Inappropriate Medication Use and Health Outcomes in the Elderly docx

A survey type of ordered probit model was then estimated to quantify the effect of inappropriate drug use on patient self-perceived health status measured using a five-point scale poor,

Inappropriate Medication Use and Health Outcomes in the Elderly

Alex Z Fu, MS, Gordon G Liu, PhD, and Dale B Christensen, PhD

Inappropriate medication use is a major problem for the

elderly Although increasing attention has been paid to

in-appropriate prescription medication use, most previous

re-search has been limited to the investigation of prevalence and

trends Few studies provide the empirical evidence for the

adverse effect of inappropriate medication use on health

outcomes at the national level This study is the first attempt

to assess the relationship between inappropriate prescription

use and health status for the elderly in the United States

Based on the 1996 Medical Expenditure Panel Survey,

inappropriate medication use in a national representative

elderly population was first identified using Beers criteria A

survey type of ordered probit model was then estimated to

quantify the effect of inappropriate drug use on patient

self-perceived health status measured using a five-point scale

(poor, fair, good, very good, and excellent) After

control-ling for a set of possible confounding factors, it was found

that individuals using inappropriate medications in Round

1 were more likely than those not using inappropriate

med-ications to report poorer health status in Round 2 Other

risk factors for poor health status include a higher number

of prescriptions, being black, having low education, and

having one or more chronic diseases

This study provides strong evidence of a significant

adverse effect of inappropriate medication use on patient

health status These findings lend partial support to the use

of Beers criteria in assessing the quality of prescribing and

the appropriateness of medication use in the elderly

pop-ulation J Am Geriatr Soc 52:1934–1939, 2004

Key words: inappropriate prescribing; Beers criteria;

MEPS; health status; ordered probit model

Previous studies have suggested that inappropriate

pre-scription drug use is a principal cause of adverse drug

events (ADEs).1–4ADEs can lead to additional physician visits, hospitalizations, injury, deterioration of body func-tioning, and death.5–9The Food and Drug Administration has estimated that the cost of hospitalizations due to inap-propriate prescription drug use averages $20 billion annu-ally.10 Inappropriateness of drug therapy is a significant issue for the elderly, because persons aged 65 and older take many more medications than younger groups The elderly constitute approximately 13% of the population but con-sume 32% of all prescription drugs in the United States.5,11

Therefore, older persons are at increased risk for inappro-priate prescription drug use and ADEs A study in 2001 reported that more than one-fifth of the community-dwell-ing elderly in the United States received some inappropriate medication in 1996.12

Inappropriate medication can be defined as drugs that pose more risks than benefits to the patients.13,14The risk of using these medications range from minor weakness, to or-thostatic hypotension, prolonged sedation, falls and frac-tures, or even life-threatening events.14Although attention has been increasingly paid to medical errors in general, it has been a challenge for researchers to quantitatively meas-ure inappropriateness of a drug for different population cohorts in different settings In 1991, Beers et al.13 devel-oped the first set of explicit criteria to measure inappropri-ateness of drugs prescribed for nursing home elderly In

1997, Beers updated the original criteria by incorporating new information for existing and new drugs.14 The Beers criteria prompted a number of empirical studies to quantify the scale of the problem Most of these studies were focused

on the prevalence and risk analyses of inappropriate drug use in the U.S elderly population.12,15–17A few studies also addressed the associations between inappropriate drug use and patient health outcome measures.18–20However, these studies were based on local populations or healthcare settings, which limited the generalizability of their study results Moreover, potential confounding factors and cross-sectional study designs that could be problematic in attempting to establish a causal relationship between inap-propriate drug use and changes in health outcome limited some of these studies.19

Otherwise, existing literature offers little evidence of how inappropriate drug use would affect patient health outcomes, especially at the national level, a critical yard-stick to fully understand the problem and ultimately to call for appropriate policy actions In response, the present

This study was supported in part by a grant from the National Institute on

Aging (R03 AG19876-01) The opinions expressed are solely those of the

authors.

Address correspondence to Gordon G Liu, PhD, Associate Professor,

Division of Pharmaceutical Policy and Evaluative Sciences, University of

North Carolina at Chapel Hill, School of Pharmacy, Beard Hall, CB #7360,

Chapel Hill, NC 27599 E-mail: ggliu@unc.edu

From the Division of Pharmaceutical Policy and Evaluative Sciences,

University of North Carolina at Chapel Hill, Chapel Hill, North Carolina.

JAGS 52:1934–1939, 2004

2004 by the American Geriatrics Society 0002-8614/04/$15.00

study is intended to help fill the gap by quantifying the

possible adverse effect of inappropriate drug use on patient

perceived health outcomes by estimating a set of health

status models in relation to inappropriate drug use in the

elderly using the national Medical Expenditure Panel

Sur-vey (MEPS)

METHODS

Data Source

Data were extracted from the 1996 MEPS data set, which

provides nationally representative estimates of healthcare

utilization, expenditures, sources of payment, and

insur-ance coverage for the noninstitutionalized population

MEPS, maintained by the Agency for Healthcare Research

and Quality, uses an overlapping panel design in which data

are collected through preliminary contact followed by a

series of five rounds of interviews per panel with a time lag

of 4 to 5 months for about 2 years

This study used the Round 1 and 2 information within

the household component of the 1996 MEPS, initially

re-leased in 2000 This data set contains records for 22,601

persons, including a person-level weight variable that

al-lows making estimates for the noninstitutionalized U.S

population for 1996 There are also 147,308 prescribed

medicine records in the data set Each record represents one

household-reported prescribed medicine that was

pur-chased or obtained during 1996, and each record contains

medical conditions associated with the prescribed medicine

The population studied was people aged 65 and older

Analytical Approaches

Outcomes Variable

The primary outcomes variable was self-perceived health

status In MEPS, respondents were asked to rate their health

status compared with other persons of their age as excellent,

very good, good, fair, or poor This self-assessment

ap-proach has been widely used as a means of gathering

in-formation on health status in epidemiological research and

has proven to have substantial reliability and predictive

validity.21

Explanatory Variables

In the health outcomes model, the key intervention variable

is whether an individual used any inappropriate prescribed

drugs Inappropriate drug use was defined as exposure to

any drug on the updated Beers list of 26 drugs or categories

of drugs.14Because MEPS lacks information on daily

dos-age or frequency of use for every prescription, only 23

cri-teria were included in the analysis Three of the drug

categories (benzodiazepines, digoxin, and iron

supple-ments) were not included in the analysis because they are

only considered inappropriate beyond certain daily dose

In addition to inappropriate prescription use, three sets

of variables are assumed in the model to affect an

individ-ual’s health status The first set, referring as predisposing

variables, comprises age, sex, race, education, employment

status, and marital status.22

The second set comprises enabling variables such as

income, census region, Metropolitan Statistical Areas, and

health insurance Because the entire sample was eligible for

Medicare, health insurance was categorized into four

groups: Medicare only, Medicare with some private insur-ance, Medicare with other public insurance including Med-icaid, and Medicare with private and other public insurance

The third set of variables was intended to control for health conditions that can determine a person’s current health status To minimize a possible endogeneity problem with health conditions in the outcomes model, Round 1 disease condition and functional limitation variables were used as additional variables to control for preperiod health status Based on International Classification of Diseases, Ninth Revision, Clinical Modification codes, nine chronic disease conditions were included: heart disease, hyperten-sion, lung disease, cancer, stroke, diabetes mellitus, more than one of these diseases, other diseases, and no diseases Functional condition variables included help needed in ac-tivities of daily living, help needed in instrumental acac-tivities

of daily living, assistant devices used, limitations in physical functioning, work/housework/school limitation, social lim-itations, and cognitive limitations

Econometric Modeling Because of the five-scale ordinal nature of the health status variable, the ordered probit model was employed to esti-mate the discrete outcome function.23 A reduced form of the outcome function can be written as follows:

PrðHi2Þ ¼ Fða þ b
Mi1þ d1
Ni1þ d2
Hi1þ d3
Zi2

þ eiÞ where Hi2is the health status for individual i in Round 2;

Mi1is a vector indicating the status of inappropriate drug use for individual i in Round 1; Ni1 represents the total number of prescription drugs for individual i in Round 1;

Hi1is a vector consisting of Round 1 health conditions as need variables, including preexisting health status and health conditions; Zi2is a vector controlling for other ob-served factors that are in the categories of predisposing and enabling variables as noted earlier; eiis a random error term

to capture all unobserved factors; and F represents the probit cumulative distribution function

It is worth noting that the model was specified to in-clude medication use pattern in Round 1 as a key interven-tion variable to influence health status in Round 2 Conceptually, inappropriate use of a drug could have a short-run and long-run effect on one’s health status If em-ploying the Round 2 medication use pattern in the model, a potential endogeneity problem would occur This is because individuals with poor health status may need more medi-cations and thus became more likely to encounter an inap-propriate medication, all other factors being equal.17

Alternatively, the link between drug use in Round 1 and health status in Round 2 was sought, with a presumption that one’s medication use pattern in Round 1 and 2 would

be correlated This presumption was supported by the em-pirical test results, suggesting a high correlation between the medication use variables in the two rounds

In estimating the model, the complex sample design of MEPS was taken into account, featuring stratification, clustering, multiple stages of selection, and disproportion-ate sampling This results in serious departures from simple random sampling assumptions To obtain valid estimates of

the standard errors for the model, the design complexities

were taken into account by specifying the variance

estima-tion strata and the variance estimaestima-tion primary sample

units in a survey type of ordered probit model.24Data

cod-ing and econometric modelcod-ing were conducted uscod-ing SAS

version 8 (SAS Institute, Inc., Cary, NC) and Stata version 6

(Stata Corp., College Station, TX), respectively

RESULTS

Descriptive Statistics

After excluding observations with missing variables, the

fi-nal sample derived from the 1996 MEPS included 2,305

elderly, representing 30,485,175 elderly nationwide, with

weight adjustment Of the original sample in Round 1,

there were 537 nonusers of any prescription drugs, 1,462

appropriate medicine users, and 306 inappropriate

medi-cine users (Number of inappropriate prescriptions ranged

from 1 to 12, with mean 2.2 and median 2.) The most

frequently inappropriately used drugs were propoxyphene

and those containing propoxyphene In Round 1, the

dis-tribution of perceived health status was excellent, 19.1%;

very good, 25.7%; good, 28.7%; fair, 18.5%; and poor,

8.0% Similarly, in Round 2, the distribution was excellent,

18.4%; very good, 26.7%; good, 28.0%; fair, 18.7%; and

poor, 8.3% Chi-square (w2) test results confirmed that there

was no difference between the two distributions (w2

(4) 5 1.036, P 5.85)

Elderly persons who used inappropriate prescriptions

in Round 1 were more likely than those who did not to

perceive fair or poor health status in Round 2 Chi-square

test indicated that the postperiod health status was

signif-icantly associated with inappropriate prescription drug use

in the preperiod (Po.001)

Modeling Strategies

In an effort to minimize the effect of potential confounding

factors, a sequential modeling strategy was employed by

adding one additional group of explanatory variables at a

time in specifying the models Model A in Table 1 reports

the ordered probit model results when the equation

includ-ed only prescription drug use The inappropriate drug use

group had significantly worse health status and the nonuse

group had significantly better health status, both compared

with the group with appropriate drug use

Models B to D sequentially added in additional control

variables After including the total number of prescriptions

in Model B, the coefficient of inappropriate drug use de-creased substantially (from  0.517 to  0.330) toward the null, indicating the strong confounding effect of the total number of prescriptions taken Adding predisposing and enabling variables in Model C, the coefficient remained virtually the same as before, indicating that these two sets of variables would not confound or modify the estimates However, after adding the need variables in Models D to F, the coefficient of inappropriate drug use decreased steadily, strongly suggesting that the need variables would otherwise bias the estimates upward if not controlled This is consist-ent with the hypothesis that no difference in health status would exist between the elderly with appropriate medica-tion use and those without any medicamedica-tion use after con-trolling for preexisting health conditions

Compared with Model D, Model E replaced the Round

1 health status with health-condition variables The de-crease in the coefficient of interest (from  0.284 to

 0.235) implies that health condition variables are more likely to control for the underlying confounding problem and therefore help minimize the endogeneity The full Model F adds in the prior health-status and health-condi-tion variables The coefficient of inappropriate prescriphealth-condi-tion drug use becomes more stabilized at  0.235 to  0.234 across the different model specifications, suggesting the relative robustness and reliability of the full Model F In specifying the full model, there was a concern about mul-ticollinearity within the right-hand side variables in the model Accordingly, a pair-wise correlation analysis was performed, which confirmed that multicollinearity is not a problem for the study results Thus the full model results are reported in the following sections (Table 2)

Explanatory Results Compared with those with appropriate prescription drug use, the elderly with inappropriate drug use in Round 1 reported significantly worse health status in Round 2 (Po.01) The total number of prescriptions was strongly associated with poorer health status (Po.01) As for the predisposing variables, older blacks had significantly

poor-er health status than whites (Po.05) Elderly with higher educational degree had better health status than those with less education (Po.05) Of the enabling variables, none was found to be predictive of health status, including health insurance

As expected, need variables were most highly

correlat-ed with perceivcorrelat-ed health status Round 1 health status was

Table 1 Effect of Inappropriate Prescription Use on Health Status

Prescription

Drug Use

Model

Coefficient Standard Error Inappropriate  0.517  0.082w  0.330  0.088w  0.347  0.087w  0.284  0.083  0.235  0.087  0.234  0.084

No use 0.508 0.062w 0.237 0.068 0.241 0.069 0.103 0.073  0.020  0.096  0.085  0.097

Note: Appropriate drug use is the reference group Control variables were total number of prescriptions for Models B, C, D, E, and F; predisposing and enabling variables for Models C, D, E, and F; self-perceived health status for Models D and F; and health condition variables for Models E and F.

Po.01; w

Po.001.

highly predictive of Round 2 health status, with a

decreas-ing trend in magnitude of the coefficients from excellent to

fair, relative to people with poor health status (Po.001)

Likewise, disease condition variables were strongly

predic-tive of health status Elderly patients with heart disease

(Po.01), hypertension (Po.001), lung disease (Po.05),

diabetes mellitus (Po.001), more than one of the

men-tioned diseases (Po.001), or any other disease conditions

(Po.01) in Round 1 were much more likely to report

poor-er health than those free of any chronic conditions

More-over, poorer health was shown to be associated with having

limitations in physical (Po.001) and cognitive (Po.01) function

The effect of inappropriate drug use on health status based on the estimated coefficient of  0.234 in Table 2 will now be focused on Because ordered probit model does not allow a straightforward interpretation of its coefficients as linear models do, the following simulation procedures were undertaken Holding all controlled variables in the model at their actual value for each individual, the probability of having each health status was predicted while varying the dummy value of inappropriate prescription use These

Table 2 Ordered Probit Model of Health Status for the Elderly

Primary interested

Prescription drug use

Appropriate (Reference)

Predisposing

Race (reference: white)

Higher degree of education (reference: no

degree)

Need

Self-perceived health status (reference: poor)

Disease condition (reference: no disease condition)

Functional and activity limitations

Note: The model was estimated using the 1996 MEPS data with 135 strata, 537 clusters, 2,305 of observations, and the sum of national weights equal to 30,485,175.

 Only variables (variable groups) with significant coefficients are listed Variables included in the model but excluded from the table include age, sex, general equivalency diploma, high school, other degree, employment status, marital status, income, census region, metropolitan statistical areas, health insurance, stroke, cancer, activities of daily living help, instrumental activities of daily living help, assistance devices used, work/housework/school limitation, social limitation Details about the statistics are available upon request from the authors.

w

American Indian, Aleut, Eskimo, Asian or Pacific Islander.

Table 3 Distribution of Health Status in Round 2 by the Use of Prescriptions in Round 1

Round 1

Round 2 Health Status

Prescription Use

Mean Percentage of Population Standard Error Inappropriate 15.9 0.50 25.3 0.38 28.6 0.32 20.2 0.40 10.0 0.46

probabilities were then averaged over the sample to provide

interpretable results, as in Table 3 Based on these results, if

the cohort of appropriate drug users used some

inappro-priate prescriptions, 15.9% of the cohort would perceive

health status as excellent and 10% poor Similarly, if all

inappropriate drug users used appropriate prescriptions,

their perceived health status would be distributed as

excel-lent for 18.7% and poor for 7.9% This provides sizable

evidence to policy-makers that, keeping other variables

constant, inappropriate drug use would lead to a decrease in

patient health status

DISCUSSION

Although previous studies suggested a widely believed

no-tion that inappropriate drug use would have adverse effects

on patient health outcomes, especially in the elderly, this

study adds to the literature by providing the first empirical

evidence documenting such an effect for the national elderly

population derived from the 1996 MEPS data After

con-trolling for a number of confounding factors in the model,

this study found strong evidence of a sizable and consistent

negative effect of inappropriate prescription drug use on

patient health status

A few other findings are noteworthy Among the

pre-disposing variables, there was a significant racial disparity

in health status, in that African Americans had poorer

health status than their white counterparts Previous

liter-ature has suggested that African Americans have higher

rates of heart disease, stroke, most cancers, pneumonia,

influenza, chronic liver disease, and diabetes mellitus than

whites.25,26 Given these findings, it is not surprising that

this study found a similar pattern of racial disparity in

self-perceived health status It was also found that higher

education was associated with better health status This

finding is consistent with the well-known notion that

ed-ucation and health, the two most important elements of

human capital, are mutually enhanced in the household

production process,27–29but healthier people tend to invest

more in education in an attempt to recoup more returns

from the investment Better-educated people in return are

likely to be more productive and efficient in maintaining

health and using healthcare resources

The model showed little dependence of health status on

enabling variables, including income, insurance type, and

regional settings It could be the case that none of the

var-iables played a substantial role in determining health status,

but it cannot be excluded that the insignificant relationships

were partly attributed to measurement errors of the

vari-ables in the database Poor measurement leads to

attenu-ation bias, but it is also possible that patients who were

prescribed inappropriate medications would receive

med-ical care below acceptable standards or quality, which could

negatively affect self-perceived health This possible effect

could not be controlled for in the model because of data

limitation

As expected, need variables were the most predictive of

health status outcomes Round 1 health status was most

highly correlated with Round 2 health status, following a

decreasing trend of the quantitative coefficients from

ex-cellent to fair, but to some extent, this variable is subject to

an endogeneity bias, as discussed earlier In contrast,

health-condition variables, including chronic disease health-conditions and functional activity limitations, were deemed to be more objective in measuring need for care and less endogenous in the determination of self-perceived health status Other un-observables associated with the preperiod health status in the error term could potentially bias the estimate, but a consistent set of coefficients was obtained across all the model specifications to indicate the strong effect of a

chron-ic condition on self-perceived health status Therefore, the potential bias due to unobservables does not seem to be a major concern Although it is difficult to fully eliminate the estimation bias, the sequential modeling strategy seemed to help reach the robustness of the estimates as the coefficient

of inappropriate drug use converges to a stable value

In addition, MEPS does not provide information on daily dosage or frequency of use for every prescription As a result, some drugs had to be excluded from the model be-cause of conditional inappropriateness In particular, ben-zodiazepines, digoxin, and iron supplements were not included in the analysis because they would be inappropri-ate only when their dosage exceeded a certain amount This limitation would lead to the underestimation of the pop-ulation size of inappropriate prescription drug use, but be-cause of the low severity of violating these criteria, the expected outcome may not be sensitive to such limitation.14

CONCLUSION This study is the first empirical research on inappropriate prescription drug use and health outcomes in the elderly based on a nationally representative population sample Strong evidence was obtained indicating that inappropriate drug use is associated with subsequent decrease in self-per-ceived health status in the elderly These results lend partial support to the Beers criteria as valid proxy indicators of poor prescribing practices leading to deterioration of health status Owing to the national representation design of MEPS, the results can be generalized to the national elderly population in the United States Previous research suggested that inappropriate drug use would cause an increased use of healthcare resources The current study provides an impor-tant addition to the literature suggesting that inappropriate drug use also can have sizable adverse effects on patient health outcomes Together, these studies suggest that inap-propriate prescription drug use can be costly in healthcare resources and health outcomes Future research can be di-rected to better address why and how inappropriate med-ication use can lead to poor outcomes With such important knowledge, better policy actions can be taken to reduce system medical errors and improve the quality of health care for the elderly.30

REFERENCES

1 Lindley CM, Tully MP, Paramsothy V et al Inappropriate medication is a major cause of adverse drug reactions in elderly patients Age Ageing 1992;21:294–300.

2 Bates DW Frequency, consequences and prevention of adverse drug events.

J Qual Clin Prac 1999;19:13–17.

3 Bates DW, Cullen DJ, Laird N et al Incidence of adverse drug events and potential adverse drug events: Implications for prevention ADE Prevention Study Group JAMA 1995;274:29–34.

4 Bates DW, Leape LL, Petrycki S Incidence and preventability of adverse drug events in hospitalized adults J Gen Intern Med 1993;8:289–294.

5 Chrischilles EA, Foley DJ, Wallace RB et al Use of medications by persons 65

and over: Data from the established populations for epidemiologic studies of

the elderly J Gerontol 1992;47:M137–M144.

6 Johnson JA, Bootman JL Drug-related morbidity and mortality and the

economic impact of pharmaceutical care Am J Health Syst Pharm 1997;54:

554–558.

7 Ray WA, Fought RL, Decker MD Psychoactive drugs and the risk of injurious

motor vehicle crashes in elderly drivers Am J Epidemiol 1992;136:873–883.

8 Col N, Fanale JE, Kronholm P The role of medication noncompliance and

adverse drug reactions in hospitalizations of the elderly Arch Intern Med

1990;150:841–845.

9 Phillips DP, Christenfeld N, Glynn LM Increase in US medication-error deaths

between 1983 and 1993 Lancet 1998;351:643–644.

10 Prescription Drugs and the Elderly FMany Still Receive Potentially Harmful

Drugs Despite Recent Improvements (Report no GAO/HEHS-95–152).

Washington, DC: U.S General Accounting Office, 1995.

11 Soumerai SB, Ross-Degnan D Inadequate prescription-drug coverage for

Medicare enrollees Fa call to action N Engl J Med 1999;340:722–728.

12 Zhan C Potentially inappropriate medication use in the community-dwelling

elderly: Findings from the 1996 Medical Expenditure Panel Survey JAMA

2001;286:2823–2829.

13 Beers MH, Ouslander JG, Rollingher I et al Explicit criteria for determining

inappropriate medication use in nursing home residents: UCLA Division of

Geriatric Medicine Arch Intern Med 1991;151:1825–1832.

14 Beers MH Explicit criteria for determining potentially inappropriate

medi-cation use by the elderly: An update Arch Intern Med 1997;157:1531–1536.

15 Willcox SM, Himmelstein DU, Woolhandler S Inappropriate drug prescribing

for the community-dwelling elderly JAMA 1994;272:292–296.

16 Golden AG, Preston RA, Barnett SD et al Inappropriate medication

prescrib-ing in homebound older adults J Am Geriatr Soc 1999;47:948–953.

17 Liu GG, Christensen DB The continued challenge of inappropriate prescribing

in the elderly: An update of the evidence J Am Pharm Assoc 2002;42:847–

857.

18 Chin MH, Wang LC, Jin L et al Appropriateness of medication selection for older persons in an urban academic emergency department Acad Emerg Med 1999;6:1232–1242.

19 Fick DM, Waller JL, Maclean JR et al Potentially inappropriate medication use in a Medicare managed care population: Association with higher costs and utilization J Manag Care Pharm 2001;7:407–413.

20 Hanlon JT Impact of inappropriate drug use on mortality and functional status in representative community dwelling elders Med Care 2002;40: 166–176.

21 LaVange LM, Silverman HA Outpatient prescription drug utilization and expenditure patterns of noninstitutionalized aged Medicare beneficiaries Natl Med Care Util Expend Surv B 1987;April:1–43.

22 Aday LA, Andersen R A framework for the study of access to medical care Health Serv Res 1974;9:208–220.

23 Greene WH Econometric Analysis Upper Saddle River, NJ: Prentice-Hall Inc., 2000.

24 Lohr SL Sampling Design and Analysis Pacific Grove: CA: Duxbury Press, 1999.

25 Pamuk E, Makuc D, Heck K et al Health, United States 1998: Socioeconomic Status and Health Chartbook (DHHS no 98–1232) Hyattsville, MD: Na-tional Center for Health Statistics, 1998.

26 Winkleby MA, Robinson TN, Sundquist J et al Ethnic variation in cardio-vascular disease risk factors among children and young adults: Findings from the Third National Health and Nutrition Examination Survey, 1988–94 JAMA 1999;281:1006–1013.

27 Grossman M On the concept of health capital and the demand for health.

J Political Econ 1972;80:223–255.

28 Schultz TP Health and schooling investments in Africa J Econ Perspect 1999; 13:67–88.

29 Bloom DE, Canning D The health and wealth of nations Science 2000;287: 1207–1209.

30 Institute of Medicine To Err Is Human: Building a Safer Health System Washington, DC: National Academic Press, 2000.