GUEST EDITORIAL Access to Data: A Contemporary Direction for Clinical Trials Henry Greenberg, MD St. Lukes-Roosevelt Hospital, N e w York, N e w York
Bringing forth information from important clinical trials is a lengthy process. The structure of a well-run trial is hierarchical in nature, which, while often cumbersome, helps ensure compliance with preset standards for recruitment, clinical treatment, and data collection. Generally in most multicenter trials there is a central computer and data-monitoring facility that is responsible for the overall conduct of the trial. These attributes ensure the quality trials we have come to take for granted. In well-run trials, the lag time between data acquisition and entry into a computer-readable format is rarely an issue. During the follow-up period of several months to several years after the end of patient enrollment, but before trial termination, any deficiencies in baseline data collection are corrected. The data file is opened on time with a complete database. During the final phase of preparation of the database, the steering committee directs the preparation of a manuscript detailing the result of the study. Once the results are revealed to the investigators, the committee finalizes the manuscript, circulates it to collaborating investigators for comment, and sends it to a journal whose editor has been primed to receive it for expedited review. The primary paper usually appears in a timely fashion after the end of a trial or premature termination because of an adverse or beneficial treatment effect. If the trial confirms the primary hypothesis of the investigators* then the lag time is short, as in the Norwegian Timalol Trial [1] where the delay was only 5 months. If the data are more complex, the lag time is longer. For the Multicenter Post Infarction Program (MPIP) [2] the lag time was 19 months; for the Multicenter Diltiazem Post Infarction Trial (MDPIT) [3] the lag time was 13 months. If the trial is stopped prematurely for either an untoward or a beneficial effect, the lag time may be short indeed even if the amount of
Address reprint requests to: Henry Greenberg, MD, Division of Cardiology, St. LukesRoosevelt Hospital, 428 West 59th St., New York, NY 10019. Received February 14, 1991; revised June 20, 1991. *The trial investigators adopt a hypothesis that treatment will produce a distinct result such as a drug reducing mortality or suppressing the capacity of silent ischemia to produce increased recurrent infarctions. This proposition is usually tested by assuming a null hypothesis that states that the treatment will have no effect. Investigators and statisticians use the same word--hypothesis--for opposing phenomena.
ControlledClinicalTrials13:93-96(1992) © ElsevierSciencePublishingCo., Inc. 1992 655Avenueof the Americas,New York,N e w
93
York
10010
0197-2456/92/$5.00
94
H. Greenberg data isquite sparse and deals with the primary and points only. In the Beta Blocker Heart Attack Trial [4] the lag time was 1 month; for the Cardiac Arrhythmia Suppression Trial (CAST) [5] the lag time was 4 months. After submission of the paper supporting the main findings, however, the efficiency of information transfer falls off dramatically. Clinical investigators in trials usually have multiple secondary hypotheses and goals that can lead to a large tier of secondary papers. These are not products of data dredging but an established and important part of clinical trials [6, 7]. Often investigator participation in trials is enhanced by, or dependent on, the opportunity to explore questions that, while important, are of secondary importance to the trial. During the planning stages, these goals are incorporated into the trial. If the requirements for attaining these goals do not increase the sample size, the duration of follow-up, the patient care costs, or the degree of effort to collect or process the data, then they enrich the trial. Presentation of this information to the medical community is a major role of clinical trials. However, since these topics rarely command the attention of the entire steering committee, the focused and efficient effort of the data-processing center is not brought to bear on each of them. Topics for these subsequent papers are then divided among either formal writing committees or groups of interested investigators. The members of these groups are often scattered around the country or, more and more frequently, around the world. For geographically dispersed writing committees, the effort to agree on the precise variables and analytical procedures can itself become a time-consuming effort. However, once those agreements have been reached the proposed analysis must enter a queue at the coordinating center. The request joins the queue composed of requests from other writing committees. If, during this time, a team of investigators wishes to amend the data for examination, the cycle is extended in time. By the time a workable dataset reaches an investigator, his or her enthusiasm, time allocation, or commitments have often changed. Would that this system could be changed. Witla the advent of personal computers and analysis programs written for the new generation of users, investigators could have a complete set of the data from the trial in a format that they can use in the comforts of their offices or homes. Once the trial sponsor, whether it be NIH or industry, receives its copy of the data tape, usually at the time of the publication of the primary paper, the tape should be made available to any collaborating investigator who wants it. In trials in which primary data may be released sequentially, as in CAST, the release of data will be guided by the degree of unmasking at each step. Those investigators who should have access to the database include members of the steering committee, directors of core testing facilities, and other senior personnel. These arrangements should be established during the formulation of the trial. The junior faculty at each site will use the data at the discretion and under the direction of the responsible member of the steering committee. Such a proposal poses risks, and they are not insignificant. With data less protected than they would be in a single, central data-processing unit, there is a risk that unauthorized investigators or competitor drug companies will gain access. Even if such major fears are never realized, there are other risks. Junior investigators with only peripheral involvement in the trial could publish
Access to Data
95
material that may overlap the analyses of the senior principal investigators. Other conflicts of major proportions can emerge. Senior collaborators working in similar areas may produce nearly identical papers with predictable unpleasant results. As updated data tapes are produced by the coordinating center, it will be difficult to ensure that each investigator will update his or her version of the database. Discrepancies in published results can discredit important findings. With database control shifted in centrifugal fashion to the enrolling centers, can these problems be avoided? They can, and must be, but this will require attention and effort. The steering committee will need to alter the role it currently assumes in multicenter trials. Increased control of investigator output will be required. In fact, without central control chaos will reign. Topics for exploration will have to be delegated to interested investigators and cooperation will need to be assured by the oversight of the steering committee or some other appropriate leadership body. All manuscripts will have to be reviewed centrally. The steering committee will have to be vigilant in ensuring that papers that are the result of "data dredging" are clearly so labeled. These are not insurmountable obstacles. The data can be assigned in a time-limited format so that annual renewal is only possible with the approval of the steering committee. If an investigator has not complied with his or her responsibilities, access to the data will be withheld. There is one multicenter study, the Coronary Artery Risk Development in Young Adults (CARDIA), that has a distributed data analysis system. Data tapes are made available to investigators in a timely fashion. The proponents feel that the system promotes greater participation and more rapid publication than the traditional model and endorse its widespread adoption [8]. Are these risks and the steps required to minimize them worth the effort? Increasingly our young investigators are computer-literate, and even sophisticated, before entering medical research. Their demands for data access will emerge simply because they are technologically competent. To guarantee a smooth transition to an inevitable future, we need to anticipate these demands. More fundamentally, such policies will permit the more rapid dispersal of important and clinically relevant data. Considering the pace of scientific progress in clinical medicine, more rapid presentation of data can help shape the direction of future inquiries with better focus and curtail expensive and time-consuming dead ends. Those topics deemed most important and timely by the steering committee can be assigned to senior investigators. With m o d e m communication techniques, draft manuscripts can be compiled rapidly and edited almost daily as observations, insights, and calculations emerge from investigator queries of the database. In addition, with the anticipation of better, and easier, data access, participation in all levels of the trial will be enhanced. During the design phase of a trial, the realization that the database will be available for individual scrutiny will stimulate more imaginative contributions on the part of members of the steering committee or core facilities. If a junior faculty person understands that the reward for working in the study will be to participate in the analysis as a coauthor with a steering committee colleague, able to query the database as an independent agent, his or her willingness to contribute from the beginning will increase immeasurably.
96
H. Greenberg The clinical trial will best serve the medical c o m m u n i t y by rapidly and accurately reporting all relevant findings. N e w c o m p u t e r technologies can play an i m p o r t a n t role in reaching this goal. This work was supported in part by grants NIH-RFP-NHLBI85-20 and NIH-HL-38702.
REFERENCES 1. Norwegian Multicenter Study Group: Timolol-induced reduction in mortality and reinfarction in patients surviving acute myocardial infarction. N Engl J Med 304:801807, 1981 2. Multicenter Postinfarction Research Group: Risk stratification and survival after myocardial infarction. N Engl J Med 309:331-336, 1983 3. Multicenter Diltiazem Postinfarction Trial Research Group: The effect of diltiazem on mortality after myocardial infarction. N Engl J Med, 319:385-392, 1988 4. Beta Blocker Heart Attack Study Group: The beta blocker heart attack trial preliminary report. JAMA 246:2073-2074, 1981 5. Cardiac Arrhythmia Suppression Trial (CAST) Investigators: Preliminary report: Effect of encainide and flecainide on mortality in a randomized trial of arrhythmia suppression after myocardial infarction. N Engl J Med, 321:406-412, 1989 6. Meinert CL, Tonascia S: Clinical Trials: Design, Conduct, and Analysis. New York, Oxford Univ. Press, 1986, 67 7. Friedman LM, Furberg CD, DeMets DL: Fundamentals of Clinical Trials, 2nd Ed. Littleton, MA, PSG 1985, 12 8. Perkins LL, Cutter GR, Wagenknecht LE, Savage PJ, Dyer AR, Birch R: Distributed data analysis in a multicenter study: The CARDIA Study. Controlled Clinical Trials 13:80-90, 1992.
Bringing forth information from important clinical trials is a lengthy process. The structure of a well-run trial is hierarchical in nature, which, while often cumbersome, helps ensure compliance with preset standards for recruitment, clinical treatment, and data collection. Generally in most multicenter trials there is a central computer and data-monitoring facility that is responsible for the overall conduct of the trial. These attributes ensure the quality trials we have come to take for granted. In well-run trials, the lag time between data acquisition and entry into a computer-readable format is rarely an issue. During the follow-up period of several months to several years after the end of patient enrollment, but before trial termination, any deficiencies in baseline data collection are corrected. The data file is opened on time with a complete database. During the final phase of preparation of the database, the steering committee directs the preparation of a manuscript detailing the result of the study. Once the results are revealed to the investigators, the committee finalizes the manuscript, circulates it to collaborating investigators for comment, and sends it to a journal whose editor has been primed to receive it for expedited review. The primary paper usually appears in a timely fashion after the end of a trial or premature termination because of an adverse or beneficial treatment effect. If the trial confirms the primary hypothesis of the investigators* then the lag time is short, as in the Norwegian Timalol Trial [1] where the delay was only 5 months. If the data are more complex, the lag time is longer. For the Multicenter Post Infarction Program (MPIP) [2] the lag time was 19 months; for the Multicenter Diltiazem Post Infarction Trial (MDPIT) [3] the lag time was 13 months. If the trial is stopped prematurely for either an untoward or a beneficial effect, the lag time may be short indeed even if the amount of
Address reprint requests to: Henry Greenberg, MD, Division of Cardiology, St. LukesRoosevelt Hospital, 428 West 59th St., New York, NY 10019. Received February 14, 1991; revised June 20, 1991. *The trial investigators adopt a hypothesis that treatment will produce a distinct result such as a drug reducing mortality or suppressing the capacity of silent ischemia to produce increased recurrent infarctions. This proposition is usually tested by assuming a null hypothesis that states that the treatment will have no effect. Investigators and statisticians use the same word--hypothesis--for opposing phenomena.
ControlledClinicalTrials13:93-96(1992) © ElsevierSciencePublishingCo., Inc. 1992 655Avenueof the Americas,New York,N e w
93
York
10010
0197-2456/92/$5.00
94
H. Greenberg data isquite sparse and deals with the primary and points only. In the Beta Blocker Heart Attack Trial [4] the lag time was 1 month; for the Cardiac Arrhythmia Suppression Trial (CAST) [5] the lag time was 4 months. After submission of the paper supporting the main findings, however, the efficiency of information transfer falls off dramatically. Clinical investigators in trials usually have multiple secondary hypotheses and goals that can lead to a large tier of secondary papers. These are not products of data dredging but an established and important part of clinical trials [6, 7]. Often investigator participation in trials is enhanced by, or dependent on, the opportunity to explore questions that, while important, are of secondary importance to the trial. During the planning stages, these goals are incorporated into the trial. If the requirements for attaining these goals do not increase the sample size, the duration of follow-up, the patient care costs, or the degree of effort to collect or process the data, then they enrich the trial. Presentation of this information to the medical community is a major role of clinical trials. However, since these topics rarely command the attention of the entire steering committee, the focused and efficient effort of the data-processing center is not brought to bear on each of them. Topics for these subsequent papers are then divided among either formal writing committees or groups of interested investigators. The members of these groups are often scattered around the country or, more and more frequently, around the world. For geographically dispersed writing committees, the effort to agree on the precise variables and analytical procedures can itself become a time-consuming effort. However, once those agreements have been reached the proposed analysis must enter a queue at the coordinating center. The request joins the queue composed of requests from other writing committees. If, during this time, a team of investigators wishes to amend the data for examination, the cycle is extended in time. By the time a workable dataset reaches an investigator, his or her enthusiasm, time allocation, or commitments have often changed. Would that this system could be changed. Witla the advent of personal computers and analysis programs written for the new generation of users, investigators could have a complete set of the data from the trial in a format that they can use in the comforts of their offices or homes. Once the trial sponsor, whether it be NIH or industry, receives its copy of the data tape, usually at the time of the publication of the primary paper, the tape should be made available to any collaborating investigator who wants it. In trials in which primary data may be released sequentially, as in CAST, the release of data will be guided by the degree of unmasking at each step. Those investigators who should have access to the database include members of the steering committee, directors of core testing facilities, and other senior personnel. These arrangements should be established during the formulation of the trial. The junior faculty at each site will use the data at the discretion and under the direction of the responsible member of the steering committee. Such a proposal poses risks, and they are not insignificant. With data less protected than they would be in a single, central data-processing unit, there is a risk that unauthorized investigators or competitor drug companies will gain access. Even if such major fears are never realized, there are other risks. Junior investigators with only peripheral involvement in the trial could publish
Access to Data
95
material that may overlap the analyses of the senior principal investigators. Other conflicts of major proportions can emerge. Senior collaborators working in similar areas may produce nearly identical papers with predictable unpleasant results. As updated data tapes are produced by the coordinating center, it will be difficult to ensure that each investigator will update his or her version of the database. Discrepancies in published results can discredit important findings. With database control shifted in centrifugal fashion to the enrolling centers, can these problems be avoided? They can, and must be, but this will require attention and effort. The steering committee will need to alter the role it currently assumes in multicenter trials. Increased control of investigator output will be required. In fact, without central control chaos will reign. Topics for exploration will have to be delegated to interested investigators and cooperation will need to be assured by the oversight of the steering committee or some other appropriate leadership body. All manuscripts will have to be reviewed centrally. The steering committee will have to be vigilant in ensuring that papers that are the result of "data dredging" are clearly so labeled. These are not insurmountable obstacles. The data can be assigned in a time-limited format so that annual renewal is only possible with the approval of the steering committee. If an investigator has not complied with his or her responsibilities, access to the data will be withheld. There is one multicenter study, the Coronary Artery Risk Development in Young Adults (CARDIA), that has a distributed data analysis system. Data tapes are made available to investigators in a timely fashion. The proponents feel that the system promotes greater participation and more rapid publication than the traditional model and endorse its widespread adoption [8]. Are these risks and the steps required to minimize them worth the effort? Increasingly our young investigators are computer-literate, and even sophisticated, before entering medical research. Their demands for data access will emerge simply because they are technologically competent. To guarantee a smooth transition to an inevitable future, we need to anticipate these demands. More fundamentally, such policies will permit the more rapid dispersal of important and clinically relevant data. Considering the pace of scientific progress in clinical medicine, more rapid presentation of data can help shape the direction of future inquiries with better focus and curtail expensive and time-consuming dead ends. Those topics deemed most important and timely by the steering committee can be assigned to senior investigators. With m o d e m communication techniques, draft manuscripts can be compiled rapidly and edited almost daily as observations, insights, and calculations emerge from investigator queries of the database. In addition, with the anticipation of better, and easier, data access, participation in all levels of the trial will be enhanced. During the design phase of a trial, the realization that the database will be available for individual scrutiny will stimulate more imaginative contributions on the part of members of the steering committee or core facilities. If a junior faculty person understands that the reward for working in the study will be to participate in the analysis as a coauthor with a steering committee colleague, able to query the database as an independent agent, his or her willingness to contribute from the beginning will increase immeasurably.
96
H. Greenberg The clinical trial will best serve the medical c o m m u n i t y by rapidly and accurately reporting all relevant findings. N e w c o m p u t e r technologies can play an i m p o r t a n t role in reaching this goal. This work was supported in part by grants NIH-RFP-NHLBI85-20 and NIH-HL-38702.
REFERENCES 1. Norwegian Multicenter Study Group: Timolol-induced reduction in mortality and reinfarction in patients surviving acute myocardial infarction. N Engl J Med 304:801807, 1981 2. Multicenter Postinfarction Research Group: Risk stratification and survival after myocardial infarction. N Engl J Med 309:331-336, 1983 3. Multicenter Diltiazem Postinfarction Trial Research Group: The effect of diltiazem on mortality after myocardial infarction. N Engl J Med, 319:385-392, 1988 4. Beta Blocker Heart Attack Study Group: The beta blocker heart attack trial preliminary report. JAMA 246:2073-2074, 1981 5. Cardiac Arrhythmia Suppression Trial (CAST) Investigators: Preliminary report: Effect of encainide and flecainide on mortality in a randomized trial of arrhythmia suppression after myocardial infarction. N Engl J Med, 321:406-412, 1989 6. Meinert CL, Tonascia S: Clinical Trials: Design, Conduct, and Analysis. New York, Oxford Univ. Press, 1986, 67 7. Friedman LM, Furberg CD, DeMets DL: Fundamentals of Clinical Trials, 2nd Ed. Littleton, MA, PSG 1985, 12 8. Perkins LL, Cutter GR, Wagenknecht LE, Savage PJ, Dyer AR, Birch R: Distributed data analysis in a multicenter study: The CARDIA Study. Controlled Clinical Trials 13:80-90, 1992.
