Peritoneal carcinomatosis does not form discrete measurable nodular lesions — it presents as diffuse implants, ascites, and omental caking. RECIST cannot measure omental caking reliably (no single bidimensional lesion) and ascites is not a RECIST-measurable lesion. Standard approach: use the Peritoneal Cancer Index (PCI) for surgical trials; for systemic therapy trials, use CT-based morphologic assessment of peritoneal implants as "non-target" lesions with overall response classified as CR/PR/SD/PD based on clinical synthesis. The protocol must specify the assessment method for peritoneal disease explicitly — if it defaults to RECIST with peritoneal implants as non-target, the progression call is clinician-dependent and highly variable.

Bone metastases are classified as non-measurable in RECIST unless they have a soft tissue component ≥10 mm. Pure lytic or blastic bone disease cannot be "measured" under RECIST 1.1. This creates a practical problem: a patient with bone-only metastatic castration-resistant prostate cancer (mCRPC) has no RECIST-measurable lesions, but bone scan and PSA are the relevant response markers. Trials in bone-predominant disease must specify bone-specific endpoints: bone scan response (PCWG3 criteria), PSA response, circulating tumor cell (CTC) counts, and — increasingly — PSMA PET criteria. Entering a bone-only patient in a RECIST-driven trial without a bone-specific response assessment plan creates an unevaluable patient for the primary endpoint.

Malignant pleural mesothelioma grows along pleural surfaces, not as a discrete nodule. RECIST cannot be applied. The modified RECIST for mesothelioma (mRECIST) measures tumor thickness perpendicular to the chest wall or mediastinum at defined anatomical levels on CT. This is a disease-specific modification that must be pre-specified in the protocol. A Medical Monitor reviewing imaging CRFs for a mesothelioma trial who sees standard RECIST measurements should flag this as a protocol deviation — the assessment methodology is wrong.

Several tumor patterns resist RECIST measurement: diffuse leptomeningeal disease (no discrete lesion); liver infiltration without dominant mass; lymphangitic carcinomatosis; and some T-cell lymphomas with diffuse organ infiltration. In these settings, response must be defined using functional markers (PET, disease-specific biomarkers), clinical criteria (symptom scales, organ function markers), or disease-specific criteria. Protocols that default to "RECIST per investigator" in these populations generate unmeasurable patients and uninterpretable primary endpoints. The time to identify this is at protocol review, not at data lock.

HBV reactivation is a potentially fatal complication of immunosuppressive therapy, including anti-CD20 agents, steroids, and to a lesser extent cytotoxic chemotherapy. All patients should be screened for HBsAg, anti-HBc, and anti-HBs before enrollment in trials involving these agents. HBsAg-positive patients require antiviral prophylaxis (tenofovir or entecavir) and hepatology co-management — they are not automatically excluded from enrollment if the protocol permits prophylaxis. Anti-HBc-positive / HBsAg-negative patients (occult HBV) have lower but real reactivation risk and also merit prophylaxis consideration depending on the immunosuppression intensity. The Medical Monitor must confirm the protocol's HBV screening and management plan is operationally feasible at all sites.

IO agents (PD-1/PD-L1 inhibitors, CTLA-4 inhibitors) can reactivate latent TB by augmenting immune responses in ways that are poorly controlled. All patients enrolling in IO trials in endemic regions should have QuantiFERON-TB Gold testing (preferred over TST in immunocompromised patients). QuantiFERON-positive patients are generally excluded from IO trials unless they have completed adequate latent TB treatment (isoniazid for 9 months or equivalent). Active TB is an absolute exclusion from IO trials. The Medical Monitor must ensure that sites in TB-endemic regions have a local TB specialist available for positive QuantiFERON workup and that the timeline for TB clearance doesn't create unacceptable enrollment delays.

Every eligibility question from a site should be evaluated with three questions asked in sequence: (1) Does this criterion address a safety risk for the patient — for example, organ function thresholds that determine whether the drug can be safely administered? (2) Does it address a data integrity concern — for example, washout periods that determine whether response to the study drug can be attributed correctly? (3) Does it address a regulatory population definition — for example, biomarker status or prior treatment requirements that define the label-relevant population?

The answer determines the risk of granting the waiver. Safety criteria: waiver risks patient harm. Data integrity: waiver risks uninterpretable endpoints. Regulatory population: waiver risks including patients outside the intended label indication — a data set that cannot support the submission.

1. Borderline lab values: ANC 1450/mm³ when criterion is ≥1500. Platelets 99,000 when criterion is ≥100,000. Creatinine clearance 48 mL/min when criterion is ≥50.

2. Treatment-free interval boundaries: Last chemotherapy dose 27 days prior to planned C1D1 when washout is 28 days. Last radiation 13 days prior when washout is 14 days.

3. Asymptomatic prior conditions: Brain metastases treated 6 months ago, asymptomatic, stable on MRI. HBV serology positive but HBV DNA undetectable. Prior malignancy treated 4.5 years ago when criterion requires 5 years disease-free.

4. ECOG reclassification requests: Patient assessed by site as ECOG 2 at screening now assessed ECOG 1 after a good week. The screening assessment was the protocol-defined baseline.

5. Historical biomarker data: HER2 IHC 3+ from a biopsy 18 months ago in metastatic breast cancer when protocol requires recent confirmation. PD-L1 CPS 1 from archival tissue when protocol requires a fresh biopsy.

Every granted waiver is a protocol deviation. It appears in the CSR deviation listing. It may be reviewed by an FDA inspector. It must be justified medically and scientifically. A pattern of granted waivers is an inspection finding — even if each individual waiver seemed clinically reasonable.

Waivers compound across the trial. One waived patient with ANC 1450 seems minor. Twenty waived patients across twelve sites creates a systematic shift in the safety population — the observed neutropenia rate will no longer be comparable to the IB-stated expected rate, and Grade 3+ neutropenia adjudication becomes ambiguous because baseline values were closer to the threshold than the protocol intended.

Waivers create precedent. Sites share information. Once an MM grants a waiver for ANC 1450 at one site, similar requests from other sites become difficult to refuse. The "narrow exception" becomes the de facto criterion. The final enrolled population no longer matches the protocol.

There are three circumstances where an MM may reasonably consider granting a waiver — all of which require documentation that would withstand regulatory inspection:

1. Transcription/timing errors: The value meets criterion; documentation error caused the discrepancy. A repeat lab the same day shows ANC 1600 — the 1450 was an analytical artifact. This is not a waiver; it is a correction.

2. Minor numerical boundary with documented medical equivalence: ANC 1480 vs. required 1500 for a trial where the safety population is heavily pretreated patients and the analytical window of the criterion is conservative. Requires MM clinical assessment AND documentation.

3. Criterion ambiguity requiring interpretation: The protocol says "no prior checkpoint inhibitor" but a patient received a CTLA-4/PD-1 bispecific in a Phase 1 trial 3 years ago. Interpretation of whether this meets the exclusion requires MM judgment.

Program-average screen failure rate establishes a baseline expectation. A site with a screen failure rate <50% of the program average has one of three explanations: (a) exceptional patient selection — unusual in oncology where eligibility is complex, (b) incomplete screening documentation — patients who should have been screen failures not documented as such, or (c) eligibility drift — patients enrolled who did not meet criteria.

A site with a screen failure rate >200% of the program average may reflect: (a) rigid interpretation that could be refined through MM guidance, (b) poor pre-screening by referring oncologists generating too many ineligible referrals, or (c) a population mismatch with the trial indication.

Central monitoring of baseline lab values can identify clustering at eligibility thresholds. If Site 14 shows 78% of enrolled patients with baseline ECOG PS exactly 1 (the upper eligibility limit) versus a program average of 45%, the signal is that the site may be systematically rounding ECOG 2 patients to ECOG 1 to maintain enrollment rate. A similar pattern in creatinine clearance, ANC, or other continuous eligibility variables indicates data integrity risk.

A site enrolling at 3× the program rate is not inherently a problem — high-performing sites exist. But velocity anomalies combined with other signals (low screen failure rate, borderline value clustering, high protocol deviation rate, underreported SAEs) create a composite pattern that warrants targeted monitoring. The converse — a site with near-zero enrollment — may indicate operational problems at the site that MM intervention can address.

An MM who tracks waiver requests by site identifies which sites are either misinterpreting criteria (suggesting additional training is needed) or systematically pushing the boundaries of eligibility (suggesting audit is needed). A site submitting 15 waiver requests when the program average is 2 per site is a pattern — regardless of whether individual waivers are reasonable. The pattern itself is the signal.

The adverse event must occur during the pre-specified DLT assessment period. Typically Cycle 1 (21-28 days, depending on the dosing schedule). An event that occurs in Cycle 3 of a patient who has been on drug for 8 weeks is an adverse event — but it is not a DLT by the standard criterion, even if Grade 4.

Critical nuance: Some protocols extend the DLT window for immune-mediated events in IO trials — because irAEs can appear weeks to months after initial dosing. IO-specific DLT windows may be 42 days or longer.

Standard (non-IO) DLT grade thresholds:

• Grade 4 hematologic toxicity (except Grade 4 neutropenia <7 days)
• Grade 3-4 non-hematologic toxicity (excluding expected, manageable, brief events)
• Grade 3 febrile neutropenia

IO-specific thresholds (lower than standard):

• Grade 2 pneumonitis = DLT
• Grade 2 myocarditis = DLT
• Grade 2 neurological irAE = DLT

IO thresholds are lower because Grade 2 irAEs can escalate rapidly to fatal Grade 5.

The event must be at least possibly related to the investigational drug. An unrelated Grade 4 event (e.g., trauma, intercurrent infection unrelated to trial drug) that occurs during the DLT window and meets the grade threshold is not a DLT if it is clearly unrelated to the study drug.

Attribution standard: "At least possibly related" means the causality assessment is anything other than "unrelated." The bar for excluding a Grade 4 event from DLT consideration on grounds of unrelatedness is high — requires specific documented alternative cause.

• Statistical inefficiency: Enroll 3, assess, decide, potentially enroll 3 more. Decision at each cohort is based on only 3 or 6 patients — large statistical uncertainty with small N.
• Binary decision rule: 0/3 DLTs = escalate. 1/3 DLTs = expand to 6. ≥2/3 DLTs = de-escalate. This binary rule does not incorporate all available prior data — each cohort decision is made as if previous cohorts don't exist.
• MTD-only output: 3+3 identifies the MTD but provides limited information about the shape of the dose-toxicity relationship. No model of the curve — just a point estimate.
• Bias toward lower doses: 3+3 tends to identify doses below the true MTD because the decision rule is conservative at each step. FDA Project Optimus flagged this as a contributor to suboptimal dose selection.

• Accumulates data across all cohorts: Each escalation/de-escalation decision incorporates the entire trial experience to date — not just the current cohort. More data = better decisions.
• Lambda thresholds: λ_e (escalation boundary) and λ_d (de-escalation boundary) define optimal intervals. If observed DLT rate falls below λ_e → escalate. Above λ_d → de-escalate. Between λ_e and λ_d → stay.
• Typical thresholds (target DLT rate 0.3): λ_e ≈ 0.236, λ_d ≈ 0.359. These are pre-computed from the target DLT rate and optimized to minimize the probability of incorrectly selecting a dose.
• FDA Aug 2024 preferred: FDA now explicitly prefers model-assisted or model-based designs over 3+3 for registrational Phase 1 programs. BOIN is the most widely adopted of these designs.
• RP2D output: BOIN supports selection of RP2D (not just MTD) by allowing the dose-toxicity curve to inform dose selection below the MTD where safety and activity may be optimized — directly aligned with Project Optimus philosophy.