Skip to Main content Skip to Navigation
Master Thesis

Développement d’un modèle prédictif de tuberculose pulmonaire parmi des patients suspects de tuberculose en zone de faible prévalence : une étude cas-témoins

Abstract : In low prevalence area for pulmonary tuberculosis infection (p-TB), predictive value of molecular-based detection method (PCR) might be low. The aim was to develop a predictive model of p-TB to identify patients at high risk of p-TB. Case-control study of consecutive patients hospitalized between 2009 and 2012, in a Parisian university hospital. Cases were patients with a culture-confirmed p-TB, regardless of smear result. Controls were patients with a suspected p-TB (direct exam and negative culture), matched to cases on a 1:1 ratio, by year and department of admission. Immunocompromised patients were not included. Factors associated with p-TB were screened using a stepwise conditional logistic regression. The score was developed using final’s model estimates. Discrimination (ROC curve, C - index / AUC), calibration (Hosmer-Lemeshow test) and internal validity using a bootstrap method were assessed. A sensitivity analysis which evaluated the score calibration in the subset of p-TB patients with negative direct exam was performed. Results: 354 patients were included. Age was forced in all analysis. Factors associated with p-TB in multivariate analysis were: age, diabetes, absence of cough +/- sputum, fever >15 days, chest X-rays findings. Score values were ranging from 0 to 8.4. The score was discriminant (0.84 C-index), well calibrated (p>0.05) and a score value higher than 3.3 can be used as a threshold (90% sensitivity, 50% specificity). The internal validity was good (0.79 corrected C-index). Score was well calibrated (p>0.05) in patients with negative direct exam. The developed score had good diagnostic performances and could be useful to determine when a PCR or a preventive respiratory isolation should be performed. An external validation is needed.
Document type :
Master Thesis
Complete list of metadatas

Cited literature [31 references]  Display  Hide  Download

https://dumas.ccsd.cnrs.fr/dumas-01305601
Contributor : Université Paris Descartes - Scd <>
Submitted on : Thursday, April 21, 2016 - 2:38:55 PM
Last modification on : Friday, March 27, 2020 - 2:34:36 AM
Long-term archiving on: : Friday, July 22, 2016 - 12:20:11 PM

Identifiers

  • HAL Id : dumas-01305601, version 1

Citation

Tristan Delory. Développement d’un modèle prédictif de tuberculose pulmonaire parmi des patients suspects de tuberculose en zone de faible prévalence : une étude cas-témoins. Médecine humaine et pathologie. 2015. ⟨dumas-01305601⟩

Share

Metrics

Record views

139

Files downloads

121