How to Conduct a Randomized Controlled Trial

Abstract

Randomized controlled trials (RCTs) are considered to produce the highest level of evidence in the original studies that informs the practice of evidence-based medicine (EBM). By manipulating an independent variable to study its impact on the outcome, RCTs establish causal relationships and provide valuable insights into clinical treatment. To improve patient outcomes and optimize the use of clinical resources, the practice of EBM plays a crucial role in designing and conducting RCTs to evaluate the effectiveness of clinical interventions. This review aims to explore the essential steps involved in conducting a rigorous and reliable RCT, ensuring the generation of high-quality evidence.

Introduction

In recent years, randomized controlled trials (RCTs) have greatly influenced the practice of evidence-based medicine (EBM) around the world.¹ RCTs are widely regarded as the best research design for evaluating the impact of clinical interventions on patient outcomes.² This experimental design involves random assignment of an intervention or standard care to individuals and measures the outcomes of interest.³ A well-designed RCT allows us to ascertain whether the study intervention is effective and the extent of its impact.

Analytical research designs can be broadly categorized into 2 groups: experimental and non-experimental (Table 1). Experimental designs, such as RCTs, manipulate the independent variable to quantify the relationship between the intervention and the outcome.⁴ Non-experimental or observational designs, on the other hand, simply observe the association between the independent and outcome variable. The major drawback of non-experimental studies is their inability to establish causality, which refers to a cause-and-effect relationship. Causality is a fundamental concept in clinical research as it helps estimate the true effect of an intervention on the outcome. Experimental study designs are superior to non-experimental designs primarily because they establish causality and are conducted in a controlled setting, minimizing the risk of selection and contamination bias.⁵ As a result, RCTs generate the highest level of evidence and occupy the top position among original studies in the hierarchy pyramid of evidence level (Fig. 1).^6,7 The evidence produced from experimental study designs has the potential to influence clinical practice and health care delivery through EBM.⁸

Table 1.

Advantages and Disadvantages of Different Study Designs

Fig. 1.

Hierarchical pyramid of evidence level. RCT = randomized controlled trial.

EBM involves integrating the best available research evidence, clinical expertise, and patient values. The term EBM was first introduced in 1990 and emphasizes the use of research evidence as the basis for clinical decision-making. In the past, clinical practice relied heavily on the clinician's intuition, unsystematic clinical experience, and pathophysiological findings.⁹ EBM has been shown to improve the quality of care provided to patients with respiratory illnesses.¹⁰ Examples include shortened mechanical ventilation duration by implementing daily spontaneous breathing trials,¹¹ decreased mortality with the utilization of lung-protective ventilation,¹² and reduced re-intubation rate with the use of prophylactic noninvasive respiratory support postextubation for patients with high risk factors.¹³ Experimental research designs play a crucial role in medical advancements and serve as a cornerstone of EBM. These designs are robust and conducted under rigorous conditions with stringent methodologies, enabling researchers to accurately address research questions.

It is essential for clinicians to be capable of designing high-quality studies and possess the skills to critically evaluate the existing literature. The primary objective of this paper is to outline the components of designing a high-quality RCT. This information is intended not just for those conducting RCTs but also for those reading the literature and relying on it for their clinical practice.

Components of an RCT

An RCT consists of several essential components that contribute to its high quality (Fig. 2). These should be carefully planned before (a priori) beginning the study (ie, subject enrollment).

Fig. 2.

Essential components of an randomized controlled trial. CONSORT = Consolidated Standards of Reporting Trials. RCT = randomized controlled trial.

Formulating the Research Question

The research question serves as the foundation of an RCT. Therefore, the first step in designing an RCT is to develop a well-defined research question that is answerable through empirical experimentation.¹⁴ A clear and specific research question is vital for ensuring the quality of the RCT. There are 2 commonly used methods for defining a research question: FINER and PICO.

FINER is an acronym that stands for feasible, interesting, novel, ethical, and relevant. It outlines the key aspects that researchers should consider when formulating a research question.^15–17 A good research question should be feasible, taking into account available resources, patient population, expertise, and time. It should also be interesting and address a research question that either fills the knowledge gap or adds scientific knowledge to the existing evidence. Ethical concerns should be considered when formulating the research question, ensuring that the study poses no harm to the study subjects. Additionally, the research question should be relevant to the field, with findings that have practical implications.

PICO is another useful strategy for developing a research question.^15,16 PICO stands for population, intervention, comparison, and outcome. It emphasizes incorporating these elements into the research question. For example, a research question using PICO could be “in obese patients (P), are recruitment maneuvers (I) more effective than standard care (C) in reducing the postextubation respiratory failure rate (O)?”¹⁸

Formulating Research Hypotheses

A research hypothesis represents the expected relationship between the independent and dependent variables. Hypothesis testing, using various statistical tests, allows researchers to draw inference about the population based on the study sample. Every research question should be accompanied by a research hypothesis, which includes null and/or alternate hypothesis. The null hypothesis states that there is no difference between the 2 study groups, whereas the alternate hypothesis posits a specific direction based on the available evidence. For example, a null hypothesis could be “there is no difference in the incidences of postextubation respiratory failure in obese patients treated with NIV compared to standard care.” On the other hand, an alternate hypothesis could be “the use of NIV reduces the incidence of postextubation respiratory failure in obese patients.” Based on the study hypothesis, RCTs can be classified into 3 types: superiority, non-inferiority, and equivalence trials (Table 2).^19,20

Table 2.

Superiority, Non-inferiority, and Equivalence Trials

Superiority RCTs are the most common trials and aim to demonstrate that a new intervention/treatment is superior to the currently available standard or placebo treatment.¹⁹ For example, a study by Amaru et al¹⁸ tested the hypothesis that recruitment maneuvers were superior to standard care in reducing the incidence of postextubation respiratory failure in obese subjects. Similarly, Nair et al²¹ investigated whether the primary use of high-flow nasal cannula (HFNC) was superior to noninvasive ventilation (NIV) in reducing the intubation rate in subjects with severe COVID-19 and reported lower intubation rate at day 7 in the HFNC group than the NIV group (20% vs 33%, P = .045).

Non-inferiority RCTs aim to demonstrate that the new intervention/treatment is not worse than the standard treatment and may offer additional benefits such as cost-effectiveness or improved patient adherence.²² These trials are typically conducted when there is an effective standard treatment. Establishing a non-inferiority margin based on previous literature is an essential step in designing such trials. For example, Zheng et al²³ tested the hypothesis that NIV was non-inferior to CPAP in reducing the postextubation failure rate in infants who underwent cardiac surgery, using a predefined non-inferiority margin of < 5% and reported that NIV (15%) was not inferior to CPAP (12%) (P = .52).

Equivalence RCTs aim to establish that a new intervention/treatment is neither worse nor better than the standard treatment within a predetermined margin. These trials are conducted to assess if the new treatment can achieve similar outcomes as the standard treatment. Wood et al²⁴ conducted an equivalence trial that demonstrated the equal effectiveness of 2 different pharmacologic agents (ketorolac and meperidine) in providing analgesia for acute renal colic.

Study Design

Among experimental controlled trials, there are 3 types of study designs: RCT, non-RCT, and randomized crossover trial (Fig. 3).

Fig. 3.

Types of randomized controlled trial designs. RCT = randomized controlled trial.

Randomized, parallel-group, controlled trial.

The purpose of randomly assigning treatments to the study groups is to assure that the study groups are similar in baseline characteristics and reduce the risk of selection bias. This methodology allows for an unbiased, true estimate of the treatment effect with greater internal and external validity. As a result, this design is usually the preferred choice for conducting controlled experiments. In a parallel-group design, each study participant is exposed to only one of the study interventions.⁶ For example, Nair et al²¹ conducted a randomized, parallel-group, controlled trial to evaluate the effectiveness of HFNC and NIV among subjects with severe COVID-19 in reducing intubation rate. In this trial, subjects were randomly assigned to receive either HFNC or NIV.

Non-randomized, parallel-group, controlled trial.

Under certain circumstances, the process of randomization may not be feasible, may delay subject enrollment, or may pose ethical concerns.²⁵ In such scenarios, conducting non-RCT provides an opportunity to address the research question. In this design, subjects are allocated to receive study treatment using non-random methods, such as assigning interventions based on subjects' medical record number or date of birth.³ The major drawback of this method is that it may introduce selection bias and imbalance the 2 study groups, potentially impacting the relationship between treatment and outcome due to confounding variables.²⁶ For example, Qian et al²⁷ conducted a non-RCT to assess the impact of awake prone positioning versus standard care on non-intubated subjects with COVID-19. In this study, the assignment was determined using the medical record numbers, with subjects having odd medical numbers receiving awake prone positioning and those with even medical record numbers assigned to usual care. This quasi-experimental design might be considered an observational study due to the lack of randomization.

Randomized crossover trial.

In this design, each study participant is randomly assigned to receive both study interventions in a predetermined sequence over a specified period. The time in between the interventions is referred to as a washout period, which allows dissipation of any carryover effect from the previous intervention. In this design, each participant acts as their own control, thereby reducing subject variability. Additionally, this design requires fewer participants than a parallel-group design because each study subject is exposed to multiple interventions. The main limitation of this design is that an insufficient washout period may lead to a carryover effect, confounding the study results. These studies are limited to those with a surrogate physiologic outcome. Clearly, this design is not possible if the outcome is mortality. Li et al²⁸ used a randomized crossover design to evaluate fugitive aerosol generation during aerosol delivery via HFNC. Similarly, Lersritwimanmaen et al²⁹ used a randomized crossover study to compare the physiologic effects of high-flow oxygen therapy and conventional oxygen therapy in subjects with a tracheostomy who received prolonged mechanical ventilation and were undergoing ventilator weaning.

Study Population

The study sample should be representative of the population to ensure external validity and generalizability. The target study population should be clearly defined with inclusion and exclusion criteria. Inclusion criteria are the specific characteristics of the sample that directly relate to answering research questions. In clinical research, inclusion criteria often encompass demographic, clinical, and geographic characteristics.³⁰ On the other hand, exclusion criteria are the sample characteristics that, although relevant to the research question, could hinder the study completion, raise ethical concerns, or pose potential risks. Narrow inclusion and exclusion criteria can lead to homogeneous study samples and minimize the impact of confounding variables on study outcomes. However, it may limit the generalizability of the study findings to the larger population.^22,31 It is, therefore, imperative to strike a balance, ensuring that the study eligibility criteria are narrow enough to address the research question effectively while also being broad enough for the results to have applicability beyond the study sample to the larger population.

Study Groups

Depending on the research question, an RCT can have 2 or more study groups, with 2 study groups/arms, control and intervention group, are the most common.

Control group.

This arm consists of subjects who are not exposed to study intervention/treatment. This group serves as a comparison group, allowing researchers to assess whether the outcomes observed with the new treatment/intervention are superior, equivalent, or non-inferior to the standard treatment/intervention.³¹ The choice of treatment/intervention for the control group depends on the presence of a preexisting and widely accepted treatment for the specific condition being studied. When a proven treatment is already established as the standard of care, researchers should utilize it as the standard treatment to address the research hypothesis.

In a study conducted by Kaur et al³² that evaluated the effect of endotracheal tube scraping on the duration of mechanical ventilation, the control group received endotracheal suctioning using standard in-line suction catheter, which represents the established standard of care. Withholding an established standard treatment from subjects for research purposes is unethical and should be avoided. Likewise, researchers must ensure that the experimental treatment is at least as effective as the established standard treatment.³¹

Alternatively, if there is no established treatment available, researchers may opt to employ a placebo or sham treatment as the control group. A placebo is an inactive substance that resembles active treatment but lacks any therapeutic effect. By utilizing a placebo, participants in the control group believe that they are receiving an active treatment, thereby eliciting the placebo effect. For example, in an RCT conducted by Beaumont et al³³ that examined the effects of local anesthetic cream compared to a placebo on the pain level during arterial blood draws, the control group received a placebo cream that closely resembled the actual local anesthetic cream. Additionally, some studies utilize sham treatments for the control groups. For instance, a study by Lacasse et al³⁴ used a sham treatment involving the delivery of ambient air through concentrators as compared to 2–4 L/min supplemental oxygen to determine if long-term nocturnal supplemental oxygen therapy affects mortality or disease progression in subjects with COPD. The standard treatment/intervention provided to the control group should be clearly defined in the research protocol.

Intervention group.

This arm consists of subjects that are actively exposed to the study intervention/treatment. The specific details of the intervention/treatment should be clearly described in the research protocol. For example, in a study by Lersritwimanmaen et al,²⁹ the intervention group received high-flow oxygen therapy at a flow of 50 L/min and F_IO2 of 0.4. Similarly, in a study by Kaur et al,³² the intervention group received endotracheal suctioning via in-line suction catheter equipped with balloon-sweeping technology.

Multiple group.

Some studies include more than 2 study arms to minimize costs and accelerate study completion by reducing sample size.³⁵ This design allows for various combinations of treatment and control arms. This study design enables the testing of multiple interventions simultaneously, leading to faster identification of effective treatments with fewer resources.^35,36 A notable example is the parallel, 3-group RCT by Perkins et al,³⁷ who tested the effectiveness of 2 interventions (CPAP and HFNC) against conventional oxygen therapy in reducing intubation rate or mortality among subjects with COVID-19.

Ethical Considerations

When designing experimental studies, it is essential to follow ethical guidelines to protect the study subjects from any undue risks. One important ethical principle is that researchers should not withhold an established proven treatment from the control group. For instance, it might be unethical to use a placebo for the control group in a bronchodilator study that is the proven treatment for treating bronchospasm. The concept of equipoise should be maintained, ensuring that there is genuine uncertainty about which treatment is more beneficial. The study should be designed in a manner that generates clinically relevant and scientifically sound information while treating each study participant with respect, beneficence, and fairness.³⁸

The study protocol must be approved by the institutional review board. Obtaining informed consent is a fundamental step to provide information and seek permission from each study participant before their participation in the study. Each participant is approached in a non-coercive manner and given ample time before making a voluntary decision about their participation. It is important to note that obtaining informed consent does not guarantee the ethical conduct of research. The study investigator must also ensure that the proposed research has scientific value and validity, allows fair selection of subjects, and promotes a favorable risk-benefit ratio; and if it poses more than minimal risk, the study needs to be monitored by an independent safety board, commonly referred to as data safety and monitoring board.³⁹

Clinical Trial Registration

Trial registration involves uploading the study details to a public platform that allows research information to be available for public use. Its primary goal is to ensure transparency and scrutiny of any trial that prospectively assigns human subjects to intervention or control groups to examine the causal relationship.⁴⁰ In the past, negative trials were less likely to be published, leading to a bias in the available research literature and hindering the information from the public to make an informed decision regarding future study participation.^41,42 By registering the trial, research information becomes accessible to the public, regardless of the trial outcome, which helps establish trust and confidence in the research process. As a result, the International Committee of Medical Journal Editors requires that all RCTs be registered on a nonprofit and publicly accessible platform.⁴¹ A trial should be registered before the beginning of the subject enrollment. Examples of platforms to register RCT include ClinicalTrials.gov, managed by the United States National Library Medicine; European Union and European Economic Area (https://euclinicaltrials.eu); China (https://www.chictr.org.cn); and World Health Organization (https://www.who.int/clinical-trials-registry-platform) Accessed June 2, 2023.

Study Outcomes

The study outcomes or end points are directly tied to the study hypothesis and should be predefined to evaluate the effectiveness of the study intervention. Study outcomes can be classified into 2 categories: patient-centered outcomes or non–patient-centered outcomes, also known as surrogate outcomes.⁴³ Patient-centered outcomes, also referred to as patient-important outcomes, are those that directly matter to the patient, such as mortality, intubation rate, pain, and quality of life. On the other hand, surrogate outcomes are used as substitutes for patient outcomes, such as blood gases, biomarkers, and radiological tests results. Notably, sometimes there is a discrepancy between surrogate outcomes (eg, blood gases) and patient-important outcomes (eg, mortality). An example is the ARDS Network study in which the high tidal volume group had better blood gases but higher mortality compared to the low tidal volume group.¹²

Primary study outcome refers to the outcome that was utilized to design the study, calculate the sample size, and will help answer the primary research question.¹⁴ This outcome is of utmost importance in evaluating the success or failure of the study intervention. Secondary outcomes are additional outcomes of interest that may provide valuable information about the intervention's effects and shed light on underlying mechanisms.⁴⁴ It is common for clinical studies to have one primary outcome and multiple secondary outcomes. The operational definitions of these outcome variables should be clearly stated by the authors to facilitate future researchers to replicate the study. For example, Amaru et al¹⁸ defined postextubation failure (the use of invasive ventilation, NIV, or HFNC within 48 h after extubation) as the primary outcome. The sample size calculation was based on detecting at least a 20% decrease in the incidence of postextubation respiratory failure with 80% power and a 5% alpha risk (P < .05). The study also included secondary outcomes such as P_aO2/F_IO2, change in breathing frequency during the spontaneous breathing trial and after extubation, ICU length of stay, hospital length of stay, and mortality.

It is important to note that the primary outcome carries more weight in assessing the study rigor and should be interpreted with less susceptibility to bias compared to secondary outcomes. In addition to the primary and secondary outcomes, some studies also report exploratory outcomes, which are primarily used to generate hypotheses or formulate research questions for future studies. Lastly, composite outcomes are also seen in some studies, where multiple related outcomes are combined into one outcome measure.⁴⁴ Composite outcomes are commonly reported as primary outcomes and offer advantages such as increased statistical power and inclusion of competing outcomes.⁴⁵ However, the main drawback of composite outcome is the inability to determine the impact of individual outcomes. For example, combining intubation and death as a composite outcome may not clearly answer whether a specific intervention prevents intubation or reduces mortality.⁴⁶

Sample Size Calculation

One of the key strengths of an RCT is its statistical power to detect true differences between the study groups and provide reliable and generalizable findings.^3,5,6 Therefore, the power analysis or sample size calculation is an important step when designing an RCT. The sample size calculation is determined by factors such as the effect size and its variability.⁴⁷ Effect size refers to the magnitude of the difference in the primary outcome between 2 study groups. This information can be derived from previously published studies that assessed similar outcomes. If no relevant data are available, pilot study data or expert opinion may be utilized to estimate the effect size. In addition to the effect size, investigators need to specify the desired level of statistical significance (α) and power (1-β) when calculating the sample size. Alpha (α) represents the probability of making a type-1 error, which is rejecting the null hypothesis when it is true. The commonly used threshold is 0.05, indicating a 5% probability of making a type-1 error. β represents the probability of making a type-2 error, which fails to reject the null hypothesis when it is false, thereby failing to detect a true difference or effect. Typically, a β of 0.2 is considered acceptable in clinical research. This results in a power of 0.8, or an 80% likelihood of detecting a difference if one exists. By specifying the effect size, level of statistical significance, and desired power, researchers can calculate the required sample size for their RCT. Larger sample sizes are generally needed to detect smaller effect sizes or to achieve higher levels of power.

Funding

Research funding encompasses the financial support available to facilitate and sustain research endeavors. Various local and national sources provide funding opportunities to support research activities. Additionally, study investigators can also apply for industry funds that support research initiatives. Whereas obtaining research funding is not mandatory, it is highly advantageous, particularly for large-scale studies that typically require substantial resource. Therefore, study investigators are encouraged to actively seek out relevant organizations that provide grants tailored to support research aligned with the study objective. By doing so, they can enhance their chances of successfully conducting and completing their research projects.

Study Implementation

In the study implementation phase of a study, several key considerations must be addressed, including the randomization method, allocation concealment, treatment blinding, protocol adherence monitoring, study sites, and study team training.

Randomization is crucial to ensure unbiased assignment of the study interventions. It involves using computer-generated random numbers or random tables to allocate participants.⁴⁸ It helps establish baseline comparability between the groups and reduces selection bias.^3,22 The random numbers are usually generated by statisticians or independent members to reduce potential bias.⁶ Randomization methods can be simple, stratified, blocked, or clustered, depending on the study design and objectives (Table 3). Simple randomization involves using random numbers to assign study treatments to participants. It is the most common and simplest method of randomization, often employed in large clinical studies. However, it may result in uneven distribution of certain baseline characteristics that may influence the study outcomes. Stratified randomization allows for stratification of participants based on specific baseline characteristics (strata) that are likely to impact the outcomes of interest. By ensuring a balanced distribution of these characteristics across study groups, stratified randomization enhances the control over potential confounders. For example, in a study comparing HFNC or NIV, participants could be stratified based on P_aCO2 to ensure balanced representation of subjects with hypercapnia in each group.

Table 3.

Randomization Methods

Blocked randomization refers to using predetermined block sizes when generating random numbers. This method ensures an equal number of participants in each group within each block. This is commonly used for smaller studies and includes block sizes of either 4, 6, 8, or 10. For instance, with a block size of 4 of a total number of 24 subjects to equally assign therapy HFNC (X) or NIV (Y) to subjects, the generated randomized numbers may be arranged as XXYY, XYXY, XYYX, YXYX, YYXX, and YXXY. This method ensures that among every 4 subjects 2 subjects will receive HFNC and the other 2 will receive NIV in a random fashion.

Cluster randomization refers to assigning study treatment to clusters or groups of participants rather than individuals. Clusters can be defined based on geographic locations (eg, different ICUs or hospitals), clinical settings, or time periods. Cluster randomization is conducted when other methods of randomization are not feasible and when the treatment effect may be impacted by uncontrolled factors within clusters. For example, to study the impact of respiratory therapist (RT)–driven, protocolized lung expansion therapy on the incidence of atelectasis, instead of randomly assigning each subject to either intervention or control arm an investigator may randomize entire ICUs (cluster). One ICU will be assigned to receive protocolized lung expansion therapy, whereas a similar ICU will serve as the control group receiving standard care. The reason for using cluster randomization in this study example is to reduce the risk of contamination bias that could influence the study outcomes. If individual subjects are randomized, there would be a concern that RTs assigned to administer the protocolized therapy to one subject may inadvertently apply the same strategy to subjects in the control group, thus contaminating the results. It is important to acknowledge that cluster randomization has its own challenges, including complex design, larger sample sizes, potential imbalance in subject characteristics, subject selection bias, and poor reporting methods.⁴⁹

Allocation concealment refers to the process of concealing the assignment of treatments from the study investigators until the subjects are ready to be enrolled.^3,6 This is done to minimize selection bias, as prior knowledge of which subjects will receive study treatment versus control can influence the enrollment process. A common method of allocation concealment is the use of opaque envelopes that contain randomization sequence numbers. These envelopes are securely stored and opened in a sequence to avoid any bias in treatment assignment. Another approach to achieve allocation concealment is through central randomization. In central randomization, after a subject is consented, the investigator contacts a central telephone number or accesses a web site to obtain the assignment information. This method provides additional protection in concealing the treatment allocation manipulation compared to the envelope method, which is susceptible to tampering.

Blinding refers to the process in which either the study participant (single blind) or both the study participant and the study investigator (double blind) are unaware of the treatment assigned to the participant. In contrast, open-label studies are conducted when both the participants and the study personnel are aware of the treatment being administered.⁵⁰ Certain studies, such as those involving drugs or subjective experiences, are particularly susceptible to subjective bias if the participant or the research team is aware of the treatment being given. For example, in an open-label study, participants assigned to a novel investigational drug may be more likely to report better symptom control compared to those receiving standard treatment, potentially impacting the observed treatment effect. Similarly, investigators who are aware of the novel investigational drug may provide better clinical attention to those receiving that therapy compared to the standard group, thus potentially influencing the study outcomes. However, it is important to recognize that not all studies can be designed as blinded trials due to various technical constraints. For example, a study assessing the impact of NIV or HFNC on intubation rate would be impossible to blind. In such cases where blinding or masking is not feasible, researchers should exercise caution and avoid relying on subjective measurements as primary outcomes to minimize potential biases.

Study sites.

A single-center RCT entails recruiting all study participants at a single research site, whereas a multi-site RCT involves recruiting participants across multiple research sites. Conducting a single-center RCT provides greater control over research activities, minimizes confounding variables, and facilitates coordination among the research team. However, single-center trials often take longer to complete due to limitations in recruiting an adequate number of subjects, and the findings may have limited generalizability to other settings. In contrast, multi-site RCTs offer the advantage of broader participant recruitment, which enhances the external validity and generalizability of the study findings. Nevertheless, conducting multi-site RCTs presents greater complexity, including scientific challenges such as ensuring adherence to study protocol, as well as operational challenges like site selection, randomization procedure, oversight, and maintaining enthusiasm among participating sites.⁵¹

Study team training.

To conduct a successful clinical trial, it is essential for the primary investigator to ensure that all study personnel are adequately trained on various aspects of the study. This includes familiarity with the study protocol, understanding of the informed consent process, proficiency in study-related procedures, knowledge of the enrollment process, and competence in data collection methods. Additionally, it is important for the principal investigator to clearly communicate the roles and responsibilities of each team member. For multi-center studies, training is usually conducted by the designated central coordinating center.

Clinical team training.

If the study implementation involves bedside staff, it is crucial to provide them with appropriate research training to ensure their understanding and collaboration.

Protocol adherence.

Another crucial aspect of study implementation is maintaining protocol adherence. The study team must closely monitor each enrolled subject to ensure that all the study steps outlined in the study protocol are followed accurately and consistently. The study team should be readily available to address any research-related queries from subjects, their families, or bedside clinicians. Any deviation from the protocol should be documented and, if required, promptly reported to the institutional review board.

Statistical Analysis

During the primary analysis of RCT results, investigators utilize various statistical approaches to derive study conclusions. Three commonly used methods are intent-to-treat (ITT), per-protocol, and per-treatment analysis.⁵² ITT analysis refers to the method of statistical analysis that includes all randomized subjects, regardless of their adherence to the study protocol, in the final analysis. The primary aim of ITT analysis is to preserve the randomization process and maintain balanced study groups, thus providing a true estimate of the treatment effect. By including all subjects, ITT analysis reflects the real-life effectiveness of the intervention, accounting for scenarios where subjects may not always adhere to the study regimen, dropout from the study, or experience external influences. This approach is usually considered conservative to avoid overinflating the study findings.⁵² However, a major limitation of ITT analysis is that the true treatment effect may be diluted due to the inclusion of subjects who did not successfully complete the study or crossed into the other group. On the other hand, per-protocol analysis refers to the statistical method that includes only those subjects who successfully completed the study without crossover into the other group in the final analysis. This approach aims to assess the effectiveness of the study intervention under ideal conditions and minimize the dilution of treatment effect by excluding subjects who did not fully adhere to the protocol. Alternatively, per-treatment analysis includes only those subjects who received treatment at the end of trial, regardless of their original randomized assignment. However, both the per-protocol and per-treatment methods substantially compromise the ability to draw definitive study conclusions due to selection bias resulting from the loss of randomization and subject dropout, which may lead to unbalanced study groups.

Reporting Results of an RCT

The goal of conducting an RCT is to disseminate the study findings to the scientific community. These findings are crucial as they have the potential to influence clinical practice and shape health care delivery. However, for RCT findings to have a meaningful impact, it is essential that they are well conducted, reported effectively, and published in a timely manner. Poor reporting of well-conducted RCT findings can lead to insufficient evidence to influence health care practice, whereas inadequately designed RCTs can provide false results that misguide the clinical practice. Therefore, RCT findings should be reported in a transparent, accurate, and holistic manner. This allows readers to assess the reliability and validity of the research findings. To promote accurate reporting of RCTs, most journal editors require investigators to follow the Consolidated Standards of Reporting Trials (CONSORT) guidelines (https://www.equator-network.org/reporting-guidelines/consort) Accessed June 11, 2023.⁵³ These guidelines offer a flow diagram to document the flow of study participants throughout the study and a checklist of all the essential components to be included in an RCT manuscript. The use of CONSORT checklist has been shown to improve the quality of reporting RCTs.⁵⁴ According to CONSORT guidelines, investigators are required to report baseline demographic and clinical characteristics for each group. Additionally, primary and secondary outcome results for each group, along with the estimated effect size and its precision, must be reported. If the outcomes are recorded on a nominal scale, the results should be presented with both absolute and relative effect sizes.⁵³

Summary

A rigorous RCT research design is widely regarded as the highest standard for assessing the effectiveness of clinical interventions and plays a pivotal role in practicing EBM. Consequently, clinicians engaged in conducting RCTs or evaluating the evidence should diligently adhere to all essential steps to generate and evaluate robust and trustworthy evidence that can inform the policies, clinical intervention, and health care delivery.