Development of a multi-phase claims-based algorithm for pregnancy research
Recommended Citation
Phillips S, Johnson K, Shen SW, Woodcroft KJ, Oliveria SA, and Simon TA. Development of a multi-phase claims-based algorithm for pregnancy research. Pharmacoepidemiol Drug Saf 2017; 26:95-96.
Document Type
Conference Proceeding
Publication Date
8-22-2017
Publication Title
Pharmacoepidemiology and Drug Safety
Abstract
Background: Medication use during pregnancy may lead to birth defects or other complications. Safety studies are critical, with registries being common but often yielding a small number of cases after years of follow-up. Administrative claims data can be used to study large numbers of women and infants more quickly and efficiently, if these claims data accurately identify the following: pregnancy outcomes, gestational age, drug exposure by trimester, and mother/infant links. No single existing algorithm uses only administrative claims data to measure all of these variables.
Objectives: Develop a multi-phase algorithm for use in administrative claims data to identify live/nonlive pregnancy outcomes (phase 1), estimate gestational age (phase 2), estimate drug exposure by trimester (phase 3), and link claims data for mothers and infants (phase 4).
Methods: A multi-phase algorithm is being developed in a phased manner among women aged ≥15 and ≤50 years with ≥1 end of pregnancy (EOP) ICD-9 code with enrollment and prescription coverage 340 days prior to the end of pregnancy in the Henry Ford Health System between 1/1/2013 to 9/30/2015. In all phases, algorithms will be developed, applied to claims data, and compared to electronic medical records for validation. The best performing algorithm will be used in the next phase. For phase 1, we developed 3 algorithms: Alg1: ≥1 definitive ICD-9 EOP code; Alg2: ≥1 ICD-9 EOP code in the primary position; and Alg3: ≥1 ICD-9 EOP code in the primary position—and—≥1 procedure code. The positive predictive value (PPV), sensitivity, and 95% confidence intervals (CI) were calculated.
Results: A total of 698 women met inclusion criteria. In phase 1, the number of women and live births (LB) were as follows: Alg1—674 women, 529 LB; Alg2—658 women, 522 LB; and Alg3—589 women, 520 LB. (Nonlive outcomes not presented.) Overall algorithm PPV and sensitivity (95% CI in parens) were: Alg1—94% (CI: 92–96%); 99% (CI: 98–100%); Alg2—91% (88–93%); 97% (CI: 95–98%); Alg3—93% (CI: 90–95%); 87% (CI: 84–89%). Live outcomes PPV and sensitivity were: Alg1—98% (CI: 97–99%); 97% (CI: 96–99%); Alg2—92% (CI: 89–94%); 98% (CI: 97–99%); Alg3—93% (CI: 90–95%); 98% (CI: 96–99%). Nonlive outcomes PPV and sensitivity were: Alg1—79% (CI: 71–85%); 99% (CI: 95–100%); Alg2—72% (CI: 64–79%); 85 % (CI: 77–91%); Alg3—72% (CI: 60–83%); 41% (CI: 32–51%).
Conclusions: End of pregnancy outcomes can be identified in claims data with a high PPV and sensitivity. Further analyses are underway in the Alg1 cohort to develop algorithms for phases 2–4.
Volume
26
First Page
95
Last Page
96
