27) 0.34 (0.21)

Gross state product per capita (USD000) 28.1 (2.92) 30.4 (4.19)

EMR-related malpractice payouts (USDm) 0.18 (0.065) 0.20 (0.057)

EMR-unrelated malpractice payouts (USDm) 0.12 (0.049) 0.16 (0.052)

Cap punitive 0.35 (0.48) 0.52 (0.50)

Cap noneconomic 0.37 (0.48) 0.042 (0.20)

Joint + several liability 0.64 (0.48) 0.92 (0.28)

Contingency fee 0.40 (0.49) 0.40 (0.49)

No. of observations 1092 2303

The sample for this table excludes the 204 hospitals (of the 3599 hospitals in our analytical sample) that appear only

in AHA surveys after 1994.


at MIT

Libraries on N

ovember 23, 2015


ownloaded from


Table A2. Robustness to Alternative Estimation Models

Linear probability model Logit model Cox proportional

hazards model

(1) (2) (3)

E-discovery rule �0.0157*** �0.526*** �0.498**

(0.00457) (0.169) (0.254)

Observations 42,106 42,106 42,106


set of hospital, state, and tort law control variables (listed in Table 4), and lagged installed base. SEs in parenth-

eses. *p<0.10, **p< 0.05, ***p< 0.01.


at MIT

Libraries on N

ovember 23, 2015


ownloaded from
