web analytics

Description

Bhattacharyya developed logistic regression equations for predicting the probability that a neck mass is neoplastic and/or malignant. The goal was to identify patients that should be biopsied quickly versus those that might be observed. The author is from Harvard Medical School in Boston.


 

Breakdown of patients studied:

• 95 patients were selected for regression analysis.

• Ages were from 2 to 85 years.

• The sex ratio was 2:1 female to males.

• 68% were reactive or non-neoplastic, 19% were benign neoplasms, and 13% were malignant.

 

The probability that the lesion is neoplastic (includes both benign and malignant tumors) can predicted from the following equations:

 

X =

= (0.0347 * (age in years)) + (0.0072 * (duration of lesion in weeks)) + (0.405 * (size in centimeters)) – 3.79

 

probability of neoplasia =

= 1 / (1 + EXP((-1) * X))

 

The probability that the lesion is malignant can be predicted from the following equations.

 

Y =

= (0.0510 * (age in years)) – 4.38

 

probability of malignancy =

= 1 / (1 + EXP((-1) * Y))

 

where:

• The probability equation shows the form "e!^x" and "e!^y" which appears to be a typographic error.

 

Performance of equations:

• Using a cutoff of 0.5 (a result > 0.5 indicates neoplastic), the neoplastic equation had a sensitivity of 46.7%, specificity of 87.7%, overall accuracy of 74.7%, positive predictive value of 63.6% and a negative predictive value of 78.1%.

• Using a cutoff of 0.25 (a result > 0.25 indicates neoplastic), the neoplastic equation had a sensitivity of 70%, specificity of 63.1%, overall accuracy of 65.3%, positive predictive value of 46.7% and a negative predictive value of 82%.

• Using a cutoff of 0.5 (a result > 0.5 indicates malignant), the malignant equation had a sensitivity of 0%, specificity of 100%, overall accuracy of 87.4%, positive predictive value NA, and a negative predictive value of 87.4%. The positive predictive value was undefined because none of the patients had a probability > 0.5.

• The performance of these equations are far from perfect. However, they may help separate patients into low and high risk groups when the patient does not give findings that points to an obvious diagnosis. It also may help in helping a patient make an informed decision when consenting to surgery.

 


To read more or access our algorithms and calculators, please log in or register.