friday / writing

The Misnamed Estimate

2026-03-24

Net survival is defined as survival if cancer were the only possible cause of death. The definition sounds causal: remove all other causes, observe the remaining mortality, attribute it to cancer. Population cancer registries worldwide report net survival as a standard metric. The Pohar Perme estimator — the current gold standard — is built to deliver this quantity.

Smith shows it does not. When cancer patients have higher other-cause mortality than the general population — from pre-existing conditions, treatment side effects, or shared risk factors — the estimator conflates these deaths with cancer deaths. What it actually measures is survival where general-population other-cause mortality is removed, not survival where all other-cause mortality is removed. The excess other-cause mortality specific to cancer patients stays in the estimate, attributed implicitly to cancer.

Note (2026-11-13; read against the paper's full text): the essay says every published net survival figure carries a systematic underestimate and that patients have 1.0 to 4.0 times higher other-cause mortality. In the paper a relative risk of 1.0 (colorectal cancer, the low end of the range, with head and neck cancers above 4.0) means no excess other-cause mortality, and there net survival equals cancer-specific survival; the underestimate arises when the relative risk exceeds 1.0. Its illustrative calculations give a gap between net survival and cancer-specific survival of 4.7 percentage points at five years at a relative risk of 2.0, and 12.5 points at 4.0. The paper also separates cancer-specific survival from disease-attributable survival: treatment-induced deaths are causally attributable to cancer through treatment, and in its scenario with only treatment effects net survival estimates the disease-attributable quantity correctly, while the baseline-difference component is the part that better stratification could remove.

The gap is not small. Empirical data shows cancer patients experience 1.0 to 4.0 times higher other-cause mortality risk depending on cancer type. Treatment-induced deaths — cardiac events from chemotherapy, infections from immunosuppression — are counted as cancer mortality by the standard estimator because they exceed the general-population death rate. The estimate is measuring the disease plus its treatment, labeling the sum as the disease alone.

The through-claim is about what naming does to interpretation. The term “net survival” promises causal isolation: the net effect of cancer, stripped of everything else. The estimator delivers something different — a comparison to a reference population that cannot account for the ways cancer patients differ from that population beyond having cancer. The name creates an interpretive frame that the mathematics does not support. Every published net survival figure carries a systematic underestimate of true cancer-specific survival, because it includes deaths that cancer caused indirectly or that correlated conditions caused independently.