High-dimensional Propensity score

Origin

(Schneeweiss et al. 2009)

Key idea

Schneeweiss et al. 2009 extended to a variety of classifications to code diagnoses (ICD), procedure (CPT), medications (eg, NDC, AHFS, ATCC), or others (PCP, LOINC).

CPT-4 (Current Procedural Terminology, 4th edition), ICD-9 (International Classification of Diseases, 9th edition), PCP visits (Primary Care Physician visits), NDC (National Drug Code), and ATC (Anatomical Therapeutic Chemical classification) are all codes or measures commonly used in healthcare and medical research.

Schneeweiss et al. 2018 Clinical Epidemiology: CC BY license

(Schneeweiss 2018)

Adjust useful proxies

In administrative data sources, the main idea of hdPS (high-dimensional propensity score) is to adjust for proxies that are empirically associated with the outcome of interest, which may not be directly measured in the data.

With hdPS, users do not need to know which unmeasured confounders are being adjusted for by proxy information.

• Adjusting for something that may not be interpretable directly with the context of the research question.

• Logic: measures from same subject should be correlated = has relevant proxy information

Schneeweiss, Sebastian. 2018. “Automated Data-Adaptive Analytics for Electronic Healthcare Data to Study Causal Treatment Effects.” Clinical Epidemiology, 771–88.

Schneeweiss, Sebastian, Jeremy A Rassen, Robert J Glynn, Jerry Avorn, Helen Mogun, and M Alan Brookhart. 2009. “High-Dimensional Propensity Score Adjustment in Studies of Treatment Effects Using Health Care Claims Data.” Epidemiology (Cambridge, Mass.) 20 (4): 512.