A carregar...
Publicação
Mortalidade a 30-dias e 1-ano após fratura da extremidade proximal do fémur : avaliação de modelos preditivos
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.33 MB | Adobe PDF |
Orientador(es)
Resumo(s)
Contexto: A fractura da extremidade proximal do fémur é considerada um problema de saúde pública no idoso, caracterizado por uma incidência crescente com impacto na mortalidade, morbilidade e custos de saúde. A aplicação de escalas e modelos preditivos para identificação de doentes de risco pode melhorar os cuidados de saúde prestados, assim como os resultados clínicos. Este trabalho tem como objectivos a apreciação da mortalidade intra-hospitalar, 30-dias e 1-ano; a avaliação dos principais modelos preditivos de mortalidade e a elaboração de um novo modelo preditivo de mortalidade a 30-dias e 1-ano após a fractura.
Métodos: Coorte retrospectiva de doentes saídos do Hospital Professor Doutor Fernando Fonseca num período de 2 anos. Dados de 934 doentes (947 fracturas) foram recolhidos e analisados. Os modelos CCI, NHFS, O-POSSUM, E-PASS, Jiang et al., Holt et al. e AHFS foram calculados e comprados quanto ao seu poder discriminatório (curva ROC) e ajustamento (teste de Hosmer-Lemeshow). Na elaboração do novo modelo os doentes foram divididos em dois grupos: teste (n=708) e validação (n=239). A selecção dos factores de risco de mortalidade a 30-dias e a 1-ano foi executada através de uma regressão logística, seguida de análise multivariável para criação dos modelos preditivos. A validação consistiu na avaliação da capacidade discriminatória, ajustamento e comparação com os modelos previamente descritos.
Resultados: A mortalidade intra-hospitalar observada foi de 6,5%, a 30-dias de 5,2% e a 1-ano de 25,4%. Na avaliação dos modelos preditivos as componentes cirúrgicas foram excluídas por não apresentarem capacidade discriminatória. Na predição de mortalidade a 30-dias o melhor poder discriminatório foi da componente POSSUM fisiológico (0,826), no entanto sem diferença estatística relativamente ao AHFS (0,819), PRS (0,819) Holt et al. (0,814), Jiang et al. (0,760) e NHFS (0,755). Apenas o AHFS e a 2ª e 3ª calibração do NHFS mostraram ajustar-se à população estudada (p>0,05). Nenhum dos modelos preditivos revelou excelente capacidade discriminatória para mortalidade a 1-ano, apresentado todos falta de ajustamento à população estudada. O novo modelo preditivo de mortalidade a 30-dias incluiu os seguintes factores de risco: idade (≥ 88 anos), sexo (masculino), proveniência à admissão (internamento de agudos), ASA (IV), alteração cognitiva, desnutrição, insuficiência cardíaca (III ou IV da NYHA), cardiopatia isquémica, doença hepática crónica (moderada ou grave), alterações nos campos pulmonares da radiografia de tórax e alterações no ECG. Na sua validação apresentou excelente poder discriminatório (0,904) e ajustamento entre a mortalidade esperada e observada (p=0,152). No modelo preditivo de mortalidade a 1-ano constaram os seguintes factores de risco: idade (≥ 90 anos), sexo (masculino), proveniência à admissão (internamento de agudos), índice desempenho ECOG (2 ou 3 e 4), ASA (IV), Parker Mobility Score (≤ 4), alteração cognitiva, desnutrição, cardiopatia isquémica, diabetes mellitus com lesão de órgão alvo, neoplasia metastizada, cardiomegalia, alterações nos campos pulmonares da radiografia de tórax e alterações no ECG. A validação constatou excelente poder discriminatório (0,841) e um ajustamento entre a mortalidade esperada e observada (p=0,240).
Conclusão: O NHFS ajustado e o AHFS, apresentaram validade externa para a predição de mortalidade a 30-dias na população estudada. Para a mortalidade a 1-ano, os modelos analisados para além de revelarem apenas uma capacidade discriminatória aceitável, mostraram falta de ajustamento para a população estudada. Os modelos preditivos de mortalidade a 30-dias e 1-ano desenvolvidos e validados parecem ser superiores aos previamente descritos na literatura, no entanto carecem de avaliação prospectiva e validação externa.
Context: Hip fractures in the elderly are a public health issue. These fractures are characterized by their growing incidence and high mortality and morbidity, as well as costs for the healthcare system. The application of scores and predictive models to identify high risk patients can improve the quality of health care provided and consequently clinical outcomes. The aim of this study is to investigate the mortality after hip fracture in inpatient and at 30-days and 1-year; to evaluate the performance of the most widely studied risk models for predicting mortality; to identify the main risk factors and develop a new Mortality Risk Score to predict the mortality at 30-days and 1-year after hip fractures. Methods: A 2-year retrospective cohort of patients discharged from Hospital Professor Doutor Fernando Fonseca. Data from 934 patients (947 fractures) were collected and analyzed. The following models were calculated and compared regarding theirs discriminatory power (ROC curve) and calibration (Hosmer-Lemeshow goodness-of-fit test): CCI, NHFS, O-POSSUM, E-PASS, Jiang et al., Holt et al. e AHFS. During the development of the new models, patients were divided in two groups: test (n=708) and validation (n=239). Risk factors for 30-days and 1-year mortality were selected through a logistic regression. Additionally, a multivariate logistic regression was used for the creation of the predictive models. The validation of these models was achieved by comparison with previously described models, as well as analysis of the discriminatory performance and calibration. Results: The inpatient mortality was 6.5%, at 30-days 5.2% and at 1-year 25.4%. During the evaluation of the predictive models, the surgical components were excluded due to the lack of discriminatory capacity. In the mortality prediction at 30-days the POSSUM physiological component (0,826) showed the best discriminatory power. However, no statistic difference was observed regarding AHFS (0.819), PRS (0.819) Holt et al. (0.814), Jiang et al. (0.760) and NHFS (0.755). Only the AHFS and 2nd and 3rd calibrations of the NHFS presented a good fit (p>0.05). None of the predictive models showed excellent discriminatory power for 1-year mortality and all demonstrated lack of fit. The new Mortality Risk Score to predict 30-days mortality included the following risk factors: age (≥ 88 years), sex (male), pre-fracture residence (acute hospital ward), ASA (IV), cognitive impairment, malnutrition, heart failure (III or IV NYHA), ischemic heart disease, liver disease (moderate or severe), radiographic pulmonary alterations and ECG abnormalities. In its validation it showed excellent discriminatory power (0.904) and good concordance between observed and predicted mortality (p=0.152). In the 1-year new Mortality Risk Score, the following risk factors were contemplated: age (≥ 90 years), sex (male), pre-fracture residence (acute hospital ward), ECOG performance score (2-3 or 4), ASA (IV), Parker Mobility Score (≤ 4), cognitive impairment, malnutrition, ischemic heart disease, diabetes mellitus with target-organ damage, metastatic neoplasia, cardiomegaly, radiographic pulmonary alterations and ECG abnormalities. The validation showed excellent discriminatory power (0.841) and a goodness-of-fit between the observed and predicted mortality (p=0.240). Conclusion: The adjusted NHFS and the AHFS showed external validation for the 30-days mortality. For the 1-year mortality, the scores analyzed revealed only an acceptable discriminatory capacity and a lack of fit for studied population. The 30-days and 1-year Mortality Risk Scores developed and validated appear to be superior to the previously described models in the literature. Nevertheless, the models require prospective and external validation.
Context: Hip fractures in the elderly are a public health issue. These fractures are characterized by their growing incidence and high mortality and morbidity, as well as costs for the healthcare system. The application of scores and predictive models to identify high risk patients can improve the quality of health care provided and consequently clinical outcomes. The aim of this study is to investigate the mortality after hip fracture in inpatient and at 30-days and 1-year; to evaluate the performance of the most widely studied risk models for predicting mortality; to identify the main risk factors and develop a new Mortality Risk Score to predict the mortality at 30-days and 1-year after hip fractures. Methods: A 2-year retrospective cohort of patients discharged from Hospital Professor Doutor Fernando Fonseca. Data from 934 patients (947 fractures) were collected and analyzed. The following models were calculated and compared regarding theirs discriminatory power (ROC curve) and calibration (Hosmer-Lemeshow goodness-of-fit test): CCI, NHFS, O-POSSUM, E-PASS, Jiang et al., Holt et al. e AHFS. During the development of the new models, patients were divided in two groups: test (n=708) and validation (n=239). Risk factors for 30-days and 1-year mortality were selected through a logistic regression. Additionally, a multivariate logistic regression was used for the creation of the predictive models. The validation of these models was achieved by comparison with previously described models, as well as analysis of the discriminatory performance and calibration. Results: The inpatient mortality was 6.5%, at 30-days 5.2% and at 1-year 25.4%. During the evaluation of the predictive models, the surgical components were excluded due to the lack of discriminatory capacity. In the mortality prediction at 30-days the POSSUM physiological component (0,826) showed the best discriminatory power. However, no statistic difference was observed regarding AHFS (0.819), PRS (0.819) Holt et al. (0.814), Jiang et al. (0.760) and NHFS (0.755). Only the AHFS and 2nd and 3rd calibrations of the NHFS presented a good fit (p>0.05). None of the predictive models showed excellent discriminatory power for 1-year mortality and all demonstrated lack of fit. The new Mortality Risk Score to predict 30-days mortality included the following risk factors: age (≥ 88 years), sex (male), pre-fracture residence (acute hospital ward), ASA (IV), cognitive impairment, malnutrition, heart failure (III or IV NYHA), ischemic heart disease, liver disease (moderate or severe), radiographic pulmonary alterations and ECG abnormalities. In its validation it showed excellent discriminatory power (0.904) and good concordance between observed and predicted mortality (p=0.152). In the 1-year new Mortality Risk Score, the following risk factors were contemplated: age (≥ 90 years), sex (male), pre-fracture residence (acute hospital ward), ECOG performance score (2-3 or 4), ASA (IV), Parker Mobility Score (≤ 4), cognitive impairment, malnutrition, ischemic heart disease, diabetes mellitus with target-organ damage, metastatic neoplasia, cardiomegaly, radiographic pulmonary alterations and ECG abnormalities. The validation showed excellent discriminatory power (0.841) and a goodness-of-fit between the observed and predicted mortality (p=0.240). Conclusion: The adjusted NHFS and the AHFS showed external validation for the 30-days mortality. For the 1-year mortality, the scores analyzed revealed only an acceptable discriminatory capacity and a lack of fit for studied population. The 30-days and 1-year Mortality Risk Scores developed and validated appear to be superior to the previously described models in the literature. Nevertheless, the models require prospective and external validation.
Descrição
Tese de mestrado, Epidemiologia, Universidade de Lisboa, Faculdade de Medicina, 2019
Palavras-chave
Fractura extremidade proximal fémur Mortalidade Factores de risco Escalas de risco Modelos preditivos Teses de mestrado - 2019