Prospective Approaches (Cochrane Handbook Ch 22)

Your Role

Guide prospective meta-analysis (PMA), trial sequential analysis (TSA), cumulative meta-analysis, and living review methods for reviews where evidence is still accumulating.

---

Prerequisites

• Completed or ongoing meta-analysis
• Information about ongoing/planned trials
• Original search dates (for cumulative MA)

---

Workflow

Step 1: Determine Which Approach Fits

| Scenario | Approach | |----------|----------| | Multiple ongoing/planned trials with compatible PICO and data-sharing commitment | PMA | | Completed review — is current evidence conclusive? | TSA | | Completed review — how did evidence accumulate over time? | Cumulative MA | | Published review — ongoing evidence surveillance | Living review |

Step 2: Prospective Meta-Analysis (PMA)

A PMA identifies trials BEFORE results are known and commits to combining them.

Protocol requirements:

• PICO and eligibility fixed before individual trial results are known
• Data-sharing agreement signed by all trial investigators
• Statistical analysis plan pre-specified
• Governance: steering committee, data monitoring, publication policy

GENERATE: sr-pma-protocol.md

Step 3: Trial Sequential Analysis (TSA)

TSA computes the required information size (RIS) and monitoring boundaries to determine if current evidence is conclusive.

RUN:

Rscript scripts/tsa_analysis.R --data ma.csv --event-rate 0.10 --relative-risk-reduction 0.25 --alpha 0.05 --power 0.80

OUTPUT:

Required Information Size (RIS): X participants
Current total: N participants (X% of RIS)
Z-curve: [crossed / not crossed] the monitoring boundary
Futility boundary: [crossed / not crossed]
Conclusion: Evidence is [conclusive / inconclusive / futile]

INTERPRET:

• Z-curve crosses monitoring boundary → sufficient evidence, stop
• Z-curve crosses futility boundary → unlikely to find effect even with more data
• Z-curve within boundaries → insufficient evidence, more trials needed

Step 4: Cumulative Meta-Analysis

GENERATE cumulative forest plot (studies added chronologically):

Rscript scripts/cumulative_ma.R --data ma.csv --date-column year

Plot shows how the pooled estimate evolved as each new study was added.

INTERPRET:

• Did the estimate stabilize over time?
• Was there an early positive result that was later attenuated?
• When did the result become statistically significant (if ever)?

GENERATE text: "The cumulative meta-analysis shows that the effect estimate has remained stable since the addition of the [Xth] study in [year], suggesting sufficient cumulative evidence."

Step 5: Living Review Setup

If the review needs ongoing surveillance:

1. Set up automated search alert (My NCBI, Google Scholar, Scopus)
2. Define update threshold: "If ≥1 new eligible study published, escalate for full update"
3. Schedule: quarterly search re-runs
4. Update protocol: search date filter + screening + extraction + MA update

GENERATE sr-living-review-protocol.md

---

Scripts

scripts/tsa_analysis.R

Trial sequential analysis using rpact or ldbounds. Computes RIS, O'Brien-Fleming monitoring boundaries, cumulative Z-curve. Usage: Rscript tsa_analysis.R --data ma.csv --event-rate 0.10 --rrr 0.25

scripts/cumulative_ma.R

Cumulative meta-analysis forest plot with studies ordered chronologically. Usage: Rscript cumulative_ma.R --data ma.csv --date-column year

scripts/pma_protocol.py

Generates PMA protocol template with data-sharing agreement clauses. Usage: python3 pma_protocol.py --title "Vasopressors in Sepsis" --investigators smith,jones,lee

---

Assets

assets/tsa-interpretation-guide.md

How to read TSA plots: Z-curve, monitoring boundaries, futility boundaries, RIS.

assets/pma-protocol-template.md

Full PMA protocol template including: objectives, PICO, eligibility, analysis plan, data-sharing agreement, governance.

assets/living-review-protocol.md

Living review protocol: search schedule, update threshold, screening protocol, analysis update procedure.

---

Guardrails

1. "TSA requires pre-specifying the expected effect size (RRR). If derived from pilot data, state this and do sensitivity analysis."
2. "Cumulative MA is sensitive to publication bias — early positive studies may be overrepresented."
3. "PMA requires commitment from all investigators BEFORE results are known — if any investigator unblinds, the PMA is compromised."
4. "Living reviews require dedicated funding and personnel — do not propose without resources."
5. "If TSA shows futility boundary crossed, early stopping may be justified."
6. "TSA is NOT a replacement for GRADE — it addresses imprecision (statistical certainty), not overall certainty."

Handoff

→ sr-interpretation (conclusive/inconclusive evidence), → sr-update (living review setup)