This EGD for acupuncture is the first such document in the field of Complementary, Alternative and Integrative Medicine (CAIM) and may serve as a model for other CAIM interventions. Here we will describe the methodological issues that impact CER in acupuncture research.
Acupuncture as a complex intervention
Acupuncture is a component of Chinese medicine, which comprises a “whole medical system” using unique diagnostic and therapeutic approaches.
] To date, only a few studies have evaluated the effectiveness of acupuncture as part of a complex Chinese medicine intervention
]. In addition, acupuncture needling itself could be seen as just one component of a complex intervention. For research purposes, in their White Paper
], the Society of Acupuncture Research Board of Directors divided the components of acupuncture treatments into the following groups:
Needling components (i.e., needle size, retaining time, depth, stimulation, location, frequeny)
Specific (acupuncture theory-related) non-needling components that are traditionally considered to have therapeutic value, for example, in Chinese Medicine the physical components such as palpation.
Generic, nonspecific non-needling components that are not unique to acupuncture treatments, such as belief and expectations of the practitioner and patient, therapeutic setting, time and attention.
For the field of complementary medicine, several authors have highlighted the importance of comparing different treatment options that reflect usual care
]. Furthermore a research strategy has been suggested aimed at establishing comparative effectiveness before assessing component efficacy
]. Using the proposed strategy will generate evidence relevant to clinical practice. It will also emphasize the important but sometimes subtle differences between CAM and conventional medical practice
]. From a translational approach, it is recommended that acupuncture should be studied “top down” in real-world, multi-component “whole-system” trials, as well as “bottom up” in mechanistic studies that focus on causal pathways and on understanding how individual treatment components work and interact
]. In this manner, clinical and basic research can inform one another, thereby benefiting clinical relevance and design of future research.
In order to provide a realistic assessment of treatment options in a usual care context, CER trials often include multiple comparison groups, more heterogeneous patients and longer-term outcomes. CER trials will also require additional time to engage stakeholders in study planning and to engage and train participating practitioners working in usual care. Cluster-randomized trials on acupuncture were recommended as one of the priorities for CER by the IOM
]. and might help overcome recruitment problems. Pilot studies are an important instrument when planning CER to determine feasibility, test outcome measures and provide information for sample size calculation.
A Bayesian (adaptive) statistical approach might be an interesting option for real-world CER where it is “noisier” than in “efficacy” studies. Standard statistical (frequentist) techniques to determining study size require increased sample sizes in effectiveness studies, because of greater variability and the fact that when comparing several active treatments relatively small differences are expected
]. Designs that use features that change or “adapt” in response to information generated during the trial can be more efficient by using formal, probabilistic statements of uncertainty based on the combination of all sources of information both from within and outside a study
]. For further information see
Research question and the efficacy-effectiveness continuum
When planning a study, it is important to determine the location on the efficacy-effectiveness continuum. In other words, should the study be designed primarily to exclude bias and increase the chance of detecting a specific effect or to reflect usual care?
Efficacy studies aim to produce results with high internal validity by reducing both bias (e.g. using blinding) and variation (e.g. narrow eligibility criteria and a standardized treatment protocol) to increase the likelihood of detecting the hypothesized difference. These studies aim to assess a treatment outcome under “ideal” conditions with highly selected patients. At the other end of the continuum, results from effectiveness studies aim to inform decision makers about the potential benefits of an intervention in a usual care setting, making the results of the study generalizable and relevant to routine care by studying patients within the context (methods and practitioners) in which the acupuncture treatment is usually deployed.
The place along the continuum is multi-dimensional and is influenced by the research question
]. Since there is a range of relevant clinical research questions on acupuncture, each study context, outcome and design (e.g. nature of the comparison group) should clearly reflect the study question.
Possible research questions are
Does acupuncture needling have acupuncture point specificity (i.e. is needling at real acupuncture points superior to a penetrating sham procedure at non-acupuncture points)?
Does acupuncture needling have a specific effect (i.e. is needling at real acupuncture points superior to a non-penetrating sham procedure at acupuncture points or non-acupuncture points)?
Does depth of needling, type of stimulation or other needling parameters have an effect on the outcome?
Is acupuncture treatment superior or non-inferior to conventional standard treatment?
Is acupuncture treatment in addition to usual care or standard care superior to usual care or standard care alone?
What dose (number of treatments) of acupuncture is needed to see an effect?
Safety and costs
What are the types and frequency of acupuncture side effects?
Is acupuncture treatment cost-effective compared to another intervention?
Studies that aim only to determine the specific effect of acupuncture are not part of CER, because sham acupuncture is not considered an “active” comparison group or as a real treatment option. Only trials that compare a minimum of two treatment options are considered to represent CER
]. However, a study that includes a standard care or usual care comparison group in addition to a sham acupuncture group can contribute to CER. There is an ongoing discussion about the challenges of using sham controls in acupuncture studies
], which should be considered when planning those studies. When comparison groups labeled as “usual care”, “standard care,” or “best practice” are used, specific details of the interventions should be described.
A helpful tool for planning randomized trials is the PRECIS (pragmatic–explanatory continuum indicator summary)
]), which includes the following ten dimensions that have an influence on the efficacy-effectiveness continuum: eligibility criteria of participants, flexibility of the intervention protocol, flexibility of the comparison group treatment protocol, expertise of the practitioners in the intervention group, expertise of the practitioners in the comparison group, adherence of practitioners, compliance of participants, follow-up intensity, outcome parameters, and analysis of study population.
Starting from these PRECIS dimensions
] as well as the six characteristics of CER developed by the IOM
], we propose to structure this guidance document a simplified list of five criteria that are most relevant for the study design of CER: participant selection/eligibility criteria, treatment protocols, practitioner expertise, outcomes, and setting in which the study is conducted.
The following paragraphs will introduce the methodological background for some designs aspects that are highly relevant for CER on acupuncture.
Developing the acupuncture and comparison group interventions
The treatment protocols for the intervention and the comparison group are central aspects of the study. The STRICTA guidelines for reporting acupuncture trials should be taken into account when planning intervention and comparison group treatment protocols
Acupuncture treatment protocol
The acupuncture intervention in CER could be a new intervention (e.g. new acupuncture style or microsystem) or an intervention that is already available (e.g. Chinese style needle acupuncture). When evaluating a new intervention, its characteristics (e.g. dosage, frequency, setting) should reflect the context in which it will be deployed in the future in a usual care context. When evaluating acupuncture as an existing intervention that is widely available for the external validity of the results, it is important that the treatment protocol of the acupuncture group reflects common practice to the extent possible. However, this could be difficult because of the large heterogeneity in acupuncture styles and expertise of the practitioners. Registries (see below) or prospective observational studies are helpful sources to inform researchers about practice in usual care. If this information is not available, cross-sectional surveys or consensus procedures with acupuncture experts are the next best option. However, when selecting the experts participating in the consensus, their expertise should reflect the heterogeneity of the acupuncture practised in usual care. If consensus is lacking before large randomized comparative effectiveness studies are performed, pilot studies comparing different treatments should be considered
]. Intervention planning, should also take into account that acupuncture styles, as well as training and experience of practitioners can vary considerable in different states and countries.
Comparison group treatment protocol
The comparison group treatment protocol should be planned and developed with the same rigor and attention to detail as the acupuncture intervention. When standard treatment or usual care is used as the comparison group, the treatment components should be described in the study protocol and detailed usage during the study should be documented and reported.
In CER, co-interventions (e.g. pain medications in pain trials) are often more heterogeneous than in efficacy trials, and for understanding the context of the study, it is relevant to document and report them.
The Agency for Health Care Research and Quality (AHRQ) in the US has defined patient registries as an organized system that uses observational study methods to collect uniform data (clinical and other) to evaluate specified outcomes for a population defined by a particular disease, condition, or exposure, and that serves a predetermined scientific, clinical, or policy purpose(s)
]. Registries are seen as a valuable complement to randomized controlled trials in determining real-world outcomes in the practice of medicine and do not generally have restrictive inclusion or exclusion criteria, nor do they specify the therapy to which the health care provider must adhere. They can be used to evaluate outcomes for diverse purposes ranging from the natural history of a disease, to the safety of drugs or devices, to the real-world effectiveness of therapies. These patient registries are designed to answer predefined questions by choosing a study design, measurable outcomes, the study population and analysis. In the registry design, potential sources of bias (systematic error) should be addressed to the extent that is practical and achievable
]. Population-based registries are the most desirable kind of registry, because they minimize selection of participants.
Another approach is to use existing data, usually from the electronic medical records of a well-defined health care delivery system, for comparative effectiveness research
]. This line of research tends to use statistical methods that differ from the clinical trial- based approaches in AHRQ-type registries, including both propensity matching for synthetic trial designs and Heckman’s selection models
], but these methods are more difficult to employ than conventional analyses and methodological problems remain.
An important reason for extending acupuncture CER research into these domains is the opportunity to evaluate this complex intervention as it is actually practiced, in the types of patients who are willing to use it, and in the settings where it is generally used.
Although CER promotes registries and observational studies in addition to randomized trials, randomized studies are still considered the most robust method for generating comparative effectiveness evidence, as they are more likely to control for bias and confounding. The randomized trial will undoubtedly remain an important component of an advanced CER framework
]. It is essential that the treatment groups are comparable before treatment. Randomization as a stochastic method has historically been the method of choice and is still the most frequently used method. To prevent selection bias it is important to conceal the allocation sequence from those assigning participants to intervention groups, until the moment of assignment. This prevents researchers from (unconsciously or otherwise) influencing which participants are assigned to a given intervention group
Dynamic allocation procedures could be an alternative to randomization procedures. Depending on the study design, they could balance treatment arms across baseline prognostic factors for clinical trials more effectively than randomization
]. Dynamic allocation uses algorithms to determine each subsequent patient’s assignment in a manner that produces the best overall balance between treatment groups. Design-adaptive allocation methods (minimization) use only patients’ baseline characteristics to determine the allocation of the next patient
]. Assignments using minimization are more predictable than with randomization, but there are ways to avoid this (see Taves for rank minimization
]). Minimization is gaining more acceptance and was recognized by CONSORT as an acceptable alternative to random assignment
Response-adaptive allocation algorithms use both previous patients’ baseline characteristics and outcomes. There is a lack of successful studies, especially in acupuncture research, using these response-adaptive allocation approaches.
Although methodologically important, allocation methods create an artificial context that may influence the outcome, e.g. they could have a direct impact on the selection of participants. Treatment preferences that are influenced by patient expectations are considered a relevant determinant of the placebo effect
]. For example, if patients’ preferences have a significant influence on whether they choose to participate in a study and how they respond to treatment, the findings of randomized trials may not apply to those patients who would avoid participating in trials and instead seek other options in usual care. Several designs have been developed to investigate whether preferences matter. The most important is probably the partially randomized patient preferences design. In such trials, eligible patients are asked whether they have a preference for one of the two (or more) treatments and whether they agree to be randomized. Those not giving consent to randomization receive the treatment of their choice, while those without strong preference are randomized
]. On an exploratory level this design provides information about whether the results observed among randomized patients are different from those among patients who were not randomized because of strong treatment preferences. A number of large acupuncture trials have used variants of this design, including incentives to motivate patients to accept randomization
Randomization methods including both simple and stratified randomization could be used to allocate patients to the treatment groups. Stratified randomization is a useful approach to control for relevant factors for imbalances in treatment groups. In settings where randomization at the patient level is not feasible, cluster randomization might be an option. Cluster trials randomize social units, such as households, primary care practices, hospital wards, classrooms, neighborhoods and entire communities, to different intervention arms. Compared with randomized trials at a patient level, cluster trials have less statistical power while using the same number of patients
]. Because cluster randomization is often applied to an entire hospital or clinic population without exclusion, it enhances recruitment and generalizability. Cluster trials allow resources to be used to best advantage and due to this they are likely to be less expensive than the “traditional” randomized controlled trial. However, cluster-randomized trials also have limitations, for example, selection bias can occur and withdrawal drop-out of whole clusters may seriously affect the validity of the results. For further guidance on cluster-randomized trials see
Blinding and expectation
Blinding reduces bias by minimizing the possibility that the results can be affected by expectations of patients and practitioners. It is commonly accepted that practitioners cannot be blinded to the acupuncture intervention. In contrast, the intent to blind patients is common practice in efficacy trials that compare acupuncture with sham acupuncture, even though data from sham-controlled acupuncture trials suggest that patients’ expectations might influence the treatment outcome
]. Further, patient blinding seems to be impossible when comparing acupuncture with usual care or standard care. Thus, in this latter type of CER trial, where neither practitioner nor patient is blinded and the outcomes are not assessed by the patient, it is important to include an assessor for the relevant outcome who is blinded to the patients’ group assignment. This is not possible in trials that use patient self-rated outcomes such as chronic pain measures or quality of life. In such cases 1) questionnaires sent to the patients’ home or 2) or blinded telephone interviewers could help to assure that the practitioners do not influence patients’ outcome assessments. Another option is to assess patients’ and practitioners’ expectations and to use them as covariates when analyzing the data to adjust for different expectations in the treatment groups.
Health economic evaluations
Over the last few years, health economic data has become increasingly relevant for decision makers faced with the challenge of reconciling the growing demand for healthcare services with the funds available
]. The number of economic evaluations of complementary and integrative medicine has increased over the last few years and the largest number has been done for acupuncture either in addition or compared to routine care
]. There are a variety of best practices for performing economic analyses that should be considered when incorporating costs into CER
Recommendations for acupuncture comparative effectiveness research
These recommendations aim to support optimal use of resources for generating evidence that will inform stakeholder decision-making as fully as possible. They are based on the assumption that it is a conscious decision of the relevant stakeholders to reduce internal validity in order to increase generalizability, relevance, feasibility and timeliness of research results. Multiple stakeholders were involved in the consensus process for this EGD to balance aspects of internal and external validity in the recommendations.
Recommendations developed for the following fields are summarized in (see also Table
Designing CER studies
Treatment protocol, expertise and setting
Study design and statistical analyses
Checklist for CER relevant aspects for acupuncture clinical studies
Designing CER studies
1) Stakeholder engagement
a) Involvement of all relevant stakeholders (e.g. practitioners, clinicians, patients, payers, researchers) is highly relevant for CER.
b) Identifying relevant topics, planning and design, and interpreting study results should involve all relevant stakeholders.
2) Study question and the efficacy-effectiveness continuum
a) The study question should be clearly phrased, and include all relevant information about study participants, interventions, comparison groups and outcome parameters. In particular, it should clarify whether the acupuncture treatment is to be assessed as an “alternative” in direct comparison, for superiority or non-inferiority, to or as an adjunctive to a usual or standard care treatment.
b) During the trial planning phase, time should be given to discuss and determine the trial’s position in the efficacy-effectiveness continuum for the following aspects: participant selection/eligibility criteria, treatment protocol, practitioner expertise, outcomes, study setting. Using the more detailed PRECIS tool to support this process is recommended
3) Eligibility criteria
a) In the context of available resources, eligibility criteria should be as broad as possible. The criteria should reflect the evidence of the pattern of usage and disease burden, and the study population should reflect all well-known relevant disease characteristics that may interact with the treatment.
b) Patients with co-morbidities should not be explicitly excluded from the study enrollment, with the exception of relevant safety aspects. Both acupuncture naïve and non-naïve patients should be enrolled as well.
c) The study disease/condition should be defined as clearly as possible.
4) Patient recruitment
a) The type of comparator group chosen should take into account its possible effect on recruitment success.
b) As far as possible, recruitment strategies should be similar for all treatment arms and recruitment should be carried out in places where the respective treatments are usually employed (e.g., both CAM and conventional clinics and practitioners).
c) Sampling of study participants should be as systematic as possible, for example, by using registries or health insurance records and inviting a random sample or systematic sample of patients seeking treatment (e.g. every 50th patient) from the health insurance record list.
d) Patients’ treatment preferences should be determined and efforts should be made to recruit both those who express strong treatment preferences and those who do not.
Treatment protocol, expertise and setting
5) Acupuncture intervention
a) The acupuncture treatment protocol should to some extent reflect real-world clinical care. If acupuncture points are predefined, the point selection should be based on broad systematic consensus (e.g., expert opinion survey, consensus meetings and/or textbook survey).
b) Different acupuncture treatment protocols (standardized, semi-standardized, fully individualized) should be compared in a design with multiple treatment arms.
c) Standardized acupuncture may take Chinese pattern diagnoses into account by:(1) including only those patients who have the most frequent pattern differentiation for the condition of interest, with treatment protocol tailored to this pattern, or (2) having different pre-established standardized treatment protocols for different Chinese pattern differentiations
6) Comparison groups
a) The treatment protocol for the comparison group should be based on broad systematic consensus (e.g., expert opinion survey, consensus meetings and/or textbook survey).
b) If the terms “usual care”, “standard care” or “best practice” are used to describe comparison group treatment, a detailed description must be provided, stating how these terms are defined within the trial. In addition, relevant references should be provided.
7) Treatment documentation
a) All treatments (interventions, co-interventions, over-the-counter self-medication) carried out in all groups should be documented. A variety of documentation methods and sources may contribute details (including medical records, case report forms, etc.).
a) If no standard outcome measures for the respective disease/condition exist, or are not suitable for CER and no pilot data is available, pilot studies should be carried out to explore the value of a broad range of outcomes that may be sensitive to the condition treated and the interventions. Qualitative and quantitative research methods could be combined to identify all relevant associated outcomes.
b) The main outcome measures should focus on patient-centered outcomes and, if appropriate, include relevant biological measures for the respective disease/condition. Diagnosis-specific validated standards for outcome measures published by professional associations (e.g., the International Headache Society on headache measures
]) or expert consensus (e.g., on back pain measures
]) should be followed, when validated for the context of use (e.g. cultural group), and whenever possible, to permit better comparison of study results.
c) Multiple primary outcomes addressing distinct dimensions may be used, if appropriate. This should be addressed in the sample size calculation and statistical analysis plan.
d) There is great value in reporting the percentage of patients with clinically meaningful response for the main outcome measures, because this information is easily interpreted by non-scientists.
e) Secondary outcome measures should capture relevant patient-centered dimensions of the respective disease/condition (both self-reported and biological) and might, if appropriate, include measures of collateral effects (i.e., positive and negative consequences of the treatment experience, often seemingly unrelated to the main outcomes).
a) In chronic diseases long-term observations (at least 1
year) are highly recommended.
b) The use of periodic intervals to document and compare the trajectory and persistence of treatment effects is recommended. However, the frequency of assessment should be balanced, so that relevant information is gained without major disruptions of treatment implementations or practice setting.
Study design and statistical analysis
a) The clinical trajectory of individual patients, which can be tracked by registries, could inform future acupuncture studies (e.g., as to intervention details, outcomes, safety, or characteristics of a usual care population that receives acupuncture treatment outside of a study situation).
b) When collecting registry data, information should be obtained from all types of providers who offer the services of interest.
c) Registries may be useful in identifying frequency of important potential covariates, including spirituality and beliefs, prior CAIM use, nutrition, exercise, etc., but this information is often not included.
d) Registries could serve as a recruitment platform for comparative studies.
e) If experience with registries in acupuncture research is lacking, the development and use of pilot registries may prove helpful.
11) Allocation methods
a) Use of appropriate allocation methods is strongly recommended. Randomization at the level of individual patients is still the method most frequently used, but dynamic allocation procedures (e.g. rank minimization
]) may be used as an alternative. The final choice depends on the design of the study as a whole.
b) Stratified randomization or adaptive allocation techniques may be used to prevent imbalances for relevant covariates and potential confounders in study arms.
c) Partially randomized patient preference designs have an advantage in that they provide additional exploratory information, whether the results observed for randomized patients are different from those not randomized because of strong treatment preferences. These designs, while adding potentially important outcomes data to a clinical trial, are often not feasible because of the need for much larger sample sizes.
d) Cluster randomization is recommended by the Institute of Medicine especially for acupuncture studies and may be used if feasible
]. When planning such a trial, it is necessary to consult the relevant literature and local institutional roles to determine from whom, when, and how informed consent must be obtained
], and to take into account that a larger sample might be needed than in patient level randomized trials
e) Standard procedures ensuring allocation concealment (e.g., central randomization or secure databases) should be employed. Recommendations for stratification will be given under point 15.
a) Blinded outcome measurement (e.g., a blinded rater) is recommended in order to reduce bias, especially for outcomes that are assessed by the practitioner (e.g. range of motion as a physical assessment). Methods to minimize the risk of unblinding (e.g., allocation concealment, rater training, standardized assessment protocol) should be employed.
b) Patient-reported trial outcomes data (e.g., migraine days, quality of life) should be kept inaccessible to the practitioner (e.g., by using sealed envelopes or preferably by sending questionnaires directly to a study office independent of the intervention site or using a blinded interviewer). This should not influence the symptoms usually reported by the patients to the practitioner for treatment follow up.
13) Patient preferences and expectation
a) Patient preferences should, if appropriate, be acknowledged in the study design, e.g., by using a partially randomized patient preference design. If such a design is not feasible, then it is important to document both the patients’ preferences regarding the treatment options available in the trial as well as the degree of their knowledge about these treatment options.
b) Patient and practitioner preferences and expectations should be assessed before the intervention begins (in randomized trials before randomization).
14) Sample size
a) Sample size depends mainly on the outcome(s) of main interest and the minimum clinically important difference (MCID) for the respective outcome(s).
b) Sample size should account for greater heterogeneity in CER study populations; because of this researchers should specifically avoid conducting small trials (< 50 patients per arm) in CER, unless there is a specific reason to do such studies (e.g., pilot studies to test feasibility and recruitment).
a) Relevant subgroups for the disease/condition under study should be identified based on existing data and the literature. Subgroup analysis should be done at least for gender, because there is preliminary evidence that acupuncture might be more effective in women than men
]. Also of interest are subgroup analyses for different Chinese pattern diagnoses and for acupuncture patients, who are naïve/non-naive to acupuncture. Further analyses could be carried out for age, ethnicity, disease severity/duration, treatment preference and recruitment site.
b) The main subgroup analyses should be pre-specified in the analysis plan and included in sample size planning for confirmatory testing. Further subgroup analyses could be done on an exploratory level, but should be stated as an objective in the study protocol.
16) Statistical analysis
a) Primary analysis for trials on superiority of acupuncture should be intention-to-treat. In order to assess real-world effectiveness of treatments, benefits and harms should be judged and compared according to the treatment to which patients were assigned.
b) Analyses should adjust for relevant potential confounders (e.g., baseline value of the outcome measure, stratification variables, expectation).
c) Especially in non-randomized studies (e.g. observational data from registries), procedures to compensate for baseline differences must be used (e.g., matching and/or adjusted analysis).
a) Comparing the effectiveness of treatment options should be the primary aim of CER, but economic evaluations should be included whenever possible as a secondary aim.
b) To allow realistic cost estimates, the setting(s) of the study should reflect as closely as possible the real-world clinical practice setting for each intervention. If a study includes a standardized and a non-standardized acupuncture arm, it would be useful to compare their cost-effectiveness.
18) Methodological approach
a) Standard methods for economic evaluations should be employed, and effectiveness measures that include both benefits and harms (e.g., utility measures based on SF-36, SF-12 or EQ-5D) should be used
b) Sensitivity analysis should include, whenever possible, different stakeholder perspectives (e.g., society, payer and patient). Because acupuncture is often paid out-of-pocket, the patient’s perspective is highly relevant.
c) Subgroup analysis should mainly follow the subgroups defined a priori for the effectiveness study. A subgroup analysis for gender is recommended, because there is preliminary evidence that gender may influence the cost-effectiveness of acupuncture treatment
d) Exploratory analyses of factors that predict a better treatment response are suggested to develop future hypotheses.
19) Observation time
a) In chronic diseases long-term observations (at least 1
year) with intermediate measurement time points are highly recommended, in order to evaluate cost-effectiveness as developed over time.
20) Existing guidelines
To ensure that CER on acupuncture will fulfill reporting standards, the relevant CONSORT guidelines should be consulted and followed:
CONSORT for parallel group randomized trials
CONSORT extension for cluster randomized trials
CONSORT extension for pragmatic trials
CONSORT for non-pharmacological trials
STRICTA as CONSORT extension for acupuncture studies
a) Publication of a detailed study protocol (design publication) should take place whenever possible prior to publication of study results.
b) The study should be registered in an international accessible trial database with as many details as possible provided.
c) Publication of the completed study should describe why and how it qualifies as CER and make clear the phase of the study.
d) The setting of the study should be described, including information about the typical usual care setting in the country where the study was performed (and if relevant in other countries). The procedure for selection of practitioners for each treatment group should be described, with an account of whether and how those included in the study differ from the average practitioner (e.g. training, experience).
e) Wording of treatment group descriptions in the informed consent should be provided.
f) If a usual-care or standard-care comparison group is used, a detailed description with citations should be included in the intervention section.
g) Detailed results of all treatments, adherence, and co-interventions in the different groups should be provided.
h) The most relevant subgroup analyses and analyses of patient characteristics that predict a better outcome should be published together with primary results. Detailed subgroup analysis and/or de-identified patient level data could be provided as online files.