Advanced analysis

Use these modules after defining the main scientific question. Availability and support vary by design; an advanced API name is not a support promotion.

Survival beyond a Cox summary

The package offers standard survival reports, diagnostics, RMST and milestone summaries, and bounded extensions for competing risks, longitudinal treatments, delayed entry and multistate settings. Check the current adapter’s data contract, time horizon, censoring assumptions and support status.

iobr_causal_survival_platform_report() and iobr_causal_survival_vnext_report() expose support plans, diagnostics, skipped and failed sections. Review Support_Status, Evidence_Class, Promotion_Blocker and the recommended next step with the numeric tables. Competing-risk and multistate paths have limited support; policy learning remains exploratory unless an exact supported configuration says otherwise.

Frozen OS mediation

Ordinary Response association and ordinary OS association are different from formal mediation. In the mechanism project governed by the frozen OS protocol, formal mediation is OS-only and HDMT is excluded.

Before executing mediation, check the gene’s saved results, cohort, endpoint, exposure/mediator definitions, covariates, model version and complete family. Reuse compatible frozen results. A missing or incompatible analysis requires a recorded reason and the same fixed framework used for the other genes.

The dedicated OS_MEDIATION_V1 adapter uses a frozen OLS mediator model and right-censored log-normal AFT outcome model with the recorded adjustment and scaling. Indirect and direct effects are on a log-survival-time scale; exponentiation gives a time ratio, not HR. Keep total, indirect and residual effects distinct. Response OR attenuation is auxiliary association evidence.

For that project, FDR below 0.10 has priority and 0.10–0.15 remains exploratory. Keep original results and contradictions. A statistical decomposition does not prove molecular function, a complete mechanism or causal identification. These are project-specific rules, not a universal restriction on every public API.

Feature selection and graphs

Feature selection prioritizes variables; DAG discovery explores possible structure under assumptions. Neither supplies biological validation or a verified adjustment set automatically. Record selection history and review same-cohort selection and self-inclusion before interpreting formal follow-up.

The function map lists the relevant public interfaces. Consult their installed R help for exact required inputs rather than adapting a small baseline example to a complex design by changing one label.