Using a gradient boosted model for case ascertainment from free-text veterinary records

Kennedy, Uttara; Paterson, Mandy; Clark, Nicholas J.

doi:10.1016/j.prevetmed.2023.105850

Cited by 1 publication

(2 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Each animal's record contained a (1) unique animal identification number, (2) demographic details such as age (as estimated at the time of entry into the shelter), neuter status on entry and gender, (3) incoming circumstance, date and region, (4) length of stay, (5) days spent in foster, (6) date and status of final outcome for that animal and (7) free-text veterinary clinical notes. In an earlier study, 17 the authors used natural language processing (NLP) on the clinical notes for each animal and generated a pattern-matching classifier that could accurately recognise and predict feline URT infections. NLP generated text-based predictors in the form of tokens (single words or two-word combinations).…”

Section: Study Setting and Subjectsmentioning

confidence: 99%

“…Preliminary work by the authors, using retrospective RSPCA Qld records has shown that using subsets, stochastic or crude approaches for case ascertainment of feline URT infections presents challenges due to incomplete or inconsistent records. To counter this limitation, the authors developed a validated pattern‐matching tool 17 to enable accurate case ascertainment from clinical data stored on RSPCA Qld's database from 1 January 2013 to 31 December 2020. The output from the pattern‐matching algorithm was a predicted infection probability for each animal, based on free‐text clinical records as stored on the RSPCA Qld database.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Epidemiological insights into the burden of feline upper respiratory tract infections in Queensland RSPCA shelters

Kennedy,

Paterson,

Clark

2023

Aust Veterinary J

Self Cite

View full text Add to dashboard Cite

IntroductionFeline upper respiratory tract infection (FURTI) is a severe problem in animal shelters where there is high turnover of populations and compromised immunity. This retrospective cohort study explores associations of potential animal‐based and environmental‐based factors with the risk of FURTI, where a previously modelled infection classification is used as the outcome of interest. The study type is a retrospective cohort and the measures of association include Odds Ratios and conditional predictions.ObjectivesTo gain epidemiological insights into variation in FURTI using retrospective data from one of Australia's leading animal shelters.MethodsWe stratified FURTI by admission and environmental variables. Predicted infection status, obtained using a machine‐learning classifier trained on clinical text (accuracy 0.95 [CI 0.92, 0.97]), was used as the outcome of interest. Prior assumptions were represented by a causal framework or a direct acyclic graph (DAG), which informed creation of multiple Bernoulli models with an observational and prior component.ResultsWe analysed 43,431 feline entries over 8 years. Males were 1.24 (95% CI 1.19 to 1.31) times more likely than females to be classified as positive, while already desexed animals were only 0.68 (95% CI 0.60 to 0.72) as likely to be classified as positive compared to those not desexed on entry. Cats (>4 months) were twice as likely (95% CI 1.91 to 2.09) as kittens (0–4 months) to be classified positive. Animals entering the shelter as seized by the inspectorate (n = 415) were more likely to be classified positive compared to animals from other sources. Predicted infection probability increased in winter and showed a linear pattern with how full the shelter was.ConclusionThis study estimates the association between animal and environmental variables of interest and FURTI classification status, thus better interpreting the distribution of disease as predicted by a previously uninterpretable model. This analysis gives much needed insight into the types of changes in an animal's environment that can impact final animal outcomes.

show abstract

Section: Study Setting and Subjectsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%