Análise de um modelo preditivo para o diagnóstico de alergia a fármacos baseado na história clínica

Abstract

8,4 a 96,4%) para o grupo dos não alérgicos, não se encontrando um limiar com bom poder discriminativo entre as duas situações (curva ROC). Conclusões: O modelo testado revelou um mau desempenho nesta população, sugerindo que os modelos preditivos necessitam de aperfeiçoamento para virem a ser uma ferramenta útil no estudo alergológico das reações a fármacos.Palavras -chave: Alergia a fármacos, hipersensibilidade a fármacos, estudo alergológico, modelos preditivos, algoritmos de decisão. ABSTRACTBackground: Allergic drug reactions represent an important public health problem associated with a significant mortality and morbidity. The allergy workup in this situation is expensive and time consuming. A predictive model using medical history would allow us to assess the risk of a positive result and might simplify the diagnostic process. Aims: To assess the performance of the Hierro Santorino predictive model in patients from our Immunoallergology department investigated for drug allergy. Methods: We included in our study patients referred to our department for drug allergy whose investigation was concluded between January 2017 and June 2018. We collected clinical data identified as the predictive factors in the mentioned model, as well as the results from the allergy work up. The Hierro Santorino model was then applied to our population and its performance was evaluated. Results: We analyzed 159 cases corresponding to 143 patients, 54 (37.8%) females, mean age of 42.1 ± 25.4 years.In 108 cases there was only 1 drug involved. In 39% of the cases a beta -lactam was the suspected drug, in 31% a nonsteroidal anti -inflammatory drug (NSAID) and in 29.6% other drug classes were suspected. In 18.2% of the cases the final diagnosis was positive for drug allergy, in 15.7% "NSAID intolerance" was established, 5% of the patients had "angiotensin -converting -enzyme inhibitor intolerance" and in 54.1% drug allergy was excluded. As the model under study was applied the mean probability of drug allergy was 73.3% (26.6% to 91.7%) in the allergic group and 68. 7% (8.4% to 96.4%) in the non -allergic group. There was no cut -off value with capacity to discriminate between the outcomes. Conclusions: The model tested in our population revealed a poor performance suggesting that predictive models still need improving to be used as a tool in the allergy workup.

Fatores preditivos de descompensação da pessoa em situação crítica no serviço de urgência

“…Este contexto justifica, portanto, a tentativa de construção de modelos preditivos. A utilização destes modelos (cuja origem provém do setor económico) tem sofrido nos últimos anos uma ampla generalização a RIIS Revista de Investigação & Inovação em Saúde quase todos os setores de atividade, incluindo a medicina (Cardoso et al, 2020).…”

“…Os modelos preditivos assumem cada vez maior importância no apoio à decisão médica na prática clínica e na gestão dos serviços (Barbini et al, 2014), e a sua aplicabilidade vai desde a área da gestão às do diagnóstico e decisão, incluindo decisões de estratégias preventivas e decisões terapêuticas individuais (Cardoso et al, 2020).…”