Perinatal & Pediatric
Maternal Characteristics by Manual versus Automated Occupation Coding Among Infants with Cleft Lip and Palate Jenil Patel* Jenil Patel Omobola O. Oluwafemi A.J. Agopian Renata H. Benjamin David Gimeno Ruiz de Porras Charles Shumate
In association analyses of large datasets, it may be tempting to restrict to subjects for whom occupation may be coded using automated software without requiring more burdensome manual coding, but there are hypothetical concerns for selection bias. Thus, we aimed to compared maternal characteristics by manual versus automated occupation coding status. Utilizing data from the Texas Birth Defects Registry, we focused on infants with cleft lip and/or palate (1999-2009). The NIOSH Industry and Occupation Computerized Coding System (NIOCCS) was employed for automated coding, with manual coding for unclassified cases. Maternal characteristics were analyzed by manual vs. automatic occupation coding status. Logistic regression was used to explore associations between major occupation groups and the occurrence of orofacial clefts.
Automatic coding could be conducted on over 90% of all mothers. After excluding nonworking mothers, there were a total of 3,865 subjects analyzed (1,063 infants with CLP, 511 infants with CP, and 2,291 controls without birth defects). There was a significant demographic difference between automatically and manually coded control mothers by race/ethnicity (p=0.001), marital status (p=0.006), and diabetes status (p=0.009), but differences were not observed by maternal age at delivery, maternal BMI, maternal education, parity, and smoking. Specific occupation groups were associated with clefts, such as building, grounds cleaning, and maintenance occupations adjusted Odds Ratio (aOR): 2.21, 95% CI: 1.30, 3.76), as well as office and administrative support occupations (aOR: 0.77, 95% CI: 0.63, 0.93), after adjustment for maternal age at delivery, education, race/ethnicity, parity, any diabetes, and smoking during pregnancy. Notably, these associations persisted even after excluding manually coded occupations.
Association analyses were consistent before and after the exclusion of manually coded data, suggesting that machine learning, specifically the NIOCCS system, can play a valuable role in facilitating occupation-related birth defects research. The findings emphasize the reliability and efficiency of automated coding systems in large-scale epidemiological studies, providing insights into potential occupational factors contributing to cleft lip and palate.