Download Does Medical Malpractice Deter? The Impact of Tort Reforms... Malpractice Standard Reforms on Healthcare Quality

Does Medical Malpractice Deter? The Impact of Tort Reforms and
Malpractice Standard Reforms on Healthcare Quality
By MICHAEL FRAKES* Despite the fundamental role of deterrence in the theoretical justification for
medical malpractice law, surprisingly little evidence has been put forth to date
bearing on its existence and scope. Using data from the 1979 to 2005 National
Hospital Discharge Surveys and drawing on an extensive set of variations in
various tort measures (e.g., damage caps) and malpractice standard-of-care rules
(Frakes 2012a), I estimate a small and statistically insignificant relationship
between malpractice forces and two metrics of health care quality emphasized by
the Agency for Healthcare Research and Quality: (1) avoidable hospitalization
rates (reflective of outpatient quality) and (2) inpatient mortality rates for
selected medical conditions. At most, the evidence implies an arguably modest
degree of malpractice-induced deterrence. For instance, at one end of the 95%
confidence interval, the lack of a non-economic damages cap (indicative of higher
malpractice pressure) is associated with only a 4% decrease in avoidable
hospitalizations.
JEL Classification: I18, K13 Keywords: malpractice; healthcare quality; deterrence; defensive medicine *
Cornell Law School, 314 Myron Taylor Hall, Ithaca, NY 14853 (email: [email protected]). I am grateful to Amy Finkelstein and Jon Gruber
for their guidance and advice and to Anupam Bapu Jena for helpful comments. I am also thankful to Negasi Beyene, Maria Owings and the staff
at the Research Data Center at the National Center for Health Statistics for their help with the National Hospital Discharge Survey files and to
Ronen Avraham for providing data on state tort laws. Funding from the National Institute on Aging, Grant Number T32-AG00186, is gratefully
acknowledged. 1 Electronic copy available at: http://ssrn.com/abstract=2090595
Among the key rationales typically invoked to support a system of medical malpractice
law is the notion that fear over medical liability may incentivize a physician to provide a level of
quality that she would not have otherwise provided absent the law. Scholarship to date,
however, has provided surprisingly little evidence to support the existence of any such
“deterrence” effect. Rather, the empirical malpractice literature has somewhat deemphasized
considerations of medical quality and medical errors and has paid significantly more attention to
the relationship between malpractice pressure and measures of health care utilization and costs.
The findings from this cost-focused literature, however, are simply not dispositive as to the
question of whether malpractice law serves as an effective deterrent (in the proper sense of the
word). In order to fully understand whether the U.S. medical malpractice system is achieving
one of its stated, fundamental goals, it is paramount that malpractice researchers give
independent evaluation to issues of health care quality.
Related to the question of whether malpractice law deters is the question of whether the
threat of medical liability induces physicians to act “defensively.” To be sure, these are not
synonymous concepts. While deterrence is meant to identify situations in which liability causes
physicians to behave optimally, defensive medicine is meant to capture the opposite. That is,
defensive medicine can be seen as deterrence gone awry, whereby liability fears cause physicians
to provide sub- or supra-optimal levels of care, with the emphasis often placed on the latter.
Those cost-focused studies alluded to above are largely concerned with probing the existence
and scope of defensive practices. However, to properly label a response as defensive requires
more than a mere understanding of whether liability encourages additional utilization of medical
care. Since a defensive response is defined with reference to an “optimal” level of treatment, this
assessment requires a determination as to whether or not any malpractice-induced expansion in
treatment is accompanied by corresponding improvements in quality or outcomes.
As such, whether the goal is to make an independent evaluation of the deterrent impact of
medical liability or to distinguish between a deterrent and a defensive response to liability, it is
necessary to estimate the impact of the malpractice system on medical errors and healthcare
quality. Evidence shows that medical errors continue to occur at potentially startling rates and
that prevailing indicators of quality are at undesirable levels, indirectly suggesting that liability
2 Electronic copy available at: http://ssrn.com/abstract=2090595
forces have not driven medical quality to its ideal levels. However, it is unclear how much
worse the state of affairs would have been absent the present system of medical liability.
While the results are somewhat mixed, the literature has generally failed to provide
systematic support for any contention that the existing medical liability system is effective in
improving health care quality. Further research, of course, is needed on this fundamental
question, though such research is confounded by a number of methodological difficulties,
including (1) the need to capture outcome variables that are more highly reflective of the quality
of healthcare received, as opposed to being more highly reflective of social, environmental and
other determinants of health and (2) the need to capture enough variation in the structure or
scope of the liability system, in order to tease out its relationship with medical outcomes.
In this paper, I estimate the relationship between medical liability and health care quality
using a data source – i.e., the National Hospital Discharge Survey (NHDS) – that is well suited to
confront these two methodological difficulties. First, the data covers a long enough period of
time – i.e., 1979 to 2005 – and a broad enough span of jurisdictions to capture a rich source of
variation in relevant malpractice and tort laws. Moreover, the NHDS records accommodate
construction of broad-based metrics promoted by the associated medical literature (and by
various agencies) as being indicative of the quality of care received during a medical encounter
itself. For these latter purposes, I consider two fundamental domains of health care quality
emphasized by the Agency for Healthcare Research and Quality: (1) avoidable hospitalizations,
reflecting the level of care provided by the associated outpatient environment and (2) inpatient
mortality rates for selected medical conditions (e.g., acute myocardial infarctions, hip fractures
and strokes, among others), reflecting the level of care provided in inpatient care.
The results of this empirical exercise generally cast doubt upon the role that current
medical liability rules play in inducing the provision of quality medical care. For instance, the
estimated relationship between avoidable hospitalization rates and malpractice pressure, as
identified by the adoption of non-economic damage caps and related tort reforms, is both
statistically insignificant and small in magnitude, with a 95% confidence interval that is
relatively tightly bound around zero. At one end of this interval, the lack of a non-economic
damages cap (which is indicative of higher malpractice pressure) is associated with only a 4%
decrease in avoidable hospitalizations. That is, at the most, the evidence implies an arguably
3 Electronic copy available at: http://ssrn.com/abstract=2090595
modest degree of deterrence (with respect to outpatient care). I derive similar findings in
exploring the impact of liability pressure on inpatient mortality rates for select conditions.
In my primary empirical approach, I investigate the malpractice-quality link by exploring
the impact of legal reforms, such as damage caps, that arguably blunt the impact of the liability
system without effectively changing its structure. As an alternative approach, following Frakes
(2012a), I consider the impact of a more structural reform – i.e., the shift from a local- to a
national-standard rule – that modifies the manner in which standards of care are determined in
medical malpractice actions and that potentially alters the law’s expectations as to how
physicians should conduct their clinical practices. The results from this alternative approach are
likewise small and statistically insignificant, suggesting, at most, a relatively modest impact of
such reforms (in the predicted direction) on the geographical patterns of avoidable
hospitalization rates and inpatient mortality rates for selected medical conditions. These results
complement the small findings from the damages cap analysis in further demonstrating a weak
connection between the law, as currently applied, and observed patterns of healthcare quality.
The paper proceeds as follows. In Section I, I discuss the indirect and direct sources of
evidence that have been offered to date regarding the deterrent impact of medical liability.
Section I likewise presents a simple model of physician decisionmaking and discusses the
various sources of ambiguity that cloud the channel of deterrence intended to be created by the
malpractice system. Section II discusses the data and empirical methodology. Section III
presents the results of the empirical deterrence analysis. Finally, Section IV concludes.
I.
Malpractice Law and Physician Behavior: Background
A. Indirect evidence of deterrence: ongoing medical errors
To explore the role of the current liability system in incentivizing higher quality care, it is
perhaps best to begin by evaluating the present healthcare landscape of the United States. Just
what level of quality are physicians currently delivering? In its influential 2000 report, To Err is
Human, the Institute of Medicine (IOM) captured remarkable attention by citing the results
reported by several studies that found that between 2.9 and 3.7 percent of hospitalizations
resulted in an adverse event and that between 6.6 and 13.6 percent of those adverse events led to
death. These results suggest that as many as 44,000 to 98,000 Americans die each year as a
4 result of medical errors, surpassing the number of Americans that die annually in motor vehicle
accidents. Considering the report’s focus on inpatient care and on mortality-related outcomes,
the full extent of the harm caused by medical errors may be even greater.
The 2000 IOM Report ushered in a wave of policy initiatives and academic studies
focused on healthcare quality, as distinct from costs and utilization patterns. Consistent with the
implications of the 2000 IOM Report, the subsequent literature painted a rather bleak picture of
the quality of care being delivered within the U.S. healthcare system. For instance, reviewing
the medical records of a random sample of U.S. adults and evaluating healthcare performance
across a range of quality indicators and clinical contexts, McGlynn et al. (2003) find that
Americans may only be receiving roughly 50% of recommended care.1
Assessing delivery quality across U.S. medical providers is, of course, complicated by the
substantial uncertainties surrounding medical care. In many circumstances, there may be little
consensus regarding the practices that physicians should be following, confounding any ability to
form quality metrics to evaluate in the first place. The quality adherence studies indicated above
are particularly troubling in so far as they demonstrate rampant misuse of care even in those
arguably limited circumstances in which best practices have been identified by the experts. In
any event, with respect to those remaining circumstances, the existence of substantial medical
uncertainties themselves casts its own form of doubt upon the state of the current healthcare
environment. A substantial body of literature in health economics and medicine has documented
striking regional variations in the care provided by physicians,2 the cause of which many have
attributed, in part, to “practice-style” variations, whereby physicians in different regions develop
divergent attitudes and beliefs regarding healthcare delivery practices (Wennberg 1984). With
beliefs diverging so remarkably, some have argued that the practice of medicine has veered far
from the scientific ideal of a “single right way of doing things” (Blumstein 2002).
All told, the evidence presented by these quality-focused studies suggests that the
medical malpractice system, if it is doing any work at all to incentive quality medicine, is falling
short of its promise. Of course, what we cannot see from these indirect sources of evidence is
1
In evaluating quality, the McGlynn et al. (2003) study focused on assessing whether the care delivered to study
participants adhered to recommended medical processes as identified by the RAND Quality Assessment Tool
system (which, in turn, is based on various established national guidelines and on the relevant medical literature).
2
See, for example, Wennberg and Cooper (1999). 5 the counterfactual. That is, how much worse off would the present state of affairs be absent the
existing set of malpractice laws? In subsection D below, I discuss those studies that have
explored this counterfactual more directly. However, I first confront another important
background inquiry and discuss the empirical evidence that has been put forth regarding the
general plausibility of the malpractice deterrence channel. To facilitate that analysis, I begin by
introducing a simple model of physician decisionmaking in the shadow of liability.
B. Model of physician decisionmaking
The determinants of physician behavior are varied and complex in nature. Given the
possible sensitivity of physicians to fears over malpractice liability, it is reasonable to suspect
that one such determinant is the standard of care clinically expected of physicians under the law.
Consider a physician faced with the decision of whether or not to undertake a precaution /
quality-improving action, A, that comes at a cost of C. Assume that patients present themselves
with a complication level, s, where higher levels of s reflect more severe case mixes and where s
follows a uniform distribution across the population. The increase in the quality of care
delivered to a patient with complication level, s, arising from the taking of action A is equal to
B(s). Assume that this benefit to the patient is increasing in s and at an increasing rate – that is,
Bs > 0 and Bss > 0. This latter condition implies diminishing benefits to taking greater
precautions as physicians face patients with more and more favorable risk profiles (i.e., lower s).
Under an optimal liability rule following the spirit of the Learned Hand Formula,3 the
physician would be held liable for failing to take this precaution so long as the benefit from
taking the precaution, B(s), exceeds the cost, C. Accordingly, the law will not expect the
physician to maximize the total level of quality. Rather, it will only expect physicians to
undertake cost-justified actions – that is, to act in the face of patients with s > s*, where s* is
such that B(s*) = C. Physicians who fail to take this precaution in the case of such high-risk
patients will be subject to liability for the expected damages sustained – i.e., the forgone benefits,
B(s). Since, B(s) is necessarily greater than C throughout the range in which this liability
standard expects that the precaution be taken, a level of damages at B(s) will ensure that
physicians exposed to this potential liability take the indicated precaution.
3
United States v. Carroll Towing Co., 159 F.2d 169 (2d. Cir. 1947). 6 The private benefit to the physician for otherwise undertaking the precaution is equal to
M(s), which is initially assumed to be less than the benefit to the patient, B(s), for all s. Thus,
without liability rules, the physician will undertake precautions for all patients with s > s**,
where s** is such that M(s**) = C. Since M(s) is less than B(s), it will intersect with C at a
point, s**, that is greater than s*. Accordingly, with no liability, the physician will take fewer
overall precautions over the full set of patients and deliver an overall lower level of quality,
relative to a situation in which the physician is exposed to potential liability for failing to comply
with the above standard. In the analysis below, I test this prediction and explore whether the
threat of greater malpractice liability leads physicians to deliver a higher average level of quality.
A positive association of this nature would be consistent with a deterrent effect of
medical malpractice law. Of course, it is possible that liability pressure may go too far and cause
physicians to form a cutoff at a point lower than optimum, s*, leading to greater precaution
taking. Often termed “defensive medicine,” this latter scenario may arise due to the liability
standard itself being set below this optimum cut-off or due to a physician’s uncertainty regarding
the location of the liability standard.4 In practice, it is difficult to determine whether any
observed increase in quality arising from greater liability pressure is reflective of a move in the
direction of the optimum, s* (i.e., deterrence), or a move away from the optimum (i.e., defensive
medicine). Of course, the returns to taking additional precautions diminish as physicians act in
the face of patients with lower and lower risk profiles, s. As such, an observation of substantial
increases in quality as precaution taking rises is perhaps more likely to be consistent with an
intended deterrence impact than it is with a wasteful defensive medicine effect.
This framework conceives of a relatively broad range of actions/precautions, including
the adoption of a quality-improving process (e.g., routine hemoglobin A1c testing for diabetes
management), the exercise of additional care during a procedure or the undertaking of a
procedure itself.5 Much of the malpractice literature has focused on estimating the impact of
4
A physician uncertain regarding the location of s* may desire to err on the side of providing too many precautions
(i.e., setting a complication cut-off less than s*) considering that the net harm from doing so (i.e., the degree to
which precaution costs exceed the benefits derived) is less than the net harm from erring too far in the opposite
direction (i.e., the degree to which the forgone benefits exceed the precaution costs). This asymmetry arises from
the fact that the absolute difference between B(s) and C is generally larger to the right of s*, considering that Bss > 0. 5
In the situation where A represents a procedure which carries various risks in its execution, the costs of the action,
C, will accordingly capture both the direct costs of the procedure and its associated risks. In this instance, it may not
be the case that an increase in utilization of A leads 1-for-1 to an increase in quality.
7 liability pressure on utilization rates of particular procedures (e.g., cesarean deliveries), often
with a goal of testing for evidence of excessive or wasteful utilization. Though such studies have
an immediate focus on the utilization rates themselves (e.g., the number of applications of
precaution, A, per person), it is necessary to also investigate the associated changes in healthcare
outcomes and healthcare quality measures (i.e., changes in patient benefits, B) in order to assess
whether the observed response represents waste or desirable deterrence.
C. Indirect evidence: plausibility of deterrence channel
With a substantial amount of ambiguity surrounding this deterrence channel, there are
many reasons to doubt whether the threat of liability will indeed induce the higher level of
quality predicted by the above model. To begin, are physicians even aware of how the law
expects them to behave? That is, are physicians aware that the law requires that precautions be
taken for all s > s*?6 Moreover, even if physicians have knowledge of such standards, it is
unclear whether they would alter their practices accordingly. Deterrent forces are potentially
blunted by the fact that a very small percentage (possibly as low as 3%) of those who are harmed
by a negligent medical error actually pursue a malpractice claim in the first place (Localio et al.
1991).7 That is, while the damages to be imposed for failure to take the expected precaution in
the above model are B(s), the physician may only assign a small fraction, α, of this amount in
weighing the costs and benefits of the proposed action. In this instance, the threat of liability
may only induce the physician to take precautions for those levels of s beyond s^, where s^ is as
such that αB(s^) = C. If the physician will already be inclined to provide the precaution at this
level – i.e., , if the physician’s provide benefit, M(s) exceeds α B(s) – then, the threat of
malpractice liability will not alter the clinical decisionmaking of the physician.
Potentially confounding this deterrent channel even further is the possibility of an
ineffective targeting of damage awards towards meritorious claims. To the extent that juries
extend liability to situations in which physicians, in fact, did not negligently harm patients (i.e.,
where either the physician did indeed undertake the required precaution, A, or where the
6
In some instances, this knowledge may come through word-of-mouth among physicians following the outcomes of
relevant malpractice proceedings or it may come through communications with their insurance providers. However,
it is unclear whether this information comes at a level fine enough to influence clinical decisions. 7
When negligently harmed patients do file claims, the evidence suggests that the expected damage awards are
considerable (Studdert et al. 2006). Of course, compensatory awards are generally not scaled upwards to account
for the low percentage of claiming by those harmed by negligence. 8 complication level of the patient, s, is as such that the precaution need not have been taken), the
law may fail to signal to physicians precisely what they must do in order to avoid liability.8
Early investigations into this targeting inquiry indeed invoked some concern, with as many as 59
percent of non-meritorious claims receiving some payment.9 In a more recent study, however,
Studdert et al. (2006) find that over 70 percent of non-meritorious claims (i.e., those involving
neither negligence nor injuries) received no compensation, while over 70 percent of those claims
that involved both injury and negligence did. Moreover, when non-meritorious claims did
receive payment, those payments were substantially lower than those extended to meritorious
claims (roughly $313,000 vs. $521,000). These findings instill some hope that the system may
be sending the proper signals to physicians, though the targeting remains far from perfect.
However, even when those harmed by a negligent error are successful in seeking
compensation, physicians may face limited immediate financial risk from the associated damage
awards considering that they are insured against such losses and that such coverage is typically
not experience rated (Currie and MacLeod 2008). Moreover, claim amounts themselves rarely
exceed malpractice insurance limits (Zeiler et al. 2007). As such, potentially blunting the
incentives posed in a given clinical encounter, physicians may not even be exposed to the
damage awards of B(s) for failing to take the indicated precaution. On the other hand, despite
the limited financial risks directly associated with the litigation, physicians may face a number of
uninsurable costs as a result of malpractice liability – e.g., reputational and psychological
damage – leaving open a pathway by which physicians may respond to liability forces.10
8
Noise in this targeting process may stem from the nature of the litigation process itself. The operable standards of
care in a given malpractice action are generally determined by a jury following the presentation of relevant
testimony by litigant-selected expert witnesses. The case-by-case outcomes resulting from this “battle of the
experts” may be relatively noisy in practice. In the face of an indeterminate and uncertain standard-setting process,
physicians may be left with little guidance as to how to coordinate their prospective clinical behaviors in order to
comply with the law. 9
Studdert et al. (2006) report this figure in surveying these earlier studies. Studdert et al. contend that these early
studies suffered from various limitations, including a reliance upon the insurer’s assessment of claim validity. In
their own investigation into this question, Studdert et al. instead rely upon the assessment of independent experts,
who were assigned to review the medical records associated with 1452 closed malpractice claims.
10
Subject to certain exceptions, payments made on behalf of physicians to settle claims or to satisfy judgments must,
under federal law, be registered in the National Practitioner Data Bank (NPDB), an electronic repository which is
made available to hospitals and certain other health care entities. The NPDB was established by the Health Care
Quality Improvement Act of 1986, as amended (42 U.S.C. 11101 et seq.). This repository may reinforce any
reputational consequences of malpractice liability.
9 In any event, if malpractice fears do arise out of this pathway, they will impact a clinical
decisionmaking process that involves a range of financial and other incentives, captured by M(s),
that may already incentivize physicians to provide quality care. For instance, physicians may
seek to avoid committing errors over fear of suffering reputational damage in the marketplace.
Physicians may likewise endeavor to maximize the quality of care that they provide in order to
be in compliance with certain professional obligations and in consideration of their affirmations
to “do no harm” pursuant to the Hippocratic Oath. Physicians may even be incentivized by the
promise of higher fees to provide a greater number of medical services, which may, in some
circumstances at least, correspond to improvements in healthcare quality.11 To the extent that
M(s) equals or exceeds B(s) (or to the extent that s** is lower than s*), liability standards will
pose no incremental contribution to the levels of precaution undertaken by physicians.
All told, a heavy cloud of uncertainty engulfs this deterrence inquiry, necessitating
empirical evaluation. A recent study by Frakes (2012a) provides empirical evidence that the
abandonment of malpractice “locality rules” (which require that physicians comply with local
customs) and contemporaneous adoptions of rules requiring physicians to follow national
standards are associated with a striking amount of convergence in regional practice patterns.
This finding arguably sheds light on the general empirical relevancy of malpractice standards of
care and suggests that physicians may indeed be aware of and responsive to the standards
expected of them under the law.12 The results presented in Frakes (2012a) are encouraging in so
far as they instill faith in the potential for malpractice law to deter undesirable behaviors.
However, Frakes’ analysis was only designed to test for a physician response to a change
in liability standards. It did not comprehensively explore whether the system as a whole is
presently delivering benefits in terms of improved safety and quality. After all, even if standards
of care have “bite” under the law and encourage the maintenance of those standards, quality
medical care may only be incentivized to the extent that those standards are properly determined
11
For evidence of physician sensitivity to financial incentives, see, for example, Gruber and Owings (1996). As indicated in Frakes (2012a), in those traditional investigations into the effect of liability pressure on physician
behavior, it is unclear whether such pressure causes physicians to alter their practices in a more general, unguided
sense or whether it causes physicians to more strongly adhere to specific clinical standards expected of them under
the law. The analysis in Frakes (2012a) arguably allows for an investigation into responses of this latter variety and
focuses more directly on variations in the structure of liability rules. 12
10 in the first place.13 As such, again, it is necessary to engage in a more direct investigation into
the malpractice-quality relationship.
D. Direct evidence of deterrence
Relatively limited evidence has been put forth to date regarding the degree to which
medical malpractice deters undesirable practices.14 Whatever has been done has largely been
conducted in connection with investigations into the presence of defensive medicine or into the
supply-related impacts of malpractice law. Thus far, the evidence offered on the impact of
malpractice law on health status and health care quality has painted a rather inconsistent picture.
Some studies hitting upon this deterrence question estimate the association between
liability pressure and a health outcome statistic that is experienced by the broader population
affected by the liability regime – i.e., not simply those patients accessing the health care system
in particular ways. For instance, Lakdawalla and Seabury (2009) find that higher county-level
malpractice pressure leads to a modest decline in county-level mortality rates. Other studies
focus on slightly more targeted populations, such as the population of newborns within a region.
For instance, Klick and Stratmann investigate the impact of malpractice reforms on prevailing
infant mortality rates and generally document no observed relationship. Similarly, Frakes
(2012b), Currie and MacLeod (2008), and Dubay, Kaestner, and Waidmann (1999) each estimate
the impact of malpractice pressure on infant APGAR scores (recognized as valid predictors of
neonatal mortality),15 generally finding no evidence consistent with any such relationship.16
One limitation of the above approaches is that such broad-based health outcome measures
are likely to be driven by many factors other than the quality of care actually delivered at
13
Indeed, considering that standards of care are largely determined based on customary practices (and not as a part
of an abstract evaluation of whether the benefits of the actions taken justify the costs) and considering the many
forces in play that may cause customary practices to deviate from optimality (e.g., physician induced demand), there
may be many reasons to doubt that customary practices generate an optimal standard. The discouraging evidence
regarding delivery quality in the U.S. presented in subsection A above lends further support to the contention that
standards of care based on customary practices may not pose the proper quality-inducing incentives upon
physicians. 14
For a general review of the evidence bearing on deterrence (up to 2002), see Mello and Brennan (2002). 15
See, for example, Casey et al. (2001). 16
Various other studies similarly calculate health outcome measures based on mortality rates that are focused on more targeted populations. For instance, Kessler and McClellan (1996) estimate a trivial relationship between liability reforms and (1) survival rates during the one year period following treatment for a serious cardiac event (e.g., acute myocardial infarction), and (2) hospital readmission rates for repeated serious cardiac events over that period. Sloan and Shadle (2009) undertake a similar analysis. 11 particular outpatient and inpatient encounters (McClellan and Staiger 1999), including variations
in access to health care in the first place and variations in a range of other factors that the public
health literature has identified as being meaningful determinants of health status – e.g.,
socioeconomic status, individual risky behaviors, living conditions, social support networks, etc.
Apgar scores and infant mortality rates are arguably more connected to a particular encounter –
i.e., the delivery itself. However, delivery outcomes are significantly shaped by environmental,
physical and medical factors throughout the full term of the pregnancy and not simply the
delivery. While outcomes such as general mortality rates and infant mortality rates are
unquestionably important, the possibility of a multitude of determinants of one’s health status
raises concerns over the ability to reliably identify the link between variations in the malpractice
environment and the indicated health outcomes. Another limitation of these mortality-focused
outcomes is, of course, that they omit a broader range of morbidity-related considerations.
A key advantage of focusing the malpractice inquiry on those measures emphasized by
the health quality literature and by the AHRQ and related agencies is that the promoted measures
are, by their very design, better reflective of the influence of the delivered health care itself.
While some of these measures – e.g., avoidable hospitalization rates – may not necessarily
capture our final outcomes of interest, they nonetheless represent intermediate measures that are
linked to those undesirable outcomes and that at least avoid some of the statistical difficulties
identified above, given their tighter connection with the healthcare provision process. Moreover,
intermediate outcomes of this nature also offer the benefit of bearing upon a broader range of
negative outcomes, beyond just those focused on mortality.
Very few malpractice studies have investigated the link between malpractice law and the
type of quality metrics emphasized by the medical literature and by agencies such as the AHRQ.
In one recent study, Greenberg et al. (2010) document a positive association between adverse
events incurred during hospitalizations (measured according to the AHRQ’s Patient Safety
Indicators) and subsequent claiming of malpractice. However, rather than exploring the way in
which liability pressure may lead ex ante to improved patient safety – i.e., deterrence – they
instead focus on studying how improvements in patient safety (perhaps arising from a range of
initiatives) can alleviate ex post liability exposure for providers. One study that perhaps most
directly explores the impact of liability pressure on a relevant set of quality metrics is Currie and
12 MacLeod (2008). The authors find that damage cap adoptions increase preventable
complications of labor and delivery (a measure in the spirit of the AHRQ’s Patient Safety
Indicators), suggesting that higher liability pressure improves patient safety.
The analysis below builds upon the Currie and MacLeod (2008) study in several
important ways. First, it expands beyond the obstetric context to explore additional domains of
health care quality, including one focused on the quality of care delivered at the outpatient level.
Despite the fact that outpatient healthcare has come to represent over 20% of the nation’s total
healthcare dollars and represents one of the fastest growing components of U.S. healthcare
spending (CMS 2011), this aspect of healthcare delivery has, to my knowledge, received no
specific attention by the malpractice deterrence literature. As explained further in Section II, the
analysis below also captures a richer degree of variation in relevant tort laws than that considered
by Currie and MacLeod, resulting in arguably more reliable estimates. Moreover, as an
alternative way to investigate the link between liability and quality, I also consider the impact of
more structural liability reforms, which alter the clinical standards expected of physicians.
II.
Data and Methodology
A. Overview
I collect data on healthcare quality from the 1979 to 2005 National Hospital Discharge
Surveys (NHDS), each of which provides a nationally-representative sample of inpatient
discharge records from short-stay, non-federal hospitals. Additional details regarding the NHDS
are provided in the Appendix. Using the relevant diagnosis codes provided by the NHDS, I
calculate, for each state and year, mean levels of various healthcare quality metrics (discussed
further below). I then take two approaches in evaluating the impact of malpractice law on
healthcare quality: (1) one approach exploring the effects of fluctuations in the general level of
pressure imposed by the malpractice system (identified by damage caps and related reforms),
effectively taking as given the structure of that system and (2) another approach exploring the
effects of more structural reforms, which alter the manner in which courts determine medical
liability standards, effectively taking as given the level of pressure imposed by the system.
13 B. Malpractice Pressure / Damage Cap Analysis
I begin this deterrence analysis by estimating the relationship between the relevant
quality measures and the prevailing level of malpractice pressure associated with the given state
and year. Consistent with much of the malpractice literature, I identify variations in malpractice
pressure using adoptions of certain tort reforms, the immediate effect of which is largely to
reduce the expected levels of damages imposed in the event of liability. This effect, in turn, may
leave plaintiffs and/or their attorneys less inclined to bring suit, thereby lessening the level of
pressure placed upon physicians. Though generally fully insured, physicians may welcome such
reduced likelihoods of suit to the extent that they face non-pecuniary costs of liability. The
reforms that I emphasize in this analysis, and that have received the most attention by the
literature, are caps on non-economic damage awards (i.e., pain and suffering awards).17
Non-economic damages represent a significant portion of the typical malpractice award.
Using a dataset of 326 closed claims in Texas for the 1988-2004 period (each with at least a
$25,000 payout), Hyman et al. (2009) document an average non-economic damages award of
$681,000, compared with $542,000 for economic damages. Non-economic damage caps
represent the tort-reform measure that has been most commonly associated with an observed
change in certain malpractice outcomes: claims severity, physician supply and malpractice
premiums.18 Twenty-eight states currently have non-economic damage cap provisions in place,
most of which were adopted during the malpractice crisis of the 1980’s.19
In light of the timing of this variation, those studies relying on post-1980s data to
evaluate the impact of non-economic damage caps (e.g., Currie and MacLeod 2008) fail to draw
on the most relevant sources of variation in malpractice law and consequently rely on few
17
A large body of related literature has explored the relationship between tort reforms and various outcomes of the
malpractice marketplace: claims frequency, claims severity, insurance premiums, and physician location. See Mello
(2006) for an extensive review of this literature. These studies suggest that non-economic damage caps are perhaps
the most relevant and most influential tort reform measures (Mello 2006). 18
See Mello (2006) for a comprehensive review of relevant studies. 19
Following Frakes (2012b), I also classify states as having non-economic damages provisions if they have laws that
place caps on total damages awards. Such laws, after all, necessarily cap non-economic damages as well. In light of
the imposition of state fixed effects, this classification only has relevance in the context of 1 state (Texas) that
adopted a total damages cap at a time when it did not have a specific non-economic damage cap in place. Only 1
additional state – i.e., Colorado – adopted a total damages cap over the sample period (2 years following the
adoption of a non-economic damages cap). With this in mind, I do not separately control for the incidence of a cap
on total damages. However, I estimate nearly identical results for the remaining coefficients when I do include this
additional covariate and treat total and non-economic damage caps separately. 14 treatment groups. Limited variation of this nature implicates concerns over the reliability of the
estimated standard errors and over the consistency of the point estimates – that is, with few
treatment groups, it is less likely that spurious correlations between the outcome variable and the
reforms of interest will average to zero (Frakes 2012; Conley and Taber 2011). The NHDS data,
supplemented with geographic identifier codes, provides inpatient discharge records from a
broad enough span of states and covering a long enough period of time to allow for a deterrence
analysis that draws on an extensive set of legislative variations.
[INSERT TABLE 1 ABOUT HERE]
Table 1 lists those states that modified their non-economic damage cap laws over the
NHDS sample period. In most specifications, I also explore the association between observed
healthcare quality and certain additional types of tort reforms, including reforms of the collateral
source rule, caps on punitive-damages awards and other “indirect” tort reforms. Traditional
collateral source rules generally prohibited defendants from introducing evidence of
compensatory payments made to plaintiffs from outside sources (e.g., insurers). Thirty-three
states currently have laws in place that eliminate this traditional rule, effectively reducing the
scope of compensatory damage awards (see Table 2 for variations in these rules over time).
Much of these reforms likewise occurred during the mid-1980s; however, there are a substantial
amount of independent reforms of each type, facilitating identification of their separate impacts.
[INSERT TABLE 2 ABOUT HERE]
Punitive damages are awarded on a much rarer basis in malpractice actions than are noneconomic damages awards (without a correspondingly large increase in average payouts).20
Thus, relative to non-economic damages, it is arguable that the threat of liability for punitive
damages will have a weaker impact on physician behavior. Nonetheless, despite the infrequent
application of such awards, considering that punitive damages are generally not insured by
liability carriers, it remains reasonable to believe that physicians may be sensitive to the threat
posed by punitive awards (Malani and Reif 2012). Finally, following the classification of
malpractice reforms introduced by Kessler and McClellan (1996), I estimate the general impact
associated with a residual reform category (labeled “indirect” reforms) that includes contingency
20
For evidence of this claim, see Cohen (2005) and Hyman et al. (2009). 15 fee limitations, requirements of periodic payment of future damages, joint and several liability
reforms, and provisions for a patients’ compensation fund.21
C. Liability-Standards Analysis
Traditional malpractice law set operable standards of care by turning to the customary
behavior of physicians practicing in the same locality as the defendant, essentially expecting that
physicians follow the practices applied by those around them. Since the 1960s and 1970s, the
majority of states have amended their substantive malpractice laws to abandon such locality rules
in favor of rules requiring physicians to follow national standards of care. In light of the rampant
regional disparities in care that persist across regions (see, for example, Wennberg and Cooper
1999), one can view the move from a local to a national-standard rule as a meaningful alteration
of the standards clinically expected of physicians (Frakes 2012a).
Accordingly, as an alternative approach to exploring the relationship between malpractice
law and healthcare quality, I explore whether this modification of the standards expected of
physicians results in a corresponding change in observed quality measures. While this
investigation may not be able to compare the present state of affairs to a counterfactual in which
liability rules place no expectations upon physicians, it nonetheless provides insight into whether
the system and its standard-setting process is at least capable of impacting prevailing measures of
health quality – i.e., that healthcare quality is sensitive to this policy lever.
[INSERT TABLE 3 ABOUT HERE]
Frakes (2012a) tests the hypothesis that, upon the adoption of a national-standard rule,
physician practices in the affected regions will converge towards the practices of the rest of the
nation, focusing on utilization rates of various obstetric and cardiac treatments and diagnostic
procedures. In this analysis, I follow Frakes (2012) and test for the impact of national-standard
adoptions by estimating whether mean quality measures in a state that uses a local-standard rule
converge towards the relevant national means when the state amends its malpractice laws to
require that physicians comply with national standards of care. Table 3 provides more details on
the evolution of malpractice-standard rules. Roughly 14 states abandoned the use of local
21
The results presented below for the damage caps and collateral source rule reform coefficients are entirely robust
to inclusion of a richer set of controls for each of the individual components of Kessler and McClellan's indirect
reform category. 16 standards in favor of national standards in the post-1978 period, along with 1 additional state
(Maryland) that retreated from a previous national-standard adoption. Sixteen states currently
retain some element of locality in their standard-of-care laws.
D. Quality measures
The acknowledgement that providers are falling short on quality has propelled initiatives
designed to fill this gap (e.g., hospital report cards), along with the associated development of
quality indicators used to implement such initiatives. Organizations developing quality
indicators include the Joint Commission on Accreditation of Healthcare Organizations, the
Hospital Quality Alliance and the Leapfrog Group. Foremost among those organizations is
perhaps the Agency for Healthcare Research and Quality (AHRQ). The AHRQ measures are
particularly useful for the present study in so far as they are designed for use with administrative
inpatient databases such as the NHDS. Accordingly, for the purposes of this study, I look to the
AHRQ for guidance in selecting quality metrics to explore. More specifically, inspired by the
AHRQ’s Prevention Quality Indicators (PQIs) and Inpatient Quality Indicators (IQIs), I calculate
two composite quality metrics: (1) avoidable hospitalization rates and (2) inpatient mortality
rates for selected medical conditions. I provide more details in the Appendix regarding the
selection of these AHRQ precedents and regarding the construction of the resulting quality
metrics. Nonetheless, below, I provide a brief overview of each such metric.
Avoidable hospitalizations. The first quality measure that I employ captures the rate of
avoidable hospitalizations (AH) within each state-year cell, a measure that is inspired by the
AHRQ’s PQIs. Though constructed using inpatient data, AH rates are meant to reflect the
quality of care prevailing in the associated outpatient / ambulatory community. Such measures
identify conditions (e.g., asthma, diabetes, malignant hypertension, etc.) with respect to which
proper outpatient care would have prevented the need for hospitalization.
To calculate avoidable hospitalization rates, I first count the number of hospitalizations
within the NHDS records for each state-year cell in which a diagnosis is indicated for any of the
relevant conditions. In the preferred specification, I count hospitalizations in which the
conditions are identified in the primary diagnosis code only. To form the relevant rate, it is of
course necessary to normalize these AH counts in some manner. Following Frakes (2012a), I
17 elect to use measures internal to the NHDS records to form the relevant denominator. The three
alternative denominators that I employ, each calculated at the state-year level, are as follows: (1)
an index of hospitalizations equal to the count of admissions associated with any of the following
conditions and events: acute myocardial infarction, stroke, gastro-intestinal bleeding or hip
fracture (i.e., the “low-variations” health index), (2) the number of hospitalizations associated
with the delivery of a child, and (3) the number of acute myocardial infarction discharges. More
details regarding the rationalization behind these choices are provided in the Appendix.
Primarily, these normalization approaches allow for a scaling of the AH count by a measure
reflective of the size of the associated state-year sample, while also offering a denominator that is
itself not likely to be significantly impacted by the prevailing malpractice environment (Frakes
2012a).
Inpatient mortality for selected conditions. Following the AHRQ’s IQIs, I next construct
a measure indicative of inpatient quality, whereby I calculate the composite rate of inpatient
mortality among a sub-sample of discharges in which the primary diagnosis code indicates
certain medical conditions (e.g., acute myocardial infarction, stroke, etc.). The indicated
conditions are both high-volume in occurrence and, for the most part, reflective of lowdiscretionary hospitalizations, in which case mortality rates among this sub-sample can be seen
as more likely reflective of the quality of care observed during the inpatient stay itself, rather
than as a result of risk selection. In the Appendix, I likewise provide more details regarding the
construction of this composite inpatient mortality rate, including a discussion of the various ways
in which I account for the potentially confounding influence of fluctuations in the proportions of
the conditions comprising this sub-sample (for instance, an increase in the proportional incidence
of hip fractures, which is a relatively low mortality condition among this sample, could result in
a reduction in the observed mortality rate with no actual improvement in quality). In the
preferred specification, I risk adjust mortality rates for these respective incidences.
[INSERT TABLE 4 ABOUT HERE]
Descriptive statistics. Means and standard deviations for the key health care quality and
legal variables are provided in Table 4. On average, each state-year cell contains roughly 1017
avoidable hospitalizations, which represents roughly 0.8 avoidable hospitalizations (based on
primary discharge codes) for each discharge associated with the delivery of a child (or a rate of
18 1.7 for the case of AH rates based on any diagnosis code) and on average 1.7 avoidable
hospitalizations (based on primary discharge codes) for each discharge in which the patient
presents with either acute myocardial infarction, stroke, hip fracture, or gastro-intestinal bleeding
(or a rate of 3.5 for the case of AH rates based on any diagnosis code). On average, each NHDS
state-year cell contains roughly 424 discharges associated with the selected conditions
comprising the sub-sample used in the second quality measure considered – that is, the
composite inpatient mortality rate. The average inpatient mortality rate among this sub-sample is
8 percent.
E. Specifications
With respect to each of the above health care quality measures, I first test for evidence of
malpractice-induced deterrence by estimating the following basic difference-in-difference
specification:
(1)
,
,
,
where s indexes state and t indexes year. CAPs,t represents an indicator variable for the presence
of a cap on non-economic damages in state s and year t. State fixed effects, γs, and year fixed
effects, λt, control for fixed differences across states and across years, respectively. Qs,t
represents the relevant healthcare quality measure – that is, either the avoidable hospitalization
rate or the composite inpatient mortality rate. The coefficient of interest in each specification is
captured by β1, representing the relationship between the relevant quality measure and the
adoption of non-economic damage caps.
I then test the robustness of the above findings to the inclusion of a range of additional
state-year factors by estimating the following specification
(2)
,
,
,
,
,
,
,
where Xs,t represents certain demographic characteristics of the state-year cell: the percentage of
patients in various age-sex categories,22 race categories (white, black and other), insurance
categories (private, government, no insurance and other), along with the percentage of patients
22
Age-sex categories are as follows: male under 30, female under 30, male 30-45, female 30-45, male 45-55, female
45-55, male 55-65, female 55-65, male 65-75, female 65-75, male over 75 and female over 75.
19 visiting hospitals of various bed sizes (0-100, 100-200, 200-300, 300-500 and 500+ beds) and of
various ownership types (proprietary, non-profit and government).23 Zs,t represents certain other
state-year characteristics (HMO penetration rate and its square, physician concentration rate, and
median household income).24 Os,t is a matrix representing a set of indicator variables for the
incidence of collateral source rule reforms, caps on punitive damages, and “indirect” tort
provisions. I include state-specific linear time trends, φs,t, to control for slowly-moving
correlations between the relevant quality measures in a state and the adoption of tort reforms by
that state.25
To explore whether prevailing health care quality is responsive to a change in standards
expected of physicians under the law (i.e., an alternative approach to exploring the malpracticequality link), I explore whether state mean rates for the relevant quality measures converge
towards their respective national mean rates as states adopt national-standard rules. For these
purposes, following Frakes (2012), I estimate the following specification:
(3)
_
,
,
,
,
,
,
,
,
where Xs,t, Zs,t, Os,t, γs, λt and φs,t are defined as above. NSs,t represents an indicator for a
national-standard law. Representing a measure of closeness between state and national quality
23
I form the incidences of the relevant demographic variables using the NHDS sample itself, though the results are
entirely robust to alternative state-year controls based off of the Census data. Following Frakes (2012a), in the
primary specifications, I form the relevant incidences using the sample of discharges in which patients present
themselves for acute myocardial infarction, stroke, gastro-intestinal bleeding or hip fracture. This sub-sample
consists of patients that will almost universally seek hospitalization upon the occurrence of the event, in which case
the sample itself is generally not sensitive to the prevailing legal environment. In any event, the results of this
exercise are also robust to the formation of the demographic covariates using the entire sample of state-year NHDS
discharges. 24
HMO penetration rates are from Interstudy Publications. Household income data is from the decennial Census
files and the American Community Surveys. Data on physician population counts are from the American Medical
Association (AMA) administrative records and were obtained from the Area Resource File. 25
Frakes (2012a) documents a relationship between the adoption of laws requiring physicians to follow national (as
opposed to local) standards and a resulting convergence in physician practices across regions. In light of the fact
that two of the damage-cap treatment states used in the defensive-medicine analysis below (Hawaii and Texas) were
dropped from the specifications estimated in Frakes (2012a) (due to an inability to classify the full history of their
standard-of-care laws), I exclude controls for national-standard laws in the damage-cap specifications estimated
below and focus instead on the traditional tort reform measures. However, the results presented below are robust to
the inclusion of controls for national-standard laws (not shown). 20 measures, Q_GAPs,t, is calculated as the absolute value of the difference between the state and
national quality measure, normalized by the national measure.26
1,
The coefficient of interest is represented by
identifying the extent to which the
adoption of a national-standard law is associated with a convergence between state and national
quality. Of course, the above-specified dependent variable may also capture sources of mean
revision in regional quality, posing a concern that
1
spuriously reflects a differential level of
mean revision between treatment and control states. I address this concern through a
falsification exercise in which I include a “lead” of the national-standard indicator, NSs,t+2, in the
base specifications, where this variable indicates, at time t, a state’s national-standard status at
time t+2. A negative estimate for
2
would suggest that the convergence associated with such
laws emerged in the period leading up to their adoptions and thus may not be reflective of a true
policy response.27 While this concern is especially paramount in the regional convergence
analysis, I likewise include lead coefficients for the damage cap variables in some of the
estimations of specifications (1) and (2) above.
III.
Results
A. Avoidable Hospitalization Rates
I begin the deterrence analysis by exploring whether malpractice pressure induces
physicians to deliver higher quality care at the outpatient or ambulatory level, where an increase
in quality is represented by a decrease in avoidable hospitalizations (AH) and where an increase
in pressure is identified by the non-incidence of a damages cap (and related reforms). Column 1
of Table 5 presents estimates of the basic difference-in-difference (DD) specification indicated in
equation (1). I estimate that the adoption of a non-economic damages cap is associated with a
roughly 0.3 percent increase in the AH rate (normalized by the low-variation health index and
calculated using primary diagnosis codes only). This estimate is statistically indistinguishable
from zero, though relatively tightly bound around zero.28 Even at the upper end of the 95
percent confidence interval, I find that the adoption of a non-economic damages cap is associated
26
In calculating this dependent variable for each state and year, the national rate is adjusted accordingly to remove
the contribution of the NHDS records from the relevant state-year cell. 27
I follow Gruber and Hungerman (2008) in observing behavior in the 2-year period leading up to the reform,
providing enough time to sufficiently assess pre-reform treatment rates.
28
Reported standard errors in Table 5 and in all subsequent tables are clustered at the state level to allow for
arbitrary within-state correlations of the error structure. 21 with only a 4.9 percent increase in the AH rate. That is, at the most, the evidence suggests that
malpractice pressure (as captured by the lack of a damages cap) leads to a modest level of
deterrence of AHs, inconsistent with the notion that the medical liability system is a substantial
driver of health care quality.
The point estimate of the non-economic damage cap coefficient falls and turns negative
with the addition of other tort reforms and state-year covariates, as demonstrated by Column 2 of
Table 5. Moreover, it remains relatively tightly bound around 0, where the upper end of the 95
percent confidence interval suggests a 3.7 percent increase in the AH rate following a noneconomic damages cap adoption. This result remains essentially unchanged with the subsequent
addition of state-specific linear time trends, as indicated by Column 3. While the above results
suggest little or no relationship between malpractice pressure and observed measures of health
care quality, I nonetheless employ a falsification exercise in which I estimate the differential in
AH rates between treatment states and control states in the period of time leading up to the
adoption of the relevant reforms, during which time one would not expect to observe any such
relationship. Consistent with these expectations, as demonstrated by Columns 4 and 5, the
estimated coefficient of the lead reform indicators for the non-economic damages cap is likewise
small, insignificant and relatively tightly bound around zero (where the upper end of the 95%
confidence interval suggests only a 5.7 percent differential AH rate between treatment and
control states in the two year period leading up the reform adoption).
The estimated coefficients of the collateral source rule reform variable and the punitive
damages cap variable are likewise small and statistically insignificant, with similarly sized
confidence intervals around zero. The estimated coefficient of the residual reform category –
i.e., the “indirect” reform category specified according to Kessler and McClellan (1996) – is
negative and bound away from zero in some specifications, counter to any expectation that such
reforms would relax malpractice pressures to the detriment of patient quality.29 In each of the
specifications estimated in Table 5, I perform F-tests for the joint significance of the various
malpractice reforms and fail to reject the hypothesis that the reforms are jointly equal to zero.
29
One component of this residual category is the reform of the joint and several liability rule. In alternative
specifications (not shown), where I include the joint and several liability reform independently, I likewise estimate
small, negative point estimates for this reform. 22 As demonstrated by Table 6, the above pattern of results does not substantially change
upon the estimation of various alternative specifications, including those that calculate avoidable
hospitalization rates by: (1) normalizing the state-year AH counts by the number of deliveries
within the state-year cell, (2) normalizing the state-year AH counts by the number of primary
acute myocardial infarction admissions in the state-year cell, and (3) counting AHs using
information from any of the diagnosis codes, as opposed to only the primary diagnosis code.30
The estimated damage cap coefficients turn more in the positive direction when normalizing AH
counts by the number of deliveries (i.e., more in the direction indicative of deterrence), though
the estimates remain small, statistically indistinguishable from zero, and relatively tightly bound
around zero. Moreover, the estimated lead coefficient suggests that any such positive
relationship may have materialized in the pre-adoption period.31
Most of the above-estimated specifications include state-year controls for physician
concentration rates. Such controls may absorb any impact of the reforms that occur through
changes in the physician population. However, these simple controls may not absorb all supplyrelated consequences of such reforms. One effect of non-economic damage cap adoptions
sometimes hypothesized is that lower-quality physicians may be attracted to the jurisdiction
subsequent to the reform (Seabury 2010), a development which could otherwise confound any
attempt to isolate the impact of malpractice pressure on the quality provided by any-given
provider (arguably the emphasis of the deterrence analysis). Of course, to the extent that noneconomic damage caps would attract low-quality physicians and lead to a decline in observed
quality measures – e.g., to an increase in AH rates – this omission could only help to explain
30
The analysis thus far has implicitly assumed that malpractice liability may only impact avoidable hospitalization
rates through the influence of liability pressure on the quality of care delivered at the outpatient level. However,
physicians may hold some degree of discretion over the decision to admit patients when presenting themselves at the
hospital with the indicated conditions. Liability may influence this discretion in various ways. For instance, fearful
over liability for failing to admit a patient, physicians may admit those marginal patients whom they may have
otherwise decided did not warrant admission. A response of this nature could confound the above results by
masking the presence of a true deterrent effect at the outpatient level. Of course, liability fears could also impact the
inpatient admission decision in the opposite direction, whereby such fears induce hospitals and providers to avoid
high risk patients (Mello et al. 2005) or to avoid some hospitalizations over liability fears stemming from the care to
be administered during the hospital admission itself. As such, it is unclear what role liability fears may play at the
admissions decision level. Further research is needed regarding this margin. In any event, the above analysis
suggests that liability pressure does not appear to be a major determinant of those measures that the health care
quality literature and the AHRQ have promoted as being indicative of the quality of care delivered at associated
outpatient environments (using inpatient data). 31
Finally (not shown), these AH estimates persist under alternative specifications that use levels of AH rates, as
opposed to logs.
23 some of the estimated positive effects of such reforms on AH rates. That is, a correction for this
bias would likely push the estimated impacts of the reforms on AH rates even lower, only
lending further support to the claim that liability pressure does not appear to be substantially
improving the quality of care being delivered by physicians at the outpatient level.
The above approach identifies the influence of malpractice law by comparing quality
across regimes marked by different levels of expected liability awards, effectively taking as
given the structure of the liability system itself. In an alternative approach to exploring the link
between malpractice and health care quality, I estimate the specification indicated by equation
(3) above and explore whether observed quality metrics are influenced by reforms that alter the
clinical standards of care expected of physicians. I present the results of this alternative
approach in Table 8. Upon the abandonment of a local standard rule and the contemporaneous
adoption of a national standard rule (which should reflect a change in the expectations being
placed upon physicians considering the significant variations in care delivered across regions), I
estimate a roughly 1 – 1.7 percentage-point increase in the percentage absolute deviation
between state and national AH rates. The direction of this estimate is inconsistent with the
hypothesized impact of national-standard rules, which would be expected to lead to a
convergence (not a divergence) in regional quality measures. As such, the direction of this
estimate is likewise inconsistent with the notion that healthcare quality measures are responsive
to the clinical standards expected of physicians under the law.
Of course, the 95 percent confidence interval for this estimate bounds zero and thus
cannot rule out a convergence impact of national-standard rules (which would otherwise suggest
that health care quality outcomes are sensitive to medical liability rules). However, the
confidence interval itself is, all things considered, relatively small. The bottom end of this
interval suggests an impact of national-standard adoptions in which the gap between state and
national avoidable hospitalization rates is closed by 2.7 percentage points, or by roughly 19.5
percent. Thus, at the most, the estimates suggest a convergence impact of national-standard
adoptions on health care quality that is less than half of the size of the mean estimated
convergence in treatment and diagnostic utilization measures documented in Frakes (2012a).
24 B. Inpatient Mortality Rates for Selected Conditions
Table 7 presents estimation results from specifications that explore the relationship
between non-economic damage caps (and related reforms) and the composite inpatient mortality
rate for selected conditions (following the AHRQ’s Inpatient Quality Indicators measure). The
pattern of results is very similar to those presented in Tables 5 and 6 for the case of avoidable
hospitalizations. Each specification estimates a small association between non-economic
damage cap adoptions and the relevant inpatient mortality rate. In the basic DD specification
without any state-year controls, I estimate that non-economic damage cap adoptions are
associated with a 0.8 percent increase in the indicated mortality rate. I find that this estimate
falls to a -3.8 percent decrease in such rates upon the addition of various state-year controls and
state-specific linear time trends, suggesting that higher liability pressure (as identified by the lack
a damages cap) may actually lead to higher mortality rates and thus lesser quality, counter to the
deterrence hypothesis.
[INSERT TABLE 7 ABOUT HERE]
The estimated coefficients of the non-economic damage cap variable in Table 7 are
statistically indistinguishable from zero and cannot rule out some amount of malpractice-induced
deterrence – that is, some amount of reduction in mortality rates associated with higher liability
pressure. However, as with the AH rates results, the estimated confidence intervals are relatively
small. Based on the specification with state-year controls and state-specific linear time trends,
the upper end of the 95 percent confidence interval suggests that the adoption of a damages cap
is associated with only a 2.4 percent increase in the composite inpatient mortality rate for the
indicated conditions (or a 0.2 percentage-point increase in the relevant rate) – that is, at the most,
a reduction (increase) in liability pressure is associated with only a minor decline (improvement)
in observed inpatient quality. 32
32
The unit of observation in the specifications estimated above is a given state-year cell. In an alternative approach
(not shown), I estimate linear probability models where the unit of observation is an individual discharge within the
sample of inpatient admissions associated with the selected conditions (e.g., acute myocardial infarctions, strokes,
etc.) and where the dependent variable is an indicator for inpatient mortality (in such models, I include controls for
the incidence of the relevant conditions). The results from this alternative approach are (perhaps not surprisingly)
nearly identical to those of the state-year specifications estimated in Table 7. 25 The non-economic damages cap results generalize to the additional tort reforms,
suggesting a generally weak relationship between inpatient quality and a broader range of
reforms. Moreover, the results of an F-test of joint significance fail to reject the hypothesis that
the coefficients of the various tort reforms are all jointly equal to zero.33 Roughly 35 state-year
observations are dropped from the above specifications as a result of having no relevant
mortalities in the associated state-year cell, in light of the logged dependent variable. However,
as demonstrated by Columns 6 and 7 of Table 7, I estimate essentially the same pattern of results
when using non-logged mortality rates. Likewise, the estimates remain nearly unchanged when I
use non-standardized mortality rates and address concerns over fluctuations in the various
conditions comprising the relevant sub-sample for this mortality analysis by including separate
covariates for the relative state-year incidences of the various conditions.34
In Panel B of Table 8, I estimate the association between the adoption of nationalstandard rules and the percentage absolute deviation between state and national inpatient
mortality rates for the selected conditions (risk adjusted). The estimated coefficient of the
national-standard variable is again small and statistically indistinguishable from zero. Though
insignificant, the point estimates suggest that national-standard rules are associated with 0.5 – 2
percentage-point reduction in the percentage absolute deviation between the relevant state and
national mortality rates, which is consistent with a convergence effect in which roughly 3 – 12
percent of the gap between such state and national rates is closed upon the legal reform. The
associated confidence intervals are slightly wider than those estimated in the AH rate
specifications, where the bottom ends of the 95 percent confidence intervals estimated across the
various specifications suggest a convergence effect in which 20 – 40 percent of the state-national
mortality gap is closed upon a national-standard adoption, an amount which is still less than the
mean treatment and diagnostic convergence findings documented by Frakes (2012a).
33
However, the p-value of the F-test in column 5 is substantially lower at 0.07; though, this result is perhaps largely
driven by the negative estimate for the punitive damages cap coefficient, a result that is of a sign opposite that
expected from a deterrent response. 34
The results are also nearly identical when estimating specifications that make no corrections for fluctuations in the
relative incidences of the respective conditions. In other specifications (not shown), I estimate the relationship
between liability reforms and the rate of hospitalization for each such condition, where this rate is calculated relative
to the total number of hospitalizations within this sub-sample of selected conditions. The results suggest very little
relationship, if any, between liability pressure and the distribution of conditions comprising this sub-sample. For
instance, at the upper end of the 95 confidence interval, the adoption of a non-economic damages cap is associated
with only a 0.7 percent increase in the relative rate of hip fracture admissions among this subsample and only a 3
percent increase in the relative rate of stroke admissions. 26 [INSERT TABLE 8 ABOUT HERE]
Moreover, though likewise statistically indistinguishable from zero, the point estimates
for the coefficients of the national-standard lead indicator variables are also negative and of a
magnitude greater than that associated with the contemporaneous indicators. This pre-period
falsification exercise suggests that any convergence in quality arising from national-standard
adoptions, even if present, may have materialized in the period leading up to the nationalstandard adoptions (which would be inconsistent with the expectation of a treatment effect). All
told, there is little evidence to suggest that the geographic standardization of medical malpractice
law is associated with a corresponding standardization of observed quality, implying that
healthcare quality may not be so responsive to liability standards (as they are currently applied).
IV.
Conclusion
The literature has shown that physicians’ clinical practices may be responsive along some
dimensions to the medical liability system, whether in showing that higher liability pressure is
associated with greater diagnostic utilization (Baicker, Fischer and Chandra 2007) or in showing
that regional obstetric and cardiac procedure rates change upon an alteration of the clinical
standards expected of that region under malpractice law (Frakes 2012a). However, these
previous findings do not necessarily entail that the system is operating as intended – that is, to
optimize the care delivered to patients and to encourage the maintenance of high quality
practices. Physicians may respond to malpractice standards, but that liability channel may only
lead to true deterrence of medical errors and low-quality practices so long as such liability
standards are themselves designed to ensure such an outcome. Presently, the liability system
largely defers to medical custom itself in determining the standards expected of physicians under
the law (as opposed to a more abstract process of determining what is reasonable and appropriate
care). As such, considering the general deficiencies of the present health care system in ensuring
high quality delivery – that is, considering the deficiencies of customary practices (McGlynn et
al. 2003 – it is perhaps reasonable to believe that the liability system, in turn, may be
insufficiently designed to deliver the desired degree of deterrence.
Focusing on composite measures of health care quality in both inpatient and outpatient
domains, the results presented above are indeed suggestive of a current malpractice system that
does not substantially improve upon the quality of care delivered by medical providers. This
27 finding, combined with the evidence offered to date regarding the remarkably low percentage of
negligently injured patients who even pursue a malpractice claim in the first instance (Localio et
al. 1991), implies that malpractice law may be falling short in meeting two of its fundamental
goals: deterring low-quality medicine and ensuring compensation for losses. However,
considering that physicians may be responsive to malpractice law in a more general sense (and
are thus not driven exclusively by non-legal influences), certain alterations of the liability system
may lead it to better achieve these objectives.
Next generation malpractice reforms currently receiving much attention include those
that provide more definitive liability safe harbors for physicians who follow specified clinical
practice guidelines (see, for example, Blumstein 2006). To the extent that such guidelines are
indeed designed to encourage the maintenance of best medical practices and are mindful of costeffectiveness considerations, they hold the promise of encouraging both high quality and low
cost medicine. Further experimentation (and subsequent empirical analysis) is of course
necessary to determine whether such reforms may live up to such promise.
REFERENCES
Baicker, Katherine, Elliot Fisher, and Amitabh Chandra. 2007. Malpractice Liability Costs And
The Practice Of Medicine In The Medicare Program. Health Affairs, 26, 841-52.
Blumstein, James. 2002. The Legal Liability Regime: How Well is it Doing in Assuring
Quality, Accounting for Costs, and Coping with an Evolving Reality in the Health Care
Marketplace. Annals of Health Law 11, 125-146.
Blumstein, James. 2006. Medical Malpractice Standard-Setting: Developing Malpractice “Safe
Harbors” as a new Role for QIOs? Vanderbilt Law Review 59, 1017 – 1049.
Carter, Mary. Variations in Hospitalization Rates among Nursing Home Residents: The Role of
Discretionary Hospitalizations. Health Services Research 38, 1177-1206.
Casey, Brian M., Donald D. McIntire, and Kenneth J. Leveno. 2001. The continuing value of
the Apgar score for the assessment of the newborn infants. New England Journal of Medicine
344, 467 –71.
Centers for Medicare and Medicaid Services. 2010. National Health Expenditure Accounts:
Tables. Available at: http://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-
28 Trends-and-Reports/NationalHealthExpendData/Downloads/tables.pdf (accessed June, 17
2012).
Chandra, Amitabh, and Douglas O. Staiger. 2007. Productivity Spillovers in Healthcare:
Evidence From the Treatment of Heart Attacks. Journal of Political Economy 115, 103-40.
Cohen, Thomas. 2005. Punitive Damage Awards in Large Counties, 2001. Bureau of Justice
Statistics Selected Findings.
Conley, Timothy, and Christopher Taber. 2011. Inference with ‘Difference-in-Differences’ with
a Small Number of Policy Changes. The Review of Economics and Statistics 1, 113-25.
Currie, Janet, and W. Bentley MacLeod. 2008. First Do No Harm? Tort Reform and Birth
Outcomes. Quarterly Journal of Economics 123, 795-830.
Dafny, Leemore, and Jonathan Gruber. 2005. Public Insurance and Child Hospitalizations:
Access and Efficiency Effects. Journal of Public Economics 89, 109-29.
Dubay, Lisa, Robert Kaestner, and Timothy Waidmann. 1999. The Impact of Malpractice Fears
on Cesarean Section Rates. Journal of Health Economics 18, 491-522.
Frakes, Michael. 2012a. The Impact of Medical Liability Standards on Regional Variations in
Physician Behavior: Evidence from the Adoption of National-Standard Rules. American
Economic Review, forthcoming.
-----------. 2012b. Defensive Medicine and Obstetric Practices. Journal of Empirical Legal
Studies, forthcoming.
Greenberg, Michael D., Amelia M. Haviland, J. Scott Ashwood, and Regan Main. 2010. Is
better patient safety associated with less malpractice activity? Evidence from California. Santa
Monica: RAND Institute for Civil Justice.
Gruber, Jonathan, and Daniel M. Hungerman. 2008. Church versus the Mall: What Happens
When Religion Faces Increased Secular Competition. Quarterly Journal of Economics 123, 83162.
Gruber, Jonathan, and Maria Owings. 1996. Physician Financial Incentives and Cesarean
Section Delivery. The RAND Journal of Economics 27, 99-123.
Hyman, David, Bernard Black, Charles Silver & William Sage. 2009. Estimating The Effect of
Damage Caps in Medical Malpractice Cases: Evidence from Texas. Journal of Legal Analysis 1,
355-409.
Institute of Medicine Committee on Quality of Health Care in America. 2000. To Err Is
Human: Building a Safer Health System. Washington, D.C.: National Academy Press.
29 Kessler, Daniel, and Mark McClellan. 1996. Do Doctors Practice Defensive Medicine?
Quarterly Journal of Economics 111, 353-90.
Klick, Jonathan, and Thomas Stratmann. 2007. Medical Malpractice Reform and Physicians in
High-Risk Specialties. Journal of Legal Studies 36, S121-S142.
Lakdawalla, Darius, and Seth Seabury. 2009. The Welfare Effects of Medical Malpractice
Liability. NBER Working Paper No. 15383.
Localio, A. Russell, Ann Lawthers, Troyen Brennan, Nan Laird, Liesi Hebert, Lynn Peterson,
Joseph Newhouse, Paul Weiler, and Howard Hiatt. 1991. Relation between malpractice claims
and adverse events due to negligence: results of the Harvard Medical Practice Study III. New
England Journal of Medicine 325, 245-51.
Malani, Anup, and Julian Reif. Accounting for Anticipation Effects: An Application to Medical
Malpractice Tort Reform. NBER Working Paper No. 16593.
McClellan, Mark, and Douglas Staiger. 1999. The Quality of Health Care Providers. NBER
Working Paper No. 7327.
McGlynn, Elizabeth A., Steven M. Asch, John Adams, Joan Keesey, Jennifer Hicks, Alison
DeCristofaro, and Eve A. Kerr. 2003. The Quality of Health Care Delivered to Adults in the
United States. New England Journal of Medicine 348, 2635-2645.
Mello, Michelle M. 2006. Medical Malpractice: Impact of the Crisis and Effect of State Tort
Reforms. Research Synthesis Report #10. New Brunswick, NJ: Robert Wood Johnson
Foundation.
Mello, Michelle M., and Troyen A. Brenan. 2002. Deterrence of Medical Errors: Theory and
Evidence for Malpractice Reform. Texas Law Review 80, 1595-1637.
Mello, Michelle M., David M. Studdert, Catherine M. DesRoches, Jordon Peugh, Kinga Zapert,
Troyen A. Brennan, and William M. Sage. 2005. Effects of a Malpractice Crisis on Specialist
Supply and Patient Access to Care. Annals of Surgery 242, 621-628.
National Center for Health Statistics. 1977-2005. National Hospital Discharge Survey. Centers
for Disease Control and Prevention (last accessed at NCHS Research Data Center on January 31,
2012).
Seabury, Seth. 2010. Does Malpractice Liability Reform Attract High Risk Doctors? RAND
Working Paper No. WR-674-ICJ.
Sloan, Frank A., and John H. Shadle. Is there empirical evidence for “Defensive Medicine”? A
reassessment. Journal of Health Economics 28, 481-491.
30 Studdert, David M., Michelle M. Mello, Atul A. Gawande, Tejal K. Ghandi, Allen Kachalia,
Catherine Yoon, Ann Louise Puopolo and Troyen A. Brennan. 2006. Claims, Errors, and
Compensation Payments in Medical Malpractice Litigation. New England Journal of Medicine
354, 2024-2033.
Wennberg, John E. 1984. Dealing With Medical Practice Variations: A Proposal for Action.
Health Affairs 3, 6-32.
Wennberg, John E., and Megan McAndrew Cooper (Eds.). 1999. The Quality of Medical Care
in the United States: A Report on the Medicare Program. The Dartmouth Atlas of Health Care in
the United States. Chicago: American Health Association Press.
Weissman, Joel S., Constantine Gatsonis, and Arnold Epstein. 1992. Rates of Avoidable
Hospitalization by Insurance Status in Massachusetts and Maryland. JAMA 268, 2388-2394.
Zeiler, Kathryn, Charles Silver, Bernard Black, David Hyman, and William Sage. 2007.
Physicians' Insurance Limits and Malpractice Payments: Evidence from Texas Closed Claims,
1990-2003. Journal of Legal Studies 36, S9-S45.
31 Table 1. Variations in Non-Economic Damage Caps (1979-2005)
State
Year Adopted
Year Dropped
State
Year Adopted
Year Dropped
Alaska
1986
Mississippi
2003
Alabama
1987
1992
Montana
1996
Colorado
1987
North Dakota
1996
Florida
2004
New Hampshire
1987 (2)
1981 (1); 1991(2)
Hawaii
1987
Ohio
2003 (2)
1992(1)
Idaho
1988
Oklahoma
2004
Illinois
1995
1998
Oregon
1988
2000
Kansas
1987
Texas
2004(2)
1988(1)
Massachusetts
1987
Utah
1988
Maryland
1987
Washington
1986
1990
Michigan
1987
Wisconsin
1986
Minnesota
1986
1990
West Virginia
1986
Missouri
1986
Notes: years of adoption and invalidation/repeal (if applicable) of laws imposing caps on non-economic damage awards in
malpractice cases (or tort cases generally) are indicated above. States are only included if their relevant malpractice laws varied
over the 1979 – 2005 period. Legislative variation is excluded from this table if it represents a situation in which an adoption and
invalidation/repeal occurred during the same year. Source: Database of State Tort Law Reforms (2nd).
Table 2. Variations in Collateral-Source Rule Reforms (1979-2005)
State
Year Adopted
Year Dropped
State
Year Adopted
Year Dropped
Alabama
1987 (1); 2001(2)
1997(1)
Montana
1988
Colorado
1987
North Dakota
1988
Connecticut
1986
New Hampshire
1981
Georgia
1988
1991
New Jersey
1988
Hawaii
1987
New York
1985
Idaho
1990
Ohio
2002(2)
1998(1)
Indiana
1987
Oklahoma
2004
Kansas
1993
Oregon
1988
Kentucky
1989
1995
Pennsylvania
2002(2)
1981(1)
Massachusetts
1987
Rhode Island
2002
Maine
1990
Utah
1987
Michigan
1987
Wisconsin
1995
Minnesota
1986
West Virginia
2003
Notes: years of adoption and invalidation/repeal (if applicable) of laws reforming traditional collateral source rules are indicated
above. States are only included if their relevant malpractice laws varied over the 1979 – 2005 period. Legislative variation is
excluded from this table if it represents a situation in which an adoption and invalidation/repeal occurred during the same year.
Source: Database of State Tort Law Reforms (2nd).
Table 3. Variations in National-Standard Rules (1979-2005)
State
Year Adopted
Year Dropped
State
Year Adopted
Year Dropped
Alabama
1980
Montana
1985
Colorado
1983
Oklahoma
1984
Connecticut
1984
Rhode Island
1998
Delaware
1999
South Carolina
1981
D.C.
1980
South Dakota
1988
Indiana
1992
West Virginia
1986
Maryland
1994
Wyoming
1981
Mississippi
1983
Notes: years of adoption and repeal (if applicable) of laws requiring that physicians follow national (as opposed to local)
standards of care in malpractice actions. States are only included if their relevant malpractice laws varied within the 1979 – 2005
period. Source: Frakes (2012a).
32 Table 4. Descriptive Statistics
Mean (Standard Deviation)
Panel A: Quality Measures (NHDS)
Avoidable Hospitalization Count: Any Diagnosis
Code
1017.84
(1390.91)
Avoidable Hospitalization Rate: Any Diagnosis
Code, Scaled by Low-Variation Health Index
3.53
(0.88)
Avoidable Hospitalization Rate: Primary Diagnosis
Code, Scaled by Low-Variation Health Index
1.70
(0.42)
Avoidable Hospitalization Rate: Any Diagnosis
Code, Scaled by Delivery Count
1.71
(0.85)
Avoidable Hospitalization Rate: Primary Diagnosis
Code, Scaled by Delivery Count
0.82
(0.40)
Composite Inpatient Mortality Rate
0.08
(0.03)
Panel B: Absolute Deviation between State and
National Quality Measures, Scaled by National
Measure
Avoidable Hospitalization Rate, Scaled by LowVariation Health Index
0.14
(0.13)
0.16
(0.15)
Composite Inpatient Mortality Rate
Panel C: Tort Variables (in NHDS sample)
Non-Economic Damage Caps
0.31
(0.46)
Collateral Source Rule Reform
0.53
(0.50)
Punitive Damage Cap
0.32
(0.47)
“Indirect” Tort Reform
0.73
(0.44)
0.60
(0.49)
Notes: Standard deviations are in parentheses. Quality measures are from a sample of 1190 state-year cells from
the 1979 – 2005 NHDS files. Quality statistics are weighted by the relevant denominator used in the state-year
quality rate (e.g., the state-year delivery count or the state-year low-variation health index), except for the case of
avoidable hospitalization counts, which are un-weighted. For illustrative purposes, I do not weight the tort variable
statistics (though they are weighted in various ways in the different specifications, depending on the denominator
used in the state-year quality rate).
Source: National Hospital Discharge Survey (1979-2005).
National-Standard Laws
33 Table 5. Relationship between Tort Reforms and Avoidable Hospitalization Rates (Logged,
Normalized by Low-Variation Health Index)
(1)
(2)
(3)
(4)
(5)
0.003
(0.023)
-0.016
(0.026)
-0.010
(0.029)
-
-
-
0.012
(0.042)
-0.015
(0.049)
0.000
(0.030)
-0.010
(0.033)
Contemporaneous Dummy
-
0.020
(0.026)
0.012
(0.032)
-
2-Year Lead Dummy
-
-
-
-
Contemporaneous Dummy
-
0.032
(0.033)
-0.012
(0.036)
-
2-Year Lead Dummy
-
-
-
-
Contemporaneous Dummy
-
-0.082
(0.049)
-0.067**
(0.032)
-
2-Year Lead Dummy
-
-
-
-
Non-Economic Damage Cap:
Contemporaneous Dummy
2-Year Lead Dummy
Collateral Source Rule Reform:
0.014
(0.027)
0.007
(0.027)
Punitive Damage Cap:
-0.041
(0.035)
0.060**
(0.028)
“Indirect” Tort Law:
-0.064*
(0.033)
-0.024
(0.037)
F-Statistic (Malpractice Variables
1.28
1.89
0.05
1.82
Jointly = 0)
Prob > F (p value)
0.29
0.13
0.95
0.10
Control Variables?
NO
YES
YES
NO
YES
State-Specific Linear Trends?
NO
NO
YES
NO
YES
N
1190
1177
1177
1190
1177
Notes: robust standard errors corrected for within-state correlation in the error term are reported in parentheses. All
regressions included state and year fixed effects and are weighted by the low-variation health index (i.e., the sum of
discharges for acute myocardial infarction, stroke, hip fracture or gastrointestinal bleeding) associated with each state-year
cell.
Source: 1979 – 2005 National Hospital Discharge Surveys.
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
* Significant at the 10 percent level.
34 Table 6. Relationship between Tort Reforms and Avoidable Hospitalization Rates (Logged):
Specification Checks
(1) (2) (3) (4) (5) (6) (7) (8) (9) 0.039
(0.059)
0.025
(0.043)
0.016
(0.039)
-0.014
(0.038)
0.012
(0.023)
-
-
-
-0.004
(0.040)
-0.010
(0.034)
0.028
(0.021)
-
0.014
(0.049)
0.030
(0.045)
-
-
0.015
(0.024)
-0.004
(0.024)
Contemporaneous Dummy
-
0.038
(0.044)
-
0.007
(0.042)
-
0.029
(0.029)
2-Year Lead Dummy
-
-
-
-
-
-
Contemporaneous Dummy
-
0.008
(0.064)
-
0.002
(0.044)
-
-0.029
(0.027)
2-Year Lead Dummy
-
-
-
-
-
-
Contemporaneous Dummy
-
0.054
(0.065)
-
-0.093*
(0.046)
-
-0.067**
(0.027)
2-Year Lead Dummy
-
-
-
-
-
-
-
1.43
-
2.31
1.31
NO
0.07
NO
0.26
NO
Non-Economic Damage Cap:
Contemporaneous Dummy
2-Year Lead Dummy
Collateral Source Rule Reform:
0.053
(0.041)
-0.025
(0.067)
0.015
(0.036)
0.002
(0.033)
0.027
(0.028)
0.008
(0.027)
Punitive Damage Cap:
-0.032
(0.070)
0.075
(0.063)
-0.039
(0.047)
0.089*
(0.049)
-0.032
(0.028)
0.007
(0.021)
“Indirect” Tort Law:
0.071
(0.057)
-0.039
(0.048)
F-Statistic (Malpractice
Variables Jointly = 0)
Prob > F (p value)
Primary Diagnosis Only?
-
0.64
0.88
YES
0.63
YES
0.53
YES
Denominator for AH Rate?
CHILDBIRTH DELIVERY COUNT
Control Variables?
State-Specific Linear Trends?
N
NO
NO
1159
YES
YES
1146
-0.079*
(0.043)
-0.060*
(0.035)
1.73
0.24
0.12
YES
YES
YES
ACUTE MYOCARDIAL
INFARCTION COUNT
NO
YES
YES
NO
YES
YES
1182
1169
1169
YES
YES
1146
-0.065**
(0.031)
-0.010
(0.039)
LOW-VARIATION INDEX
NO
NO
1190
YES
YES
1177
YES
YES
1177
Notes: robust standard errors corrected for within-state correlation in the error term are reported in parentheses. All regressions
included state and year fixed effects and are weighted by the denominator associated with the avoidable hospitalization rate.
Source: 1979 – 2005 National Hospital Discharge Surveys.
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
* Significant at the 10 percent level.
35 Table 7: Relationship between Tort Reforms and Inpatient Mortality Rate for Selected
Conditions
(1) (2) (3) (4) (5) (6) (7) (8) (9) 0.008
(0.030)
-0.012
(0.028)
-0.038
(0.030)
-0.003
(0.003)
0.007
(0.025)
-0.035
(0.030)
-
-
-0.012
(0.041)
-0.049
(0.041)
0.002
(0.003)
-
0.020
(0.031)
-0.020
(0.036)
-
-
-
-
Contemporaneous Dummy
-
0.015
(0.025)
0.014
(0.041)
-
-
0.003
(0.004)
-
0.013
(0.039)
2-Year Lead Dummy
-
-
-
-
-
-
-
-
Contemporaneous Dummy
-
-0.001
(0.038)
0.006
(0.047)
-
-
0.000
(0.004)
-
0.007
(0.046)
2-Year Lead Dummy
-
-
-
-
-
-
-
-
Contemporaneous Dummy
-
0.009
(0.027)
0.003
(0.027)
-
-
0.001
(0.002)
-
0.005
(0.028)
2-Year Lead Dummy
-
-
-
-
-
-
-
-
-
0.15
0.49
0.25
-
0.44
-
0.44
Non-Economic Damage Cap:
Contemporaneous Dummy
2-Year Lead Dummy
Collateral Source Rule Reform:
-0.007
(0.047)
0.052
(0.051)
Punitive Damage Cap:
-0.053*
(0.029)
0.033
(0.050)
“Indirect” Tort Law:
F-Statistic (Malpractice
Variables Jointly = 0)
Prob > F (p value)
Risk-Adjusted Mortality
Rates?
Mortality Rates Logged?
Control Variables?
State-Specific Linear Trends?
N
0.000
(0.034)
-0.010
(0.040)
1.99
-
0.96
0.74
0.78
0.07
-
0.78
-
0.78
YES
YES
YES
YES
YES
YES
YES
NO
NO
YES
NO
NO
1154
YES
YES
NO
1141
YES
YES
YES
1141
YES
NO
NO
1154
YES
YES
YES
1141
NO
NO
NO
1189
NO
YES
YES
1176
YES
NO
NO
1154
YES
YES
YES
1141
Notes: robust standard errors corrected for within-state correlation in the error term are reported in parentheses. All regressions
included state and year fixed effects and are weighted by the number of admissions (for the relevant state and year) in the subsample of discharges associated with the selected conditions (i.e., the sum of discharges for acute myocardial infarction, heart
failure, acute stroke, gastrointestinal bleeding, hip fracture or pneumonia). Mortality rates in Columns 1-7 are risk-adjusted for
the incidence (among the sub-sample) of each of the conditions comprising the sub-sample of selected conditions. Columns 8
and 9 use un-adjusted mortality rates, but include controls for each of these incidences.
Source: 1979 – 2005 National Hospital Discharge Surveys.
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
* Significant at the 10 percent level.
36 Table 8. Relationship between National‐Standard Laws and Absolute Deviation between State and National Quality Measures (1)
(2)
(3)
(4)
(5)
Panel A. Dependent variable: absolute deviation between state and national avoidable hospitalization rates, scaled by
national rate (where avoidable hospitalization rates are normalized by low-variation health index)
National Standard Law
Contemporaneous Dummy
2-Year Lead Dummy
N
0.010
(0.022)
1166
-0.012
(0.030)
0.030
(0.029)
1166
-0.002
(0.030)
0.024
(0.029)
1155
0.003
(0.022)
0.034
(0.031)
1155
0.017
(0.022)
1155
Panel B. Dependent variable: absolute deviation between state and national inpatient mortality rates for selected
conditions, scaled by national rate
National Standard Law
-0.006
-0.001
0.003
-0.014
-0.020
Contemporaneous Dummy
(0.016)
(0.028)
(0.024)
(0.026)
(0.025)
-0.007
-0.022
-0.015
2-Year Lead Dummy
(0.025)
(0.026)
(0.027)
N
1165
1165
1154
1154
1154
Control Variables?
NO
NO
YES
YES
YES
State-Specific Linear Trends?
NO
NO
NO
YES
YES
Notes: robust standard errors corrected for within-state correlation in the error term are reported in parentheses. All regressions
included state and year fixed effects and are weighted by the low-variation health index (i.e., the sum of discharges for acute
myocardial infarction, stroke, hip fracture or gastrointestinal bleeding) in Panel A or weighted by the number of hospitalizations
in the selected conditions sub-sample (i.e., the sum of discharges for acute myocardial infarction, heart failure, acute stroke,
gastrointestinal bleeding, hip fracture or pneumonia) in Panel B.
Source: 1979 – 2005 National Hospital Discharge Surveys.
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
* Significant at the 10 percent level.
37 Appendix: Data Source and Quality Measures
National Hospital Discharge Survey
Healthcare quality data is collected from the National Hospital Discharge Survey (NHDS), a
nationally-representative sample of inpatient discharge records from short-stay, non-federal
hospitals. For approximately 260,000 inpatient records per year, the NHDS contains information
on, among other things: (a) primary and secondary diagnosis and procedure codes, (b) certain
demographic characteristics of the patient, and (c) certain characteristics of the hospital. I
supplement the public NHDS files with geographic identifiers (restricted-use variables) received
pursuant to an agreement with the Research Data Center at the National Center for Health
Statistics (NCHS). The resulting sample covers the years 1979 to 2005.
Healthcare Quality Measures
For the purposes of the above study, I look to the AHRQ for guidance in selecting quality
metrics. The AHRQ measures are particularly useful for the present study in so far as they are
designed for use with administrative inpatient databases such as the NHDS. The AHRQ’s
quality indicators are classified into 4 modules: (1) Prevention Quality Indicators (PQIs),
identifying admissions that could have been avoided through access to high-quality outpatient
care, (2) Inpatient Quality Indicators (IQIs), reflecting the quality of care inside hospitals
including inpatient mortality for certain medical conditions, (3) Patient Safety Indicators (PSIs),
focusing on potentially avoidable complications during inpatient care and (4) Pediatric Quality
Indicators, applying the other three modules to the case of children and neonates.
For the purposes of this analysis, I utilize quality metrics inspired by the first two modules: PQIs
and IQIs. While the NHDS does sample a large number of hospitalizations for each state and
year, it nonetheless may not provide enough power to explore the rather rare adverse events that
comprise some of the PSIs, along with the quality of care delivered only to the child population.
Moreover, the richness of the classification codes associated with the PSIs, in light of the need to
track changes to these classifications over the entire 27-year period of the NHDS sample,
complicates the ability to apply such indicators to the NHDS sample (further complicating this
application are the lack of DRG codes in the NHDS records, which the ARHQ recommends
using in construction their PSIs). The PQIs and IQIs, under proper modifications are more
readily approachable using the NHDS records.
Avoidable hospitalizations. First, I calculate a rate of avoidable hospitalizations (AH) within
each state-year cell, a measure inspired by the AHRQ’s PQIs. AH rates, generally, and the PQIs,
specifically, are measures that are constructed using inpatient data, though meant to reflect the
quality of care prevailing in the associated outpatient / ambulatory community. Such measures
identify conditions (e.g., asthma, diabetes, malignant hypertension, etc.) with respect to which
38 proper outpatient care would have prevented the need for hospitalization. According to the
AHRQ, their PQIs grew out of research in the early 1990s by Joel Weissman and colleagues.35
The Weissman et al. (1992) AH classification scheme is designed in slightly more general terms
than the PQIs and thus arguably lends itself to easier codification using a set of NHDS records
that span several decades (considering the complexity associated with tracking variations in ICD
classifications over time).36 For this reason, and in light of the fact that Weismann et al.
developed their classification during the middle of the period in which the NCHS sampled
physicians to compile the NHDS (unlike the PQIs, which came later), I elect to construct an AH
rate for this analysis using the Weissman et al. classification.
To calculate avoidable hospitalization rates for each state and year in the sample, I first count the
number of hospitalizations within the NHDS records for that state-year cell in which a diagnosis
is indicated for any of the conditions included in the Weissman et al. (1992) classification. I
perform such counts under two alternative approaches: one in which the conditions are identified
in any one of the indicated diagnosis codes and one in which the conditions are identified in the
primary diagnosis code only. To form the relevant rate, it is of course necessary to normalize
these AH counts in some manner. Following Frakes (2012a), I elect to use measures internal to
the NHDS records to form the relevant denominator for each state-year AH rate, taking several
alternative approaches to this normalization.37 In one approach, for example, I normalize each
AH count by the number of hospitalizations associated with the delivery of a child found in the
NHDS records for the relevant state and year. This approach allows for a scaling of the AH
count by a measure reflective of the size of the associated state-year sample, while also offering
a denominator that is itself not likely to be significantly impacted by the prevailing malpractice
environment (allowing for a focus on the influence of malpractice on the AH count comprising
the numerator, our margin of interest).
Primarily, however, based on the same premise as the delivery approach and following Frakes
(2012a), I normalize each state-year AH count by an index of hospitalizations equal to the count
of admissions associated with any of the following conditions and events: (1) acute myocardial
infarction, (2) stroke, (3) gastro-intestinal bleeding or (4) hip fracture. Such events represent
situations characterized by relatively little variation across regions (see, for example, Wennberg
35
See http://www.qualityindicators.ahrq.gov/Downloads/Modules/PQI/PQI%20Summary%20Report.pdf.
Those conditions represented in the Weissman et al. (1992) classification include: ruptured appendix, asthma,
cellulitis, congestive heart failure, diabetes, gangrene, hypokalemia, immunizable conditions, malignant
hypertension, pneumonia, pyelonephritis, and perforated or bleeding ulcer. 37
The NHDS weights are not designed to generate representative state-specific estimates. Of course, observing
within-state changes over time in the set of records included in the state-year cells nonetheless affords the ability to
identify the intended relationships (Dafny and Gruber 2005). In any event, though noisier, the results of this
exercise generally persist under alternative approaches that either (1) multiply observations by the NHDS sample
weights and form AH rates by dividing weighted AH counts by the total population of that state (yet another
normalization approach), or (2) forming dependent variables based on the natural log of the state-year AH counts
(i.e., under no normalization at all). The primary approaches taken, however, soften some of the sampling
variability that occurs within states over time, while normalizing by a measure that is more directly reflective of the
scale of the hospital sampled.
36
39 1984 and Wennberg and Cooper 1999), even in the face of environments that impose varying
legal and financial incentives (i.e., where such hospitalizations are better seen as proxies for the
underlying disease environment, as opposed to reflections of immediate healthcare utilization
decisions). As such, this index likewise affords an appropriate scaling of the numerator count
with arguably little concern over the malpractice environment impacting the scaling metric.38 In
yet another alternative approach, I simply normalize by the count of acute myocardial infarction
discharges (primary diagnosis only) for the relevant state and year.
Inpatient Mortality for selected conditions. Following the AHRQ’s IQIs, I next construct a
quality measure in which I calculate the composite rate of inpatient mortality among a subsample of discharges in which the primary diagnosis code indicates any one of the following
conditions: acute myocardial infarction, heart failure, acute stroke, gastrointestinal bleeding, hip
fracture or pneumonia. Such events are generally high volume in occurrence, allowing for robust
sample sizes. It is worth noting that such conditions, for the most part, also represent lowdiscretionary hospitalizations, whereby inpatient admissions generally follow upon their
occurrence.39 With this in mind, mortality rates among this sub-sample of admissions can be
seen as more likely reflective of the quality of care observed during the inpatient stay itself,
rather than as a result of risk selection by providers or patients.
Of course, a concern arises regarding fluctuations in the proportions of the various conditions
comprising this selected-conditions sub-sample. That is, a reduction in the composite mortality
rate could arise from a relative increase in the rate of hip fracture admissions (where mortality
rates are lower for such admissions relative to the other selected conditions), as opposed to
reductions in mortalities that would actually be attributable to improvements in quality. I take
two approaches to dealing with this concern. First, in some specifications, I include state-year
controls for the proportion of this sub-sample made up of each of the respective conditions. In
the primary approach, however, I follow the AHRQ and standardize the composite mortality rate
for state-year changes in the various incidences of the conditions.
To risk adjust mortality rates, I employ an indirect standardization approach, in which I first
predict the mortality rate that a national sample of patients would be expected to experience if
they faced the relevant patient characteristics of each state-year cell. I generate such predictions
based on the estimated coefficients from national, annual regressions of mortality incidence on
the incidence of the relevant set of conditions. I then calculate the standardized mortality rate by
(1) taking the ratio between the observed state-year composite mortality rate and this predicted
national mortality rate and (2) multiplying this ratio by the observed national mortality rate.
38
See Frakes (2012a) for empirical support over the contention that the incidences of these low-variation conditions
are not sensitive to medical liability standards. Note that higher quality outpatient care may be effective at reducing
some amount of hospitalizations for the above-indicated low-variation conditions, though likely to an extent less
than quality care may reduce the incidence of the Weissman et al. (1992) avoidable conditions, in which case the
proposed avoidable hospitalization rate nonetheless identifies a relative quality measure. 39
For a discussion of the selection of low-discretionary hospitalization categories, see Carter (2003). Congestive
heart failure admissions, however, are arguably more discretionary than the others.
40