Background: Congenital malformations (CMs) are key outcomes to evaluate drug safety during pregnancy. Valid algorithms are critical for their ascertainment in large scale real-world administrative data sources, where direct case confirmation is not always feasible.
Objectives: To systematically review the validity of CM algorithms in United States (US) claims, registry, and major surveillance databases reported in the literature.
Methods: A PubMed search was performed for studies published between 1/1/2010-1/20/2023. Citations from reviewed studies were also searched for inclusion. Eligible studies included algorithm(s) on CMs overall, CMs by organ system, or specific CMs. The algorithms must be based on structured coding systems such as International Classification of Diseases (ICD) and Current Procedural Terminology (CPT), and apply to pregnancies, fetuses, stillbirths, livebirths, and/or infants. The study must generate at least one quantitative validity estimate (e.g., positive predictive value [PPV]) in a US claims, registry, or surveillance database.
Results: The PubMed search yielded 467 studies, of which 118 passed abstract screen, and 14 were included after full-text review. Eight of the included studies were in commercial or Medicaid claims data, 4 in state or military registries, and 2 in Vaccine Safety Datalink (VSD). Twelve studies used algorithms based on ICD-9, with or without CPT, and 2 on ICD-10; very few algorithms apply to stillbirths. All studies used chart review as the gold standard. Sixty-eight validity estimates based on a sample of at least 10 patients were abstracted, with a largest sample being 4,772 patients. Specific CMs including oral cleft (6 estimates, 80-100%), gastroschisis (2 estimates, 96-100%), intestinal stenosis (2 estimates, 82-93%), ventricular septal defect (3 estimates, 76-95%), and hypospadias (3 estimates, 76-96%) consistently showed relatively high PPV. By organ system, orofacial (2 estimates, 91-97%) and cardiac defects (6 estimates, 68-91%) consistently showed moderate to high PPV. CMs of other organ systems and CMs overall showed widely varying PPV across studies (19 estimates, 17-100%). Within the same study, using a more stringent algorithm (e.g., requiring multiple occurrences of code, restricting to inpatient records) reduced false positive rate, but missed more true cases.
Conclusions: Among US claims, registry, and major surveillance databases, algorithms for CMs that are clinically apparent at birth or require postnatal repair consistently showed moderate to high PPV (i.e., at least around 70%) across data sources. Apart from orofacial and cardiac defects, algorithms for CMs of other organ systems and CMs overall may require further refinement and/or be supplemented by direct case confirmation.
