SEARCH

SEARCH BY CITATION

Keywords:

  • Cost-effectiveness;
  • Heart failure;
  • Angiotensin converting-enzyme inhibitor

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. Acknowledgements
  8. References

Objective

A cost-effectiveness analysis of high and low doses of the angiotensin-converting enzyme (ACE) inhibitor lisinopril in the treatment of chronic heart failure.

Methods

A cost-effectiveness analysis using data from a randomized controlled trial, ATLAS, where 3164 patients with chronic heart failure were allocated to a high-dose (daily target dose 32.5–35 mg) or low-dose strategy (daily target dose 2.5–5.0 mg) of lisinopril. Differential costs were based on resource use data collected in the trial costed using UK unit costs. Cost-effectiveness analysis related differential costs to differential life-years during a 4-year trial follow-up.

Results

The mean total number of hospital in-patient days per patient was 18.5 in the high dose group and 22.5 in the low dose group. Over the whole duration of the trial, the mean (S.D.) daily dose of lisinopril in the high-dose group was 22.5 mg (15.7mg) compared to 3.2 mg (2.5 mg) in the low-dose group. The mean difference in cost per patient was £397 lower in the high-dose group [95% CI (high-dose–low-dose) −£1263 to £436]. Mean life-years per patient were 0.085 years higher in the high-dose group [95% CI (high-dose–low-dose) −0.0074 to 0.1706). Based on mean costs and life-years, high-dose therapy dominates low-dose (less costly and more effective). Allowing for uncertainty in mean costs and life-years, the probability of high-dose therapy being less costly than low dose was 82%. If a decision maker is willing to pay at least £3600 per life-year gained, the probability of high-dose being more cost-effective was 92%.

Conclusions

The ATLAS Study showed that the treatment of heart failure with high-doses of lisinopril has a high probability of being more cost-effective than low-dose therapy.


1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. Acknowledgements
  8. References

Large clinical trials have shown angiotensin converting-enzyme (ACE) inhibitors to reduce the risk of death and hospitalization in chronic heart failure [[1],[2]]. There is also evidence that the use of ACE inhibitors represents a cost-effective use of resources [[3]]. Despite guidelines recommending their use in patients with heart failure due to left ventricular systolic dysfunction [[4][6]], many patients using ACE inhibitors are prescribed smaller dosages than those used in the trials that established their effectiveness [[7],[8]].

The Assessment of Treatment with Lisinopril and Survival (ATLAS) study was an international trial undertaken to compare low doses and high doses of the ACE inhibitor lisinopril (Zestril™) in the treatment of chronic heart failure. The study randomized 3164 heart failure patients to double-blind treatment with either high (32.5–35.0 mg) or low (2.5–5.0 mg) doses of lisinopril and found a trend towards lower mortality (an 8% relative risk reduction) and a statistically significantly lower risk of death or hospitalization in the high-dose group (a 12% relative risk reduction) with no important increase in adverse events [[9]].

Although the trial indicates a strong clinical case for high doses of ACE inhibitors, it is important to assess their cost-effectiveness. In addition to mortality and morbidity data, information on the key health service resources consumed by ATLAS study patients was also collected. The results of an economic sub-study, where these resource-use data were used to estimate the differential cost and cost-effectiveness of high- and low-dose lisinopril on the basis of UK health care costs, are presented here. The analysis is based on data collected over 4 years of patient follow-up.

2. Methods

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. Acknowledgements
  8. References

2.1. Trial design

The design, baseline characteristics and clinical results of the ATLAS study have been published elsewhere [[9]]. In brief, the trial was undertaken in 19 countries with the USA recruiting the most patients (38%). Patients with New York Heart Association (NYHA) Class II–IV heart failure and left ventricular ejection fraction equal to or below 30% were recruited. Exclusion criteria included acute myocardial infarction, unstable angina or a revascularization procedure in the preceding 2 months; the presence of symptomatic ventricular tachycardia; unstable congestive heart failure; and the use of various negatively or positively isotropic drugs. Details of the key baseline characteristics of patients in the trial are given in Table 1.

Table 1. Baseline characteristics
CharacteristicHigh doseLow dose
 (n=1568)(n=1596)
  • a

    Due to rounding, these figures do not sum to 100%.

Mean (S.D.) age (years) 63.6 (10.5) 63.6 (10.3)
Male n (%)1251 (79.8)1265 (79.3)
New York Heart Association  
class [n (%)]a  
 Class II 262 (16.7) 231 (14.5)
 Class III1194 (76.1)1252 (78.4)
 Class IV 112 (7.1) 113 (7.0)
Aetiology of heart failure  
[n (%)]  
 Coronary artery disease 999 (63.7)1036 (64.9)
Mean (S.D.) left ventricular 22.6 (5.7) 22.6 (5.6)
 ejection fraction (%)  
Mean (S.D.) systolic blood 125.3 (19.3) 125.9 (19.6)
 pressure (mmHg)  
Previous use of ACE1390 (88.6)1420 (89.0)
 inhibitor [n (%)]  

Following an initial open-label period to establish tolerance of a daily dose of 12.5 to 15 mg of the ACE inhibitor, lisinopril, 3164 patients were randomly allocated, double-blind, to a high-dose strategy (daily target dose 32.5–35.0 mg, n=1568) or low-dose strategy (daily target dose 2.5–5.0 mg, n=1596) of lisinopril. Physicians could prescribe open-label ACE inhibitor if a patient's condition deteriorated, after which dosing with the study medication could be maintained, reduced or discontinued at the investigator's discretion. After a median follow-up of 46 months (range 39–58 months), there was an 8% lower risk of death in the high-dose group (P=0.128). Patients in the high dose group also had a statistically significant 12% lower risk of death or hospitalization for any reason (P=0.002). Although patients in the high-dose group experienced more events related to hypotension and renal impairment, these adverse events were generally successfully managed by adjusting therapy. The proportions of patients stopping medication due to adverse events were similar (17% high-dose; 18% low-dose). Heart failure symptoms, as measured by NYHA class, improved with both high- and low-dose therapy, with no between-group differences during the study.

2.2. Resource-use measurement

The study adopted a health service perspective and focused on two key areas of resource-use that were expected to drive cost differences: days in hospital and drug use. Data were collected on the number and cause of hospital in-patient days and day-case visits (defined as visits to hospital without an overnight stay). Details of each patient's dose of study medication were collected over the full period of follow-up, enabling the calculation of the total amount of study medication taken. Information on concomitant ACE inhibitor use was also collected. Given the absence of any clinically important difference in the incidence of adverse events, use of concomitant medication or degree of symptomatic relief [[9]], the cost of other concomitant drugs has not been included in this analysis.

2.3. Valuing resource-use

The differential cost of managing patients in the two arms of the trial has been estimated using the resource-use measured in the trial and 1997–8 unit costs from English hospitals. On the basis of the reason(s) for hospitalization, each day in hospital was allocated to a specific speciality (e.g. if a patient was admitted to hospital due to renal failure, those days were allocated to nephrology). Days in hospital have then been costed using the average cost per in-patient day — or the average out-patient cost per attendance for day-case visits — in English hospitals based on their financial returns in 1997–8 [[10]]. These unit costs cover all resources consumed, on average, during a day in hospital, including diagnostic and therapeutic procedures. When there was more than one reason for hospitalization, the days were allocated to the highest cost speciality.

The doses of study drug, lisinopril, have been costed, using British National Formulary prices [[11]], in terms of the minimum number of tablets needed to achieve the dose prescribed: £0.27 per 2.5 mg tablet, £0.34 per 5 mg tablet, £0.42 per 10 mg tablet and £0.48 per 20 mg tablet. Although a 30 mg tablet is available in the USA and some other European countries, this is not the case in the UK, so this dose has been costed on the basis of a 20 mg and a 10 mg tablet reflecting likely dosing in the NHS. The use of open label ACE inhibitors was costed in a similar manner depending on the ACE inhibitor used. Patients in the high-dose group were assumed to need three additional visits to a general practitioner for dose titration; these additional visits have been costed at £14 each [[12]].

2.4. Analysis

Patients in the trial were followed up for differential periods, so costs have been analysed using the method developed by Lin et al. [[13]]. Mean costs are presented, by study group, for hospitalizations (in-patient and day-case), study drug, open-label ACE inhibitors and in total. Mean differences in costs are also presented with 95% CI based on the 2.5 and 97.5 percentiles generated by a bootstrapping exercise [[14]].

For the cost-effectiveness analysis, estimates of the mean life expectancy in the two arms of the trial are required. This was based on the Kaplan–Meier survival curves as mean time until death, with confidence intervals again based on the 2.5 and 97.5 percentiles of a bootstrapping exercise.

To allow for differential timing of resource consumption and clinical events, discounting is used in economic evaluation to generate a present value of future costs and effects [[15]]. For this purpose, the UK Department of Health's recommended annual discount rate of 6% for costs and 2% for effects is used [[15]]. Sensitivity analysis is used to explore the implications of using a 6% discount rate for costs and effects and of no discounting.

2.5. Cost-effectiveness analysis

One treatment can be defined as more cost-effective than its comparator if one of the following four conditions apply: (a) it is less costly and at least as effective; (b) it is more effective and no more costly; (c) it is more costly and more effective but its additional cost per extra unit of effectiveness is considered worth paying by decision makers; and (d) it is less costly and less effective but the additional cost per extra unit of effectiveness of its comparator is not considered worth paying by decision makers.

Here, cost-effectiveness is assessed by relating the differential cost per patient of alternative treatment strategies based on high- and low-dose lisinopril to their differential effectiveness in terms of life-years per patient measured during the trial. As a first step, a deterministic analysis is undertaken where the differences (between high- and low-dose) in mean costs and mean life-years are compared to consider which of the four conditions listed above applies. As a second stage, a stochastic analysis considers the uncertainty around the estimates of mean costs and mean effects. The focus of the stochastic analysis is to present the probability that high-dose therapy is more cost-effective than low-dose. This is achieved by the use of bootstrapping which re-samples a large number of times (here 1000) from the patient-specific cost and life-year data in the trial. This generates a distribution of mean differences in cost and mean differences in life-years between the high- and low-dose therapies. From this distribution, the probability that high-dose therapy satisfies one of the four conditions detailed above is calculated subject to a range of possible maximum values that a decision maker might be willing to pay for an additional life-year in this patient group. These results are presented in the form of a cost-effectiveness acceptability curve which shows the probability of high-dose therapy being more cost-effective than low-dose [[16],[17]]. This is a Bayesian approach to the presentation of cost-effectiveness data [[18]], although a full Bayesian analysis has not been undertaken.

3. Results

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. Acknowledgements
  8. References

3.1. Resource-use

A total of 28 941 in-patient days were spent in hospital by patients in the high-dose group compared to 35 906 in the low-dose group. The mean total number of in-patient days in hospital per patient was 18.5 in the high-dose group and 22.5 in the low-dose group; with a median of 6 (inter-quartile range 3–11) in both groups. Table 2 shows how these were broken down by specialty, showing that cardiology in general and heart failure in particular were the main reasons for hospitalization and where the largest difference between the two forms of therapy emerged. The table also indicates the cost per in-patient day that has been used in the costing for each of the main specialties. In addition, the total number of day-cases was 600 in the high-dose group and 698 in the low-dose group (mean number per patient 0.38 and 0.44, respectively). Over the whole duration of the trial, the mean (S.D.) and median daily dose, respectively, of lisinopril in the high-dose group was 22.5 mg (15.7 mg) and 32.5 mg, compared with 3.2 mg (2.5 mg) and 5.0 mg in the low-dose group.

Table 2. Details of the number and unit costs of days in hospital by speciality
SpecialityaCost perMean days in hospital per patient (median,
 day (£)inter-quartile range days per hospitalization)b
    
  High doseLow dose
  (n=1568)(n=1596)
  • a

    Only specialities with a mean number of days in hospital of 0.5 or greater are shown. For patients hospitalized for more than one reason, the event has been allocated to the most costly specialty (per patient-day).

  • b

    The mean days per patient include patients with no hospitalization. The median and inter-quartile range figures relate to hospitalizations rather than patients.

  • c

    Individual costs were assigned to each speciality in this group.

  • d

    Due to rounding, the totals are not the exact sum of the numbers in the columns.

Cardiology (total)393 9.5 (6, 3–11)12.2 (6, 3–10)
 Heart failure393 6.0 (7, 4–12) 8.1 (6, 4–11)
 Other cardiology393 3.5 (5, 2–9) 4.1 (4, 2–9)
Other medical161 2.3 (5, 2–9) 3.0 (5, 3–11)
General surgery257 2.3 (6, 3–12) 2.5 (6, 3–12)
Thoracic medicine211 1.1 (6, 4–10.5) 1.5 (6, 3–11)
Orthopaedics255 0.6 (8, 4–15) 0.7 (7, 4–15)
Cardiothoracic523 0.5 (8, 4–12) 0.3 (9, 3.5–14)
OthersVariousc 2.2 (5, 2–11) 2.4 (5, 3–13)
TotaldVariousc18.5 (6, 3–11)22.5 (6, 3–11)

3.2. Costs and effects

Table 3 shows the estimated costs per patient in the two arms of the trial. Although the mean cost of lisinopril was higher in the high-dose group, this additional mean cost was more than offset by the lower number of hospital in-patient days in that group. The mean difference in cost per patient was £397 lower (that is, a saving) in the high-dose group (95% CI high-dose minus low-dose: −£1263 to £436).

Table 3. Mean costs and mean cost differences in the high and low dose groups
ComponentMean cost per patient (£)Mean cost difference per
    
 High doseLow dosePatient (95% CI) (£)
   (high-dose minus low-dose)
  • a

    This includes the cost of a visit to a general practitioner for dose titration in the high-dose group.

  • b

    Due to rounding, the total is not the exact sum of the column.

Days in hospital57696917−1148 (−2013 to −305)
Day casesa 63 27  36 (27 to 43)
Trial drug (lisinopril)1004 277  727 (697 to 758)
Open label ACE 32 45 −13 (−26 to −2)
Totalb68677264 −397 (−1263 to 436)

These results are not sensitive to the choice of discount rate employed: if a 3% rate is used, as has been recommended in the USA [[19]], the mean cost difference increases to £414; if costs are not discounted the figure is £465.

On the basis of a 2% discount rate, the mean life-years per patient in the high-dose group were 2.980 compared with 2.896 in the low-dose group, a mean difference of 0.085 years [95% CI (high-dose–low-dose) −0.0074 to 0.1706].

3.3. Cost-effectiveness

On the basis of a deterministic comparison of mean costs and mean life-years, high-dose therapy is both less costly and more effective than low-dose. On this basis, high-dose would be considered a dominant treatment and unequivocally more cost-effective.

However, mean costs and life-years are estimated with uncertainty. To reflect this uncertainty, Fig. 1 presents a scatterplot of the mean differences per patient in cost and life years between high- and low-dose therapies estimated by repeated sampling as part of the bootstrapping exercise. The dotted lines divide the graph into four quadrants. The concentration of points in the bottom right quadrant shows the relatively high probability (79%) of high-dose therapy both being less costly and more effective than low-dose (i.e. dominant).

image

Figure 1. Scatterplot showing the mean differences in cost and in life-years based on 1000 re-samples from the trial data generated by the bootstrapping exercise.

Download figure to PowerPoint

On the basis of the bootstrapped data shown in Fig. 1, Fig. 2 presents cost-effectiveness acceptability curves which indicate the probability of high-dose lisinopril being more cost-effective than low-dose. The figure shows that, for the base-case analysis where costs are discounted at 6% and effects at 2% per annum, the probability of high-dose being less costly than low-dose (that is, the probability of being cost-effective when the decision maker is unwilling to pay anything extra for an additional life-year), is 82%. If a decision maker is willing to pay at least £3600 per life-year gained, the probability of high-dose being more cost-effective increases to 92% with base-case discount rates. As the figure shows, the acceptability curves do not alter appreciably when alternative discount rates are used.

image

Figure 2. Cost-effectiveness acceptability curves showing, for alternative discount rates, the probability of high-dose therapy being more cost-effective than low-dose therapy for a given willingness to pay for an additional life-year.

Download figure to PowerPoint

4. Discussion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. Acknowledgements
  8. References

On the basis of the results presented here, high-dose therapy results in an overall mean cost saving of £397 per patient over 4 years follow up. Although the drug cost of using high-dose lisinopril was higher, this was offset by a reduction in the costs of hospitalizations. In the paper reporting the clinical results of the ATLAS study [[9]], an 8% lower risk of death was reported in the high-dose group, although this did not reach statistical significance (P=0.128). When this mortality effect is translated into life-years gained over a 4-year period of follow-up (i.e. the area between the two survival curves observed in the trial) with an annual discount rate of 2%, high-dose therapy was found to generate 0.085 mean additional years of life.

How should these data be interpreted by health service decision makers? Clinical decision makers will take note of the superior outcomes resulting from the higher dose. In particular, the statistically significant reduction in death or hospitalization [[9]]. From a resource allocation perspective, a comparison of mean costs and effects indicates that high-dose therapy dominates low-dose (lower costs, higher life-years). In the face of uncertainty in mean costs and life-years, decision makers are likely to be interested in the probability that high-dose therapy is more cost-effective than low-dose. The analysis presented here suggests that the overall probability of high-dose being more cost-effective is high. If decision makers are only interested in costs, and they do not value an improvement in patients’ life expectancy, the probability of being cost-effective (i.e. cost saving) is 82%. If the mean cost saving were to apply to the UK in general and assuming a 1% prevalence of heart failure and that 25% of these patients receive an ACE inhibitor [[20]], this would suggest an aggregate saving of approximately £60 million to the health service over 4 years.

Given the purpose of the health service, we know that decision makers do value additional life-years. No monetary value has officially been stated in the UK, although values of between Can$20 000 (£8000) and Can$100 000 (£40 000) per quality-adjusted life-year gained were suggested in Canada in 1992 [[21]]. The National Health Service has funded many interventions with implied values within this range [[22]]. The analysis shown in Fig. 2 indicates that, if decision makers place a value on a life-year gained of at least £3600, the probability of high-dose therapy being cost-effective is 92%. Even if decision makers are unclear what value they place on an additional life-year, the fact that the probability of high-dose being cost saving is appreciably higher than 50% would suggest that this is the optimum therapy for the health service. This is because the statistical distribution underlying Fig. 2 is normally distributed, so a 50% decision rule is consistent with maximizing expected health gain (in terms of life-years) from limited resources.

This analysis is based on UK unit costs to value resource consumption observed in trial patients. Whilst the USA and some European countries have a 30 mg lisinopril tablet available, this is not the case in the UK. Therefore, the acquisition cost of a 30 mg dose is based on the prices of a 20 mg and 10 mg tablet. This is likely to result in a higher cost of study medication in the high-dose group than if a competitively priced 30 mg tablet was available.

As an international clinical trial, patients were recruited from centres in 19 different countries. Hence, health care was delivered in a range of different health care systems, and it is recognized that important variations exist between countries in practice patterns and hence resource-use [[23]]. The focus of this paper was to assess the implications of the ATLAS trial for resource allocation in the UK health service, but an issue in this form of analysis is whether there are systematic differences between trial centres and countries in resource-use and hence costs. Given the dominance of the USA in recruiting 38% of patients into the trial, with the remainder shared among 18 other countries, the numbers of patients recruited in the majority of countries are too small to justify country-specific analyses. Although there may be some risk of heterogeneity between centres in health care resource-use and costs, the absence of treatment-centre interactions in the clinical analysis and the high levels of probability of high-dose therapy being the more cost-effective reported here suggests that this is unlikely to affect the conclusions of the analysis.

In conclusion, the ATLAS Study showed that the treatment of heart failure with high-doses of lisinopril has a high probability of being more cost-effective than low-dose therapy.

Acknowledgements

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. Acknowledgements
  8. References

This study was supported by AstraZeneca. Dr Andrew Briggs, Karina Hornby, Dr David Meddis and Professor Tony O'Hagan provided valuable input into the analysis but all views, and any errors, are the responsibility of the authors.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. Acknowledgements
  8. References
  • 1
    The Consensus Trial Study GroupEffects of enalapril on mortality in severe congestive heart failure: results of the Cooperative North Scandanavian Enalapril Survival StudyNew Engl J Med198731614291435
  • 2
    The SOLVD InvestigatorsEffect of enalapril on survival in patients with reduced left ventricular ejection fractions and congestive heart failureNew Engl J Med1991325292302
  • 3
    GlickH.CookJ.KinosianB. et al. Costs and effects of enalapril therapy in patients with symptomatic heart failure: an economic analysis of the Studies of Left Ventricular Dysfunction (SOLVD) treatment trialJ Cardiac Failure19951371380
  • 4
    The Task Force of the Working Group on Heart Failure of the European Society of CardiologyThe treatment of heart failureEur Heart J199718736753
  • 5
    EcclesM.FreemantleN.MasonJ.North of England evidence based development project: guideline for angiotensin converting enzyme inhibitors in primary care management of adults with symptomatic heart failureBr Med J199831613691375
  • 6
    Advisory Council to Improve Outcomes Nationwide in Heart FailureConsensus recommendations for heart failureAm J Cardiol199983Suppl 2A1A38
  • 7
    PackerM.Do angiotensin-converting enzyme inhibitors prolong life in patients with heart failure treated in clinical practice?J Am College Cardiol199633316701676
  • 8
    ClelandJ.G.Poole-WilsonP.A.ACE inhibitors for heart failure: a question of doseBr Heart J199472Suppl 3S106S110
  • 9
    PackerM.Poole-WilsonP.A.ArmstrongP.W. et al. Comparative effects of low doses and high doses of the angiotensin converting-enzyme inhibitor, lisinopril, on mortbidity and mortality in chronic heart failureCirculation199910023122318
  • 10
    The Chartered Institute of Public Finance and Accountancy (CIPFA)The Health Service Database 19981998CroydonCIPFA
  • 11
    British National Formulary, Number 361998LondonBritish Medical Association and the Royal Pharmaceutical Society of Great Britain
  • 12
    NettenA.DennettJ.KnightJ.Unit Costs of Health and Social Care1998CanterburyPSSRU, University of Kent
  • 13
    LinD.Y.FeuerE.J.EtzioniR.WaxY.Estimating medical costs from incomplete follow-up dataBiometrics199753419434
  • 14
    BriggsA.GrayA.The distribution of health care costs and their statistical analysis for economic evaluationJ Health Services Res Policy19983233245
  • 15
    Department of HealthPolicy Appraisal and Health1995LondonDepartment of Health
  • 16
    Van HoutB.A.AlM.J.GordonG.S.RuttenF.F.H.Costs, effects and c/e-ratios alongside a clinical trialHealth Econ19943309319
  • 17
    UK Prospective Diabetes Study GroupCost effectiveness analysis of improved blood pressure control in hypertensive patients with type 2 diabetes: UKPDS 40Br Med J1998317720726
  • 18
    BriggsA.H.A Bayesian approach to stochastic cost-effectiveness analysisHealth Econ19998257262
  • 19
    GoldM.R.SiegelJ.E.RussellL.B.WeinsteinM.C.Cost-effectiveness analysis in health and medicine1996New YorkOxford University Press
  • 20
    North of England Evidence Based Guideline Development Project. ACE Inhibitors in the Primary Care Management of Adults with Symptomatic Heart Failure1998Newcastle upon TyneCentre for Health Services Research, University of Newcastle upon Tyne
  • 21
    LaupacisA.FeenyD.DetskyA.S.TugwellP.X.How attractive does a new technology have to be to warrant adoption and utilization? Tentative guidelines for using clinical and economic evaluationsCan Med Assoc J1992146473481
  • 22
    Department of HealthRegister of cost-effectiveness studies1994LondonDepartment of Health, Economics and Operational Research Division
  • 23
    McGlynnE.A.NaylorD.AndersonG.M. et al. Comparison of the appropriateness of coronary angiography and coronary artery bypass graft surgery between Canada and New York stateJ Am Med Assoc1994272934941