OHDSI Symposium Lightning Talk - Nega-ve Controls for Self-Controlled Case Series Design - James Weaver

Aug 26, 2026

Lightning talk by James Weaver (Janssen R&D) at the Observational Health Data Sciences and Informatics Symposium outlining OHDSI’s goal of generating reliable evidence. The presentation explains population-level effect estimation using causal inference criteria, then introduces the self-controlled case series (SCCS) design that compares outcomes within individuals under differing exposure. It details replication study specification for influenza, defining exposure and risk intervals, and assessing multiple negative control outcomes. Results include reported IRRs for diabetes and acute myocardial infarction plus a replication estimate for acute myocardial infarction.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages14

Reading Time35 min

AccessFree to download

Purpose of OHDSI

Causal effect identification and population-level effects

Self-controlled case series (SCCS) design

Assumptions: consistency, positivity, exchangeability

Replication study specification

Exposure and risk intervals

Outcomes and negative control outcomes

Results and replication findings

The value of nega-ve controls for the self-‐controlled case series design James Weaver Epidemiology Analy-cs, Janssen R&D Observa-onal Health Data Sciences and Informa-cs Symposium Bethesda, MD 12 October 2018 The purpose of OHDSI • To generate reliable evidence for the beneﬁt of pa-ents, providers, researchers, health care systems, industry, and government agencies The purpose of OHDSI • To generate reliable evidence for the beneﬁt of pa-ents, providers, researchers, health care systems, industry, and government agencies The purpose of OHDSI • To generate reliable evidence for the beneﬁt of pa-ents, providers, researchers, health care systems, industry, and government agencies IRR: 6.05 (95% CI 3.86, 9.50) Popula-on-‐level eﬀect es-ma-on • Methods for iden-fying causal rela-onships • AYempts to es-mate unbiased, average eﬀect of an exposure on an outcome • Compares outcome occurrence between comparable popula-ons • Causal eﬀect iden-ﬁca-on criteria • Consistency • Posi-vity • Exchangeability Self-‐controlled case series design (SCCS) • Compares outcomes within persons during-me periods of diﬀering exposure status • Consistency: • Time-‐stamped risk intervals and outcome events • Posi-vity: • Pa-ents must have both exposed-me and unexposed-me • Exchangeability: • Unexposed-me = counterfactual approxima-on Replica-on study speciﬁca-on • Exposure (inﬂuenza) • Dx code, +lab result • OP, any visit • No inﬂuenza past 60 days, no outcome during visit • Risk interval • Inﬂuenza visit end date +1 day to visit start date +7 days • Outcomes (ﬁrst MI + 31 nega-ve control outcomes ) • IP, any visit • Primary posi-on • Analy-c variants • Simple SCCS • Pre-‐exposure window • Age and seasonality adjustment • Event-‐dependent observa-on period correc-on Nega-ve control outcomes | Lung cancer | Inﬂammatory disorder of diges4ve system | | --- | --- | | HIV | Infec4ous disease of genitourinary system | | Depression | Infec4ous disease of abdomen | | Ingrowing nail | Func4onal ﬁnding of gastrointes4nal tract | | So@ 4ssue lesion | Edema | | Skin or mucosa lesion | Diabetes mellitus | | Organic mental disorder | Chronic inﬂammatory disorder | | Observa4on of colon | Chronic disease of genitourinary system | | Neoplas4c disease | Bleeding | | Musculoskeletal and connec4ve 4ssue disorder | Acute genitourinary disorder | | Mass of trunk | Abscess | | Mass of respiratory structure | Schizophrenia | | Mass of lymphore4cular structure | Bipolar disorder | | Lymphoprolifera4ve disorder | Substance abuse | | Lesion of brain | Trauma4c injury | | Inﬂammatory disorder of extremity | | NEJM results Diabetes mellitus IRR: 1.35 (95% CI 0.50, 3.62) Acute myocardial infarc4on IRR: 6.05 (95% CI 3.86, 9.50) Replica-on results Inﬂuenza diagnosis code (any visit), outcome (IP) Acute myocardial infarc4on IRR: 4.85 (95% CI 4.06, 5.74)

FAQs

What is the self-controlled case series (SCCS) design comparing?

SCCS compares outcomes within the same person during periods with different exposure status. This enables within-person comparisons across risk and non-risk intervals.

Which key criteria are listed for causal effect identification in the talk?

The talk highlights consistency, positivity, and exchangeability as causal effect identification criteria for estimating unbiased average effects of exposure on outcomes.

How are risk intervals and exposure defined in the influenza replication specification?

The exposure is defined using an influenza diagnosis code and lab result at an appropriate visit setting, with risk intervals from one day after the influenza visit end date to seven days before the visit start date. Additional rules exclude prior influenza and outcomes during the exposure visit.

Aug 26, 2026

Uploaded by wps_ap_test_251126_0180