Cancer & Oncology Meta-Analysis Guide

Table of Contents

1. Why Meta-Analysis Matters in Oncology
2. Defining Your Oncology Research Question (PICO)
3. Choosing the Right Effect Size for Cancer Endpoints
4. Survival Endpoints: HR Meta-Analysis in Practice
5. Binary Endpoints: ORR, DCR, and Adverse Events
6. Data Extraction Checklist for Oncology Studies
7. Handling Heterogeneity in Cancer Trials
8. Subgroup Analysis by Biomarker and Treatment Line
9. Step-by-Step: Cancer Meta-Analysis in MetaReview
10. Common Pitfalls in Oncology Meta-Analysis

Why Meta-Analysis Matters in Oncology

Cancer research produces thousands of clinical trials each year, but individual trials are often limited by small sample sizes, short follow-up, or narrow patient populations. Meta-analysis addresses these limitations by:

• Increasing statistical power — Pooling data across trials reveals treatment effects that individual studies may miss, especially for uncommon endpoints or subgroups.
• Resolving conflicting results — When trials disagree on whether a treatment works, meta-analysis provides a quantitative synthesis weighted by study quality and precision.
• Enabling subgroup exploration — With sufficient studies, you can examine whether treatment effects differ by biomarker status (PD-L1, EGFR, HER2), tumor stage, or treatment line.
• Informing clinical guidelines — Organizations like NCCN, ESMO, and ASCO rely on meta-analyses to grade evidence and formulate treatment recommendations.

Defining Your Oncology Research Question (PICO)

A well-structured PICO framework is essential for a focused, reproducible oncology meta-analysis.

Search Strategy Tips for Oncology

• Databases: PubMed, Embase, Cochrane CENTRAL, ClinicalTrials.gov (for unpublished results)
• MeSH terms: "Neoplasms"[Mesh], "Antineoplastic Agents"[Mesh], specific cancer terms
• Conference abstracts: ASCO (asco.org), ESMO (esmo.org), AACR (aacr.org) — critical for capturing recent unpublished data
• Trial registries: ClinicalTrials.gov status filter = "Completed" to find unreported studies

Choosing the Right Effect Size for Cancer Endpoints

This is the most critical methodological decision in your cancer meta-analysis. The wrong effect size invalidates the entire analysis.

Survival Endpoints: HR Meta-Analysis in Practice

Why HR Is the Gold Standard for Survival Data

In oncology, survival endpoints (OS, PFS, DFS) are the most important efficacy measures. The Hazard Ratio is the correct effect size because:

• It uses all available follow-up time, not just events at a single time point
• It correctly handles censored patients (those lost to follow-up or alive at data cutoff)
• It is the standard output of Cox regression, the dominant survival analysis method in oncology trials

Three Scenarios for HR Extraction

Interpreting the Pooled HR

• HR = 0.70: Treatment reduces the event hazard by 30% compared to control
• HR = 1.00: No difference between treatment and control
• HR = 1.25: Treatment increases the event hazard by 25% (harmful)
• 95% CI excludes 1.0: The result is statistically significant at p < 0.05

Binary Endpoints: ORR, DCR, and Adverse Events

Many cancer meta-analyses also pool binary outcomes alongside survival endpoints. Common binary endpoints include:

Objective Response Rate (ORR)

Defined as CR + PR per RECIST 1.1 criteria. For each study, extract:

• Events: Number of patients achieving CR or PR in each arm
• Total: Number of evaluable patients in each arm

Use OR when ORR is rare (<20%) or RR when ORR is common (>20%). OR overestimates relative effects when baseline rates are high.

Adverse Event Rates

Grade ≥3 treatment-related adverse events are commonly pooled to compare safety profiles. Extract events/totals from each arm. Report separately for specific AE types (pneumonitis, hepatotoxicity, skin toxicity) when data permits.

Data Extraction Checklist for Oncology Studies

Use a standardized extraction form to ensure consistency. For each included study, record:

Study Characteristics

• First author, publication year, journal
• Study design (RCT, single-arm, retrospective cohort)
• Phase (II, III), registration number (NCT ID)
• Region / Country (single-center vs multicenter, international vs regional)

Patient Population

• Cancer type and histological subtype
• TNM stage at enrollment
• Treatment line (first-line, second-line, ≥third-line)
• Biomarker status (PD-L1 TPS/CPS, EGFR/ALK/ROS1/BRAF mutations, HER2, MSI/dMMR)
• ECOG performance status distribution
• Sample size per arm
• Median age, sex distribution

Treatment Details

• Exact regimen (drug names, doses, schedules)
• Treatment duration / cycles
• Crossover allowed? (critical for OS interpretation)

Outcomes Data

• HR + 95% CI for OS, PFS (or events/totals for binary endpoints)
• Median follow-up duration
• Data maturity (e.g., "mature at 60% OS events")
• Adjusted vs unadjusted HR (document which)

Handling Heterogeneity in Cancer Trials

Heterogeneity is almost inevitable in oncology meta-analyses due to differences in cancer biology, patient selection, and treatment protocols.

Sources of Heterogeneity in Oncology

Quantifying Heterogeneity

• I²: 0% = no heterogeneity, 25% = low, 50% = moderate, 75% = high
• Q-test p-value: p < 0.10 indicates significant heterogeneity (lower threshold because Q-test has low power with few studies)
• τ²: Between-study variance (useful for comparing heterogeneity across analyses)

Strategies When I² Is High

1. Pre-specified subgroup analysis — by tumor type, biomarker, treatment line
2. Meta-regression — test continuous moderators (publication year, median age, sample size)
3. Sensitivity analysis — leave-one-out, restrict to Phase III RCTs only, exclude studies with high risk of bias
4. Narrative synthesis — if heterogeneity remains unexplained and I² > 85%, consider whether pooling is appropriate at all

Subgroup Analysis by Biomarker and Treatment Line

Subgroup analysis is arguably the most clinically relevant component of an oncology meta-analysis. Modern cancer treatment is increasingly biomarker-driven, and pooled subgroup data can inform precision medicine decisions.

Common Pre-Specified Subgroups

Other Important Subgroups

• Treatment line: First-line vs second-line vs later lines (benefit often differs)
• Geographic region: East Asian vs Western (especially for EGFR mutations in NSCLC)
• Performance status: ECOG 0-1 vs ≥2
• Age: <65 vs ≥65 (elderly patients may have different risk-benefit profiles)
• Brain metastases: Present vs absent (affects PFS and treatment selection)

Statistical Considerations

• Use Q-between test (interaction test) to determine whether treatment effects significantly differ across subgroups
• Never treat subgroup HRs from the same study as independent observations — this causes unit-of-analysis error
• Pre-specify all subgroup analyses in your PROSPERO registration to avoid post-hoc data dredging accusations
• With <3 studies per subgroup, results are exploratory only and should not drive clinical conclusions

Step-by-Step: Cancer Meta-Analysis in MetaReview

Step 1: Select Effect Measure

Open MetaReview and choose the appropriate effect measure from the dropdown:

• Hazard Ratio for OS/PFS/DFS analysis
• Odds Ratio or Risk Ratio for ORR/DCR/AE analysis

Step 2: Enter Study Data

For HR analysis: enter Study name, Year, HR, CI Lower, CI Upper.

For OR/RR analysis: enter Study name, Year, Events and Total for both Treatment and Control arms.

Step 3: Assign Subgroups

Use the "Subgroup" column to assign biomarker status, cancer type, or treatment line to each study. This enables stratified analysis.

Step 4: Run the Analysis

Click "Run Meta-Analysis". Results appear within seconds, including:

• Forest plot with individual study effects and pooled estimate
• Subgroup forest plot (if subgroups assigned) with Q-between test
• Funnel plot for publication bias assessment
• Heterogeneity statistics (I², Q, τ²)
• Sensitivity analysis (leave-one-out)

Step 5: Advanced Diagnostics

• Egger's / Begg's test — Quantitative publication bias assessment
• Trim-and-Fill — Estimate the impact of potentially missing studies
• Meta-Regression — Test whether a continuous moderator (publication year, sample size) explains heterogeneity
• Influence diagnostics — Cook's distance, DFFITS, covariance ratio to identify influential studies
• GRADE assessment — Rate the certainty of evidence across 5 domains

Step 6: Export Report

Generate a complete HTML or DOCX report with all figures, tables, auto-generated Methods paragraph (PRISMA 2020 format), and narrative interpretation.

Common Pitfalls in Oncology Meta-Analysis

Pitfall 1: Mixing Effect Sizes

Combining HR (survival) with OR (response) in a single pooled analysis is the most common and most serious error. They measure different aspects of treatment effect and use different statistical frameworks.

Solution: Conduct separate analyses for each endpoint type. Present OS (HR), PFS (HR), and ORR (OR/RR) as distinct results.

Pitfall 2: Ignoring Crossover Effects on OS

Many oncology RCTs allow control-arm patients to receive the experimental treatment upon progression. This blurs the OS difference between arms. An HR for PFS of 0.50 may translate to an OS HR of only 0.85 due to crossover.

Solution: Report both PFS and OS results. Look for crossover-adjusted OS analyses (RPSFT, IPCW methods). Discuss crossover as a limitation.

Pitfall 3: Including Immature Data

Early data cutoffs may show impressive PFS but immature OS. Subsequent data updates may reveal different results. Including both interim and final analyses from the same trial causes double-counting.

Solution: Use the most mature data available for each trial. If multiple publications exist, extract from the latest data cutoff. Never include both interim and final results from the same trial.

Pitfall 4: Overlooking Single-Arm Studies

Phase II single-arm trials report only the experimental arm (no comparator). These cannot be directly included in a comparative meta-analysis.

Solution: Restrict inclusion to comparative studies (RCTs or well-designed cohorts with control arms). Single-arm studies can be separately pooled to estimate single-arm ORR if needed.

Pitfall 5: Biomarker Subgroup Double-Counting

A trial reporting HR for PD-L1 ≥50%, PD-L1 1-49%, and PD-L1 <1% is one study with three subgroups, not three independent studies.

Solution: If your meta-analysis targets the overall effect, use the overall HR. If it targets a specific biomarker subgroup, extract only that subgroup HR.

Pitfall 6: Publication Bias in Positive Oncology Trials

Pharmaceutical-sponsored trials with positive results are more likely to be published quickly and in high-impact journals. Negative trials may be delayed or published as abstracts only.

Solution: Search ClinicalTrials.gov for completed but unreported studies. Include conference abstracts. Use Egger's test and Trim-and-Fill. Discuss potential bias transparently.