Sleep and Breathing

, Volume 15, Issue 3, pp 431–437

Diagnostic characteristics of clinical prediction models for obstructive sleep apnea in different clinic populations

Authors

    • Division of Respiratory and Critical Care Medicine, University Medicine ClusterNational University Health System
  • Hze-Khoong Poh
    • Department of Oral and Maxillofacial Surgery, Faculty of DentistryNational University of Singapore
  • Yiong-Huak Chan
    • Biostatistics Unit, Yong Loo Lin School of MedicineNational University of Singapore
  • Wang-Jee Ngerng
    • Division of Respiratory and Critical Care Medicine, University Medicine ClusterNational University Health System
  • Dong-Xia Shi
    • Division of Respiratory and Critical Care Medicine, University Medicine ClusterNational University Health System
  • T. K. Lim
    • Division of Respiratory and Critical Care Medicine, University Medicine ClusterNational University Health System
Original Article

DOI: 10.1007/s11325-010-0354-3

Cite this article as:
Khoo, S., Poh, H., Chan, Y. et al. Sleep Breath (2011) 15: 431. doi:10.1007/s11325-010-0354-3

Abstract

Purpose

As predictive factors and their diagnostic values are affected by the characteristics of the population studied, clinical prediction model for obstructive sleep apnea (OSA) may exhibit different diagnostic characteristics in different populations. We aimed to compare the diagnostic characteristics of clinical prediction models developed in two different populations.

Methods

One hundred seventeen consecutive clinic patients (group 1) were evaluated to develop a clinical prediction model for OSA (local model). The diagnostic characteristics of this local model were compared with those of a foreign model by applying both models to another group of 52 patients who were referred to the same clinic (group 2). All patients underwent overnight polysomnography.

Results

The local model had an area under receiver operator characteristics curve of 79%. A cutoff of 0.6 was associated with sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of 77.9%, 72.5%, 84.5%, and 63.0%, respectively. The overall diagnostic accuracy, sensitivity, specificity, PPV, and NPV of the local model using data from patients in group 2 were 69.0%, 78.1%, 45.0%, 69.4%, and 56.3%, respectively. The foreign model had an overall diagnostic accuracy of 64.0% when applied to data from patients in group 2. At the optimal cutoff of 17, the foreign model was associated with sensitivity of 38.2%, specificity of 83.3%, NPV of 41.7% and PPV of 81.3%.

Conclusions

Clinical prediction model for OSA derived from a foreign population exhibits markedly different diagnostic characteristics from one that is developed locally, even though the overall accuracy is similar. Our findings challenge the predictive usefulness and the external validity of clinical prediction models.

Keywords

Clinical prediction modelsObstructive sleep apneaDiagnostic characteristicsDifferent populations

Abbreviations

AHI

Apnea–hypopnea index

BMI

Body mass index

CPAP

Continuous positive airway pressure

NPV

Negative predictive value

OSA

Obstructive sleep apnea

PPV

Positive predictive value

PSG

Polysomnography

ROC

Receiver operating characteristic

Copyright information

© Springer-Verlag 2010