Natural History, Not Lead Time

Jonas et al. analyze longitudinal data from a sample of individuals with schizophrenia who suffered delayed access (median DUP > 300 days) to, and “intermittent and inconsistent” treatment after, a first admission for psychosis. Despite inadequate aftercare, shorter delays from psychosis onset to first admission predicted better outcomes at 2, but unsurprisingly not 10 or 20 years later(1). We were thus puzzled with the authors’ conclusion: “The association between DUP and psychosocial function may be an artifact of early detection creating the illusion that early intervention is associated with improved outcomes.” We believe this incorrect inference reveals a conceptual confusion about lead time and other sources of bias.

Lead-time is usually conceptualized as the interval by which diagnosis is advanced to an earlier point in the natural history of a disease. Such early detection (ED) can occur via screening for asymptomatic disease or proactive case identification of already manifest illness(2). Lead-time bias is the spurious attribution of benefit to intervention offered early (during the lead time, EI) vs. later (upon usual presentation to care). Textbook examples include the illusory benefit of increased 5-year survival in individuals whose asymptomatic tumors (e.g. lung, breast) were identified earlier by screening programs, but who suffered similarly shortened life expectancies as those identified and treated later. For EI to meaningfully improve outcomes, the disease must be identifiable at a stage before it is usually recognized, and the ensuing intervention should be more effective when applied earlier in the illness course(2). This has been demonstrated for psychotic disorders: when DUP was successfully reduced, and followed with a model of care closer to modern standards, outcomes up to 10 years later were measurably improved, compared to samples not exposed to ED(3). Such prospective and controlled tests of ED can minimize lead time bias, but also other important sources of systematic error e.g. length-time bias (those who accept intervention may differ in illness duration and prognosis from those who do not) and compliance bias (those who accept intervention are prognostically different from those who do not) (4).

Observational studies are methodologically more vulnerable to such biases, as they are unable to manipulate the key variable of treatment timing, and must rely on data from patients who happen to present for care. Jonas et al. report on a self-selected or convenience sample not subjected to early detection (no lead-time was gained, so no such bias can logically ensue), but is instead more likely biased in the other ways outlined above. The results are better framed as approximating the natural history of schizophrenia in pre-modern systems of care i.e. the long-term outcomes of individuals who navigated to a first-admission after unacceptably long and likely aversive pathways to care, with which they subsequently engaged only erratically (e.g. self-reported antipsychotic use, 25% of the time). More than two decades ago such observations motivated innovations in specialty team based and youth oriented care models that have survived experimental, and inferentially stronger, tests of demonstrated improvements in both access(5) and care quality(6). Modern early intervention services that wed these two elements of ED with comprehensive care now offer the prospect of durable impact(7). While skepticism about such claims is necessary, it should motivate more studies that manipulate the relevant variables of timing and treatment quality.