Screening for asthma by population ranking: a validation study.

ABSTRACT

PURPOSE: A neural network system was previously developed to rank a population in order of asthma probability based on responses to a postal questionnaire. Respondents ranked higher than a percentage point screening threshold are offered clinical review. The present study validates this system in a new population that had not been involved in system development. METHODS: The system was used to rank respondents to a community survey and to predict positive predictive value (PPV) for percentage point thresholds between the top 1% and 10% of the ranking. Respondents in the top 10% were invited for clinical review. Review information and expert opinion was used to designate respondents as clinically "asthmatic" or "non-asthmatic." PPV prediction for each threshold was compared with clinical status of respondents. RESULTS: As the threshold increased from 1% to 10%, the additional yield of clinical asthmatics decreased, indicating a ranking in asthma probability order (all 7 in the top 1% were clinical asthmatics compared with 91% of the top 5% and 83% of the top 10%). Percentages of clinical asthmatics were generally slightly higher than system PPV predictions. CONCLUSIONS: The system ranked the population in asthma probability order and estimated PPV conservatively, enabling health-care providers to predict resource implications of a screening program.