{"doi":"10.1002/sim.9492","title":"Explicit underdose control based on toxicity: Four points to consider","abstract":"There are many design options for toxicity-driven phase 1 dose-finding in oncology drug development. There are methods that are simple, traditional and transparent, such as the 3 + 31 or rolling 6 design2 that do not target a dose limiting toxicity (DLT)-rate and are focused on exploring escalating doses subject to limits on patient risk. On the other end of the spectrum are designs that focus on a DLT target-rate and require sophisticated tools throughout the conduct of the study, are less transparent, and require statistical support. Two examples are the CRM design3, 4 and the BLRM design.5 There are also designs between these two extremes. Closest to the 3 + 3 and rolling 6 are queue-based variations of those designs6 or other A + B designs7 that also explore doses subject to pre-specified rules to limit patient risk. Closer to the CRM or BLRM are statistically motivated methods such as the mTPI,8 TEQR,9 and BOIN,10 where there is a target DLT rate and/or range, but simple decision rules and tables reduce the absolute requirement for sophisticated tools or a statistician except in the design and final analysis. Entering into this challenging area is the current effort11 to provide a new twist to the BLRM. The key change is to augment the BLRM with an underdose control rule to be pitted against overdose control. The popularity of this method will likely be limited, as this method is being introduced exactly at the time when the “more is better” paradigm is less accepted.12 Project Optimus, led by the FDA, is specifically concerned with the assumption that the maximum tolerated dose provides the best toxicity-efficacy trade-off, leading to a focus on selecting doses with relatively high toxicity. Instead, low toxicity is not generally an issue to be used to eliminate certain doses for possible consideration for future study. Nevertheless, there are increasingly rare, but important, clinical settings where underdose control concerns persists and this is exactly where such a tool can be considered. Primarily, these are short-term therapies with curative intent. Once restricted to such a setting, is this BLRM with both overdose control and underdose control superior to other designs that seek to find a DLT rate within a pre-specified range? The authors make the reasonable case (see Table 7 of Reference 11) that in some settings, depending on the details of the dose-toxicity curve and the dose increment, this method may have some advantages over some alternative DLT-target rate designs. In other settings, it may not. Knowing when and how to use a new tool is critical for statisticians in the field, and to that end, we will consider 4 separate points. The first point is the concern expressed in Table 1 (see Reference 11), where 0/3 DLTs are observed at each dose level of 10 mg, 25 mg, 50 mg and 100 mg, and yet 200 mg exceeds the overdose control rule so the next patients are not treated at that dose. At first view, this might appear to convey a problem since no escalation is permitted with 0 DLTs on multiple dose levels. However, it is very uncommon in a traditional 3 + 3 design to see dose doubling between a fourth and fifth dose level. The BLRM with overdose control may be pointing to this concern in the design when encountering the 100 mg increment in this case. The authors did not present the full BLRM model behind this hypothetical data scenario, but would a lesser increment, such as a 40 mg increment, representing a 40% increase, have been allowed? This is not to detract from the main role of underdose control, but simply to highlight that the concern about the conservative nature of the BLRM design with overdose control may have been overstated and this may influence the choice of method. The second point to highlight relates to the details on how to use this modified BLRM method. The BLRM method is usually written as a guide while providing certain constraints on decision-making. It presents a permissible range, and a guid","journal":"Statistics in Medicine","year":2022,"id":296090,"datarank":0.0,"base_score":0.0,"endowment":0.0,"self_citation_contribution":0.0,"citation_network_contribution":0.0,"self_endowment_contribution":0.0,"citer_contribution":0.0,"corpus_percentile":null,"corpus_rank":null,"citation_count":1,"citer_count":0,"citers_with_citation_signal":0,"citers_with_endowment":0,"datacite_reuse_total":0,"is_dataset":false,"is_dataset_confidence":0.9535,"is_data_producer":false,"deposit_databanks":null,"is_oa":true,"file_count":0,"downloads":0,"has_version_chain":false,"published_date":"2022-01-01","fair_score":null,"fair_percentile":null,"algorithm_id":"datarank_citation_only_1hop_v6","ranking_scope":"data_only","authors":[{"id":350610,"name":"Elizabeth Garrett‐Mayer","orcid":"0000-0003-4709-0333","position":1,"is_corresponding":false},{"id":413360,"name":"Mark Krailo","orcid":"0000-0002-7608-0698","position":2,"is_corresponding":false},{"id":380347,"name":"Paul Frankel","orcid":"0000-0003-2231-6121","position":0,"is_corresponding":true}],"reference_count":16,"raw_metadata":null,"created_at":"2026-07-19T00:31:12.531553Z","pmid":"36394105","pmcid":null,"fwci":null,"citation_percentile":null,"influential_citations":0,"oa_status":null,"license":null,"views":0,"total_file_size_bytes":0,"version_count":0,"fair_f":null,"fair_a":null,"fair_i":null,"fair_r":null,"fair_zscore":null,"fair_rationale":null,"fair_model":null,"fair_agent_version":null,"fair_fulltext_source":null,"fair_has_llm":null,"fair_computed_at":null,"clinical_trials":[],"software_tools":[],"db_accessions":[],"linked_datasets":[],"topics":[]}