Epidemiologic Exposure Lagging¶
Exclude a declared recent interval of exposure before an outcome index time when analyzing older exposure under a timing assumption.
Core Idea¶
Epidemiologic exposure lagging excludes a declared recent interval of recorded exposure before a health outcome or matched index date, then analyzes the eligible older exposure.[1][2] The analyst chooses the interval for a stated timing reason, such as allowing a minimum induction period or reducing possible protopathic bias.[1][2] The choice is an analysis rule. It does not measure an individual's biological latency, prove that the excluded exposure was harmless, or establish that bias was removed.
Five parts identify the method: a dated exposure history, an outcome or matched index time, a recent interval excluded by a cutoff, older exposure retained for analysis, and a reason for treating the recent exposure differently. Remove the cutoff and the history is not lagged; remove the index and “recent” has no anchor. The general cutoff operation instantiates the Prime Truncation, while the health-outcome timing rationale keeps this entry specific to epidemiology.
Scope of Application¶
INWORKS estimated cumulative colon dose from external photon monitoring for a cancer-mortality study.[1] Its main analysis excluded the most recent ten years of dose to allow a minimum induction and latency period; five- and fifteen-year alternatives tested sensitivity.[1] Coupland and colleagues studied anticholinergic prescription records and dementia.[2] Their main exposure window counted prescriptions one to eleven years before a dementia diagnosis or matched control index date, omitting the latest year to reduce possible protopathic bias.[2] Those studies use different exposure measures and reasons for their cutoff; neither interval should be copied automatically into the other study.
Clarity¶
Say exactly which index date, exposure-record unit, recent interval, and older window the analysis uses. A prescription record documents prescribed exposure, not confirmed dispensing or ingestion.[2] The cutoff's equality and any record spanning it need a study-specific convention. With the same index, oldest bound, records, and boundary convention, a longer recent exclusion retains a subset of the records kept by a shorter exclusion. Coupland's reported three-to-thirteen- and five-to-twenty-year sensitivity windows also change the older bound, so the published windows are not themselves a fixed-bound nesting test.[2]
Manages Complexity¶
A longitudinal record may contain many exposures before one outcome. The declared cutoff makes eligibility reproducible: dated contributions on the older permitted side enter the exposure summary; recent ones do not. The discarded records are absent from that predictor, not necessarily erased from the source database or proved causally irrelevant. A longer lag can discard relevant information as well as potentially reduce attribution of exposure too close to disease onset. Report alternate windows and the reason for each choice.
Abstract Reasoning¶
First choose the outcome or matched index. Place eligible atomic exposure contributions on the time axis, declare an excluded recent interval, and retain the older contributions under a stated boundary convention. Then calculate the study's exposure summary from the retained records and compare results under defensible alternative windows. INWORKS aggregated estimated dose; Coupland aggregated standardized daily doses derived from prescription records.[1][2] Computing that summary is a later step than the cutoff. A fixed lag can bias an effect estimate when individual latency varies; selecting the lag that maximizes an observed association can also bias the estimate.[3]
Knowledge Transfer¶
The five-part audit transfers between occupational radiation records and medication prescriptions even though their outcome, exposure unit, and timing premise differ. The lag length does not transfer without new justification. The portable boundary–retain–discard skeleton belongs to Truncation; the epidemiologic attribution problem belongs to this specialist entry.
Example¶
In the INWORKS main analysis, the dated history is estimated external-photon colon dose, the later outcome is cancer mortality, the excluded interval is the recent ten years, and the retained dose is the older cumulative history.[1] The purpose is an assumed minimum induction and latency allowance, not proof of any worker's exact latency.[1] In Coupland's main analysis, the dated history is anticholinergic prescriptions, the index is dementia diagnosis or a matched control date, the excluded interval is the final year, and the retained measure uses prescriptions one to eleven years before that index.[2] Concern about treatment of early, undiagnosed symptoms supplies the rationale, not a causal conclusion about those medicines.[2]
Relationships to Other Abstractions¶
Current abstraction Epidemiologic Exposure Lagging Domain-specific
Parents (1) — outgoing typed relations
-
Epidemiologic Exposure Lagging is a kind of Truncation Prime
Exposure lagging applies a time cutoff to retain older exposure records and omit a recent terminal interval from the analyzed history.
Hierarchy path (1) — routes to 1 parentless root
- Epidemiologic Exposure Lagging → is a kind of Truncation → is a kind of Transformation → is a kind of, typical Function (Mapping)
Neighborhood in Abstraction Space¶
Epidemiologic Exposure Lagging sits in a sparse region of the domain-specific corpus (61st percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Clinical Trial Design & Drug Safety (23 abstractions)
Nearest neighbors
- Kaplan–Meier estimator — 0.87
- Continuous Individualized Risk Index — 0.85
- Spectrum Bias — 0.85
- Survival Analysis — 0.84
- Cumulative Dose — 0.84
Computed from structural-signature embeddings · 2026-10-25
Not to Be Confused With¶
- Biological latency: an actual delay between exposure and effect; the analysis cutoff is chosen, not observed in each participant.
- Distributed-lag regression: it can weight several time-specific predictors without omitting the latest interval.
- Moving an outcome date: Poulsen and colleagues describe subtracting a year from diagnosis or censoring dates, with exceptions; that changes the analysis endpoint rather than directly removing recent exposure records from a fixed-index predictor.[4]
- Proof of bias removal: sensitivity to the cutoff reveals dependence on an assumption, not the absence of reverse causation or other bias.
References¶
[1] Richardson and colleagues, “Risk of cancer from occupational exposure to ionising radiation: retrospective cohort study of workers in France, the United Kingdom, and the United States (INWORKS)”, BMJ 351 (2015), h5359, especially Statistical methods (estimated colon dose, ten-year lag, five- and fifteen-year sensitivity analyses), Table 2 note, and Results sensitivity paragraph. Full original cohort study. The dose is estimated from monitoring records. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g
[2] Carol A. C. Coupland et al., “Anticholinergic Drug Exposure and the Risk of Dementia: A Nested Case-Control Study”, JAMA Internal Medicine 179(8) (2019), 1084–1093, doi:10.1001/jamainternmed.2019.0677, especially Methods > Selection of Cases and Controls, Exposures, Statistical Analysis, and Discussion limitations. Full original publisher article; prescription records are not confirmed ingestion. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i
[3] David B. Richardson, Stephen R. Cole, Haitao Chu, and Bryan Langholz, “Lagging Exposure Information in Cumulative Exposure-Response Analyses”, American Journal of Epidemiology 174(12) (2011), 1416–1422, doi:10.1093/aje/kwr260. Publisher abstract inspected for fixed-lag and lag-selection cautions; full publisher text was not inspected for this draft. registry ↩
[4] Poulsen and colleagues, “Proton pump inhibitors and risk of gastric cancer: a population-based cohort study”, British Journal of Cancer 100(9) (2009), 1503–1507, doi:10.1038/sj.bjc.6605024, especially printed p. 1504 Statistical analysis and p. 1505 Table 2 note. Full original publisher PDF; cited solely for the endpoint-date near-miss, including its stated exceptions. registry ↩