Understanding the role of Tables & Listings in oncology data review
What Tables & Listings are — and what they are notFoundation
Tables (T) are aggregate summaries — they count, group, and compare across patients and time. Listings (L) are patient-level data outputs — every row is one patient or one event. Figures (F) are graphical representations of the same data. Together, TLFs are the analytical backbone of every clinical study report (CSR), DSMB package, and regulatory submission.
The reviewer's job is not to re-derive statistics. It is to read T&L as a clinician — looking for signals, inconsistencies, and patterns that numbers alone do not surface. A table showing 12% grade 3 ALT elevations is a number. A reviewer who understands that this is 3× the protocol-expected rate, that it clusters in the first 8 weeks, and that it correlates with a specific dose level — that is a signal.
The reviewer's job is not to re-derive statistics. It is to read T&L as a clinician — looking for signals, inconsistencies, and patterns that numbers alone do not surface. A table showing 12% grade 3 ALT elevations is a number. A reviewer who understands that this is 3× the protocol-expected rate, that it clusters in the first 8 weeks, and that it correlates with a specific dose level — that is a signal.
The four questions to ask of every tableFramework
1. Is the denominator correct?The denominator defines who is included. "All treated patients" vs "safety population" vs "per-protocol population" — these are different denominators giving different rates. The reviewer must verify which population is being used and whether it is appropriate for the question being answered.
2. Does the number make biological sense?A 0% incidence of nausea in a cisplatin-containing regimen is not favorable — it is biologically implausible. Expected class effects absent from tables signal systematic under-collection, not a clean safety profile.
3. What is the rate relative to expectation?The reviewer compares observed rates against: (a) rates stated in the Investigator's Brochure, (b) rates from the published literature for this drug class, (c) rates in the control arm if present. A rate 2× the expected rate is a signal regardless of the absolute number.
4. What does the listing show behind the number?Every aggregate table should be paired with its listing for any rate that is unexpected. The listing shows the individual patients — when did the event start, how long did it last, what was the outcome, what was done about it. The table asks "how many?" — the listing answers "who, when, and what happened?"
ICH E3 structure — where T&L live in the CSRRegulatory context
Section 12 — EfficacyPrimary and secondary endpoint tables. Response rate tables. Survival analysis tables (KM). Subgroup analyses. Forest plots.
Section 12.4 — SafetyAE summary tables. SAE listings. Deaths. Discontinuations due to AEs. Dose modifications. Laboratory summaries and shift tables. Vital signs. ECG summaries.
Appendix 16 — ListingsPatient-level data for all AEs, labs, ECGs, efficacy assessments, concomitant medications, protocol deviations. These are the source behind every number in the main body tables.
Reviewer priority: Start with the summary tables to identify signals. Then go directly to the listings for every signal. Never accept a table number without understanding what the listing shows behind it.
AE summary table — what to verifyCore safety table
The AE summary table is the first table a reviewer and a regulator looks at. It sets the entire safety narrative. Every number here must be internally consistent with the listings and with the SAE table.
- Any patient with AE (%) — verify denominator is Safety Population (all patients who received at least one dose). A denominator mismatch inflates or deflates all rates.
- Grade 3–4 AE rate — compare to IB-stated expected rate. A rate >2× expected for any preferred term requires investigation and narrative explanation in the CSR.
- AEs leading to dose reduction — must be consistent with the exposure table. If 18% had dose reductions but only 12% are listed here, the discrepancy must be resolved.
- AEs leading to discontinuation — must match the discontinuation table exactly. Any numerical discrepancy between tables in the same CSR is a data integrity flag.
- Class-expected AEs present — verify that known on-target effects appear at plausible rates. Their absence is a signal of under-reporting, not favorable safety.
- Treatment-arm comparison — if controlled trial, compare rates between arms. Statistically similar rates for a class-specific AE (e.g., rash in EGFR arm vs control) is implausible.
- MedDRA coding consistency — same clinical event may be coded differently across sites (e.g., "nausea" vs "abdominal discomfort"). Verify that coding is consolidated at HLT/PT level for aggregate accuracy.
Reading the AE listing behind the tableGo to listing when:
Any grade 3–4 eventPull the listing for every grade 3–4 event. Verify: start/stop dates clinically plausible, action taken documented (dose mod, hospitalization, concomitant medication), outcome confirmed (resolved, ongoing, sequelae).
Rate higher than expectedPull listing for that PT. Check clustering: do events concentrate in a specific cycle, dose level, site, or patient subgroup? Clustering is a signal that aggregate rates obscure.
Causality "not related" >50%If the majority of grade 3–4 AEs are investigator-assessed as unrelated to study drug, pull the listing. Is this clinically defensible given the mechanism of the drug? Or is this systematic under-attribution?
Resolution rate <80%Unresolved AEs at study end are clinically significant. Pull listing to identify which patients have ongoing AEs, what the event is, and whether the protocol's follow-up requirements were met.
SAE table and deaths listing — critical reviewHighest scrutiny
Every SAE and every death must have a complete narrative in the listing. The reviewer reads these as a clinician, not as a data checker.
- SAE count in table matches SAE count in listing — a numerical discrepancy between the summary table and the individual listing is never acceptable and requires resolution before submission.
- Every death has a complete narrative — last dose date, symptom onset, hospitalization date, cause of death, attribution. "Progressive disease" as cause of death requires supporting evidence (imaging, clinical deterioration timeline).
- Deaths within 30 days of last dose — all must be reviewed regardless of attributed cause. A death attributed to "disease progression" in a patient with concurrent grade 4 hepatotoxicity requires investigation.
- Unexpected SAEs — any SAE not listed in the IB as a known risk requires a detailed narrative and assessment of expectedness. These drive SUSAR obligations.
- SAE reporting timeliness — verify that the date of SAE occurrence vs date of sponsor receipt vs date of regulatory report filing is within ICH E2A timelines (7 days fatal, 15 days other serious unexpected).
- Causality distribution — if >80% of SAEs are assessed as "not related," this is a systematic pattern requiring medical review. Compare investigator causality vs sponsor causality assessments.
Deaths listing is never a checkbox exercise. Read every death narrative as if you are the FDA reviewer. If the narrative does not tell a coherent, clinically credible story — it will be questioned at review. Fix it before submission.
Discontinuation table — reasons and consistencyRetention signal
Reasons for discontinuationStandard categories: disease progression, AE, patient withdrawal, investigator decision, death, protocol deviation, lost to follow-up. Every patient must be assigned exactly one primary reason. Mixed or ambiguous reasons require resolution.
AE-driven discontinuation rateCompare to IB and comparable agents in the class. A rate significantly higher than historical data for the drug class is a safety signal. In IO trials, verify whether AE-driven discontinuations were appropriately managed before withdrawal — some reflect inadequate irAE management rather than intolerable toxicity.
Discontinuation before first response assessmentPatients who discontinue before the first tumor assessment are excluded from the efficacy-evaluable population. A high rate of early discontinuations affects the denominator of ORR and must be discussed in the benefit-risk section.
Lost to follow-upAny patient lost to follow-up must have documentation of the last confirmed contact date. "Lost to follow-up" without a confirmed last contact date is a GCP compliance issue, not just a data gap.
Dose modification table — exposure integrityDose-safety link
Dose reduction rateCompare observed reduction rate vs protocol-anticipated rate stated in the SAP. A much higher rate signals cumulative toxicity; a lower rate in a highly toxic drug may signal under-documentation of the clinical justification.
Reason for modificationEvery dose modification must be linked to a documented clinical reason (AE, lab finding, investigator decision). Modifications with "other" or missing reasons require source data query. These drive the narrative of the exposure section in the CSR.
Relative dose intensity (RDI)Mean and median RDI must be reported. RDI <85% across the cohort means patients received significantly less drug than planned — this affects both safety interpretation (AEs at reduced dose) and efficacy interpretation (sub-therapeutic exposure). The CSR must address this explicitly.
Dose at time of key AEsCross-reference: were severe AEs occurring at full dose or at reduced dose? AEs at reduced dose are a stronger safety signal than the same AE at full dose.
The dose modification table is the bridge between safety and efficacy. It answers whether patients received enough drug to be evaluable for both endpoints. A study where 40% of patients had significant dose reductions requires an explicit sub-group analysis by RDI in both the safety and efficacy sections.
Objective response rate (ORR) table — what to verifyPrimary efficacy
Evaluable population denominatorORR is calculated over the efficacy-evaluable population (patients with at least one post-baseline tumor assessment). The reviewer must verify: how many patients were enrolled, how many received treatment, how many had at least one assessment, and why the rest did not. A large gap between treated and evaluable patients requires explanation.
Confirmed vs unconfirmed responsesMost protocols require confirmed response (two consecutive assessments ≥4 weeks apart meeting CR or PR criteria). The table must specify confirmed response rate separately. Unconfirmed responses in the primary ORR table will be challenged by FDA.
Investigator vs IRC assessmentIf an independent review committee (IRC) exists, both investigator and IRC ORR must be reported. A large discordance (>10–15% absolute difference) between investigator and IRC ORR requires narrative explanation. IRC is the primary efficacy endpoint in most registration-intent trials.
Best overall response listingPull the best overall response listing (one row per patient). Verify: every CR and PR patient has a response confirmation date ≥4 weeks from first response. Any patient with PR showing <4 weeks to confirmation is non-confirmatory and must be recategorized as SD or unconfirmed.
PFS and OS tables — Kaplan-Meier reviewSurvival endpoints
Median PFS/OS with 95% CIThe 95% CI width reflects the precision of the estimate. A very wide CI (e.g., PFS median 8.2 months, 95% CI 4.1–NE) means the estimate is imprecise — likely because too few events occurred or censoring is heavy. This must be disclosed and contextualized in the CSR.
Censoring rules verificationThe listing must show for each patient: last tumor assessment date, progression/death date (if applicable), and censoring date. Verify that censoring rules in the SAP were applied consistently: patients who missed assessments, patients who started new anticancer therapy, patients lost to follow-up.
PFS assessment window complianceCross-reference the PFS listing with the tumor assessment dates. Assessments done outside the protocol-specified window affect PFS calculation. Flag any site where >20% of assessments are outside window — this is an operational quality signal.
OS follow-up completenessOS requires survival status for every patient. Patients with last contact >6 months before data cut and listed as "alive" are a data currency problem. The listing must show the date of last confirmed contact for every censored observation.
Subgroup analyses — forest plots and tablesExploratory
Pre-specified vs post-hocEvery subgroup analysis must be clearly labeled as pre-specified (in the SAP) or post-hoc (exploratory). Post-hoc subgroups have inflated Type I error and cannot drive regulatory conclusions. FDA will question any unplanned subgroup presented as if it were confirmatory.
Small subgroup sample sizesSubgroups with n <15–20 patients produce estimates with very wide confidence intervals. The point estimate is numerically unreliable. These subgroups should be presented with explicit caveats and should not drive labeling claims without additional confirmatory data.
Biomarker subgroups in IOPD-L1 high vs low, TMB high vs low, MSI-H vs MSS — these are the subgroups with the highest regulatory significance in IO. Verify that biomarker status was collected prospectively, that the assay was pre-specified, and that the cut-off was defined before unblinding. Post-hoc cut-off optimization is not acceptable.
Forest plot consistencyThe subgroup ORR or HR values in the forest plot must match the corresponding rows in the subgroup table exactly. Discrepancies between figures and tables in the same submission are a quality signal that triggers FDA queries.
Subgroup analyses are exploratory by default. A favorable subgroup does not rescue an unfavorable primary endpoint. An unfavorable subgroup does not override a favorable primary endpoint. Both must be presented transparently and interpreted within the pre-specified statistical analysis framework.
IO-specific efficacy table considerationsIO trials
iRECIST response tableMust report iUPD rate separately from iCPD rate. Patients with iUPD who subsequently achieved response must be identifiable in the listing — these are pseudoprogressors. The table should show: iCR, iPR, iSD, iUPD, iCPD rates, and the proportion of iUPD patients who converted to response vs confirmed progression.
Duration of response (DoR)In IO trials, DoR is frequently longer than in chemotherapy — this is a key efficacy differentiator. The DoR table must report: number of responders, median DoR with 95% CI, and proportion of responders with DoR >6, >12, >18 months. The listing must confirm response start and end dates for every responder.
Landmark survival rates12-month, 18-month, and 24-month OS and PFS rates are as important as median OS in IO. IO survival curves frequently show a "plateau" — a subset of long-term survivors. The table must present these landmark rates with 95% CIs. Regulators use landmark rates to assess the tail of the curve, which median OS does not capture.
Corticosteroid subgroup analysisEfficacy in patients who received systemic corticosteroids (for irAE management) vs those who did not is a key IO subgroup. If steroids reduce efficacy, this has labeling implications. This analysis must be pre-specified in the SAP if it is to carry regulatory weight.
Lab shift tables — how to read themCore lab summary
A shift table shows how many patients moved from one toxicity grade to another (e.g., from Grade 0 at baseline to Grade 2 on-treatment). The cells that matter most are the upper-right quadrant — patients who worsened. The reviewer must read shifts, not just worst-grade summaries.
Baseline to worst-on-treatmentThe primary shift table for each lab parameter. Patients who went from Grade 0 → Grade 3 are the most clinically significant. These must be individually reviewed in the listing — what were the clinical consequences? Was the event managed? Did it resolve?
Worst CTCAE grade summaryComplement to shift table. Shows the maximum grade each patient reached for each parameter. Compare Grade 3–4 rates to IB-expected rates. Any parameter with Grade 3–4 rate >5% in a non-hematologic parameter warrants narrative discussion.
Clinically notable valuesMany protocols define "clinically notable" thresholds separate from CTCAE grades (e.g., ALT >3× ULN regardless of grade, QTcF >500ms). These must appear in the shift table and be consistent with the Hy's Law analysis and the ECG summary table.
Shift table reading rule: The diagonal (same grade baseline → same grade on-treatment) is reassuring. Movement toward the upper-right (worsening) is the signal. Movement toward the lower-left (improvement) in already-abnormal patients suggests the drug may have beneficial off-target effects — also worth noting.
Hy's Law listing — mandatory for every hepatic safety reviewFDA expects this
The Hy's Law listing is a patient-level output that identifies every patient who met or approached Hy's Law criteria. FDA expects this listing in every NDA/BLA submission involving a drug with hepatic metabolism or known hepatic signal.
- Listing contains all patients with peak ALT or AST ≥3× ULN — regardless of whether bilirubin was elevated. This is the pool from which Hy's Law cases are identified.
- For each patient in the pool: concurrent peak total bilirubin value and date. Were the peaks concurrent (within 30 days of each other)?
- R-value calculated and documented: (ALT/ULN) ÷ (ALP/ULN). Hepatocellular pattern (R>5) carries higher DILI risk than cholestatic pattern (R<2).
- Criterion 3 exclusion documented for every patient with ALT ≥3× ULN + bilirubin ≥2× ULN: viral serology, imaging, concomitant hepatotoxins, metastatic involvement of liver.
- Hy's Law scatter plot present: peak ALT/ULN (x-axis) vs peak bilirubin/ULN (y-axis) for all patients. Reference lines at 3× (ALT) and 2× (bili) define the four quadrants. Every point in upper-right quadrant must be individually narratived.
- Temple's Corollary documented: total number and proportion of patients with ALT ≥3× ULN (the denominator that predicts DILI risk at the population level).
Hematology lab listing — key parametersHematologic safety
ANC nadir listingOne row per patient showing baseline ANC, nadir ANC value, nadir date (cycle and day), ANC at recovery, and duration of grade 3–4 neutropenia. The pattern across patients (nadir timing, depth, recovery duration) defines whether toxicity is expected class effect or signal.
Febrile neutropenia listingMust identify: all patients with ANC <1000/mm³ who had concurrent fever ≥38.3°C. Confirm each case was coded as febrile neutropenia in the AE domain. Cross-reference with hospitalization data. Missing febrile neutropenia AE entries for patients who clearly met criteria = data quality failure.
Platelet nadir and transfusionsPlatelet listing should include whether transfusion was required. Transfusion requirement is a clinical outcome, not just a lab value — it must appear in the AE domain as a serious event if it meets seriousness criteria.
Hemoglobin — relative change from baselineThe listing should show baseline Hgb and on-treatment nadir per patient. A patient-level view of relative change (not just absolute values) captures clinically meaningful anemia in patients with already-low baseline Hgb.
Chemistry lab listing — oncology-specific parametersChemistry safety
Creatinine trend listingShow creatinine values by cycle per patient. A pattern of gradual increase (0.2 mg/dL per cycle) that never crosses CTCAE Grade 2 threshold but shows consistent upward trend across 6 cycles is clinically significant and should be flagged in the narrative.
Electrolytes (K, Mg, Phos)Hypomagnesemia listing is mandatory in EGFR inhibitor trials. Hypokalemia listing relevant in any trial with QTc signal. Hypophosphatemia listing required in FGFR inhibitor trials. These are rarely highlighted in summary tables but carry clinical significance.
Glucose in IO trialsNew-onset hyperglycemia >250 mg/dL in a patient with no prior diabetes history must be individually reviewed — this is immune-mediated Type 1 diabetes until proven otherwise. The listing must show baseline glucose, HbA1c if collected, and the trajectory of glucose elevation.
Thyroid function in IOTSH and free T4 per patient per cycle. The listing must show baseline, nadir/peak values, and whether thyroid replacement therapy was initiated. TSH elevation without documented management is an irAE management gap, not just a lab finding.
Exposure summary table — the foundation of dose-response analysisExposure integrity
The exposure table is the bridge between the protocol dose and what patients actually received. Without understanding exposure, neither safety nor efficacy data can be correctly interpreted.
Duration of treatmentMean, median, and distribution of treatment duration. Number of patients treated <1 cycle, 1–3 cycles, 3–6 cycles, >6 cycles. Patients who received fewer than 2 cycles are rarely evaluable for efficacy — their proportion affects the efficacy-evaluable denominator.
Number of cycles / dosesMean and median cycles received vs planned. Compare to the protocol-specified number of cycles. A mean of 4.2 cycles in a study designed for 6 cycles means the average patient received 70% of planned treatment — clinically and analytically significant.
Relative dose intensity (RDI)Mean and median RDI. Distribution of patients by RDI category (<75%, 75–85%, 85–95%, >95%). If RDI <85% in >30% of patients, the study may not be testing the drug at the intended therapeutic dose. This requires explicit discussion in the benefit-risk section.
Dose modifications frequencyProportion of patients with at least one reduction, at least one delay, and at least one interruption. These are separate categories — a patient may have all three. Cross-reference with the safety table to confirm all modifications are linked to documented clinical reasons.
Exposure listing — patient-level dose historyPer patient
Per-cycle dose listingShows planned dose, administered dose, and percent of planned dose for every cycle of every patient. The reviewer cross-references this with the AE listing to verify: was a dose modification documented at the same time as a grade ≥2 AE? Is the modification clinically justified by the concurrent toxicity?
Dose at time of responseIn targeted therapy and IO trials, dose at time of first response is informative — did responders achieve response at full dose or at reduced dose? This informs the dose-response relationship and has implications for dose optimization in future trials.
Dose at time of progressionWas the patient on full dose, reduced dose, or off drug at the time of progression? Progression on reduced dose is a different scenario than progression on full dose — it may reflect sub-therapeutic exposure rather than true drug resistance.
Exposure analysis rule: Safety rates calculated over patients at full dose vs patients at reduced dose may be very different. Any aggregate safety table that does not stratify by dose level or RDI category is hiding a potential dose-toxicity signal.
Demographics and baseline characteristics tableStudy population
The demographics table defines who was studied. It is the foundation for assessing generalizability and for interpreting all downstream efficacy and safety data.
ECOG PS distributionThe single most important baseline characteristic in oncology. Verify that the distribution matches the protocol eligibility criteria. If the protocol allowed ECOG 0–2 but 95% of enrolled patients were ECOG 0–1, the study may not be representative of the real-world population the drug will be used in — a generalizability gap.
Prior lines of therapyNumber of prior treatment lines and specific prior therapies (especially prior IO in IO trials). Verify consistency with eligibility criteria. In a study requiring ≥1 prior line, a patient with 0 prior lines in the listing is an eligibility deviation. Prior lines also define the treatment context for interpreting response rates — 1st line vs 3rd line ORR are not comparable.
Tumor histology and molecular markersFor targeted therapies, verify that all patients have documented biomarker status (EGFR mutation, ALK rearrangement, PD-L1 score, MSI, TMB) that meets eligibility criteria. Missing biomarker data at baseline in a biomarker-selected study is a significant protocol deviation.
Hepatic and renal function at baselineProportion of patients with mild, moderate, severe hepatic impairment and renal impairment. This defines the safety population context and determines whether specific organ-function subgroup analyses are needed (and pre-specified in the SAP).
Baseline lab valuesMean and median baseline ALT, AST, creatinine, ANC, platelets. Any patient with already-abnormal baseline values defines a higher-risk subgroup. The listing of baseline labs must be reviewed to confirm eligibility for patients with borderline values.
Treatment arm balance (RCT)In randomized trials, verify that key prognostic factors are balanced between arms. Statistical imbalance in ECOG PS, number of prior lines, or biomarker status can confound efficacy comparisons and must be addressed in the statistical analysis (stratified analysis, sensitivity analysis).
Demographic listing — eligibility verificationProtocol compliance
- Every patient meets all inclusion criteria — verify screening vs baseline values for age, ECOG, organ function labs. Document any patient who had borderline values and how eligibility was confirmed.
- No patient violated any exclusion criterion — cross-reference concomitant medication listing against prohibited medications in the exclusion criteria. Check for prior IO in trials excluding prior checkpoint inhibitor use.
- Informed consent date precedes first dose date for every patient — a consent date after the first dose date is a GCP violation, not a data entry error.
- Stratification factors at randomization (RCTs) — verify that the stratification factors recorded in the randomization system match the baseline characteristics table. Discordance between stratification strata and actual patient characteristics affects the validity of stratified analyses.
Concomitant medication summary tableContext for safety
The concomitant medication table is the most under-reviewed table in most oncology data packages. It is the essential context for interpreting every AE causality assessment.
Most frequent concomitantsThe top 10–20 most frequent concomitant medications by ATC class. In an oncology trial, this typically includes antiemetics, G-CSF, analgesics, antifungals, antibiotics, and corticosteroids. The reviewer verifies that the frequency of each class is consistent with the AE profile — high antiemetic use should correspond to reported nausea rates.
QTc-prolonging agentsIn trials with a QTc signal, the CM table must show the proportion of patients receiving concurrent QTc-prolonging medications. This is required to contextualize the QTc findings — a high rate of QTc prolongation in a population with 60% concurrent fluoroquinolone use is a different risk signal than the same rate without concurrent prolongers.
Hepatotoxic agentsIn trials with hepatic signals or Hy's Law analysis, the CM table must show the proportion of patients receiving concurrent hepatotoxic medications (acetaminophen, amoxicillin-clavulanate, statins, azoles). This is part of the Criterion 3 exclusion evidence at the population level.
Corticosteroids in IOSystemic corticosteroid use must be tabulated by: indication (irAE vs prophylaxis vs other), dose (prednisone equivalent), and duration. The proportion of patients receiving corticosteroids for irAE management defines the clinical management burden and is a key safety metric for IO trials.
Concomitant medication listing — per-patient reviewPer patient
Indication documentedEvery concomitant medication entry must have a documented indication. "Unknown" or blank indication is a data quality failure. The indication is the primary tool for identifying unreported AEs (drug started for a condition that should have been an AE entry).
Prohibited medicationsCross-reference every patient's concomitant list against protocol exclusion criteria and prohibited medication list. Any prohibited medication present from the start of the study (not just at screening) is a protocol deviation that must be reported and may affect the patient's eligibility for the per-protocol analysis set.
Start date relative to AEsFor patients with significant AEs, review the concomitant medication start dates in the 4 weeks prior to the AE. Any hepatotoxic, nephrotoxic, or QTc-prolonging medication started in this window is a potential contributing factor to the AE — it must be documented in the individual patient narrative.
Critical red flags across all T&L domains
!
Numerical discrepancies between tables in the same report
The AE summary table shows 23 patients with SAEs. The SAE listing has 21 rows. The discontinuation table shows 19 patients discontinued due to AEs. These three numbers must be reconcilable. Any discrepancy between tables in the same CSR or submission package triggers FDA queries and signals data integrity problems.
!
Absence of biologically expected findings
A 0% rate of any class-expected AE is not a favorable finding — it is a data collection failure. An EGFR inhibitor trial with 0% rash, a cisplatin regimen with 0% nausea, an anti-VEGF study with 0% hypertension. The reviewer must flag these and require explanation before any submission.
!
Rates significantly higher than IB or published literature
Any AE rate >2× the IB-stated expected rate requires a narrative explanation in the CSR. Any SAE rate >2× the published literature rate for the drug class is a signal that must be contextualized. These are not administrative findings — they are the primary signal detection triggers for the medical team.
!
High proportion of "other" or missing reasons
Dose modifications with "other" reason >15% of modifications. Discontinuations with "other" reason >10%. Missing or ambiguous reasons in any category require source data queries before finalization. Regulators will not accept "other" as a dominant category in any key table.
!
Wide confidence intervals in primary efficacy estimates
A very wide 95% CI on the primary ORR or median PFS means the estimate is imprecise and may not be reliable for regulatory decision-making. This occurs when sample size is too small, too few events occurred, or censoring is too heavy. It must be explicitly addressed in the benefit-risk discussion.
!
Systematic site-level differences in AE or efficacy rates
If Site A reports 5% grade 3 AEs and Site B (same protocol, same drug) reports 25% grade 3 AEs — this is not patient variability, it is a site-level reporting problem. Similarly, if a single site contributes 40% of all responders, the overall ORR is driven by that site's data quality. Site-level data review is mandatory before any data lock.
✓
The T&L review is complete when:
Every number in every summary table is traceable to the corresponding listing. Every red flag has been resolved or explained in writing. The clinical narrative and the statistical data tell the same story. No unexplained discrepancies exist between any two tables in the package.