Validation of lactational mastitis diagnosis codes in electronic health care data

Background: Lactational mastitis is an adverse event that impacts the wellbeing of lactating persons and their infants. Human safety data on lactational mastitis following perinatal medication and vaccine exposures are needed. Electronic health record (EHR) data are an underutilized data source for conducting lactation-related research that may be useful for studying mastitis.

Objectives: To assess the validity of lactational mastitis diagnosis codes in automated EHR data by chart review.

Methods: We conducted a retrospective study using EHR data from three large integrated health care delivery systems in the United States. Of 19,660 patients who met inclusion criteria (i.e., had a live birth between December 2020-September 2022, linked to their infant who had ≥1 well-visit before 60 days of life, and had EHR documentation of ever breastfeeding in infant or maternal records), we identified those with an inpatient or outpatient International Classification of Diseases, Tenth Revision Clinical Modification (ICD-10)-CM diagnosis code for mastitis (N61.0 or O91.2x) in the 6 months after delivery. We performed medical record reviews on a random sample of ~10% of those with a mastitis diagnosis code to determine whether they met definitions for “possible mastitis” defined as having breast symptoms (e.g., breast pain) without systemic symptoms (e.g., fever) and no alternative diagnosis; “probable mastitis” defined as having breast symptoms, systemic symptoms, and no alternative diagnoses; or neither. For all patients with a mastitis code, we assessed whether they had an outpatient pharmacy dispensing for a systemic antibiotic typically used to treat mastitis within 3 days of the code. We report the prevalence of mastitis codes and positive predictive values (PPV) with 95% Wilson Confidence Intervals (CI) for having possible or probable mastitis.

Results: Among all individuals, 1,023 (5.2%) had a diagnosis code for mastitis and 4.2% had a code plus antibiotic dispensing. In the chart review sample (n=120), the PPV for probable mastitis was 75.3% (95% CI: 66.6, 82.3) and for probable or possible mastitis it was 95.9% (95% CI: 90.7, 98.3). Restricting the sample to those with an antibiotic (n=87), PPVs improved to 81.3% (95% CI: 71.5, 88.3) for probable mastitis and 100% (95% CI: 95.8, 100) for probable or possible mastitis.

Conclusions: Diagnosis codes alone may be used to identify lactational mastitis from EHR data, although PPV improves when requiring an antibiotic dispensing. Investigators should consider the intended application, the impact of the magnitude of misclassification on results, and the availability of antibiotic dispensing data when selecting an approach to identify lactational mastitis in electronic health data.