Table 2 Further examples of applying the strategies to different intercurrent events

Example 1: A trial aims to estimate the effect of providing individuals access to an IBI in comparison to no treatment in a scenario where no antidepressants are available. Some individuals in the IBI arm will start antidepressants in parallel (=ICE), initiated outside the trial. For these individuals, an antidepressant-free assessment will not be available. A hypothetical strategy could be employed to model a scenario in which all participants have completed the IBI without taking antidepressants. Researchers must assume that data from participants who did not start antidepressants is sufficient to model symptom development under the hypothetical scenario. However, individuals who start antidepressants may differ from those who do not initiate antidepressants; covariates to adjust for this may not be observed. If the MAR assumption does not hold, the estimates will be biased. If starting antidepressants in parallel is common clinical practice, the estimate lacks clinical relevance.

Example 2: A study aims to evaluate the effect of a treatment regimen that comprises an IBI targeting depression in patients undergoing cancer treatment. Some participants may discontinue the intervention (=ICE) due to a deteriorating somatic condition unrelated to the treatment. First, suppose the depressive severity at a fixed time point is of interest (e.g., 3 months after randomization). One may model the missing assessments by focusing on the hypothetical scenario in which individuals could have continued the treatment (hypothetical strategy). It is necessary to assume that the available information collected among individuals who completed the IBI is sufficient to recover the symptom course under this hypothetical scenario. The estimate quantifies a scenario in which the somatic condition did not deteriorate. Second, researchers could also try to follow up with individuals despite their deteriorating somatic symptoms and consider deteriorations followed by discontinuation a natural characteristic of the target population (treatment policy strategy). The latter approach assumes that individuals can benefit from the psychotherapeutic techniques they learn when working with IBI, even after they discontinue the treatment due to the worsening somatic condition. However, this becomes problematic when the worsening somatic condition leads to death. Lastly, one may use a while-on-treatment approach, focusing on the symptom while individuals are using the treatment.

Example 3: In a trial investigating an IBI for adults with depression, some individuals will report elevated levels of suicidality, necessitating acute inpatient treatment. First, one could treat hospitalization as part of the treatment regimen and attempt to follow up with all participants (treatment policy strategy). This seems to provide a clinically relevant estimate. However, the resulting estimate is a blend of the IBI and hospitalization. If the hospitalization rates differ between treatment arms, problems can arise in interpreting the effect of the treatment. Alternatively, hospitalization can be defined as a “treatment failure” and used as a component of a composite endpoint (composite strategy). However, the composite strategy assumes that hospitalization has the same clinical relevance as other “treatment failure” indicators merged in the composite.

Example 4: A study compares an IBI to face-to-face psychotherapy (f2f-PT). The depression severity 12 months after randomization is the endpoint. The stakeholder expected that some participants in the IBI group may start psychotherapy between the end of the IBI and the 12-month assessment (ICE). First, researchers could decide to assess all individuals regardless of whether they have started f2f-PT after the IBI (treatment policy strategy). This approach would effectively compare IBI with the possibility of face-to-face psychotherapy if needed, to standard f2f-PT. Second, researchers could discard the assessments of individuals who began f2f-PT after IBI and attempt to model a scenario in which these individuals did not access f2f-PT (hypothetical strategy). Applying the hypothetical strategy can be challenging when no follow-up information is available for individuals without access to f2f-PT. Then, it would be necessary to rely solely on assumptions about symptom development in this hypothetical scenario. If, however, such information is available, one could assume that symptom development is similar to that of those who did not access f2f-PT, although their follow-up measurements suggest that they may have benefited from participating in f2f-PT. The same restriction as in Example 1 applies: if the MAR condition is violated, the estimates will be biased. Third, individuals who entered f2f-PT after IBI could be considered “treatment failures” in a composite measure (composite strategy). Again, the merged indicators of treatment failure must have the same clinical relevance. Lastly, one could focus on the effect among patients who would use the assigned treatment as intended (i.e., either f2f-PT or IBI without subsequent f2f-PT), regardless of their assigned treatment (principal stratum). However, the researchers need to assume that sufficient data is available to recover the latent strata with sufficient accuracy.