BlackCat Bio — Statistical Analysis Plans · Regulatory Biostatistics
We write Statistical Analysis Plans for trials that borrow strength from data outside the randomized comparison: single-arm studies anchored to an external control, and randomized trials augmented with registry or historical patient-level data.
External and augmented control designs let a sponsor strengthen the evidence from a small or hard-to-enroll trial by drawing on patient-level data from prior clinical trials, registries, or other real-world sources. These designs add precision and partially correct for confounding, but they also invite the biases regulators worry about if the methodology isn't locked down in advance.
The Statistical Analysis Plan is where this gets decided. A credible plan specifies, before anyone looks at outcomes, how external patients are selected, how the comparison is constructed, how residual bias is detected and limited, and how much weight the borrowed data are allowed to carry. We write that plan, grounded in designs that have already cleared FDA review.
Every plan begins with analysis plans that regulators have already accepted. We fit them to the specifics of your study.
We begin from SAP templates drawn from trials that led to FDA approval in the same indication, ideally trials that used the same family of external- or augmented-control methodology.
Starting from designs already vetted in review gives the plan a defensible spine: the estimand framing, control-arm construction, bias diagnostics, and decision rules all come from language that has cleared the bar.
We then map that template onto your protocol: its endpoints and estimands, eligibility criteria, visit schedule, available external data sources, and the operating characteristics you need for the decision at hand.
The result is a plan tailored to your study yet rooted in precedent. Every adaptation is a deliberate, documented departure from a known-acceptable design.
These are the components of a plan for an external or augmented control design, from selecting source patients through anchoring inference back to the randomized result.
We screen external or registry patients to closely match your protocol's eligibility criteria, defining the index visit, requiring aligned follow-up, and applying temporal eligibility so background therapy and assay methods are consistent with the trial era.
Intercurrent events are handled per the protocol's estimand. Missing endpoint values are addressed with pre-specified rules: multiple imputation, and tightly bounded interpolation only where flanking measurements permit.
We balance source and trial patients with propensity-score methods: matching where feasible, inverse-probability weighting as fallback, plus trimming and caps so no single patient dominates. We report the resulting effective sample size.
We compare the external control to the randomized control to estimate residual bias, then apply a graduated correction. The plan borrows more when bias is small and less when it is large, so borrowed data never override the randomized signal.
The primary estimate combines the randomized comparison with the partially bias-corrected external-control comparison, weighted by the information each contributes. The combined result stays anchored to the randomized trial.
All methods, thresholds, and decision rules are finalized in the SAP before any outcome data are examined. We document the regulatory grounding: FDA RWD/RWE guidance, ICH E10, and established methods literature.
Our principal led the creation of the Synthetic Control Arm (SCA) database, a novel approach for the analysis of single-arm trials that statistically matches experimentally treated patients to standard-of-care patients drawn from completed trials across sponsors.
The offering leveraged a clinical data repository spanning thousands of oncology trials to form a control arm of patients whose baseline disease and demographic characteristics match those of the experimentally treated patients in a present-day single-arm trial.
Non-small cell lung cancer (NSCLC) case study examining whether results in a randomized control arm are replicated by a synthetic control arm (SCA). Journal of Clinical Oncology, 37(15_suppl), 9108. View abstract →
The broader Friends of Cancer Research SCA project is described in the 2018 white paper. Read the white paper →
Separate, more recent project work applying external- and augmented-control methods beyond the original SCA program.
Designed an augmented control arm strategy for an ILD study, supplementing a randomized placebo arm with matched registry patients while retaining the randomized comparison as an internal benchmark for bias detection and partial correction.
Used propensity-score matching to compare 30-day mortality between patients who received andexanet alfa for life-threatening bleeding related to direct oral anticoagulants (from the single-arm ANNEXA-4 trial) and patients who received prothrombin complex concentrate (from the ORANGE registry).
Send us your protocol and the decision you need to make. We'll tell you whether an external or augmented control design fits.