Trials within cohorts (TwiCs): A novel design to efficiently embed pragmatic randomised trials into cohort studies

TwiCs: Why and how? 

Randomised clinical trials are the gold standard for causal inference in medical research. However, randomised clinical trials often face various challenges, including high costs, slow participant recruitment, limited generalisability, burdensome consent procedures, and a disappointment bias that may occur in open-label trials if participants and providers change their behaviour when participants are not allocated to their preferred group.^1–3 

In recent years, trials within cohorts (TwiCs) has emerged as a pragmatic trial design with the potential to overcome these challenges.^4–9 Studies with the TwiCs design involve recruiting participants with a condition of interest into a prospective cohort. At enrolment, not only is consent obtained for regular prospective data collection, but participants are also informed about randomisation into future trials nested within the cohort. In a future trial using the TwiCs design, participants are approached only if they are randomised to the intervention group and are then given the option to accept or decline the proposed intervention. The participants randomised to the control group are not informed about the intervention being offered to other cohort participants but continue usual care and regular data collection as part of the cohort (see Figure 1). This consent procedure mimics usual care in that individuals are informed about new treatment options but not about treatments they may not receive. Additionally, the TwiCs design offers a comparison to a real-life control group, allows researchers to recruit efficiently from a well-described cohort, and embeds outcome collection efficiently within the cohort’s follow-up structure.

TwiCs: What do we know?

In a recent scoping review, Amstutz, Schönenberger, Gerber, et al. identified 46 trials in 14 different countries that were conducted with a TwiCs design up to December 2022.¹⁰ The most common medical fields in which the design was applied were oncology (24%), infectious diseases (17%), and mental health (15%). A typical trial with a TwiCs design was investigator-initiated, was publicly funded, and recruited outpatients. The TwiCs in the review evaluated various types of interventions – mostly behavioural, psychological, or complementary interventions (42%) – as well as drugs (13%) and radiotherapy (9%). 

Based on how ethics committees in three different countries guided trialists who implemented the TwiCs design, three major consent patterns have emerged (see Figure 2). In the Dutch pattern, there are three separate consent steps for cohort participation, randomisation, and intervention; in the French pattern, there is combined consent for cohort participation and randomisation and separate intervention consent; and in the UK pattern, there is consent only for cohort participation and intervention (randomisation consent is not mentioned). Among the 46 trials with the TwiCs design, the UK pattern was the most common (41%), followed by the Dutch pattern (37%) and the French pattern (22%).

TwiCs: What are the design’s limitations and mitigation strategies?

The TwiCs design presents several challenges and limitations. First, the consent procedure involves multiple stages with tailored information provided at each stage, which requires training the participating sites and carefully communicating consent information when initially implementing the TwiCs design. Nevertheless, once cohort consent and randomisation consent are part of routine cohort enrolment procedures, participants will only be asked for intervention consent in all future TwiCs. Because intervention consent is closer to routine clinical decision-making, the consent process in these trials promises to be less burdensome, less complex, and less distressful than the consent process in traditional trials.^11–13 

Second, the control group in a TwiCs study is, by design, always receiving usual cohort care. Consequently, a placebo-controlled comparison is not possible, and participants and providers are aware of the intervention received/provided. In most pragmatic trials, however, a usual care comparator is the option of choice. Therefore, the TwiCs design may even offer a comparison group that is closer to reality since the randomised groups do not know there are other groups (masked allocation). To mitigate undesired open-label effects, researchers may choose clinical endpoints that are hard to modify (e.g. survival) or blinded outcome assessors.

Third, trials with a TwiCs design are embedded in a cohort, and data collection is strictly dictated by the type and frequency of the routine follow-up visits in the overarching cohort. Since the control group remains unaware of the trial, additional assessments and visits are generally not possible. However, if the cohort is built up with the first TwiCs study in mind, as was the case for more than 50% of the trials with a TwiCs design reviewed by Amstutz, Schönenberger, Gerber, et al., the follow-up can be tailored to meet the necessary data collection frequency and endpoints. This was demonstrated in some radiotherapy TwiCs conducted in Utrecht and some COVID-19 drug TwiCs conducted in Paris.^14–19

Fourth, while in the control group all eligible participants are included by design, some eligible participants will decline the proposed intervention (non-uptake), resulting in an imbalance of uptake across the groups. Across all the trials with a TwiCs design that were reviewed, non-uptake was highly variable, ranging from 0% to 75%. If non-uptake is high, the intention-to-treat estimand will not reflect a direct intervention effect but merely an offer-of-intervention effect. Moreover, non-uptake should be accounted for in the sample size calculation, which only 37% of TwiCs in the review did.¹⁰ While an intention-to-treat estimand is of interest to policymakers, it may have limited value for participants and treating physicians.²⁰ Instrumental variable and inverse probability weighting can be applied to estimate per protocol estimands accounting for non-uptake, but they depend on the data available and the type of non-uptake (time-varying versus one-time) and require careful consideration of the underlying assumptions of such observational causal inference approaches.^21–27

TwiCs: Has the design been used in Switzerland?

The Swiss HIV Cohort Study (SHCS) is the first Swiss cohort – and, notably, the first HIV cohort worldwide – to implement the TwiCs design. Over a ten-month period with various stakeholder meetings, the SHCS worked closely with patient representatives to adapt the cohort protocol to reflect the Dutch consent pattern. The SHCS obtained ethics approval for the amended protocol, and in August 2024 it started rolling out randomisation consent across its sites in order to prepare for the implementation of future trials using the TwiCs design. The first such trial is to be started by the end of 2024 and will test the effect of a preference-based choice of different nicotine replacement products on smoking cessation in people living with HIV in Switzerland. The TwiCs design may enable researchers to efficiently generate high-quality, randomised evidence using existing cohort infrastructure in Switzerland and elsewhere. Early insights from the pioneering roll-out of the design and the first TwiCs study in the SHCS will determine if the anticipated benefits of the TwiCs design – such as a more realistic comparator, less burdensome consent procedures, and improved recruitment efficiency – outweigh its limitations. 

To follow developments related to the TwiCs design or receive information about and support with the approach, researchers may visit the TwiCs network’s website (www.twics.global).

References

1. Young RC (2010) Cancer clinical trials: A chronic but curable crisis. The New England Journal of Medicine 363(4):306–309. doi: 10.1056/nejmp1005843
2. S and Zwarenstein M (2009) Making trials matter: Pragmatic and explanatory trials and the problem of applicability. Trials 10(1):37. doi: 10.1186/1745-6215-10-37
3. Speich B et al. (2022) Nonregistration, discontinuation, and nonpublication of randomized trials: A repeated metaresearch analysis. PLoS Medicine 19(4):e1003980. doi: 10.1371/journal.pmed.1003980
4. Relton C, Torgerson D, O’Cathain A, and Nicholl J (2010) Rethinking pragmatic randomised controlled trials: Introducing the “cohort multiple randomised controlled trial” design. BMJ 340:c1066. doi: 10.1136/bmj.c1066
5. Young-Afat DA et al. (2016) Brief report: Staged-informed consent in the cohort multiple randomized controlled trial design. Epidemiology 27(3):389–392. doi: 10.1097/EDE.0000000000000435
6. Kim SY, Flory J, and Relton C (2018) Ethics and practice of trials within cohorts: An emerging pragmatic trial design. Clinical Trials 15(1):9–16. doi: 10.1177/1740774517746620
7. Couwenberg AM et al. (2020) The trials within cohorts design facilitated efficient patient enrollment and generalizability in oncology setting. Journal of Clinical Epidemiology 120:33–39. doi: 10.1016/j.jclinepi.2019.12.015
8. van der Velden JM et al. (2017) The cohort multiple randomized controlled trial design: A valid and efficient alternative to pragmatic trials? International Journal of Epidemiology 46(1):96–102. doi: 10.1093/ije/dyw050
9. Vickers AJ et al. (2018) Just-in-time consent: The ethical case for an alternative to traditional informed consent in randomized trials comparing an experimental intervention with usual care. Clinical Trials 15(1):3–8. doi 10.1177/1740774517746610
10. Amstutz A, Schönenberger CM, Gerber F et al. (2024) Characteristics, consent patterns, and challenges of randomized trials using the trials within cohorts (TwiCs) design: A scoping review. Journal of Clinical Epidemiology 174:111469. doi: 10.1016/j.jclinepi.2024.111469
11. Young-Afat DA et al. (2021) Oncology patients were found to understand and accept the trials within cohorts design. Journal of Clinical Epidemiology 130:135–142. doi: 10.1016/j.jclinepi.2020.10.015
12. Vickers AJ et al. (2023) A randomized comparison of two-stage versus traditional one-stage consent for a low-stakes randomized trial. Clinical Trials 20(6):642–648. doi: 10.1177/17407745231185058
13. Verweij ME et al. (2022) Most patients reported positively or neutrally of having served as controls in the trials within cohorts design. Journal of Clinical Epidemiology 148:39–47. doi: 10.1016/j.jclinepi.2022.04.015
14. Couwenberg AM et al. (2020) Efficacy of dose-escalated chemoradiation on complete tumor response in patients with locally advanced rectal cancer (RECTAL-BOOST): A phase 2 randomized controlled trial. International Journal of Radiation Oncology, Biology, Physics 108(4):1008–1018. doi: 10.1016/j.ijrobp.2020.06.013
15. Mariette X et al. (2022) Sarilumab in adults hospitalised with moderate-to-severe COVID-19 pneumonia (CORIMUNO-SARI-1): An open-label randomised controlled trial. The Lancet Rheumatology 4(1):E24–E32. doi: 10.1016/S2665-9913(21)00315-5
16. The CORIMUNO-19 Collaborative group (2021) Effect of anakinra versus usual care in adults in hospital with COVID-19 and mild-to-moderate pneumonia (CORIMUNO-ANA-1): A randomised controlled trial. The Lancet Respiratory Medicine 9(3):295–304. doi: 10.1016/S2213-2600(20)30556-7
17. Hermine O et al. (2022) Tocilizumab plus dexamethasone versus dexamethasone in patients with moderate-to-severe COVID-19 pneumonia: A randomised clinical trial from the CORIMUNO-19 study group. eClinicalMedicine 46:101362. doi: 10.1016/j.eclinm.2022.101362
18. Hermine O et al. (2021) Effect of tocilizumab vs usual care in adults hospitalized with COVID-19 and moderate or severe pneumonia: A randomized clinical trial. JAMA Internal Medicine 81(1):32–40. doi: 10.1001/jamainternmed.2020.6820
19. Lacombe K et al. (2023) Use of COVID-19 convalescent plasma to treat patients admitted to hospital for COVID-19 with or without underlying immunodeficiency: Open label, randomised clinical trial. BMJ Medicine 2(1):e000427. doi: 10.1136/bmjmed-2022-000427
20. Murray EJ et al. (2018) Patients and investigators prefer measures of absolute risk in subgroups for pragmatic randomized trials. Journal of Clinical Epidemiology 103:10–21. doi: 10.1016/j.jclinepi.2018.06.009
21. Mansournia MA et al. (2017) Biases in randomized trials: A conversation between trialists and epidemiologists. Epidemiology 28(1):54–59. doi: 10.1097/ede.0000000000000564
22. Hernán MA and Robins JM (2017) Per-protocol analyses of pragmatic trials. The New England Journal of Medicine 377(14):1391–1398. doi: 10.1056/nejmsm1605385
23. Murray EJ and Hernán MA (2016) Adherence adjustment in the Coronary Drug Project: A call for better per-protocol effect estimates in randomized trials. Clinical Trials 13(4):372–378. doi: 10.1177/1740774516634335
24. Hernán MA and Scharfstein D (2018) Cautions as regulators move to end exclusive reliance on intention to treat. Annals of Internal Medicine 168(7):515–516. doi: 10.7326/m17-3354
25. Hewitt CE, Torgerson DJ, and Miles JNV (2006) Is there another way to take account of noncompliance in randomized controlled trials? Canadian Medical Association Journal 175(4):347. doi: 10.1503/cmaj.051625
26. Hernán MA and Hernández-Díaz S (2012) Beyond the intention-to-treat in comparative effectiveness research. Clinical Trials 9(1):48–55. doi: 10.1177/1740774511420743
27. Peugh JL et al. (2017) Beyond intent to treat (ITT): A complier average causal effect (CACE) estimation primer. Journal of School Psychology 60:7–24. doi: 10.1016/j.jsp.2015.12.006