- Email: [email protected]
Recruitment experience in the nocturnal oxygen therapy trial
13 Recruitment Experience in the Nocturnal Oxygen Therapy Trial George W. Williams, Sandy M. Snedecor, David L. DeMets, and the NOTT Research Group
ABSTRACT: The Nocturnal Oxygen Therapy Trial was a randomized controlled clinical trial
sponsored by the National Heart, Lung, and Blood Institute and carried out by six clinics. The primary objective was to compare, in patients with advanced chronic obstructive pulmonary disease, the effectiveness of 24 hours of oxygen administration to that of 12 hours of oxygen administration including the patient's usual period of sleep. Some patients entering the period of baseline observation were not eligible for randomization at the end of the baseline period. Such attrition is discussed relative to setting goals for the number of patients to enter such an observation period. The impact of an enrollment rate less than what was originally projected is discussed relative to changes in the eligibility criteria and relative to the decision as to whether to extend the recruitment period. KEY WORDS: randomized multicentered clinical trial, recruitment, screening, sample size, eligibility requirements. INTRODUCTION The Nocturnal Oxygen Therapy Trial (NOTr) was a randomized controlled clinical trial sponsored by the National Heart, Lung, and Blood Institute (NHLBI) and carried out by six clinics (see Appendix). The primary objective was to compare, in patients with advanced chronic obstructive pulmonary disease (COPD), the effectiveness of 24 hours of oxygen administration to that of 12 hours of oxygen administration including the patient's usual period of sleep. The primary outcome variables used to assess effectiveness were pulmonary and cardiac function, quality of life, neuropsychological function, and survival. The rationale for and results of this study have been described previously [1,2]. In this article, the recruitment experience of NOTT is described and we amplify on previously reported results of that experience [3,4]. Table 1 contains a brief timetable for this trial. METHODS
The original design called for 300 patients to be randomized to the two treatment groups [5] over a 2-year period. The sample size was influenced Controlled Clinical Trials 8:121S--130S (1987) © Elsevier Science Publishing Co., Inc. 1987 52 Vanderbilt Ave., New York, New York 10017
121S 0197-2456/87/$3.50
122S
G.W. Williams et al. Table 1 Timetable for the NOTT January 1976 June 1976 May 1977 June 1979 June 1980 September 1980
Request for contract proposal issued NIH contract awards Patient enrollment begins Patient enrollment ends Patient follow-up ends Publication of primary results
by a number of factors. The mortality, based on small studies [6,7], was estimated to be 10% for 6 months and 20% for 12 months. Unless the mortality would differ substantially in the two groups, say by at least 40% at 12 months, the required number of patients would be approximately 1000 for a 5% significance level and 90% power. At the planning stages, such dramatic differences in mortality between 12 and 24 hours of oxygen administration were not anticipated. The financial resources of NHLBI and the anticipated availability of COPD patients from large established pulmonary research centers suggested that a few hundred patients might be recruited. Because, with a few hundred patients, the power to detect large differences in mortality would be small, attention turned to morbidity outcome variables. Based on the data from two small studies of patients with COPD [8,9], selected quality of life, neuropsychological status, and pulmonary and cardiac function variables were used to calculate sample size. It appeared that 250-300 patients would be adequate to detect differences of clinical interest with reasonable power (80%90%) for many of these outcome variables. A major concern was that the study have adequate power, such that differences of medical interest would most likely not be missed. If the trial showed no differences and had adequate power, the less expensive and more convenient 12-hour oxygen therapy would likely be the recommended treatment. From these considerations, a sample size of 300 was established, recognizing that this estimate was based on
Table 2 Entry and Exclusion Criteria Entry Criteria Clinical diagnosis of chronic obstructive lung disease Hypoxemia Pao2 ~< 55 mm Hg Pao2 ~< 59 plus one of the following: Edema Hematocrit/> 55% P pulmonale on ECG: 3 mm in leads II, III, aVf Lung Functiona FEV1/FVC < 70% after inhaled bronchodilator TLC ~ 80% predicted Age > 35 Exclusion Criteria Previous 02 therapy: 12 hours/day for 30 days during previous 2 months. Other disease that might be expected to influence mortality, morbidity, compliance with therapy, or ability to give informed consent ~FEV1,forcedexpiratoryvolumein 1 second;FVC, forcedvital capacity;TLC,totallung capacity.
Recruitment Experience in the NOTT
123S
imprecise estimates of the variability for most of the variables and thus was quite approximate. Entry and exclusion criteria are shown in Table 2. The most important entry criterion was hypoxemia. The protocol required that criterion be satisfied on at least two occasions more than I week apart during a 3-week observation period while the patient was free of exacerbations and was managed without supplemental oxygen and with intensive bronchodilator therapy. In practice, patients were recruited in a preliminary way and identified as fitting the entry but not the exclusion criteria. If the patient still met these criteria after a 3-week period of observation to ensure stability, informed consent was obtained, and the patient was hospitalized for I week of baseline studies. At the end of these studies, each patient was randomly allocated by the Data Coordinating Center to either continuous oxygen or nocturnal oxygen therapy. In this trial, the patients were to be followed for a minimum of 6 months and evaluated at periodic intervals. The major time points for evaluation were to be at 6 and 12 months of follow-up.
RESULTS Overview of NOTT During 1977-1980, six clinics randomized 203 patients, 102 to nocturnal oxygen, and followed them for a minimum of I year, with an average followup of 19.3 months. The two groups were comparable at baseline with respect to all major pulmonary, neuropsychological, and quality-of-life baseline measurements. Compliance to the two therapeutic regimens was felt to be excellent, with the continuous oxygen group averaging at least 17.7 hours/day of oxygen usage and the nocturnal oxygen group averaging 12.0 hours/day. The results of the study suggested that the mortality was nearly twice as great for patients in the nocturnal oxygen group than in those in the continuous oxygen group (41 of 102 versus 23 of 101).
The Results of Screening The 203 patients who were randomized were selected from 1043 patients who were screened for the study. Reasons for not entering patients eligible for the trial (in terms of age, diagnosis, and the criteria for hypoxemia) were recorded, although the diligence with which such information was obtained varied among clinics. Of those screened, 809 patients never entered the baseline observation period described in the Methods section: 31% of them were found to have other major disease, 15% refused, 12% were discovered to have previously received chronic oxygen therapy, and 21% had increases in arterial Po2 so that the patient was no longer eligible (Table 3). Thirty-one patients entered the baseline observation period but were ineligible for randomization at the end of that period. The failure to meet the criteria for randomization was chiefly due to a rise in arterial PO2 during the baseline observation period to a value such that the patient was then ineligible. Perhaps the most interesting reason a patient was not enrolled in the NOTT
124S
G.W. Williams et al. Table 3 Reasons for not Entering Patients Into Stabilization Phase Reasona
Po2 out of range Patient refusal Present use of O2/cannot be without 0 2 for stabilization phase Miscellaneous diseases or conditions Travel time excessive Heart disease Other admission requirements not met Neoplasm Multiple illnesses Mental disorders Alcoholism Uncooperative or unreliable patient Acute exacerbation or death No third-party payer Miscellaneous TOTAL
Total
172 123 101 73 64 51 38 35 34 29 25 25 20 11 37 838
aMore than one reason may be given for a patient. A total of 809 patients listed 838 reasons.
was that his or her Pao2 i m p r o v e d to a value above 60 m m H g simply by waiting for an additional 3 weeks w i t h o u t s u p p l e m e n t a l o x y g e n therapy. This observation was m a d e possible b y including an initial 3-week baseline observation period, or "stabilization p h a s e , " in the protocol design before randomization. In the NOTT, the stabilization phase h e l p e d e n s u r e that only patients with COPD w e r e enrolled into the trial a n d that s p o n t a n e o u s r e s p o n d e r s were excluded. Moreover, such a stabilization period m a y help select more compliant patients for r a n d o m i z a t i o n into the trial. In fact, compliance with each o x y g e n regimen was g o o d in this trial. Ten patients, after failure to m e e t the eligibility criteria for r a n d o m i z a t i o n after a first a t t e m p t at stabilization, were able to complete a second a t t e m p t at the 4-week observation period. Friedman et al. [11] have w a r n e d against the inclusion of such " s e c o n d chance" patients. H o w e v e r , there was n o evidence in the N O T T that these " s e c o n d chance" patients differed in terms of their disease status or their compliance from other patients in this trial. Careful attention s h o u l d be given at the design stage of a clinical trial as to w h e t h e r patients should be given a " s e c o n d chance" to be enrolled into the trial. It is of interest to note that less t h a n 5% of patients enrolled in the N O T T were "second chance" patients.
The Results of Enrollment
Figure 1 shows the actual experience of all N O T T clinics c o m p a r e d with the recruitment goals established at the initiation of the study. The overall enrollment rate was consistently less than the rate required by the original protocol during the recruitment period. As n o t e d by others [11], it is of interest to examine the recruitment experience of individual clinics (Fig. 2). Information such as that p r e s e n t e d in Figures 1 and 2 was routinely p r o v i d e d to each clinic d u r i n g the course of the recruitment phase in an effort to provide
125S
Recruitment Experience in the NOTT
PATIENTS
300-
250
. . . .
'
. . . .
I
'
'
'
'
. . . . .
I
5
. . . . . . . . .
I
10
. . . . . . . . .
I
i~
. . . . . . . . .
20
I
'
'
'
'
"
'
'
'
25
"
I
SO
MONTHS
Figure 1 NOTT randomization. Ordinate is cumulative number of patients randomized into the N o T r ; abscissa is time from the initiation of randomization in months; + + + = actual randomization experience; m = the projected r a n d o m i z a t i o n experience required to meet the original protocol recruitment goals.
PATIENTS
50
40
30'
i0.
0 . . . . .
0
'
'
'
'
r 5
. . . . . . . . .
f 10
'
'
. . . . .
'
'
I 15
'
'
'
'
'
'
'
'
' 2{)
I
'
'
. . . . . . .
I 25
'
'
'
'
'
'
'
'
'
I 3O
MONTHS
Figure 2 N o T r r a n d o m i z a t i o n by clinic. Ordinate is cumulative n u m b e r of patients r a n d o m i z e d into the N o T r ; abscissa is time from the initiation of randomization in months; separate curves represent the actual randomization experience of three clinics participating in the NOTT. Clinic 1 = * * *; Clinic 2 = [ ] [ ] O; Clinic 3 = Z~/~Z~.
126S
G.W. Williams et al. feedback on the status of individual recruitment efforts. Clinic 1 shows the progress of a clinic that started patient recruitment on schedule and maintained its pace during the recruitment period. Clinic 2 shows the record of one that started slowly but steadily improved with time. However, considerable effort was required for this improvement in performance. Efforts in that clinic included use of letters to local referring physicians, presentations at the County Medical Society, and presentations at Grand Rounds. In contrast, clinic 3 shows the record of one that started slowly and never improved its performance. A variety of individual recruitment strategies were used by each of the participating clinics. However, each participating clinic had been originally selected because of its large referral practice, and most clinics relied primarily upon their referral base for patients to enroll in the NOTT. Information on referral source was collected. A large majority of patients enrolled were referred by study physicians (46%) or other physicians (32%). Less than 25% of the patients were either self-referred or were introduced to the trial by other means.
CHANGES IN ELIGIBILITY CRITERIA
Five ~nonths after the start of randomization (24 patients randomized), the eligibility criterion for hypoxemia was relaxed to include Pao2 of 59 mm or lower, Hg plus one of the following: edema, hematocrit of 55% or higher, or P pulmonale on ECG. It is interesting to note that in the final NOTT study population, 13% of the patients who would not have been eligible by the original definition of hypoxemia met this expanded criterion for hypoxemia. Hence, additional patients were enrolled into the NOTT by a relaxation of entry criteria. As noted by others [11], relaxing inclusion criteria can adversely affect the study. For example, relaxing inclusion criteria can affect the study design if the change affects the event rates. Moreover, the expected difference in response rate may not be as great as in the original patients. It is interesting to note that patients in the NOTT with higher Pao2 measurements had a lower mortality, but the effect of the extent of oxygen therapy did not vary by level of Pao2. THE DECISION WHETHER TO EXTEND RECRUITMENT
After 1½ years of performance, the NOTT was randomizing patients at a rate that, when projected over the planned and budgeted 2-year period of recruitment, implied a final number of about 220 patients, compared with the targeted 300 patients. The Policy Advisory Board (PAB) thus had to consider whether the recruitment phase should be extended or whether the projected number would be adequate. Extension of the recruitment phase meant increased financial cost. On the other hand, if the NOTT ultimately reported no significant difference, the PAB and the NOTT investigators had to be able to argue persuasively that no difference of clinical interest between nocturnal and continuous oxygen had been missed due to low power. In addition, this decision on accrual had to be made sufficiently far in advance (several months
Recruitment Experience in the NOTT
127S
in fact) to allow for administrative accommodation (e.g., budgeting modifications). To aid in this decision, for each of the primary outcome variables, power calculations were redone with sample sizes of 220 and 300, using the more reliable estimates of variability based on available data and the protocol specified differences in key variables at 6 months of follow-up. The differences used in the power calculations ranged from 0.2 to 0.4 standard deviation units. It is important to note that trends in the treatment differences that were available at that time did not enter into these calculations. Increasing the sample size by something less than 50%, i.e., from 220 to 300 subjects, would not improve the power appreciably. Power to detect the differences specified by the study investigators was 50% or less for 13 of the 34 key variables for either of the two sample sizes. These variables were clustered in the quality-of-life measures. The PAB felt that improving the power for these variables would require too large an increase in sample size. However, power to detect clinically relevant differences was greater than 70% for the majority of the outcome variables, even at the projected smaller sample size of 220 subjects. In March 1979, the PAB recommended not to extend recruitment to achieve 300 randomized patients. As it turned out, the mean values of only two of the variables (hematocrit and pulmonary vascular resistance) were significantly (p < 0.05) different for the two treatment groups, but the final estimates of variability on the total sample size were very close to those used at the time of the extension decision. Thus, the power of the study met the anticipated levels. In addition, conditional power analyses also were performed. These analyses were similar to what is now described as stochastic curtailing [12]. Specifically, the conditional probability of rejecting the null hypothesis after the enrollment of 220 and 300 patients into the trial was computed assuming the null hypothesis to be true and given the current data. Results similar to those noted above were obtained. A variety of statistical methods that were not available at the time this decision was made by the NOTT are currently available to aid PABs in making decisions regarding the extension of a recruitment period. The article by DeMets and Lan [13] summarizes several of the new techniques. The recruitment period was modified to include 26 months of recruitment, as opposed to the 24 months originally planned. Actually, the exact length of the extension varied by clinic and was proportional to the amount of unexpended funds by clinic during the originally planned enrollment period. The problem of whether recruitment should be extended appears in many trials. Strict adherence to the protocol design is not always possible or practical. Because a particular clinical trial may not be repeated, special importance is placed on protecting the scientific effort in the trial being conducted. Decisions about modifications in the design, such as the number of patients randomized, to compensate for inaccuracies in the planning phase are difficult. Preferably, these decisions should be made as early as possible. It is interesting to note that the NOTT Steering Committee quickly recognized the potential difficulty in meeting recruitment goals. Approximately 15% of the way through the recruitment phase, this issue was a topic of some discussion by the Steering Committee.
128S
G.W. Williams et al. It is of interest to note that the minimum follow-up period for each patient was extended from the 6 months indicated in the original protocol to 1 year. However, the decision to extend the minimum follow-up period in the NOTT was not related to the decision to reduce the sample size for the trial. A preliminary report [14] for a trial under way at that time in the United Kingdom had suggested that treatment differences were more marked at 1 year than at 6 months of follow-up.
DISCUSSION
The originally planned recruitment goals for the NOTT were not met. It appears that several factors were responsible. First, a 4-week baseline period before randomization was part of the trial design. More patients were excluded from the trial during this baseline period than originally projected. However, that baseline period had distinct advantages in that truly chronic patients likely to be compliant to the protocol were enrolled in the trial. Second, there is no evidence that attempts to relax entrance criteria and to allow "second chance" patients to be enrolled into the trial altered events rates or compliance rates. However, these mechanisms to increase enrollment did not, in fact, dramatically do so. Third, clinics were selected for this trial partially based upon their large referral populations that were used for patient recruitment. In retrospect, it would appear that additional attention could have been given to developing a wider net for patient recruitment by the inclusion of additional clinics or by more active advertisement by the existing clinics. Fourth, a deliberate decision was made not to extend the recruitment period markedly. Sample size calculations were recomputed based upon information regarding estimates of variability not available at the planning stages of the trial.
ACKNOWLEDGMENTS This study was supported by contracts NO1-HR-6-2942, 2943, 2944, 2945, 2946, and 2947 from the Divisionof Lung Diseases, NationalHeart, Lung, and BloodInstitute, NationalInstitutes of Health, U.S. Department of Health and Human Services.
REFERENCES 1. Nocturnal Oxygen Therapy Group: Continuous or nocturnal oxygen therapy in hypoxemic chronic obstructive lung disease. Ann Intern Med 93:391-398, 1980 2. Lenfant C: Twelve or 24 hour oxygen therapy: Why a clinical trial? JAMA 243:551552, 1980 3. DeMets DL, Williams GW, Brown BW, NOTT Research Group: A case report of data monitoring experience: The Nocturnal Oxygen Therapy Trial. Controlled Clin Trials 3:113-124, 1982 4. Timms RM, Kvale PA, Antonisen NR, Boylen CT, Cugell DW, Petty TL, Williams GW: Selection of patients with chronic obstructive pulmonary disease for longterm oxygen therapy. JAMA 245:2514--2515, 1981 5. National Heart, Lung, and Blood Institute, Division of Lung Diseases: Protocol for Nocturnal Oxygen Therapy Collaborative Program. Bethesda, MD: National
Recruitment Experience in the NOTT
6. 7. 8. 9. 10. 11. 12. 13. 14.
129S
Institutes of Health, 1978. (Available from the National Heart, Lung, and Blood Institute.) Neff TA, Petty TL: Long-term continuous oxygen therapy in chronic airway obstruction: Mortality in relationship to corpulmonale, hypoxia and hypoxemia. Ann Intern Med 72:621, 1970 Stewart BM, Hood CI, Block AJ: Long-term results of continuous oxygen therapy at sea level. Chest 68:486, 1975 Stark RD, Finnegan P, Bishop JM: Daily requirements of oxygen to reverse pulmonary hypertension in patients with chronic bronchitis. Br Med J 3:724, 1972 Block AJ, Castle JR, Keith AS: Chronic oxygen therapy: Treatment of chronic obstructive pulmonary disease at sea level. Chest 65:279, 1974 Cox DR: Regression models and lifetables. J R Stat Soc B 35:187-200, 1972 Friedman LM, Furberg CD, DeMets DL: Fundamentals of Clinical Trials, 2nd ed. Littleton, MA: PSG, 1985 Lan KKG, Simon R, Halperin M: Stochastically curtailed tests in long-term clinical trials. Commun Star C1:207-219, 1982 DeMets DL, Lan KKG: An overview of sequential methods and their application in clinical trials. Commun Stat A13:2315-2338, 1984 Flenley DC, Douglas NJ, Lamb D: Nocturnal hypoxemia and long term domiciliary oxygen in blue and bloated bronchitis. Chest 77:305-307, 1980
APPENDIX: NOTT RESEARCH GROUP
Participants in the Nocturnal Oxygen Therapy Trial were the following: Clin-
ical Centers Henry Ford Hospital, Detroit, MI: Paul A. Kvale, M.D., William A. Conway, M.D. Northwestern University, Chicago, IL: David W. Cugell, M.D., Norman Solliday, M.D. University of Manitoba, Winnipeg, Manitoba: Nicholas R. Antonisen, M.D. University of California and Scripps Clinic Research Foundation, San Diego, CA: Richard M. Timms, M.D. University of Colorado, Denver, CO: Thomas L. Petty, M.D. University of Southern California, Los Angeles, CA: C. Thomas Boylen, M.D.
Data Center University of Michigan, Ann Arbor, MI: George W. Williams, Ph.D., Sandy M. Snedecor, M.L.S.
Neuropsychology Centers Veterans Administration Medical Center and University of California, San Diego, CA: Igor Grant, M.D. University of Colorado, Denver, CO: Robert H. Heaton, Ph.D.
Quality-of-Life Assessment University of West Virginia, Morgantown, WV: A. John McSweeny, Ph.D.
Pathology Center University of Manitoba, Winnipeg, Manitoba: William M. Thurlbeck, M.D.
130S
G.W. Williams et al. Advisory Board Marvin A. Sackner, M.D. (Chairman), Stephen M. Ayers, M.D., B. William Brown, Ph.D., Millicent Higgins, M.D., Dr.P.H., Philip Kimbel, M.D., Jeanne K. Malchon, Harold Menkes, M.D., William F. Miller, M.D., and Louis Vachon, M.D. National Heart, Lung, and Blood Institute Lynn H. Blake, Ph.D., David L. DeMets, Ph.D., Claude Lenfant, M.D., Hannah Peavy, M.D., and Richard Sohn, Ph.D.
ABSTRACT: The Nocturnal Oxygen Therapy Trial was a randomized controlled clinical trial
sponsored by the National Heart, Lung, and Blood Institute and carried out by six clinics. The primary objective was to compare, in patients with advanced chronic obstructive pulmonary disease, the effectiveness of 24 hours of oxygen administration to that of 12 hours of oxygen administration including the patient's usual period of sleep. Some patients entering the period of baseline observation were not eligible for randomization at the end of the baseline period. Such attrition is discussed relative to setting goals for the number of patients to enter such an observation period. The impact of an enrollment rate less than what was originally projected is discussed relative to changes in the eligibility criteria and relative to the decision as to whether to extend the recruitment period. KEY WORDS: randomized multicentered clinical trial, recruitment, screening, sample size, eligibility requirements. INTRODUCTION The Nocturnal Oxygen Therapy Trial (NOTr) was a randomized controlled clinical trial sponsored by the National Heart, Lung, and Blood Institute (NHLBI) and carried out by six clinics (see Appendix). The primary objective was to compare, in patients with advanced chronic obstructive pulmonary disease (COPD), the effectiveness of 24 hours of oxygen administration to that of 12 hours of oxygen administration including the patient's usual period of sleep. The primary outcome variables used to assess effectiveness were pulmonary and cardiac function, quality of life, neuropsychological function, and survival. The rationale for and results of this study have been described previously [1,2]. In this article, the recruitment experience of NOTT is described and we amplify on previously reported results of that experience [3,4]. Table 1 contains a brief timetable for this trial. METHODS
The original design called for 300 patients to be randomized to the two treatment groups [5] over a 2-year period. The sample size was influenced Controlled Clinical Trials 8:121S--130S (1987) © Elsevier Science Publishing Co., Inc. 1987 52 Vanderbilt Ave., New York, New York 10017
121S 0197-2456/87/$3.50
122S
G.W. Williams et al. Table 1 Timetable for the NOTT January 1976 June 1976 May 1977 June 1979 June 1980 September 1980
Request for contract proposal issued NIH contract awards Patient enrollment begins Patient enrollment ends Patient follow-up ends Publication of primary results
by a number of factors. The mortality, based on small studies [6,7], was estimated to be 10% for 6 months and 20% for 12 months. Unless the mortality would differ substantially in the two groups, say by at least 40% at 12 months, the required number of patients would be approximately 1000 for a 5% significance level and 90% power. At the planning stages, such dramatic differences in mortality between 12 and 24 hours of oxygen administration were not anticipated. The financial resources of NHLBI and the anticipated availability of COPD patients from large established pulmonary research centers suggested that a few hundred patients might be recruited. Because, with a few hundred patients, the power to detect large differences in mortality would be small, attention turned to morbidity outcome variables. Based on the data from two small studies of patients with COPD [8,9], selected quality of life, neuropsychological status, and pulmonary and cardiac function variables were used to calculate sample size. It appeared that 250-300 patients would be adequate to detect differences of clinical interest with reasonable power (80%90%) for many of these outcome variables. A major concern was that the study have adequate power, such that differences of medical interest would most likely not be missed. If the trial showed no differences and had adequate power, the less expensive and more convenient 12-hour oxygen therapy would likely be the recommended treatment. From these considerations, a sample size of 300 was established, recognizing that this estimate was based on
Table 2 Entry and Exclusion Criteria Entry Criteria Clinical diagnosis of chronic obstructive lung disease Hypoxemia Pao2 ~< 55 mm Hg Pao2 ~< 59 plus one of the following: Edema Hematocrit/> 55% P pulmonale on ECG: 3 mm in leads II, III, aVf Lung Functiona FEV1/FVC < 70% after inhaled bronchodilator TLC ~ 80% predicted Age > 35 Exclusion Criteria Previous 02 therapy: 12 hours/day for 30 days during previous 2 months. Other disease that might be expected to influence mortality, morbidity, compliance with therapy, or ability to give informed consent ~FEV1,forcedexpiratoryvolumein 1 second;FVC, forcedvital capacity;TLC,totallung capacity.
Recruitment Experience in the NOTT
123S
imprecise estimates of the variability for most of the variables and thus was quite approximate. Entry and exclusion criteria are shown in Table 2. The most important entry criterion was hypoxemia. The protocol required that criterion be satisfied on at least two occasions more than I week apart during a 3-week observation period while the patient was free of exacerbations and was managed without supplemental oxygen and with intensive bronchodilator therapy. In practice, patients were recruited in a preliminary way and identified as fitting the entry but not the exclusion criteria. If the patient still met these criteria after a 3-week period of observation to ensure stability, informed consent was obtained, and the patient was hospitalized for I week of baseline studies. At the end of these studies, each patient was randomly allocated by the Data Coordinating Center to either continuous oxygen or nocturnal oxygen therapy. In this trial, the patients were to be followed for a minimum of 6 months and evaluated at periodic intervals. The major time points for evaluation were to be at 6 and 12 months of follow-up.
RESULTS Overview of NOTT During 1977-1980, six clinics randomized 203 patients, 102 to nocturnal oxygen, and followed them for a minimum of I year, with an average followup of 19.3 months. The two groups were comparable at baseline with respect to all major pulmonary, neuropsychological, and quality-of-life baseline measurements. Compliance to the two therapeutic regimens was felt to be excellent, with the continuous oxygen group averaging at least 17.7 hours/day of oxygen usage and the nocturnal oxygen group averaging 12.0 hours/day. The results of the study suggested that the mortality was nearly twice as great for patients in the nocturnal oxygen group than in those in the continuous oxygen group (41 of 102 versus 23 of 101).
The Results of Screening The 203 patients who were randomized were selected from 1043 patients who were screened for the study. Reasons for not entering patients eligible for the trial (in terms of age, diagnosis, and the criteria for hypoxemia) were recorded, although the diligence with which such information was obtained varied among clinics. Of those screened, 809 patients never entered the baseline observation period described in the Methods section: 31% of them were found to have other major disease, 15% refused, 12% were discovered to have previously received chronic oxygen therapy, and 21% had increases in arterial Po2 so that the patient was no longer eligible (Table 3). Thirty-one patients entered the baseline observation period but were ineligible for randomization at the end of that period. The failure to meet the criteria for randomization was chiefly due to a rise in arterial PO2 during the baseline observation period to a value such that the patient was then ineligible. Perhaps the most interesting reason a patient was not enrolled in the NOTT
124S
G.W. Williams et al. Table 3 Reasons for not Entering Patients Into Stabilization Phase Reasona
Po2 out of range Patient refusal Present use of O2/cannot be without 0 2 for stabilization phase Miscellaneous diseases or conditions Travel time excessive Heart disease Other admission requirements not met Neoplasm Multiple illnesses Mental disorders Alcoholism Uncooperative or unreliable patient Acute exacerbation or death No third-party payer Miscellaneous TOTAL
Total
172 123 101 73 64 51 38 35 34 29 25 25 20 11 37 838
aMore than one reason may be given for a patient. A total of 809 patients listed 838 reasons.
was that his or her Pao2 i m p r o v e d to a value above 60 m m H g simply by waiting for an additional 3 weeks w i t h o u t s u p p l e m e n t a l o x y g e n therapy. This observation was m a d e possible b y including an initial 3-week baseline observation period, or "stabilization p h a s e , " in the protocol design before randomization. In the NOTT, the stabilization phase h e l p e d e n s u r e that only patients with COPD w e r e enrolled into the trial a n d that s p o n t a n e o u s r e s p o n d e r s were excluded. Moreover, such a stabilization period m a y help select more compliant patients for r a n d o m i z a t i o n into the trial. In fact, compliance with each o x y g e n regimen was g o o d in this trial. Ten patients, after failure to m e e t the eligibility criteria for r a n d o m i z a t i o n after a first a t t e m p t at stabilization, were able to complete a second a t t e m p t at the 4-week observation period. Friedman et al. [11] have w a r n e d against the inclusion of such " s e c o n d chance" patients. H o w e v e r , there was n o evidence in the N O T T that these " s e c o n d chance" patients differed in terms of their disease status or their compliance from other patients in this trial. Careful attention s h o u l d be given at the design stage of a clinical trial as to w h e t h e r patients should be given a " s e c o n d chance" to be enrolled into the trial. It is of interest to note that less t h a n 5% of patients enrolled in the N O T T were "second chance" patients.
The Results of Enrollment
Figure 1 shows the actual experience of all N O T T clinics c o m p a r e d with the recruitment goals established at the initiation of the study. The overall enrollment rate was consistently less than the rate required by the original protocol during the recruitment period. As n o t e d by others [11], it is of interest to examine the recruitment experience of individual clinics (Fig. 2). Information such as that p r e s e n t e d in Figures 1 and 2 was routinely p r o v i d e d to each clinic d u r i n g the course of the recruitment phase in an effort to provide
125S
Recruitment Experience in the NOTT
PATIENTS
300-
250
. . . .
'
. . . .
I
'
'
'
'
. . . . .
I
5
. . . . . . . . .
I
10
. . . . . . . . .
I
i~
. . . . . . . . .
20
I
'
'
'
'
"
'
'
'
25
"
I
SO
MONTHS
Figure 1 NOTT randomization. Ordinate is cumulative number of patients randomized into the N o T r ; abscissa is time from the initiation of randomization in months; + + + = actual randomization experience; m = the projected r a n d o m i z a t i o n experience required to meet the original protocol recruitment goals.
PATIENTS
50
40
30'
i0.
0 . . . . .
0
'
'
'
'
r 5
. . . . . . . . .
f 10
'
'
. . . . .
'
'
I 15
'
'
'
'
'
'
'
'
' 2{)
I
'
'
. . . . . . .
I 25
'
'
'
'
'
'
'
'
'
I 3O
MONTHS
Figure 2 N o T r r a n d o m i z a t i o n by clinic. Ordinate is cumulative n u m b e r of patients r a n d o m i z e d into the N o T r ; abscissa is time from the initiation of randomization in months; separate curves represent the actual randomization experience of three clinics participating in the NOTT. Clinic 1 = * * *; Clinic 2 = [ ] [ ] O; Clinic 3 = Z~/~Z~.
126S
G.W. Williams et al. feedback on the status of individual recruitment efforts. Clinic 1 shows the progress of a clinic that started patient recruitment on schedule and maintained its pace during the recruitment period. Clinic 2 shows the record of one that started slowly but steadily improved with time. However, considerable effort was required for this improvement in performance. Efforts in that clinic included use of letters to local referring physicians, presentations at the County Medical Society, and presentations at Grand Rounds. In contrast, clinic 3 shows the record of one that started slowly and never improved its performance. A variety of individual recruitment strategies were used by each of the participating clinics. However, each participating clinic had been originally selected because of its large referral practice, and most clinics relied primarily upon their referral base for patients to enroll in the NOTT. Information on referral source was collected. A large majority of patients enrolled were referred by study physicians (46%) or other physicians (32%). Less than 25% of the patients were either self-referred or were introduced to the trial by other means.
CHANGES IN ELIGIBILITY CRITERIA
Five ~nonths after the start of randomization (24 patients randomized), the eligibility criterion for hypoxemia was relaxed to include Pao2 of 59 mm or lower, Hg plus one of the following: edema, hematocrit of 55% or higher, or P pulmonale on ECG. It is interesting to note that in the final NOTT study population, 13% of the patients who would not have been eligible by the original definition of hypoxemia met this expanded criterion for hypoxemia. Hence, additional patients were enrolled into the NOTT by a relaxation of entry criteria. As noted by others [11], relaxing inclusion criteria can adversely affect the study. For example, relaxing inclusion criteria can affect the study design if the change affects the event rates. Moreover, the expected difference in response rate may not be as great as in the original patients. It is interesting to note that patients in the NOTT with higher Pao2 measurements had a lower mortality, but the effect of the extent of oxygen therapy did not vary by level of Pao2. THE DECISION WHETHER TO EXTEND RECRUITMENT
After 1½ years of performance, the NOTT was randomizing patients at a rate that, when projected over the planned and budgeted 2-year period of recruitment, implied a final number of about 220 patients, compared with the targeted 300 patients. The Policy Advisory Board (PAB) thus had to consider whether the recruitment phase should be extended or whether the projected number would be adequate. Extension of the recruitment phase meant increased financial cost. On the other hand, if the NOTT ultimately reported no significant difference, the PAB and the NOTT investigators had to be able to argue persuasively that no difference of clinical interest between nocturnal and continuous oxygen had been missed due to low power. In addition, this decision on accrual had to be made sufficiently far in advance (several months
Recruitment Experience in the NOTT
127S
in fact) to allow for administrative accommodation (e.g., budgeting modifications). To aid in this decision, for each of the primary outcome variables, power calculations were redone with sample sizes of 220 and 300, using the more reliable estimates of variability based on available data and the protocol specified differences in key variables at 6 months of follow-up. The differences used in the power calculations ranged from 0.2 to 0.4 standard deviation units. It is important to note that trends in the treatment differences that were available at that time did not enter into these calculations. Increasing the sample size by something less than 50%, i.e., from 220 to 300 subjects, would not improve the power appreciably. Power to detect the differences specified by the study investigators was 50% or less for 13 of the 34 key variables for either of the two sample sizes. These variables were clustered in the quality-of-life measures. The PAB felt that improving the power for these variables would require too large an increase in sample size. However, power to detect clinically relevant differences was greater than 70% for the majority of the outcome variables, even at the projected smaller sample size of 220 subjects. In March 1979, the PAB recommended not to extend recruitment to achieve 300 randomized patients. As it turned out, the mean values of only two of the variables (hematocrit and pulmonary vascular resistance) were significantly (p < 0.05) different for the two treatment groups, but the final estimates of variability on the total sample size were very close to those used at the time of the extension decision. Thus, the power of the study met the anticipated levels. In addition, conditional power analyses also were performed. These analyses were similar to what is now described as stochastic curtailing [12]. Specifically, the conditional probability of rejecting the null hypothesis after the enrollment of 220 and 300 patients into the trial was computed assuming the null hypothesis to be true and given the current data. Results similar to those noted above were obtained. A variety of statistical methods that were not available at the time this decision was made by the NOTT are currently available to aid PABs in making decisions regarding the extension of a recruitment period. The article by DeMets and Lan [13] summarizes several of the new techniques. The recruitment period was modified to include 26 months of recruitment, as opposed to the 24 months originally planned. Actually, the exact length of the extension varied by clinic and was proportional to the amount of unexpended funds by clinic during the originally planned enrollment period. The problem of whether recruitment should be extended appears in many trials. Strict adherence to the protocol design is not always possible or practical. Because a particular clinical trial may not be repeated, special importance is placed on protecting the scientific effort in the trial being conducted. Decisions about modifications in the design, such as the number of patients randomized, to compensate for inaccuracies in the planning phase are difficult. Preferably, these decisions should be made as early as possible. It is interesting to note that the NOTT Steering Committee quickly recognized the potential difficulty in meeting recruitment goals. Approximately 15% of the way through the recruitment phase, this issue was a topic of some discussion by the Steering Committee.
128S
G.W. Williams et al. It is of interest to note that the minimum follow-up period for each patient was extended from the 6 months indicated in the original protocol to 1 year. However, the decision to extend the minimum follow-up period in the NOTT was not related to the decision to reduce the sample size for the trial. A preliminary report [14] for a trial under way at that time in the United Kingdom had suggested that treatment differences were more marked at 1 year than at 6 months of follow-up.
DISCUSSION
The originally planned recruitment goals for the NOTT were not met. It appears that several factors were responsible. First, a 4-week baseline period before randomization was part of the trial design. More patients were excluded from the trial during this baseline period than originally projected. However, that baseline period had distinct advantages in that truly chronic patients likely to be compliant to the protocol were enrolled in the trial. Second, there is no evidence that attempts to relax entrance criteria and to allow "second chance" patients to be enrolled into the trial altered events rates or compliance rates. However, these mechanisms to increase enrollment did not, in fact, dramatically do so. Third, clinics were selected for this trial partially based upon their large referral populations that were used for patient recruitment. In retrospect, it would appear that additional attention could have been given to developing a wider net for patient recruitment by the inclusion of additional clinics or by more active advertisement by the existing clinics. Fourth, a deliberate decision was made not to extend the recruitment period markedly. Sample size calculations were recomputed based upon information regarding estimates of variability not available at the planning stages of the trial.
ACKNOWLEDGMENTS This study was supported by contracts NO1-HR-6-2942, 2943, 2944, 2945, 2946, and 2947 from the Divisionof Lung Diseases, NationalHeart, Lung, and BloodInstitute, NationalInstitutes of Health, U.S. Department of Health and Human Services.
REFERENCES 1. Nocturnal Oxygen Therapy Group: Continuous or nocturnal oxygen therapy in hypoxemic chronic obstructive lung disease. Ann Intern Med 93:391-398, 1980 2. Lenfant C: Twelve or 24 hour oxygen therapy: Why a clinical trial? JAMA 243:551552, 1980 3. DeMets DL, Williams GW, Brown BW, NOTT Research Group: A case report of data monitoring experience: The Nocturnal Oxygen Therapy Trial. Controlled Clin Trials 3:113-124, 1982 4. Timms RM, Kvale PA, Antonisen NR, Boylen CT, Cugell DW, Petty TL, Williams GW: Selection of patients with chronic obstructive pulmonary disease for longterm oxygen therapy. JAMA 245:2514--2515, 1981 5. National Heart, Lung, and Blood Institute, Division of Lung Diseases: Protocol for Nocturnal Oxygen Therapy Collaborative Program. Bethesda, MD: National
Recruitment Experience in the NOTT
6. 7. 8. 9. 10. 11. 12. 13. 14.
129S
Institutes of Health, 1978. (Available from the National Heart, Lung, and Blood Institute.) Neff TA, Petty TL: Long-term continuous oxygen therapy in chronic airway obstruction: Mortality in relationship to corpulmonale, hypoxia and hypoxemia. Ann Intern Med 72:621, 1970 Stewart BM, Hood CI, Block AJ: Long-term results of continuous oxygen therapy at sea level. Chest 68:486, 1975 Stark RD, Finnegan P, Bishop JM: Daily requirements of oxygen to reverse pulmonary hypertension in patients with chronic bronchitis. Br Med J 3:724, 1972 Block AJ, Castle JR, Keith AS: Chronic oxygen therapy: Treatment of chronic obstructive pulmonary disease at sea level. Chest 65:279, 1974 Cox DR: Regression models and lifetables. J R Stat Soc B 35:187-200, 1972 Friedman LM, Furberg CD, DeMets DL: Fundamentals of Clinical Trials, 2nd ed. Littleton, MA: PSG, 1985 Lan KKG, Simon R, Halperin M: Stochastically curtailed tests in long-term clinical trials. Commun Star C1:207-219, 1982 DeMets DL, Lan KKG: An overview of sequential methods and their application in clinical trials. Commun Stat A13:2315-2338, 1984 Flenley DC, Douglas NJ, Lamb D: Nocturnal hypoxemia and long term domiciliary oxygen in blue and bloated bronchitis. Chest 77:305-307, 1980
APPENDIX: NOTT RESEARCH GROUP
Participants in the Nocturnal Oxygen Therapy Trial were the following: Clin-
ical Centers Henry Ford Hospital, Detroit, MI: Paul A. Kvale, M.D., William A. Conway, M.D. Northwestern University, Chicago, IL: David W. Cugell, M.D., Norman Solliday, M.D. University of Manitoba, Winnipeg, Manitoba: Nicholas R. Antonisen, M.D. University of California and Scripps Clinic Research Foundation, San Diego, CA: Richard M. Timms, M.D. University of Colorado, Denver, CO: Thomas L. Petty, M.D. University of Southern California, Los Angeles, CA: C. Thomas Boylen, M.D.
Data Center University of Michigan, Ann Arbor, MI: George W. Williams, Ph.D., Sandy M. Snedecor, M.L.S.
Neuropsychology Centers Veterans Administration Medical Center and University of California, San Diego, CA: Igor Grant, M.D. University of Colorado, Denver, CO: Robert H. Heaton, Ph.D.
Quality-of-Life Assessment University of West Virginia, Morgantown, WV: A. John McSweeny, Ph.D.
Pathology Center University of Manitoba, Winnipeg, Manitoba: William M. Thurlbeck, M.D.
130S
G.W. Williams et al. Advisory Board Marvin A. Sackner, M.D. (Chairman), Stephen M. Ayers, M.D., B. William Brown, Ph.D., Millicent Higgins, M.D., Dr.P.H., Philip Kimbel, M.D., Jeanne K. Malchon, Harold Menkes, M.D., William F. Miller, M.D., and Louis Vachon, M.D. National Heart, Lung, and Blood Institute Lynn H. Blake, Ph.D., David L. DeMets, Ph.D., Claude Lenfant, M.D., Hannah Peavy, M.D., and Richard Sohn, Ph.D.