Its very exciting that a research team in which we work have another academic paper published using our Natural Language Processing approach to unlock the power of clinical narrative. Our objective was to identify childhood respiratory tract-related illness presentation rates and service utilisation in primary care by interrogating free text and coded data from electronic medical records. We used a Retrospective cohort study and interrogation clinical narrative using a natural language processing software inference algorithm. We used data from 36 primary care practices in New Zealand collected from January 2008 to December 2013. We had records from 77,582 children that we reviewed to estimate the presentation of childhood respiratory illness and service utilisation. This cohort represented 268,919 person-years of data and over 650,000 unique consultations.
Our main aim was to describe Childhood respiratory illness presentation to primary care practice, with description of seasonal and yearly variation. Respiratory conditions constituted 46% of all child-general practitioner consultations with a stable year-on-year pattern of seasonal peaks. Upper respiratory tract infection was the most common respiratory category accounting for 21.0% of all childhood consultations, followed by otitis media (12.2%), wheeze-related illness (9.7%), throat infection (7.4%) and lower respiratory tract infection (4.4%). Almost 70% of children presented to their general practitioner with at least one respiratory condition in their first year of life; this reduced to approximately 25% for children aged 10–17.
This is the first study to assess the primary care incidence and service utilisation of childhood respiratory illness in a large primary care cohort by interrogating electronic medical record free text. The study identified the very high primary care workload related to childhood respiratory illness, especially during the first 2 years of life. These data can enable more effective planning of health service delivery. The findings and methodology have relevance to many countries, and the use of primary care ‘big data’ in this way can be applied to other health conditions.
The richness of the data sets contained in general practice clinical narrative is extraordinary. This paper demonstrates how we can use this narrative to identify multiple different conditions in historic clinical records; free of reliance on structured clinical coding.
This paper outlines the results. For more in depth detail of our method and algorithm, you can read our methodology paper.