A multiphase claims-based algorithm for live pregnancy outcomes research

Document Type

Conference Proceeding

Publication Date


Publication Title

Pharmacoepidemiology and Drug Safety


Background: Safety studies on outcomes of medication exposure during pregnancy are critical. Administrative claims data could be used however, these studies are often unfeasible due to the difficulty in identifying pregnancies, estimating gestational age (GA), estimating exposure during pregnancy and linking mothers to infants using only claims data.

Objectives: To develop a multiphase claims-based algorithm that identifies pregnancy outcomes (Phase [Ph]1), estimates GA (Ph2), estimates medication exposure (Ph3) and links mothers to infants (Ph4).

Methods: A multiphase algorithm was developed for use in women aged ≥15 and ≤50 years with ≥1 end of pregnancy (EOP) ICD-9 code and enrolment and prescription coverage 340 days prior to EOP in the Henry Ford Health System in the United States between 1 January 2013 and 30 September 2015. Algorithms were developed, applied to claims data and validated against electronic medical records. Positive predictive value (PPV), sensitivity (Sens) and 95% CI were calculated. The best-performing algorithm in each phase was used in the next phase. Ph1 results were presented at ICPE 2017 (abstract 154). Two Ph2 algorithms estimated GA at live birth as pre-term (245 days), term (273 days) or post-term (294 days) using EOP ICD-9 code dates adjusted with the dates of obstetric procedure codes. Three Ph3 algorithms evaluated exposure to long-term (eg, antidepressants) and short-term (antibiotics) medications overall and by trimester (using Ph2 estimated GA) using fill date and one of the following: days' supply (DS), DS plus a 14-day grace period or DS plus medication-specific grace periods. Three Ph4 algorithms assessed mother/infant linkage using combinations of birth/delivery dates, 3-digit ZIP codes, caesarean birth codes, delivery status (eg, singleton) and a screening hierarchy.

Results: Ph2 algorithms performed identically for term births (PPV: 94% [95% CI: 91, 96] Sens: 81% [76, 84]) and post-term births (PPV: 40% [32, 49] Sens: 90% [79, 96]). Algorithm 2 performed better for pre-term births (PPV: 94% [79, 99] Sens: 68% [52, 81]). Using Ph2 Algorithm 2, all Ph3 algorithms performed identically overall (PPV: 100% [98, 100] Sens: 100% [98, 100]). The lowest PPV for long-or short-term drug exposure by trimester was 91% (77, 98) (Sens: 100% [89, 100]). Ph4 algorithms had approximately 60% PPV. Work is ongoing to improve Ph4 algorithm performance.

Conclusions: A multiphase algorithm can be used with claims data to estimate GA at live birth and exposure to medications during pregnancy.




Supplement 2

First Page