(224) Validity of Healthcare Administrative Data to Identify Infants with Low Birth Weight

Publication Number: 555

Background: Low birth weight (LBW; < 2500 grams) is an important risk factor for infant mortality and childhood morbidity and is therefore frequently included as an outcome of interest in pregnancy drug safety studies. Healthcare utilization databases are increasingly used in perinatal pharmacoepidemiology. However, the validity of studying LBW based on recorded diagnosis is not well documented.

Objectives: The aim of this study was to evaluate the accuracy of the International Classification of Diseases, Ninth Revision (ICD-9) LBW codes recorded in healthcare claims.

Methods: We identified a nationwide sample of 462 infants born 2009-2014 in the Medicaid Analytic eXtract, who had Medicaid coverage at birth. Infants were sampled stratified by preterm status (N=201 preterm, N= 261 term). Medical delivery hospitalization records were retrieved, and birth weight as documented in these records was considered the gold standard. LBW was determined present in the claims if there was ≥1 ICD-9 code with a specified weight range < 2500 grams (765.01-08, 765.11-18, V21.31-35) within 30 days after birth. Performance characteristics – sensitivity (Se), specificity (Sp), positive predictive value (PPV) and negative predictive value (NPV) – of LBW codes against LBW classification based on weight documented in the medical records were calculated. Results were stratified by preterm status, and consistency between claims-based weight range and true weight was assessed.

Results: Among 201 infants born preterm were 100 with ≥1 LBW code (95 true positives) and 101 with no such code (82 true negatives), leading to a Se of 83.3%, Sp of 94.3%, PPV of 95.0% and NPV of 81.2%. Among 261 infants born at term, 11 had ≥1 code (10 true positives) and 250 had no LBW code (239 true negatives), resulting in a Se of 47.6%, Sp of 99.6%, PPV of 90.6% and NPV of 95.6%. Estimates overall, accounting for the stratified sampling, were 65.7%, 99.2%, 93.5% and 94.6% respectively. When looking at the actual weight, there was very good correspondence between weight range recorded in claims and weight documented in medical records, with 103 of all 105 true positives having claims-based weight ranges including the true weight. Twenty-five of the 30 false negatives had a true weight close to the 2500-gram cut point.

Conclusions: The high PPV for LBW codes in healthcare claims indicates that this outcome can be validly studied, and relative measures of association should be unbiased when misclassification is non-differential. Investigators should be aware that absolute risks and risk differences will be underestimated given the lower Se, in particular among term births. The weight ranges included in the ICD code description might be useful to assess LBW more granularly.