Is there evidence of implicit exclusion criteria for elderly subjects in randomized trials? Evidence from the GUSTO-1 study

Is there evidence of implicit exclusion criteria for elderly subjects in randomized trials? Evidence from the GUSTO-1 study Harlan M. Krumholz, MD,a,b,d,e Cary P. Gross, MD,b,f Eric D. Peterson, MD,g Hal V. Barron, MD,h,i Martha J. Radford, MD,a,c,d Lori S. Parsons, BS,j and Nathan R. Every, MDk New Haven and Middletown, Conn, Durham, NC, San Francisco, Calif, Chicago, Ill, and Seattle, Wash

Background Some experts have raised concerns about the ability to generalize randomized trials, emphasizing that patients who participate in these studies are often not representative of those seen in clinical practice, particularly in the case of elderly patients. To determine the effect of implicit exclusion criteria on a trial study sample, we compared data from the Global Utilization of Streptokinase and t-PA for Occluded Coronary Arteries (GUSTO) trial with data from a retrospective registry from selected hospitals, the National Registry of Myocardial Infarction (NRMI), and a nationally representative study of myocardial infarction care, the Cooperative Cardiovascular Project (CCP). Methods We compared GUSTO subjects aged 65 years and older who were enrolled in the United States with similarily aged patients in the 2 observational studies who met the trial’s eligibility criteria. We examined baseline characteristics, clinical presentation, treatments, procedures, clinical events, and in-hospital mortality rates. Results We found modest, although significant, differences between patients in NRMI, CCP, and GUSTO in demographic and clinical characteristics, treatment, and outcome. For example, GUSTO patients were significantly younger (73.1 ⫾ 5.7 vs 74.7 ⫾ 6.8 for NRMI and 75.8 ⫾ 7.2 for CCP), less likely to have Killip class III/IV at presentation (3.1% vs 6.2% for NRMI and 32.7% for CCP), and more likely to receive aspirin (95.5% vs 86.3% for NRMI and 86.5% for CCP) and ␤-blockers (71.9% vs 43.5% for NRMI and 52.7% for CCP). Overall, NRMI and CCP patients had a lower risk of 30-day mortality after adjustment for demographic, clinical, and hospital characteristics than patients in GUSTO (odds ratio, 0.79; 95% CI, 0.73– 0.86 for NRMI; odds ratio, 0.65; 95% CI, 0.59 – 0.71 for CCP). Conclusions Older patients enrolled in a randomized trial without an age restriction had many similarities compared with patients seen in clinical practice. The higher mortality rate of the GUSTO patients does not support the hypothesis that the trial enrolled a healthier cohort than is seen in practice. (Am Heart J 2003;146:839 – 47.) Many reports have raised concerns about the ability to generalize randomized trials, emphasizing that patients who participate in these studies are often not representative of those seen in clinical practice.1– 8 In

response to such concerns, the US Congress included requirements for the inclusion of women and minorities in the National Institutes of Health (NIH) Revitalization Act of 1993 (Public Law 103-43).9,10 Several

From the aSection of Cardiovascular Medicine, Department of Medicine, Yale University School of Medicine, New Haven, Conn, bYale-New Haven Hospital Center for Outcomes Research and Evaluation, New Haven, Conn, cYale-New Haven Health Center for Outcomes Research and Evaluation, New Haven, Conn, dQualidigm, Middletown, Conn, eSection of Health Policy and Administration, Department of Epidemiology and Public Health, Yale University School of Medicine, New Haven, Conn,

not necessarily reflect the views or policies of the Department of Health and Human Services, nor does the mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government. The author assumes full responsibility for the accuracy and completeness of the ideas presented. This article is a direct result of the Health Care Quality Improvement Program initiated by the Centers for Medicare and Medicaid Services, which has encouraged identification of quality improvement projects derived from analysis of patterns of care, and therefore required no special funding on the part of this contractor. Ideas and contributions to the author concerning experience in engaging with issues presented are welcomed. Guest editor for this manuscript was A. Michael Lincoff, MD, Cleveland Clinic Foundation, Cleveland, Ohio. Submitted October 8, 2002; accepted April 7, 2003. Reprint requests: Harlan M. Krumholz, MD, Yale University School of Medicine, PO Box 208088, New Haven, CT 06520-8088. E-mail: [email protected] © 2003, Mosby, Inc. All rights reserved. 0002-8703/2003/$30.00 ⫹ 0 doi:10.1067/S0002-8703(03)00408-3

f

Department of Medicine, Yale University School of Medicine, New Haven, Conn, Duke Clinical Research Institute, Durham, NC, hDepartments of Epidemiology and Biostatistics and Medicine (Cardiology), University of California, San Francisco, Calif, g

i Department of Medical Affairs, Genentech Inc, South San Francisco, Calif, jOvation Research Group, Chicago, Ill, and kFrazier and Company, Seattle, Wash.

Dr Gross was supported by a Cancer Prevention, Control and Population Sciences Career Development Award (1K078CA-90402) and the Claude D. Pepper Older Americans Independence Center at Yale (P30AG21342). The analyses on which this publication is based were performed under contract number 500-99-CT01, titled “Utilization and Quality Control Peer Review Organization for the State of Connecticut,” sponsored by the Centers for Medicare and Medicaid Services, Department of Health and Human Services. The content of this publication does

American Heart Journal November 2003

840 Krumholz et al

analyses have suggested that the sex and ethnicity of patients enrolled in many federally sponsored trials were representative of the general population,3,11,12 although a recent study of cardiovascular trials suggested only a modest increase in the enrollment of women except in single-sex trials.13 Elderly patients were not mentioned in the NIH Revitalization Act. Reviews have documented that the proportion of older research subjects enrolled in clinical trials is disproportionately lower than the number of elderly patients in the population from which these subjects are drawn.3,14 Partially in response to these concerns, investigators have designed large, simple trials that use broad eligibility criteria and do not explicitly exclude elderly subjects.16 However, physician/investigator preference or the type of sites involved in recruitment may unwittingly produce implicit exclusion criteria that influence the type of patients enrolled and thus affect the ability to generalize the findings. Whether large, simple randomized trials effectively enroll representative samples of older patients is unknown. To investigate this issue, we evaluated the Global Utilization of Streptokinase and t-PA for Occluded Coronary Arteries (GUSTO) trial, a large, simple trial with no age exclusion.17 This important trial has generated ⬎150 articles relevant to the care and outcomes of patients with acute myocardial infarction (AMI). Because participants in this trial received thrombolysis, a therapy with an age-related risk of major complications, we postulated that implicit exclusion criteria might have yielded a highly selected group of participants which differed from the general population.17 To examine the effects of explicit and implicit exclusion criteria on the trial study sample, we analyzed data from 2 registries, the Cooperative Cardiovascular Project (CCP)18 and the National Registry of Myocardial Infarction (NRMI).19

Methods Data sources GUSTO. The GUSTO trial compared thrombolytic strategies for the treatment of AMI; all 4 treatment arms included at least 1 thrombolytic agent. The trial’s eligibility criteria required that patients come to a participating hospital ⬍6 hours after the onset of symptoms, with chest pain lasting at least 20 minutes and accompanied by electrocardiographic signs of ⱖl 0.1 mV of ST-segment elevation in ⱖ2 limb leads or ⱖl 0.2 mV in ⱖ2 contiguous precordial leads. The criteria for exclusion were previous stroke, active bleeding, previous treatment with streptokinase or anistreplase, recent trauma or major surgery, previous participation in the trial, or noncompressible vascular punctures. The trial enrolled 41,021 patients from 1081 hospitals in 15 countries from December 27, 1990, through February 22, 1993. CCP. The CCP is a national quality improvement project that involved the abstraction of medical records of Medicare

patients hospitalized at non-governmental acute care hospitals in the United States with a principal discharge diagnosis of AMI (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] code 410) in 1994 and 1995. The CCP did not sample admissions that were unrelated to the acute care of an AMI (in which the fifth digit of the ICD-9-CM code is 2). The file includes all patients treated in the fee-for-service plans, but does not include all patients treated as part of Medicare managed care risk contracts. NRMI. NRMI 2, initiated on June 1, 1994, is a prospective, observational study of patients admitted to the hospital with AMI. Hospitals, invited by Genentech Inc. to participate in the registry, enrolled consecutive patients with an AMI. The methods of case ascertainment and data acquisition have been validated. By January 31, 1998, 691,995 patients from 1658 US hospitals had been enrolled. For this study, we restricted the analysis to patients hospitalized between June 1, 1994, and December 31, 1995, the period overlapping with the CCP.

Comparison groups We compared GUSTO subjects enrolled in the United States with patients in the 2 registries who met the trial’s eligibility criteria. Although we used the same inclusion criteria as GUSTO, we excluded only patients with previous stroke, active bleeding, recent trauma, or major surgery. We could not, because of a lack of information in the registries, exclude patients with the other GUSTO exclusion criteria (previous treatment with streptokinase or anistreplase or non-compressible vascular punctures). For all groups, we restricted the sample to patients who were ⱖ65 years of age. We also excluded patients who were transferred to another institution, because we were comparing in-hospital mortality rates.

Statistical analysis We compared differences in baseline characteristics and clinical presentation among the 3 groups with the ␹2 test for categorical variables and analysis of variance for continuous variables. We compared treatments, use of cardiac procedures, clinical events, and in-hospital mortality rates in all 3 study populations. We also compared in-hospital mortality rates after adjusting for age, sex, blood pressure, body weight, Killip class, heart rate, MI location, history of prior MI, diabetes mellitus, smoking, bypass grafting surgery, hospital region, hospital volume, and use of aspirin and ␤-blockers. All statistical calculations were performed with SAS version 6.12 statistical procedures software (SAS Institute, Cary, NC).

Results Patient characteristics When we applied the GUSTO selection criteria as aforementioned, we found that 20,647 of the NRMI patients and 17,157 of the CCP patients would have been eligible for participation in GUSTO (Table I). The most common reasons why patients in CCP and NRMI would have been excluded from GUSTO were presen-

American Heart Journal Volume 146, Number 5

Krumholz et al 841

Table I. Exclusion criteria No. Exclusions* Non-US patients Exclusions* Transfer-outs Age ⬍65† NRMI data ⬎ 12/31/95 Total after exclusions Exclusions* Presentation ⬎6 hours after symptoms† Variable coded and ⬎6 hours Variable missing No chest pain on admission† No ST segment elevation in:† ⱖ0.1 mV in ⱖ2 limb leads or ⱖ0.2 mV in ⱖ2 cont. precordial leads Previous stroke† Contraindications to TTx Active bleed Internal bleeding Bleeding disorder History of cerebrovascular accident Recent trauma Recent major surgery Patient refusal Previous streptokinase or anistreptolase Previous participation in the trial Non-compressible vascular punctures Total after exclusions

%

(All GUSTO, n ⫽ 41,021) 17,916 43.70 (US GUSTO, n ⫽ 23,105) 5,642 24.76 14,139 61.19 X n ⫽ 6811 (GUSTO, n ⫽ 6811)‡

No.

(NRMI, n ⫽ 772,586) X (NRMI, n ⫽ 772,586) 159,547 20.70 326,106 42.20 491,834 63.70 n ⫽ 133,545 (NRMI, n ⫽ 133,545)‡ 68,347 51.20 23,213 17.40 45,134 33.80 49,958 37.40 86,103 64.50

15,027 11,893

n ⫽ 6811

%

11.30 8.90

361 0.27 X X X n ⫽ 20,647

No.

%

(CCP, n ⫽ 234,769) X (CCP, n ⫽ 234,769) 39,028 16.60 17,593 7.49 X n ⫽ 181,777 (CCP, n ⫽ 181,777)‡ 63,153 34.70 41,917 23.10 21,235 11.70 64,643 35.60 137,755 75.80

26,291 25,776

14.50 14.20

X X X X n ⫽ 17,157

*Percent based on contents of entire dataset; groups not mutually exclusive. †Or missing data. ‡Percent based on study population; groups not mutually exclusive. CCP, Cooperative Cardiovascular Project; GUSTO, Global Utilization of Streptokinase and t-PA for Occluded Coronary Arteries; NRMI, National Registry of Myocardial Infarction; TTx, thrombolytic therapy.

tation ⬎6 hours after symptom onset and the absence of chest pain or ST-segment elevation on admission. Of the 41,021 patients who participated in GUSTO, many were ⬍65 years of age (60%) or were treated in hospitals outside the United States (44%). After excluding inter-hospital transfers, 6811 (16.6%) GUSTO patients remained in the study sample. We then compared these GUSTO patients with the subset of CCP and NRMI patients who were trial-eligible (Table II). Differences were generally modest, but present for demographic and clinical characteristics. Patients in GUSTO were significantly younger and more likely to be white than either CCP or NRMI patients. GUSTO patients were generally more stable than either CCP or NRMI patients, because they were less likely to have a higher Killip class at presentation. GUSTO patients were less likely than CCP or NRMI patients to have a history of AMI, coronary artery bypass graft surgery, or percutaneous transluminal coronary angioplasty. Compared with CCP patients, GUSTO patients were also less likely to have hypertension or diabetes mellitus. NRMI patients were significantly

more likely than GUSTO or CCP patients to receive care at higher volume hospitals (Table III).

Patterns of care The comparison of patterns of care for the groups is shown in Table IV. As expected per protocol, almost all patients (96.6%) in GUSTO were given thrombolytic therapy, compared with 65.9% of NRMI patients and 43.6% of CCP patients. ␤-blocker therapy was given to 71.9% of GUSTO patients, compared with 43.5% of NRMI patients and 52.7% of CCP patients. GUSTO patients were also significantly more likely to have received aspirin than NRMI or CCP patients (95.5% vs 86.3% vs 86.5%, respectively). CCP patients had the highest likelihood of receiving calcium channel-blockers or angiotensin-converting enzyme inhibitors, followed by GUSTO patients and then NRMI patients. In GUSTO, more patients received a pacemaker or underwent mechanical ventilation than those in NRMI or CCP.

American Heart Journal November 2003

842 Krumholz et al

Table II. Demographic and clinical characteristics NRMI to 12.31.95

GUSTO No. Total patients Sex Male Female Age Mean ⫾ SD 25, 50, 75 Percentile Race White Nonwhite Symptoms to door (h) Mean ⫾ SD Minimum, maximum 25, 50, 75 Percentile No. missing Systolic blood pressure (mean ⫾ SD) Diastolic blood pressure (mean ⫾ SD) Pulse (mean ⫾ SD) Killip class (reference group class I) 1 2 3 4 MI location (reference group anterior) Anterior Inferior Other Hypertension No Yes Diabetes No Yes Current smoker No Yes Previous MI No Yes Previous angina No Yes Previous CABG No Yes Previous PTCA No Yes Number of leads ST elevation (mean ⫾ SD) LBBB No Yes Peak CPK (IU)

%

6811 4079 2732

%

20,647 59.9 40.1

73.1 ⫾ 5.67 68.5, 72.1, 76.8 6350 455

No.

93.3 6.7

%

GUSTO vs NRMI P value

GUSTO vs CCP P value

NRMI vs CCP P value

52.4 47.6

⬍.001

⬍.001

⬍.001

⬍.001

⬍.001

⬍.001

CCP No. 17,157

11,668 8979

56.5 43.5

74.7 ⫾ 6.78 69.2, 73.6, 79.2 18,294 1522

92.3 7.7

8995 8162

75.8 ⫾ 7.23 70.0, 75.0, 81.0 15,701 1451

91.5 8.5

.0070

⬍.001

.0060

1.90 ⫾ 1.20 0.02, 6.00 1.00, 1.58, 2.53 503 139.7 ⫾ 31.94

1.98 ⫾ 1.33 0.05, 5.98 1.00, 1.58, 2.70 0 140.8 ⫾ 32.71

⬍.001

141.3 ⫾ 32.73

82.5 ⫾ 18.04

80.6 ⫾ 17.98

80.4 ⫾ 18.67

⬍.001

⬍.001

75.4 ⫾ 18.28

78.1 ⫾ 21.30

82.7 ⫾ 23.94

⬍.001

⬍.001

⬍.001

.0161

.0006

.1381 .2357

5561 995 120 87

(82.2) (14.7) (1.8) (1.3)

16,075 3118 826 448

(78.5) (15.2) (4.0) (2.2)

9148 2395 5061 553

(53.3) (14.0) (29.5) (3.2)

.0415 ⬍.001 ⬍.001

⬍.001 ⬍.001 ⬍.001

⬍.001 ⬍.001 ⬍.001

2741 3829 206

(40.5) (56.5) (3.0)

7930 10,993 1724

(38.4) (53.2) (8.3)

8646 6321 1475

(52.6) (38.4) (9.0)

.7915 ⬍.001

⬍.001 ⬍.001

⬍.001 ⬍.001

3397 3382

(50.1) (49.9)

10,415 10,232

(50.4) (49.6)

7217 9940

(42.1) (57.9)

.6347

⬍.001

⬍.001

5386 1401

(79.4) (20.6)

16,205 4442

(78.5) (21.5)

12,620 4537

(73.6) (26.4)

.1281

⬍.001

⬍.001

5290 1445

(78.5) (21.5)

16,866 3781

(81.7) (18.3)

14,203 2954

(82.8) (17.2)

⬍.001

⬍.001

5402 1372

(79.7) (20.3)

15,923 4724

(77.1) (22.9)

11,969 5188

(69.8) (30.2)

⬍.001

⬍.001

⬍.001

4251 2496

(63.0) (37.0)

16,969 3678

(82.2) (17.8)

9183 7192

(56.1) (43.9)

⬍.001

⬍.001

⬍.001

6316 477

(93.0) (7.0)

18,754 1893

(90.8) (9.2)

15,373 1784

(89.6) (10.4)

⬍.001

⬍.001

6432 (94.9) 348 (5.1) 4.2 ⫾ 1.80

19,069 1578

(92.4) (7.6) 3.6 ⫾ 1.39

15,890 (92.6) 1267 (7.4) 4.1 ⫾ 1.72

⬍.001 ⬍.001

⬍.001 .1553

.3438 ⬍.001

14,889 2268

⬍.001

⬍.001

⬍.001

5561 53

99.1 0.9

19.953 547

97.3 2.7

86.8 13.2

.0056

.0001

American Heart Journal Volume 146, Number 5

Krumholz et al 843

Table II. continued NRMI to 12.31.95

GUSTO No. Mean ⫾ SD Minimum, maximum 25, 50, 75 Percentile No. missing Peak CK-MB (%) Mean ⫾ SD Minimum, maximum 25, 50, 75 Percentile No. missing

%

No.

CCP %

1876 ⫾ 1588 10.0, 8,279.0 662.0, 1,441.0, 2,679.0 189

383.0, 958.0, 2,026.0 424

15.4 ⫾ 9.61 0.0, 99.5 9.8, 14.0, 19.0 4066

13.5 ⫾ 11.10 0.0, 99.9 7.7, 11.3, 16.0 2478

No.

%

GUSTO vs NRMI P value

1455 ⫾ 1464 3.4, 8,379.0

GUSTO vs CCP P value

NRMI vs CCP P value

⬍.001

⬍.001

BP, Blood pressure; CABG, coronary artery bypass graft surgery; CPK, creatine phosphokinase; LBBB, left bundle branch block; MI, myocardial infarction; PTCA, percutaneous transluminal coronary angioplasty.

Table III. Hospital characteristics NRMI to 12.31.95

GUSTO No. Total patients United States census region Mountain West South Central East South Central South Atlantic West North Central East North Central Middle Atlantic New England Pacific Hospital patient volume/year* Mean ⫾ SD Minimum, maximum 25, 50, 75 Percentile No. missing (reference group high volume) ⬍20 20–40 ⬎40

%

6811 641 275 146 1513 581 1195 1133 720 607

No.

%

20,647 (9.4) (4.0) (2.1) (22.2) (8.5) (17.5) (16.6) (10.6) (8.9)

1754 1541 1234 3537 2071 3999 2425 953 3126

No.

%

GUSTO vs NRMI P value

GUSTO vs CCP P value

NRMI vs CCP P value

⬍.001

⬍.001

⬍.001

⬍.001 ⬍.001

⬍.001 ⬍.001

⬍.001 ⬍.001

17,157 (8.5) (7.5) (6.0) (17.1) (10.0) (19.4) (11.7) (4.6) (15.1)

43.0 ⫾ 22.93 1.2, 116.4 25.8, 38.5, 55.6 0

74.5 ⫾ 53.19 1.0, 287.4 34.7, 61.2, 97.6 0

1018 2534 3259

1612 4624 14411

(14.9) (37.2) (47.8)

CCP

(7.8) (22.4) (69.8)

1037 1761 1054 3151 1202 2797 2811 1086 2044

(6.1) (10.4) (6.2) (18.6) (7.1) (16.5) (16.6) (6.4) (12.1)

25.4 ⫾ 15.44 1.0, 243.5 15.4, 22.9, 32.5 0

7079 7617 2461

(41.3) (44.4) (14.3)

For GUSTO, total days in study based on date first patient enrolled and assumes hospital did not drop out of study. For NRMI and CCP, total days in study based on date first patient enrolled and date last patient enrolled. *Volume ⫽ total patients hospital enrolled/total days hospital in study ⫻ 365.25.

Outcomes The comparison of patient outcomes among the groups is shown in Table V. Patients in CCP had a higher unadjusted risk of in-hospital mortality (16.5% in CCP vs 12.9% in NRMI and 13.8% in GUSTO). NRMI and CCP patients had a lower risk of 30-day mortality after adjustment for demographic, clinical, and hospital characteristics than patients in GUSTO (odds ratio

[OR], 0.79; 95% CI, 0.73– 0.86 for NRMI; OR, 0.65; 95% CI, 0.59 – 0.71 for CCP).

Subgroup analysis The results of the various subgroup analyses, including patients treated with thrombolytic therapy, patients treated at GUSTO recruitment sites, patients not transferred into the hospital, and patients without left

American Heart Journal November 2003

844 Krumholz et al

Table IV. Treatments and procedures NRMI to 12.31.95

GUSTO No. Total patients Thrombolytic therapy No Yes If TTx, streptokinase given No Yes If TTx, t-PA given No Yes If TTx, other TTx given No Yes If TTx, door to drug (h)* Mean ⫾ SD Minimum, maximum 25, 50, 75 Percentile No. missing Aspirin No Yes IV ␤-blocker No Yes Oral ␤-blocker No Yes Any ␤-blocker No Yes ACE inhibitor No Yes Lidocaine No Yes Calcium-channel blocker No Yes Pacemaker No Yes Ventilator No Yes Intraortic balloon pump No Yes CABG No Yes Catheterization No Yes If catheterization, type Elective Emergency Protocol PTCA No Yes

%

6811

No.

CCP %

20,647

No.

%

GUSTO vs NRMI P value

GUSTO vs CCP P value

NRMI vs CCP P value

17,157

232 6578

(3.4) (96.6)

7050 13,595

(34.1) (65.9)

9681 7476

(56.4) (43.6)

⬍.001

⬍.001

⬍.001

1722 4849

(26.2) (73.8)

11,345 2250

(83.4) (16.6)

5801 1675

(77.6) (22.4)

⬍.001

⬍.001

⬍.001

3186 3384

(48.5) (51.5)

2381 11,214

(17.5) (82.5)

1762 5714

(23.6) (76.4)

⬍.001

⬍.001

⬍.001

13,443 152

(98.9) (1.1)

7436 40

(99.5) (0.5)

1.38 ⫾ 0.91 0.03, 19.26 0.87, 1.18, 1.63 497

0.96 ⫾ 1.02 0.02, 23.42 0.50, 0.75, 1.12 802

1.39 ⫾ 3.02 0.02, 47.00 0.53, 0.78, 1.25 366 2312 14,814

(13.5) (86.5)

⬍.001 ⬍.001

.7612

⬍.001

305 6476

(4.5) (95.5)

2824 17,823

(13.7) (86.3)

3643 3158

(53.6) (46.4)

16,780 3867

(81.3) (18.7)

⬍.001

2357 4446

(34.6) (65.4)

13,805 6842

(66.9) (33.1)

⬍.001

1916 4892

(28.1) (71.9)

11,662 8985

(56.5) (43.5)

8118 9039

(47.3) (52.7)

⬍.001

⬍.001

⬍.001

5155 1610

(76.2) (23.8)

18,397 2250

(89.1) (10.9)

10,412 6714

(60.8) (39.2)

⬍.001

⬍.001

⬍.001

5159 1634

(75.9) (24.1)

15181 5466

(73.5) (26.5)

11,533 5624

(67.2) (32.8)

⬍.001

⬍.001

4375 2420

(64.4) (35.6)

17,808 2839

(86.2) (13.8)

10,510 6647

(61.3) (38.7)

⬍.001

⬍.001

⬍.001

6061 742

(89.1) (10.9)

19,203 1444

(93.0) (7.0)

11,912 380

(96.9) (3.1)

⬍.001

⬍.001

⬍.001

5551 1256

(81.5) (18.5)

17,033 3614

(82.5) (17.5)

11,064 1228

(90.0) (10.0)

⬍.001

⬍.001

6413 391

(94.3) (5.7)

18,961 1686

(91.8) (8.2)

11,398 894

(92.7) (7.3)

⬍.001

6093 709

(89.6) (10.4)

17,619 3028

(85.3) (14.7)

15,772 1385

(91.9) (8.1)

⬍.001

⬍.001

⬍.001

3035 3769

(44.6) (55.4)

7569 13,078

(36.7) (63.3)

9893 7264

(57.7) (42.3)

⬍.001

⬍.001

⬍.001

2829 590 352

(75.0) (15.6) (9.3)

5435 1358

(80.0) (20.0)

14,133 6514

(68.5) (31.5)

13,371 3786

(77.9) (22.1)

⬍.001

⬍.001

.0001

.0758

⬍.001

.0001

.0004

.6162

.0035

⬍.001

ACE, Angiotensin-converting enzyme; CABG, coronary artery bypass graft surgery; IV, intravenous; PTCA, percutaneous transluminal coronary angioplasty; t-PA, tissue plasminogen activator. *Values of 0 and ⬎48 hours set to missing.

American Heart Journal Volume 146, Number 5

Krumholz et al 845

Table V. Outcomes/complications NRMI to 12.31.95

GUSTO No. Total patients Hypotension No Yes 2nd or 3rd degree heart block No Yes Heart failure No Yes Reinfarction No Yes Recurrent ischemia No Yes Stroke No Yes Discharge destination Home LTC Deceased Hospital LOS (days) (mean ⫾ SD)* 24-Hour death Alive Dead Inhospital death Alive Dead 30-Day death* Alive Dead 1-Year death* Alive Dead

%

6811

No.

%

20,647

CCP No.

%

GUSTO vs NRMI P value

GUSTO vs CCP P value

NRMI vs CCP P value

⬍.001

17,157

5503 1303

(80.9) (19.1)

16,195 4452

(78.4) (21.6)

11,852 5304

(69.1) (30.9)

⬍.001

⬍.001

5988 815

(88.0) (12.0)

19,276 1371

(93.4) (6.6)

16,093 1064

(93.8) (6.2)

⬍.001

⬍.001

5070 1725

(74.6) (25.4)

17,145 3502

(83.0) (17.0)

10,260 6896

(59.8) (40.2)

⬍.001

⬍.001

⬍.001

6531 261

(96.2) (3.8)

19,870 777

(96.2) (3.8)

15,506 868

(94.7) (5.3)

⬍.001

⬍.001

5468 1321

(80.5) (19.5)

17,616 3031

(85.3) (14.7)

12,145 5011

(70.8) (29.2)

⬍.001

⬍.001

6621 186

(97.3) (2.7)

20,177 470

(97.7) (2.3)

16,659 478

(97.2) (2.8)

.7658

⬍.001

.0326

.8092

5701 (84.1) 176 (2.6) 900 (13.3) 9.83 ⫾ 6.93

16,557 (81.3) 1149 (5.6) 2654 (13.0) 8.15 ⫾ 6.31

13,338 (77.7) 989 (5.8) 2830 (16.5) 7.77 ⫾ 6.07

⬍.001

6405 406

(94.0) (6.0)

19,655 940

(95.4) (4.6)

15,901 1256

(92.7) (7.3)

⬍.001

5868 936

(86.2) (13.8)

17,993 2654

(87.1) (12.9)

14,327 2830

(83.5) (16.5)

5836 965

(85.8) (14.2)

13,890 3246

(81.1) (18.9)

⬍.001

5446 1365

(80.0) (20.0)

11,999 5030

(70.5) (29.5)

⬍.001

.0555

⬍.001

.0002

⬍.001

.0837

.0015

⬍.001

⬍.001

⬍.001

LOS, Length of stay; LTC, long-term care. *If discharged or died on same day as admission, LOS set to 0.5 days.

bundle-branch block, were not substantially different from those of the overall group. The analyses in which CCP patients with evidence of frailty were excluded also did not change the result.

Discussion Our comparison of patients in GUSTO with those in 2 large, national observational studies demonstrates several important findings. First, the proportion of elderly patients in the community setting who would have been eligible for the GUSTO trial is relatively small (10%–15% of potentially eligible patients), primarily because of 2 exclusion criteria: delayed presentation and the absence of suitable electrocardiographic

findings. Among the subgroup of patients in NRMI and CCP who would have met these eligibility criteria, we found that the GUSTO trial had succeeded in enrolling a substantial number of older patients who were similar in many ways to the patients in CCP and NRMI. The use of CCP as a comparison group is particularly informative because CCP has a population-based database, whereas NRMI is heavily weighted toward larger medical centers that were more likely to have on-site catheterization facilities and on-site bypass grafting surgery.20 In comparison with CCP, the GUSTO trial’s major difference is it had fewer patients with Killip class III/IV, left bundle branch block, and prior MI. The differences persisted in analyses restricting the

American Heart Journal November 2003

846 Krumholz et al

cohort by the type of hospitals participating in GUSTO or by the spectrum of patients treated with thrombolytic therapy in the registries. Prior work has suggested that trial participants may receive higher quality care than community patients.21,22 This finding is consistent with our data; GUSTO patients were more likely than CCP or NRMI patients to be treated with aspirin and ␤-blockers during hospitalization. Also, many patients in CCP and NRMI who were eligible for GUSTO did not receive reperfusion therapy. After adjustment for demographic, clinical and treatment factors, the risk of mortality was significantly lower for patients in CCP and NRMI than those in GUSTO. Thus, the findings do not support the idea that patients enrolled in GUSTO were less ill. The adjusted analysis suggests that there may be some unmeasured severity factors that were associated with being enrolled in GUSTO, because it is unlikely that the trial itself caused a worse outcome. It is plausible that among those who were eligible, physicians only enrolled patients who they were certain were sustaining an infarction, and all things being equal, these patients may have been at higher risk of mortality than patients with otherwise similar clinical characteristics. Few studies have examined whether subjects enrolled in large, simple trials are representative of patients seen in practice. Our findings contrast with the report of Jha and colleagues, who compared patients enrolled in Canadian thrombolytic therapy trials with non-trial patients with AMI.23 However, their use of administrative data precluded identification of a comparison group of non-trial patients who would have been candidates for thrombolytic therapy. Consequently, the differences they observed may have been attributable to differences in patients with and patients without indications for thrombolytic therapy, rather than to the difference between patients who were and were not enrolled in a clinical trial. The enrollment of many older patients in GUSTO was an achievement, and there may have been improvements in enrolling representative trial populations since that time. Nevertheless, our study represents the most extensive assessment of implicit eligibility criteria, and we expect that our findings are likely applicable to other randomized clinical trials. The 3 data sources used in our study were not identical, and although this raises the possibility that coding or abstraction variation may account for differences between trial and registry populations, many of the most important characteristics (eg, age, sex, vital status) are unlikely to be influenced by such variations. The timing of hospitalization of the patients from the 3 data sources was also not precisely aligned (GUSTO patients were hospitalized earlier than the patients in the 2 registries). With secular improvements in mortal-

ity rates, this difference in study periods may make the outcomes of registry patients appear more favorable compared with those of trial patients. However, we would not expect substantial differences to arise in the 1-year difference between CCP/NRMI and GUSTO. In addition, the databases have slight differences in case definitions. The GUSTO trial enrolled patients with AMI diagnosed at the time of presentation, whereas NRMI and CCP included patients with a diagnosis of an AMI made on the basis of information available from the hospitalization. We believe that the impact of this difference is probably small, because only 2.4% of GUSTO patients did not meet diagnostic criteria for AMI. Finally, the GUSTO trial did not collect detailed information about frailty. However, comparing GUSTO subjects with CCP subjects without documented dementia, urinary incontinence or mobility did not alter the main results. What are the practical implications of this study? The study shows that a reasonably representative sample of older patients can be enrolled in a large simple trial without an age exclusion. The success of the GUSTO investigators in this regard is noteworthy and should be commended. However, even with that success, there were some differences between the trial patients and the more representative study of older patients in CCP and NRMI. The reasons for these differences are not clear. Investigators need to continue to evolve strategies to enroll even more representative groups of patients and to collect information to support the ability to generalize their cohort.24 We thank the patients, investigators, and hospitals that participated in GUSTO, NRMI, and CCP.

References 1. Robinson D, Woerner MG, Pollack S. Subject selection biases in clinical trials: data from a multicenter schizophrenia treatment study. J Clin Psychopharmacol 1996;16:170 – 6. 2. Wentzer RL, Gouliaev G, Vestergaard P, et al. Generalisability of results from randomized drug trials: a trial on animanic treatment. Br J Psychiatry 1997;170:264 –7. 3. Hutchins LF, Unger JM, Crowley JJ, et al. Underrepresentation of patients 65 years of age or older in cancer-treatment trials. N Engl J Med 1999;341:2061–7. 4. Schneider LS, Olin JT, Lyness SA, et al. Eligibility of Alzheimer’s disease clinic patients for clinical trials. J Am Geriatr Soc 1997; 45:923– 8. 5. Britton A, McKee M, Black N, et al. Threats to applicability of randomised trials: exclusions and selective participation. J Health Serv Res Policy 1999;4:112–21. 6. Greer AL. The state of the art versus the state of the science. Int J Technol Assess 1988;4:5–26. 7. Horwitz RI. Complexity and contradiction in clinical trials research. Am J Med 1987;82:498 –506.

American Heart Journal Volume 146, Number 5

8. Gurwitz JH, Col NF, Avorn J. The exclusion of the elderly and women from clinical trials in acute myocardial infarction. JAMA 1992;268:1417–22. 9. Freedman LS, Simon R, Foukes MA, et al. Inclusion of women and minorities in clinical trials and the NIH revitalization act of 1993— the perspective of NIH Clinical Trialists. Control Clin Trials 1995; 16:286 – 8. 10. NIH Revitalization Act. Subtitle B. In: Sec. 131–3; 1993. 11. Bleyer WA, Tejeda HA, Murphy SB, et al. Equal participation rates of minority patients in U.S. national pediatric clinical trials. J Ped Hematol Oncol 1997;19:423–7. 12. Tejeda HA, Green SB, Trimble EL, et al. Representations of African-Americans, Hispanics, and Whites in National Cancer Institute Cancer Center treatment trials. J Natl Cancer Inst 1996;88:812– 6. 13. Harris DJ, Douglas PS. Enrollment of women in cardiovascular clinical trials funded by the National Heart, Lung, and Blood Institute. N Engl J Med 2000;343:475– 80. 14. Trimble EL, Carter CL, Cain D, et al. Representation of older patients in cancer treatment trials. Cancer 1994;74:2208 –14. 15. Calabresi P, Freeman H. Concerns of special populations. Cancer 1997;80:1258 – 60. 16. Yusuf S, Collins R, Peto R. Why do we need large, simple randomized trials? Stat Med 1984;3:409 –20. 17. The GUSTO Investigators. An international randomized trial comparing four thrombolytic strategies for acute myocardial infarction. N Engl J Med 1993;329:673– 82.

Krumholz et al 847

18. Krumholz HM, Radford MJ, Wang Y, et al. National use and effectiveness of beta-blockers for the treatment of elderly patients after acute myocardial infarction: National Cooperative Cardiovascular Project. JAMA 1998;280:623–9. 19. Rogers WJ, Bowlby LJ, Chandra NC, et al. Treatment of myocardial infarction in the United States (1990 to 1993): observations from the National Registry of Myocardial Infarction. Circulation 1994;90:2103–14. 20. Every NR, Frederick PD, Robinson M, et al. A comparison of the National Registry of Myocardial Infarction 2 with the Cooperative Cardiovascular Project. J Am Coll Cardiol 1999;33:1886 –94. 21. Davis S, Wright PW, Schulman SF, et al. Participants in prospective randomized clinical trials for resected non-small lung cancer have improved survival compared with non-participants in such trials. Cancer 1985;56:1710 – 8. 22. Karjalainen S, Palva I. Do treatment protocols improve end results? A study of survival of patients with multiple myeloma in Finland. BMJ 1989;299:1069 –72. 23. Jha P, Deboer D, Sykora K, et al. Characteristics and mortality outcomes of thrombolysis trial participants and nonparticipants: a population-based comparison. J Am Coll Cardiol 1996;27:1335– 42. 24. Gross CP, Mallory R, Heiat A, et al. Reporting the recruitment process in clinical trials: who are these patients and how did they get there? Ann Intern Med 2002;137:10 – 6.