Development and validation of parental alert and reminder assistant: a prenatal expert system prototype

The Prenatal Alert and Reminder Assistant (PARA) was developed at the LDS Hospital at Salt Lake City, Utah, as part of the HELP hospital information system. PARA is the third expert system prototype to computerize the prenatal record and provide decision support to the obstetric team with patient specific alerts and reminders. Its development and validation are presented, as well as the studies into (1) the control of error propagation and (2) data modeling of pregnancy outcome. The experimental knowledge base with 25 rules was validated using a retrospective study of 50 prenatal charts (25 normal and 25 hypertensive pregnancies). PARA's knowledge base was validated with a mean score of 98.9% correct as compared to 97.8% for the obstetric team (p = 0.0216). The difference of 1.1% represents a 50% reduction in the error rate (2.2%) as compared to the obstetric team. Propagation of erroneous decision-support messages in a large clinical database was controlled through the use of Truth Maintenance techniques. These techniques included tagging messages as 'active' and 'inactive,' limiting the reasoning to active messages, and requiring each Medical Logic Module (MLM) in PARA's knowledge base to have rules that would assert the message was false, unless it could be asserted true. The patient charge for a labor-and-delivery admission was used as a crude index of pregnancy outcomes. A model of the relationships of certain pregnancy complications to this pregnancy outcome variable was developed. The regression model's 17 independent variables had a strong, positive correlation (R = 0.70) with the dependent variable, log hospital charges. The association was statistically significant (p = 0.0001). Forty-eight percent of the variance in the hospital charges were explained by the 17 variables (adjusted multiple R-square = 0.482).