《学术英语（医学）》拓展阅读资料：fundamentals of clinical research for radiology randomized controlled trials

Fundamentals of clinical research for Radiologists Harald o. Stolb Randomized controlled trials Geoffrey Norman Isabelle Trop P receding articles in this series The 19th century saw many major ad- have provided a great deal of in- vances in clinical trials. In 1836, the editor of formation concerning research the American Journal of Medical Sciences design and methodology, including research wrote an introduction to an article that he protocols, statistical analyses, and assess- considered most important medi- ment of the clinical importance of radiologic cal works of the present century, marking the esearch studies. Many methods of research start of a new era of science, "and stated that design have already been presented, includ the article was"the first formal exposition of ing descriptive studies(.g, case reports, the results of the only true method of inves case series, and cross-sectional surveys), and gation in regard to the therapeutic value of some analytical designs (e.g, cohort and remedial agents. " The article that evoked ase-control studies) such effusive praise was the French study on Case-control and cohort studies are also bloodletting in treatment of pneumonia by called observational studies, which distin- P. C. A. Louis [2, 3 guishes them from interventional (experi- Credit for the modern randomized trial is ived June 14, 2004: accepted after revision o.ental)studies because the decision to seek usually given to Sir Austin Bradford Hill [4] ne treatment or another, or to be exposed to The Medical Research Council trials on strep. ditors:Nancy Obuchowski.C.Craig Blackmore, one risk or another, was made by someone tomycin for pulmonary tuberculosis are rightly other than the experimenter. Consequently, regarded as a landmark that ushered in a new adiologist, andthe (ACR,, the Canadian Association of the researcher's role is one of observing t era of medicine. Since Hill's pioneering orise 22 angenology. outcome of these exposures. By contrast, in achievement, the methodology of the random- is experimental studies, the researcher(experi- ized controlled trial has been increasingly ac nasic principles to a nevel of considerable sophistication. powerful type of experimental study is the controlled trials reported has grown exponen- software that permits the user towork with what he or she randomized controlled trial. The basic prin- tially. The Cochrane Library already lists more has learned, which is available on the ACR Web site ciples of randomized controlled trials will be than 150,000 such trials, and they have be- (www.acrorg) discussed in this article come the underlying basis for what is currently called"evidence-based medicine" on Research and Technology Assessment: History of Randomized Controlled Trials aff coordinator: Jonathan H Sunshine. Senior Director for Research. ACR The history of clinical trials dates back to General Principles of Randomized [1] conducted what is probably the earliest re- The randomized controlled trial is one of the Address correspondence to H 0. Stolberg corded clinical trial. He compared the health simplest but most powerful tools of research. In epartment of Clinical Epidemiology and Biostatistics, effects of the vegetarian diet with those of a essence, the randomized controlled trial is McMaster University, Hamiton, ON LBN 345, Canada royal babylonian diet over a 10-day period. study in which people are allocated at random apartment of Radiology Hopital Saint-Luc, 1058 S enis st. Montreal. oC H2X 3J4. Canada. The trial had obvious deficiencies by contem- to receive one of several clinical interventions AR2004183:1539-1544 porary medical standards(allocation bias, as- [2]. On most occasions, the term"intervention certainment bias, and confounding by divine refers to treatment, but it should be used in a intervention), but the report has remained in- much wider sense to include any clinical ma fluential for more than two millennia [2] neuer offered to study participants that may AJR: 183. December 2004 1539

AJR:183, December 2004 1539 Randomized Controlled Trials receding articles in this series have provided a great deal of in￾formation concerning research design and methodology, including research protocols, statistical analyses, and assess￾ment of the clinical importance of radiologic research studies. Many methods of research design have already been presented, includ￾ing descriptive studies (e.g., case reports, case series, and cross-sectional surveys), and some analytical designs (e.g., cohort and case-control studies). Case-control and cohort studies are also called observational studies, which distin￾guishes them from interventional (experi￾mental) studies because the decision to seek one treatment or another, or to be exposed to one risk or another, was made by someone other than the experimenter. Consequently, the researcher’s role is one of observing the outcome of these exposures. By contrast, in experimental studies, the researcher (experi￾menter) controls the exposure. The most powerful type of experimental study is the randomized controlled trial. The basic prin￾ciples of randomized controlled trials will be discussed in this article. History of Randomized Controlled Trials The history of clinical trials dates back to approximately 600 B.C. when Daniel of Judah [1] conducted what is probably the earliest re￾corded clinical trial. He compared the health effects of the vegetarian diet with those of a royal Babylonian diet over a 10-day period. The trial had obvious deficiencies by contem￾porary medical standards (allocation bias, as￾certainment bias, and confounding by divine intervention), but the report has remained in- fluential for more than two millennia [2]. The 19th century saw many major ad￾vances in clinical trials. In 1836, the editor of the American Journal of Medical Sciences wrote an introduction to an article that he considered “one of the most important medi￾cal works of the present century, marking the start of a new era of science,” and stated that the article was “the first formal exposition of the results of the only true method of investi￾gation in regard to the therapeutic value of remedial agents.” The article that evoked such effusive praise was the French study on bloodletting in treatment of pneumonia by P. C. A. Louis [2, 3]. Credit for the modern randomized trial is usually given to Sir Austin Bradford Hill [4]. The Medical Research Council trials on strep￾tomycin for pulmonary tuberculosis are rightly regarded as a landmark that ushered in a new era of medicine. Since Hill’s pioneering achievement, the methodology of the random￾ized controlled trial has been increasingly ac￾cepted and the number of randomized controlled trials reported has grown exponen￾tially. The Cochrane Library already lists more than 150,000 such trials, and they have be￾come the underlying basis for what is currently called “evidence-based medicine” [5]. General Principles of Randomized Controlled Trials The randomized controlled trial is one of the simplest but most powerful tools of research. In essence, the randomized controlled trial is a study in which people are allocated at random to receive one of several clinical interventions [2]. On most occasions, the term “intervention” refers to treatment, but it should be used in a much wider sense to include any clinical ma￾neuver offered to study participants that may Harald O. Stolberg1 Geoffrey Norman2 Isabelle Trop3 Received June 14, 2004; accepted after revision July 2, 2004. Series editors: Nancy Obuchowski, C. Craig Blackmore, Steven Karlik, and Caroline Reinhold. This is the 12th in the series designed by the American College of Radiology (ACR), the Canadian Association of Radiologists, and the American Journal of Roentgenology. The series, which will ultimately comprise 22 articles, is designed to progressively educate radiologists in the methodologies of rigorous clinical research, from the most basic principles to a level of considerable sophistication. The articles are intended to complement interactive software that permits the user to work with what he or she has learned, which is available on the ACR Web site (www.acr.org). Project coordinator: G. Scott Gazelle, Chair, ACR Commission on Research and Technology Assessment; staff coordinator: Jonathan H. Sunshine, Senior Director for Research, ACR. 1 Department of Radiology, McMaster University Medical Centre, 1200 Main St. W, Hamilton, ON L8N 3Z5, Canada. Address correspondence to H. O. Stolberg. 2 Department of Clinical Epidemiology and Biostatistics, McMaster University, Hamilton, ON L8N 3Z5, Canada. 3 Department of Radiology, Hôpital Saint-Luc, 1058 St. Denis St., Montreal, QC H2X 3J4, Canada. AJR 2004;183:1539–1544 0361–803X/04/1836–1539 © American Roentgen Ray Society P Fundamentals of Clinical Research for Radiologists

Stolberg et al have an effect on their health status. Such clini- cases, randomized controlled trials may not be procedure. They must first define the rules cal maneuvers include prevention strategies, feasible because of financial constraints or be- that will govern allocation and then follow reening programs, diagnostic tests, interven- cause of the expectation of low compliance or those rules strictly throughout the entire tional procedures, the setting in which health high drop-out rates tudy [2]. The crucial issue is that after the care is provided, and educational models [2]. Many randomized controlled trials involve procedure for randomization is determined Randomized controlled trials in radiology can large sample sizes because many treatments it should not be modified at any point during play a major role in the assessment of screen- have relatively small effects. The size of the ex- the study. There are many adequate methods ing programs, diagnostic tests, and procedures pected effect of the intervention is the main de- of randomization, but their common element in interventional radiology [6-13 terminant of the sample size necessary to is that no one should be able to determine Randomized controlled trials are used to conduct a successful randomized controlled ahead of time to which group a given patient examine the effect of interventions on particu- trial. Obtaining statistically significant differ- will be assigned. Detailed discussion of rar lar outcomes such as death or the recurrence ences between two samples is easy if large dif- domization methods is beyond the scope of of disease. Some consider randomized con- ferences are expected. However, the smaller the this article trolled trials to be the best of all research de- expected efiect of the intervention, the larger the Numerous methods are also available to en signs [14], or"the most powerful tool in sample size needed to be able to conclude, with sure that the sample of patients is balanced modern clinical research"[15], mainly be- enough power, that the differences are unlikely whenever a small predetermined number of cause the act of randomizing patients to re- to be due to chance. For example, let us assume patients have been enrolled. Unfortunately, the eive or not receive the intervention ensures that we wish to study two groups of patients who methods of allocation in studies described as that, on average, all other possible causes are will undergo different interventions, one of randomized are poorly and infrequently re- equal between the two groups. Thus, any sig- which is a new procedure. We expect a 10%de- ported [2, 28]. As a result, it is not possible to nificant differences between groups in the out- crease in the morbidity rate with the new proce- determine, on most occasions, whether the in- ome event can tributed to the dure. To be able to detect this difference with a vestigators used proper methods to generate intervention and not to some other unidenti- probability(power)of 80%, we need 80 patients random sequences of allocation [2] fied factor. However, randomized controlled in each treatment arm. If the expected difference trials are not a panacea to answer all clinical in effect between the two groups increases to questions, for example, the effect of a risk fac- 20%, the number of patient required per arm de- Bias in Randomized Controlled Trials tor such as smoking cannot ethically be ad- creases to 40. Conversely, if the difference be- The main appeal of the randomized con. dressed with randomized controlled trials. tween the groups is expected to be only 1%, the trolled trial in health care derives from its po- Furthermore, in many situations randomized study population must increase to 8,000 per tential for reducing allocation bias [2]. No controlled trials are not feasible, necessary, treatment arm. The sample size required to other study design allows researchers to bal- important problems [2]. Randomized con- tional to the treatment effect squared [23]. Stan- Random allocation does not, however, pro- trolled trials are not appropriate for cancer dard formulas are available to calculate the tect randomized controlled trials against screening, a situation in which the outcome is approximate sample size necessary when de- other types of bias. during the past 10 years, rare and frequently occurs only after a long signing a randomized controlled trial [24-26 randomized controlled trials have been th delay. Thus, although the test for appraising subject rather than the tool of important, al- the ultimate value of a diagnostic test may be beit isolated, research efforts usually de- a large well-designed randomized controlled Randomization: The Strength of the signed to generate empiric evidence to trial that has patient outcomes as the end point Randomized Controlled Trial improve the design, reporting, dissemina [16], the trial should presumably be per- The randomization procedure gives the ran- tion, and use of randomized controlled trials formed after other smaller studies have exam- domized controlled trial its strength. Random in health care [28]. Such studies have shown ined the predictive value of the test against allocation means that all participants have the that randomized controlled trials are vulnera- some accepted standard. same chance of being assigned to each of the ble to multiple types of bias at all stages of An excellent example of the controversies study groups [27]. The allocation, therefore, is their workspan. a detailed discussion of bias that can arise with randomized controlled tri- not determined by the investigators, the clini- in randomized controlled trials was offered als is an overview of the publications on cians, or the study participants [2]. The pur- by Jadad [2] tant references concern the article by Miet- assure that the characteristics of the partici- are quantitative, comparative controlled. o ammography screening. The most impor- pose of random allocation of participants is to In summary, randomized controlled tria tinen et al. [17 linking screening for breast pants are as likely to be similar as possible periments in which a group of investigators ancer with mammography and an appar- across groups at the start of the comparison studies two or more interventions by admin- ently substantial reduction in fatalities and (also called the baseline). If randomization is istering them to groups of individuals who the responses that it elicited [18-22 done properly, it reduces the risk of a serious have been randomly assigned to receive each Randomized controlled trials may not be imbalance in known and unknown factors that intervention. Alternatively, each individual the assessment of interventions could influence the clinical course of the of interventi that have rare outcomes or effects that take a ticipants. No other study design allows investi- dom order(crossover design) if the outcome ong time to develop In such instances, other gators to balance these factors can be uniquely associated with each inter study designs such as case-control studies or The investigators should follow two rules vention, through, for example, use of a cohort studies are more appropriate. In other to ensure the success of the randomization "washout"period. This step ensures that the AR- 183. December 2004

1540 AJR:183, December 2004 Stolberg et al. have an effect on their health status. Such clini￾cal maneuvers include prevention strategies, screening programs, diagnostic tests, interven￾tional procedures, the setting in which health care is provided, and educational models [2]. Randomized controlled trials in radiology can play a major role in the assessment of screen￾ing programs, diagnostic tests, and procedures in interventional radiology [6–13]. Randomized controlled trials are used to examine the effect of interventions on particu￾lar outcomes such as death or the recurrence of disease. Some consider randomized con￾trolled trials to be the best of all research de￾signs [14], or “the most powerful tool in modern clinical research” [15], mainly be￾cause the act of randomizing patients to re￾ceive or not receive the intervention ensures that, on average, all other possible causes are equal between the two groups. Thus, any sig￾nificant differences between groups in the out￾come event can be attributed to the intervention and not to some other unidenti- fied factor. However, randomized controlled trials are not a panacea to answer all clinical questions; for example, the effect of a risk fac￾tor such as smoking cannot ethically be ad￾dressed with randomized controlled trials. Furthermore, in many situations randomized controlled trials are not feasible, necessary, appropriate, or even sufficient to help solve important problems [2]. Randomized con￾trolled trials are not appropriate for cancer screening, a situation in which the outcome is rare and frequently occurs only after a long delay. Thus, although the test for appraising the ultimate value of a diagnostic test may be a large well-designed randomized controlled trial that has patient outcomes as the end point [16], the trial should presumably be per￾formed after other smaller studies have exam￾ined the predictive value of the test against some accepted standard. An excellent example of the controversies that can arise with randomized controlled tri￾als is an overview of the publications on mammography screening. The most impor￾tant references concern the article by Miet￾tinen et al. [17] linking screening for breast cancer with mammography and an appar￾ently substantial reduction in fatalities and the responses that it elicited [18–22]. Randomized controlled trials may not be appropriate for the assessment of interventions that have rare outcomes or effects that take a long time to develop. In such instances, other study designs such as case-control studies or cohort studies are more appropriate. In other cases, randomized controlled trials may not be feasible because of financial constraints or be￾cause of the expectation of low compliance or high drop-out rates. Many randomized controlled trials involve large sample sizes because many treatments have relatively small effects. The size of the ex￾pected effect of the intervention is the main de￾terminant of the sample size necessary to conduct a successful randomized controlled trial. Obtaining statistically significant differ￾ences between two samples is easy if large dif￾ferences are expected. However, the smaller the expected effect of the intervention, the larger the sample size needed to be able to conclude, with enough power, that the differences are unlikely to be due to chance. For example, let us assume that we wish to study two groups of patients who will undergo different interventions, one of which is a new procedure. We expect a 10% de￾crease in the morbidity rate with the new proce￾dure. To be able to detect this difference with a probability (power) of 80%, we need 80 patients in each treatment arm. If the expected difference in effect between the two groups increases to 20%, the number of patient required per arm de￾creases to 40. Conversely, if the difference be￾tween the groups is expected to be only 1%, the study population must increase to 8,000 per treatment arm. The sample size required to achieve power in a study is inversely propor￾tional to the treatment effect squared [23]. Stan￾dard formulas are available to calculate the approximate sample size necessary when de￾signing a randomized controlled trial [24–26]. Randomization: The Strength of the Randomized Controlled Trial The randomization procedure gives the ran￾domized controlled trial its strength. Random allocation means that all participants have the same chance of being assigned to each of the study groups [27]. The allocation, therefore, is not determined by the investigators, the clini￾cians, or the study participants [2]. The pur￾pose of random allocation of participants is to assure that the characteristics of the partici￾pants are as likely to be similar as possible across groups at the start of the comparison (also called the baseline). If randomization is done properly, it reduces the risk of a serious imbalance in known and unknown factors that could influence the clinical course of the par￾ticipants. No other study design allows investi￾gators to balance these factors. The investigators should follow two rules to ensure the success of the randomization procedure. They must first define the rules that will govern allocation and then follow those rules strictly throughout the entire study [2]. The crucial issue is that after the procedure for randomization is determined, it should not be modified at any point during the study. There are many adequate methods of randomization, but their common element is that no one should be able to determine ahead of time to which group a given patient will be assigned. Detailed discussion of ran￾domization methods is beyond the scope of this article. Numerous methods are also available to en￾sure that the sample of patients is balanced whenever a small predetermined number of patients have been enrolled. Unfortunately, the methods of allocation in studies described as randomized are poorly and infrequently re￾ported [2, 28]. As a result, it is not possible to determine, on most occasions, whether the in￾vestigators used proper methods to generate random sequences of allocation [2]. Bias in Randomized Controlled Trials The main appeal of the randomized con￾trolled trial in health care derives from its po￾tential for reducing allocation bias [2]. No other study design allows researchers to bal￾ance unknown prognostic factors at baseline. Random allocation does not, however, pro￾tect randomized controlled trials against other types of bias. During the past 10 years, randomized controlled trials have been the subject rather than the tool of important, al￾beit isolated, research efforts usually de￾signed to generate empiric evidence to improve the design, reporting, dissemina￾tion, and use of randomized controlled trials in health care [28]. Such studies have shown that randomized controlled trials are vulnera￾ble to multiple types of bias at all stages of their workspan. A detailed discussion of bias in randomized controlled trials was offered by Jadad [2]. In summary, randomized controlled trials are quantitative, comparative, controlled ex￾periments in which a group of investigators studies two or more interventions by admin￾istering them to groups of individuals who have been randomly assigned to receive each intervention. Alternatively, each individual might receive a series of interventions in ran￾dom order (crossover design) if the outcome can be uniquely associated with each inter￾vention, through, for example, use of a “washout” period. This step ensures that the

Fundamentals of Clinical Research for Radiologists effects from one test are not carried over to daily practice. Although both explanatory and ventions is determined at random. This de- the next one and subsequently affect the in- pragmatic approaches are reasonable, and sign, obviously, is appropriate only for dependent evaluation of the second test ad- even complementary, it is important to under- chronic conditions that are fairly stable over ministered. Apart from random allocation to stand that they represent extremes of a spec- time and for interventions that last a short comparison groups, the elements of a ran- trum, and most randomized controlled trials time within the patient and that do not inter- omized controlled trial are no different combine elements of both fere with one another. Otherwise. false con- from those of any other type of prospective, Efficacy or effectiveness trials-Random- clusions about the effectiveness of an comparative, quantitative study ized controlled trials are also often described in intervention could be drawn 291 terms of whether they evaluate the efficacy or Factorial design. -A randomized con- Types of Randomized Controlled Trials effectiveness of an intervention. Efficacy refers trolled trial has a factorial design when two or to interventions carried out under ideal circum- more experimental interventions are not only As Jadad observed in his 1998 book Ran- stances, whereas effectiveness evaluates the ef- evaluated separately but also in combination mised Controlled Trials [2] fects of an intervention under circumstances and against a control [2]. For example, a2 x2 similar to those found in daily practice. factorial design generates four sets of data to Over the years, multiple terms have Phase 1, 2, 3, and 4 trials.-These terms analyze: data on patients who received none been used to describe different types of describe the different types of trials used for of the interventions, patients who received randomized controlled trials. This termi- he introduction of a new intervention, tradi- treatment A, patients who received treatment nology has evolved to the point of tionally a new drug, but could also encom- B, and patients who received both A and B becoming real jargon. This jargon is not pass trials used for the evaluation of a new More complex factorial designs, involving easy to understand for those who are embolization material or type of prosthesis, multiple factors, are occasionally used. The starting their careers as clinicians or se I studies are usually con- strength of this design is that it provides more researchers because there is no single ducted after the safety of the new interven- information than parallel designs. In addition ce with clear and simple definitions tion has been documented in animal to the effects of each treatment, factorial de of all these terms esearch, and their purpose is to document sign allows evaluation of the interaction that the safety of the intervention in humans. may exist between two treatments. Because The best classification of frequently Phase I studies are usually performed on randomized controlled trials are generally ex- terms was offered by Jadad [2], and healthy volunteers. Once the intervention pensive to conduct, the more answers that can based our article on his work According to Jadad, randomized con- intervention is given to a small group of real trolled trials can be classified as to the as- patients, and the purpose of this study is to Randomized Controlled Trials Classified According to pects of intervention that investigators want evaluate the efficacy of different modes of the Number of Participants to explore, the way in which the participants administration of the intervention to patients. Randomized controlled trials can be per e exposed to the intervention, the number Phase 2 studies focus on efficacy while still formed in one or many centers and can in- of participants included in the study, whether providing information on safety. Phase 3 clude from one to thousands of participants, the investigators and participants know studies are typically effectiveness trials, and they can have fixed or variable(sequen- which intervention is being assessed, and which are performed after a given procedure tial)numbers of participants whether the preference of nonrandomized in- has been shown to be safe with a reasonable " N-of-one trials "Randomized con- dividuals and participants has been taken chance of improving patients' conditions. trolled trials with only one participant are into account in the design of the study. In the Most phase 3 trials are randomized con- called"n-of-one trials"or"individual patient context of this article, we can offer only a trolled trials. Phase 4 studies are equivalent trials "Randomized controlled trials with brief discussion of each of the different types to postmarketing studies of the intervention; simple design that involve thousands of pa- of randomized controlled trials they are performed to identify and monitor tients and limited data collection are called possible adverse events not yet documented. "megatrials [30, 31]. Usually, megatrial require the participation of many investiga Randomized controlled trials used to eval Randomized Controlled Trials Classified Acording to tors from multiple centers and from different the Different Aspects of Interventions Evaluated ate different interventions include explanatory Intervention Sequential trials.-A sequential trial is a or pragmatic trials, efficacy or equivalence tri- These types of randomized controlled trials study with parallel design in which the number als; and phase 1, 2, 3, and 4 trials. nclude parallel, crossover, and factorial designs. of participants is not specified by the investiga Explanatory or pragmatic trials-Explan- Parallel design.-Most randomized con- tors beforehand. Instead, the investigator atory trials are designed to answer a simple trolled trials have parallel designs in which continue recruiting participants until a clear question: Does the intervention work? If it each group of participants is exposed to only benefit of one of the interventions is observed does, then the trial attempts to establish how one of the study interventions until they become convinced that there it works. Pragmatic trials, on the other hand, rossover design- Crossover design re- no important differences between the inter re designed not only to determine whether fers to a study in which each of the partici- ventions [27. This element applies to the the intervention works but also to describe all pants is given all of the study interventions in comparison of some diagnostic interventions the consequences of the intervention and its successive periods. The order in which the and some procedures in interventional radiol under circumstances corresponding to participants receive each of the study inter- ogy. Strict rules govern when trials can be R: 183. December 2004 541

Fundamentals of Clinical Research for Radiologists AJR:183, December 2004 1541 effects from one test are not carried over to the next one and subsequently affect the in￾dependent evaluation of the second test ad￾ministered. Apart from random allocation to comparison groups, the elements of a ran￾domized controlled trial are no different from those of any other type of prospective, comparative, quantitative study. Types of Randomized Controlled Trials As Jadad observed in his 1998 book Ran￾domised Controlled Trials [2]: Over the years, multiple terms have been used to describe different types of randomized controlled trials. This termi￾nology has evolved to the point of becoming real jargon. This jargon is not easy to understand for those who are starting their careers as clinicians or researchers because there is no single source with clear and simple definitions of all these terms. The best classification of frequently used terms was offered by Jadad [2], and we have based our article on his work. According to Jadad, randomized con￾trolled trials can be classified as to the as￾pects of intervention that investigators want to explore, the way in which the participants are exposed to the intervention, the number of participants included in the study, whether the investigators and participants know which intervention is being assessed, and whether the preference of nonrandomized in￾dividuals and participants has been taken into account in the design of the study. In the context of this article, we can offer only a brief discussion of each of the different types of randomized controlled trials. Randomized Controlled Trials Classified According to the Different Aspects of Interventions Evaluated Randomized controlled trials used to evalu￾ate different interventions include explanatory or pragmatic trials; efficacy or equivalence tri￾als; and phase 1, 2, 3, and 4 trials. Explanatory or pragmatic trials.—Explan￾atory trials are designed to answer a simple question: Does the intervention work? If it does, then the trial attempts to establish how it works. Pragmatic trials, on the other hand, are designed not only to determine whether the intervention works but also to describe all the consequences of the intervention and its use under circumstances corresponding to daily practice. Although both explanatory and pragmatic approaches are reasonable, and even complementary, it is important to under￾stand that they represent extremes of a spec￾trum, and most randomized controlled trials combine elements of both. Efficacy or effectiveness trials.—Random￾ized controlled trials are also often described in terms of whether they evaluate the efficacy or effectiveness of an intervention. Efficacy refers to interventions carried out under ideal circum￾stances, whereas effectiveness evaluates the ef￾fects of an intervention under circumstances similar to those found in daily practice. Phase 1, 2, 3, and 4 trials.—These terms describe the different types of trials used for the introduction of a new intervention, tradi￾tionally a new drug, but could also encom￾pass trials used for the evaluation of a new embolization material or type of prosthesis, for example. Phase 1 studies are usually con￾ducted after the safety of the new interven￾tion has been documented in animal research, and their purpose is to document the safety of the intervention in humans. Phase 1 studies are usually performed on healthy volunteers. Once the intervention passes phase 1, phase 2 begins. Typically, the intervention is given to a small group of real patients, and the purpose of this study is to evaluate the efficacy of different modes of administration of the intervention to patients. Phase 2 studies focus on efficacy while still providing information on safety. Phase 3 studies are typically effectiveness trials, which are performed after a given procedure has been shown to be safe with a reasonable chance of improving patients’ conditions. Most phase 3 trials are randomized con￾trolled trials. Phase 4 studies are equivalent to postmarketing studies of the intervention; they are performed to identify and monitor possible adverse events not yet documented. Randomized Controlled Trials Classified According to Participants’ Exposure and Response to the Intervention These types of randomized controlled trials include parallel, crossover, and factorial designs. Parallel design.—Most randomized con￾trolled trials have parallel designs in which each group of participants is exposed to only one of the study interventions. Crossover design.— Crossover design re￾fers to a study in which each of the partici￾pants is given all of the study interventions in successive periods. The order in which the participants receive each of the study inter￾ventions is determined at random. This de￾sign, obviously, is appropriate only for chronic conditions that are fairly stable over time and for interventions that last a short time within the patient and that do not inter￾fere with one another. Otherwise, false con￾clusions about the effectiveness of an intervention could be drawn [29]. Factorial design.—A randomized con￾trolled trial has a factorial design when two or more experimental interventions are not only evaluated separately but also in combination and against a control [2]. For example, a 2 × 2 factorial design generates four sets of data to analyze: data on patients who received none of the interventions, patients who received treatment A, patients who received treatment B, and patients who received both A and B. More complex factorial designs, involving multiple factors, are occasionally used. The strength of this design is that it provides more information than parallel designs. In addition to the effects of each treatment, factorial de￾sign allows evaluation of the interaction that may exist between two treatments. Because randomized controlled trials are generally ex￾pensive to conduct, the more answers that can be obtained, the better. Randomized Controlled Trials Classified According to the Number of Participants Randomized controlled trials can be per￾formed in one or many centers and can in￾clude from one to thousands of participants, and they can have fixed or variable (sequen￾tial) numbers of participants. “N-of-one trials.”—Randomized con￾trolled trials with only one participant are called “n-of-one trials” or “individual patient trials.” Randomized controlled trials with a simple design that involve thousands of pa￾tients and limited data collection are called “megatrials.” [30, 31]. Usually, megatrials require the participation of many investiga￾tors from multiple centers and from different countries [2]. Sequential trials.—A sequential trial is a study with parallel design in which the number of participants is not specified by the investiga￾tors beforehand. Instead, the investigators continue recruiting participants until a clear benefit of one of the interventions is observed or until they become convinced that there are no important differences between the inter￾ventions [27]. This element applies to the comparison of some diagnostic interventions and some procedures in interventional radiol￾ogy. Strict rules govern when trials can be

Stolberg et al topped on the basis of cumulative in a randomized controlled trial but have a limitations of the research methods of ran and important statistical considerations clear preference for one of the study inter- domized controlled trials is growing. A ma ventions. At least three types of randomized jor barrier hindering the assessment of trial Fixed trials.Alternatively, in a fixed controlled trials take into account the prefer- quality is that, in most cases, we must rely on trial, the investigators establish deductively ences of eligible individuals as to whether or the information contained in the written re- the number of participants(sample size)that not they take part in the trial. These are port. A trial with a biased design, if well re- will be studied. This number can be decided called preference trials because they include ported, could be judged to be of high quality, arbitrarily or can be calculated using statisti- at least one group in which the participants whereas a well-designed but poorly reported cal methods. The latter is a more commonly are allowed to choose their preferred treat- trial could be judged to be of low quality used method. Even in a fixed trial, the design ment from among several options offered Recently, efforts have been made to im of the trial usually specifies whether there 32, 33]. Such trials can have a Zelen design, prove the quality of randomized controlled will be one or more interim analyses of data. comprehensive cohort design, or Wennbergs trials. In 1996, a group of epidemiologists If a clear benefit of one intervention over the design 33-36]. For a detailed discussion of biostatisticians, and journal editors published other can be shown with statistical signifi- these designs of randomized controlled tri -"CONSORT( Consolidated Standards of Re- cance before all participants are recruited, it als, the reader is directed to the excellent de- porting Trials)"[38], a statement that re may not be ethical to pursue the trial, and it tailed discussion offered by Jadad [2] process to improve the standards of written reports of randomized controlled trials. The Randomized Controlled Trials Classified According to The Ethics of Randomized Controlled CONSORT Statement was revised in 2001 Trials 39]. It was designed to assist the reporting In addition to randomization, the investi- Despite the claims of some enthusiasts for of randomized controlled trials with two gators can incorporate other methodologic randomized controlled trials, many important groups and those with parallel designs.Some trategies to reduce the risk of other biases. aspects of health care cannot be subjected to a modifications will be required to report These strategies are known as"blinding. randomized trial for practical and ethical rea- crossover trials and those with more than two The purpose of blinding is to reduce the risk sons. A randomized controlled trial is the best groups [40]. Although the CONSORT state- of ascertainment and observation bias. An way of evaluating the effectiveness of an inter- ment was not evaluated before its publica- open randomized controlled trial is one in vention, but before a randomized controlled tion, it was expected that it would lead to an which everybody involved in the trial knows trial can be conducted, there must be equ- improvement in the quality of reporting of which intervention is given to each partici- poise-genuine doubt about whether one randomized controlled trials, at least in th pant. Many radiology studies are open ran- course of action is better than another [16]. journals that endorse it [41] domized controlled trials because blinding is Equipoise then refers to that state of knowl- Recently, however, Chan et al. [42] not feasible or ethical. One cannot, for exam- edge in which no evidence exists that shows pointed out that the interpretation of the re- ple, perform an interventional procedure that any intervention in the trial is better than sults of randomized controlled trials has em- with its associated risks without revealing to another and that any intervention is better than phasized statistical significance rather than the patient and the treating physician to those in the trial. It is not ethical to build a trial clinical importance which group the patient has been random- in which, before enrollment, evidence suggests ized. A single-blinded randomized controlled that patients in one arm of the study are more The lack of emphasis on clinical trial is one in which a group of individuals likely to benefit from enrollment than patients involved in the trial (usually patients)does in the other arm Equipoise thus refers to the ceptions and disagreements regarding not know which intervention is given to each fine balance that exists between being hopeful the interpretation of the results of clinical participant. A double-blinded randomized a new treatment will improve a condition and trials and a tendency to equate statistical controlled trial, on the other hand, is one in having enough evidence to know that it does the trial ( usually patients and treating physi- can be planned only in areas of uncertainty results may not be clinically important concerns that are results do not completely rule out the physicians, and study investigators) and as other research designs will be addressed in quadruple-blinded randomized controlled subsequent articles in this series. Hellman trials(blinding of patients, treating physi- and Hellman [37] offered a good discussion Randomized Controlled Trials cians, study investigators, and statisticians) on this subject The evaluation of the methodologic qual have been described but are rarely used ity of randomized controlled trials is centra to the appraisal of individual trials, the con- Trials9 of Randomized Controlled Reporti duct of unbiased sy ic reviews, and the Nonrandomized Participant Preferences performance of evidence-based health care igible individuals may refuse to partici- The Quality of Randomized Controlled Trial Repor ting However, important methodologic details pate in a randomized controlled trial. Other Awareness conceming the quality of re- may be omitted from published reports, and eligible individuals may decide to participate porting randomized controlled trials and the the quality of reporting is, therefore, often AR- 183. December 2004

1542 AJR:183, December 2004 Stolberg et al. stopped on the basis of cumulative results, and important statistical considerations come into play. Fixed trials.—Alternatively, in a fixed trial, the investigators establish deductively the number of participants (sample size) that will be studied. This number can be decided arbitrarily or can be calculated using statisti￾cal methods. The latter is a more commonly used method. Even in a fixed trial, the design of the trial usually specifies whether there will be one or more interim analyses of data. If a clear benefit of one intervention over the other can be shown with statistical signifi- cance before all participants are recruited, it may not be ethical to pursue the trial, and it may be prematurely terminated. Randomized Controlled Trials Classified According to the Level of Blinding In addition to randomization, the investi￾gators can incorporate other methodologic strategies to reduce the risk of other biases. These strategies are known as “blinding.” The purpose of blinding is to reduce the risk of ascertainment and observation bias. An open randomized controlled trial is one in which everybody involved in the trial knows which intervention is given to each partici￾pant. Many radiology studies are open ran￾domized controlled trials because blinding is not feasible or ethical. One cannot, for exam￾ple, perform an interventional procedure with its associated risks without revealing to the patient and the treating physician to which group the patient has been random￾ized. A single-blinded randomized controlled trial is one in which a group of individuals involved in the trial (usually patients) does not know which intervention is given to each participant. A double-blinded randomized controlled trial, on the other hand, is one in which two groups of individuals involved in the trial (usually patients and treating physi￾cians) do not know which intervention is given to each participant. Beyond this, tri￾ple-blinded (blinding of patients, treating physicians, and study investigators) and quadruple-blinded randomized controlled trials (blinding of patients, treating physi￾cians, study investigators, and statisticians) have been described but are rarely used. Randomized Controlled Trials Classified According to Nonrandomized Participant Preferences Eligible individuals may refuse to partici￾pate in a randomized controlled trial. Other eligible individuals may decide to participate in a randomized controlled trial but have a clear preference for one of the study inter￾ventions. At least three types of randomized controlled trials take into account the prefer￾ences of eligible individuals as to whether or not they take part in the trial. These are called preference trials because they include at least one group in which the participants are allowed to choose their preferred treat￾ment from among several options offered [32, 33]. Such trials can have a Zelen design, comprehensive cohort design, or Wennberg’s design [33–36]. For a detailed discussion of these designs of randomized controlled tri￾als, the reader is directed to the excellent de￾tailed discussion offered by Jadad [2]. The Ethics of Randomized Controlled Trials Despite the claims of some enthusiasts for randomized controlled trials, many important aspects of health care cannot be subjected to a randomized trial for practical and ethical rea￾sons. A randomized controlled trial is the best way of evaluating the effectiveness of an inter￾vention, but before a randomized controlled trial can be conducted, there must be equi￾poise—genuine doubt about whether one course of action is better than another [16]. Equipoise then refers to that state of knowl￾edge in which no evidence exists that shows that any intervention in the trial is better than another and that any intervention is better than those in the trial. It is not ethical to build a trial in which, before enrollment, evidence suggests that patients in one arm of the study are more likely to benefit from enrollment than patients in the other arm. Equipoise thus refers to the fine balance that exists between being hopeful a new treatment will improve a condition and having enough evidence to know that it does (or does not). Randomized controlled trials can be planned only in areas of uncertainty and can be carried out only as long as the un￾certainty remains. Ethical concerns that are unique to randomized controlled trials as well as other research designs will be addressed in subsequent articles in this series. Hellman and Hellman [37] offered a good discussion on this subject. Reporting of Randomized Controlled Trials The Quality of Randomized Controlled Trial Reporting Awareness concerning the quality of re￾porting randomized controlled trials and the limitations of the research methods of ran￾domized controlled trials is growing. A ma￾jor barrier hindering the assessment of trial quality is that, in most cases, we must rely on the information contained in the written re￾port. A trial with a biased design, if well re￾ported, could be judged to be of high quality, whereas a well-designed but poorly reported trial could be judged to be of low quality. Recently, efforts have been made to im￾prove the quality of randomized controlled trials. In 1996, a group of epidemiologists, biostatisticians, and journal editors published “CONSORT (Consolidated Standards of Re￾porting Trials)” [38], a statement that re￾sulted from an extensive collaborative process to improve the standards of written reports of randomized controlled trials. The CONSORT statement was revised in 2001 [39]. It was designed to assist the reporting of randomized controlled trials with two groups and those with parallel designs. Some modifications will be required to report crossover trials and those with more than two groups [40]. Although the CONSORT state￾ment was not evaluated before its publica￾tion, it was expected that it would lead to an improvement in the quality of reporting of randomized controlled trials, at least in the journals that endorse it [41]. Recently, however, Chan et al. [42] pointed out that the interpretation of the re￾sults of randomized controlled trials has em￾phasized statistical significance rather than clinical importance: The lack of emphasis on clinical importance has led to frequent miscon￾ceptions and disagreements regarding the interpretation of the results of clinical trials and a tendency to equate statistical significance with clinical importance. In some instances, statistically significant results may not be clinically important and, conversely, statistically insignificant results do not completely rule out the possibility of clinically important effects. Limitations of the Research Methods Used in Randomized Controlled Trials The evaluation of the methodologic qual￾ity of randomized controlled trials is central to the appraisal of individual trials, the con￾duct of unbiased systematic reviews, and the performance of evidence-based health care. However, important methodologic details may be omitted from published reports, and the quality of reporting is, therefore, often

Fundamentals of Clinical Research for Radiologists used as a proxy measure for methodologic randomized controlled trials has significantly 8. Fontana RS, Sanderson DR, Woolner LB, et al quality. High-quality reporting may hide im- increased in both diagnostic and interven- Screening for lung cancer. a critique of the Mayo ant differences in methodologic quality, tional radiology. Examples of randomized Lung Project. Cancer 1991: 67 suppl 4: 1 155-ll and well-conducted trials may be reported controlled trials in diagnostic imaging in- on survival and health-related quality of life in badly [43]. As Devereaux et al. [41] ob- clude the works of Gottlieb et al. [48] and breast served, [h]ealth care providers depend upon Kaiser et al. [149]. Examples of interventional controlled triakthe GIvIO Investigators JAMA authors and editors to report essential meth- randomized controlled trials are the studies odological factors in randomized controlled by Pinto et al. [50] and Lencioni et al. [51]. 10. Jarvik JG, Maravilla KR, Haynor DR, Levitz M trials(RCTs)to allow determination of trial Randomized controlled trials are equally Deyo RA. Rapid MR imaging versus plain radi- alidity(i.e, likelihood that the trials'results important in screening for disease. Our ini- graphy in patients with low back pain: initial re- tial experience with breast screening was ur of a randomized study. Radiology 997:204:447-454 The most important limitations of re- fortunate, and controversy over this issue 11. Kinnison ML, Powe NR, Steinberg EP. Results search methods include the following continues to this day [52, 53]. On the other Insufficient power-A survey of 71 ran- hand, positive developments have occurred, osmolality contrast media. Radiology 1989, 170 domized controlled trials showed that most such as the work of the American College of 381-389 of these trials were too small (i.e, had insuf- Radiology Imaging Network. Writing for Rosselli M. Palli D. Cariddi A. Ciatto S. Pacini P. ferences)and that the authors of these trials tary on the rationale for a trial of screening 27:159-/>ane diagnostic follow-up after ficient power to detect important clinical dif- this group, Berg [54] has offered a commen- seemed unaware of these facts [44] breast sonography 13. Swingler GH, Hussey GD, Zwarenstein M. Ran Poor reporting of randomication-A study Radiologists have a great deal to learn about mised controlled trial of clinical outcome aft of 206 randomized controlled trials showed randomized controlled trials. Academic radio- chest radiograph in ambulatory acute lower-respl- that randomization, one of the main design fea- ogists who perform research and radiologists ation infection in children. Lancet 1998: 351 tures necessary to prevent bias in randomized who translate research results into practice 14. Cochrane Library Web site. Available at update-software. com/abstracts/ Other limitations -Additional limitations these trials, including those conducted for di- ab001877. htm. Accessed September 10,2004 identified by Chalmers [46] were inadequate agnostic tests and interventional procedures. 15. Nystrom L, Rutqvist LE, Wall S, et al. Breast can- randomization failure to blind the assessors Radiologists also must be aware of the limita- cer screening with mammography: overview of to the outcomes, and failure to follow up all tions and problems associated with the meth- Swedish randomised trials. Lancet 1993: 341 patients in the trials 973-978 odologic quality and reporting of the trials. It is 16. Dutty sw. Interpretation of the breast screening ope that this article proves to be a valu- able source of information about randomized Intent to Treat Gotzsche and Olsen. Breast 2001: 10: 209-212 controlled trials 17. Miettinen OS. Henschke CL. Pasmantier MW. A method to correct for differential drop- Smith JP, Libby DM, Yankelevitz DF. Mammo- out rates between patients from one arm of aphic screening: no reliable supporting evi- the study and another is to analyze data by Acknowledgments dence? ancer2002;359:404-405 he intent to treatthat is, data are analyzed We thank Alejandro Jadad for his support 18. Tabar L, Vitak B, Chen HHT, Yen MF, Dutty sw the way patients were randomized, regard- and Monika Ferrier for her patience and Smith RA. Beyond randomized controlled trials less of whether or not they received the in- port in keeping us on track and for preparing ganized mammographic screening substantially reduces breast carcinoma mortality. Cancer 2001 ended intervention. The intent to treat the manuscript. 9l:1724-1731 correction is a form of protection against bias nography save lives? CMA and strengthens the conclusions of a study. A 2002;166:l187-1188 detailed discussion of the assessment of the 20. Norman GR, Streiner DL. Biostatistics: the bare References quality of randomized controlled trials wa essentials, 2nd ed. Hamilton, ON, Canada: B C. offered by Jadad [2] I. Book of Daniel 1: 1-21 Decker. 200 In the appraisal of randomized controlled 2. Jadad AR. Randomised controlled trials a user's 21. Silerman WA. Gnosis an gnide. London, England: BMJ Books, 1998 ol Clin Trials 1981: 2: trials, a clear distinction should be made be. 3. Louis pca. Research into the effects of blood- 22. Gray jAM. evidence-base tween the quality of the reporting letting in some infammatory diseases and on the burgh, Scotland: churchill e,l997 quality of methodology of the trials [43]] pneumonitis. Am J Med Sct 1836: 18: 102-111 Duxbury, England: Thomson Learning, 2000 4. Hill AB. The clinical trial. N Engl J Med 1952; 24. Norman GR, Streiner DL. PDO statistics, 2nd ed St Louis, MO: Mosby, 19 Recent Randomized Controlled Trials in 5. Cochrane Library Web site. Available at: 25. Altman DG. Machin D. Bagant TN. Gardner MJ Radiology update-software. com/cochrane. Accessed Septem- Statistics with ce, 2nd ed. London, Eng In recent years, randomized controlled tri- land: BMJ Books, 2000 als have become increasingly popular in ra- 6. Bree RL, Kazeroon EA, Katz S. Eftect of man- 26. Moher D, Dulberg CS, Wells GA. Statistical power diology research In 1997, for instance, there ing use. JAMA 1996276-1595-1598 were only a few good randomized studies in 7. DeVore GR. The routine antenatal diagnostic im- 27. Altman DG. Practical statistics for diagnostic imaging, such as the one by Jarvik aging with ultrasound study: another perspectiv search. London, England: Chapman& Hall, 1991 et al. [47]. Since 2000, the number of good Obstet Gynecol1994:84: 622-626 28. Jadad AR. Rennie D. The randomized controlled R: 183. December 2004 543

Fundamentals of Clinical Research for Radiologists AJR:183, December 2004 1543 used as a proxy measure for methodologic quality. High-quality reporting may hide im￾portant differences in methodologic quality, and well-conducted trials may be reported badly [43]. As Devereaux et al. [41] ob￾served, “[h]ealth care providers depend upon authors and editors to report essential meth￾odological factors in randomized controlled trials (RCTs) to allow determination of trial validity (i.e., likelihood that the trials’ results are unbiased).” The most important limitations of re￾search methods include the following: Insufficient power.—A survey of 71 ran￾domized controlled trials showed that most of these trials were too small (i.e., had insuf- ficient power to detect important clinical dif￾ferences) and that the authors of these trials seemed unaware of these facts [44]. Poor reporting of randomization—A study of 206 randomized controlled trials showed that randomization, one of the main design fea￾tures necessary to prevent bias in randomized controlled trials, was poorly reported [45]. Other limitations.—Additional limitations identified by Chalmers [46] were inadequate randomization, failure to blind the assessors to the outcomes, and failure to follow up all patients in the trials. Intent to Treat A method to correct for differential drop￾out rates between patients from one arm of the study and another is to analyze data by the intent to treat—that is, data are analyzed in the way patients were randomized, regard￾less of whether or not they received the in￾tended intervention. The intent to treat correction is a form of protection against bias and strengthens the conclusions of a study. A detailed discussion of the assessment of the quality of randomized controlled trials was offered by Jadad [2]. In the appraisal of randomized controlled trials, a clear distinction should be made be￾tween the quality of the reporting and the quality of methodology of the trials [43]. Recent Randomized Controlled Trials in Radiology In recent years, randomized controlled tri￾als have become increasingly popular in ra￾diology research. In 1997, for instance, there were only a few good randomized studies in diagnostic imaging, such as the one by Jarvik et al. [47]. Since 2000, the number of good randomized controlled trials has significantly increased in both diagnostic and interven￾tional radiology. Examples of randomized controlled trials in diagnostic imaging in￾clude the works of Gottlieb et al. [48] and Kaiser et al. [49]. Examples of interventional randomized controlled trials are the studies by Pinto et al. [50] and Lencioni et al. [51]. Randomized controlled trials are equally important in screening for disease. Our ini￾tial experience with breast screening was un￾fortunate, and controversy over this issue continues to this day [52, 53]. On the other hand, positive developments have occurred, such as the work of the American College of Radiology Imaging Network. Writing for this group, Berg [54] has offered a commen￾tary on the rationale for a trial of screening breast sonography. Radiologists have a great deal to learn about randomized controlled trials. Academic radiol￾ogists who perform research and radiologists who translate research results into practice should be familiar with the different types of these trials, including those conducted for di￾agnostic tests and interventional procedures. Radiologists also must be aware of the limita￾tions and problems associated with the meth￾odologic quality and reporting of the trials. It is our hope that this article proves to be a valu￾able source of information about randomized controlled trials. Acknowledgments We thank Alejandro Jadad for his support and Monika Ferrier for her patience and sup￾port in keeping us on track and for preparing the manuscript. References 1. Book of Daniel 1:1–21 2. Jadad AR. Randomised controlled trials: a user’s guide. London, England: BMJ Books, 1998 3. Louis PCA. Research into the effects of blood￾letting in some inflammatory diseases and on the influence of tartarized antimony and vesication in pneumonitis. Am J Med Sci 1836;18:102–111 4. Hill AB. The clinical trial. N Engl J Med 1952; 247:113–119 5. Cochrane Library Web site. Available at: www. update-software.com/cochrane. Accessed Septem￾ber 10, 2004 6. Bree RL, Kazerooni EA, Katz SJ. Effect of man￾datory radiology consultation on inpatient imag￾ing use. JAMA 1996;276:1595–1598 7. DeVore GR. The routine antenatal diagnostic im￾aging with ultrasound study: another perspective. Obstet Gynecol 1994;84:622–626 8. Fontana RS, Sanderson DR, Woolner LB, et al. Screening for lung cancer: a critique of the Mayo Lung Project. Cancer 1991;67[suppl 4]:1155–1164 9. [No authors listed]. Impact of follow-up testing on survival and health-related quality of life in breast cancer patients: a multicenter randomized controlled trial—the GIVIO Investigators. JAMA 1994;271:1587–1592 10. Jarvik JG, Maravilla KR, Haynor DR, Levitz M, Deyo RA. Rapid MR imaging versus plain radi￾ography in patients with low back pain: initial re￾sults of a randomized study. Radiology 1997;204:447–454 11. Kinnison ML, Powe NR, Steinberg EP. Results of randomized controlled trials of low-versus high￾osmolality contrast media. Radiology 1989;170: 381–389 12. Rosselli M, Palli D, Cariddi A, Ciatto S, Pacini P, Distante V. Intensive diagnostic follow-up after treatment of primary breast cancer. JAMA 1994; 271:1593–1597 13. Swingler GH, Hussey GD, Zwarenstein M. Ran￾domised controlled trial of clinical outcome after chest radiograph in ambulatory acute lower-respi￾ration infection in children. Lancet 1998;351: 404–408 14. Cochrane Library Web site. Available at: www.update-software.com/abstracts/ ab001877.htm. Accessed September 10, 2004 15. Nystrom L, Rutqvist LE, Wall S, et al. Breast can￾cer screening with mammography: overview of Swedish randomised trials. Lancet 1993;341: 973–978 16. Duffy SW. Interpretation of the breast screening trials: a commentary on the recent paper by Gotzsche and Olsen. Breast 2001;10:209–212 17. Miettinen OS, Henschke CI, Pasmantier MW, Smith JP, Libby DM, Yankelevitz DF. Mammo￾graphic screening: no reliable supporting evi￾dence? Lancet 2002;359:404–405 18. Tabar L, Vitak B, Chen HHT, Yen MF, Duffy SW, Smith RA. Beyond randomized controlled trials: organized mammographic screening substantially reduces breast carcinoma mortality. Cancer 2001; 91:1724–1731 19. Hoey J. Does mammography save lives? CMAJ 2002;166:1187–1188 20. Norman GR, Streiner DL. Biostatistics: the bare essentials, 2nd ed. Hamilton, ON, Canada: B. C. Decker, 2000 21. Silerman WA. Gnosis and random allotment. Control Clin Trials 1981;2:161–164 22. Gray JAM. Evidence-based health care. Edin￾burgh, Scotland: Churchill Livingstone, 1997 23. Rosner B. Fundamentals of biostatistics, 5th ed. Duxbury, England: Thomson Learning, 2000 24. Norman GR, Streiner DL. PDQ statistics, 2nd ed. St. Louis, MO: Mosby, 1997 25. Altman DG, Machin D, Bagant TN, Gardner MJ. Statistics with confidence, 2nd ed. London, Eng￾land: BMJ Books, 2000 26. Moher D, Dulberg CS, Wells GA. Statistical power, sample size, and their reporting in randomized con￾trolled trials. JAMA 1994;22:122–124 27. Altman DG. Practical statistics for medical re￾search. London, England: Chapman & Hall, 1991 28. Jadad AR, Rennie D. The randomized controlled

Stolberg et al ial gets a middle-aged checkup. JAMA 1998, als). The CONSORT statement: revised 99548:67-70 79:319320 mendations for improving the quality of I 47. Jarvik JG, Maravilla KR, Haynor DR, Levitz M, 29. Louis TA, Lavori Pw, Bailar JC Ill, Polansky M parallel-group randomized trials. Deyo RA. Rapid MR imaging versus Crossover and self-controlled designs in clinical re. 2001;285:1987-1991 raphy in patients with low back pai arch. In: Bailar JC Ill, Mosteller F. eds. Medical 40. Altman DG. Better reporting of randomized ults of a randomized study. Radiology 1997: 204 Tses of statistics, 2nd ed. Boston, MA: New Eng- trolled trials: the CONSORt statement. BM/ land Medical Joumal Publications, 1992: 83-104 996,313:570-571 48. Gottlieb RH, Voci SL, Syed L, et al. randomized 30. Woods KL. Megatrials and management of acute 41. Devereaux PJ, Manns BI, Ghali WA, Quan H, prospective study comparing routine versus selec- myocardial infarction. Lancer 1995, 346: 611-614 Guyatt GH. The reporting of methodological fac- tive use of sonography of the complete calf in pa- 31. Charlton BG. Megatrials: methodological issues rs in randomized controlled trials and the asso tients with suspected deep venous thrombosis and clinical implications. Coll Phys Lond 1995; AJR2003;180:241-245 29:96-100 adherence to the Consolidated Standards of Re- 49. Kaiser S, Frenckner B, Jorulf HK Suspected ap- 2. Till JE, Sutherland H, Meslin EM. Is there a role porting Trials( CONSORT) checklist. Control pendicitis in children: US and CT-a prospective for performance assessments in research on Clin Trials2002;23:380-388 andomized study. Radiology 2002: 223: 633-638 of life in oncology? Qualiry Life Res g 42. Chan KBY, Man-Son-Hing M, Molnar FJ, Laupa- 50. Pinto L, Chimeno P, Romo A, et al. Uterine fi- cis A. How well is the clinical importance of study braids: uterine artery embolization versus abdom- 33. Silverman WA, Altman DG. Patient preferences sults reported? an assessment of randomized co and randomized trials. Lancer 1996, 347: 171-174 olled trials. CMA/ 2001: 165: 1197-120 randomized, and controlled clinical trial. Radio/- 34. Zelen M. A new design for randomized clinical 43. Huwiler-Mantener K, Juni P, Junker C, Egger M. ogy 2003: 226: 425-431 Med1979300:1242-1245 juality of reporting of randomized trials as a 51. Lencioni RA, Allgaier HP, Cioni D, et al. Small 35. Olschewski M, Scheurlen H Comprehensive Co- neasure of methodologic quality. JAMA hepatocellular carcinoma in cirrhosis: random- ort Study: an alternative to randomized consent 2002: 287: 2801-2804 ized comparison of radio-frequency thermal abla- design in a breast preservation trial Methods Inf 44. Freiman JA, Chalmers TC, Smith H, Kuebler RR. tion versus percutaneous ethanol injection. led198524:13 The importance of Beta, the type 2 error, and Radiology2003228235-240 36. Brewin CR, Bradley C Patient preferences and ran- sample size in design and interpretation of 52, Dean PB. Gotzsche's quixotic antiscreening cam- omized clinical trials. BM/ 1989: 299: 684-685 lomized controlled trials. N Engl J Med paign: nonscientific and contrary to Cochrane 37. Hellman S. Hellman DS. Of mice but not men 978;29969069 principles. JACR 2004: 1: 3-7 roblems of the randomized trial. N Engl J Med 45. Schulz KF, Chalmers I, Hayes R, Altman DJ. 53. Gotzsche PC. The debate on breast cancer screen- 1991;324:1585-1592 mpirical evidence of bia tant. JACR ethodologic quality associat JAMA 54. Berg WA. Rationale for a trial of screening breast the CoNSORt statement. JAMA 1996, 276: 7-9 995,273:408412 ultrasound: American College of Radiology Imag 39. Moher D, Schulz KF, Altman DG, CONSORT 46. Chalmers I Applying overviews and meta-an ing Network (ACRIN) 6666. AR 2003: 180: roup( Consolidated Standards of Reporting Tri- is at the bedside: discussion. J Clin Epide 1225-1228 The reader's attention is directed to earlier articles in the fundamentals of clinical research series 1. Introduction, which appeared in February 2001 8. Exploring and Summarizing Radiologic Data, January 2003 2. Framework, April 2001 9. Visualizing Radiologic Data, March 2003 3. Protocol. June 2001 10. Introduction to Probability Theory and Sampling 4. Data Collection, October 2001 Distributions, April 200 5. Population and Sample, November 2001 11. Observational Studies in Radiology, November 2004 6. Statistically Engineering the Study for Success, July 2002 7. Screening for Preclinical Disease: Test and Disease Characteristics, October 2002 1544 UR- 183. December 2004

1544 AJR:183, December 2004 Stolberg et al. trial gets a middle-aged checkup. JAMA 1998; 279:319–320 29. Louis TA, Lavori PW, Bailar JC III, Polansky M. Crossover and self-controlled designs in clinical re￾search. In: Bailar JC III, Mosteller F. eds. Medical uses of statistics, 2nd ed. Boston, MA: New Eng￾land Medical Journal Publications, 1992:83–104 30. Woods KL. Megatrials and management of acute myocardial infarction. Lancet 1995;346:611–614 31. Charlton BG. Megatrials: methodological issues and clinical implications. Coll Phys Lond 1995; 29:96–100 32. Till JE, Sutherland HJ, Meslin EM. Is there a role for performance assessments in research on qual￾ity of life in oncology? Quality Life Res 1992; 1:31–40 33. Silverman WA, Altman DG. Patient preferences and randomized trials. Lancet 1996;347:171–174 34. Zelen M. A new design for randomized clinical trials. N Engl J Med 1979;300:1242–1245 35. Olschewski M, Scheurlen H. Comprehensive Co￾hort Study: an alternative to randomized consent design in a breast preservation trial. Methods Inf Med 1985;24:131–134 36. Brewin CR, Bradley C. Patient preferences and ran￾domized clinical trials. BMJ 1989;299:684–685 37. Hellman S, Hellman DS. Of mice but not men: problems of the randomized trial. N Engl J Med 1991;324:1585–1592 38. Begg C, Cho M, Eastwood S, et al. Improving the quality of reporting of randomized controlled trials: the CONSORT statement. JAMA 1996;276:7–9 39. Moher D, Schulz KF, Altman DG, CONSORT Group (Consolidated Standards of Reporting Tri￾als). The CONSORT statement: revised recom￾mendations for improving the quality of reports of parallel-group randomized trials. JAMA 2001;285:1987–1991 40. Altman DG. Better reporting of randomized con￾trolled trials: the CONSORT statement. BMJ 1996;313:570–571 41. Devereaux PJ, Manns BJ, Ghali WA, Quan H, Guyatt GH. The reporting of methodological fac￾tors in randomized controlled trials and the asso￾ciation with a journal policy to promote adherence to the Consolidated Standards of Re￾porting Trials (CONSORT) checklist. Control Clin Trials 2002;23:380–388 42. Chan KBY, Man-Son-Hing M, Molnar FJ, Laupa￾cis A. How well is the clinical importance of study results reported? an assessment of randomized con￾trolled trials. CMAJ 2001;165:1197–1202 43. Huwiler-Müntener K, Jüni P, Junker C, Egger M. Quality of reporting of randomized trials as a measure of methodologic quality. JAMA 2002;287:2801–2804 44. Freiman JA, Chalmers TC, Smith H, Kuebler RR. The importance of Beta, the type 2 error, and sample size in design and interpretation of ran￾domized controlled trials. N Engl J Med 1978;299:690–694 45. Schulz KF, Chalmers I, Hayes RJ, Altman DJ. Empirical evidence of bias: dimensions of the methodologic quality associated with estimates of treatment efforts in controlled trials. JAMA 1995;273:408–412 46. Chalmers I. Applying overviews and meta-analy￾sis at the bedside: discussion. J Clin Epidemiol 1995;48:67–70 47. Jarvik JG, Maravilla KR, Haynor DR, Levitz M, Deyo RA. Rapid MR imaging versus plain radi￾ography in patients with low back pain: initial re￾sults of a randomized study. Radiology 1997;204: 447–454 48. Gottlieb RH, Voci SL, Syed L, et al. Randomized prospective study comparing routine versus selec￾tive use of sonography of the complete calf in pa￾tients with suspected deep venous thrombosis. AJR 2003;180:241–245 49. Kaiser S, Frenckner B, Jorulf HK. Suspected ap￾pendicitis in children: US and CT–a prospective randomized study. Radiology 2002;223:633–638 50. Pinto I, Chimeno P, Romo A, et al. Uterine fi- broids: uterine artery embolization versus abdom￾inal hysterectomy for treatment—a prospective, randomized, and controlled clinical trial. Radiol￾ogy 2003;226:425–431 51. Lencioni RA, Allgaier HP, Cioni D, et al. Small hepatocellular carcinoma in cirrhosis: random￾ized comparison of radio-frequency thermal abla￾tion versus percutaneous ethanol injection. Radiology 2003;228:235–240 52. Dean PB. Gotzsche’s quixotic antiscreening cam￾paign: nonscientific and contrary to Cochrane principles. JACR 2004;1:3–7 53. Gotzsche PC. The debate on breast cancer screen￾ing with mammography is important. JACR 2004;1:8–14 54. Berg WA. Rationale for a trial of screening breast ultrasound: American College of Radiology Imag￾ing Network (ACRIN) 6666. AJR 2003;180: 1225–1228 1. Introduction, which appeared in February 2001 2. Framework, April 2001 3. Protocol, June 2001 4. Data Collection, October 2001 5. Population and Sample, November 2001 6. Statistically Engineering the Study for Success, July 2002 7. Screening for Preclinical Disease: Test and Disease Characteristics, October 2002 8. Exploring and Summarizing Radiologic Data, January 2003 9. Visualizing Radiologic Data, March 2003 10. Introduction to Probability Theory and Sampling Distributions, April 2003 11. Observational Studies in Radiology, November 2004 The reader’s attention is directed to earlier articles in the Fundamentals of Clinical Research series: