Quantifying diagnostic performance: The log odds method as a practical option

Abstract 

Purpose

To investigate the most practical option available when statistically analysing small sample size studies.

Method

The log odds ratio was applied to a study which took a retrospective approach and involved 50 patients who had undergone a ventilation and perfusion scan in the period between 01 January 2002 and 31 December 2002. Radiology reports and the patients' medical records were obtained, so that the clinical indications for the examination and the patients' medical history could be transferred onto the report form. Four radiographers who are routinely rotated into the Nuclear Medicine Department independently reported the same 50 ventilations and perfusion scans under identical conditions to the reporting radiologist.

Results

Taking the radiologist reports as the gold standard the radiographers had an accuracy rate of 78%. The radiographers recorded an average sensitivity value of 94% and specificity value of 63%, with an A_z reading of 0.71.

Conclusion

As the sample size was too small for a receiver operating characteristic (ROC) curve to be drawn, the log odds ratio provided a value, which represents an area under the receiver operating characteristic (ROC) curve, which in turn ensured the results were statistically reliable. The log odds ratio enables researchers to carry out small-scale research without harbouring the data collected by unreliable analysis.

Keywords: Radiographer reporting, Diagnostic performance test, ROC analysis, Log odds ratio, Ventilation and perfusion scans